codes.lstp.rec

Overview

This repo contains the LSTPR model and the seven baselines. Also, the Amazon-beauty dataset is included.

├── data (directory for original graph files, LSTP graph files, and field files)
├── emb (directory for embedding files)
├── preprocess.py (generate LSTP graph files)
├── field.py  (generate field files for HOP-Rec and LSTPR)
├── lfm-bpr.py (baseline: BPR)
├── lfm-warp.py (baseline: WARP)
├── smore (baseline: HOP-Rec)
├── SkewOPT (baseline: SkewOPT)
├── LightGCN (baseline: LightGCN)
├── Caser (baseline: Caser)
├── CosRec (baseline: CosRec)
├── predict.py (evaluation code for BPR, WARP, HOP-Rec, SkewOPT, and HOP-Rec)
├── utils.py (contains evaluation metrics)
├── run.sh (script file for preprocessing and all methods' usages)

Abstract

Considering the temporal order of user-item interactions for recommendation forms a novel class of recommendation algorithms in recent years, among which sequential recommendation models are the most popular approaches. Although, theoretically, such fine-grained modeling should be beneficial to the recommendation performance, these sequential models in practice greatly suffer from the issue of data sparsity as there are a huge number of combinations for item sequences. To address the issue, we propose LSTPR, a graph-based matrix factorization model that incorporates both high-order graph information and long short-term user preferences into the modeling process. LSTPR explicitly distinguishes long-term and short-term user preferences and enriches the sparse interactions via random surfing on the user-item graph. Experiments on three recommendation datasets with temporal user-item information demonstrate that the proposed LSTPR model achieves significantly better performance than the seven baseline methods.

Usages

All usages are written in the run.sh file. Use bash run.sh to run the script.

For parameter tuning and requirements of each model, please follow the corresponding README.md or reference. Note that the default parameters are same as those in the LSTPR paper.

• LightFM: https://github.com/lyst/lightfm
• SMORe: https://github.com/cnclabs/smore
• LightGCN: https://github.com/gusye1234/LightGCN-PyTorch
• Caser: https://github.com/graytowne/caser_pytorch
• CosRec: https://github.com/zzxslp/CosRec

#Generate the LSTP graph file and the field file for LSTPR
python3 preprocess.py
python3 field.py

#BPR
python3 lfm-bpr.py --train ./data/beauty_train.txt --save ./emb/beauty_lfm-bpr.txt --dim 100 --worker 16
python3 predict.py --emb_file ./emb/beauty_lfm-bpr.txt --dataset beauty --K 10
python3 predict.py --emb_file ./emb/beauty_lfm-bpr.txt --dataset beauty --K 20

#WARP
python3 lfm-warp.py --train ./data/beauty_train.txt --save ./emb/beauty_lfm-warp.txt --dim 100 --worker 16
python3 predict.py --emb_file ./emb/beauty_lfm-warp.txt --dataset beauty --K 10
python3 predict.py --emb_file ./emb/beauty_lfm-warp.txt --dataset beauty --K 20

#HOP-Rec
./smore/cli/hoprec -train ./data/beauty_train.txt -field ./data/beauty_field.txt -save ./emb/beauty_hoprec.txt -dimensions 100 -threads 16 -sample_times 200
python3 predict.py --emb_file ./emb/beauty_hoprec.txt --dataset beauty --K 10
python3 predict.py --emb_file ./emb/beauty_hoprec.txt --dataset beauty --K 20

#Skew-OPT
./SkewOPT/cli/SkewOPT -train ./data/beauty_train.txt -save ./emb/beauty_skewopt.txt -dimensions 100 -threads 16 -sample_times 200
python3 predict.py --emb_file ./emb/beauty_skewopt.txt --dataset beauty --K 10
python3 predict.py --emb_file ./emb/beauty_skewopt.txt --dataset beauty --K 20

#LightGCN
cd LightGCN/code/ && python3 main.py --decay=1e-4 --lr=0.001 --layer=3 --seed=2020 --dataset="beauty" --topks="[10,20]" --recdim=100

#Caser
cd Caser/ && python3 train_caser.py

#CosRec
cd CosRec/ && python3 train.py --dataset=beauty --d=100 --fc_dim=150 --l2=1e-6 --n_iter=2000 --learning_rate=1e-4

#LSTPR
./smore/cli/hoprec -train ./data/beauty_lstp_5.txt -field ./data/beauty_lstp_field.txt -save ./emb/beauty_lstpr.txt -dimensions 100 -threads 16 -sample_times 200
python3 predict.py --emb_file ./emb/beauty_lstpr.txt --dataset beauty --K 10
python3 predict.py --emb_file ./emb/beauty_lstpr.txt --dataset beauty --K 20