This repository provides a reference implementation of the model in the following paper:

PARSRec: Explainable Personalized Attention-fused Recurrent Sequential Recommendation Using Session Partial Actions Ehsan Gholami, Mohammad Motamedi, Ashwin Aravindakshan 2022.

%Open access link to the paper coming soon

This code, provides tools to:

1. Generate synthetic dataset described in the paper
2. Generate Dataset and Dataloader for any dataset with specified file format
3. Train and evaluate PARSRec model on the dataset

For inqueries please contact Ehsan Gholami (contact: [email protected]). Please feel free to file an issue if an issue is detected.

If you find this code/data useful for your research, please consider citing the following paper:

Ehsan Gholami, Mohammad Motamedi, and Ashwin Aravindakshan. 2022. PARSRec: Explainable Personalized Attention-fused Recurrent Sequential Recommendation Using Session Partial Actions. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD ’22), August 14–18, 2022, Washington, DC, USA. ACM, New York, NY, USA, 11 pages. https://doi.org/10.1145/3534678.3539432

To generate synthetic dataset, set the arguments in the 'synthetic.sh' file and run this command in terminal:

./synthetic.sh

The dataset will be generated in the folder 'data/' in default mode

To train PARSRec on dataset, set the arguments in the 'PARSRec.sh' file and run this command in terminal:

./PARSRec.sh

The output files will be saved in the folder 'output/' in default mode

Create two files under folder 'data/' with following format:

<dataset value in PARSRec.sh>.txt

<dataset value in PARSRec.sh>_columns_dtype.json

Header line:

user_id,time,session_length,session_items

Data lines (one line per session-user):

#e.g. instance (user_id:1 ,time:5 ,session_length:6,session_items:3,7,4,8,12,34) is:

1,5,6,3,7,4,8,12,34

NOTE: item_ids must start from 2, (0 and 1 are reserved for SOB and EOB, respectively)

Please refer to 'data/synthetic.txt' for example file.

The dtype of your content for dataset generator. Always use "object" for "session_items" and "history". The rest can be user defined dtypes.

{
    "session_length":"int16",
    "session_items":"object", 
    "history":"object", 
    "time":"int", 
    "user_id":"int32"
}

And finally, set the flag --convert-dataset2binary on in PARSRec.sh file and run the ./PARSRec.sh in terminal. You only need to activate it once. After a single run, the binary files are generated and saved in folder 'data' and you can deactivate the flag for next runs.

This is output for sample synthetic dataset of 1024 users, 100 sessions per user, and 2000 items:

The emerging meta- and multi-verse landscape is yet another step towards the more prevalent use of already ubiquitous online markets. In such markets, recommender systems play critical roles by offering items of interest to the users, thereby narrowing down a vast search space that comprises hundreds of thousands of products. Recommender systems are usually designed to learn common user behaviors and rely on them for inference. This approach, while effective, is oblivious to subtle idiosyncrasies that differentiate humans from each other. Focusing on this observation, we propose an architecture that relies on common patterns as well as individual behaviors to tailor its recommendations for each person. Simulations under a controlled environment show that our proposed model learns interpretable personalized user behaviors. Our empirical results on Nielsen Consumer Panel dataset indicate that the proposed approach achieves up to 27.9% performance improvement compared to the state-of-the-art.