Author: Alonso M. Guerrero Castañeda

Link to GitHub repo: https://github.com/alonsomgc/AIPI531_HW3

Objectives:

1. Train different session (contextual, sequential) based product recommendation recommenders for E-commerce use cases and compare the performances of the recommenders.

2. Include item and/or user features as side information for cold items/users.

Methodology:

We have adapted the source code to enable the inclusion of item properties/features. Thus, we compute cross-entropy logits when using only item numbers (original) and only item features/properties (addition). Following prof Ma's recommendation, we incorporate both cross-entropy logits by computing a weighted average of the two and introducing a new hyper-parameter lambda (lam) that acts as the weight. Lambda takes values from 0 to 1. lam=0 means that only the original item numbers are included. lam=1 on the contrary means that only the new item features are included. Values in between are a weighted average of the two.

Please, see SNQN.py for the modified version of the original code: https://github.com/alonsomgc/AIPI531_HW3/blob/main/SNQN.py

Please, see Colab notebook for implementation and output: https://colab.research.google.com/drive/1HC6By8GPJZg8ZkngOzoQqqXkQW9YsX6F#scrollTo=CSSIjQYh7-fa