qizhyuan / dsc291project

DSC291Project

Dataset Generation

Source Domain - MovieLens

The data is available at https://grouplens.org/datasets/movielens/. We use ml-latest.zip to construct our cross-domain source data, which can be download at https://files.grouplens.org/datasets/movielens/ml-latest.zip. It contains 27,000,000 ratings and 1,100,000 tag applications applied to 58,000 movies by 280,000 users.

Target Domain - Netflix Prize Data

The data is available at https://www.kaggle.com/datasets/netflix-inc/netflix-prize-data, which contains 17770 movies and 480189 users.

The generated data is provided in dataset.rar file. The notebook dataset_generator_ml_nf.ipynb contains the code to generate the data for cross-domain recommendation.

The generated dataset has three csv files, and each file has the following format: user, item, rating

Since our task is to predict the interactions, we can simply ignore the rating data (treat them as positive instances).

Dataset Statistics

Target Domain:

Train:
User Num: 29,818
Item Num: 4,799
Rating Num: 227,355
Average User Degree: 7.62
Average Item Degree: 47.37

Test:
Num: 15152

Source Domain:

User Num: 1,2481
Item Num: 4,799
Rating Num: 737,822
Average User Degree: 59.11
Average Item Degree: 153.74

Setup

Dependencies

• Numpy
• PyTorch
• scikit-learn

Start

Unzip the file dataset.rar at first. Then, use the following command to run MF and EMCDR models.

python main.py

The experimental results will be saved in the file result.json in the current dictionary.

To run the baseline models, download the notebook BaselineCDR.ipynb and run all the code blocks.

Note that this branch does not contain the SOTA methods. For SOTA method, please swith to the sota branch.

Experimental Result

MovieLens (S) -> Netflix Prize Data (T)

Model	Hits@1	Hits@5	Hits@10	Hits@15
Popularity	0.0487	0.2866	0.5560	0.7292
Jaccard Similarity	0.0419	0.3944	0.6954	0.8330
Cosine Similarity	0.0180	0.2014	0.4224	0.5837
MF	0.1197	0.4634	0.6713	0.7748
MF-Mixed	0.1209	0.4362	0.6398	0.7503
EMCDR	0.1570	0.4854	0.6463	0.7259

Model	NDCG@1	NDCG@5	NDCG@10	NDCG@15
MF	0.1197	0.2932	0.3607	0.3881
MF-Mixed	0.1209	0.2793	0.3453	0.3746
EMCDR	0.1570	0.3253	0.3775	0.3986

Netflix Prize Data (T) -> MovieLens (S) [Switch the Source and Target Domain]

Model	Hits@1	Hits@5	Hits@10	Hits@15
MF	0.3762	0.6911	0.8154	0.8745
MF-Mixed	0.3378	0.6689	0.8051	0.8676
EMCDR	0.2781	0.5822	0.7194	0.7909

Model	NDCG@1	NDCG@5	NDCG@10	NDCG@15
MF	0.3762	0.5424	0.5828	0.5985
MF-Mixed	0.3378	0.5125	0.5567	0.5733
EMCDR	0.2781	0.4364	0.4809	0.4999

Visualization

For convenience purpose, we stored and provided the detailed results of our experiments in result.json file. You may also run the main.py script to reduce the results. The visualization relies on the result.json file. The visualization code is presented in the notebook Visualization.ipynb.

For visualizing dataset statistics and results of baseline model, download and open the notebook BaselineCDR.ipynb, which contains the live demo of dataset statistics we listed above. Also, the baseline poplarity and Collabortive Filtering model is implemented at the end, running those code blocks will print the hit score of the baseline models.