Recommendations in text-rich heterogeneous networks

In this work, we study the problem of performing recommendations in a Heterogeneous Network which has auxiliary text information present with the nodes. We cover the relevant background and illustrate heterogeneous networks along with related tasks through the help of examples. We choose the setting of Bibliographic Heterogeneous Network and devise a Citation Recommendation system that integrates the various sources of information present in the network. We utilize specific similarity matrices to compare the query paper with the set of candidate papers which enables us to capture the query-specific context of the candidate papers. Our proposed approach employs suitably transformed embeddings to create the similarity matrices and follows-up with convolution neural networks. We demonstrate the effectiveness of our solution over two popular datasets where our method outperforms several network and/or text-based methods. We also perform a thorough qualitative analysis based on sample queries to show the effectiveness of our model in holistically combining heterogeneous information.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2020

Author dc:creator

• Raj, Jeetu

Contributors dc:contributor

• Han, Jiawei

• Network Mining
• Recommendation
• Deep Learning
• Heterogeneous Networks
• Text-Rich Networks

Statement dc:rights

• Copyright 2020 Jeetu Raj

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/108339

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/108339