KGQR

Framework for interactive recommendation system based on reinforcement learning taking leverage of knowledge graph

Contents table

1. Baseline - KGQR
2. Getting stated
3. File Tree
   
   

Baseline - KGQR

KGQR is the first trial to use knowledge graph prior on interactive recommendation system based on reinforcement learning.

"Interactive Recommender System via Knowledge Graph-enhanced Reinforcement Learning"

Getting stated

python >= 3.7

1. Download ratings.csv to raw_data folder.

   In raw_data folder, there should be ratings.csv, kg.txt, item_index2enity_id.txt files

2. Preprocess rating file and knowledge graph file

   This will generate data folder and sorting.csv

      python preprocess.py  

3. Train the model

      python train.py 

File Tree (After preprocessing)

• raw_data

  • movie
    • kg.txt: knowledge graph file;
    • ratrings.csv: raw rating file of Movielens-20M dataset;
    • item_index2entity_id.txt: the mapping 3from item indices in the raw rating file to entity IDs in the KG;
    • sorted.csv(generated by preprocess.py): sorted (by user id and timestamp) rating file of Movielens-20M dataset;

• data(generated by preprocess.py): preprocessed files

  • movie
    • *_vocab.pkl: Change indicators in raw_data files to index used in this project
    • n_hop_kg.pkl: save entity's ids in 2 hop;

        n_hop_kg.pkl[entity_id][1] = list of 1-hop entity ids
        n_hop_kg.pkl[entity_id][2] = list of 2-hop entity ids

    • kg_adj_mat.npy: Knowledge graph's adjacent nodes information;
    • *_data_dict.pkl: can refer user's rating history sorted in chronological order by user_id;