KLQR

Introduction

Knowledge-aware Logical Query-based Recommendation (KLQR) is a knowledge-aware recommendation framework, which consists of four components: (1) Query Generation Module, (2) Embedding Module (3) Logical Reasoning Module, (4) Prediction Module.

Files

• data/

  • BookCrossing/: raw dataset of Book-Crossing

    • 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 BookCrossing

  • MovieLens1M_0.1/,MovieLens1M_0.3/,MovieLens1M_0.5/: preprocessed sparse dataset of MovieLens1M with respective sparsity ratio 10%, 30%, 50%

    • train.txt: train file
    • valid.txt: valid file
    • test.txt: test file
    • kg_final.txt: KG file

  • Yelp_0.1/: Preprocessed sparse dataset of Yelp

    • train.txt: train file
    • valid.txt: valid file
    • test.txt: test file
    • kg_final.txt: KG file

• src/

  • data_preparation.py: proprocess raw data files
  • dataloader.py: dataloader for training and testing
  • models.py: colleciton of models (Embedding Module, Logical Reasoning Module, Prediction Module) in KLQR
  • path_extration.py: path extraction code for Query Generation Module
  • util.py: collection of helper funtions

• main.py: main code

Running the codes with Reproducibility

• BookCrossing

  $ python src/data_preparation.py -dataset BookCrossing
  $ python src/path_extraction.py -dataset BookCrossing
  $ CUDA_VISIBLE_DEVICES=0 python main.py --cuda --do_train --do_valid --do_test --valid_steps 100 --data_path data/BookCrossing -n 128 -b 128 -d 32 -g 15 -lr 0.0001 --l2_lambda 1e-2 --geo beta --beta_mode "(800,2)" --tasks "1p.2p.3p"
  

• MovieLens1M_0.1

  $ python src/path_extraction.py -dataset MovieLens1M_0.1 
  $ CUDA_VISIBLE_DEVICES=0 python main.py --cuda --do_train --do_valid --do_test --valid_steps 100 --data_path data/MovieLens1M_0.1 -n 128 -b 128 -d 32 -g 15 -lr 0.0001 --l2_lambda 1e-2 --geo beta --beta_mode "(400,2)" --tasks "1p.2p.3p"
  

• MovieLens1M_0.3

  $ python src/path_extraction.py -dataset MovieLens1M_0.3
  $ CUDA_VISIBLE_DEVICES=0 python main.py --cuda --do_train --do_valid --do_test --valid_steps 100 --data_path data/MovieLens1M_0.3 -n 128 -b 128 -d 32 -g 6 -lr 0.01 --l2_lambda 1e-2 --geo beta --beta_mode "(1600,2)" --tasks "1p.2p.3p"
  

• MovieLens1M_0.5

  $ python src/path_extraction.py -dataset MovieLens1M_0.5
  $ CUDA_VISIBLE_DEVICES=0 python main.py --cuda --do_train --do_valid --do_test --valid_steps 100 --data_path data/MovieLens1M_0.5 -n 128 -b 128 -d 32 -g 6 -lr 0.01 --l2_lambda 1e-2 --geo beta --beta_mode "(1600,2)" --tasks "1p.2p.3p"
  

• Yelp_0.1

  $ python src/path_extraction.py -dataset Yelp_0.1
  $ CUDA_VISIBLE_DEVICES=0 python main.py --cuda --do_train --do_valid --do_test --valid_steps 100 --data_path data/Yelp_0.1 -n 128 -b 128 -d 32 -g 6 -lr 0.01 --l2_lambda 1e-2 --geo beta --beta_mode "(1600,2)" --tasks "1p.2p.3p"