Information network analysis for recommendation

Abstract

Algorithms for recommendation systems are always attracting considerable attention due to its promising commercial/industrial usage and the rapid development of machine learning. Heterogeneous information network(HIN), as a natural and powerful data structure, is more capable of representing the data as well as the relationships among data objects, and the node embedding has been proved to be successful in describing the features in HIN. Therefore, the recommendation based on HIN could be more attractive to researchers. This project aims to develop an intelligent recommendation system based on node embedding by using aspect embedding (AspEm) learning algorithm in HIN. The recommendation system is used to recommend the most appropriate examiners for graduate students in the School of Electrical & Electronic Engineering(EEE). In this project, we investigated different embedding algorithms in both natural language processing(NLP) and information network(IN) approaches and chose AspEm to implement the system. AspEm is a novel method that takes advantage of different types of nodes in HIN to allow the user to learn node embeddings from any aspect of their interest. We performed the information network analysis on the constructed HIN and optimized the system by examining and selecting the most appropriate aspects of the HIN. Since the project is developed from scratch, necessary text data preprocessing, HIN data construction, and software engineering practices are required in this project.