How Bibliographic Features Contribute to Scientific Topic Prediction

Huo, Chaoguang; Liu, Xiaozhong; Ma, Shutian

Permalink

https://hdl.handle.net/2142/109666 Copy

Description

Title

How Bibliographic Features Contribute to Scientific Topic Prediction

Author(s)

• Huo, Chaoguang
• Liu, Xiaozhong
• Ma, Shutian

Issue Date

2021-03-17

Keyword(s)

• Scientific Topic Prediction
• Bibliographic Knowledge Graph
• Node2vec
• SVR
• PageRank

Abstract

Scientific topic prediction has become an important research topic for innovation forecasting and knowledge discovery. This study proposes the use of heat as an indicator of scientific topic prediction and a tool to mine topics’ features from a dynamic bibliographic knowledge graph (BKG) to improve prediction accuracy. Using 4.5 million PubMed Central papers and the knowledge graph Medical Subject Heading (MeSH) as examples, a dynamic novel BKG is constructed via time slices, and the heat of all bibliographic entities at each time slice is computed by utilizing modified PageRank algorithms (TopicRank, PaperRank, AuthorRank, and VenueRank). Then, node2vec is used to embed the graphs and mine features for each topic at each time slice to derive the heat time series of a topic and its features, which transform the scientific topic prediction problem into multivariate time series forecasting. The experimental results show that the dynamic BKG can significantly improve the accuracy of scientific topic prediction. Two type features in particular, paper features and author features, largely affect the predictions of topic. The duration of 10 years can be used as a basis to analyze the evolution of bibliographic entities.

Publisher

iSchools

Type of Resource

text

Language

eng

Permalink

http://hdl.handle.net/2142/109666

Copyright and License Information

Copyright 2021 is held by Chaoguang Huo, Xiaozhong Liu, and Shutian Ma. Copyright permissions, when appropriate, must be obtained directly from the authors.

Owning Collections