ATCS: Recommender Systems - Assignment
Pre-requisite
To run the scripts you have to put transactions.csv and Customer_product_purchases.csv in root project directory.
ALS
The ALS Script uses spark.mllib to train an ALS model on Customer_product_purchase.csv. It writes top-10 and top-100 products recommendations (in ALS/recommendations) for 10 random customers.
Item CF
The collaborative_filtering.py script uses pandas, numpy, scipy and sklearn to compute the purchase matrix and the similarity matrix. It writes top-10 and top-100 products recommendations (in item_CF/recommendations) for 10 random users.
Context
prefiltering.py
This script is used to pre-filter the transaction.csv for each context. It writes in data/ csv files for period day contexts (morning, afternoon, evening) and for 10 random stations.
contextual_filtering_ALS.py
This module contains functions to train ALS models and to get top recommendations.
contextual_filtering_CF.py
This module contain the function to get top recommendations using Collaborative Filtering.
period_day.py
This script is used to compute top-10 recommendations (for period day contexts) for ten random users using ALS and CF.
N.B. : to run the period_day.py script you'll need to run prefiltering.py
stations.py
This script is used to compute top-10 recommendations (for 10 station contexts) for ten random customers using ALS and CF.
N.B. : to run the stations.py script you'll need to run prefiltering.py
Results
The results for ALS analysis are recommendations_ALS directory. The results for CF analysis are recommendations_CF directory.
In general, the results for the same customer are different with the two analysis. However, the rusults are comparable. We also notice that some stations sell only the 9999 product and that some customers only buy pruducts during certain periods day.