Hybrid recommendation system based on collaborative-filtering and user-based features for "Data-like" hackathon dataset made with LightFM Python library.
The project aims to solve several tasks:
- Transform given datasets into format acceptable by LightFM
- Train LightFM model
- Predict recommendation of shopping category suitable for ceratin customer_ids
The raw data-like dataset has been preprocessed and saved in two csv-files, containing encoded customers' features and information about customers' transactions.
customer.csv Contain personal information about customer in numerical form.
| customer_id | gender_cd | marital_status_cd | children_cnt | job_position_cd | job_title | first_session_year | first_session_month | first_session_day | first_session_hour |
|---|---|---|---|---|---|---|---|---|---|
| ... | ... | ... | ... | ... | ... | ... | ... | ... | ... |
transactions.csv Weight - number of transactions made by customer in appropriate MCC category.
| customer_id | merchant_mcc | weight |
|---|---|---|
| ... | ... | ... |
The example of system usage can be viewed in Recommendtation-system demonstration notebook.
- Update dataset storage links
- Add initial dataset preprocessing code
- Improve output format