Meta Recommendation with Robustness Improvement

Quick Start

Step 1: Download the project

First of all, download our project RobMeta.zip from Google Drive and unzip the file. The file includes both codes and datasets.

Step 2: Create the running environment

Create Python 3.8 enviroment and install the packages that the project requires.

• numpy==1.22.3
• PyYAML==6.0
• scikit_learn==1.1.2
• torch==1.11.0

You can install the packages with the following command.

    pip install -r requirements.txt

Step 3: Run the project

Choose a dataset to run (e.g. MovieLens-1M) with the following command.

    cd movielens-1m

Choose a model (e.g. MeLU) to run with the following command.

    python run.py MeLU hiddenDim1 32 hiddenDim2 32 localLr 0.1 lr 0.1 gamma 0.99 batchSize 20 epoch 200 userNum 20000

You can also use quick_start.py to run the project conveniently.

    python quick_start.py

You can also change the hyper-parameters as you want. The necessary hyper-parameters for each model have been recorded in the hyperParam dictionary.

Step 4: Check the performance

The performance will be saved in performance.csv. The first column is the name of models. The second column is the number of testing sets (with different β). The third to the sixth columns are metrices NDCG@10, Recall@10, Precision@10, F1@10 in order.

5 Hyper-parameters search range

We tune hyper-parameters according to the following table.

Hyper-parameter	Explain	Range
taskWeightLr	user weights lr	{0.00001, 0.0001, 0.001, 0.005, 0.01, 0.05, 0.1}
sampleWeightLr	interaction weights lr	{0.00001, 0.0001, 0.001, 0.005, 0.01, 0.05, 0.1}
localLr	adapt lr	{0.001, 0.01, 0.05, 0.1}
lr	model lr	{0.001, 0.01, 0.05, 0.1}
batchSize	batch size	{20, 100, 200, 400}
gamma	learning rate decay	{0.95, 0.97, 0.99}
hiddenDim1	hidden layer1 dim	{16, 32, 64, 128}
hiddenDim2	hidden layer2 dim	{16, 32, 64, 128}
indHiddenDim	index embedding dim	{16, 32, 64, 128}

As different base models have different hyper-paramerters to tune, you can view the details in corresponding model files.