Learning Robust Sequential Recommenders through Confident Soft Labels

Paper summary

Framework of CSRec, with (a) model-level teachers, (b) data-level teachers, and (c) training-level teachers

In this work, we propose CSRec, a new learning framework for sequential recommenders. The key insight is to introduce confident soft labels to provide robust guidance when learning from useritem interactions. More precisely, CSRec contains a teacher module that generates high-quality and confident soft labels and a student module that acts as the target recommender and is trained on the combination of dense, soft labels and sparse, one-hot labels.

We propose and compare three approaches to constructing the teacher module: (i) model-level, (ii) data-level, and (iii) training-level. To evaluate the effectiveness and generalization ability of CSRec, we conduct experiments using various state-of-the-art sequential recommendation models as the target student module on four benchmark datasets. Our experimental results demonstrate that CSRec is effective in training better performing sequential recommenders.

Running experiments

Requirements

This code is written in PyTorch. Any version later than 1.10 is expected to work with the provided code. Please refer to the official website for an installation guide.

We recommend to use conda for installing the requirements. If you haven't installed conda yet, you can find instructions here. The steps for installing the requirements are:

Create a new environment
```
conda create env -n CSRec
```
In the environment, a python version >= 3.8 should be used.
Activate the environment
```
conda activate CSRec
```
Install the requirements within the environment via pip:
```
pip install -r requirements.txt
```

Datasets

Last.FM,
Yelp and
two categories (e.g., Electronics and Movies and TV) of Amazon datasets to assess our methods.

We use the preprocessed dataset from Recbole, and further process them according to our paper (remove users and items with less than five interactions for all datasets). The final version of dataset is put in Data.

Training

[note]: In the code, dvae denotes CSRec-T.

The config files are in corresponding directories (e.g., config2.electronics, etc.)

The most useful arguments are T, alpha and dvae_alpha, which corresponds to $\mathcal{T}$, $\beta$ and $\alpha$ in the paper.

For CSRec-M, the reference configurations are called config_distill_ensemble.json, you can directly run it which will automatically train the teachers and student.

python main.py --config_file="config_distill_ensemble.json"

For CSRec-D, the reference configurations are called config_distill_partial.json. Currently you need to train teachers with sub-datesets manually, with proper configurations,

python main.py --config_file="config_partial.json"

then,

python main.py --config_file="config_distill_partial.json"

For CSRec-T, the reference configurations are called config_dvae.json. The training process is similar to CSRec-D. First you need trained teachers using sub-datasets, then,

python main.py --config_file="config_dvae.json"

Testing

python main.py --config_file="config_test.json"