Yuchen Yan

Zhejiang University

Beijing Normal University

Intelligent Analysis and Reasoning for Cross Media Content Management

- Multimodal retrieval on a knowledge base of tens of millions of products

- Efficient extraction of product text and visual features using CLIP

- Using Faiss to construct a vector retrieval index, with a retrieval time of less than 0.5 seconds and an accuracy rate of no less than 85%

Multi-dimensional indexing of scientific papers for domain knowledge organization

- Design functional vocabulary recognition algorithm and subject word hyponym relationship calculation method

- Automatic functional indexing and subject indexing of 900000 scientific papers in 1203 technological fields

- Build a knowledge graph of scientific papers, including 1.9 million nodes and 16.4 million relationships

- Design a knowledge graph based scientific paper retrieval system to support scholars' exploratory retrieval needs

- Publication: Yan, Y., & Chen, C. SciGraph: A Knowledge Graph Constructed by Function and Topic Annotation of Scientific Papers[C]. EEKE@JCDL2022.

Research on Domain Term Recognition and Terminology Standardization for Technical Themes

- Building a Terminological Co-occurrence Network and a Terminological semantic network

- Form a hierarchical conceptual system of terminology based on the generated network

- Based on a hierarchical concept system, sort out the granularity of concepts and complete the standardization of terminology concepts

Research on Multidimensional Academic Expertise Identification and Attribute Measurement for Quantitative Evaluation of Scientific Researchers

- Responsible for semantic based knowledge entity extraction and entity linking

Research and development of models and algorithms related to technological innovation data services

- Conduct tasks such as Chinese image classification, topic word recommendation, and entity recommendation on 5.5 million scientific and technological papers

Research on Network Rumor Recognition Model and Characteristics of Domestic Public Health Emergencies Based on Deep Learning Method

- Crawl online rumors related to the COVID-19, analyze the characteristics of rumors, and propose deep characteristics such as public trust index and microblog influence

- Build a text model based on BERT and integrate it with emotional models, user and communication models, with an F1 value of 0.93

- Publication: 颜聿辰,郑婧林,李轶楠,陈翀. 基于多维特征和集成学习的网络谣言识别研究. 2021年中国情报学年会.