Automated Sequential Recommendation

Loading...
Thumbnail Image

Date

2021-08-09

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Most traditional recommendation systems aim to discover users' preferences by predicting top scores on items. However, this recommendation system might not fully capture users' preferences because they do not involve any sequential information. Moreover, developing a sequential recommendation system is a complex task because the settings for each sequential scenario are very different. We present an automated sequential recommendation, AutoSRec, to unify such intricate tasks to address this problem. Each sequential model shares some good attributes, and they have useful features to deal with their situations. However, a fine-grained neural model is not able to tackle all sequential recommendation problems. Our work aims to build upon three concepts: unification of sequential models, automation of recommendation systems, and user-friendly framework. To achieve this, we extract critical components of recommendation models and combine them into what we call: hyper-interactions. Then, with the help of automated machine learning, AutoML, our AutoSRec can achieve the best model in the hyper-interaction search space. Lastly, AutoSRec is based on the TensorFlow API ecosystem, where even non-experts can understand quickly. Experiments on multiple data sets show the effectiveness of our approach.

Description

Keywords

Recommendation System, Automated Machine Learning

Citation