Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks334 from public data
- Attention-based LSTM for Aspect-level Sentiment Classification2,397
This paper reveals that the sentiment polarity of a sentence is not only determined by the content but is also highly related to the concerned aspect, and proposes an Attention-based Long Short-Term Memory Network for aspect-level sentiment classification.
- ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools1,810
This report primarily focuses on the GLM-4 language series, which includes GLM-4, GLM-4-Air, and GLM-4-9B, which represent the most capable models that are trained with all the insights and lessons gained from the preceding three generations of ChatGLM.
- AgentBench: Evaluating LLMs as Agents1,460
AgentBench is presented, a multi-dimensional benchmark that consists of 8 distinct environments to assess LLM-as-Agent's reasoning and decision-making abilities and shows that poor long-term reasoning, decision-making, and instruction following abilities are the main obstacles for developing usable LLM agents.
- Pre-trained models: Past, present and future1,134
A deep look into the history of pre-training, especially its special relation with transfer learning and self-supervised learning, is taken to reveal the crucial position of PTMs in the AI development spectrum.
- Large Language Models Are Not Robust Multiple Choice Selectors620
This work proposes a label-free, inference-time debiasing method, called PriDe, which separates the model's prior bias for option IDs from the overall prediction distribution, and demonstrates that it achieves interpretable and transferable debiasing with high computational efficiency.
- Towards Emotional Support Dialog Systems535
The Emotional Support Conversation task is defined and an ESC Framework is proposed, which is grounded on the Helping Skills Theory, to show the importance of support strategies in providing effective emotional support and the utility of ESConv in training more emotional support systems.
- Commonsense Knowledge Aware Conversation Generation with Graph Attention525
This is the first attempt that uses large-scale commonsense knowledge in conversation generation, and unlike existing models that use knowledge triples (entities) separately and independently, this model treats each knowledge graph as a whole, which encodes more structured, connected semantic information in the graphs.
- GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models474
GLM-4.5 is presented, an open-source Mixture-of-Experts (MoE) large language model with 355B total parameters and 32B activated parameters, featuring a hybrid reasoning method that supports both thinking and direct response modes that ranks 3rd overall among all evaluated models and 2nd on agentic benchmarks.
- Learning to identify review spam440
This paper exploits machine learning methods to identify review spam and provides a twoview semi-supervised method, co-training, to exploit the large amount of unlabeled data and shows that the proposed method is effective.
- TransG : A Generative Model for Knowledge Graph Embedding369
This paper addresses a new issue of multiple relation semantics that a relation may have multiple meanings revealed by the entity pairs associated with the corresponding triples, and proposes a novel Gaussian mixture model for embedding, TransG, which can discover latent semantics for a relation and leverage a mixture of relation component vectors for embeddedding a fact triple.
- Structure-Aware Review Mining and Summarization362
This paper proposes a new machine learning framework based on Conditional Random Fields that can employ rich features to jointly extract positive opinions, negative opinions and object features for review sentences and shows that structure-aware models outperform many state-of-the-art approaches to review mining.
- Challenges in Building Intelligent Open-domain Dialog Systems359
This article reviews the recent work on neural approaches that are devoted to addressing three challenges in developing intelligent open-domain dialog systems: semantics, consistency, and interactiveness.
- Reinforcement Learning for Relation Classification From Noisy Data355
Experiment results show that the proposed novel model can deal with the noise of data effectively and obtains better performance for relation classification at the sentence level.
- ToRA: A Tool-Integrated Reasoning Agent for Mathematical Problem Solving331
This paper proposes ToRA a series of Tool-integrated Reasoning Agents designed to solve challenging mathematical problems by seamlessly integrating natural language reasoning with the utilization of external tools, thereby amalgamating the analytical prowess of language and the computational efficiency of tools.
- Agent-SafetyBench: Evaluating the Safety of LLM Agents311
This paper introduces Agent-SafetyBench, a comprehensive benchmark designed to evaluate the safety of LLM agents, and summarizes two fundamental safety defects in current LLM agents: lack of robustness and lack of risk awareness.
- Augmenting End-to-End Dialogue Systems With Commonsense Knowledge282
This model represents the first attempt to integrating a large commonsense knowledge base into end-to-end conversational models and suggests that the knowledge-augmented models are superior to their knowledge-free counterparts.
- A Knowledge-Enhanced Pretraining Model for Commonsense Story Generation269
A knowledge-enhanced pretraining model to utilize commonsense knowledge from external knowledge bases to generate reasonable stories that can generate more reasonable stories than state-of-the-art baselines, particularly in terms of logic and global coherence.
- A Hierarchical Framework for Relation Extraction with Reinforcement Learning253
This paper presents a novel paradigm to deal with relation extraction by regarding the related entities as the arguments of a relation and applies a hierarchical reinforcement learning (HRL) framework in this paradigm to enhance the interaction between entity mentions and relation types.
- Sentiment Analysis with Global Topics and Local Dependency250
This paper proposes a major departure from the previous approaches by assuming that the sentiments are related to the topic in the document, and puts forward a joint sentiment and topic model, i.e. Sentiment-LDA, and shows that exploiting the sentiment dependency is clearly advantageous, and that the Dependency-Sentiment- LDA is an effective approach for sentiment analysis.
- Discovering patterns to extract protein–protein interactions from full texts248
A robust and powerful methodology to mine protein-protein interactions from biomedical texts by using a dynamic programming algorithm to compute distinguishing patterns by aligning relevant sentences and key verbs that describe protein interactions.
- SSP: Semantic Space Projection for Knowledge Graph Embedding with Text Descriptions196
This paper proposes the semantic space projection (SSP) model, a model which jointly learns from the symbolic triples and textual descriptions to discover semantic relevance and offer precise semantic embedding.
- 189
- COLD: A Benchmark for Chinese Offensive Language Detection183
The factors that influence the offensive generations are investigated, and it is found that anti-bias contents and keywords referring to certain groups or revealing negative attitudes trigger offensive outputs easier.
- The gene normalization task in BioCreative III182
Evaluating teams using the gold standard and inferred ground truth shows that the EM algorithm allows team submissions to be differentiated while keeping the manual annotation effort feasible, and shows measures of comparative performance between teams.
- Densely Connected CNN with Multi-scale Feature Attention for Text Classification180
A densely connected CNN with multi-scale feature attention with competitive performance against state-of-the-art baselines on five benchmark datasets and reveals the model's ability to select proper n-gram features for text classification.
- A Pre-Training Based Personalized Dialogue Generation Model with Persona-Sparse Data179
A pre-training based personalized dialogue model that can generate coherent responses using persona-sparse dialogue data and outperforms state-of-the-art methods for generating more coherent and persona consistent responses with persona-Sparse data is proposed.
- Linguistically Regularized LSTM for Sentiment Classification179
Results show that the proposed simple models are able to capture the linguistic role of sentiment words, negation words, and intensity words in sentiment expression.
- Story Ending Generation with Incremental Encoding and Commonsense Knowledge169
A novel model for story ending generation that adopts an incremental encoding scheme to represent context clues which are spanning in the story context and can generate more reasonable story endings than state-of-the-art baselines1.
- VisionReward: Fine-Grained Multi-Dimensional Human Preference Learning for Image and Video Generation165
This work presents VisionReward, a general framework for learning human visual preferences in both image and video generation that employs a hierarchical visual assessment framework to capture fine-grained human preferences, and leverages linear weighting to enable interpretable preference learning.
- PsyQA: A Chinese Dataset for Generating Long Counseling Text for Mental Health Support157
This paper proposes PsyQA, a Chinese dataset of psychological health support in the form of question and answer pair, and annotates a portion of answer texts with typical strategies for providing support, and evaluates the performance of generating counseling answers with the generative pretrained models.
- 155
- Learning Structured Representation for Text Classification via Reinforcement Learning154
Results show that the proposed reinforcement learning method can learn task-friendly representations by identifying important words or task-relevant structures without explicit structure annotations, and thus yields competitive performance.
- Assigning Personality/Profile to a Chatting Machine for Coherent Conversation Generation154
Manual and automatic evaluation shows that the model can deliver more coherent, natural, and diversified responses to profile-coherent responses.
- Learning to Link Entities with Knowledge Base153
This paper proposes a learning to rank algorithm that effectively utilizes the relationship information among the candidates when ranking and achieves 18.5% improvement in terms of accuracy over the classification models for those entities which have corresponding entries in the Knowledge Base.
- A Large-Scale Chinese Short-Text Conversation Dataset152
A large-scale cleaned Chinese conversation dataset, LCCC, which contains a base version (6.8million dialogues) and a large version (12.0 million dialogues), and pre-training dialogue models which are trained on LCCC-base and LCCC -large respectively.
- Recommending MeSH terms for annotating biomedical articles152
A novel approach to facilitate MeSH indexing, a challenging task of assigning MeSH terms to MEDLINE citations for their archiving and retrieval, that makes the most accurate MeSH predictions to date, which suggests its great potential in making a practical impact on Me SH indexing.
- ERICA: Improving Entity and Relation Understanding for Pre-trained Language Models via Contrastive Learning143
Experimental results demonstrate that ERICA can improve typical PLMs (BERT and RoBERTa) on several language understanding tasks, including relation extraction, entity typing and question answering, especially under low-resource settings.
- SentiLARE: Sentiment-Aware Language Representation Learning with Linguistic Knowledge143
A novel language representation model called SentiLARE is proposed, which introduces word-level linguistic knowledge including part-of-speech tag and sentiment polarity (inferred from SentiWordNet) into pre-trained models.
- Out-of-Domain Detection for Natural Language Understanding in Dialog Systems142
A novel model is proposed to generate high-quality pseudo OOD samples that are akin to IN-Domain (IND) input utterances and thereby improves the performance of OOD detection and is demonstrated to be effective in NLU.
- Language Models Learn to Mislead Humans via RLHF140
This work asks time-constrained human subjects to evaluate the correctness of model outputs and calculate humans' accuracy against gold labels and highlights an important failure mode of RLHF, which makes the model harder to evaluate.
- KdConv: A Chinese Multi-domain Dialogue Dataset Towards Multi-turn Knowledge-driven Conversation136
This paper proposes a Chinese multi-domain knowledge-driven conversation dataset, KdConv, which grounds the topics in multi-turn conversations to knowledge graphs, and provides several benchmark models to facilitate the following research.
- JointGT: Graph-Text Joint Representation Learning for Text Generation from Knowledge Graphs135
A graph-text joint representation learning model called JointGT is proposed which devise a structure-aware semantic aggregation module which is plugged into each Transformer layer to preserve the graph structure during encoding and shows new state-of-the-art performance on various KG-to-text datasets.
- CPM: A large-scale generative Chinese Pre-trained language model135
CPM, with 2.6 billion parameters and 100GB Chinese training data, is the largest Chinese pre-trained language model, which could facilitate several downstream Chinese NLP tasks, such as conversation, essay generation, cloze test, and language understanding.
- AugESC: Dialogue Augmentation with Large Language Models for Emotional Support Conversation132
This work leverages large language models for dialogue augmentation in the task of emotional support conversation (ESC) to prompt a fine-tuned language model to complete full dialogues from available dialogue posts of various topics, which are then postprocessed based on heuristics.
- CrossWOZ: A Large-Scale Chinese Cross-Domain Task-Oriented Dialogue Dataset126
The large size and rich annotation of CrossWOZ make it suitable to investigate a variety of tasks in cross-domain dialogue modeling, such as dialogue state tracking, policy learning, user simulation, etc.
- ConvLab-2: An Open-Source Toolkit for Building, Evaluating, and Diagnosing Dialogue Systems121
ConvLab-2, an open-source toolkit that enables researchers to build task-oriented dialogue systems with state-of-the-art models, perform an end-to-end evaluation, and diagnose the weakness of systems is presented.
- Encoding Syntactic Knowledge in Neural Networks for Sentiment Classification121
This article proposes to learn tag-specific composition functions and tag embeddings in recursive neural networks, and proposes to utilize POS tags to control the gates of tree-structured LSTM networks.
- 120
- Language Generation with Multi-Hop Reasoning on Commonsense Knowledge Graph120
This paper proposes Generation with Multi-Hop Reasoning Flow (GRF) that enables pre-trained models with dynamic multi-hop reasoning on multi-relational paths extracted from the external commonsense knowledge graph and empirically shows that the model outperforms existing baselines on three text generation tasks that require reasoning over Commonsense knowledge.
- A Deep Sequential Model for Discourse Parsing on Multi-Party Dialogues120
A deep sequential model for parsing discourse dependency structures of multi-party dialogues by predicting dependency relations and constructing the discourse structure jointly and alternately is presented.
- Long and Diverse Text Generation with Planning-based Hierarchical Variational Model119
A planning-based Hierarchical Variational Model where a global planning latent variable models the diversity of reasonable planning and a sequence of local latent variables controls sentence realization is proposed.
- ConvLab: Multi-Domain End-to-End Dialog System Platform117
ConvLab is presented, an open-source multi-domain end-to-end dialog system platform that enables researchers to quickly set up experiments with reusable components and compare a large set of different approaches, ranging from conventional pipeline systems to end- to-end neural models, in common environments.
- 114
- 114
- Safety Assessment of Chinese Large Language Models113
A Chinese LLM safety assessment benchmark is developed based on the LLM's strong evaluation ability and develops it as a safety evaluator by prompting, which finds that instruction attacks are more likely to expose safety issues of all LLMs.
- ChID: A Large-scale Chinese IDiom Dataset for Cloze Test112
A large-scale Chinese cloze test dataset ChID is proposed, which studies the comprehension of idiom, a unique language phenomenon in Chinese, in which the idioms in a passage are replaced by blank symbols and the correct answer needs to be chosen from well-designed candidate idioms.
- Guided Dialog Policy Learning: Reward Estimation for Multi-Domain Task-Oriented Dialog106
Guided Dialog Policy Learning is proposed, a novel algorithm based on Adversarial Inverse Reinforcement Learning for joint reward estimation and policy optimization in multi-domain task-oriented dialog.
- Learning Tag Embeddings and Tag-specific Composition Functions in Recursive Neural Network105
Two models are proposed, Tag Guided RNN (TGRNN for short) which chooses a composition function according to the part-ofspeech tag of a phrase, and Tag Embedded RNN/RNTN (TE-RNN/ RNTN for long) which learns tag embeddings and then combines tag and word embedDings together.
- Synthetic Prompting: Generating Chain-of-Thought Demonstrations for Large Language Models103
Synthetic prompting is introduced, a method that leverages a few handcrafted examples to prompt the model to generate more examples by itself, and selects effective demonstrations to elicit better reasoning.
- Meta-Learning for Low-resource Natural Language Generation in Task-oriented Dialogue Systems103
This paper proposes a generalized optimization-based approach (Meta-NLG) based on the well-recognized model-agnostic meta-learning (MAML) algorithm, and defines a set of meta tasks, and directly incorporates the objective of adapting to new low-resource NLG tasks into the meta- learning optimization process.
- CPM-2: Large-scale cost-effective pre-trained language models102
A suite of cost-effective techniques for the use of PLMs to deal with the efficiency issues of pre-training, fine-tuning, and inference and validate the efficiency of InfMoE when conducting inference of large-scale models having tens of billions of parameters on a single GPU.
- PatentMiner97
This paper designs and implements a general topic-driven framework for analyzing and mining the heterogeneous patent network, and proposes a dynamic probabilistic model to characterize the topical evolution of these objects within the patent network.
- 96
- 96
- Long Text Generation by Modeling Sentence-Level and Discourse-Level Coherence95
A long text generation model, which can represent the prefix sentences at sentence level and discourse level in the decoding process, and two pretraining objectives to learn the representations by predicting inter-sentence semantic similarity and distinguishing between normal and shuffled sentence orders.
- 93
- GAKE: Graph Aware Knowledge Embedding93
This paper proposes a graph aware knowledge embedding method (GAKE), which formulates knowledge base as a directed graph, and learns representations for any vertices or edges by leveraging the graph’s structural information.
- 93
- Learning to Ask Questions in Open-domain Conversational Systems with Typed Decoders91
Two typed decoders are devised in which a type distribution over the three types is estimated and the type distribution is used to modulate the final generation distribution and it is observed that a good question is a natural composition of interrogatives, topic words, and ordinary words.
- CritiqueLLM: Towards an Informative Critique Generation Model for Evaluation of Large Language Model Generation90
This paper proposes a simple yet effective method called Eval-Instruct, which can first acquire pointwise grading critiques with pseudo references and then revise these critiques via multi-path prompting to obtain informative evaluation data in different tasks and settings, including pointwise grading and pairwise comparison with / without references.
- Knowledge graph embedding by flexible translation89
A new principle to allow flexible translation between entity and relation vectors is proposed and results show that the principle can remarkably improve the performance compared with several state-of-the-art baselines.
- 88
- Aspect-level Sentiment Analysis using AS-Capsules87
This paper proposes the aspect-level sentiment capsules model (AS-Capsules), which is capable of performing aspect detection and sentiment classification simultaneously, in a joint manner, and achieves state-of-the-art performances on a benchmark dataset for aspect- level sentiment analysis.
- Directed Acyclic Transformer for Non-Autoregressive Machine Translation86
Experiments show that DA-Transformer substantially outperforms previous NATs by about 3 BLEU on average, which is the first NAT model that achieves competitive results with autoregressive Transformers without relying on knowledge distillation.
- Generating Informative Responses with Controlled Sentence Function86
This model utilizes a continuous latent variable to capture various word patterns that realize the expected sentence function, and introduces a type controller to deal with the compatibility of controlling sentence function and generating informative content.
- OpenMEVA: A Benchmark for Evaluating Open-ended Story Generation Metrics83
OpenMEVA provides a comprehensive test suite to assess the capabilities of metrics, including the correlation with human judgments, the generalization to different model outputs and datasets, the ability to judge story coherence, and the robustness to perturbations.
- UNION: An Unreferenced Metric for Evaluating Open-ended Story Generation82
UNION is a learnable unreferenced metric for evaluating open-ended story generation, which measures the quality of a generated story without any reference, which correlates better with human judgments and is more generalizable than existing state-of-the-art metrics.
- CoMAE: A Multi-factor Hierarchical Framework for Empathetic Response Generation81
A multi-factor hierarchical framework, CoMAE, is proposed, which models the above three key factors of empathy expression in a hierarchical way and can generate more empathetic responses than previous methods.
- 81
- MultiWOZ 2.3: A Multi-domain Task-Oriented Dialogue Dataset Enhanced with Annotation Corrections and Co-Reference Annotation80
This paper introduces MultiWOZ 2.3, in which it differentiate incorrect annotations in dialogue acts from dialogue states, and identifies a lack of co-reference when publishing the updated dataset, to ensure consistency between dialogue acts and dialogue states.
- GeneTUKit: a software for document-level gene normalization80
GeneTUKit is developed, a document-level gene normalization software for full-text articles that employs both local context surrounding gene mentions and global context from the whole full- text document.
- CR-Walker: Tree-Structured Graph Reasoning and Dialog Acts for Conversational Recommendation77
The proposed CR-Walker is a model that performs tree-structured reasoning on a knowledge graph, and generates informative dialog acts to guide language generation, which links how entities are selected and expressed.
- Learning to Collaborate: Multi-Scenario Ranking via Multi-Agent Reinforcement Learning74
This paper proposes a novel model named Multi-Agent Recurrent Deterministic Policy Gradient (MA-RDPG) which has a communication component for passing messages, several private actors (agents) for making actions for ranking, and a centralized critic for evaluating the overall performance of the co-working actors.
- MMConv: An Environment for Multimodal Conversational Search across Multiple Domains70
The Multimodal Multi-domain Conversational dataset (MMConv) is introduced, a fully annotated collection of human-to-human role-playing dialogues spanning over multiple domains and tasks and adopted the state-of-the-art methods for these tasks respectively.
- On Prompt-Driven Safeguarding for Large Language Models69
It is demonstrated that DRO remarkably improves the safeguarding performance of human-crafted safety prompts and outperforms strong baselines, as evaluated on out-of-domain benchmarks, without compromising the general model capability.
- Uncertainty in Natural Language Processing: Sources, Quantification, and Applications67
This survey provides a comprehensive review of uncertainty-relevant works in the NLP field and categorizes the sources of uncertainty in natural language into three types, including input, system, and output.
- Acceleration of Federated Learning with Alleviated Forgetting in Local Training67
FedReg is proposed, an algorithm to accelerate FL with alleviated knowledge forgetting in the local training stage by regularizing locally trained parameters with the loss on generated pseudo data, which encode the knowledge of previous training data learned by the global model.
- Depression Detection in Clinical Interviews with LLM-Empowered Structural Element Graph66
A structural element graph (SEGA) is proposed, which transforms the clinical interview into an expertise-inspired directed acyclic graph for comprehensive modeling and further empower SEGA by devising novel principle-guided data augmentation with large language models (LLMs) to supplement high-quality synthetic data and enable graph contrastive learning.
- Chat More: Deepening and Widening the Chatting Topic via A Deep Model63
A novel deep scheme consisting of three channels, namely global, wide, and deep, which trains a Multi-layer Perceptron model to select some keywords for an in-depth discussion to generate desired responses in open-domain multi-turn dialog systems.
- EVA2.0: Investigating Open-domain Chinese Dialogue Systems with Large-scale Pre-training62
This work proposes EVA2.0, a large-scale pre-trained open-domain Chinese dialogue model with 2.8 billion parameters, and will make the models and codes publicly available.
- Multi-Agent Task-Oriented Dialog Policy Learning with Role-Aware Reward Decomposition61
This work proposes Multi-Agent Dialog Policy Learning, which regards both the system and the user as the dialog agents and uses the actor-critic framework to facilitate pretraining and improve scalability.
- 56
- Towards Identifying Social Bias in Dialog Systems: Framework, Dataset, and Benchmark56
This paper proposes a novel Dial-Bias Frame for analyzing the social bias in conversations pragmatically, which considers more comprehensive bias-related analyses rather than simple dichotomy annotations, and introduces CDail-B bias Dataset that is the first well-annotated Chinese social bias dialog dataset.
- Robustness Testing of Language Understanding in Task-Oriented Dialog56
This paper proposes a model-agnostic toolkit LAUG to approximate natural language perturbations for testing the robustness issues in task-oriented dialog, and introduces three important aspects related to language understanding in real-world dialog systems, namely, language variety, speech characteristics, and noise perturbation.
- User Attention-guided Multimodal Dialog Systems56
Experimental results demonstrate that the proposed model outperforms the existing state-of-the-art methods by integrating the multimodal utterances and encoding the visual features based on the users' attribute-level attention.
- 55
- CPED: A Large-Scale Chinese Personalized and Emotional Dialogue Dataset for Conversational AI50
This work proposes CPED, a large-scale Chinese personalized and emotional dialogue dataset, which consists of multi-source knowledge related to empathy and personal characteristic, to be widely adopted by the NLP community as a new open benchmark for conversational AI research.
- CTRLEval: An Unsupervised Reference-Free Metric for Evaluating Controlled Text Generation48
Experimental results show that the proposed unsupervised reference-free metric, CTRLEval, has higher correlations with human judgments than other baselines, while obtaining better generalization of evaluating generated texts from different models and with different qualities.
- Difference-aware Knowledge Selection for Knowledge-grounded Conversation Generation48
Automatic, human observational, and interactive evaluation shows that the proposed difference-aware knowledge selection method is able to select knowledge more accurately and generate more informative responses, significantly outperforming the state-of-the-art baselines.
- Memory-Augmented Dialogue Management for Task-Oriented Dialogue Systems48
MAD, a novel memory-augmented dialogue management model that employs a memory controller and two additional memory structures that augments the representation of hidden states of traditional recurrent neural networks by storing more context information, is proposed.
- Fine Granular Aspect Analysis using Latent Structural Models47
The model is able to find informative sentences that are closely related to some respects in a review, which may be helpful for aspect-level sentiment analysis such as aspect-oriented summarization and preliminary experiments demonstrate that the model obtains promising performance.
- Data Selection via Optimal Control for Language Models46
This work introduces PMP-based Data Selection (PDS), a framework that approximates optimal data selection by solving the PMP conditions and shows that the PDS-selected corpus accelerates the learning of LMs and constantly boosts their performance on a wide range of downstream tasks across various model sizes.
- 46
- Answering opinion questions with random walks on graphs45
An Opinion PageRank model and an Opinion HITS model are proposed to fully explore the information from different relations among questions and answers, answers and Answers, and topics and opinions to extract the opinion oriented answers.
- ASETF: A Novel Method for Jailbreak Attack on LLMs through Translate Suffix Embeddings42
An Adversarial Suffix Embedding Translation Framework (ASETF), aimed at transforming continuous adversarial suffix embeddings into coherent and understandable text, that significantly reduces the computation time of adversarial suffixes and achieves a much better attack success rate than existing techniques, while significantly enhancing the textual fluency of the prompts.
- A Weakly Supervised Method for Topic Segmentation and Labeling in Goal-oriented Dialogues via Reinforcement Learning42
A reinforcement learning (RL) method for topic segmentation and labeling in goal-oriented dialogues, which aims to detect topic boundaries among dialogue utterances and assign topic labels to the utterances, is proposed.
- A Survey of Post-Training Scaling in Large Language Models41
This paper presents a comprehensive survey of post-training scaling, an emergent paradigm aiming to relieve the limitations of traditional pre-training by focusing on the alignment phase, which traditionally accounts for a minor fraction of the total training computation.
- On the Learning of Non-Autoregressive Transformers41
It is shown that simply training NAT by maximizing the likelihood can lead to an approximation of marginal distributions but drops all dependencies between tokens, where the dropped information can be measured by the dataset's conditional total correlation.
- Mining microblog user interests based on TextRank with TF-IDF factor41
A framework for mining and analyzing personal interests from microblog text with a new algorithm which integrates term frequency-inverse document frequency (TF-IDF) with TextRank is presented.
- CharacterGLM: Customizing Chinese Conversational AI Characters with Large Language Models40
The CharacterGLM is designed for generating Character-based Dialogues (CharacterDial), which aims to equip a conversational AI system with character customization for satisfying people's inherent social desires and emotional needs.
- Rethinking Conversational Agents in the Era of LLMs: Proactivity, Non-collaborativity, and Beyond39
This tutorial will introduce the recent advances on the design of agent’s awareness of goals in a wide range of Conversational systems, including proactive, non-collaborative, and multi-goal conversational systems.
- Rethinking and Refining the Distinct Metric39
This work refine the calculation of distinct scores by scaling the number of distinct tokens based on their expectations, and shows that the proposed metric, Expectation-Adjusted Distinct (EAD), correlates better with human judgment in evaluating response diversity.
- 39
- Model Extrapolation Expedites Alignment38
This work proposes a straightforward method called ExPO (model extrapolation) to expedite LLMs' alignment with human preferences by improving the implicit optimization objective of alignment training by simply amplifying the parameter change based on a first-order approximation, without any additional training overhead.
- Tailoring Language Generation Models under Total Variation Distance38
The total variation distance (TVD) is leveraged with its robustness to outliers, and practical bounds are developed to apply it to language generation and the TaiLr objective is introduced that balances the tradeoff of estimating TVD.
- 38
- Diversifying Dialog Generation via Adaptive Label Smoothing38
An Adaptive Label Smoothing (AdaLabel) approach that can adaptively estimate a target label distribution at each time step for different contexts, which outperforms various competitive baselines in producing diverse responses.
- MiniPLM: Knowledge Distillation for Pre-Training Language Models37
This work proposes MiniPLM, a KD framework for pre-training LMs by refining the training data distribution with the teacher LM's knowledge, and reveals that MiniPLM supports KD across model families and enhances the pre-training data utilization.
- 37
- Neural Multimodal Belief Tracker with Adaptive Attention for Dialogue Systems37
The first neural multimodal belief tracker (NMBT) is presented to demonstrate how multi-modal evidence can facilitate semantic understanding and dialogue state tracking.
- ChatGLM-RLHF: Practices of Aligning Large Language Models with Human Feedback36
This paper presents the ChatGLM-RLHF pipeline -- a reinforcement learning from human feedback (RLHF) system -- designed to enhance ChatGLM's alignment with human preferences, and presents the practices of aligning LLMs with human preferences.
- 36
- A Self-Training Method for Machine Reading Comprehension with Soft Evidence Extraction36
A Self-Training method (STM), which supervises the evidence extractor with auto-generated evidence labels in an iterative process, and demonstrates the improvement on existing MRC models.
- Towards automatic generation of gene summary35
An extractive system that automatically generates gene summaries from the biomedical literature by exploiting gene-specific information and similarity relationships between sentences and significantly outperforms the other two methods with regard to all ROUGE metrics.
- Task-Adaptive Tokenization: Enhancing Long-Form Text Generation Efficacy in Mental Health and Beyond34
Through extensive experiments on psychological question-answering tasks in both Chinese and English, this work finds that the task-adaptive tokenization approach brings a significant improvement in generation performance while using up to 60% fewer tokens.
- Stylized Story Generation with Style-Guided Planning34
This study proposes a novel generation model that first plans the stylized keywords and then generates the whole story with the guidance of the keywords, and demonstrates that this model can controllably generateemo-tion-driven orevent-driven stories based on the ROCStories dataset.
- Dialogue Intent Classification with Long Short-Term Memory Networks34
A deep hierarchical LSTM model is presented to classify the intent of a dialogue utterance and is able to recognize and classify user’s dialogue intent in an efficient way and introduces a memory module to the hierarchical L STM model so that the model can utilize more context information to perform classification.
- Mining physical protein-protein interactions from the literature34
A text-mining framework to extract physical protein-protein interactions from the literature is presented and three key issues are addressed, namely filtering irrelevant articles, identifying protein names and normalizing them to molecule identifiers, and extracting protein- protein interactions.
- Human Decision-making is Susceptible to AI-driven Manipulation33
Human susceptibility to AI-driven manipulation is examined, highlighting a critical vulnerability in human-AI interactions, and the need for ethical safeguards and regulatory frameworks to protect human autonomy is highlighted.
- Unlocking Reasoning Potential in Large Langauge Models by Scaling Code-form Planning33
This work introduces CodePlan, a scalable framework that empowers LLMs to generate and follow pseudocode -- pseudocode that outlines high-level, structured reasoning processes that enables it to scale up efficiently and improve LLM's reasoning capabilities across diverse scenarios.
- FIVES: Feature Interaction Via Edge Search for Large-Scale Tabular Data33
This work proposes a novel method named Feature Interaction Via Edge Search (FIVES), which formulates the task of interactive feature generation as searching for edges on the defined feature graph and instantiates this search strategy by optimizing both a dedicated graph neural network (GNN) and the adjacency tensor associated with thedefined feature graph.
- 33
- Metadata-Aware Measures for Answer Summarization in Community Question Answering33
This paper presents a framework for automatically processing information coming from community Question Answering portals with the purpose of generating a trustful, complete, relevant and succinct summary in response to a question and shows that the information contained in the best answers voted by users of cQA portals can be successfully complemented by the method.
- 32
- Youling: an AI-assisted Lyrics Creation System32
Youling, an AI-assisted lyrics creation system, designed to collaborate with music creators, supports traditional one pass full-text generation mode as well as an interactive generation mode, which allows users to select the satisfactory sentences from generated candidates conditioned on preceding context.
- A Unified Active Learning Framework for Biomedical Relation Extraction32
This paper describes a unified active learning framework that systematically addresses some practical issues during active learning process, including a strategy for selecting informative data, a data diversity selection algorithm, an active feature acquisition method, and an informative feature selection algorithm in order to meet the challenges due to the immense amount of complex and diverse biomedical text.
- Robustness to Modification with Shared Words in Paraphrase Identification31
This paper studies the robustness of paraphrase identification models from a new perspective – via modification with shared words, and shows that the models have significant robustness issues when facing such modifications.
- MoralDial: A Framework to Train and Evaluate Moral Dialogue Systems via Moral Discussions30
This paper proposes a framework, MoralDial, to train and evaluate moral dialogue systems, and proposes a novel evaluation method under the framework, which evaluates the multiple aspects of morality by judging the relation between dialogue responses and human values in discussions.
- Stylized Dialogue Response Generation Using Stylized Unpaired Texts30
This paper proposes a stylized dialogue generation method that can capture stylistic features embedded in unpaired texts and produce dialogue responses that are both coherent to the given context and conform to the target style.
- Context-aware Natural Language Generation for Spoken Dialogue Systems.30
A Context-Aware LSTM model for NLG is proposed, which is completely driven by data without manual designed templates or rules, and obtains state-of-the-art performance.
- 29
- Dialogue Distillation: Open-Domain Dialogue Augmentation Using Unpaired Data29
Automatic and manual evaluation indicates that the proposed novel data augmentation method for training open-domain dialogue models by utilizing unpaired data can produce high-quality dialogue pairs with diverse contents, and can improve the performance of competitive baselines.
- A Comparative Study on Ranking and Selection Strategies for Multi-Document Summarization29
A systematic study of comparing different learning-to-rank algorithms and comparing different selection strategies on two key problems existing in extractive summarization: the ranking problem and the selection problem is presented.
- A Unified Dialogue User Simulator for Few-shot Data Augmentation28
A unified dialogue user simulation model is built by pre-training on several publicly available datasets and can be tuned on a target domain with few-shot data to bring remarkable performance increases through data augmentation.
- 28
- Learning Goal-oriented Dialogue Policy with opposite Agent Awareness28
An opposite behavior aware framework for policy learning in goal-oriented dialogues that estimates the opposite agent’s policy from its behavior and uses this estimation to improve the target agent by regarding it as part of the target policy.
- A Corpus for Understanding and Generating Moral Stories27
A retrieval-augmented algorithm that effectively exploits related concepts or events in training sets as additional guidance to improve performance on two understanding tasks and two generation tasks to assess the abilities of machines.
- 27
- 26
- 26
- Results of the Multi-Domain Task-Completion Dialog Challenge26
An overview of the “Multi-domain Task Completion” track (Track 1) at the 8th Dialog System Technology Challenge (DSTC-8) is provided, seeking to develop models that predict user responses when only limited in-domain data is available.
- Product Review Summarization by Exploiting Phrase Properties26
This work formalizes the phrase selection procedure as an optimization problem and solve it using integer linear programming (ILP), an aspect-based bigram language model is used for generating the final summary with the selected phrases.
- 25
- DiscoDVT: Generating Long Text with Discourse-Aware Discrete Variational Transformer25
The proposed DiscoDVT, a discourse-aware discrete variational Transformer, learns a discrete variable sequence that summarizes the global structure of the text and applies it to guide the generation process at each decoding step, and introduces an auxiliary objective to model the discourse relations within the text.
- Generating Commonsense Explanation by Extracting Bridge Concepts from Reasoning Paths25
This work proposes a method that first extracts the underlying concepts which are served as bridges in the reasoning chain and then integrates these concepts to generate the final explanation.
- Recognizing Biomedical Named Entities Using Skip-Chain Conditional Random Fields25
A novel study of capturing long distance dependency by defining two principles of constructing skip-edges for a skip-chain CRF: linking similar words and linking words having typed dependencies.
- Does RLHF Scale? Exploring the Impacts From Data, Model, and Method24
Findings show that increasing data diversity and volume improves reward model performance, helping process-supervision models scale better and propose strategies to optimize RLHF performance within computational limits.
- When does Further Pre-training MLM Help? An Empirical Study on Task-Oriented Dialog Pre-training24
DAPT is beneficial in the low-resource setting, but as the fine-tuning data size grows, DAPT becomes less beneficial or even useless, and scaling the size of D APT data does not help.
- ARAML: A Stable Adversarial Training Framework for Text Generation24
Experiments show that the proposed novel framework, Adversarial Reward Augmented Maximum Likelihood (ARAML), can outperform state-of-the-art text GANs with a more stable training process.
- 23
- 23
- Exploring the Interactions of Storylines from Informative News Events23
This paper addresses the coherence between news articles which is not considered in traditional similarity-based methods, and discovers salient storyline interactions to form a clear, global picture of the news topic.
- 23
- Domain-Constrained Advertising Keyword Generation22
This work investigates to use generative neural networks for keyword generation in sponsored search and shows that generative models can improve coverage, click-through rate, and revenue per mille (RPM) substantially insponsored search.
- Finding nuggets in IP portfolios22
This paper addresses the unique patent vocabulary usage problem, which is not considered in traditional word-based statistical methods, and proposes a topic-based temporal mining approach to quantify a patent's novelty and influence.
- CharacterGLM: Customizing Social Characters with Large Language Models21
By dissecting well-rounded social characters composed of both inherent social profiles and external social behaviors, this work manually collect a large-scale Chinese corpus featuring characters with diverse categories and behaviors, and develop CharacterGLM models alongside well-designed refinement methods.
- Goal Awareness for Conversational AI: Proactivity, Non-collaborativity, and Beyond21
This tutorial will introduce the recent advances on the design of agent’s awareness of goals in a wide range of conversational systems and suggest ways to empower the system to handle more complicated conversation tasks that involve strategical and motivational interactions.
- 21
- Semi-Supervised Multinomial Naive Bayes for Text Classification by Leveraging Word-Level Statistical Constraint21
A novel method is proposed to augment MNB-EM by leveraging the word-level statistical constraint to preserve the class distribution on words to improve MNB and outperforms state-of-art baselines remarkably.
- Incorporating domain and sentiment supervision in representation learning for domain adaptation21
This work addresses two key factors to guide the representation learning process for domain adaptation of sentiment classification -- one is domain supervision, enforcing the learned representation to better predict the domain of an input, and the other is sentiment supervision which utilizes the source domain sentiment labels to learn sentiment-favorable representations.
- Quality-biased Ranking of Short Texts in Microblogging Services21
This work proposes a regression approach that incorporates various features to recommend short-text documents from Twitter, with a bias toward quality perspective, built on top of a linear regression model which includes a regularization factor inspired from the content conformity hypothesis.
- ONBIRES: Ontology-Based Biological Relation Extraction System21
A two-level pattern learning algorithm is proposed that is unified and able to extract several kinds of relations such as gene-disease, gene-gene, and protein-protein interactions etc.
- Enhanced large language models for effective screening of depression and anxiety20
A pipeline for synthesizing clinical interviews is introduced, resulting in 1,157 interactive dialogues (PsyInterview), and EmoScan, an LLM-based emotional disorder screening system is presented, revealing that EmoScan effectively screens for emotional disorders with high accuracy and can provide detailed explanations, enhancing transparency and trust in automated mental health assessments.
- SPaR: Self-Play with Tree-Search Refinement to Improve Instruction-Following in Large Language Models20
SPaR is introduced, a self-play framework integrating tree-search self-refinement to yield valid and comparable preference pairs free from distractions, and demonstrates promising scalability, greatly enhancing models like GLM-4-9B and LLaMA3-70B.
- NAST: A Non-Autoregressive Generator with Word Alignment for Unsupervised Text Style Transfer20
A non-autoregressive generator for unsupervised text Style Transfer (NAST), which focuses on the connections between aligned words, which learns the word-level transfer between styles and can significantly improve the overall performance and provide explainable word alignments.
- 20
- Internal Value Alignment in Large Language Models through Controlled Value Vector Activation19
A Controlled Value Vector Activation (ConVA) method is introduced that directly aligns the internal values of LLMs by interpreting how a value is encoded in their latent representations and modifies relevant activations to ensure consistent values in LLMs.
- Transferable Persona-Grounded Dialogues via Grounded Minimal Edits19
Focusing on personas, Grounded Minimal Editor (GME), which learns to edit by disentangling and recombining persona-related and persona-agnostic parts of the response, is proposed, and experimental results show that GME outperforms competitive baselines by a large margin.
- Aggregating E-commerce Search Results from Heterogeneous Sources via Hierarchical Reinforcement Learning19
This paper decomposes the task of aggregating search results from heterogeneous sources in an E-commerce environment into two subtasks and builds a model with hierarchical reinforcement learning that obtains remarkable improvements in search performance metrics, and achieves a higher user satisfaction.
- 18
- 18
- Robust Reading Comprehension With Linguistic Constraints via Posterior Regularization18
The over-confidence issue and the over-sensitivity issue existing in current RC models simultaneously are addressed simultaneously with the help of external linguistic knowledge, and linguistic constraints induce more reasonable predictions for both semantic different and semantic equivalent adversarial examples.
- A New Approach for Multi-Document Update Summarization18
A novel approach for multi-document update summarization which closely correlates with the human summaries and outperforms other programs such as LexRank in many categories under the ROUGE evaluation criterion.
- Extract interaction detection methods from the biological literature18
A generative topic model, the Correlated Method-Word model (CMW model), is developed to extract the detection methods from the literature and shows promising results, which outperform the best result reported in the BioCreative II challenge evaluation.
- 17
- 17
- A Roadmap for Big Model17
This paper discusses not only the BM technologies themselves but also the prerequisites for BM training and applications with BMs, dividing the BM review into four parts: Resource, Models, Key Technologies and Application.
- MAPS: Advancing Multi-Modal Reasoning in Expert-Level Physical Science16
Validated using the collected college-level circuit analysis problems, MAPS significantly improves reasoning accuracy of MLLM and outperforms all existing models, and the results confirm MAPS offers a promising direction for enhancing multi-modal scientific reasoning ability of MLLMs.
- AISafetyLab: A Comprehensive Framework for AI Safety Evaluation and Improvement16
AISafetyLab is introduced, a unified framework and toolkit that integrates representative attack, defense, and evaluation methodologies for AI safety and features an intuitive interface that enables developers to seamlessly apply various techniques while maintaining a well-structured and extensible codebase for future advancements.
- 16
- 16
- Clustering Aspect-related Phrases by Leveraging Sentiment Distribution Consistency16
Through formalizing sentiment distribution consistency as soft constraint, a novel unsupervised model in the framework of Posterior Regularization (PR) to cluster aspectrelated phrases is proposed and demonstrates that it outperforms baselines remarkably.
- CharacterBench: Benchmarking Character Customization of Large Language Models15
This work proposes CharacterBench, the largest bilingual generative benchmark, with 22,859 human-annotated samples covering 3,956 characters from 25 detailed character categories, and develops CharacterJudge model, a cost-effective and stable evaluations of LLMs’ character customization capability.
- SocialSim: Towards Socialized Simulation of Emotional Support Conversation15
This paper introduces SocialSim, a novel framework that simulates ESC by integrating key aspects of social interactions: social disclosure and social awareness, and constructs SSConv, a large-scale synthetic ESC corpus of which quality can even surpass crowdsourced ESC data.
- 15
- 15
- A Semantic-based Method for Unsupervised Commonsense Question Answering15
This paper presents a novel SEmantic-based Question Answering method (SEQA) for unsupervised commonsense question answering that first generates a set of plausible answers with generative models, and then uses these plausible answers to select the correct choice by considering the semantic similarity between each plausible answer and each choice.
- 15
- Advancing Collaborative Debates with Role Differentiation through Multi-Agent Reinforcement Learning14
The core idea of MLC is to initialize role assignments randomly and then allow the role embeddings to be learned jointly with the down-stream task, which enhances collaboration and expertise, which collaboratively addresses multi-agent tasks.
- JPS: Jailbreak Multimodal Large Language Models with Collaborative Visual Perturbation and Textual Steering14
This work proposes JPS, Jailbreak MLLMs with collaborative visual Perturbation and textual Steering, which achieves jailbreaks via corporation of visual image and textually steering prompt, and proposes the Malicious Intent Fulfillment Rate (MIFR) metric.
- Directed Acyclic Transformer Pre-training for High-quality Non-autoregressive Text Generation14
This paper proposes Pre-trained Directed Acyclic Transformer (PreDAT) and a novel pre-training task to promote prediction consistency in NAR generation and shows that PreDAT remarkably outperforms existing pre-trained NAR models and achieves better results than pre- trained autoregressive baselines in n-gram-based metrics.
- 14
- AdvExpander: Generating Natural Language Adversarial Examples by Expanding Text14
AdvExpander, a method that crafts new adversarial examples by expanding text by utilizing linguistic rules and constrain the modifications with a heuristic rule, is presented, promising to reveal new robustness issues of models.
- 14
- Exploring weakly supervised latent sentiment explanations for aspect-level review analysis14
A new concept for aspect-level review analysis is explored, latent sentiment explanations, which are defined as a set of informative aspect-specific sentences whose polarities are consistent with that of the review, which best represents a review in terms of both aspect and polarity.
- DC-Instruct: An Effective Framework for Generative Multi-intent Spoken Language Understanding13
DC-Instruct is proposed, a novel generative framework based on Dual-task Inter-dependent Instructions (DII) and Supervised Contrastive Instructions (SCI) that markedly outperforms current generative models and state-of-the-art methods, demonstrating its effectiveness in enhancing dialogue language understanding and reasoning.
- 13
- HyKnow: End-to-End Task-Oriented Dialog Modeling with Hybrid Knowledge Management13
A TOD system with hybrid knowledge management, HyKnow, is proposed, which extends the belief state to manage both structured and unstructured knowledge, and is the first end-to-end model that jointly optimizes dialog modeling grounded on these two kinds of knowledge.
- Using First-Order Logic to Compress Sentences13
A novel sentence compression model based on first-order logic, using Markov Logic Network is presented, able to incorporate local linguistic features and to capture global dependencies between word deletion operations.
- Knowledge Semantic Representation: A Generative Model for Interpretable Knowledge Graph Embedding.13
A semantic representation method for knowledge graph (KSR) is proposed, which imposes a two-level hierarchical generative process that globally extracts many aspects and then locally assigns a specific category in each aspect for every triple.
- From One Point to A Manifold: Orbit Models for Knowledge Graph Embedding.13
The orbit-based embedding model, OrbitE, is proposed, a well-posed algebraic system that expands the position of golden triples from one point in current models to a manifold and achieves substantial improvements against the state-of-the-art baselines, especially for precise prediction.
- Estimating feature ratings through an effective review selection approach13
This paper proposes a novel approach to accurately estimate feature ratings of products, and selects user reviews that extensively discuss specific features of the products (called specialized reviews), using information distance of reviews on the features.
- 13
- Exploiting and integrating rich features for biological literature classification13
A novel feature value schema TF*ML is proposed, features covering from lower level domain independent “string feature’ to higher level domain dependent “semantic template feature”, and proper integrations among the features, which outperforms the best performance achieved in BioCreAtIvE 2006.
- BlackDAN: A Black-Box Multi-Objective Approach for Effective and Contextual Jailbreaking of Large Language Models12
Experimental results demonstrate that BlackDAN outperforms traditional single-objective methods, yielding higher success rates and improved robustness across various LLMs and multimodal LLMs, while ensuring jailbreak responses are both relevant and less detectable.
- 12
- SS-GEN: A Social Story Generation Framework with Large Language Models11
This work develops a constraint-driven sophisticated strategy named StarSow to hierarchically prompt LLMs to generate Social Stories at scale, followed by rigorous human filtering to build a high-quality dataset, and introduces quality assessment criteria to evaluate the effectiveness of these generated stories.
- 11
- AutoCAD: Automatically Generating Counterfactuals for Mitigating Shortcut Learning11
AutoCAD is presented, a fully automatic and task-agnostic CAD generation framework that consistently and significantly boosts the out-of-distribution performance of powerful pre-trained models across different NLU tasks, which is comparable or even better than previous state-of the-art human-in-the-loop or task-specific CAD methods.
- Learning to Annotate Scientific Publications11
An automatic approach of annotating scientific publications based on a machine learning algorithm and a set of novel features shows significantly improved performance when compared to other methods.
- Scenario-independent Uncertainty Estimation for LLM-based Question Answering via Factor Analysis10
A plug-and-play scenario-independent framework to enhance unsupervised UE in LLMs by removing scenario-related noise and focusing on semantic information is introduced and leverages only the existing UE methods' outputs.
- Ψ-Arena: Interactive Assessment and Optimization of LLM-based Psychological Counselors with Tripartite Feedback10
PsychoArena is proposed, an interactive framework for comprehensive assessment and optimization of LLM-based counselors, featuring three key characteristics: Realistic arena interactions that simulate real-world counseling through multi-stage dialogues with psychologically profiled NPC clients, and closed-loop optimization that iteratively improves LLM counselors using diagnostic feedback.
- ShieldVLM: Safeguarding the Multimodal Implicit Toxicity via Deliberative Reasoning with LVLMs: ShieldVLM10
ShieldVLM is built, a model which identifies implicit toxicity in multimodal statements, prompts and dialogs via deliberative cross-modal reasoning and outperforms existing strong baselines in detecting both implicit and explicit toxicity.
- 10
- Language Model Decoding as Direct Metrics Optimization10
This work frames decoding from a language model as an optimization problem with the goal of strictly matching the expected performance with human texts measured by multiple metrics of desired aspects simultaneously, and proves that this induced distribution is guaranteed to improve the perplexity on human texts.
- 10
- KuiLeiXi: a Chinese Open-Ended Text Adventure Game10
In KuiLeiXi, players need to interact with the AI until the pre-determined plot goals are reached, and players have a stronger incentive to explore ways to reach plot goals, while the AI’s abilities are not abused to generate harmful contents.
- ExpanRL: Hierarchical Reinforcement Learning for Course Concept Expansion in MOOCs10
Employing a two-level HRL mechanism of seed selection and concept expansion, ExpanRL is more feasible to adjust the expansion strategy to find new concepts based on the students’ feedback on expansion results.
- An Operation Network for Abstractive Sentence Compression.10
Operation Network is proposed, a neural network approach for abstractive sentence compression, which combines the advantages of both delete-based and generate-based sentence compression models and outperforms state-of-the-art baselines.
- Question routing in community based QA: incorporating answer quality and answer content10
This paper addresses a key issue about cQA systems: routing newly asked questions to appropriate users that may potentially provide answer with high quality, and incorporates answer quality and answer content to build a probabilistic question routing model.
- BARREL: Boundary-Aware Reasoning for Factual and Reliable LRMs9
This work identifies two pathological reasoning patterns characterized by overthinking that contribute to the overconfident and incorrect answers: last-minute guessing and second-thought spiraling and proposes BARREL-a novel framework that promotes concise and boundary-aware factual reasoning.
- Thoughts to Target: Enhance Planning for Target-driven Conversation9
A novel two-stage framework, named EnPL, is proposed to improve the LLMs’ capability in planning conversations towards designated targets, including distilling natural language plans from target-driven conversation corpus and generating new plans with demonstration-guided in-context learning.
- 9
- How Well Do Large Language Models Serve as End-to-End Secure Code Agents for Python?9
A lightweight tool is developed that prompts LLMs to construct safer source code through an iterative repair procedure based on the insights gained from the study of GPT-3.5 and GPT-4's capability to identify and repair vulnerabilities in the code generated by four popular LLMs.
- Guided Structure-Aware Review Summarization9
A novel framework to generate guided summaries for product reviews that attempts to maximize expected aspect satisfaction during summary generation and is modeled using Labeled Latent Dirichlet Allocation.
- Towards Optimal Learning of Language Models8
A theory for the optimal learning of LMs is presented, and it is empirically verified that the optimal learning of LMs essentially stems from the improvement of the coefficients in the scaling law of LMs, indicating great promise and significance for designing practical learning acceleration methods.
- 8
- Automatic Comment Generation for Chinese Student Narrative Essays8
A planning-based generation model is proposed, which first plans a sequence of keywords, and then expands these keywords into a complete comment, which outperforms strong baselines significantly and exerts explicit control on the model to generate comments to describe the strengths or weaknesses of inputs.
- End-to-End Task-Oriented Dialog Modeling With Semi-Structured Knowledge Management8
This article proposes a TOD system with semi-structured knowledge management, SeKnow, which extends the belief state to manage knowledge with both structured and unstructured contents, and introduces two implementations of SeKnow based on a non-pretrained sequence-to-sequence model and a pretrained language model.
- Tsinghua University at TAC 2009: Summarizing Multi-documents by Information Distance8
This paper presents the extractive summarization systems at the update summarization track of TAC 2009, based on the newly developed document summarization framework under the theory of conditional information distance among many objects.
- Learning Domain-Specific Knowledge from Context--THUIR at TREC 2005 Genomics Track.8
This work designed and implemented a serious of methods encompassed learning domain-specific knowledge from context that achieved success in the four subtasks of Ad Hoc Retrieval Task and Categorization Task.
- Bridging the Gap between Synthetic and Natural Questions via Sentence Decomposition for Semantic Parsing7
This paper proposes a decomposition-based method to unify the sentence structures of questions, which benefits the generalization to natural questions and extends to the task of sentence embedding learning, and observes substantial improvements on sentence retrieval.
- 7
- A Sentence Interaction Network for Modeling Dependence between Sentences7
A Sentence Interaction Network (SIN) is proposed for modeling the complex interactions between two sentences by introducing “interaction states” for word and phrase pairs and obtaining improvements on Answer Selection and Dialogue Act Analysis without any feature engineering.
- 6
- 6
- Enhancing Offensive Language Detection with Data Augmentation and Knowledge Distillation6
AugCOLD (Augmented Chinese Offensive Language Dataset), a large-scale unsupervised dataset containing 1 million samples gathered by data crawling and model generation, is introduced and a multiteacher distillation framework is employed to enhance detection performance with unsuper supervised data.
- 6
- 6
- WSpeller: Robust Word Segmentation for Enhancing Chinese Spelling Check6
WSpeller, a CSC model that takes into account word segmentation, is presented and is superior to state-of-the-art baselines on SIGHAN13 and SIGHAN15 and maintains equal performance on SIGHAN14.
- 6
- Overview of the NLPCC 2017 Shared Task: Emotion Generation Challenge6
The details of this challenge to investigate how well a chatting machine can express emotion by generating a textual response to an input post, including task definition, data preparation, annotation schema, submission statistics, and evaluation results are reported.
- Clustering Sentiment Phrases in Product Reviews by Constrained Co-clustering6
A two-sided hidden Markov random field HMRF model is applied on clustering two-dimensional sentiment phrases in product reviews and this method outperforms baselines remarkably.
- 6
- 6
- SEMI-SUPERVISED PATTERN LEARNING FOR EXTRACTING RELATIONS FROM BIOSCIENCE TEXTS6
A semisupervised model is presented to combine both unlabeled and labeled data for the pattern learning procedure, showing that labeled data, when used in conjunction with the inexpensive unlabeling data, can considerably improve the learning accuracy.
- 5
- 5
- Be Careful When Fine-tuning On Open-Source LLMs: Your Fine-tuning Data Could Be Secretly Stolen!5
These comprehensive experiments suggest that the extraction performance can be strikingly high: in practical settings, as much as 76.3% downstream fine-tuning data (queries) out of a total 5,000 samples can be perfectly extracted, and the success rate can increase to 94.9% in more ideal settings.
- 5
- 5
- 5
- Multi-domain Task-oriented Dialog Challenge II5
This proposal extends the tasks by incorporating new datasets, creating new sub-tasks, and providing a new development platform, and specifically focuses on two aspects of dialog systems: language portability and end-to-end system complexity.
- Sentiment Extraction by Leveraging Aspect-Opinion Association Structure5
This paper investigates the aspect-opinion association structure, and proposes a "first clustering, then extracting" unsupervised model to leverage properties of the structure for sentiment extraction, and shows that cluster-cluster association is a much stronger signal to distinguish aspect (opinion) words from non-aspect (non-operative) words.
- Ranking Sentiment Explanations for Review Summarization Using Dual Decomposition5
This paper addresses the problem of ranking sentiment explanations by formulating the process as two subproblems: sentence informativeness ranking and structural sentiment analysis.
- Finding nuggets in patent portfolios: core patent mining and its applications5
The work addresses the unique patent vocabulary usage which is not considered in traditional word-based statistical methods with a topic-based temporal mining approach that quantifies a patent's novelty and influence through topic activeness variations.
- Summarizing Documents by Measuring the Importance of a Subset of Vertices within a Graph5
A novel method of generating extractive summaries for multiple documents by developing a method to measure the importance of a subset of vertices by adding a super-vertex into the original graph and proposing a heuristic algorithm to find the best summary.
- An Ontology-Based Mining System for Competitive Intelligence in Neuroscience5
An ontology-based mining system, which is able to probe Competitive Intelligence in Neuroscience by using a well-defined ontology, able to support decision making by searching neuroscientific discoveries semantically and tracking new trends statistically is presented.
- 4
- 4
- When Smiley Turns Hostile: Interpreting How Emojis Trigger LLMs' Toxicity4
A comprehensive exploration of emoji-triggered LLM toxicity generation is begun by automating the construction of prompts with emojis to subtly express toxic intent, and model-level interpretations spanning semantic cognition, sequence generation and tokenization suggest that emojis can act as a heterogeneous semantic channel to bypass the safety mechanisms.
- Seeker: Towards Exception Safety Code Generation with Intermediate Language Agents Framework4
This work proposes Seeker, a multi-agent framework inspired by expert developer strategies for exception handling that uses agents: Scanner, Detector, Predator, Ranker, and Handler to assist LLMs in detecting, capturing, and resolving exceptions more effectively.
- 4
- 4
- 4
- 4
- 4
- 4
- Construction and Evaluation of QOL Specialized Dictionary SqolDic Using Vocabulary Meaning and QOL Scale4
This study created a QOL-specific dictionary called SqolDic, which is based on large-scale Japanese textual data and created a system that outputs the time-series variation of a user’s conversation content and the QOL scores based on it, and proposed a new text-based estimation method.
- 4
- Sentiment Analysis with Multi-source Product Reviews4
A binary classification framework based on the aspects of product is proposed and both explicit and implicit aspect is considered and multiple kinds of feature weighing and classifiers are compared in this framework.
- 3
- 3
- AGD: Adversarial Game Defense Against Jailbreak Attacks in Large Language Models3
It is argued that adversarial game-based approaches promise a solution for conflicts be-tween the two objectives of helpfulness and harmlessness and proposed AGD, an adversarial game-based defense method that dynamically adjusts LLMs’ internal representations to achieve a balanced trade-off between helpfulness and harmlessness.
- Battling against Tough Resister: Strategy Planning with Adversarial Game for Non-collaborative Dialogues3
GAIA, a G ame-based adversarial self-play training paradigm, is proposed, which constructs an adversarial two-player (a persuader and a resister) zero-sum game and guides the game to approximate Nash Equilibrium (NE) via reinforcement learning (RL) for the non-collaborative dialogues.
- 3
- 3
- Re3Dial: Retrieve, Reorganize and Rescale Conversations for Long-Turn Open-Domain Dialogue Pre-training3
The Retrieve, Reorganize and Rescale framework (Re$^3$Dial), which can automatically construct billion-scale long-turn dialogues by reorganizing existing short-turn ones, is proposed and significantly improves the dialogue model's ability to utilize long-range context and thus generate more sensible and informative responses.
- 3
- 3
- 3
- 3
- 3
- Independence-aware Advantage Estimation3
The recognized independence property can be naturally utilized to construct a novel importance sampling advantage estimator with close-to-zero variance even when the Monte-Carlo return signal yields a large variance.
- Scalable Functional Dependencies Discovery from Big Data3
This paper proposes an efficient functional dependencies discovery algorithm, for mining functional dependencies from distributed big data, and shows that the proposed algorithm is effective on real dataset and synthetic dataset.
- 3
- The superalignment of superhuman intelligence with large language models2
A conceptual framework for superalignment is presented, which comprises three modules: an attacker which generates the adversary queries trying to expose the weaknesses of a learner model, a learner which refines itself by learning from scalable feedbacks generated by a critic model with minimal human experts, and a critic which generates critics or explanations for a given query-response pair.
- 2
- DPRM: A Dual Implicit Process Reward Model in Multi-Hop Question Answering2
The DPRM (Dual Implicit Process Reward Model) trains two implicit PRMs for CoT and KG reasoning in MHQA tasks and introduces a consistency constraint between CoT and KG reasoning steps, making the two PRMs mutually verify and collaboratively optimize the reasoning paths.
- 2
- 2
- A Benchmark for Understanding and Generating Dialogue between Characters in Stories2
This is the first study to explore whether machines can understand and generate dialogue in stories, which requires capturing traits of different characters and the relationships between them, and proposes two new tasks including Masked Dialogue Generation and Dialogue Speaker Recognition.
- Chat-Capsule: A Hierarchical Capsule for Dialog-level Emotion Analysis2
A Context-based Hierarchical Attention Capsule~(Chat-Capsule) model, which models both utterance-level and dialog-level emotions and their interrelations and is able to predict user satisfaction and emotion curve category.
- 2
- Leveraging Large Data with Weak Supervision for Joint Feature and Opinion Word Extraction2
This paper proposes a data-driven approach to represent product features and opinion words as a list of corpus-level syntactic relations, which captures rich language structures and builds a simple yet robust unsupervised model with prior knowledge incorporated to extract new feature and opinionWords, which obtains high performance robustly.
- Combining Convolution Kernels Defined on Heterogeneous Sub-structures2
This paper studies the problem of sub-structure selection for constructing convolution kernels by combining heterogeneous kernels defined on different levels of substructures, and analyzes from the perspective of feature space why combined kernels exhibit potential advantages.
- 1
- HCDS: Hierarchical Clustering for Cold-Start Few-Shot Data Selection1
Experimental results on six public datasets, including both balanced and imbalanced ones, show that HCDS achieves state-of-the-art performance, particularly with limited and imbalanced data.
- 1
- 1
- 1
- 1
- 1
- 1
- Many-Class Text Classification with Matching1
This work forms a simple yet effective framework named TCM that takes advantage of the fine-grained semantic information of the classification labels, which helps distinguish each class better when the class number is large, especially in low-resource scenarios.
- Manual-Guided Dialogue for Flexible Conversational Agents1
This paper introduces three dialogue modeling subtasks: instruction matching, argument filling, and response generation, and reduces the dependence of dialogue models on fine-grained domain ontology, and makes them more flexible to adapt to various domains.
- Adversarial Examples with Difficult Common Words for Paraphrase Identification1
A novel algorithm is proposed to generate a new type of adversarial examples to study the robustness of deep paraphrase identification models and it is shown that adversarial training with generated adversarialExamples can improve model robustness.
- –
- –
- A Differentially Private Collaborative Intelligence Structure for Trust Aware Secure Data Exchange in Distributed Computing Platforms–
The proposed framework combines adaptive differential privacy optimization and the trust-aware collaboration management to enable secure and reliable information sharing and achieves better privacy protection, less communication overhead and better collaborative intelligence performance than traditional privacy-aware exchange mechanisms.
- –
- –
- Self-Supervised Sentence Polishing by Adding Engaging Modifiers–
This work presents the first study on automatic sentence polishing by adding modifiers, and designs a retrieval augmentation algorithm to prompt the model to add suitable modifiers.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- A Novel Kernel-Based Approach for Predicting Binding Peptides for HLA Class II Molecules–
The substitution score between amino acids in BLOSUM matrix is incorporated into computing the similarity between two binding peptides, which exhibits more biological meaning over traditional string kernels, and the promising results of this approach show advantages than other methods.
- –
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.