arXiv subjects

Zexuan Li

Publications and source records attributed to Zexuan Li.

7 recordsLinked to original sources

arXiv2026

AnnotateThis: Analyzing a human-LLM system for annotating social media data with the concept of climate change mitigation pessimism

Large language models (LLMs) are increasingly being integrated into research workflows. However, LLMs have been shown to struggle with difficult and nuanced concepts such as those found in computational social science (CSS) research. Within the CSS community, there has been a call for new systems to be developed which center humans in LLM-supported scientific workflows. We develop AnnotateThis, a human-centered system for inspecting and improving LLM annotations, a process we refer to as LLM grounding for a target concept. AnnotateThis is developed with both computational and social scientists to reflect existing workflows for data annotation. It includes a range of information features for users to interrogate the quality and reliability of LLM annotations. We evaluate our system in two settings. In the first, we assume a researcher may not have access to ground truth data and that users of AnnotateThis have limited prior knowledge of the concept they would like an LLM to annotate. That is, they may be conducting concept specification and LLM grounding simultaneously. In the second setting, we assume access to ground truth labels and that the concept is specified for a given annotation task; here, the task of LLM grounding is more straightforward. We find that in both settings users can improve the quality of LLM annotations with AnnotateThis and that their final annotations far surpass those created without human intervention. For example, when we evaluate with ground truth labels, we see an absolute improvement of 0.15 in F-Measure and 0.23 in accuracy over a fully automated state-of-the-art method for prompt refinement.

cs.CY↗

arXiv2026

Are eHMIs always helpful? Investigating how eHMIs interfere with pedestrian behavior on multi-lane streets: An eye-tracking virtual reality experiment

Appropriate communication is crucial for efficient and safe interactions between pedestrians and autonomous vehicles (AVs). External human-machine interfaces (eHMIs) on AVs, which can be categorized as allocentric or egocentric, are considered a promising solution. While the effectiveness of eHMIs has been extensively studied, in complex environments, such as unsignalized multi-lane streets, their potential to interfere with pedestrian crossing behavior remains underexplored. Hence, a virtual reality-based experiment was conducted to examine how different types of eHMIs displayed on AVs affect the crossing behavior of pedestrians in multi-lane streets environments, with a focus on the gaze patterns of pedestrians during crossing. The results revealed that the presence of eHMIs significantly influenced the cognitive load on pedestrians and increased the possibility of distraction, even misleading pedestrians in cases involving multiple AVs on multi-lane streets. Notably, allocentric eHMIs induced higher cognitive loads and greater distraction in pedestrians than egocentric eHMIs. This was primarily evidenced by longer gaze time and higher proportions of attention for the eHMI on the interacting vehicle, as well as a broader distribution of gaze toward vehicles in the non-interacting lane. However, misleading behavior was mainly triggered by eHMI signals from yielding vehicles in the non-interacting lane. Under such asymmetric signal configurations, egocentric eHMIs resulted in a higher misjudgment rate than allocentric eHMIs. These findings highlight the importance of enhancing eHMI designs to balance the clarity and consistency of the displayed information across different perspectives, especially in complex multi-lane traffic scenarios. This study provides valuable insights regarding the application and standardization of future eHMI systems for AVs.

cs.HC↗

arXiv2025

M-BRe: Discovering Training Samples for Relation Extraction from Unlabeled Texts with Large Language Models

For Relation Extraction (RE), the manual annotation of training data may be prohibitively expensive, since the sentences that contain the target relations in texts can be very scarce and difficult to find. It is therefore beneficial to develop an efficient method that can automatically extract training instances from unlabeled texts for training RE models. Recently, large language models (LLMs) have been adopted in various natural language processing tasks, with RE also benefiting from their advances. However, when leveraging LLMs for RE with predefined relation categories, two key challenges arise. First, in a multi-class classification setting, LLMs often struggle to comprehensively capture the semantics of every relation, leading to suboptimal results. Second, although employing binary classification for each relation individually can mitigate this issue, it introduces significant computational overhead, resulting in impractical time complexity for real-world applications. Therefore, this paper proposes a framework called M-BRe to extract training instances from unlabeled texts for RE. It utilizes three modules to combine the advantages of both of the above classification approaches: Relation Grouping, Relation Extraction, and Label Decision. Extensive experiments confirm its superior capability in discovering high-quality training samples from unlabeled texts for RE.

cs.CL↗

arXiv2025

ARC-Hunyuan-Video-7B: Structured Video Comprehension of Real-World Shorts

Real-world user-generated short videos, especially those distributed on platforms such as WeChat Channel and TikTok, dominate the mobile internet. However, current large multimodal models lack essential temporally-structured, detailed, and in-depth video comprehension capabilities, which are the cornerstone of effective video search and recommendation, as well as emerging video applications. Understanding real-world shorts is actually challenging due to their complex visual elements, high information density in both visuals and audio, and fast pacing that focuses on emotional expression and viewpoint delivery. This requires advanced reasoning to effectively integrate multimodal information, including visual, audio, and text. In this work, we introduce ARC-Hunyuan-Video, a multimodal model that processes visual, audio, and textual signals from raw video inputs end-to-end for structured comprehension. The model is capable of multi-granularity timestamped video captioning and summarization, open-ended video question answering, temporal video grounding, and video reasoning. Leveraging high-quality data from an automated annotation pipeline, our compact 7B-parameter model is trained through a comprehensive regimen: pre-training, instruction fine-tuning, cold start, reinforcement learning (RL) post-training, and final instruction fine-tuning. Quantitative evaluations on our introduced benchmark ShortVid-Bench and qualitative comparisons demonstrate its strong performance in real-world video comprehension, and it supports zero-shot or fine-tuning with a few samples for diverse downstream applications. The real-world production deployment of our model has yielded tangible and measurable improvements in user engagement and satisfaction, a success supported by its remarkable efficiency, with stress tests indicating an inference time of just 10 seconds for a one-minute video on H20 GPU.

cs.CV↗

arXiv2025

Generating Diverse Training Samples for Relation Extraction with Large Language Models

Using Large Language Models (LLMs) to generate training data can potentially be a preferable way to improve zero or few-shot NLP tasks. However, many problems remain to be investigated for this direction. For the task of Relation Extraction (RE), we find that samples generated by directly prompting LLMs may easily have high structural similarities with each other. They tend to use a limited variety of phrasing while expressing the relation between a pair of entities. Therefore, in this paper, we study how to effectively improve the diversity of the training samples generated with LLMs for RE, while also maintaining their correctness. We first try to make the LLMs produce dissimilar samples by directly giving instructions in In-Context Learning (ICL) prompts. Then, we propose an approach to fine-tune LLMs for diversity training sample generation through Direct Preference Optimization (DPO). Our experiments on commonly used RE datasets show that both attempts can improve the quality of the generated training data. We also find that comparing with directly performing RE with an LLM, training a non-LLM RE model with its generated samples may lead to better performance.

cs.CL↗

arXiv2025

Divisi: Interactive Search and Visualization for Scalable Exploratory Subgroup Analysis

Analyzing data subgroups is a common data science task to build intuition about a dataset and identify areas to improve model performance. However, subgroup analysis is prohibitively difficult in datasets with many features, and existing tools limit unexpected discoveries by relying on user-defined or static subgroups. We propose exploratory subgroup analysis as a set of tasks in which practitioners discover, evaluate, and curate interesting subgroups to build understanding about datasets and models. To support these tasks we introduce Divisi, an interactive notebook-based tool underpinned by a fast approximate subgroup discovery algorithm. Divisi's interface allows data scientists to interactively re-rank and refine subgroups and to visualize their overlap and coverage in the novel Subgroup Map. Through a think-aloud study with 13 practitioners, we find that Divisi can help uncover surprising patterns in data features and their interactions, and that it encourages more thorough exploration of subtypes in complex data.

cs.HC↗

arXiv2023

A High-efficient Battery Charging System for Electric Vehicle

Nowadays, automobile is facing the trend of electrification. Lithium-ion batteries is widely used as their power supplies. Lithium-ion battery has complex characteristics, as a result, Lithium-ion battery needs optimal charging strategies to make sure it is charged safely and efficiently. This paper focuses on development of a high-efficient charging method for lithium-ion battery. To test different charging strategies, the electric vehicle charging system consisting of a dual active bridge DC-DC converter and a Thevenin battery model is implemented. Multistage constant current charging (MSCC) and multistage constant current reflex charging (MSCC with reflex charging) were proposed. Compared with the traditional constant voltage constant current (CC-CV) charging method, MSCC can reduce 12% of the charging time and 1.1% of the battery loss; MSCC with reflex charging has a 10.45% and a 1.54% reduction of charging time and battery loss separately.

eess.SY↗