# Considering user profile and product information in biased product review classification: A Dual Memory Network Model

Outstanding Project Award (only 1)
in Best Capstone Project Award Competition of Department of Computing
Related paper
to appear in EMNLP 2018 9th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis
Outstanding Work by Students
in PolyU Institutional Research Archive

## Abstract

In sentiment analysis (SA) of product reviews, both user and product information are proven to be useful. Current tasks handle user proﬁle and product information in a uniﬁed model which may not be able to learn salient features of users and products effectively. In this work, we propose a dual user and product memory network (DUPMN) model to learn user proﬁles and product reviews using separate memory networks. Then, the two representations are used jointly for sentiment prediction. The use of separate models aims to capture user proﬁles and product information more effectively. Compared to state-of-the-art uniﬁed prediction models, the evaluations on three benchmark datasets, IMDB, Yelp13, and Yelp14, show that our dual learning model gives performance gain of 0.6%, 1.2%, and 0.9%, respectively. The improvements are also deemed very signiﬁcant measured by p-values.

## Capstone Project Report

Download Report (PolyU ID needed)

## Capstone Project Presentation

Download Slides

## Capstone Project Poster

Download Poster

## Code

Code on GitHub

## Related Publications

### Dual Memory Network Model for Biased Product Review Classiﬁcation

In sentiment analysis (SA) of product reviews, both user and product information are proven to be useful. Current tasks handle user pro- file and product information in a unified model which may not be able to learn salient fea- tures of users and products effectively. In this work, we propose a dual user and product memory network (DUPMN) model to learn user profiles and product reviews using sepa- rate memory networks. Then, the two repre- sentations are used jointly for sentiment pre- diction. The use of separate models aims to capture user profiles and product information more effectively. Compared to state-of-the- art unified prediction models, the evaluations on three benchmark datasets, IMDB, Yelp13, and Yelp14, show that our dual learning model gives performance gain of 0.6%, 1.2%, and 0.9%, respectively. The improvements are also deemed very significant measured by p-values.

### A Sentiment Analysis Method to Better Utilize User Profile and Product Information

Sentiment analysis which aims to predict users’ opinions is a huge need for many industrial services. In recent years, many methods based on neural network achieve great performance on sentiment analysis such as Tang et al. (2014); Socher et al. (2013); Mikolov et al. (2013b). However, most of existing methods only focus on local documents and do not consider related user profile and product information (Kim, 2014; Yang et al., 2016; Long et al., 2017). Even some works attempt to combine those background information in sentiment analysis, they normally treat user profile and product information as a united entity and ignore that different background information may cause different aspects of influences to users’ sentiments(Tang et al., 2015b; Chen et al., 2016; Dou, 2017). To address these issues, we proposed a new Joint User and Product Memory Network (JUPMN) utilizing user profile and product information in separate ways into sentiment classification. Inspired by the successful utilization of memory network (Weston et al., 2014; Sukhbaatar et al., 2015), our model first creates document representations using hierarchical LSTM model and then feeds the document vectors into new carefully designed user and product memory networks to reflect corresponding features. The evaluation of JUPMN on three benchmark review datasets IMDB, Yelp13 and Yelp14 shows that JUPMN outperforms the state-of-the-art model and further analysis of experimental results is employed.