Comment Grading for Recommendation System

This project focuses on user comments with high quality but few reviews due to release time. We implement NLP methods on prediction of the new-coming comments. Therefore, we are able to select the comments that are likely to be useful, and give them more exposure to other users.

Checkout our project on github.

In this project, we finished:

• A network that can predict the helpfulness of a new comment based on information like text, user, product ID.
• A mathematical model(recommendation system) that can place a new comment at a predicted place instead of the bottom(baseline model). This function will largely decrease the cost to rerank a helpful comment and increase the probability that a potentially helpful new comment can be seen by more users.

I am responsible for:

• statistical work on dataset
• build the multimodal part of the network
• help design the recommendation system
• do simulation and analysis experiments

Here is our presentation pdf, including illustrations showing our network architecture, recommendation system design, and experiment results.