Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/DyGRec/ASReP
Released code of SIGIR2021 Augmenting Sequential Recommendation with Pseudo-Prior Items via Reversely Pre-training Transformer.
https://github.com/DyGRec/ASReP
augmented-sequences item-prediction recommender-system sequential-recommendation transformer
Last synced: about 2 months ago
JSON representation
Released code of SIGIR2021 Augmenting Sequential Recommendation with Pseudo-Prior Items via Reversely Pre-training Transformer.
- Host: GitHub
- URL: https://github.com/DyGRec/ASReP
- Owner: DyGRec
- Created: 2021-04-15T17:10:26.000Z (almost 4 years ago)
- Default Branch: main
- Last Pushed: 2021-11-05T23:29:52.000Z (about 3 years ago)
- Last Synced: 2024-08-09T13:18:34.939Z (5 months ago)
- Topics: augmented-sequences, item-prediction, recommender-system, sequential-recommendation, transformer
- Language: Python
- Homepage:
- Size: 28.4 MB
- Stars: 50
- Watchers: 2
- Forks: 12
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
- StarryDivineSky - DyGRec/ASReP - start）问题。为了解决该问题，我们提出需要对冷启动对应的短序列（short sequence）进行增广（Augmentation），从而能够补全信息而避免冷启动的问题。 (其他_推荐系统 / 网络服务_其他)
README
# ASReP
This is our TensorFlow implementation for the paper:
SIGIR'21 [PDF](./Pre_sequence_Generation.pdf) . You may also find it on [Arxiv](https://arxiv.org/abs/2105.00522)
Please cite our paper if you use the code:
```bibtex
@inproceedings{liu2021augmenting,
title={Augmenting Sequential Recommendation with Pseudo-Prior Items via Reversely Pre-training Transformer},
author={Liu, Zhiwei and Fan, Ziwei and Wang, Yu and Yu, Philip S.},
booktitle={Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval},
year={2021}
}
```
## Paper Abstract
Sequential Recommendation characterizes the evolving patterns by modeling item sequences chronologically. The essential target of it is to capture the item transition correlations. The recent developments of transformer inspire the community to design effective sequence encoders, \textit{e.g.,} SASRec and BERT4Rec. However, we observe that these transformer-based models suffer from the cold-start issue, \textit{i.e.,} performing poorly for short sequences. Therefore, we propose to augment short sequences while still preserving original sequential correlations. We introduce a new framework for \textbf{A}ugmenting \textbf{S}equential \textbf{Re}commendation with \textbf{P}seudo-prior items~(ASReP). We firstly pre-train a transformer with sequences in a reverse direction to predict prior items. Then, we use this transformer to generate fabricated historical items at the beginning of short sequences. Finally, we fine-tune the transformer using these augmented sequences from the time order to predict the next item. Experiments on two real-world datasets verify the effectiveness of ASReP. The code is available on [this page](https://github.com/DyGRec/ASReP).
## Code introduction
The code is implemented based on Tensorflow version of [SASRec](https://github.com/kang205/SASRec).
## Environment Setup
The code is tested under a Linux desktop (w/ GTX 1080 Ti GPU) with TensorFlow 1.12 and Python 3.6.
Create the requirement with the requirements.txt
## Datasets
We use the Amazon Review datasets Beauty and Cell_Phones_and_Accessories. The data split is done in the
leave-one-out setting. Make sure you download the datasets from the [link](https://jmcauley.ucsd.edu/data/amazon/).
### Data Preprocessing
Use the DataProcessing.py under the data/, and make sure you change the DATASET variable
value to your dataset name, then you run:
```
python DataProcessing.py
```
You will find the processed dataset in the directory with the name of your input dataset.
## Baby Dataset Pre-training and Prediction
### Reversely Pre-training and Short Sequence Augmentation
Pre-train the model and output 20 items for sequences with length <= 20.
```
python main.py --dataset=Beauty --train_dir=default --lr=0.001 --hidden_units=128 --maxlen=100 --dropout_rate=0.7 --num_blocks=2 --l2_emb=0.0 --num_heads=4 --evalnegsample 100 --reversed 1 --reversed_gen_num 20 --M 20
```
### Next-Item Prediction with Reversed-Pre-Trained Model and Augmented dataset
```
python main.py --dataset=Beauty --train_dir=default --lr=0.001 --hidden_units=128 --maxlen=100 --dropout_rate=0.7 --num_blocks=2 --l2_emb=0.0 --num_heads=4 --evalnegsample 100 --reversed_pretrain 1 --aug_traindata 15 --M 18
```
## Cell_Phones_and_Accessories Dataset Pre-training and Prediction
### Reversely Pre-training and Short Sequence Augmentation
Pre-train the model and output 20 items for sequences with length <= 20.
```
python main.py --dataset=Cell_Phones_and_Accessories --train_dir=default --lr=0.001 --hidden_units=32 --maxlen=100 --dropout_rate=0.5 --num_blocks=2 --l2_emb=0.0 --num_heads=2 --evalnegsample 100 --reversed 1 --reversed_gen_num 20 --M 20
```
### Next-Item Prediction with Reversed-Pre-Trained Model and Augmented dataset
```
python main.py --dataset=Cell_Phones_and_Accessories --train_dir=default --lr=0.001 --hidden_units=32 --maxlen=100 --dropout_rate=0.5 --num_blocks=2 --l2_emb=0.0 --num_heads=2 --evalnegsample 100 --reversed_pretrain 1 --aug_traindata 17 --M 18
```