Graph Convolutional Networks and Text Integration for Recommender Systems

Anonymous

TL;DR: Proposal of new model for recommender systems based on keyword-centered and graph convolutional neural networks.

Abstract: We present a new model KeywordSage which consists of integration in text data and Graph Convolutional Networks for recommender systems. This model extracts keyword in most efficient way from user reviews text using language model based on Transformer and then Graph Convolutaionl Networks is efficiently trained to learn about user-item interactions by utilizing extracted keywords. This makes it possible to reflect meaningful information from users and utilize it for representing the user-item interaction. We prove that our model is more efficient showing that KeywordSage result in better performance even with significantly fewer learning steps compared to existing models. Our approach is to be a meaningful contribution in that it proposes a new recommender systems by combining Natural Language Processing and a graph-based neural networks, suggesting a direction for covering research in both fields.

Paper Type: long

Research Area: Information Extraction

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Approaches low compute settings-efficiency

Languages Studied: English