This repository is the implementation of KGPL (arXiv):

Alleviating Cold-Start Problems in Recommendation through Pseudo-Labelling over Knowledge Graph Riku Togashi, Mayu Otani, Shin'ichi Satoh.
In Proceedings of The 14th ACM International Conference on Web Search and Data Mining (WSDM 2021)

KGPL is a method to alleviate cold-start problems in personalised recommendation tasks. It expoits users' implicit feedback and a knowledge graph to enhance the sampling for pseudo-labelling. The codes are mainly based on the implementation of KGNN-LS by authors (here).

• data/
  • movie/
    • item_index2entity_id.txt: the mapping from item indices in the raw rating file to entity IDs in the KG;
    • kg.txt: knowledge graph file;
    • ratings.txt: raw rating file of MovieLens-1M;
  • music/
    • item_index2entity_id.txt: the mapping from item indices in the raw rating file to entity IDs in the KG;
    • kg.txt: knowledge graph file;
    • user_artists.dat: raw rating file of Last.FM;
  • book/
    • item_index2entity_id.txt: the mapping from item indices in the raw rating file to entity IDs in the KG;
    • kg.txt: knowledge graph file;
    • BX-Book-Ratings.csv: raw rating file of Book-Crossing;
• conf/:
  • dataset: "dataset_name.yaml" contains the path configurations;
  • model: "model_name.yaml" contains the model configurations;
  • config.yaml: default config file;
  • preprocess.yaml: default config file for path finding;
• preprocess:
  • preprocess.py: preprocessing for raw rating files;
  • split_data.py: script for splitting data;
  • make_path_list.py: script for path finding;
• models:
  • train.py: the implementation of training process;
  • aggregators.py: the aggregator of KGNN-LS;
  • kgpl.py: the implementation of training process;
  • utils.py: utilities;

• Music

  $ python preprocess/preprocess.py -d music
  $ python preprocess/make_path_list.py lp_depth=6 dataset=music kg_path=data/music/kg_final.npy rating_path=data/music/ratings_final.npy num_neighbor_samples=32
  $ python preprocess/split_data.py dataset=music rating_path=data/music/ratings_final.npy
  $ python main.py model=kgplcot_music log.experiment_name=music optimize.n_epochs=40 dataset=music evaluate.user_num_topk=2000 log.show_loss=True
  

• Movie

  $ python preprocess/preprocess.py -d movie
  $ python preprocess/make_path_list.py lp_depth=5 dataset=movie kg_path=data/movie/kg_final.npy rating_path=data/movie/ratings_final.npy num_neighbor_samples=32
  $ python preprocess/split_data.py dataset=movie rating_path=data/movie/ratings_final.npy
  $ python main.py model=kgplcot_movie log.experiment_name=movie optimize.n_epochs=40 dataset=movie evaluate.user_num_topk=2000 log.show_loss=True
  

• Book

  $ python preprocess/preprocess.py -d book
  $ python preprocess/make_path_list.py lp_depth=6 dataset=book kg_path=data/book/kg_final.npy rating_path=data/book/ratings_final.npy num_neighbor_samples=8
  $ python preprocess/split_data.py dataset=book rating_path=data/book/ratings_final.npy
  $ python main.py model=kgplcot_book log.experiment_name=book optimize.n_epochs=40 dataset=book evaluate.user_num_topk=2000 log.show_loss=True
  

You can suppress the logs for each training step (e.g. losses and gradients) by setting log.show_loss=False. See config.yaml for further detailed configurations.