Distributionally Robust Sequential Recommendation

Anonymous Author

1.Abstract

Modeling user sequential behaviors have been proven effective in promoting the recommendation performance. While previous work has achieved remarkable successes, they mostly assume that the training and testing distributions are consistent, which may contradict with the dynamic nature of the user preference, and lead to lowered recommendation performance. To alleviate this problem, in this paper, we propose a robust sequential recommender framework to overcome the distribution shift. In specific, we first simulate different distributions by reweighting the training samples. Then, the maximum loss induced by the various sample distributions is minimized to optimize the 'worst-case' model for improving the robustness. Considering that there can be too many sample weights, which may introduce too much flexibility and be hard to converge, we cluster the training samples based on both hard and soft strategies, and assign each cluster with a unified weight. At last, we analyze our framework by presenting the generalization error bound of the above minimax objective, which is expected to better understand our framework from the theory perspective. We conduct extensive experiments based on three real-world datasets to demonstrate the effectiveness of our proposed framework. Empirical results suggest that our framework can on average improve the performance by about 2.27% and 3.51% on Recall and NDCG respectively.

2.Motivating Examples

Figure 1: (a) Example of the user preference shift from digital products to sports items.

(b) and (c) Examples of the user preference changes on the item brands and colors

3.Contributions

In a summary, the main contributions of this paper are presented as follows:

We propose a robust framework for sequential recommendation, which to the best of our knowledge, is the first time in the recommendation domain.
To achieve the above idea, we firstly design a basic model by tuning the sample weights. Then, we improve the basic model by clustering the training samples based on both soft and hard strategies.
We provide theoretical analysis on the designed framework based on the probably approximately correct (PAC) learning theory.
We conduct extensive experiments to evaluate our framework, and for promoting this direction, we have released our project at https://anonymousrsr.github.io/RSR/ .

4.Code and Datasets

4.1 Code files link:One Drive

datasets	Dataset folder
models	Basemodel and our framework program
utility	Utils program
run_RSR.py	Quick start program
main.py	Training program

4.1 Datasets overview link:One Drive

Dataset	# User	# Item	# Interactions	Ave.Sl	Sparsity
Sports	35,598	18,357	286,207	8.02	99.95%
Toys	19,412	11,924	156,072	8.04	99.93%
Yelp	30,431	20,033	282,399	9.28	99.95%

5.Usage

Download the code.
Download the dataset, and put it into the datasets folder.
run run_RSR.py file.
For example use soft-clustering framework and GRU4Rec as the basemodel on Sports with 10% noisy ratio.


xxxxxxxxxx
python run_RSR.py -d sports -base GRU4REC -t l_w -noisy 10

6.Detailed parameter search ranges

We tune hyper-parameters according to the following table.

Hyper-parameter	Explanation	Range
lr_base	Learning rate of basemodel	{0.1, 0.01, 0.001, 0.0001, 0.00001}
lr_w	Learning rate of weight vector	{0.1, 0.01, 0.001, 0.0001, 0.00001}
lr_f	Learning rate of cluster component	{0.1, 0.01, 0.001, 0.0001, 0.00001}
BatchSize	Batch Size	{32, 64, 128, 256, 512, 1024}
K	Class number	{4, 8, 16, 32, 64, 128}
temperature	Temperature coefficient in softmax	{0.01, 0.1, 0.2, 0.4, 0.6, 0.8, 1, 2, 100}
lambda	Scale weight of Clustering Loss	{0.0001, 0.0005,0.001,0.005, 0.01, 0.05}

7.Runtime Environment

System:Linux 20.04.2-Ubuntu
CPU: 96 Intel(R) Xeon(R) Gold 5318Y CPU @ 2.10GHz
CPU-Memory:256G
GPU:NVIDIA Corporation NVIDIA A40
Pytorch: 1.11.0
CUDA:11.6