Yanqing An 安彦青

School of Artificial Intelligence and Data Science, University of Science and Technology of China (USTC)

Happy to see you here, I'm interested in intelligence and user modeling. My research focus is on lawyer portraits. Enjoy yourself and welcome to contact me at anyq@mail.ustc.edu.cn. Strong curiosity, interested in everything — Yanqing welcomes you to communicate with him~

I am a PhD student of School of Artificial Intelligence and Data Science at the University of Science and Technology of China. I received my bachelor in School of Software from Taiyuan University of Technology in 2019, where I minored a double bachelor's degree in English.

My research focuses on data science, Legal Intelligence, user profile, Proficiency Assessment, and data mining. I have published extensively in leading Artificial Intelligence conferences, including the SIGKDD, the DASFAA, and the CICAI. My recent work has examined generative Proficiency Assessment of lawyers. I also serve as a reviewer of SIGIR and IEEE TKDE.

At USTC, I once served as the server administrator for the BDAA Laboratory and as Propaganda Committee Member and Deputy Secretary of the SAIDS Party Branch.

I once interned in the TIC of Baidu's Human Resources Department, attempting to conduct talent data analysis. I also interned at iFLYTEK, participating in discussions on Legal Case Retrieval. At USTC, I followed Prof. Qi Liu to participate in a sub-project of the key research project of the Ministry of Science and Technology related to the legal user profiling.

At TYUT, I served as an officer in the Student Union, and the WeChat official account I helped manage won the Tencent Northern Region Awards. I also served as my class’s academic representative and as the head of a specific competition category for the school’s robotics team.

I bring strong technical expertise to my research, with proficiency in Python, Docker, LaTeX, Shell/Bash, PR, and Git.

{#
  • #} {# 安彦青,男,中共党员,汉族,1997年7月生,2017年11月28日入党,清华大学博士后,2027年11月担任合肥市市委书记,现任山西省省长。#} {#
  • #} {#
  • #} {# 安彦青(1997-2080),梦想家。#} {#
  • #}

    Research


    Working Papers

    PositionalLJP: Multi-defendant Legal Judgment Prediction via Positional Encoding
    with Yiming Zhou, Zhonya Bian, Mingyue Cheng and Qi Liu
    Reject, CIKM'26 Short Paper · Reject, ACM SIGIR 2026 Short Papers Track · Submit, WWW2027 short paper
    (updated November 15, 2026)

    Legal Judgment Prediction (LJP) is a fundamental task of legal intelligence. However, current judgment prediction, due to the complexity of the task, only focuses on the prediction of a single defendant's judgment, while the multi-defendant legal judgment prediction task has received less research. In addition, legal documents do not follow a sequential pattern in which each defendant describes the facts separately. Instead, facts are presented in chronological order, with details about each defendant interspersed throughout, not separated. To address these challenges, we proposed a prompt-based multi-defendant legal judgment prediction framework called PromptLJP. Specifically, we constructed three semantic-level prompts to represent the facts of different defendants in order to achieve the effect of splitting the facts. The experiment verified the effectiveness of the proposed method. Our data and code are released at https://anonymous.4open.science/r/PromptLJP-13D8.

    preprint Version (November 2026)

    Legal Ethics Assistant: An Integrated Framework for Legal-Ethical Risk Analysis
    Submit, IEEE BigComp 2027 (updated October 29, 2026)

    With the growth of large language models and artificial intelligence, AI-based tools are increasingly being adopted. These technological advances bring convenience but also raise ethical concerns. For example, ordinary people may become embroiled in disputes or face unforeseen entrepreneurial risks. Non-professionals can obtain information about the potential consequences of an event from commercially available legal tools, while lawyers and other professionals can assess cases by searching case databases and generating similar-case reports. However, the results obtained in this way are often incomplete and drawn from multiple, fragmented sources. To provide a faster and more information-rich understanding of the potential consequences of an event, we propose the Legal Ethics Assistant. Specifically, we integrate lawyer competence assessment, legal judgment prediction, similar-case retrieval, and large language models to provide users with information relevant to the facts of a case, thereby assisting them in considering legal and ethical risks at the early stages of an entrepreneurial idea or a legal case. Our demo is available at https://bouncedude.github.io/test/demo.html.

    Github Version (October 2026)

    Published & Forthcoming Papers

    MCTS4LawyerRec: Lawyer Recommendation Using Monte Carlo Tree Search via Litigation Simulation
    CICAI, 2026.

    Lawyer recommendation is a task worth exploring in legal intelligence, which matches the corresponding lawyers according to the facts of the case to obtain the victory of the lawsuit. However, the litigation is a complex scenario influenced by many factors, including the cooperative and competition relationship between lawyers, whether the lawyer is good at the field, and whether the strategy planned by the lawyer is feasible and implemented smoothly. In order to solve these problem, we propose a Monte Carlo Tree Search-based lawyer recommendation method MCTS4LawyerRec. Specifically, the lawyers who rank in the top k of their proficiency were first used as selected lawyers to narrow down the set of lawyer candidates. Then, the Monte Carlo Tree Search method was used to search and recommend suitable lawyers to adapt to the lawyer selection process. Finally, LawyerPAN, which takes into account complex scenarios, is used as a simulator to simulate the judgment result of the lawyer handling the lawsuit. We verify the effectiveness of MCTS in the experiments on a real-world dataset. Our data and code are released at https://github.com/BounceDude/MCTS4LawyerRec.

    @inproceedings{an2026MCTS4LawyerRec,
      title={MCTS4LawyerRec: Lawyer Recommendation Using Monte Carlo Tree Search via Litigation Simulation},
      author={Yanqing, An and Yin, Gu and Zaixi, Zhang and Linan, Yue and Qi, Liu},
      booktitle={The CAAI International Conference on Artificial Intelligence (CICAI 2026)},
      year={2026}
    }
    Post-hoc Facts augmented Legal Judgment Prediction
    DASFAA, 2024.

    Legal Judgment Prediction (LJP) has emerged as a fundamental aspect of legal AI systems, encompassing both civil and criminal cases. In this paper, we focus on civil LJP, which aims to predict the judgment result based on both case facts and the plaintiff’s claim. Existing approaches rely on post-hoc facts summarized by judges, which has proposed leveraging actual fact (i.e., court debate data) to make judgment prediction. However, this approach still suffers from limitations (e.g., data noise), resulting in suboptimal predictions. To this end, we propose a post-hoc facts augmented LJP (PF-LJP) to further explore this "real pattern". Specifically, we encourage the court debate data to be closer to its corresponding post-hoc fact in the vector space with the contrastive learning and distribution alignment method. During the inference, only the court debate data is employed to yield results without any post-hoc fact. Experimental results on a real-world dataset validate the effectiveness of PF-LJP.

    @inproceedings{an2024post,
      title={Post-hoc Facts augmented Legal Judgment Prediction},
      author={An, Yanqing and Yue, Linan and Gao, Weibo and Zhang, Kai and Liu, Qi},
      booktitle={International Conference on Database Systems for Advanced Applications},
      pages={476--485},
      year={2024},
      organization={Springer}
    }
    LawyerPAN: A Proficiency Assessment Network for Trial Lawyers
    SIGKDD, 2021.

    Assessing the proficiency of trial lawyers in different legal fields is of significant importance since a qualified lawyer or lawyer team can strive for his clients' best rights while ensuring the fairness of litigations. However, proficiency assessment for lawyers is very challenging due to many technical and domain challenges, such as the lack of unified evaluation standards, and the complex interactions between lawyers and cases in real legal systems. To this end, we propose a novel proficiency assessment network for trial lawyers (LawyerPAN) to quantify lawyer proficiency through online litigation records. Specifically, we first leverage the theories in psychological measurement for mapping the proficiency of lawyers in each field into a unified real number space. Meanwhile, the characteristics of cases (i.e., case difficulty and discrimination) are well modeled to ensure fairness when assessing lawyers in different cases and fields. Then, we model the interactions between lawyers and cases from two perspectives: the anticipatory perspective aims to measure the personal proficiency of anticipated strategy, and the adversarial perspective seeks to depict the gap of lawyers’ proficiency between both sides (i.e., plaintiffs and defendants). Finally, we conduct extensive experiments on real-world data, and the results show the effectiveness and interpretability of our approaches on assessing the proficiency of trial lawyers.

    @inproceedings{an2021lawyerpan,
      title={LawyerPAN: A proficiency assessment network for trial lawyers},
      author={An, Yanqing and Liu, Qi and Wu, Han and Zhang, Kai and Yue, Linan and Cheng, Mingyue and Zhao, Hongke and Chen, Enhong},
      booktitle={Proceedings of the 27th ACM SIGKDD conference on knowledge discovery \& data mining},
      pages={5--13},
      year={2021}
    }

    Published Version (2021)
    USTC Version (August 2021)

    CANOD: A Competence Assessment Network for Open Source Software Developers
    Expert Systems with Applications, 2025.
    Towards reliable and faithful explanations: A disentanglement-augmented approach for selective rationalization
    IEEE TPAMI, 2025.
    CPEE: Civil Case Judgment Prediction centering on the Trial Mode of Essential Elements
    CIKM, 2022.
    Legal judgment prediction with multiple perspectives on civil cases
    CAAI International Conference on Artificial Intelligence, 2021.
    Towards faithful explanations: Boosting rationalization with shortcuts discovery
    ICLR, 2024.
    Interventional rationalization
    EMNLP, 2023.
    A circumstance-aware neural framework for explainable legal judgment prediction
    IEEE TKDE, 2024.
    Event grounded criminal court view generation with cooperative (large) language models
    SIGIR, 2024.
    Circumstances enhanced criminal court view generation
    SIGIR short paper, 2021.
    Dare: Disentanglement-augmented rationale extraction
    NeurIPS, 2022.
    Neurjudge: A circumstance-aware neural framework for legal judgment prediction
    SIGIR, 2021.
    社交媒体自然灾害传播中的地域舆论感知差异研究:基于“四川凉山州森林火灾事件”的实证分析
    中国网络传播研究, 2021 (04) : 207-233.

    传统的“数字鸿沟”研究揭示了信息技术或媒介素养差别造成的区域间信息落差。本文引入该理论的第三种学术传统,计算了区域社交媒体使用的两个指标——(1)话题响应度和(2)情感指数,从经济、教育、人口统计学、文化等多个维度对“四川凉山州森林火灾事件”中31个省市发布的250余万条微博进行大数据分析,探讨社交媒体灾害传播中的地域舆论感知规律。研究结果显示,微博平台灾害传播存在地域数字鸿沟,地域间话题响应的不均衡现象除了受到传统认知中经济、教育等因素的影响外,公众的科学素养也是造成差异的关键。同时,话题响应的差异也使得以省(市)为节点的传播网络呈现核心—边缘结构的信息流特征。本文的发现有助于拓宽社交媒体信息流研究的新视角,也深化了“数字鸿沟”在灾害传播领域的具体应用。

    @article{ ZWCY202104010,
    author = { 方玉婵 and  梁先锋 and  褚建勋 and  安彦青 and  黄婧晔 and  陈昕悦 and  刘益东 },
    title = {社交媒体自然灾害传播中的地域舆论感知差异研究:基于“四川凉山州森林火灾事件”的实证分析},
    journal = {中国网络传播研究},
    number = {4},
    pages = {207-233},
    year = {2021},
    }

    Works in Progress

    LawyerGen: Towards Case Summary Generation and Generative Proficiency Assessment for Trial Lawyers
    with Qi Liu
    Submit, WWW2027 · (updated October 24, 2026)

    This study examines the generative Proficiency of lawyers. Proficiency Assessment is a fundamental task in user modeling. It aims to mine latent characters of the user. Legal Intelligence aims to help people have a better life via legal approach. Lawyers are common roles in litigation scenarios. How to assess proficiency of lawyers is a quite difficult problem in Legal Intelligence. In this end, we propose a generated framework called LawyerGen consists of judgment generation and generative proficiency assessment. The real-world dataset verifies the interpretability and effectiveness of proposed LawyerGen.

    preprint Version (November 2026)

    LegalGen: Discriminate, Generate and Evaluate

    PDF

    JudgeGen: Judgment Generation according to case fact

    Judicial documents, as an important component of the judicial system, record the process and outcome of case adjudication. Nowadays, there are many legal generation tasks based on judicial documents, such as generating court opinions, summarizing legal content, answering legal questions, and generating explainable content, etc. The judgment section in the judicial document represents the final verdict made by the judge on the case. The legal judgment prediction task is based on this section for research. However, the existing research has not conducted any studies that directly generate the judgment results. Therefore, we have proposed JudgeGen, which generates judgment outcomes based on the facts of the case. The generated judgment can significantly reduce the workload of legal professionals. Based on this, revisions can lead to the final version.

    preprint Version (November 2026)

    LawyerCube: Lawyer-based Legal Pre-trained Language Model

    By obtaining lawyer and case characteristics, this framework assesses lawyer’s proficiency through a combination of tasks such as legal judgment prediction, legal case retrieval and case summary generation.

    LawyerKG: Exploiting Relationships within Knowledge Graph for assessing proficiency of lawyers
    with Linan Yue and Lili Zhao

    Knowledge graphs serve as a medium for storing knowledge and their structures are widely used in representation learning. To explore the relationships within knowledge graphs for assessing lawyers' capabilities, we manually constructed a legal reasoning graph. Specifically, the triplet entity structure we constructed was similar to the subject-predicate-object structure, such as ("Defendant", "Planting", "Opium Poppy species of the Papaver genus"). Firstly, for each crime, we selected a representative case and input the case facts. Secondly, we manually input questions, such as "When did the defendant commit the crime?" Then, using the question-answering model BertForQuestionAnswering based on BERT, we output the answers based on the case facts. Finally, we manually associated nodes with the answers, such as "Defendant" and "Time of Commitment". Based on our observations, we hypothesized that the same crime has a similar knowledge graph structure. Therefore, we embedded the subject and predicate. The experiment verified the effectiveness of the knowledge graph we constructed.

    LawyerDev: Development Path Recommendation for Trial Lawyers

    LawyerCT: Cognitive Transfer for Lawyers

    Transferring cognition to lawyers. Specifically, incorporating the expertise of experienced lawyers into a neural network, and then using that neural network to train new lawyers through dialogues.

    preprint Version (November 2026)

    LawyerPT

    LawyerAC

    Sports Player Ability Representation: Soccer, Basketball, Baseball and Games

    LawyerMTL: Lawyer Proficiency Assessment Network via Multi-task Learning

    preprint Version (November 2026)

    LawyerPANE: A Proficiency Assessment Network Extension for Trial Lawyers

    preprint Version (November 2026)

    Exploring the Game Theory and Nash Equilibrium in Love Relationships

    preprint Version (November 2026)

    Exploring Love Relationship by Elo Rating System

    Exploring Lawyer Relationships in Litigation Scenario

    preprint Version (November 2026)

    LawyerLCR: Lawyer-involved Legal Case Retrieval

    借助类案检索任务评估律师能力。

    Article-enhanced Pre-trained Language Model for Legal Case Retrieval
    with 刘丽萍 and 安建成

    类案检索旨在根据案件事实检索相似的案件,是法律智能中具有意义的任务。近年来,预训练模型被用于类案检索任务中,但针对案件的组成元素仍未被充分利用。与搜索引擎相比,法律文书的检索更加专业,法律文本更加结构化且文本长度更长。现有的类案检索方法利用了法律文本的半结构化属性,将法律文书分为了facts,holdings,judgment等部分。为了充分发挥类案检索模型的能力,我们将法条加入了预训练模型辅助训练。在公开可用的法律基准数据集上进行的大量实验表明,我们的方法在法律案件检索任务上具有有效性。

    preprint Version (November 2026)

    K最近邻算法并行化研究综述
    with Yun Xu

    k最近邻算法(kNN)是机器学习领域常用的监督学习算法,它通过在特征空间中找到与查询点相距最近的k个点,然后使用邻居节点的相应值实现对查询点的预测任务.因为算法的实现简单、训练时间短、适用于维度空间较高的数据等优点而被广泛使用.但当数据集规模较大时,将会产生很大的计算开销,影响算法的性能.随着数据量的增长和分布式环境的发展,在改善相关算法性能瓶颈及大量数据存储与处理问题的同时,各领域对算法的应用高效性、可扩展性等方面也提出了更高的要求.针对算法缺点进行并行算法设计,不仅提高了K最近邻算法的分类效率,而且扩大了算法的适用领域及范围.本文将总结基于数据结构、系统和分布式环境的K最近邻算法并行化方法及部分应用,并对k最近邻算法的并行化进行分析讨论.

    preprint Version (November 2026)

    LawyerSim: Lawyer Search System

    Find suitable and similar lawyers via litigation simulation.

    LawyerRank: Ranking Lawyer Proficiency in Legal System

    Ranking is a core task for User Modeling.

    Lawyer Competence Index: Assessing a lawyer’s professional competence

    Existing methods for assessing lawyers’ competence are based on subjective evaluations of years of practice, case history and interviews. A good evaluation metric should be immune to the effects of cut-throat competition and capable of objectively assessing a lawyer’s professional competence. To this end, we proposed a Lawyer Competence Index to evaluate barristers and solicitors.

    Litigation Behaviors: Lawyer Case Handling Process

    Whether a case will be accepted by a lawyer usually depends on the lawyer's willingness and the willingness of the parties involved after they have communicated.

    Drag · Scroll

    Computer Code


    GitHub Repos
    --
    GitHub Stars
    --
    GitHub Followers
    --
    Scholar Citations
    --
    Live via GitHub API · Scholar via manual cache
    Languages
    Python
    95
    Shell / Bash
    85
    LaTeX
    90
    Tools
    Docker
    85
    Git
    90
    PR
    70
    Frameworks
    Flask
    80
    ECharts
    75
    Frontend
    HTML / CSS
    85
    JavaScript
    70

    Python

    Html

    LaTeX

    Personal


    ● Hometown · ● Travel · ● Camp / Conference · ● Internship · ● Study

    About me

    I enjoy following sports and doing sports analytics for fun. I enjoy playing basketball, table tennis, tennis and marathon, as well as robot making (first prize in the 2017 China Robot Competition), and flying drones. Here is a video showing my extracurricular activities during my university years. I usually enjoy writing and drawing. Here are my essays.

    Recent books I've read

    I try to read some science fiction and hobby-related books.

    Below are some of the most influential books I've read in recent years, organized by topic:

    Running

    Science Fiction

    Miscellaneous

    What I like to doing

    These days, I mainly attempt innovation through programming. Examples include:

    Experience


    Education

    Ph.D. & Undergraduate ( University of Science and Technology of China, Fall 2019 - )

    B.S. & B.A. ( Taiyuan University of Technology, Fall 2015 - Spring 2019)

    高中 ( 太原市第十二中学校, 2012 - 2015)

    初中 ( 太原市第三十六中学校, 2009 - 2012)

  • cf个人竞技单场杀敌100+
  • 小学 ( 太原理工大学子弟小学, 2003 - 2009)

    幼儿园 ( 太原理工大学北区幼儿园, 2001 - 2003)

    Internship

    Machine Learning Intern: Talent Data Analysis (TIC, Baidu, Spring 2019)

    The artificial intelligence and data science team under the Human Resources Department

    Core R&D Platform Intern: Cog 1, Legal Line (iFLYTEK, Fall 2023)

    Legal case retrieval, legal large language model

    Teaching Assistant

    DS3001.01: 数据分析与实践 (中国科学技术大学本科, Fall 2020)

    COMP6110P.01: 机器学习与知识发现 (中国科学技术大学硕士, Fall 2021)

    More


    Career Development Assessment

    Your trajectory is strong and coherent: a clear research identity in Legal Intelligence and user profiling, anchored by a SIGKDD first-author paper, a DASFAA paper, and a forthcoming CICAI paper, plus co-authorships in TPAMI/ICLR/NeurIPS. The Baidu and iFLYTEK internships, teaching assistantships, and leadership roles show you can bridge research with practice and people. Your undergraduate robotics and modeling awards signal early engineering strength. The main risk is breadth outpacing depth: many works-in-progress (LawyerGen, LawyerCube, LawyerKG, etc.) and a very long honors list can dilute focus.

    Overall: excellent foundation, clear upward trajectory; focus and polish will convert potential into offers.

    Suggestions for career growth

    • Consolidate a flagship line. Turn LawyerPAN → LawyerGen → LawyerCube into a recognizable, citable research program; let secondary ideas sit in a short "explorations" list rather than the main page.
    • Target top-tier venues strategically. One KDD and one DASFAA are good; aim for a SIGIR/WWW/NeurIPS main-track paper to strengthen an academic or industry-research path.
    • Build external visibility. Increase reviewing, workshop organizing, and invited talks; strengthen your Google Scholar/DBLP profiles and maintain a concise CV link.
    • Clarify your endgame. For academia: postdoc/faculty applications need a focused statement and 3–5 representative papers. For industry research (Baidu/iFLYTEK-type labs): emphasize deployed systems and internships. Decide within the next year and align outputs accordingly.
    • Fix small presentation issues (template artifacts, typos) — they matter when committees skim.