<MY> ON THE WAY <LIFE>

Wei Peng (Albert)

Phone：010 612 16123

Birth：1996.02

English：CET-6

Other pages

Educational background

  • Institute of Information Engineering, Chinese Academy of Sciences | PhD
      Time：2018-2023
      Research：NLP, Dialogue system, Question Answering
  • Chang'an University | Undergraduate education (211) Rank First
      Time：2014-2018
      Major： Computer science and technology

Programming Language

PYTHON LINUX C++ PyTorch

Published

The conference and journal

  • Accepted 22 papers (SIGIR, IJCAI, EMNLP, AAAI, et al.), first/corresponding author 11 papers, SCI District 1, Chinese Academy of Sciences 2 papers，CCF-A 3 paper，CCF-B 5 papers，CCF-C 1 paper
  • |Paper Link| Wei Peng, Wanshui Li, Yue Hu et al. Leader-Generator Net: Dividing Skill and Implicitness for Conquering FairytaleQA. SIGIR, 2023. CCF-A.
  • |Paper Link| Wei Peng, Ziyuan Qin, Yue Hu, Yuqiang Xie, Yunpeng Li. FADO: Feedback-Aware Double COntrolling Network for Emotional Support Conversation. Knowl. Based Syst, 2023. SCI District 1, Chinese Academy of Sciences.
  • |Paper Link| Wei Peng, Yue Hu, Luxi Xing, Yuqiang Xie, Yajing Sun et al. Control Globally, Understand Locally: A Global-to-Local Hierarchical Graph Network for Emotional Support Conversation. IJCAI, 2022. CCF-A.
  • |Paper Link| Wei Peng, Yue Hu, Jing Yu et al. APER: AdaPtive Evidence-driven Reasoning Network for Machine Reading Comprehension. Knowl. Based Syst, 2021. SCI District 1, Chinese Academy of Sciences, Impact Factor 8.038.
  • |Paper Link| Wei Peng, Yue Hu, Yajing Sun et al. Modeling Intention, Emotion and External World in Dialogue Systems. ICASSP, 2022. CCF-B.
  • |Paper Link| Jingcheng Deng, Wei Peng*. IRRGN: An Implicit Relational Reasoning Graph Network for Multi-turn Response Selection. EMNLP,2022. CCF-B.
  • |Paper Link| Yunpeng Li, Wei Peng, et al. Learning to Know Myself: A Coarse-to-Fine Persona-aware Training Framework for Personalized Dialogue Generation. AAAI,2023. CCF-A.
  • |Paper Link| Wei Peng, Yue Hu, Luxi Xing et al. Do You Know My Emotion? Emotion-Aware Strategy Recognition towards a Persuasive Dialogue System. ECML-PKDD, 2022. CCF-B.
  • |Paper Link| Wei Peng, Yue Hu , Luxi Xing, Yuqiang Xie, Jing Yu, Yajing Sun, Xiangpeng Wei. Bi-directional Cognitive Thinking Network for Machine Reading Comprehension. COLING, 2020. CCF-B.
  • |Paper Link| Wei Peng, Yue Hu, Jing Yu, Yajing Sun et al. MCR-NET: A Multi-Step Co-Interactive Relation Network for Unanswerable Questions on Machine Reading Comprehension. ICASSP, 2021. CCF-B.
  • |Paper Link| Wei Peng, Yue Hu, Luxi Xing, Yuqiang Xie, Yajing Sun. CogIntAc: Modeling the Relationships between Intention, Emotion and Action in Interactive Process from Cognitive Perspective. IJCNN, 2022. CCF-C.
  • |Paper Link| Luxi Xing, Yuqiang Xie, Wei Peng. IIE-NLP-NUT at SemEval-2020 Task 4: Guiding PLM with Prompt Template Reconstruction Strategy for ComVE. SemEval, 2020.
  • |Paper Link| Yuqiang Xie, Yue Hu, Luxi Xing, Wei Peng. CLSEG: Contrastive Learning of Story Ending Generation. ICASSP, 2022. CCF-B.
  • |Paper Link| Luxi Xing, Yue Hu, Yuqiang Xie, Wei Peng. Coarse-to-Careful: Seeking Semantic-Related Knowledge for Open-domain Commonsense Question Answering. ICASSP, 2021. CCF-B.
  • |Paper Link| Yuqiang Xie, Wei Peng, Yue Hu et al. IIE-NLP-Eyas at SemEval-2021 Task 4: Enhancing PLM for ReCAM with Special Tokens, Re-Ranking, Siamese Encoders, Back Translation. SemEval, 2021.

    • Posting

  • --

    • Award

    Competition

  • 2022 Natural Language Generation and Intelligent Writing Conference - small model (avg 4.8%=60M/1233M) TOP1
  • 2020 SemEval Task4: Commonsense Validation and Explanation TOP 3
  • 2020 Language and Intelligence Technology Competition – Machine Reading Comprehension Tasks 2%
  • 2021 SemEval Task4: Machine Reading Comprehension TOP 8
  • 2017 National Electronic Design Competition Shaanxi Province Undergraduate Group The Third Prize
  • 2017 The 11th Extracurricular Academic Works Competition Challenge Cup of Shaanxi Province The Second Prize
  • 2016 National College Students Mathematical Contest in Modeling The Second Prize (Shaanxi Province)
  • 2016 The fourth Program Design Competition for College students in Shaanxi Province The Third Prize

    • Scholarship

  • 2020-2021 Institute of Information Engineering, Chinese Academy of Sciences, Institute-level Excellence Scholarship
  • 2019-2020 Institute of Information Engineering, Chinese Academy of Sciences, First Class Scholarship
  • 2018-2019 Outstanding Merit Student of Chinese Academy of Sciences, Second Class Scholarship
  • 2016-2017 National Inspirational Scholarship
  • 2015-2016 National Scholarship
  • 2014-2015 National Scholarship

    • Patent soft and others

  • Two invention patents
  • Three utility model patents
  • A software copyright mobile terminal application system of Internet of vehicles, combined with wireless wifi indoor positioning, navigation, location fingerprint matching, Bluetooth communication and other technologies, uses mobile app to dynamically master parking space information, so as to facilitate users' travel and improve the utilization rate of indoor parking lots
  • More than 16 awards for cultural and sports activities and 8 awards for organizational work

    • Internship

    2021.07-2021.09

    MSRA | NLC Group | Algorithmic intern

  • Common sense knowledge generation, try to generate common sense knowledge using the Method of Non Autoregressive (Bang) on the CommonGen dataset. At the same time, the concept-related images retrieved also provide the basis for subsequent research.
  • We investigate edit-based operation generation methods and try to experiment on sentence decomposition, sentence fusion, dialogue summary and grammar correction tasks.

    • 2020.07-2020.09

    Tencent | KeenLab AI Lab | Algorithmic intern

  • Research the code similarity detection of programming languages, the main work involves model improvement and experimental iteration.
  • Applied to CNN, DPCNN, Transformer and pre-trained language model and other technologies.
  • On the task of measuring the similarity of code versions between different libraries (JPEG, Zlib, etc.), the running efficiency of the model is improved through code optimization.
  • By improving the scoring function and increasing the penalty factor (considering the string matched between versions), as far as the FREETYPE library is concerned, the accuracy of the identification of TOP3 has increased from 79.4% to 98.5%, and some progress has been made for subsequent product launches.

    • 2019.02-2019.06

    Didi Chuxing | Intelligent Customer Service AI Lab | Algorithmic intern

  • Research automatic conversation summary generation.
  • Considering the redundancy of conversations, sentence level statements are filtered based on the gate mechanism, which is improved by 1.21% in rouge-L.
  • An automatic tag construction method is designed, which uses MRC's Rouge indicator to calculate the most relevant subsummaries.
  • A multi-task learning method is proposed. By considering the relationship between sub-summaries and sentences, multiple tasks are jointly learned.
  • Wrote a patent on - A summary generation method based on dialog tag information.

    • Projects

    Multi-choice Machine Reading Comprehension

    2021.03 - 2021.04

    Converting the multi-choice machine reading comprehension task into a multi-classification problem, mainly considers the use of pre-trained language models, data enhancement, back translation, adversarial training, and model fusion to improve the effect of the model and improve the accuracy of the model.

    GIT

    Bi-directional Cognitive Thinking Network

    2020.02 - 2020.06

    Inspired by the cognitive complementary learning system, the reverse thinking network is proposed, which considers the inverse relationship between the passages and the answers, so that the machine can learn to take the initiative to ask questions, and decouple the two-way knowledge, to assist the answer generation.

    GIT

    MRC method based on Multi-step Reasoning

    2019.07 - 2020.01

    The evidence extraction module is designed for unanswerable questions in DuReader and SQuAD 2.0, and the key evidence are extracted by multi-step reasoning, which adaptively determines whether the question is answerable or not, with an improvement of 2.2% in rouge-L and 1.8% in F1 compared to the two datasets, respectively.

    GIT

    2018.10 - 2019.01

    Classification based on hierarchical Attention    
  • Combined with the GRU neural network and the Attention mechanism, considering the hierarchical structure between words, sentences and documents, the hierarchical Attention is introduced to capture word order and semantic information to a certain extent, and trained on a subset of the Tsinghua THUCNews dataset, the accuracy rate is 91.95 % increased to 93.75%.

    • 2017.12 - 2018.04

    Named Entity Recognition(NER) based on CRF+LSTM    
  • When the LSTM neural network performs NER, it only considers the local optimum, and the effect is poor, and CRF has the problem of feature engineering. Combined with the method of automatic feature extraction by neural network and global normalization of conditional random field. In the Chinese NER task, the model is improved by 4.65%, 5.34% and 5.78% compared with the CRF on person, location and recognition.

    • Society

    Live platform

  • 2021.4 Bilibili, the Haihua Research Institute Machine Reading Comprehension Competition was broadcast live, viewed by 800+.
  • 2021.3 Bilibili, live thesis – emotion and behavior recognition in conversations, viewed by 500+.
  • 2020.7 Bilibili, live thesis – Select, Answer and Explain: Interpretable Multi-hop Reading Comprehension over Multiple Documents，viewed by 1000+.
  • 2019.3 Post-Baoyan HPY, lecture on The Difference Between Graduate and Undergraduate，viewed by 2300+.

    • Academic impact

  • CSDN blog, cumulative reading 18W+，ranking 10,047 / 48W.
  • 2021.12 Invited by the Student Union, to share scientific research and academic experience in the "Peer Piloting Exchange Meeting" of the Institute of Information Engineering.
  • 2021.1 COLING-2020 The machine reading comprehension model based on Bi-directional Cognitive Thinking has been reported by many media such as Zhuanzhi, Sohu and PaperWeekly.
  • 2020.10 SemEval-2020 has been reported by AI Technology, Tencent and other media.

    • <MY> ON THE WAY <LIFE>

    彭伟 Albert

    电话：010 612 16123

    出生：1996.02

    英语：六级

    政治面貌：中共党员

    其他主页

    教育经历

    • 中国科学院大学信工所 | 博士
        时间：2018-2023
        方向：NLP 对话系统 问答
    • 长安大学 | 本科 (211) 专业排名第一
        时间：2014-2018
        方向：计算机科学与技术

    熟练语言 & 网络框架

    PYTHON LINUX C++ PyTorch

    发表论文

    会议期刊

  • 已发表/接受论文22篇(SIGIR, IJCAI, EMNLP, AAAI, et al.)；其中一作/通讯11篇，包含一区期刊2篇(影响因子8.032)，CCF-A3篇，CCF-B5篇，CCF-C1篇
  • |Paper Link| Wei Peng, Wanshui Li, Yue Hu et al. Leader-Generator Net: Dividing Skill and Implicitness for Conquering FairytaleQA. SIGIR, 2023. CCF-A.
  • |Paper Link| Wei Peng, Yue Hu, Luxi Xing, Yuqiang Xie, Yajing Sun et al. Control Globally, Understand Locally: A Global-to-Local Hierarchical Graph Network for Emotional Support Conversation. IJCAI, 2022. CCF-A.
  • |Paper Link| Wei Peng, Yue Hu, Jing Yu et al. APER: AdaPtive Evidence-driven Reasoning Network for Machine Reading Comprehension. Knowl. Based Syst, 2021. SCI District 1, Chinese Academy of Sciences, Impact Factor 8.038.
  • |Paper Link| Jingcheng Deng, Wei Peng*. IRRGN: An Implicit Relational Reasoning Graph Network for Multi-turn Response Selection. EMNLP,2022. CCF-B.
  • |Paper Link| Yunpeng Li, Wei Peng, et al. Learning to Know Myself: A Coarse-to-Fine Persona-aware Training Framework for Personalized Dialogue Generation. AAAI,2023. CCF-A.
  • |Paper Link| Wei Peng, Yue Hu, Yajing Sun et al. Modeling Intention, Emotion and External World in Dialogue Systems. ICASSP, 2022. CCF-B.
  • |Paper Link| Wei Peng, Yue Hu, Luxi Xing et al. Do You Know My Emotion? Emotion-Aware Strategy Recognition towards a Persuasive Dialogue System. ECML-PKDD, 2022. CCF-B.
  • |Paper Link| Wei Peng, Yue Hu , Luxi Xing, Yuqiang Xie, Jing Yu, Yajing Sun, Xiangpeng Wei. Bi-directional Cognitive Thinking Network for Machine Reading Comprehension. COLING, 2020. CCF-B.
  • |Paper Link| Wei Peng, Yue Hu, Jing Yu, Yajing Sun et al. MCR-NET: A Multi-Step Co-Interactive Relation Network for Unanswerable Questions on Machine Reading Comprehension. ICASSP, 2021. CCF-B.
  • |Paper Link| Wei Peng, Yue Hu, Luxi Xing, Yuqiang Xie, Yajing Sun. CogIntAc: Modeling the Relationships between Intention, Emotion and Action in Interactive Process from Cognitive Perspective. IJCNN, 2022. CCF-C.
  • |Paper Link| Luxi Xing, Yuqiang Xie, Wei Peng. IIE-NLP-NUT at SemEval-2020 Task 4: Guiding PLM with Prompt Template Reconstruction Strategy for ComVE. SemEval, 2020.
  • |Paper Link| Yuqiang Xie, Yue Hu, Luxi Xing, Wei Peng. CLSEG: Contrastive Learning of Story Ending Generation. ICASSP, 2022. CCF-B.
  • |Paper Link| Luxi Xing, Yue Hu, Yuqiang Xie, Wei Peng. Coarse-to-Careful: Seeking Semantic-Related Knowledge for Open-domain Commonsense Question Answering. ICASSP, 2021. CCF-B.
  • |Paper Link| Yuqiang Xie, Wei Peng, Yue Hu et al. IIE-NLP-Eyas at SemEval-2021 Task 4: Enhancing PLM for ReCAM with Special Tokens, Re-Ranking, Siamese Encoders, Back Translation. SemEval, 2021.

    • 在投论文

  • --

    • 获奖经历

    学科竞赛

  • 2022 自然语言生成与智能写作大会 小模型（avg 4.8%=60M/1233M） TOP1
  • 2020年 SemEval国际语义评测⼤赛中Task4: Commonsense Validation and Explanation取得TOP 3
  • 2020年 语言与智能技术竞赛——机器阅读理解任务排名前2%，其中参赛队伍1514支
  • 2021年 SemEval国际语义评测⼤赛中Task4:机器阅读理解任务1取得TOP 8
  • 2017年 全国电子设计竞赛陕西省本科组三等奖
  • 2017年 第十一届课外学术作品竞赛挑战杯省二等奖
  • 2016年 首届"中国高校计算机大赛——团体程序设计天梯赛"优秀奖
  • 2016年 全国大学生数学建模竞赛省二等奖
  • 2016年 陕西省第四届大学生程序设计竞赛三等奖

    • 奖学金

  • 2023北京市优秀毕业生
  • 2022-2023学年中国科学院院长奖学金
  • 2021-2022学年中国科学院信息工程研究所所长优秀奖
  • 2020-2021学年中国科学院信息工程研究所所级优秀奖学金
  • 2019-2020学年中国科学院信息工程研究所一等奖学金
  • 2018-2019学年中国科学院优秀三好学生，信息工程研究所二等奖学金
  • 2016-2017学年度国家励志奖学金
  • 2015-2016学年度国家奖学金
  • 2014-2015学年度国家奖学金

    • 专利、软著与其他

  • 两项发明专利
  • 三项实用新型专利
  • 一项软件著作权——车联网移动终端应用系统，结合无线wifi室内定位导航、位置指纹匹配以及蓝牙通信等技术，使用手机app动态掌握停车场车位信息，方便用户出行，提高室内停车场利用率。
  • 文体活动奖项16余项，组织工作奖项8项

    • 实习经历

    2021.07-2021.09

    MSRA | NLC Group | 算法实习生

  • 常识知识的生成，在CommonGen数据集上尝试使用Non autoregressive（Bang）的方法对常识知识进行生成。同时也检索的和概念相关的image为后续的研究做基础。
  • 调研了基于EDIT的操作生成方法。尝试在句子分解，句子融合，对话摘要和语法纠错任务上面进行实验。

    • 2020.07-2020.09

    腾讯 | 科恩实验室 AI Lab | 算法实习生

  • 研究编程语言的代码相似性检测，主要工作涉及到模型改进和实验迭代。
  • 运用到CNN,DPCNN,Transformer以及预训练语言模型等技术。
  • 最终在衡量不同库（jpeg，zlib等）之间的版本代码的相似性上，通过代码优化提升模型的运行效率；
  • 通过改进打分函数，增加惩罚因子（考虑版本间匹配的字符串），就FREETYPE库而言，TOP3的识别的准确率从79.4%->98.5%，取得了一定的进展，用于后续产品上线。

    • 2019.02-2019.06

    滴滴 | 智能客服AI Lab | 算法实习生

  • 研究自动对话摘要生成；
  • 考虑对话的冗余性，基于门机制对句子级的语句进行过滤，在ROUGE-L提升1.21%；
  • 设计了一种自动构造标签的方法，借鉴MRC的思路，通过Rouge指标计算与子摘要最相关的对话话术；
  • 提出了一种多任务的学习方法，通过考虑子摘要和句子之间的关系，多个任务联调学习；
  • 书写了一篇专利《一种基于对话标签信息的摘要生成方法》。

    • 项目经验

    多选型MRC任务

    2021.03 - 2021.04

    将多选机器阅读理解任务转化为多分类问题，主要考虑使用预训练语言模型、数据增强、Back Translation、对抗训练以及模型融合等角度对模型的效果进行改善，提升模型的准确率。

    GIT链接

    基于双向思维网络MRC生成模型

    2020.02 - 2020.06

    受认知互补学习系统的启发，提出逆向思维网络，考虑答案的文本之间的逆关系，让机器学会主动提出问题。同时将双向知识进行解耦合，利用逆向思维知识，辅助模型的答案生成。

    GIT链接

    基于自适应多步推理的MRC方法

    2019.07 - 2020.01

    针对DuReader和SQuAD 2.0中的不可回答问题设计证据提取模块，以多步推理逐渐提炼关键片段，自适应地决定问题是否可回答。相比两个数据集在ROUGE-L和F1分别提升2.2%和1.8%。

    GIT链接

    2018.10 - 2019.01

    基于层级Attention文本分类    
  • 结合GRU神经网络和Attention机制，考虑词汇、句子以及文档之间的层级结构，引入层级Attention，在一定程度上捕获词序和语义信息，在清华THUCNews数据集的一个子集上训练，准确率由91.95%提高到了93.75%。

    • 2017.12 - 2018.04

    基于CRF+LSTM的命名实体识别    
  • LSTM神经网络进行NER时，由于只考虑了局部最优，效果较差，而CRF又存在特征工程问题，结合神经网络自动提取特征和条件随机场全局归一化的方法，对中文文本进行命名实体识别，在人名、地名和机构名相比CRF提高了4.65%，5.34%和5.78%。

    • 社会传播

    平台直播

  • 2021.4 B站 海华研究院机器阅读理解竞赛直播，观看人数达800+；
  • 2021.3 B站 论文直播—对话中的情感和行为识别，观看人数达500+；
  • 2020.7 B站 论文直播—Select, Answer and Explain: Interpretable Multi-hop Reading Comprehension over Multiple Documents，观看人数达1000+；
  • 2019.3 后保研HBY 《研究生与本科生的区别》讲座，在线人数达2300+。

    • 学术影响

  • CSDN 博客 累计阅读量 23W+，粉丝1000+；
  • 2023.7 参与SIGIR国际会议进行口头学术报告分享；
  • 2023.7 参与上海 IJCAI Yes青年精英分享大会，并且进行口头学术报告分享；
  • 2022.10 在中国中文信息学会进行2022 EMNLP 论文（长文）报告分享；
  • 2021.12月受学生会邀请，在信工所《朋辈引航交流会》活动中进行科研学术经验分享；
  • 2021.1 COLING-2020基于双向思维的机器阅读理解模型被专知，搜狐，PaperWeekly等多家媒体报导；
  • 2020.10 SemEval-2020国际语义评测⼤赛被AI科技、腾讯网等多家媒体报导；