Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks389 from public data
- CodeBERT: A Pre-Trained Model for Programming and Natural Languages4,421
This work develops CodeBERT with Transformer-based neural architecture, and trains it with a hybrid objective function that incorporates the pre-training task of replaced token detection, which is to detect plausible alternatives sampled from generators.
- A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions3,965
A thorough overview of hallucination detection methods and benchmarks is presented and the promising research directions on LLM hallucinations are highlighted, including hallucination in large vision-language models and understanding of knowledge boundaries in LLM hallucinations.
- SemEval-2016 Task 5: Aspect Based Sentiment Analysis1,740
This paper describes the SemEval 2016 shared task on Aspect Based Sentiment Analysis (ABSA), a continuation of the respective tasks of 2014 and 2015, which attracted 245 submissions from 29 teams and provided 19 training and 20 testing datasets.
- Document Modeling with Gated Recurrent Neural Network for Sentiment Classification1,547
A neural network model is introduced to learn vector-based document representation in a unified, bottom-up fashion and dramatically outperforms standard recurrent neural network in document modeling for sentiment classification.
- Learning Sentiment-Specific Word Embedding for Twitter Sentiment Classification1,229
Three neural networks are developed to effectively incorporate the supervision from sentiment polarity of text (e.g. sentences or tweets) in their loss functions and the performance of SSWE is improved by concatenating SSWE with existing feature set.
- Aspect Level Sentiment Classification with Deep Memory Network1,046
A deep memory network for aspect level sentiment classification that explicitly captures the importance of each context word when inferring the sentiment polarity of an aspect, and is also fast.
- Effective LSTMs for Target-Dependent Sentiment Classification1,010
Two target dependent long short-term memory models, where target information is automatically taken into account, are developed, which achieve state-of-the-art performances without using syntactic parser or external sentiment lexicons.
- Learning Semantic Representations of Users and Products for Document Level Sentiment Classification378
By combining evidence at user-, product and documentlevel in a unified neural framework, the proposed model achieves state-of-the-art performances on IMDB and Yelp datasets1.
- 335
- HuaTuo: Tuning LLaMA Model with Chinese Medical Knowledge311
The experimental results demonstrate that HuaTuo generates responses that possess more reliable medical knowledge, and a LLaMA-based model that has been supervised-fine-tuned with generated QA (Question-Answer) instances is proposed.
- Learning Semantic Hierarchies via Word Embeddings289
This paper proposes a novel and effective method for the construction of semantic hierarchies based on word embeddings, which can be used to measure the semantic relationship between words.
- A language-independent neural network for event detection282
A language-independent neural network is developed to capture both sequence and chunk information from specific contexts and use them to train an event detector for multiple languages without any manually encoded features.
- Navigate through Enigmatic Labyrinth A Survey of Chain of Thought Reasoning: Advances, Frontiers and Future280
This paper systematically investigate relevant research, summarizing advanced methods through a meticulous taxonomy that offers novel perspectives, and delve into the current frontiers and delineate the challenges and future directions, thereby shedding light on future research.
- Coooolll: A Deep Learning System for Twitter Sentiment Classification251
A neural network with hybrid loss function 1 is developed to learn SSWE, which encodes the sentiment information of tweets in the continuous representation of words, which can be easily re-implemented with the publicly available sentiment-specific word embedding.
- Deep learning for sentiment analysis: successful approaches and future challenges207
An overview of the successful deep learning approaches for sentiment analysis tasks is provided, lay out the remaining challenges and provide some suggestions to address these challenges.
- Building Large-Scale Twitter-Specific Sentiment Lexicon : A Representation Learning Approach192
The sentiment lexicon (TS-Lex) is built with a dedicated neural architecture and integrates the sentiment information of text into its hybrid loss function for learning sentiment-specific phrase embedding (SSPE).
- Examining Inter-Consistency of Large Language Models Collaboration: An In-depth Analysis via Debate171
This work introduces a formal debate framework (FORD) to conduct a three-stage debate among LLMs with real-world scenarios alignment: fair debate, mismatched debate, and roundtable debate to examine whether LLMs can collaborate effectively to achieve a consensus for a shared goal.
- User modeling with neural network for review rating prediction163
The lexical semantic composition models are extended and a userword composition vector model (UWCVM) is introduced, which effectively captures how user acts as a function affecting the continuous word representation.
- Document representation and feature combination for deceptive spam review detection154
A neural network based model is presented to learn the representation of reviews and makes a hard attention through the composition from sentence representation into document representation, compute the importance weights of each sentence and incorporate them into the composition process of document representation.
- A Text-Centered Shared-Private Framework via Cross-Modal Prediction for Multimodal Sentiment Analysis151
This paper proposes a text-centered shared-private framework (TCSP) for multimodal fusion, which consists of the cross-modal prediction and sentiment regression parts and provides a new way to utilize the unlabeled data for multimodal sentiment analysis.
- 145
- A Survey on Dialogue Summarization: Recent Advances and New Frontiers127
A thorough review of dialogue summarization according to the characteristics of each domain, covering meeting, chat, email thread, customer service and medical dialogue, as well as an overview of publicly available research datasets.
- Molweni: A Challenge Multiparty Dialogues-based Machine Reading Comprehension Dataset with Discourse Structure125
The Molweni dataset is presented, a machine reading comprehension (MRC) dataset with discourse structure built over multiparty dialog with large-scale annotations in a modified Segmented Discourse Representation Theory (SDRT; Asher et al., 2016) style.
- Effective Deep Memory Networks for Distant Supervised Relation Extraction122
This paper introduces a novel neural approach for distant supervised RE with special focus on attention mechanisms, which includes two major attention-based memory components, which are capable of explicitly capturing the importance of each context word for modeling the representation of the entity pair.
- Target-Dependent Sentiment Classification with Long Short Term Memory.122
Two target dependent long short-term memory models, where target information is automatically taken into account, are developed, which achieve state-of-the-art performances without using syntactic parser or external sentiment lexicons.
- 114
- Language Model as an Annotator: Exploring DialoGPT for Dialogue Summarization114
DialoGPT, a pre-trained model for conversational response generation, can be developed as an unsupervised dialogue annotator, which takes advantage of dialogue background knowledge encoded in DialoG PT.
- TableGPT: Few-shot Table-to-Text Generation with Table Structure Reconstruction and Content Matching110
This work proposes TableGPT, a model that outperforms existing systems on most few-shot settings for table-to-text generation and exploits multi-task learning with two auxiliary tasks that preserve table’s structural information by reconstructing the structure from GPT-2's representation and improving the text's fidelity with content matching task.
- Is ChatGPT Equipped with Emotional Dialogue Capabilities?99
While ChatGPT's performance on emotional dialogue understanding may still lag behind that of supervised models, it exhibits promising results in generating emotional responses, and suggests potential avenues for future research directions.
- e-CARE: a New Dataset for Exploring Explainable Causal Reasoning97
Experimental results show that generating valid explanations for causal facts still remains especially challenging for the state-of-the-art models, and the explanation information can be helpful for promoting the accuracy and stability of causal reasoning models.
- Ensemble Learning for Heterogeneous Large Language Models with Deep Parallel Collaboration94
A training-free ensemble framework DeePEn, fusing the informative probability distributions yielded by different LLMs at each decoding step, which achieves consistent improvements across six benchmarks covering subject examination, reasoning, and knowledge.
- Constructing and Embedding Abstract Event Causality Networks from Text Snippets93
Experimental results demonstrate that the abstract causality network is effective for discovering high-level causality rules behind specific causal events, and the embedding models perform better than state-of-the-art link prediction techniques in predicting events.
- Semantic Parsing with Syntax- and Table-Aware SQL Generation83
The quality of the generated SQL query is significantly improved through learning to replicate content from column names, cells or SQL keywords; and improving the generation of WHERE clause by leveraging the column-cell relation.
- Dialogue Discourse-Aware Graph Model and Data Augmentation for Meeting Summarization81
A Dialogue Discourse-Dware Meeting Summarizer (DDAMS) to explicitly model the interaction between utterances in a meeting by modeling different discourse relations and a relational graph encoder, where the utterances and discourse relations are modeled in a graph interaction manner.
- Hierarchical Attention Flow for Multiple-Choice Reading Comprehension78
This paper presents the hierarchical attention flow to adequately leverage candidate options to model the interactions among passages, questions and candidate options and explicitly model the option correlations with attention mechanism to obtain better option representations.
- Large language models meet text-centric multimodal sentiment analysis: a survey77
A comprehensive review of recent research in text-centric multimodal sentiment analysis tasks is presented, the potential of LLMs for text-centric multimodal sentiment analysis is examined, and the challenges and potential research directions for multimodal sentiment analysis in the future are explored.
- Parsing Tweets into Universal Dependencies77
It is shown that it is challenging to deliver consistent annotation due to ambiguity in understanding and explaining tweets and proposed a new method to distill an ensemble of 20 transition-based parsers into a single one that achieves an improvement of 2.2 in LAS over the un-ensembled baseline and outperforms parsers that are state-of-the-art on other treebanks in both accuracy and speed.
- Topic-to-Essay Generation with Neural Networks72
A multi-topic aware long short-term memory network that learns the weight of each topic and is sequentially updated during the decoding process and has the ability to generate essays that are not only coherent but also closely related to the input topics.
- Multilingual dependency-based syntactic and semantic parsing72
The authors' CoNLL 2009 Shared Task system includes three cascaded components: syntactic parsing, predicate classification, and semantic role labeling, which wins the first place in the joint task, including both the closed and open challenges.
- Face-Sensitive Image-to-Emotional-Text Cross-modal Translation for Multimodal Aspect-based Sentiment Analysis70
A face-sensitive image-to-emotional-text translation (FITE) method, which focuses on capturing visual sentiment cues through facial expressions and selectively matching and fusing with the target aspect in textual modality, which is the first that explicitly utilize the emotional information from images in the multimodal aspect-based sentiment analysis task.
- An Iterative Emotion Interaction Network for Emotion Recognition in Conversations65
An iterative emotion interaction network is proposed, which uses iteratively predicted emotion labels instead of gold emotion labels to explicitly model the emotion interaction, and can effectively retain the performance advantages of explicit modeling.
- Table-to-Text Generation with Effective Hierarchical Encoder on Three Dimensions (Row, Column and Time)65
This work develops a table cell fusion gate to combine representations from row, column and time dimension into one dense vector according to the saliency of each dimension’s representation.
- Improving Low Resource Named Entity Recognition using Cross-lingual Knowledge Transfer65
A lexicon extension strategy is designed to address out-of lexicon problem by automatically learning semantic projections and regard word-level entity type distribution features as an external language-independent knowledge and incorporate them into the neural architecture.
- 63
- 59
- Adaptive Multi-pass Decoder for Neural Machine Translation59
A novel architecture called adaptive multi-pass decoder is proposed, which introduces a flexible multi- pass polishing mechanism to extend the capacity of NMT via reinforcement learning and adopts an extra policy network to automatically choose a suitable and effective number of decoding passes.
- Retrieval-Generation Synergy Augmented Large Language Models58
This work proposes an iterative retrieval-generation collaborative framework that is not only able to leverage both parametric and non-parametric knowledge, but also helps to find the correct reasoning path through retrieval- generation interactions, which is very important for tasks that require multi-step reasoning.
- Incorporating Commonsense Knowledge into Abstractive Dialogue Summarization via Heterogeneous Graph Networks56
A novel multi-speaker dialogue summarizer to demonstrate how large-scale commonsense knowledge can facilitate dialogue un-derstanding and summary generation and a Dialogue Heterogeneous Graph Network for modeling both information.
- Trends in Integration of Knowledge and Large Language Models: A Survey and Taxonomy of Methods, Benchmarks, and Applications55
A review to discuss the trends in integration of knowledge and large language models, including taxonomy of methods, benchmarks, and applications is proposed, with the intention of inspiring future research endeavors.
- The Role of "Condition"55
A semi-supervised Multi-Input Multi-Output sequence labeling model that learns complex dependencies between the sequence tags from multiple signals and generates output sequences for fact and condition tuples and has a self-training module of multiple strategies to leverage the massive scientific data for better performance when manual annotation is limited.
- 54
- Multimodal machine learning enables AI chatbot to diagnose ophthalmic diseases and provide high-quality medical responses52
This study developed and evaluated the ChatGPT-powered Intelligent Ophthalmic Multimodal Interactive Diagnostic System (IOMIDS) to enable patient self-diagnosis and self-triage to enable patient self-diagnosis and self-triage.
- Generalizing Syntactic Structures for Product Attribute Candidate Extraction51
A novel approach is proposed by generalizing syntactic structures of the product attributes with two strategies: intuitive heuristics and syntactic structure similarity, which shows that the proposed approach is effective.
- Knowledge-tuning Large Language Models with Structured Medical Knowledge Bases for Trustworthy Response Generation in Chinese48
Experimental results show that the LLMs which are knowledge-tuned with cMedKnowQA can exhibit higher levels of accuracy in response generation compared with vanilla instruction-tuning and offer a new trustworthy way for the domain adaptation of LLMs.
- Cross-Lingual Machine Reading Comprehension48
A novel model called Dual BERT is proposed, which takes advantage of the large-scale training data provided by rich-resource language and learn the semantic relations between the passage and question in bilingual context, and then utilize the learned knowledge to improve reading comprehension performance of low- resource language.
- Recent advances in deep learning based sentiment analysis47
A brief introduction to the recent advance of the deep learning-based methods in these sentiment analysis tasks, including summarizing the approaches and analyzing the dataset is given.
- Chisco: An EEG-based BCI dataset for decoding of imagined speech46
The Chinese Imagined Speech Corpus (Chisco), including over 20,000 sentences of high-density EEG recordings of imagined speech from healthy adults, is presented, representing the largest dataset per individual currently available for decoding neural language to date.
- Length Extrapolation of Transformers: A Survey from the Perspective of Positional Encoding46
A survey of extrapolatable PEs that have dominated this research field, covering position interpolation and randomized position methods, aims to enable the reader to gain a deep understanding of existing methods and provide stimuli for future research.
- How Does Selective Mechanism Improve Self-Attention Networks?46
Assessing the strengths of selective SANs (SSANs), which are implemented with a flexible and universal Gumbel-Softmax, shows that SSANs consistently outperform the standard SANs.
- Learning to Ask Unanswerable Questions for Machine Reading Comprehension46
A pair-to-sequence model for unanswerable question generation, which effectively captures the interactions between the question and the paragraph, and a way to construct training data for question generation models by leveraging the existing reading comprehension dataset is presented.
- Collocation Polarity Disambiguation Using Web-based Pseudo Contexts46
This paper proposes an unsupervised three-component framework to expand some pseudo contexts from web, to help disambiguate a collocation's polarity.
- Social sentiment sensor: a visualization system for topic detection and topic sentiment analysis on microblog45
A Social Sentiment Sensor (SSS) system on Sina Weibo to detect daily hot topics and analyze the sentiment distributions toward these topics and produces a real-time visualization system to monitor social sentiments, which is offering the public a new and timely perspective on the dynamics of the social topics.
- An AMR Aligner Tuned by Transition-based Parser44
A new rich resource enhanced AMR aligner which produces multiple alignments and a new transition system for AMR parsing along with its oracle parser and a transition-based AMR parser that parses a sentence into its AMR graph directly are proposed.
- A cascaded syntactic and semantic dependency parsing system43
The CoNLL 2008 Shared Task system includes two cascaded components: a syntactic and a semantic dependency parsers, which simplify the traditional four stages semantic role labeling into two: a maximum entropy based argument classification and an ILP-based post inference.
- 42
- 40
- 39
- Distilled Dual-Encoder Model for Vision-Language Understanding39
DiDE, a framework that distills the knowledge of the fusion-encoder teacher model into the dual-encoding student model, and encourages the student not only to mimic the predictions of teacher, but also to calculate the cross-modal attention distributions and align with the teacher.
- Aligning Semantic in Brain and Language: A Curriculum Contrastive Method for Electroencephalography-to-Text Generation37
A Curriculum Semantic-aware Contrastive Learning strategy (C- SCL), which effectively recalibrates the subject-dependent EEG representation to the semantic- dependent EEG representation, thereby reducing the discrepancy.
- ExCAR: Event Graph Knowledge Enhanced Explainable Causal Reasoning37
This work presents an Event graph knowledge enhanced explainable CAusal Reasoning framework (ExCAR), and proposes the Conditional Markov Neural Logic Network (CMNLN) that combines the representation learning and structure learning of logical rules in an end-to-end differentiable manner.
- DADgraph: A Discourse-aware Dialogue Graph Neural Network for Multiparty Dialogue Machine Reading Comprehension37
A discourse-aware dialogue graph neural network, DADgraph, is presented, which explicitly constructs the dialogue graph using discourse dependency links and discourse relations, and achieves statistically significant improvements compared against strong neural network MRC baselines.
- Enhancing Neural Data-To-Text Generation Models with External Background Knowledge37
This paper enhances neural data-to-text models with external knowledge in a simple but effective way to improve the fidelity of generated text by attends to relevant external knowledge, encoded as a temporary memory, and combines this knowledge with the context representation of data before generating words.
- Transition-Based Syntactic Linearization37
This work extends the transition system so that word ordering is performed in addition to syntactic parsing, resulting in a linearization system that runs in guaranteed quadratic time and runs an order of magnitude faster than a state-of-the-art baseline using best-first search.
- Learning to Rewrite for Non-Autoregressive Neural Machine Translation36
An architecture named RewriteNAT is proposed to explicitly learn to rewrite the erroneous translation pieces by utilizing a locator module to locate the erroneous ones, which are then revised into the correct ones by a revisor module.
- Learning Event Graph Knowledge for Abductive Reasoning36
A variational autoencoder based model ege-RoBERTa is proposed, which employs a latent variable to capture the necessary commonsense knowledge from event graph for guiding the abductive reasoning task.
- 36
- Contradiction Detection with Contradiction-Specific Word Embedding35
A tailored neural network to learn contradiction-specific word embedding (CWE), which can separate antonyms in the opposite ends of a spectrum and performs comparably with the top-performing system in accuracy of three-category classification.
- Cognitive distortion based explainable depression detection and analysis technologies for the adolescent internet users on social media33
This study proposes a new task, explainable depression detection, to not only automatically detect depression but also try to give clues to depression based on cognitive distortion theory, and constructs a multi-task learning model based on a pre-trained model to detect depression and identify cognitive distortion.
- Learning Document Representation for Deceptive Opinion Spam Detection33
A neural network is presented to learn document-level representation that not only learns to represent each sentence but also represent the whole document of the review and shows the better performance in deceptive spam detection.
- Integrating Intra- and Inter-document Evidences for Improving Sentence Sentiment Classification33
A graph-based propagation approach is presented to incorporate inside and outside sentence features to improve the sentence sentiment classification performance, and the experimental results on camera domain show that the proposed approach performs better than the approaches without usingoutside sentence features, and outperforms other representational previous approaches.
- A Generative Model for End-to-End Argument Mining with Reconstructed Positional Encoding and Constrained Pointer Mechanism32
This paper proposes a generative framework, in which the expected outputs of AM are framed as a simple target sequence, and employs a pre-trained sequence-to-sequence language model with a constrained pointer mechanism to model the clues for all the subtasks of AM in the light of the target sequence.
- A novel approach to update summarization using evolutionary manifold-ranking and spectral clustering32
A novel content selection framework based on evolutionary manifold-ranking and normalized spectral clustering that aims to capture the temporal characteristics and relay propagation of information in dynamic data stream and user need is proposed.
- Negation Scope Detection with Recurrent Neural Networks Models in Review Texts31
This paper proposes different models based on recurrent neural networks (RNNs) and word embedding that can be successfully applied to such tasks without any task-specific feature engineering efforts and shows that RNNs, without using any hand-crafted features, outperform feature-rich CRF-based model.
- Development and evaluation of multimodal AI for diagnosis and triage of ophthalmic diseases using ChatGPT and anterior segment images: protocol for a two-stage cross-sectional study29
The development of a smartphone-based multimodal AI system using ChatGPT and anterior segment images for diagnosing and triaging ophthalmic diseases and its performance is assessed through a two-stage cross-sectional study.
- 29
- Content-based table retrieval for web queries29
This work presents a ranking based approach to content-based table retrieval, and implements both carefully designed features and neural network architectures to measure the relevance between a query and the content of a table.
- MSAMSum: Towards Benchmarking Multi-lingual Dialogue Summarization28
A multi-lingual dialogue summarization dataset, namely MSAMSum, which covers dialogue-summary pairs in six languages using sophisticated translation techniques and employs two methods to ensure the integral translation quality and summary factual consistency.
- 28
- CausalTriad27
A factor graph model to incorporate three clues to discover causation expressions in the text data is proposed, called CausalTriad, which uses textual and structural knowledge to infer pseudo causal relations from the triad structures.
- Generating Chinese named entity data from parallel corpora27
A general framework to generate large-scale NER training data from parallel corpora using an English-Chinese parallel corpus is presented and can collect high-quality labeled data and can help improve Chinese NER.
- Learning Sentence Representation for Emotion Classification on Microblogs27
Experimental results in the supervised learning framework show that using the pseudo-labeled data, the representation learned by Deep Belief Network outperforms the Principal Components Analysis based and Latent Dirichlet Allocation based representations.
- 26
- A survey of discourse parsing26
This survey introduces several different kinds of discourse parsing tasks, mainly including RST-style discourse parsing, PDTB-style discusses, and discourse parsing for multiparty dialogue, and discusses the future trends of the task.
- Transforming Wikipedia Into Augmented Data for Query-Focused Summarization26
This paper uses Wikipedia to automatically collect a large query-focused summarization dataset (named WikiRef) of more than 280,000 examples, which can serve as a means of data augmentation and develops a BERT-based query- focused summarization model (Q-BERT) to extract sentences from the documents as summaries.
- Neural Semantic Parsing in Low-Resource Settings with Back-Translation and Meta-Learning26
This work aims to learn a neural semantic parser when only prior knowledge about a limited number of simple rules is available, without access to either annotated programs or execution results, and improves the accuracy and stability on examples uncovered by rules.
- Microblog Entity Linking by Leveraging Extra Posts25
This paper proposes to use extra posts for the microblog entity linking task and significantly improves the linking accuracy over traditional methods by 83% and 7.5% respectively.
- Dialogue Discourse-Aware Graph Convolutional Networks for Abstractive Meeting Summarization.24
A Dialogue Discourse-Aware Graph Convolutional Networks (DDA-GCN) for meeting summarization is developed by utilizing dialogue discourse, which is a dialogue-specific structure that can provide pre-defined semantic relationships between each utterance.
- Hierarchical Attention based Neural Network for Explainable Recommendation24
This work proposes an approach for rating prediction using a hierarchical attention-based network named HANN, which can distinguish the importance of reviews at both word level and review level for explanations automatically and achieves an improvement in prediction compared to several state-of-the-art approaches.
- Exploiting Multiple Sources for Open-Domain Hypernym Discovery24
This paper proposes a simple yet effective distant supervision framework for Chinese open-domain hypernym discovery, and presents a heuristic strategy to build a large-scale noisy training data for the model without human annotation.
- 23
- 23
- A Joint Segmentation and Classification Framework for Sentiment Analysis23
A log-linear model is used to score each segmentation candidate, and exploit the phrasal information of top-ranked segmentations as features to build the sentiment classifier.
- Event type recognition based on trigger expansion23
This paper describes a method based on automatic expansion of the event triggers for event type recognition that improves the unbalanced data problem in training models and the data sparseness problem with a small corpus.
- Deciphering the Impact of Pretraining Data on Large Language Models through Machine Unlearning22
This work systematically analyzes the impact of 48 datasets from 5 major categories of pretraining data of LLMs and measures their impacts on LLMs using benchmarks about nine major categories of model capabilities to provide insights into the organization of data to support more efficient pretraining of LLMs.
- Leveraging Multi-modal Interactions among the Intermediate Representations of Deep Transformers for Emotion Recognition22
The RILA model achieves the state-of-the-art performance, benefiting from fully exploiting the multi-modal interactions among the intermediate representations of deep pre-trained transformers for end-to-end emotion recognition.
- Enhancing Content Planning for Table-to-Text Generation with Data Understanding and Verification22
This paper proposes to enhance neural content planning by understanding data values with contextual numerical value representations that bring the sense of value comparison into content planning; and verifying the importance and ordering of the selected sequence of records with policy gradient.
- 22
- Distilling Knowledge for Search-based Structured Prediction22
Experimental results show that distillation can effectively improve the single model’s performance and the final model achieves improvements of 1.32 in LAS and 2.65 in BLEU score on these two tasks respectively and it outperforms the greedy structured prediction models in previous literatures.
- Open-categorical text classification based on multi-LDA models22
This work proposes a novel approach to construct the categorization system and classify documents based on multi-latent Dirichlet allocation (LDA) models, which outperforms the state-of-the-art supervised and semi-supervised text classification methods.
- Towards Comprehensive Post Safety Alignment of Large Language Models via Safety Patching21
This work introduces SafePatching, a novel framework for comprehensive PSA, where two distinct safety patches are developed on the harmful data to enhance safety and mitigate over-safety concerns, and then seamlessly integrated into the target LLM backbone without compromising its utility.
- 21
- 21
- Masking and Generation21
This work explores a masking and generation paradigm in the context to extract the context incongruities for learning sarcastic expression and uses unsupervised contrastive learning to improve the sentence representation based on the standard dropout.
- 20
- Learning to Describe for Predicting Zero-shot Drug-Drug Interactions20
Leveraging textual information from online databases like DrugBank and PubChem, an innovative approach TextDDI with a language model-based DDI predictor and a reinforcement learning~(RL)-based information selector is proposed, enabling the selection of concise and pertinent text for accurate DDI prediction on new drugs.
- Improving Controllable Text Generation with Position-Aware Weighted Decoding20
A novel framework based on existing weighted decoding methods called CAT-PAW is proposed, which introduces a lightweight regulator to adjust bias signals from the controller at different decoding positions to solve the control strength/fluency trade-off problem.
- A syntactic path-based hybrid neural network for negation scope detection20
A syntactic path-based hybrid neural network architecture, a novel approach to identify the scope of negation in a sentence, that combines a bidirectional long short-term memory (Bi-LSTM) network and a convolutional neural network (CNN).
- Encoding World Knowledge in the Evaluation of Local Coherence20
This paper introduces semantic relatedness as an enrichment to the original graph-based model of Guinaudeau and Strube (2013) and incorporates semantic relatednesses as additional features into the popular entity-based models of Barzilay and Lapata (2008).
- Improving Candidate Generation for Entity Linking20
This paper proposes a novel candidate generation approach to generate high recall candidate set with small size and shows that the candidate generation recall achieves more than 93%.
- 19
- Neural Natural Logic Inference for Interpretable Question Answering19
This paper investigates a neural-symbolic QA approach that integrates natural logic reasoning within deep learning architectures, towards developing effective and yet explainable question answering models.
- 19
- 18
- 18
- 18
- Truth discovery with memory network18
This work proposes memory-network-based models that adopt a memory mechanism to learn the reliability of sources for truth prediction and shows that these models outperform state-of-the-art methods for truth discovery.
- 17
- Learning to share by masking the non-shared for multi-domain sentiment classification17
The BERTMasker model is proposed, which explicitly masks domain-related words from texts, learns domain-invariant sentiment features from these domain-agnostic texts and uses those masked words to form domain-aware sentence representations.
- Information retrieval: a view from the Chinese IR community17
A key result is that information retrieval researchers need to open their mind to embrace a broader IR field by rethink the definition of information, retrieval, user, system, and evaluation of IR.
- Multi-Input Multi-Output Sequence Labeling for Joint Extraction of Fact and Condition Tuples from Scientific Text17
A new sequence labeling framework (as well as a new tag schema) to jointly extract the fact and condition tuples from statement sentences and improves F1 score relatively by 4.2% on BioNLP2013 and by 6.2%" on a new bio-text dataset for tuple extraction.
- HIT-SCIR at WASSA 2023: Empathy and Emotion Analysis at the Utterance-Level and the Essay-Level16
This paper introduces the participation of team HIT-SCIR to the WASSA 2023 Shared Task on Empathy Detection and Emotion Classification and Personality Detection in Interactions and designs three different models to address them separately.
- KwaiAgents: Generalized Information-seeking Agent System with Large Language Models16
This paper introduces KwaiAgents, a generalized information-seeking agent system that employs LLMs as its cognitive core, and proposes the Meta-Agent Tuning framework, designed to ensure even an open-sourced 7B or 13B model performs well among many agent systems.
- From Artificially Real to Real: Leveraging Pseudo Data from Large Language Models for Low-Resource Molecule Discovery16
This paper first introduces a retrieval-based prompting strategy to construct high-quality pseudo data, then explores the optimal method to effectively leverage this pseudo data to address the low-resource challenge by utilizing artificially-real data generated by Large Language Models (LLMs).
- 15
- GEDIT15
A Geographic-Enhanced and Dependency-guIded sequence Tagging (GEDIT) model to concurrently address the two challenges of newly-coined words and a neural sequence tagging model by integrating and feeding the previously pre-learned representations into a CRF layer.
- Heterogeneous graph knowledge enhanced stock market prediction15
A novel heterogeneous neural network is presented to aggregate multi-grained information within financial text, which can be useful for context knowledge supplement and predictive information selection, and then improve the performance of stock market prediction.
- 14
- 14
- Knowledge-Aware Conversational Semantic Parsing over Web Tables14
A knowledge-aware semantic parser to improve parsing performance by integrating various types of knowledge, including grammar knowledge, expert knowledge, and external resource knowledge, which is based on a decomposable model.
- Person Name Disambiguation of Searching Results Using Social Network14
On the corpus of Chinese person names, experimental results show that the whole performance achieves high level and graph clustering algorithm benefits improving disambiguation effect from further dividing the connecting social network.
- 13
- CFSP: An Efficient Structured Pruning Framework for LLMs with Coarse-to-Fine Activation Information13
This work introduces an efficient structured pruning framework named CFSP, which leverages both Coarse and Fine-grained activation information as an importance criterion to guide pruning, and outperforms existing methods on diverse models across various sparsity budgets.
- 13
- 13
- A Multi-View–Based Collective Entity Linking Method13
This article proposes a multi-view–based collective entity linking algorithm, which combines several views of entities into an objective function for entity linking and can acquire higher accuracy than many state-of-the-art entity linking methods.
- English-Chinese Knowledge Base Translation with Neural Network13
A neural network approach is presented, which learns continuous triple representation with a gated neural network, which improves translation accuracy compared with baseline methods and shows that adaptive composition model improves standard solution such as neural tensor network in terms of translation accuracy.
- Appraisal Expression Recognition Based on Syntactic Path13
Experimental results on the camera and MP3 player domains show that syntactic paths are helpful for appraisal expression recognition, and an edit distance based syntactic path matching method can further improve the performance of appraisal expressions recognition.
- Improving cross-task generalization with step-by-step instructions12
Extensive experiments on Sup-NatInst reveal that high-quality step-by-step instructions can enhance cross-task generalization across different model sizes and the significance of the order of steps in the proposed instruction for improvement is indicated.
- Meaningful Learning: Enhancing Abstract Reasoning in Large Language Models via Generic Fact Guidance12
This work tailor an abstract reasoning dataset (AbsR) together with a meaningful learning paradigm to teach LLMs how to leverage generic facts for reasoning purposes, and shows that this approach not only boosts the general reasoning performance of LLMs but also makes considerable strides towards their capacity for abstract reasoning.
- Pretraining without wordpieces: learning over a vocabulary of millions of words12
This work explores to develop a BERT-style pretrained model over a vocabulary of words instead of wordpieces, called WordBERT, which makes significant improvements on cloze test and machine reading comprehension and has comparable time cost to BERT in natural language understanding tasks.
- Kuaipedia: a Large-scale Multi-modal Short-video Encyclopedia12
Kuaipedia is proposed, a large-scale multi-modal encyclopedia consisting of items, aspects, and short videos lined to them, which was extracted from billions of videos of Kuaishou (Kwai), a well-known short-video platform in China.
- A comparison of directed functional connectivity among fist-related brain activities during movement imagery, movement execution, and movement observation12
To better understand the brain activity changes and the communications across various brain regions under MO, ME, and MI, this study conducted the fist experiment and investigated the brain networks using spectral Granger causality.
- Enhanced Neural Machine Translation by Joint Decoding with Word and POS-tagging Sequences12
This paper improves the performance of neural machine translation with shallow syntax of target language, which has better accuracy and latency than deep syntax such as dependency parsing, and takes less parameters and runtime than other complex machine translation models.
- HIT-SCIR at SemEval-2020 Task 5: Training Pre-trained Language Model with Pseudo-labeling Data for Counterfactuals Detection12
This paper explores multiple pre-trained models to learn basic features and then fine-tune models with counterfactual data and pseudo-labeling data, which helps in detecting causation in texts.
- Deep Contextualized Word Embeddings for Universal Dependency Parsing12
Based on ELMo’s advantage on OOV, experiments that simulate low-resource settings are conducted and the results show that deep contextualized word embeddings are effective for data-insufficient tasks where the OOV problem is severe.
- Sentence Compression for Target-Polarity Word Collocation Extraction12
This work applies a discriminative conditional random field model, with some special sentimentrelated features, in order to automatically compress sentiment sentences.
- 12
- Simulating Before Planning: Constructing Intrinsic User World Model for User-Tailored Dialogue Policy Planning11
The User-Tailored Dialogue Policy Planning (UDP) framework is proposed, which integrates an Intrinsic User World Model to capture user traits and feedback and introduces an active learning approach that prioritizes challenging user personas during training to enhance robustness and adaptability.
- MACSA: A multimodal aspect-category sentiment analysis dataset with multimodal fine-grained aligned annotations11
This paper introduces a new task, named Multimodal Aspect-Category Sentiment Analysis (MACSA), with the goal of predicting sentiment polarities of image-text pairs with respect to given aspect categories and proposes a novel Multimodal Graph-based Aligned Network (MGAM) model for this task.
- 11
- GraCoRe: Benchmarking Graph Comprehension and Complex Reasoning in Large Language Models11
GraCoRe is presented, a benchmark for systematically assessing LLMs' graph comprehension and reasoning, which reveals that OpenAI o1 model has amazing comprehension and reasoning capabilities, semantic enrichment enhances reasoning performance, node ordering impacts task success, and the ability to process longer texts does not necessarily improve graph comprehension or reasoning.
- 11
- 11
- Adapter-Based Selective Knowledge Distillation for Federated Multi-Domain Meeting Summarization11
An adapter-based summarization model where two adapters cooperatively facilitate learning using fewer parameters to reduce communication costs is developed, and a selective knowledge distillation strategy is devised, assisting clients in robustly handling domain-focused modelling on their own data, while leveraging global parameters based on non-IID data.
- 11
- Self-Supervised Logic Induction for Explainable Fuzzy Temporal Commonsense Reasoning11
This work proposes to augment LMs with the temporal logic induction ability, which frames the temporal reasoning by defining three modular components: temporal dependency inducer and temporal concept defuzzifier and logic validator, which can yield great improvements over the traditional language model for temporal reasoning.
- Cross-Lingual Text-Rich Visual Comprehension: An Information Theory Perspective10
Experimental results on the XT-VQA demonstrate that MVCL-MI effectively reduces the visual-text cross-lingual performance disparity while preserving the inherent capabilities of LVLMs, shedding new light on the potential practice for improving LVLMs.
- Exploring & exploiting high-order graph structure for sparse knowledge graph completion10
A novel framework, LR-GCN, is presented that is able to automatically capture valuable long-range dependency among entities to supplement insufficient structure features and distill logical reasoning knowledge for sparse KGC.
- Cortico-muscular coherence of time–frequency and spatial characteristics under movement observation, movement execution, and movement imagery10
The results show that the cortico-muscular beta-lh band plays a critical role in the synchronous coupling between cortex and muscle and the primary motor cortex (M1), dorsolateral prefrontal cortex (DLPFC), supplementary motor area (SMA), and premotor cortex (PMC) are the specific regions of MO, ME, and MI.
- 10
- 10
- 10
- Retrieve, Discriminate and Rewrite: A Simple and Effective Framework for Obtaining Affective Response in Retrieval-Based Chatbots10
A simple and effective Retrieve-Discriminate-Rewrite framework that replaces the reranking mechanism with a new discriminate-andrewrite mechanism, which predicts the affect label of the retrieved high-quality response via discrimination module and further rewrites the affect unsatisfied response via rewriting module.
- 10
- Local Contexts Are Effective for Neural Aspect Extraction10
Surprisingly, it is found that a purely window-based neural network obtain comparable performance with a LSTM-RNN approach, which reveals the importance of local contexts for aspect extraction.
- Encoding syntactic representations with a neural network for sentiment collocation extraction10
The experimental results show that the proposed method can accurately capture the latent semantic features hidden behind the syntactic paths that neither the common feature-based methods nor the syntactical-path-based method can handle, and that it significantly outperforms numerous baselines and previous methods.
- Triple based Background Knowledge Ranking for Document Enrichment10
This study proposes a document enrichment framework which automatically extracts “argument1,predicate,argument2” triple from any text corpus as background knowledge, so that to ensure the compatibility with any resource and improve the enriching accuracy.
- Semantic-aware Contrastive Learning for Electroencephalography-to-Text Generation with Curriculum Learning9
A Curriculum Semantic-aware Contrastive Learning strategy (C-SCL), which effectively re-calibrates the subject-dependent EEG representation to the semantic- dependent EEG representation, thus reducing the discrepancy.
- 9
- Hansel: A Chinese Few-Shot and Zero-Shot Entity Linking Benchmark9
Hansel is presented, a new benchmark in Chinese that fills the vacancy of non-English few-shot and zero-shot EL challenges, and a strong baseline is established that achieves competitive results on TAC-KBP2015 Chinese Entity Linking task.
- Debiasing Stance Detection Models with Counterfactual Reasoning and Adversarial Bias Learning9
A novel counterfactual inference framework is leverage, which enables the dataset bias in the text part to be captured as the direct causal effect of the text on stances and reduced by subtracting the direct text effect from the total causal effect.
- ReCo: Reliable Causal Chain Reasoning via Structural Causal Recurrent Neural Networks9
By injecting reliable causal chain knowledge distilled by ReCo, BERT can achieve better performances on four downstream causal-related tasks than BERT models enhanced by other kinds of knowledge.
- 8
- 8
- 8
- 8
- 8
- Less Is More: Domain Adaptation with Lottery Ticket for Reading Comprehension8
Experimental results show that the method outperforms the full model adaptation on four out of four domains when only a small amount of annotated data available for adaptation, and introducing self-attention attribution reserves more parameters for important attention heads in the lottery subnetwork and improves the target domain model performance.
- Collective entity linking: a random walk-based perspective8
A novel collective entity linking algorithm that constructs a complete entity graph by connecting any two related entities, and the relationship between two entities is measured via a random walk-based calculating way.
- Intra-document Coreference Resolution: The state of the art.8
It is pointed out that the current key problems and research trends are global optimization, the use of deep linguistical and background knowledge, and the combination of linguistical models and statistical models.
- Toward Secure Tuning: Mitigating Security Risks from Instruction Fine-Tuning7
This work introduces a novel secure-tuning strategy called SWAT, which can be seamlessly integrated with pre-training and post-training methods, leading to greater improvements in security risks and task performance.
- NLP-AKG: Few-Shot Construction of NLP Academic Knowledge Graph Based on LLM7
NLP-AKG, an academic knowledge graph for the NLP domain, is developed and NLP-AKG, an academic knowledge graph for the NLP domain, is developed by extracting 620,353 entities and 2,271,584 relations from 60,826 papers in ACL Anthology.
- MA-GTS: A Multi-Agent Framework for Solving Complex Graph Problems in Real-World Applications7
MA-GTS (Multi-Agent Graph Theory Solver), a multi-agent framework that decomposes these complex problems through agent collaboration into clear, structured graph representations and dynamically selects the most suitable algorithm based on problem constraints and graph structure scale is proposed.
- 7
- Towards Stable Natural Language Understanding via Information Entropy Guided Debiasing7
A debiasing framework IEGDB is proposed that comprehensively detects the dataset biases to induce a set of biased features, and then purifies the biased features with the guidance of information entropy and can consistently improve the stability of performance on OOD datasets for aSet of widely adopted NLU models.
- 7
- Don't Ignore Dual Logic Ability of LLMs while Privatizing: A Data-Intensive Analysis in Medical Domain7
This work introduces the CALLA dataset to probe Large Language Models’ interactive knowledge acquisition from Chinese medical literature and concludes that IFT data highly correlated with the medical literature corpus serves as a potent catalyst for LLMs, enabling themselves to skillfully employ the medical knowledge acquired during the pre-training phase within interactive scenarios, enhancing accuracy.
- Biomedical evidence engineering for data-driven discovery7
A BERT-based evidence extraction model is proposed to extract evidence from literature in response to queries, and a dataset with 1 million examples of biomedical evidence is created, 10,000 of which are manually annotated.
- Learning to refine source representations for neural machine translation7
A novel encoder–refiner–decoder framework, which dynamically refines the source representations based on the generated target-side information at each decoding step, and a policy network to decide when to refine at specific decoding steps is proposed.
- Prompt Combines Paraphrase: Teaching Pre-trained Models to Understand Rare Biomedical Words7
Experimental results show that the proposed simple yet effective approach to helping models learn rare biomedical words during tuning with prompt can achieve up to 6% improvement in biomedical natural language inference task without any extra parameters or training steps using few-shot vanilla prompt settings.
- Unifying the Convergences in Multilingual Neural Machine Translation7
A novel training strategy named LSSD (LanguageSpecific Self-Distillation), which can alleviate the convergence inconsistency and help MNMT models achieve the best performance on each language pair simultaneously.
- Exploring Answer Stance Detection with Recurrent Conditional Attention7
The Recurrent Conditional Attention (RCA) model is introduced which incorporates a conditional attention structure into the recurrent reading process and iteratively guides the distillation of question semantic with answer information and collects stance-oriented text relating to question, further revealing mutual relationship among stance, answer and question.
- Babbling - The HIT-SCIR System for Emotional Conversation Generation7
This paper describes the HIT-SCIR emotional response agent “Babbling” to the NLPCC 2017 Shared Task 4 on emotional conversation generation and employs the emotion embeddings to guide emotionalizing process.
- Multi-level cross-lingual attentive neural architecture for low resource name tagging7
This paper combines multi-level cross-lingual knowledge as attention into a neural architecture, which guides low resource name tagging to achieve a better performance and is able to yield improvements in name tagging over all previous baselines.
- Product Aspect Clustering by Incorporating Background Knowledge for Opinion Mining7
Experimental results obtained demonstrate that the proposed product aspect clustering method based on two types of backgroundknowledge performs better than the baseline approach without the use of background knowledge, and indicate that expanding the available background knowledge using the Web is feasible.
- Encoding Distributional Semantics into Triple-Based Knowledge Ranking for Document Enrichment7
Evaluated as a ranking problem, this model significantly outperforms multiple strong baselines and conducts a task-based evaluation by incorporating these triples as additional features into document classification and enhances the performance by 3.02%.
- Topical key concept extraction from folksonomy through graph-based ranking7
The proposed method for extracting topical key concepts from folksonomy is applied to different datasets and is found to outperform the state-of-the-art baselines significantly.
- Clustering Product Aspects Using Two Effective Aspect Relations for Opinion Mining7
A hierarchical clustering algorithm is designed to cluster these aspects into different groups, in which aspect similarity computation is conducted with the relevant aspect set and irrelevant aspect set of each product aspect.
- Building Chinese Event Type Paradigm Based on Trigger Clustering7
A novel approach of building an event type para- digm by clustering event triggers based on the trigger clusters, which can be built automatically and can be easily adapted to new domains.
- HITIR's Update Summary at TAC 2008: Extractive Content Selection for Language Independence.7
The evolutionary manifold-ranking algorithm is proposed, and further integrate the sub-topics partition with spectral clustering to have a content selection, which is completely language independence.
- 6
- When Correct Beliefs Collapse: Epistemic Resilience of LLMs under Clinical Pressure6
This work proposes Med-Stress, a targeted stress test framework that evaluates belief stability under escalating pressure, and proposes a lightweight inference-time defense, a training-time approach that internalizes evidence-based resistance to pressure.
- 6
- CogGPT: Unleashing the Power of Cognitive Dynamics on Large Language Models6
Recognizing the inherent static nature of LLMs, CogGPT is introduced for the task, which features an innovative iterative cognitive mechanism aimed at enhancing lifelong cognitive dynamics.
- 6
- 6
- Quantifying the effects of long-term news on stock markets on the basis of the multikernel Hawkes process6
A novel multikernel-powered Hawkes process framework is proposed, which uses multiple kernels to model different time-decaying rates, thus alleviating the instability of the proposed Hawke process based prediction model.
- Stress Assessment of Vestibular Endurance Training for Civil Aviation Flight Students Based on EEG6
C3 and C4 areas are specific brain regions of the stress response of anti-vertigo endurance training, and the absolute power of the α wave can be used as a parameter for identifying the degree of motion sickness (MS).
- 6
- 6
- Table-to-Text Generation via Row-Aware Hierarchical Encoder6
Three extractive evaluation metrics and human evaluation show that the proposed neural model has the ability to select salient records and the generated game summary is more accurate.
- 6
- ContextCare: Incorporating Contextual Information Networks to Representation Learning on Medical Forum Data6
A novel and general representation learning method CONTEXTCARE is proposed for human generated health-related data, which learns latent relationship between symptoms and diseases from the symptom-disease diagnosis network for disease prediction, disease category prediction and disease clustering.
- Constructing Semantic Hierarchies via Fusion Learning Architecture6
A fusion learning architecture based on word embeddings for constructing semantic hierarchies, composed of discriminative generative fusion architecture and a very simple lexical structure rule for assisting, gets an F1-score of 74.20% with 91.60% precision-value, outperforming the state-of-the-art methods on a manually labeled test dataset.
- 6
- Aspect-Object Alignment with Integer Linear Programming in Opinion Mining6
A novel and important sentiment analysis task, termed aspect-object alignment, to solve the "object neglect" problem and designs a two-step framework for this task, which incorporates three sets of features, namely, the basic, relational, and special target features.
- How do Language Models Reshape Entity Alignment? A Survey of LM-Driven EA Methods: Advances, Benchmarks, and Future5
A novel taxonomy is proposed that categorizes methods in three key stages: data preparation, feature embedding, and alignment, to provide researchers and practitioners with a clear and comprehensive understanding of how language models reshape the field of entity alignment.
- 5
- Learning to Select Bi-Aspect Information for Document-Scale Text Content Manipulation5
An unsupervised neural model with interactive attention mechanism is presented, used for learning the semantic relationship between records and reference texts to achieve better content transfer and better style preservation.
- Unsupervised Coreference Resolution with HyperGraph Partitioning5
A hypergraph partitioning algorithm is introduced that divides mentions directly into equivalence classes representing individual entities and outperforms previous unsupervised methods on the ACE dataset.
- 5
- Improved Diffusion-based Generative Model with Better Adversarial Robustness4
The training objective of DPMs is analyzed and it is theoretically demonstrated that this mismatch can be alleviated through Distributionally Robust Optimization (DRO), which is equivalent to performing robustness-driven Adversarial Training (AT) on DPMs.
- Beyond Snapshots: A Multimodal User-Level Dataset for Depression Detection in Dynamic Social Media Streams4
The Multimodal User-level Depression Detection Dataset (MUD3) is introduced, which includes the long-term video histories of depressed users on social media platforms, containing user mental states across multiple videos and treating the video histories as a continuous social media stream.
- 4
- 4
- 4
- UNIMO-3: Multi-granularity Interaction for Vision-Language Representation Learning4
The UNIMO-3 model is proposed, which has the capacity to simultaneously learn the multimodal in-layer interaction and cross- layer interaction and can establish effective connections between different layers in a cross-modal encoder, and adaptively capture the interaction between two modalities at different levels.
- GTR: A Grafting-Then-Reassembling Framework for Dynamic Scene Graph Generation4
A Grafting-Then-Reassembling framework, which explicitly extracts intra-frame spatial information and inter-frame temporal information in two separate stages to decouple spatio-temporal contextual information and achieves the state-of-the-art performance on Action Genome dataset.
- 4
- 4
- Interactive Gated Decoder for Machine Reading Comprehension4
An Interactive Gated Decoder (IG Decoder), which focuses on modeling the interactions between the answer span prediction and no-answer prediction with a gating mechanism, and a simple but effective approach for automatically generating pseudo training data, which aims to enrich the training data of the unanswerable questions.
- Zero-shot Aspect-level Sentiment Classification via Explicit Utilization of Aspect-to-Document Sentiment Composition4
This work achieves zero-shot aspect-level sentiment classification by only using document-level reviews and proposes the AF-DSC method to explicitly model such sentiment composition in reviews, which outperforms previous work utilizing millions of data.
- 4
- Coreference Resolution System using Maximum Entropy Classifier4
The system relies on a maximum entropy-based classifier for pairs of mentions, and adopts a rich linguisitically motivated feature set, and experiment with alternaive resolution process, preprocessing tools, and classifiers.
- Cascaded Regression Analysis Based Temporal Multi-document Summarization4
A cascaded regression analysis based macro-micro importance discriminative model for the content selection of TMDS is presented, which mines the temporal characteristics at different levels of topical detail in order to provide the cue for extracting the important content.
- Improving the Downstream Performance of Mixture-of-Experts Transformers via Weak Vanilla Transformers3
This work introduces the concept of transfer capability distillation, positing that although vanilla models have weaker performance, they are effective teachers of transfer capability, and develops a specific distillation method and conducts experiments using the BERT architecture, showing significant improvements in the downstream performance of MoE models.
- 3
- 3
- Diagnosing and Remedying Knowledge Deficiencies in LLMs via Label-free Curricular Meaningful Learning3
A label-free curricular meaningful learning framework that efficiently and effectively diagnoses and remedies knowledge deficiencies in LLMs, improving various LLMs across seven out-of-distribution reasoning and language understanding benchmarks, and surpasses methods that rely on labeled datasets for deficiency diagnosis.
- 3
- 3
- The Role of Summarization in Generative Agents: A Preliminary Perspective3
This report presents an integrated perspective on comprehendingGenerative agents through summarization, since it believes summarization is the most fundamental and indispensable capacity of generative agents manifested across diverse scenarios.
- 3
- TAGNet: a tiny answer-guided network for conversational question generation3
TAGNet is introduced, a lightweight network based on the lightweight module (Bi-LSTM) for CQG that takes the target answers as input, which interacts with the passages and conversation history in the encoder and guides the question generation through the gated attention mechanism in the decoder.
- 3
- Improved Neural Machine Translation with POS-Tagging Through Joint Decoding3
Three NMT decoding models are presented to jointly predict target word and POS tag sequences and it is shown that the fully shared decoder can acquire the best performance.
- 3
- A Joint Model for Sentiment Classification and Opinion Words Extraction3
This work proposes a hierarchical network for jointly modeling aspect-level sentiment classification and word-level opinion words extraction and achieves comparable results in sentiment polarity classification on two datasets from SemEval 2014.
- 3
- BUEES: a bottom-up event extraction system3
This paper presents BUEES, a bottom-up event extraction system, which extracts events from the web in a completely unsupervised way, and compares it to a state-of-the-art Chineseevent extraction system.
- What Causes Different Emotion Distributions of a Hot Event? A Deep Event-Emotion Analysis System on Microblogs3
This paper presents a deep Event-Emotion analysis system on Microblogs to reveal what causes different emotions of a hot event and tracks the amount treads and emotion treads of the hot event, and presents the deep analysis based on the user profile.
- Improving Web Search Ranking by Incorporating Structured Annotation of Queries3
A method to improve web search ranking by detecting Structured Annotation of queries based on top search results, which involves splitting the original query into different units associated with semantic attributes in the corresponding domain.
- ENPMR-Bench: Benchmarking Proactive Memory Retrieval for Emotional Support Agents2
ENPMR-Bench is introduced, a benchmark for evaluating Emotional Need-aware Proactive Memory Retrieval (ENPMR), a core capability that enables agents to infer users'latent emotional needs and proactively retrieve appropriate memories to support empathetic interaction.
- MoGU V2: Toward a Higher Pareto Frontier Between Model Usability and Security2
The MoGU framework is proposed, in which the intra-layer router dynamically allocates weights by sensing hidden states, thereby balancing the contributions of security-optimized and usability-optimized variants, and an improved MoGU_v2 framework is proposed that establishes a tighter coupling between the routers and hidden states.
- 2
- Natural Logic at the Core: Dynamic Rewards for Entailment Tree Generation2
This work proposes a method that integrates natural logic principles into reinforcement learning, enabling dynamic reward computation to guide entailment tree generation, ensuring logical consistency across reasoning steps while improving inter-pretability and generalization.
- 2
- 2
- 2
- 2
- 2
- 2
- VEM$$^2$$L: an easy but effective framework for fusing text and structure knowledge on sparse knowledge graph completion2
This paper proposes VEM2, a joint learning framework that incorporates structure and relevant text information to supplement insufficient features for sparse KGs, and proposes a novel co-distillation strategy using the Variational EM algorithm on unobserved queries.
- 2
- Data Uncertainty-Aware Learning for Multimodal Aspect-based Sentiment Analysis2
A novel data uncertainty-aware multimodal aspect-based sentiment analysis approach, UA-MABSA, which weighted the loss of different samples by the data quality and difficulty, thus enabling the model to pay more attention to high-quality and challenging samples.
- 2
- Dialogue Context Modelling for Action Item Detection: Solution for ICASSP 2023 Mug Challenge Track 52
- 2
- GLS-CSC: A Simple but Effective Strategy to Mitigate Chinese STM Models' Over-Reliance on Superficial Clue2
This study aims to investigate STM models' over-reliance on the edit distance feature, commonly used to measure the semantic similarity of Chinese text pairs, and proposes a novel resampling training strategy called Gradually Learn Samples Containing Superficial Clue (GLS-CSC).
- 2
- BigCilin: An Automatic Chinese Open-domain Knowledge Graph with Fine-grained Hypernym-Hyponym Relations2
BigCilin is presented, the first Chinese open-domain knowledge graph with fine-grained hypernym-hyponym relations which are extracted automatically from multiple sources for Chinese named entities and owns flexible semantic hierarchical structure.
- 2
- An Efficient End-to-End Transformer with Progressive Tri-modal Attention for Multi-modal Emotion Recognition2
The proposed multi-modal end-to-end transformer (ME2ET), which can effectively model the tri- modal features interaction among the textual, acoustic, and visual modalities at the low-level and high-level, and the proposed progressive tri-modAL attention, which can help the model to achieve better performance while significantly reducing the computation and memory cost.
- 2
- 2
- Document-Level Event Subject Pair Recognition2
A global document-level vector based on sentence-level vectors which are encoded from BERT aims to cover the information carried by the entire document and is utilized to guide the extraction process conducted sentence by sentence.
- 2
- Kernelised supervised context hashing2
This study proposes a novel hashing method dubbed kernelised supervised context hashing, which considers the hashing codes interrelation to reduce the quantisation and achieves better performance than several other state-of-the-art methods.
- Encoding Dependency Representation with Convolutional Neural Network for Target-Polarity Word Collocation Extraction2
A dependency representation is introduced to explore the most useful semantic features behind the syntactic rules and a framework based on a convolutional neural network (CNN) is adopted to extract the T-P collocations.
- 2
- Generating Triples Based on Dependency Parsing for Contradiction Detection2
The dp-triple extraction method is proposed to extract more contradiction pairs than present shallow semantic extraction methods like open IE or SRL and it is demonstrated the joint method is effective in detecting contradiction.
- A Novel Heuristic Error-Driven Learning for Recognizing Chinese Time Expression.2
A heuristic error-driven learning framework is proposed for recognizing Chinese time expression, which integrates the heuristic search strategy * A algorithm into error- driven learning.
- 1
- 1
- 1
- Exploring Cross-Lingual Latent Transplantation: Mutual Opportunities and Open Challenges1
This paper introduces and investigates cross-lingual latent transplantation, a probing framework which aims to further exploit the model’s internalized multilingual knowledge during inference and examine its effects on the multilingual capability and cultural adaptability of LLMs.
- 1
- 1
- 1
- Necessary and sufficient knowledge enhanced collaborative logical reasoning in LLMs1
This work contributes to remedy the inherent limitations of the logical reasoning paradigms in LLMs and lays the foundation for modeling human cognitive thinking.
- CrossICL: Cross-Task In-Context Learning via Unsupervised Demonstration Transfer1
A new ICL paradigm CrossICL is explored to study how to utilize existing source task demonstrations in the ICL for target tasks, thereby obtaining reliable guidance without any additional manual effort.
- 1
- META-RAG: Meta-Analysis-Inspired Evidence-Re-Ranking Method for Retrieval-Augmented Generation in Evidence-Based Medicine1
This work employs a combination of several EBM methods to emulate the meta-analysis, which includes reliability analysis, heterogeneity analysis, and extrapolation analysis, and enables RAG to extract higher-quality and more reliable evidence from the PubMed dataset.
- Think straight or think again? Continual joint learning of deduction, abduction and induction1
This paper designs a three-level learning procedure for deduction, abduction and induction that can achieve comparable performance with GPT-3.5 in human evaluation and introduces induction as a fact retriever to support and guide the above dual learning.
- 1
- Towards Generalizable and Faithful Logic Reasoning over Natural Language via Resolution Refutation1
This work proposes a novel framework, named Generalizable and Faithful Reasoner (GFaiR), which introduces the paradigm of resolution refutation, and demonstrates that the system outperforms previous works by achieving state-of-the-art performances in complex scenarios while maintaining performances in simple scenarios.
- How does Architecture Influence the Base Capabilities of Pre-trained Language Models? A Case Study Based on FFN-Wider and MoE Transformers1
This work attempts to explain and reverse the decline in base capabilities caused by the architecture of FFN-Wider Transformers, and finds the contribution ratio of Multi-Head Attention to pre-trained language modeling is a key factor affecting base capabilities.
- MolFusion: Multimodal Fusion Learning for Molecular Representations via Multi-granularity Views1
Experimental results show that MolFusion effectively utilizes complementary multimodal information, leading to significant improvements in performance across various classification and regression tasks.
- Supervised Fine-Tuning Achieve Rapid Task Adaption Via Alternating Attention Head Activation Patterns1
A gradient-based method is employed, to dissect the process that the SFT process adapts LLMs to downstream tasks via the perspective of attention patterns, finding that LLMs selectively activate task-specific attention heads during SFT and activation patterns for complex tasks are combinations of basic task patterns.
- Ensuring Consistency for In-Image Translation1
This work introduces a novel two-stage framework named HCIIT (High-Consistency In-Image Translation) which involves text-image translation using a multimodal multilingual large language model in the first stage and image backfilling with a diffusion model in the second stage.
- Leveraging Psychiatric Scale for Suicide Risk Detection on Social Media1
This work proposes a Scale-based Neural Network architecture that aims to extract content associated with scales from the posting history of social media users to predict their suicide risk level and provides scale-based explanations for the model's predictions.
- 1
- 1
- 1
- 1
- Improving Affective Event Classification with Multi-perspective Knowledge Injection1
A novel Multi-perspective Knowledge-injected Interaction Network (MKIN) is proposed to fully understand the event and accurately predict its sentiment by injecting multi-Perspective knowledge.
- Expression analysis of nel during zebrafish embryonic development1
This study provides new evidence to better understand the role of nel in zebrafish embryo development and indicates that nell2a and nell 2b genes are expressed in the nervous system and eyes of zebra fish embryo, while nell3 is expressed in different regions inThe nervous system.
- Learning number reasoning for numerical table-to-text generation1
This paper proposes a neural architecture called Neural Table Reasoning Generator (NTRG), with an additional switching gate as well as a specifically designed equation decoder for generating mathematical equations adaptively and presents a pre-training strategy similar to the mask language model.
- A Passage-Level Text Similarity Calculation1
This paper aims to construct a passage-level event connection graph to model the relations among the events mentioned by one text and acquires superior results than unsupervised methods by a large margin and even comparable results with some popular supervised neuron based methods.
- Text correlation calculation based on passage-level event representation1
A passage-level event representation method based on sentence- level event extraction that outperforms conventional text correlation calculation methods and can acquire text correlation.
- 1
- Tag recommendation based on topic hierarchy of folksonomy1
The experimental results show that the proposed method significantly outperforms state-of-the-art methods (t-test, p-value < 0.0001) and demonstrates effectiveness with respect to data sources on tag recommendation.
- 1
- 1
- Deep Learning Based Document Theme Analysis for Composition Generation1
Two deep learning based models under the proposed hierarchical neural network framework to address theme analysis problem in Chinese college entrance examination and two transfer learning strategies are tried to deal with the lack of large training data for composition theme analysis problems.
- Topic hierarchy construction from heterogeneous evidence1
This method combines heterogeneous evidence from multiple sources including folksonomy and encyclopedia, separately in both initial topic hierarchy construction and topic hierarchy improvement, and significantly outperforms state-of-theart methods.
- Aspect-Object Alignment Using Integer Linear Programming1
This paper proposes a novel and important sentiment analysis task: aspect-object alignment, which aims to obtain the correct corresponding object for each aspect, to solve the “object ignoring” problem.
- Topical Key Concept Extraction from Folksonomy1
This paper proposes a novel framework to extract topical key concepts from folksonomy and a novel link weight function which combines the local structure information and global semantic similarity is proposed in importance score propagation.
- 1
- 1
- Iterative Feedback Based Manifold-Ranking for Update Summary1
A novel manifold-ranking frame based on iterative feedback mechanism to this summary task that can naturally make use of both the relationships among all the sentences in the documents and relationships between the topic and the sentences.
- –
- SL-BiLEM: Structured Learnable Behavior-in-the-Loop Epidemic Modeling for Forecasting and Policy Evaluation–
The proposed SL-BiLEM (Structured Learnable Behavior-in-the-Loop Epidemic Model), leveraging physical constraints as regularization for robust extrapolation, is established as an interpretable tool for public health decision-makers seeking accurate prediction and principled intervention planning.
- –
- –
- –
- –
- S HARING B EYOND D ECISION : Deep Collaboration between Large Language Models via Representation Ensemble–
This work proposes Representation Ensemble (RISE), a novel ensemble framework that enables cross-LLM representation sharing for richer information exchange and introduces a representation alignment method based on relational similarity measures and an orthogonal latent-space transformation.
- –
- –
- –
- –
- Focus Like a Human: Efficient GUI Grounding via Coarse-to-Fine Visual Attention and Parallel Verification–
This work proposes a coarse-to-fine GUI grounding framework termed FastFocus, which exploits the stability of low-resolution attention to efficiently generate candidate grounding regions and validates that a hierarchical integration of low-resolution guidance and high-resolution verification constitutes an effective and robust paradigm for GUI grounding.
- –
- –
- Shared-private vision-to-text connector for grounded multimodal named entity recognition with synergistic global–local alignment–
A Vision-to-Text Connector (VTC) framework that projects visual features into the text feature space and integrates a relevance-aware coupling strategy, which leverages global image-text semantic consistency to adaptively weight the local alignment term, thus effectively alleviating noise from weak sparse supervision.
- –
- –
- LangGPS: Language Separability Guided Data Pre-Selection for Joint Multilingual Instruction Tuning–
LangGPS is proposed, a lightweight two-stage pre-selection framework guided by language separability—a signal that quantifies how well samples in different languages can be distinguished in the model’s representation space that improves their effectiveness and generalizability in multilingual training.
- –
- Rethinking Thinking Steps in Overthinking: Towards More Effective Reasoning–
This work proposes a training-free, interpretable framework that selects thinking words via attention heads to guide LRMs toward more effective reasoning, and consistently outperforms the strong baseline DEER.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- CultureRL: Internalizing Cultural Principles in Large Language Models via Norm-Driven Reinforcement Learning–
This work proposes CultureRL, a culture-norm-driven reinforcement learning framework that directly encodes cultural principles into model behavior and provides normative feedback during training, enabling the model to internalize high-level cultural rules.
- –
- –
- –
- Ensuring Responses Contain Appropriate Images: Timing Judgment for Multimodal Responses–
This paper proposes AMI (Automated Multimodal Insertion), a novel framework for dynamic, context-aware multimodal supplementation in dialogue, and introduces RID (Response with Appropriate Image Dataset), a bilingual (Chinese-English) multimodal multi-turn dialogue dataset designed to train and evaluate systems on this capability.
- –
- –
- –
- Investigating and Enhancing the Robustness of Large Multimodal Models Against Temporal Inconsistency–
- –
- –
- –
- –
- –
- SkillNet-X: A Multilingual Multitask Model with Sparsely Activated Skills–
A general multilingual multitask model, named SkillNet-X, which enables a single model to tackle many different tasks from different languages, and shows that SkillNet-X performs better than task-specific and two multitask learning baselines.
- –
- DiffuD2T: Empowering Data-to-Text Generation with Diffusion–
This paper uses the diffusion model to improve the prefix tuning for data-to-text generation and proposes a look-ahead guiding loss to supervise the iterative refinement process for better text generation.
- –
- –
- Assessment of pilot stress based on functional near-infrared techniques–
The cerebral blood oxygen characteristics of pilots before and after stress and the relationship between this characteristic and their antiglare performance by functional near-infrared techniques (fNIRs) are analyzed to establish an index for the evaluation of pilots' stress response.
- –
- –
- –
- –
- –
- HC-search for incremental parsing–
Inspired by recently proposed HC-search framework, this work decomposes the incremental parsing algorithm into two steps: first searching a set of high-quality outputs with beam-search, and second selecting the best output with a ranking model.
- –
- An Instance based Dynamic Generalization Model for Coreference Resolution–
This work presents an instance based dynamic generalization algorithm (IDGen), which constructs an exclusive model level by level automatically for every single instance with the generation pointreated from the features of the model.
- –
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-12. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.