Visual generative models have achieved remarkable progress in synthesizing photorealistic images and videos, yet aligning their outputs with human preferences across critical dimensions remains a persistent challenge. Though reinforcement learning from human feedback offers promise for preference alignment, existing reward models for visual generation face limitations, including black-box scoring without interpretability and potentially resultant unexpected biases. We present VisionReward, a general framework for learning human visual preferences in both image and video generation. Specifically, we employ a hierarchical visual assessment framework to capture fine-grained human preferences, and leverages linear weighting to enable interpretable preference learning. Furthermore, we propose a multi-dimensional consistent strategy when using VisionReward as a reward model during preference optimization for visual generation. Experiments show that VisionReward can significantly outperform existing image and video reward models on both machine metrics and human evaluation. Notably, VisionReward surpasses VideoScore by 17.2% in preference prediction accuracy, and text-to-video models with VisionReward achieve a 31.6% higher pairwise win rate compared to the same models using VideoScore.

Jiale Cheng, Yusen Liu, Xinyu Zhang, Yulin Fei, Wenyi Hong, Ruiliang Lyu, Weihan Wang, Zhe Su, Xiaotao Gu, Xiao Liu, Yushi Bai, Jie Tang, Hongning Wang, Minlie Huang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Andrew Zhuoer Feng, Cunxiang Wang, Yu Luo, Lin Fan, Irene Zhou, Zikang Wang, Xiaotao Gu, Jie Tang, Hongning Wang, Minlie Huang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Yida Lu, Jiale Cheng, Zhexin Zhang, Shiyao Cui, Cunxiang Wang, Xiaotao Gu, Yuxiao Dong, Jie Tang, Hongning Wang, Minlie Huang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

In recent years, vision language models (VLMs) have made significant advancements in video understanding. However, a crucial capability — fine-grained motion comprehension — remains under-explored in current benchmarks. To address this gap, we propose MotionBench, a comprehensive evaluation benchmark designed to assess the fine-grained motion comprehension of video understanding models. MotionBench evaluates models’ motion-level perception through six primary categories of motion-oriented question types and includes data collected from diverse sources, ensuring a broad representation of real-world video content. Experimental results reveal that existing VLMs perform poorly in understanding fine-grained motions. To enhance VLM’s ability to perceive fine-grained motion within a limited sequence length of LLM, we conduct extensive experiments reviewing VLM architectures optimized for video feature compression and propose a novel and efficient Through-Encoder (TE) Fusion method. Experiments show that higher frame rate inputs and TE Fusion yield improvements in motion understanding, yet there is still substantial room for enhancement. Our benchmark aims to guide and motivate the development of more capable video understanding models, emphasizing the importance of fine-grained motion comprehension. Project page: https://motion-bench.github.io.

Video generation models have achieved remarkable progress in text-to-video tasks. These models are typically trained on text-video pairs with highly detailed and carefully crafted descriptions, while real-world user inputs during inference are often concise, vague, or poorly structured. This gap makes prompt optimization crucial for generating high-quality videos. Current methods often rely on large language models (LLMs) to refine prompts through in-context learning, but suffer from several limitations: they may distort user intent, omit critical details, or introduce safety risks. Moreover, they optimize prompts without considering the impact on the final video quality, which can lead to suboptimal results. To address these issues, we introduce VPO, a principled framework that optimizes prompts based on three core principles: harmlessness, accuracy, and helpfulness. The generated prompts faithfully preserve user intents and, more importantly, enhance the safety and quality of generated videos. To achieve this, VPO employs a two-stage optimization approach. First, we construct and refine a supervised fine-tuning (SFT) dataset based on principles of safety and alignment. Second, we introduce both text-level and video-level feedback to further optimize the SFT model with preference learning. Our extensive experiments demonstrate that VPO significantly improves safety, alignment, and video quality compared to baseline methods. Moreover, VPO shows strong generalization across video generation models. Furthermore, we demonstrate that VPO could outperform and be combined with RLHF methods on video generation models, underscoring the effectiveness of VPO in aligning video generation models. Our code and data are publicly available at https://github.com/thu-coai/VPO.

Recent progress in multimodal large language models has markedly enhanced the understanding of short videos (typically under one minute), and several evaluation datasets have emerged accordingly. However, these advancements fall short of meeting the demands of real-world applications such as embodied intelligence for long-term decision-making, in-depth movie reviews and discussions, and live sports commentary, all of which require comprehension of long videos spanning several hours. To address this gap, we introduce LVBench, a benchmark specifically designed for long video understanding. Our dataset comprises publicly sourced videos and encompasses a diverse set of tasks aimed at long video comprehension and information extraction. LVBench is designed to challenge multimodal models to demonstrate long-term memory and extended comprehension capabilities. Our extensive evaluations reveal that current multimodal models still underperform on these demanding long video understanding tasks. Through LVBench, we aim to spur the development of more advanced models capable of tackling the complexities of long video comprehension. Our data and code are publicly available at: https://lvbench.github.io.

Instruction-following is a fundamental capability of language models, requiring the model to recognize even the most subtle requirements in the instructions and accurately reflect them in its output. Such an ability is well-suited for and often optimized by preference learning. However, existing methods often directly sample multiple independent responses from the model when creating preference pairs. Such practice can introduce content variations irrelevant to whether the instruction is precisely followed (e.g., different expressions about the same semantic), interfering with the goal of teaching models to recognize the key differences that lead to improved instruction following. In light of this, we introduce SPaR, a self-play framework integrating tree-search self-refinement to yield valid and comparable preference pairs free from distractions. By playing against itself, an LLM employs a tree-search strategy to refine its previous responses with respect to the instruction while minimizing unnecessary variations. Our experiments show that a LLaMA3-8B model, trained over three iterations guided by SPaR, surpasses GPT-4-Turbo on the IFEval benchmark without losing general capabilities. Furthermore, SPaR demonstrates promising scalability, greatly enhancing models like GLM-4-9B and LLaMA3-70B. We also identify how inference scaling in tree search would impact model performance. Our code and data are publicly available at https://github.com/thu-coai/SPaR.

Large Multimodal Models (LMMs) have ushered in a new era in artificial intelligence, merging capabilities in both language and vision to form highly capable \textbf{Visual Foundation Agents} that are postulated to excel across a myriad of tasks. However, existing benchmarks fail to sufficiently challenge or showcase the full potential of LMMs as visual foundation agents in complex, real-world environments. To address this gap, we introduce VisualAgentBench (VAB), a comprehensive and unified benchmark specifically designed to train and evaluate LMMs as visual foundation agents across diverse scenarios in one standard setting, including Embodied, Graphical User Interface, and Visual Design, with tasks formulated to probe the depth of LMMs' understanding and interaction capabilities. Through rigorous testing across 9 proprietary LMM APIs and 9 open models (18 in total), we demonstrate the considerable yet still developing visual agent capabilities of these models. Additionally, VAB explores the synthesizing of visual agent trajectory data through hybrid methods including Program-based Solvers, LMM Agent Bootstrapping, and Human Demonstrations, offering insights into obstacles, solutions, and trade-offs one may meet in developing open LMM agents. Our work not only aims to benchmark existing models but also provides an instrumental playground for future development into visual foundation agents. Code, train, and test data are available at \url{https://github.com/THUDM/VisualAgentBench}.

We present CogVideoX, a large-scale text-to-video generation model based on diffusion transformer, which can generate 10-second continuous videos that align seamlessly with text prompts, with a frame rate of 16 fps and resolution of 768 x 1360 pixels. Previous video generation models often struggled with limited motion and short durations. It is especially difficult to generate videos with coherent narratives based on text. We propose several designs to address these issues. First, we introduce a 3D Variational Autoencoder (VAE) to compress videos across spatial and temporal dimensions, enhancing both the compression rate and video fidelity. Second, to improve text-video alignment, we propose an expert transformer with expert adaptive LayerNorm to facilitate the deep fusion between the two modalities. Third, by employing progressive training and multi-resolution frame packing, CogVideoX excels at generating coherent, long-duration videos with diverse shapes and dynamic movements. In addition, we develop an effective pipeline that includes various pre-processing strategies for text and video data. Our innovative video captioning model significantly improves generation quality and semantic alignment. Results show that CogVideoX achieves state-of-the-art performance in both automated benchmarks and human evaluation. We publish the code and model checkpoints of CogVideoX along with our VAE model and video captioning model at https://github.com/THUDM/CogVideo.

Xiao Liu, Xuanyu Lei, Shengyuan Wang, Yue Huang, Andrew Feng, Bosi Wen, Jiale Cheng, Pei Ke, Yifan Xu, Weng Lam Tam, Xiaohan Zhang, Lichao Sun, Xiaotao Gu, Hongning Wang, Jing Zhang, Minlie Huang, Yuxiao Dong, Jie Tang. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Recently years have witnessed a rapid development of large language models (LLMs). Despite the strong ability in many language-understanding tasks, the heavy computational burden largely restricts the application of LLMs especially when one needs to deploy them onto edge devices. In this paper, we propose a quantization-aware low-rank adaptation (QA-LoRA) algorithm. The motivation lies in the imbalanced degrees of freedom of quantization and adaptation, and the solution is to use group-wise operators which increase the degree of freedom of quantization meanwhile decreasing that of adaptation. QA-LoRA is easily implemented with a few lines of code, and it equips the original LoRA with two-fold abilities: (i) during fine-tuning, the LLM's weights are quantized (e.g., into INT4) to reduce time and memory usage; (ii) after fine-tuning, the LLM and auxiliary weights are naturally integrated into a quantized model without loss of accuracy. We apply QA-LoRA to the LLaMA and LLaMA2 model families and validate its effectiveness in different fine-tuning datasets and downstream scenarios. The code is made available at https://github.com/yuhuixu1993/qa-lora.

Instruction following is one of the fundamental capabilities of large language models (LLMs). As the ability of LLMs is constantly improving, they have been increasingly applied to deal with complex human instructions in real-world scenarios. Therefore, how to evaluate the ability of complex instruction-following of LLMs has become a critical research problem. Existing benchmarks mainly focus on modeling different types of constraints in human instructions while neglecting the composition of different constraints, which is an indispensable constituent in complex instructions. To this end, we propose ComplexBench, a benchmark for comprehensively evaluating the ability of LLMs to follow complex instructions composed of multiple constraints. We propose a hierarchical taxonomy for complex instructions, including 4 constraint types, 19 constraint dimensions, and 4 composition types, and manually collect a high-quality dataset accordingly. To make the evaluation reliable, we augment LLM-based evaluators with rules to effectively verify whether generated texts can satisfy each constraint and composition. Furthermore, we obtain the final evaluation score based on the dependency structure determined by different composition types. ComplexBench identifies significant deficiencies in existing LLMs when dealing with complex instructions with multiple constraints composition.

Different knowledge graphs for the same domain are often uniquely housed on the Web. Effectively linking entities from different graphs is critical for building an open and comprehensive knowledge graph. However, linking entities across different sources has thus far faced various challenges, including the increasingly large-scale volume of the data, the heterogeneity of the graphs, and the ambiguity of real-world entities. To address them, we propose a unified framework LinKG. Specifically, we decouple the problem into different linking tasks based on the unique properties of each type of entity. To link word sequence based entities, we propose an LSTM-based method to capture word dependencies. To link entities of large scale, we utilize the hashing technique and convolutional neural networks for scalable and accurate linking. To link ambiguous entities, we propose heterogeneous graph attention networks to leverage heterogeneous structural information. Finally, to validate the design choices of different LinKG modules, we characterize the relationships between different tasks based on the single-domain and multi-domain transfer models. Extensive experiments demonstrate the effectiveness of LinKG with an overall F1-score of 95.15%, based on which we deploy and release the Open Academic Graph (OAG)—the largest publicly available heterogeneous academic graph to date.

Recent studies have achieved inspiring success in unsupervised grammar induction using masked language modeling (MLM) as the proxy task.Despite their high accuracy in identifying low-level structures, prior arts tend to struggle in capturing high-level structures like clauses, since the MLM task usually only requires information from local context.In this work, we revisit LM-based constituency parsing from a phrase-centered perspective.Inspired by the natural reading process of human readers, we propose to regularize the parser with phrases extracted by an unsupervised phrase tagger to help the LM model quickly manage low-level structures.For a better understanding of high-level structures, we propose a phrase-guided masking strategy for LM to emphasize more on reconstructing nonphrase words.We show that the initial phrase regularization serves as an effective bootstrap, and phrase-guided masking improves the identification of high-level structures.Experiments on the public benchmark with two different backbone models demonstrate the effectiveness and generality of our method.

Identifying and understanding quality phrases from context is a fundamental task in text mining. The most challenging part of this task arguably lies in uncommon, emerging, and domain-specific phrases. The infrequent nature of these phrases significantly hurts the performance of phrase mining methods that rely on sufficient phrase occurrences in the input corpus. Context-aware tagging models, though not restricted by frequency, heavily rely on domain experts for either massive sentence-level gold labels or handcrafted gazetteers. In this work, we propose UCPhrase, a novel unsupervised context-aware quality phrase tagger. Specifically, we induce high-quality phrase spans as silver labels from consistently co-occurring word sequences within each document. Compared with typical context-agnostic distant supervision based on existing knowledge bases (KBs), our silver labels root deeply in the input domain and context, thus having unique advantages in preserving contextual completeness and capturing emerging, out-of-KB phrases. Training a conventional neural tagger based on silver labels usually faces the risk of overfitting phrase surface names. Alternatively, we observe that the contextualized attention maps generated from a transformer-based neural language model effectively reveal the connections between words in a surface-agnostic way. Therefore, we pair such attention maps with the silver labels to train a lightweight span prediction model, which can be applied to new input to recognize (unseen) quality phrases regardless of their surface names or frequency. Thorough experiments on various tasks and datasets, including corpus-level phrase ranking, document-level keyphrase extraction, and sentence-level phrase tagging, demonstrate the superiority of our design over state-of-the-art pre-trained, unsupervised, and distantly supervised methods.

Xiaotao Gu, Liyuan Liu, Hongkun Yu, Jing Li, Chen Chen, Jiawei Han. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Walk-based models have shown their advantages in knowledge graph (KG) reasoning by achieving decent performance while providing interpretable decisions.However, the sparse reward signals offered by the KG during traversal are often insufficient to guide a sophisticated walk-based reinforcement learning (RL) model.An alternate approach is to use traditional symbolic methods (e.g., rule induction), which achieve good performance but can be hard to generalize due to the limitation of symbolic representation.In this paper, we propose RuleGuider, which leverages high-quality rules generated by symbolicbased methods to provide reward supervision for walk-based agents.Experiments on benchmark datasets show that RuleGuider improves the performance of walk-based models without losing interpretability. 1

The recent success of Distant Supervision (DS) brings abundant labeled data for the task of fine-grained entity typing (FET) without human annotation. However, the heuristically generated labels inevitably bring a significant distribution gap, namely dataset shift, between the distantly labeled training set and the manually curated test set. Considerable efforts have been made to alleviate this problem from the label perspective by either intelligently denoising the training labels, or designing noise-aware loss functions. Despite their progress, the dataset shift can hardly be eliminated completely. In this work, complementary to the label perspective, we reconsider this problem from the model perspective: Can we learn a more robust typing model with the existence of dataset shift? To this end, we propose a novel regularization module based on virtual adversarial training (VAT). The proposed approach first uses a self-paced sample selection function to select suitable samples for VAT, then constructs virtual adversarial perturbations based on the selected samples, and finally regularizes the model to be robust to such perturbations. Experiments on two benchmarks demonstrate the effectiveness of the proposed method, with an average 3.8%, 2.5%, and 3.2% improvement in accuracy, Macro F1 and Micro F1 respectively compared to the next best method.

Millions of news articles are published online every day, which can be overwhelming for readers to follow. Grouping articles that are reporting the same event into news stories is a common way of assisting readers in their news consumption. However, it remains a challenging research problem to efficiently and effectively generate a representative headline for each story. Automatic summarization of a document set has been studied for decades, while few studies have focused on generating representative headlines for a set of articles. Unlike summaries, which aim to capture most information with least redundancy, headlines aim to capture information jointly shared by the story articles in short length and exclude information specific to each individual article.

Xiyuan Yang, Xiaotao Gu, Sheng Lin, Siliang Tang, Yueting Zhuang, Fei Wu, Zhigang Chen, Guoping Hu, Xiang Ren. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

Linking entities from different sources is a fundamental task in building open knowledge graphs. Despite much research conducted in related fields, the challenges of linkinglarge-scale heterogeneous entity graphs are far from resolved. Employing two billion-scale academic entity graphs (Microsoft Academic Graph and AMiner) as sources for our study, we propose a unified framework --- LinKG --- to address the problem of building a large-scale linked entity graph. LinKG is coupled with three linking modules, each of which addresses one category of entities. To link word-sequence-based entities (e.g., venues), we present a long short-term memory network-based method for capturing the dependencies. To link large-scale entities (e.g., papers), we leverage locality-sensitive hashing and convolutional neural networks for scalable and precise linking. To link entities with ambiguity (e.g., authors), we propose heterogeneous graph attention networks to model different types of entities. Our extensive experiments and systematical analysis demonstrate that LinKG can achieve linking accuracy with an F1-score of 0.9510, significantly outperforming the state-of-the-art. LinKG has been deployed to Microsoft Academic Search and AMiner to integrate the two large graphs. We have published the linked results---the Open Academic Graph (OAG)\footnote\urlhttps://www.openacademic.ai/oag/ , making it the largest publicly available heterogeneous academic graph to date.

We present a novel end-to-end reinforcement learning approach to automatic taxonomy induction from a set of terms.While prior methods treat the problem as a two-phase task (i.e., detecting hypernymy pairs followed by organizing these pairs into a tree-structured hierarchy), we argue that such two-phase methods may suffer from error propagation, and cannot effectively optimize metrics that capture the holistic structure of a taxonomy.In our approach, the representations of term pairs are learned using multiple sources of information and used to determine which term to select and where to place it on the taxonomy via a policy network.All components are trained in an end-to-end manner with cumulative rewards, measured by a holistic tree metric over the training taxonomies.Experiments on two public datasets of different domains show that our approach outperforms prior state-ofthe-art taxonomy induction methods up to 19.6% on ancestor F1. 1

Many efforts have been made to facilitate natural language processing tasks with pre-trained language models (LMs), and brought significant improvements to various applications.To fully leverage the nearly unlimited corpora and capture linguistic information of multifarious levels, large-size LMs are required; but for a specific task, only parts of these information are useful.Such large-sized LMs, even in the inference stage, may cause heavy computation workloads, making them too time-consuming for large-scale applications.Here we propose to compress bulky LMs while preserving useful information with regard to a specific task.As different layers of the model keep different information, we develop a layer selection method for model pruning using sparsityinducing regularization.By introducing the dense connectivity, we can detach any layer without affecting others, and stretch shallow and wide LMs to be deep and narrow.In model training, LMs are learned with layerwise dropouts for better robustness.Experiments on two benchmark datasets demonstrate the effectiveness of our method.

Recent advances in deep neural models allow us to build reliable named entity recognition (NER) systems without handcrafting features.However, such methods require large amounts of manually-labeled training data.There have been efforts on replacing human annotations with distant supervision (in conjunction with external dictionaries), but the generated noisy labels pose significant challenges on learning effective neural models.Here we propose two neural models to suit noisy distant supervision from the dictionary.First, under the traditional sequence labeling framework, we propose a revised fuzzy CRF layer to handle tokens with multiple possible labels.After identifying the nature of noisy labels in distant supervision, we go beyond the traditional framework and propose a novel, more effective neural model AutoNER with a new Tie or Break scheme.In addition, we discuss how to refine distant supervision for better NER performance.Extensive experiments on three benchmark datasets demonstrate that AutoNER achieves the best performance when only using dictionaries with no additional human effort, and delivers competitive results with state-of-the-art supervised benchmarks.

The study of Web user profiling can be traced back to 30 years ago, with the goal of extracting “semantic”-based user profile attributes from the unstructured Web. Despite slight differences, the general method is to first identify relevant pages of a specific user and then use machine learning models (e.g., CRFs) to extract the profile attributes from the page. However, with the rapid growth of the Web volume, such a method suffers from data redundancy and error propagation between the two steps. In this paper, we revisit the problem of Web user profiling in the big data era, trying to deal with the new challenges. We propose a simple but very effective approach for extracting user profile attributes from the Web using big data. To avoid error propagation, the approach processes all the extraction subtasks in one unified model. To further incorporate human knowledge to improve the extraction performance, we propose a Markov logic factor graph (MagicFG) model. The MagicFG model describes human knowledge as first-order logics and combines the logics into the extraction model. Our experiments on a real data set show that the proposed method significantly improves (+4-6%; p ≪ 0.01, t-test) the extraction performance in comparison with several baseline methods.