SoRecGAT: Leveraging Graph Attention Mechanism for Top-N Social Recommendation

This is our tensorflow implementation of the paper "SoRecGAT: Leveraging Graph Attention Mechanism for Top-N Social Recommendation SoRecGAT implementation".

Environment

tensorflow-gpu == 1.12.0

numpy == 1.16.0

Python == 2.7.12

scipy == 1.2.0

Example to Run the Codes

For Music dataset

python Main.py --method sorecgatitem --path ../data/music/ --dataset music --res_path ../results/ --epochs 60 --batch_size 1024 --valid_batch_siz 256 --lr 0.0004 --initializer xavier --stddev 0.02 --optimizer rmsprop --loss ce --num_factors 64 --num_negatives 1 --keep_prob 0.5 --attn_keep 1.0 --ffd_keep 1.0 --proj_keep 1.0 --hid_units [32] --n_heads [12,6] --at_k 5 --num_thread 8

For Art dataset

python Main.py --method sorecgatuser --path ../data/art/ --dataset art --res_path ../results/ --epochs 60 --batch_size 1024 --valid_batch_siz 256 --lr 0.0001 --initializer xavier --stddev 0.02 --optimizer rmsprop --loss ce --num_factors 64 --num_negatives 1 --keep_prob 0.5 --attn_keep 1.0 --ffd_keep 1.0 --proj_keep 0.7 --hid_units [32] --n_heads [12,6] --at_k 5 --num_thread 8

Dataset Description

Music Dataset

• music.train

  • Train file
  • Each line contains UserID, ItemID, ImplicitRating and Some extra details (tab separated)

• music.valid

  • Valid file
  • Each line contains (UserID, ItemID, extra detail)::negative item list

• music.test

  • Test file
  • Each line contains (UserID, ItemID, extra detail)::negative item list (space separated)

• music.social

  • Social interaction file
  • Each line contains ItemID, SocialEntityID (tab separated)

• music.user_embed.final

  • User embedding file
  • Each line contains UserID::Initial User Embedding (space separated)

• music.item_embed.final

  • Item embedding file
  • Each line contains ItemID::Initial Item Embedding (space separated)

• music.socialitem_embed.final

  • Social entity embedding file
  • Each line contains SocialEntityID::Initial SocialEntity Embedding (space separated)

Art Dataset

• art.train

  • Train file
  • Each line contains UserID, ItemID, ImplicitRating and some extra details (tab separated)

• art.valid

  • Valid file
  • Each line contains (UserID, ItemID, extra details)::negative item list (space separated)

• art.test

  • Test file
  • Each line contains (UserID, ItemID, extra details)::negative item list (space separated)

• art.social

  • Social interaction file
  • Each line contains UserID, SocialEntityID (tab separated)

• art.user_embed.final

  • User embedding file
  • Each line contains UserID::Initial User Embedding (space separated)

• art.item_embed.final

  • Item embedding file
  • Each line contains ItemID::Initial Item Embedding (space separated)

• art.socialitem_embed.final

  • Social entity embedding file
  • Each line contains SocialEntityID::Initial SocialEntity Embedding (space separated)

Link for all the datasets: https://www.dropbox.com/sh/3bkratvwuhgzctw/AABIA2GxPy4KZbmX0Do4S8b5a?dl=0