CV

Education

• Fudan University, Master of Applied Statistics, 2023 (expected)
  • Lab: FudanDISC (interested in Dialogue System, supervised by Prof. Zhongyu Wei)
• Zhejiang University, B.S of Management Information Systems, 2021
  • Honors: Outstanding Graduates, The First Prize Scholarship, Honorable Mention in MCM/ICM

Professional experience

• ByteDance, Recommendation System Research Intern
  • May. in 2022-Sept. in 2022
  • Participated in Toutiao’s push business, optimized CTR of push messages through cutting-edge recall, ranking, and frequency control algorithms, thereby directly increased DAU of Toutiao.
  • Enlarged article candidates, added WeiToutiao articles and provincial articles as enlarged candidates, carried out ab tests to prove that enlarging candidates effectively increase CTR+0.835%, DAU+0.2%.
  • Used batch softmax loss, adding real negative examples and other methods to debias hot articles in the recall stage, CTR+0.3% on the basis of effectively reducing the ratio of hot articles.
• CETHIK Group Corporation, NLP Research Intern
  • Nov. in 2020-Apr. in 2021
  • Participated in the Key R&D Project of Zhejiang Province “Artificial Intelligence-based Digital Diagnosis and System Development”, developed a medical dialogue system for ophthalmic diseases.
  • Used Scrapy to get ophthalmic entity information and relationship information from website, used Neo4j to build a medical knowledge graph, and built a simple dialogue system on the graph.

Research experience

• EASED: An Emotion-Aware Stratified Encoder-Decoder, Lead Author

  • Mar. in 2022-Sept. in 2022

  • COLING 2022 Short Paper (Submitted); The paper proposes EASED, An Emotion-Aware Stratified Encoder-Decoder for Dialogue Strategy Prediction.

  • The model adopts a multi-task framework to explicitly model emotional states of seekers, takes into account the interactive information in dialogues, and hierarchically models contexts, seekers and supporters.

  • The F1 score of EASED on ESConv dataset reaches 36.71, which is about 21.9% higher than the state-of-the-art model in Emotional Support Dialog Systems.

• Fudan-CLP Jinxin Research Center: Intelligent Call, Developer

  • Sept. in 2021-Jun. in 2022

  • An automatic dialogue system implemented with natural language processing technology.
  • Responsible for the development of Rule Engine, including Intent Recognition Module, data augmentation, Rule Logical Tree, API packaging (Python Flask).
  • Implemented a NER model in banking scenarios. In few-shot learning scenario, built BERT+CRF model to identify 8 named entities with data augmentation, and the offline test F1 score is 97.45.

• Research on Emotion Recognition in Conversations, Author

  • Sept. in 2020-Jun. in 2021
  • Bachelor thesis. Implemented DialogueRNN, DialogueGCN, BERT and Graph Attention Network with Relational Positional Encoding (RGAT) to recognize emotion in conversations on IEMOCAP dataset.
  • Proposed a new relational position encoding for RGAT. Experiments showed that RGAT with proposed encoding has better performance than RGAT with other positional encoding.

Skills

• Familiar with Python.
• Having experience R, SQL, C++, d3.js in courses or internship projects.
• English as working language, have earned 100 TOEFL score.