Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks101 from public data
- 660
- A Survey on Legal Judgment Prediction: Datasets, Metrics, Models and Challenges127
This study aims to provide a comprehensive review for NLP researchers and legal professionals to understand the advances in LJP over the past years, and to facilitate further joint efforts towards improving the performance of LJP models.
- Improving Multi-turn Dialogue Modelling with Utterance ReWriter125
This paper proposes rewriting the human utterance as a pre-process to help multi-turn dialgoue modelling and introduces a Transformer-based utterance rewriting architecture using the pointer network, which achieves remarkably good performance on the utterance rewrite task.
- A Conditional Variational Framework for Dialog Generation120
This paper proposes a framework allowing conditional response generation based on specific attributes, which can be either manually assigned or automatically detected and validated on two different scenarios, where the attribute refers to genericness and sentiment states respectively.
- Improving Variational Encoder-Decoders in Dialogue Generation104
A separate VED model is developed that learns to autoencode discrete texts into continuous embeddings and generalize latent representations by reconstructing the encoded embedding using Gaussian noise and multi-layer perceptrons.
- Large Language Models Empowered Personalized Web Agents90
A Personalized User Memory-enhanced Alignment (PUMA) framework is proposed to adapt LLM-empowered personalized Web agents to adapt LLMs to the personalized Web agent task, and extensive experiments validate the superiority of PUMA over existing Web agents on PersonalWAB.
- LawBench: Benchmarking Legal Knowledge of Large Language Models89
Using LawBench, the first evaluation benchmark composed of 20 tasks aimed to assess the ability of Large Language Models (LLMs) to perform Chinese legal-related tasks, this work presents a comprehensive evaluation of 21 popular LLMs and the first comparative analysis of the empirical results.
- 81
- State-of-health estimation based on real data of electric vehicles concerning user behavior80
A SOH estimation method based on real data of electric vehicles concerning user behavior and Bayesian optimization algorithm is used to ensure the generalization of the model when different electric vehicles produce different user behaviors.
- Logical Fallacy Detection70
It is found that existing pretrained large language models perform poorly on this task, and it is shown that a simple structure-aware classifier outperforms the best language model by 5.46% on Logic and 4.51% on logicClimate.
- EasyAug: An Automatic Textual Data Augmentation Platform for Classification Tasks59
EasyAug is a data augmentation platform that provides several augmentation approaches, such that with minimal effort a new method can comprehensively be compared with the baselines and can easily choose the most suitable one for their own dataset.
- On Training Instance Selection for Few-Shot Neural Text Generation52
This work proposes a simple selection strategy with K-means clustering in few-shot neural text generation, and shows that even with the naive clustering-based approach, the generation models consistently outperform random sampling on three text generation tasks: data-to-text generation, document summarization and question generation.
- Neural Data-to-Text Generation via Jointly Learning the Segmentation and Correspondence52
This work proposes to explicitly segment target text into fragment units and align them with their data correspondences to maintain the same expressive power as neural attention models, while being able to generate fully interpretable outputs with several times less computational cost.
- Diversifying Dialogue Generation with Non-Conversational Text48
This paper collects a large-scale non-conversational corpus from multi sources including forum comments, idioms and book snippets and presents a training paradigm to effectively incorporate these text via iterative back translation.
- 47
- Neural Data-to-Text Generation with LM-based Text Augmentation46
This weakly supervised training paradigm is able to outperform fully supervised sequence-to-sequence models with less than 10% of the training set and can boost the performance of a standard sequence- to-sequence model by over 5 BLEU points, establishing a new state-of-the-art on both datasets.
- RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining44
RoCBert is a pretrained Chinese Bert that is robust to various forms of adversarial attacks like word perturbation, synonyms, typos, etc, and is pretrained with the contrastive learning objective which maximizes the label consistency under different synthesized adversarial examples.
- Learning Fine-Grained Fact-Article Correspondence in Legal Cases38
This work creates a corpus with manually annotated fact-article correspondences and treats the learning as a text matching task and proposes a multi-level matching network to address it, which can improve the recommendation accuracy by a large margin.
- CHiP: Cross-modal Hierarchical Direct Preference Optimization for Multimodal LLMs37
This work introduces a visual preference optimization module within the DPO framework, enabling MLLMs to learn from both textual and visual preferences simultaneously, and proposes a hierarchical textual preference optimization module that allows the model to capture preferences at multiple granular levels, including response, segment, and token levels.
- Multi-Layer Visual Feature Fusion in Multimodal LLMs: Methods, Analysis, and Best Practices36
This paper systematically investigates two core aspects of multi-layer visual feature fusion, selecting the most effective visual layers and identifying the best fusion approach with the language model.
- 36
- AST-Transformer: Encoding Abstract Syntax Trees Efficiently for Code Summarization36
Experiments show that AST-Transformer outperforms the state-of-arts by a substantial margin while being able to reduce 90 ~ 95% of the computational complexity in the encoding process.
- MovieChats: Chat like Humans in a Closed Domain34
This work takes a close look at the movie domain and presents a large-scale high-quality corpus with fine-grained annotations in hope of pushing the limit of movie-domain chatbots.
- Estimation of Gap Between Current Language Models and Human Performance32
Evaluating LMs in a generative setting and running a human evaluation on the generated sentences imply that LMs need about 10 to 20 more years of research before human performance is reached.
- Improving Latent Alignment in Text Summarization by Generalizing the Pointer Generator31
This paper allows the model to “edit” pointed tokens instead of always hard copying them, and shows the model is able to capture more latent alignment relations than exact word matches, improve word alignment accuracy, and generate higher-quality summaries validated by both qualitative and quantitative evaluations.
- NEXUS Network: Connecting the Preceding and the Following in Dialogue Generation31
It is argued that a good response should smoothly connect both the preceding dialogue history and the following conversations, and an auxiliary continuous code space is introduced to sidestep the non-differentiability of discrete natural language tokens.
- 30
- Select and Attend: Towards Controllable Content Selection in Text Generation29
This paper decoupled content selection from the decoder, paving the way for controllable content selection in text generation and proposes an effective way to trade-off between performance and controllability with a single adjustable hyperparameter.
- Multimodal Language Models See Better When They Look Shallower28
This work offers the first principled study of visual layer selection in MLLMs, showing that MLLMs can often see better when they look shallower, and proposes a lightweight feature fusion method that strategically incorporates shallower layers, achieving consistent improvements over both single-layer and specialized fusion baselines.
- Dialogue Generation With GAN27
A Generative Adversarial Network (GAN) is presented to model multiturn dialogue generation, which trains a latent hierarchical recurrent encoder-decoder simultaneously with a discriminative classifier that make the prior approximate to the posterior.
- StableMask: Refining Causal Masking in Decoder-only Transformer26
StableMask is proposed: a parameter-free method that introduces pseudo-attention values to balance attention distributions and encodes absolute positional information via a progressively decreasing mask ratio and naturally supports efficient extrapolation without special tricks such as StreamingLLM and easy integration with existing attention optimization techniques.
- From Calculation to Adjudication: Examining LLM judges on Mathematical Reasoning Tasks25
This analysis uncovers a strong correlation between judgment performance and the candidate model task performance, indicating that judges tend to favor higher-quality models even if their answer is incorrect.
- Low-Resource Dense Retrieval for Open-Domain Question Answering: A Comprehensive Survey25
A thorough structured overview of mainstream techniques for low-resource DR is provided, which divides the techniques into three main categories: only documents are needed; only documents and questions are needs; and (3) documents and question-answer pairs are needed.
- 25
- Data Augmentation for Multiclass Utterance Classification – A Systematic Study23
A comprehensive survey regarding data augmentation approaches for text classification, including simple random resampling, word-level transformations, and neural text generation to cope with imbalanced data, shows that the effectiveness of differentData augmentation schemes depends on the nature of the dataset under consideration.
- 19
- Question Rewriting for Open-Domain Conversational QA19
While conversation history modeling with dense representations outperforms QR, it is shown the advantages to apply both jointly, as QR boosts the performance especially when limited history turns are considered.
- 18
- DART: A Lightweight Quality-Suggestive Data-to-Text Annotation Tool18
In a simulation experiment performed on annotating large quantities of structured data, DART has been shown to reduce the total number of annotations needed with active learning and automatically suggesting relevant labels.
- 18
- StreamingThinker: Large Language Models Can Think While Reading17
Experimental results show that the StreamingThinker preserves performance comparable to batch thinking, while yielding an 80% reduction in token waiting before the onset of reasoning and a more than 60% reduction in time-level latency for producing the final answer, demonstrating the effectiveness of the streaming paradigm for LLM reasoning.
- 17
- Unsupervised Rewriter for Multi-Sentence Compression17
This work presents a neural rewriter for multi-sentence compression that achieves comparable results upon automatic evaluation and improves the grammaticality of compression based on human evaluation.
- SkipGPT: Dynamic Layer Pruning Reinvented with Token Awareness and Module Decoupling15
SkipGPT is introduced, a dynamic layer pruning framework designed to optimize computational resource allocation through two core innovations: (1) global token-aware routing to prioritize critical tokens, and (2) decoupled pruning policies for MLP and self-attention components.
- xPQA: Cross-Lingual Product Question Answering across 12 Languages14
It is found that in-domain data is essential as cross-lingual rankers trained on other domains perform poorly on the PQA task, and that translation-based approaches are most effective for candidate ranking while multilingual finetuning works best for answer generation.
- 14
- Product Answer Generation from Heterogeneous Sources: A New Benchmark and Best Practices14
A benchmark with annotations for both evidence selection and answer generation covering 6 information sources is built and a novel data augmentation method is proposed to iteratively create training samples for answer generation, which achieves close-to-human performance with only a few thousandannotations.
- Beyond Content Relevance: Evaluating Instruction Following in Retrieval Models13
This study evaluates the instruction-following capabilities of various retrieval models beyond content relevance, including LLM-based dense retrieval and reranking models and develops InfoSearch, a novel retrieval evaluation benchmark spanning six document-level attributes, and introduces novel metrics.
- Fine-Grained and Multi-Dimensional Metrics for Document-Level Machine Translation12
The results show that this method improves translation quality compared to translating sentences separately, even without document-level fine-tuning, and demonstrates that instruction-tuned LLMs can effectively leverage document context for translation.
- 12
- 12
- Gated Neural Network for Sentence Compression Using Linguistic Knowledge11
Experimental results on two popular datasets show that the proposed gated neural network equipped with selectively fused linguistic features leads to better compressions upon both automatic metric and human evaluation, compared with a previous competitive compression system.
- A Comprehensive Evaluation of Parameter-Efficient Fine-Tuning on Software Engineering Tasks10
Evaluating the effectiveness of five PEFT methods on eight PTMs and four SE downstream tasks and the efficiency of PEFT methods, concerning the costs of required training time and GPU resource consumption is discussed.
- semiPQA: A Study on Product Question Answering over Semi-structured Data10
This work presents semiPQA: a dataset to benchmark PQA over semi-structured data, which contains 11,243 written questions about json-formatted data covering 320 unique attribute types and provides baseline results and a deep analysis on the successes and challenges of leveraging semi- Structured data for PZA.
- The SelectGen Challenge: Finding the Best Training Samples for Few-Shot Neural Text Generation9
A shared task on training instance selection for few-shot neural text generation to help make the most use of the annotation budget in downstream tasks and better benchmark few- shot text generative models.
- UTPTrack: Towards Simple and Unified Token Pruning for Visual Tracking8
Extensive evaluations demonstrate that UTPTrack achieves a new state-of-the-art in the accuracy-efficiency trade-off for pruning-based trackers, pruning 65.4% of vision tokens in RGB-based tracking and 67.5% in unified tracking while preserving 99.7% and 100.5% of baseline performance, respectively.
- Preventing Author Profiling through Zero-Shot Multilingual Back-Translation8
This paper proposes a simple, zero-shot way to effectively lower the risk of author profiling through multilingual back-translation using off-the-shelf translation models and achieves the best overall performance while requiring no training data.
- Experimental Evaluation of Parameter-Efficient Fine-Tuning for Software Engineering Tasks7
An empirical study on six PEFT methods, eight PTMs, and four SE tasks, which reveals several noteworthy findings and provides valuable insights to guide future researchers in effectively applying PEFT methods to SE tasks.
- 7
- FocusQA: Open-Domain Question Answering with a Context in Focus7
Strong baselines for answer sentence selection that outperform the precision of state-of-the-art models for AS2 up to 21.3% absolute points are introduced.
- 7
- A comprehensive study: Sentence compression with linguistic knowledge-enhanced gated neural network7
A gating mechanism is introduced and a gated neural network that selectively exploits linguistic knowledge for deletion-based sentence compression is proposed that leads to better compression upon both automatic metrics and human evaluation, compared to previous competitive compression methods.
- Neural Ranking with Weak Supervision for Open-Domain Question Answering : A Survey6
This work provides a structured overview of standard WS signals used for training a NR model and divides them into three main categories, based on their required resources, which are summarized in this work.
- 5
- LLaSO: A Foundational Framework for Reproducible Research in Large Language and Speech Model5
By releasing the complete stack of data, benchmarks, and models, LLaSO establishes a foundational open standard to unify research efforts and accelerate community-driven progress in LSLMs.
- 5
- 5
- Rethinking Soft Actor-Critic in High-Dimensional Action Spaces: The Cost of Ignoring Distribution Shift5
A novel action sampling method is presented that directly identifies and selects the most probable actions within the transformed distribution of the tanh transformation, thereby addressing this issue of hindering the selection of the most probable actions.
- 4
- Deep Latent-Variable Models for Text Generation4
This dissertation presents how deep latent-variable models can improve over the standard encoder-decoder model for text generation.
- 4
- 4
- Towards Better Variational Encoder-Decoders in Seq2Seq Tasks4
This paper thoroughly analyzes the reason behind this training difficulty, compares different ways of alleviating it and proposes a new framework that helps significantly improve the overall performance.
- 3
- 3
- 3
- xPQA: Cross-Lingual Product Question Answering in 12 Languages3
It is found that in-domain data is essential as cross-lingual rankers trained on other domains perform poorly on the PQA task, and that translation-based approaches are most effective for candidate ranking while multilingual finetuning works best for answer generation.
- 3
- 2
- 2
- Wake-Sleep Variational Autoencoders for Language Modeling2
The training pattern is similar to the wake-sleep algorithm: these two steps are trained alternatively until an equilibrium is achieved and the results demonstrate significant improvement over the current state-of-the-art latent variable models.
- SkipOPU: An FPGA-based Overlay Processor for Large Language Models with Dynamically Allocated Computation1
This work proposes SkipOPU, an FPGA-based overlay processor that dynamically allocates computation across tokens and layers with high flexibility through a lightweight routing mechanism, and designs a processing element (PE) array that efficiently supports float–fixed hybrid execution.
- 1
- 1
- Is Translation Helpful? An Empirical Analysis of Cross-Lingual Transfer in Low-Resource Dialog Generation1
Experiments show that leveraging English dialog corpora can indeed improve the naturalness, relevance and cross-domain transferability in Chinese, and suggest future research should rather focus on utilizing the original English data for cross-lingual transfer in dialog generation.
- 1
- 1
- An Application of Extreme Programming in Development of Management Information System1
Through the introduction of the characteristics and application on real estate enterprise management information system, the advantage of XP which being applied on requirements variable small and medium-sized information management software development project is discussed.
- AutoDataset: A Lightweight System for Continuous Dataset Discovery and Search–
AutoDataset is introduced, a lightweight, automated system for real-time dataset discovery and retrieval that adopts a paper-first approach by continuously monitoring arXiv to detect and index datasets directly from newly published research.
- –
- Learning A Bank of Transferable Prompts for Vision-Language Models–
A prompt transfer strategy that reconstructs class-specific prompts for novel classes, based on their semantic similarity with base classes, is introduced, which consistently improves both adaptation and generalization, achieving state-of-the-art performance.
- E^2-LLM: Bridging Neural Signals and Interpretable Affective Analysis–
This work establishes a new paradigm combining physiological signals with LLM reasoning capabilities, showing that model scaling improves both recognition accuracy and interpretable emotional understanding in affective computing.
- –
- –
- MCM-DPO : Multifaceted Cross-Modal Direct Preference Optimization for Alt-text Generation–
Experimental results show that the proposed Multi-faceted Cross-modal Direct Preference Optimization (MCM-DPO) method consistently outperforms both DPO and SFT, establishing a new state of the-art in alt-text generation.
- –
- –
- –
- –
- –
- –
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.