Improved Convolutional Matrix Factorization for Recommendation System

Overview

The extreme sparsity of the rating data seriously affects the recommendation quality of the recommendation system. In order to alleviate the problem of data sparsity, some convolutional neural network (CNN)-based models make full use of text data to improve the recommendation accuracy. However, due to the inherent limitations of the traditional convolutional network, they have difficulty in effectively using the contextual information of the document, resulting in an insufficient understanding of the document. This paper improves the convolutional matrix factorization model by structural innovations. Our extensive evaluation of two real data sets shows that even with rating data that is extremely sparse, the performance of the improved model far exceeds the original model.

All code in the project is original, refer to: https://github.com/WZ-ZXY/NNDL_PJ

Our selected paper: https://dl.acm.org/doi/10.1145/2959100.2959165

Requirements

• Python 3.7.6
• Tensorflow 2.2.0
• Keras 2.24

How to Run

Note: Run python main.py in bash shell. You will see the process.

File Explanation

• Input File

  The input data is saved in dataset folder

• Output File

  • The w2v folder stores the embedding information of the training model
  • The result folder stores the weights of the training model
  • The loss folder stores the loss of the output of each training model

• Code File

  • main.py: Responsible for running and training models of the main program
  • data_preprocessing.py: Responsible for data preprocessing, return and save data in proper format
  • model.py: Responsible for saving model code
  • attention.py: Responsible for the realization of the attention module
  • configuration.py: Configuration file, which mainly stores the hyperparameters of the model
  • utils.py: Contains some other needed functions

File tree is shown as following:

dataset/

1m/

0-fold

...

10m/

ml-10m_movies.dat

...

TV_R_m2u.pkl

..

result/

w2vmodel/

logs/

total_state_.log

...

configuration.py

data_preprocessing.py

attention.py

mode.py

main.py

utils.py

Notes

The folds are empty, if you wanna acquire these data and configuration file, please contact me: [email protected] or [email protected].