Congratulations to Ling Chenfan on His Graduation!
Ling Chenfan received his bachelor’s degree from Wuhan Institute of Bioengineering and began pursuing an academic master’s degree in the School of Computer Engineering and Science at Shanghai University in 2022. After joining the research group, under the guidance of Professors Zhang Rui, Han Yuexing, and Chen Qiaochuan, he conducted research in the field of natural language processing and completed the following work:
To address the issue of insufficient semantic information in long-span representations within existing named entity recognition (NER) methods, he proposed a span representation enhancement module based on even convolution. This module effectively captured semantic information within entity spans, strengthened long-span feature representations, and thereby improved the model’s recognition performance on long spans.
To overcome the limitation of single-dimensional contrastive objectives in existing contrastive learning-based NER frameworks, he introduced a dual contrastive learning objective based on both span and category dimensions. Compared with methods optimized only along the span dimension, this approach learned more discriminative representations for entity categories. Additionally, a lightweight trainable embedding layer was used to replace the pretrained language model at the category level, improving model efficiency while maintaining performance.
To address the insufficient interaction between subtasks in information extraction methods, he proposed a knowledge transfer approach that transfers parameters from the contrastive learning-based NER model to the relation extraction model. Specifically, this method reused the text encoder trained during the NER stage, enabling more effective encoding of entity contextual information. At the same time, a category-aware fusion module based on category encoder transfer was introduced to strengthen the relation extraction model’s awareness of entity categories. The synergy of these two transfer mechanisms enhanced the performance of entity relation extraction.
After graduation, Ling Chenfan joined SF Technology Co., Ltd. as a software developer. During his postgraduate studies at Shanghai University, he was diligent and dedicated, continuously improving his professional expertise and research capabilities. He was fortunate to work alongside excellent mentors and peers, gaining invaluable experiences. We wish him to always uphold his ideals, stay grounded, face challenges fearlessly, and move forward with courage on his future journey.
Essay: Research on Information Extraction Methods Based on Contrastive Learning and Knowledge Transfer
Code: https://github.com/han-yuexing/2025-thesis-lcf-code
