LI Xinran

LI Xinran

Ph.D. applicant in NLP and Large Language Models.

LI Xinran

๐Ÿ‘‹ About Me

I am LI Xinran (ๆŽๆฌฃ็„ถ), a dual-degree M.S. student at Waseda University (Japan) and Dalian University of Technology (China).

My current research focuses on Large Language Model (LLM)-based Multi-Agent Systems and Social Simulation, where I study information diffusion, opinion evolution, and collective behaviors through cognitively diverse LLM agents.

Previously, my research centered on Affective Computing, including Emotion Recognition in Conversation (ERC), Aspect-Based Sentiment Analysis (ABSA), and Machine Translation, with particular interests in curriculum learning and large language model alignment.

I am currently actively applying for Ph.D. positions (2027 Fall Intake) โ€” please feel free to contact me!

๐Ÿ”ฅ News

  • Jul. 2026 Joined Fudan University as a Research Assistant working on LLM-based Social Simulation.
  • May. 2026 One first-author paper accepted to IJCAI 2026.
  • Mar. 2026 Joined bilibili as an Algorithm Intern in the Game AI Application Technical Group.
  • Nov. 2025 One first-author paper accepted to AAAI 2026.
  • Sep. 2025 Started my dual-degree M.S. program at Waseda University, Japan โ€” my first international academic experience.
  • Jul. 2025 One first-author paper accepted to ECAI 2025.
  • Nov. 2024 One first-author paper accepted to IEEE BIBM 2024.

๐ŸŽ“ Education

  • Waseda University, Graduate School of Information, Production and Systems โ€” Dual M.S. Degree (Sep. 2025 โ€“ Expected Mar. 2027)
    Research on Dialogue-level Aspect-Based Sentiment Analysis (DiaASQ) and LLM-based Aspect-Based Sentiment Analysis (ABSA).
    Kitakyushu Academic Research City Scholarship Recipient.

  • Dalian University of Technology, School of Software โ€” Dual M.S. Degree (Sep. 2024 โ€“ Expected Jun. 2027)
    Research on Emotion Recognition in Conversation (ERC) using Large Language Models, Graph Neural Networks, and Curriculum Learning.
    Masterโ€™s GPA: 88.6/100
    Outstanding Masterโ€™s Student Award (Top 20/300).

  • Dalian University of Technology, School of Software โ€” B.Eng. in Software Engineering (Sep. 2020 โ€“ Jun. 2024)
    Undergraduate research in Natural Language Processing (NLP) and Medical Image Analysis.
    Undergraduate GPA: 87.8/100
    Direct Admission to the M.S. Program (Top 17%).

๐Ÿง  Research Interests

Current Research

  • Large Language Model (LLM)-based Multi-Agent Systems
  • Social Simulation
  • Large Language Models (LLMs)

Previous Research

  • Emotion Recognition in Conversation (ERC)
  • Aspect-Based Sentiment Analysis (ABSA)
  • Machine Translation
  • Curriculum Learning

๐Ÿ’ผ Professional Experience

bilibili (Shanghai, China)

Algorithm Intern, Game AI Application Technical Group
Mar. 2026 โ€“ Jun. 2026

  • Built a multilingual translation pipeline for a well-known Japanese mobile game using Retrieval-Augmented Generation (RAG) with terminology and translation memory databases.
  • Developed a multi-stage translation quality estimation (QE) system integrating rule-based methods, lightweight classifiers, and LLM review.
  • Fine-tuned Qwen3-8B using SFT, GRPO, and a customized reward mechanism for game-domain machine translation.

๐Ÿงฉ Publications

1. Do LLMs Feel? Teaching Emotion Recognition with Prompts, Retrieval, and Curriculum Learning
AAAI 2026 (CORE A*, CCF A) โ€” First Author
Authors: Xinran Li, Yu Liu, Jiaqi Qiao, Xiujuan Xu

We propose PRC-Emo, a new ERC training framework that integrates Prompt engineering, demo Retrieval, and Curriculum learning to investigate whether LLMs can effectively perceive emotions in conversations. PRC-Emo introduces emotion-sensitive prompt templates capturing both explicit and implicit emotional cues, constructs the first ERC-specific demonstration retrieval repository, and incorporates curriculum strategies into LoRA fine-tuning via weighted emotional shifts. Experiments on IEMOCAP and MELD achieve new SOTA results, demonstrating the strong generalizability of our approach.
Paper (DOI:10.1609/aaai.v40i38.40446) | Code (GitHub)

2. TCDA: Thread-Constrained Discourse-Aware Modeling for Conversational Sentiment Quadruple Analysis
IJCAI 2026 (CORE A*, CCF B) โ€” First Author
Authors: Xinran Li, Xinze Che, Yifan Lyu, Zhiqi Huang, Xiujuan Xu

We propose TCDA, a novel framework for Conversational Aspect-based Sentiment Quadruple Analysis (DiaASQ) that addresses structural noise and temporal sequence challenges in multi-turn dialogues. The framework introduces Thread-Constrained Directed Acyclic Graph (TC-DAG) to filter cross-thread noise while maintaining global connectivity, and Discourse-Aware Rotary Position Embedding (D-RoPE) to align multi-layer semantics and alleviate the Distance Dilution problem. Experimental results on two benchmark datasets demonstrate that our approach achieves new state-of-the-art performance.
Paper (arxiv) | Code (GitHub)

3. Long-Short Distance Graph Neural Networks and Improved Curriculum Learning for Emotion Recognition in Conversation
ECAI 2025 (CORE A, CCF B) โ€” First Author
Authors: Xinran Li, Xiujuan Xu, Jiaqi Qiao

Developed a novel LSDGNN + ICL framework combining long- and short-distance GNNs for ERC tasks, improving emotion classification on IEMOCAP and MELD datasets.
Paper (DOI:10.3233/FAIA251292) | Code (GitHub)

4. CheX-DS: Improving Chest X-ray Image Classification with Ensemble Learning Based on DenseNet and Swin Transformer
IEEE BIBM 2024 (CCF B) โ€” First Author
Authors: Xinran Li, Xiujuan Xu, Yu Liu, Xiaowei Zhao

Designed CheX-DS, combining DenseNet and Swin Transformer for long-tail medical image classification, achieving 83.76% AUC on NIH ChestX-ray14 dataset.
Paper (IEEE BIBM 2024)

5. Extensible Multi-Granularity Fusion Network and Transferable Curriculum Learning for Aspect-based Sentiment Analysis
Under Review โ€” First Author
Authors: Xinran Li, Xiaowei Zhao, Yubo Zhu, Zhiheng Zhang, Zhiqi Huang, Hongkun Song, Jinglu Hu, Xinze Che, Yifan Lyu, Yong Zhou, Xiujuan Xu

Proposed an Extensible Multi-Granularity Fusion (EMGF) network that unifies dependency syntax, constituency syntax, attentional semantics, and external knowledge graph information. It achieves efficient feature collaborative modeling through multi-anchor triplet learning and orthogonal projection, incorporating a transferable curriculum learning strategy to enhance model generalization.
Paper (arXiv:2402.07787)

6. A Unified Framework for Emotion Recognition and Sentiment Analysis via Expert-Guided Multimodal Fusion with Large Language Models
Under Review โ€” Third Author (Second among students)
Authors: Jiaqi Qiao, Xiujuan Xu, Xinran Li, Yu Liu

Proposed an Expert-Guided Multimodal Fusion (EGMF) framework integrating multimodal cues via LLMs and LoRA fine-tuning, achieving superior results on MELD, CHERMA, MOSEI, and SIMS-V2 datasets.
Paper (arXiv:2601.07565)

๐Ÿค Professional Service

Program Committee Member

  • AAAI 2027

๐Ÿ… Honors & Awards

  • Kitakyushu Academic Research City Scholarship, Waseda University (2025)
  • Outstanding Masterโ€™s Student Award, DUT (2025)
  • Direct Admission to Masterโ€™s Program, DUT (2024)
  • Second-Class Scholarship for Academic Excellence, DUT (2021, 2022)

๐Ÿ› ๏ธ Skills

Programming & Frameworks: Python, Java, PyTorch, HuggingFace Transformers, Git, LaTeX
Research Areas: Large Language Models (LLMs), Multi-Agent Systems, Social Simulation, Natural Language Processing (NLP), Curriculum Learning, Machine Translation
Standardized Tests: TOEFL iBT 95, GRE 322

๐Ÿ“ซ Contact:

Email:

I am currently seeking Ph.D. opportunities for the 2027 Fall intake. I am always happy to discuss research ideas and potential collaborations. Feel free to reach out via email.