Di Weng

Qiming Shi, Zhaolu Kang, Yunfan Zhou et al.

Large language models are increasingly deployed as tool-augmented agents to acquire information beyond parametric knowledge. While recent work has improved long-horizon tool-use reasoning, most approaches focus on tasks with a single correct answer. In contrast, many real-world queries require discovering a comprehensive set of valid answers, a setting known as Multi-Answer QA. This setting raises two challenges: fine-grained credit assignment over long search trajectories and reward alignment for sustained exploration beyond easy high-frequency entities. We propose SPADER, a reinforcement learning framework for long-horizon tool use in Multi-Answer QA. SPADER includes Step-wise Peer Advantage (SPA), a critic-free step-level credit assignment mechanism that aligns parallel trajectories by decision step and estimates advantages from peer returns. It also includes a diversity-aware exploration reward that promotes long-tail entity discovery by upweighting rare findings and downweighting redundant ones. Experiments on QAMPARI, Mintaka, WebQSP, and QUEST show that SPADER generally improves recall and overall F1 over prompting-based agents, outcome-supervised RL methods, and recent step-level supervision approaches. Our code and model weights are available at https://github.com/KhanCold/spader.

Zhenning Chen, Hanbei Zhan, Yanwei Huang et al.

Large Language Models (LLMs) demonstrate exceptional capabilities in factual question answering, yet they sometimes provide incorrect responses. To address this issue, knowledge editing techniques have emerged as effective methods for correcting factual information in LLMs. However, typical knowledge editing workflows struggle with identifying the optimal set of model layers for editing and rely on summary indicators that provide insufficient guidance. This lack of transparency hinders effective comparison and identification of optimal editing strategies. In this paper, we present KEditVis, a novel visual analytics system designed to assist users in gaining a deeper understanding of knowledge editing through interactive visualizations, improving editing outcomes, and discovering valuable insights for the future development of knowledge editing algorithms. With KEditVis, users can select appropriate layers as the editing target, explore the reasons behind ineffective edits, and perform more targeted and effective edits. Our evaluation, including usage scenarios, expert interviews, and a user study, validates the effectiveness and usability of the system.

Xuming He, Zehao Fan, Hengjia Li et al.

Recent advances in video generation have enabled the synthesis of videos with strong temporal consistency and impressive visual quality, marking a crucial step toward vision foundation models. To evaluate these video generation models, existing benchmarks primarily focus on factors related to visual perception and understanding, like visual aesthetics, instruction adherence, and temporal coherence. However, the rule-based reasoning capabilities of video generation models remain largely unexplored. Although recent studies have carried out preliminary explorations into whether video models can serve as zero-shot learners, they still lack a fine-grained decomposition of reasoning capabilities and a comprehensive evaluation protocol. To address this gap, we introduce RULER-Bench, a benchmark designed to evaluate the reasoning ability of video generation models from the perspective of cognitive rules. Built upon two fundamental paradigms: text-to-video and image-to-video, RULER-Bench covers 40 representative tasks spanning six rule categories with 622 high-quality annotated instances. For the evaluation of each generated video, we construct a checklist covering four metrics and leverage GPT-o3 to assign scores to each question, achieving 85% alignment with human judgements. Extensive experiments show that the state-of-the-art model achieves only 48.87% on the rule coherence metric, highlighting significant room for improvement in the reasoning capability of next-level video models. We expect that the insight obtained from RULER-Bench will facilitate further development of reasoning-aware video generation, advancing video generation models toward vision foundation intelligence.

Di Weng, Chengbo Zheng, Zikun Deng et al.

Bus routes are typically updated every 3-5 years to meet constantly changing travel demands. However, identifying deficient bus routes and finding their optimal replacements remain challenging due to the difficulties in analyzing a complex bus network and the large solution space comprising alternative routes. Most of the automated approaches cannot produce satisfactory results in real-world settings without laborious inspection and evaluation of the candidates. The limitations observed in these approaches motivate us to collaborate with domain experts and propose a visual analytics solution for the performance analysis and incremental planning of bus routes based on an existing bus network. Developing such a solution involves three major challenges, namely, a) the in-depth analysis of complex bus route networks, b) the interactive generation of improved route candidates, and c) the effective evaluation of alternative bus routes. For challenge a, we employ an overview-to-detail approach by dividing the analysis of a complex bus network into three levels to facilitate the efficient identification of deficient routes. For challenge b, we improve a route generation model and interpret the performance of the generation with tailored visualizations. For challenge c, we incorporate a conflict resolution strategy in the progressive decision-making process to assist users in evaluating the alternative routes and finding the most optimal one. The proposed system is evaluated with two usage scenarios based on real-world data and received positive feedback from the experts.

Di Weng, Jichang Zhao

Negative reviews, the poor ratings in postpurchase evaluation, play an indispensable role in e-commerce, especially in shaping future sales and firm equities. However, extant studies seldom examine their potential value for sellers and producers in enhancing capabilities of providing better services and products. For those who exploited the helpfulness of reviews in the view of e-commerce keepers, the ranking approaches were developed for customers instead. To fill this gap, in terms of combining description texts and emotion polarities, the aim of the ranking method in this study is to provide the most helpful negative reviews under a certain product attribute for online sellers and producers. By applying a more reasonable evaluating procedure, experts with related backgrounds are hired to vote for the ranking approaches. Our ranking method turns out to be more reliable for ranking negative reviews for sellers and producers, demonstrating a better performance than the baselines like BM25 with a result of 8% higher. In this paper, we also enrich the previous understandings of emotions in valuing reviews. Specifically, it is surprisingly found that positive emotions are more helpful rather than negative emotions in ranking negative reviews. The unexpected strengthening from positive emotions in ranking suggests that less polarized reviews on negative experience in fact offer more rational feedbacks and thus more helpfulness to the sellers and producers. The presented ranking method could provide e-commerce practitioners with an efficient and effective way to leverage negative reviews from online consumers.