Explainable Recommendation System for Solving Review Loss

This is our implementation for the paper:

Explainable Recommendation System for Solving Review Loss

Author: Sean Chen ([email protected])

ABSTRACT

We proposed a review-base recommender system named HANN-Plus, a hierarchical attention neural network to model user’s preference and product’s preference. HANN-Plus composed of two sub-models. The first one is rating prediction model named HANN-RPM, we adjust the calculation method of attention mechanism to improve the reviews’ extraction performance of model. The second one is review generation model named HANN-RGM, which is based on encoder-decoder architecture and can be used to generate the representation for making user aware of why such products are recommended.

Training flow

Rating Prediction Model

Review Generation Model

Environments

• Python 3
• Pytoch
• tqdm
• gensim
• numpy
• pymysql

Dataset

In this experiments, we use the datasets from Amazon prodct data. (http://jmcauley.ucsd.edu/data/amazon/)