search

540117253 / Review-based-Collaborative-Filtering

A model zoo that contains different review-based collaborative filtering models of rating prediction task for recommendation
3 stars 1 forks source link

readme

Review-based-Collaborative-Filtering

    A model zoo that contains different review-based collaborative filtering models of rating prediction task for recommendation.

Definition of Review-based Rating Prediction

image

Requirements

1. Models

1.1 WordVectorBased Models

  Model Name
1 DeepCoNN
2 DARR
3 MACF

1.2 BertBased Models

  Model Name
1 DeepCLFM
2 NCEM

2. Code Description

2.1 DataSet and DataPreprocess

The experimental dataset is Amazon Review DataSet, which can be downloaded here.

The pre-trained word vector is glove.6B, which can be downloaded here.

According to the mapping method of review text, there are two strategies to preprocess the dataSet.

  DataPreprocessor Description Directory
1 wordVectorBased_data_preprocess.py split train, valid and test dataSet. Represent the review by word vectors. You may choose different dataSet and modify the preprocess params in the first serveral lines of this code /Model_Zoo
2 bertBased_data_preprocess.py split train, valid and test dataSet. Represent the review by BERT's embedding. You may choose  different dataSet and modify the preprocess params in the first serveral lines of this code /Model_Zoo

2.2 Models Definition

2.2.1 WordVectorBased Models
  Model Name Directory
1 DeepCoNN.py Model_Zoo/models/
2 DARR.py Model_Zoo/models/
3 MACF.py Model_Zoo/models/
2.2.2 BertBased Models
  Model Name Directory
1 DeepCLFM.py Model_Zoo/models/
2 NCEM.py Model_Zoo/models/

2.3 Runner

  Runner Description Directory
1 run_wordVectorBased_model.py rnn the wordVectorBased models, you may choose the specific model by commenting out codes nearly line 478 /Model_Zoo
2 run_bertBased_model.py rnn the wordVectorBased models, you may choose the specific model by commenting out codes nearly line 425 /Model_Zoo

3. How to Use

  1. Make sure the glove pre-trained word vector (download). and Amazon Review DataSet (download) in the following directory:

    • /Model_Zoo/data/
    • Automotive_5
    • Automotive_5.json
    • glove.6B
    • glove.6B.50d.txt
    • glove.6B.100d.txt
    • glove.6B.200d.txt
    • glove.6B.300d.txt

  2. If you run the WordVectorBased Models, first to run the wordVectorBased_data_preprocess.py, and then run the run_wordVectorBased_model.py.

  3. If you run the BertBased Models, first to make sure the bert_service is started (How to start bert_service can read here). Run the bertBased_data_preprocess.py (My bert_service machine's IP is 10.1.63.214, so you can change the IP of your bert_service machine in the first serveral lines of this code), and then run the run_bertBased_model.py.