otto2023_giba

OTTO – Multi-Objective Recommender System

Factorization Machine, predicts next click.

This repo is part of team G.B.D.T. #3 place solution. https://www.kaggle.com/competitions/otto-recommender-system/leaderboard

Run notebooks in this order:

1. preprocess_data.ipynb: read competition data and convert to multiple parquet files

2. fm-seq2seq.ipynb: train a FM model using last 10 items + hour. Write embeddings table to disk.

├── preprocess_data.ipynb        # preprocess data
├── fm-seq2seq.ipynb             # train FM model and extract embeddings
└── README.md