Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileAcademic lineage
Possible advisorsa guess from early papers, not confirmed
- Ting LiuSuggested from co-authorship
- Ming ZhouSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works102 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.
- Document Modeling with Gated Recurrent Neural Network for Sentiment Classification1,519
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.
- Adaptive Recursive Neural Network for Target-dependent Twitter Sentiment Classification1,074
AdaRNN adaptively propagates the sentiments of words to target depending on the context and syntactic relationships between them and it is shown that AdaRNN improves the baseline methods.
- Aspect Level Sentiment Classification with Deep Memory Network1,041
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.
- scBERT as a large-scale pretrained deep language model for cell type annotation of single-cell RNA-seq data682
A pretrained deep neural network-based model, single-cell bidirectional encoder representations from transformers (scBERT), to overcome the challenges of existing annotation methods, and is validated on cell type annotation, novel cell type discovery, robustness to batch effects and model interpretability.
- K-Adapter: Infusing Knowledge into Pre-Trained Models with Adapters648
K-Adapter is proposed, which remains the original parameters of the pre-trained model fixed and supports continual knowledge infusion and captures richer factual and commonsense knowledge than RoBERTa.
- Improving Question Answering by Commonsense-Based Pre-training489
Results show that incorporating commonsense-based function improves the baseline on three question answering tasks that require commonsense reasoning and leverages useful evidence from an external commonsense knowledge base, which is missing in existing neural network models.
- 400
- Learning Semantic Representations of Users and Products for Document Level Sentiment Classification365
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
- Question Generation for Question Answering299
Experimental results show that, by using generated questions as an extra signal, significant QA improvement can be achieved.
- Compare to The Knowledge: Graph Neural Fake News Detection with External Knowledge266
A novel end-to-end graph neural model called CompareNet is proposed, which compares the news to the knowledge base (KB) through entities for fake news detection and significantly outperforms state-of-the-art methods.
- ToolACE: Winning the Points of LLM Function Calling253
This paper presents ToolACE, an automatic agentic pipeline designed to generate accurate, complex, and diverse tool-learning data, which leverages a novel self-evolution synthesis process to curate a comprehensive API pool of 26,507 diverse APIs.
- 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.
- Graph-Based Reasoning over Heterogeneous External Knowledge for Commonsense Question Answering232
This work proposes to automatically extract evidence from heterogeneous knowledge sources, and answer questions based on the extracted evidence, and achieves the state-of-the-art accuracy on the CommonsenseQA dataset.
- 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).
- Reasoning Over Semantic-Level Graph for Fact Checking191
This work proposes two mechanisms to exploit the structure of evidence while leveraging the advances of pre-trained models like BERT, GPT or XLNet, and is the state-of-the-art system in terms of both official evaluation metrics, namely claim verification accuracy and FEVER score.
- 183
- 163
- 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.
- CoSQA: 20,000+ Web Queries for Code Search and Question Answering162
A contrastive learning method dubbed CoCLR is introduced to enhance text-code matching, which works as a data augmenter to bring more artificially generated training instances to bring better semantic matching between query and code.
- WhiteningBERT: An Easy Unsupervised Sentence Embedding Approach153
This work conducts a thorough examination of pretrained model based unsupervised sentence embeddings and concludes that an easy whitening-based vector normalization strategy with less than 10 lines of code consistently boosts the performance.
- 152
- 147
- 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.
- Table-to-Text: Describing Table Region With Natural Language116
A generative model to generate a natural language sentence describing a table region, e.g., a row, and a flexible copying mechanism that selectively replicates contents from the table in the output sequence is presented.
- Dialog-to-action: conversational question answering over a large-scale knowledge base110
An approach to map utterances in conversation to logical forms, which will be executed on a large-scale knowledge base, and shows that the semantic parsing-based approach outperforms a memory network based encoder-decoder model by a huge margin.
- Logic-Driven Context Extension and Data Augmentation for Logical Reasoning of Text94
The results show that the proposed logic-driven data augmentation algorithm achieves state-of-the-art performance on both datasets, and even surpasses human performance on the ReClor dataset.
- Asking Clarification Questions in Knowledge-Based Question Answering88
A new clarification dataset, CLAQUA, with nearly 40K open-domain examples is constructed and a coarse-to-fine model for clarification question generation is introduced, showing that the proposed model achieves better performance than strong baselines.
- Multi-Task Learning for Conversational Question Answering over a Large-Scale Knowledge Base85
This work proposes an innovative multi-task learning framework where a pointer-equipped semantic parsing model is designed to resolve coreference in conversations, and naturally empower joint learning with a novel type-aware entity detection model.
- 85
- Neural Deepfake Detection with Factual Structure of Text83
Results of experiments show that the proposed graph-based model for deepfake detection significantly improves strong base models built with RoBERTa and indicates that the model can distinguish the difference in the factual structure between machine-generated text and human-written text.
- 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.
- Question Generation from SQL Queries Improves Neural Semantic Parsing68
This study conducts a study on WikiSQL, the largest hand-annotated semantic parsing dataset to date, and demonstrates that question generation is an effective method that empowers us to learn a state-of-the-art neural network based semantic parser with thirty percent of the supervised training data.
- LogicalFactChecker: Leveraging Logical Operations for Fact Checking with Graph Module Network63
This work proposes LogicalFactChecker, a neural network approach capable of leveraging logical operations for fact checking, which achieves the state-of-the-art performance on TABFACT, a large-scale, benchmark dataset built for verifying a textual statement with semi-structured tables.
- Sentiment-Specific Representation Learning for Document-Level Sentiment Analysis60
It is argued that learning sentiment-specific semantic representations of documents is crucial for document-level sentiment analysis and decomposes the document semantics into four cascaded constitutes.
- Syntax-Enhanced Pre-trained Model56
It is demonstrated that infusing automatically produced syntax of text improves pre-trained models, and global syntactic distances among tokens bring larger performance gains compared to local head relations between contiguous tokens.
- Coupling Retrieval and Meta-Learning for Context-Dependent Semantic Parsing51
An approach to incorporate retrieved datapoints as supporting evidence for context-dependent semantic parsing, such as generating source code conditioned on the class environment, and shows that both the context-aware retriever and the meta-learning strategy improve accuracy.
- Learning to Collaborate for Question Answering and Asking51
A training algorithm is presented that generalizes both Generative Adversarial Network (GAN) and Generative Domain-Adaptive Nets (GDAN) under the question answering scenario and finds that the performance of a QG model could be easily improved by a QA model via policy gradient.
- 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.
- SWE-Factory: Your Automated Factory for Issue Resolution Training Data and Evaluation Benchmarks42
This paper introduces SWE-Builder, a multi-agent system that automates evaluation environment construction, which employs four specialized agents that work in a collaborative, iterative loop and leverages an environment memory pool to enhance efficiency, and introduces a standardized, exit-code-based grading method that eliminates the need for manually writing custom parsers.
- 42
- Deep Learning in Sentiment Analysis40
An overview of the successful deep learning approaches sentiment analysis tasks at different levels is given, including sentiment classification, opinion extraction, fine-grained sentiment analysis, etc.
- UserAdapter: Few-Shot User Learning in Sentiment Analysis39
This work introduces a lightweight approach dubbed UserAdapter, which clamps hundred millions of parameters of the Transformer model and optimizes a tiny user-specific vector and offers an efficient way to produce a personalized Transformer model with less than 0.5% parameters added for each user.
- One Model, Multiple Modalities: A Sparsely Activated Approach for Text, Sound, Image, Video and Code33
It is found that pretraining significantly improves the performance of SkillNet on five modalities, on par with or even better than baselines with modality-specific pretraining.
- Analytical Reasoning of Text31
This paper studies the challenge of analytical reasoning of text and collects a new dataset consisting of questions from the Law School Admission Test from 1991 to 2016, and presents an approach dubbed ARM, which outperforms pre-trained models significantly.
- 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.
- 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.
- 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.
- 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.
- Assertion-Based QA With Question-Aware Open Information Extraction20
A new dataset called WebAssertions is introduced, which includes hand-annotated QA labels for 358,427 assertions in 55,960 web passages and shows that ABQA features significantly improve the accuracy on passage-based QA.
- Effidit: An Assistant for Improving Writing Efficiency18
Effidit (Efficient and Intelligent Editing), a digital writing assistant that facilitates users to write higher-quality text more efficiently through the use of Artificial Intelligence (AI) and Natural Language Processing (NLP) technologies is introduced.
- Mixture of Lookup Experts15
Experiments show that, with the same FLOPs and VRAM usage, MoLE achieves inference speeds comparable to dense models and significantly faster than MoE with experts offloading, while maintaining performance on par with MoE.
- Machine Reasoning: Technology, Dilemma and Future15
This tutorial will describe the motivation of this tutorial, introduce typical machine reasoning frameworks, including symbolic reasoning, probabilistic reasoning, neural-symbolic reasoning and neural-evidence reasoning, and show their successful applications in real-world scenarios.
- 15
- 14
- 14
- Evidence-Aware Inferential Text Generation with Vector Quantised Variational AutoEncoder14
The proposed approach is an approach that automatically finds evidence for an event from a large text corpus, and leverages the evidence to guide the generation of inferential texts, and provides state-of-the-art performance on both Event2mind and Atomic datasets.
- 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.
- 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.
- 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.
- Effidit: Your AI Writing Assistant12
Effidit (Efficient and Intelligent Editing), a digital writing assistant that facilitates users to write higher-quality text more efficiently by using artificial intelligence (AI) technologies, is introduced.
- Advances and Frontiers of LLM-based Issue Resolution in Software Engineering: A Comprehensive Survey10
This paper provides a comprehensive analysis of methodologies, spanning training-free frameworks with their modular components to training-based techniques, including supervised fine-tuning and reinforcement learning, covering automated collection and synthesis approaches.
- Current World Models Lack a Persistent State Core10
This work introduces WRBench, the first systematic diagnostic benchmark that treats camera motion as an intervention on observability and resolves evaluation into a human-calibrated chain that asks whether the camera executes the requested interaction, whether the scene stays continuous and identifiable while in view, and whether a returning target remains consistent with the event that was set in motion.
- 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.
- 9
- Not All Models Suit Expert Offloading: On Local Routing Consistency of Mixture-of-Expert Models8
This paper proposes two metrics to measure local routing consistency of MoE models and reveals that domain-specialized experts contribute more to routing consistency than vocabulary-specialized ones, and that most models balance between cache effectiveness and efficiency with cache sizes approximately twice the active experts.
- 8
- SWE Data Construction, Automatically!7
This paper introduces SWE-Builder, an LLM-based agentic system that automates evaluation environment construction, and introduces a standardized, exit-code-based log parsing method to automatically extract test status, enabling a fully automated fail2pass validation.
- SkillNet-NLU: A Sparsely Activated Model for General-Purpose Natural Language Understanding7
This paper presents a Pathways approach to handle many tasks at once with the ability of stitching together existing skills to learn new tasks more effectively, and calls this model SkillNet, which outperforms baseline systems when being extended to new tasks.
- Leveraging Declarative Knowledge in Text and First-Order Logic for Fine-Grained Propaganda Detection7
Experiments show that the method achieves superior performance, demonstrating that injecting declarative knowledge expressed in both natural language and first-order logic can help the model to make more accurate predictions.
- 7
- 7
- Entity disambiguation with memory network7
Experimental results demonstrate that the proposed computational approach empirically surpasses strong feature based and neural network based methods for entity disambiguation.
- HardcoreLogic: Challenging Large Reasoning Models with Long-tail Logic Puzzle Games6
The introduction of HardcoreLogic, a challenging benchmark of over 5,000 puzzles across 10 games, designed to test the robustness of LRMs on the"long-tail" of logical games, exposes the limitations of current LRMs and establishes a benchmark for advancing high-level logical reasoning.
- 6
- 6
- 6
- Skillnet-NLG: General-Purpose Natural Language Generation with a Sparsely Activated Approach5
Results show that, with only one model file, SkillNet-NLG outperforms previous best performance methods on four of five tasks and surpasses baseline systems when adapted to new tasks.
- 5
- Entity disambiguation with decomposable neural networks5
This work decomposes the problem into subproblems and develops various neural network architectures, all of which are purely data‐driven and capable of learning continuous representations of the mention and the entity from data.
- 4
- 4
- 3
- Pretraining Chinese BERT for Detecting Word Insertion and Deletion Errors3
A simple and effective Chinese pretrained model is presented to enable the model to determine whether a word exists at a particular position, by introducing a special token \texttt{[null]}, the prediction of which stands for the non-existence of a word.
- AutoTool: Automatic Scaling of Tool-Use Capabilities in RL via Decoupled Entropy Constraints2
This work proposes a novel training paradigm that first employs warm-up supervised fine-tuning to help models distinguish between simple and complex problems, followed by RL that enable models to automatically determine appropriate reasoning trajectories, and introduces an entropy-based long-short reasoning fusion RL strategy.
- 2
- Unshackling Context Length: An Efficient Selective Attention Approach through Query-Key Compression2
Efficient Selective Attention (ESA), a novel approach that extends context length by efficiently selecting the most critical tokens at the token level to compute attention, reduces the computational complexity of token selection by compressing query and key vectors into lower-dimensional representations.
- 2
- 2
- 2
- 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.
- Emage: Non-Autoregressive Text-to-Image Generation1
This work explores non-autoregressive text-to-image models that efficiently generate hundreds of image tokens in parallel and runs 16 times to generate images of competitive quality with an order of magnitude lower inference latency.
- 1
- –
- 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.
- –
- –
- 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.
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 2026-10-14. 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.