This is an implementation for the paper titled "The FacT: Taming Latent Factor Models for Explainability with Factorization Trees" published at SIGIR 2019.
The FacT model aims at explaining latent factor based recommendation algorithms with rule-based explanations. It integrates regression trees to guide the learning of latent factor models for recommendation, and uses the learned tree structure to explain the resulting latent factors. With user-generated reviews, regression trees on users and items are built respectively, and each node on the trees are asscoiated with a latent profile to represent users and items. The detailed algorithm can be found in the paper.
To run the code to generate experimental results like those found in our papers, you will need to run a command in the following format, using Python 2:
$ cd code
$ python main.py [-h] [--train_file] [--test_file] [--num_dim NUM_DIM]
[--max_depth MAX_DEPTH] [--lambda_u LAMBDA_U] [--lambda_v LAMBDA_V]
[--lambda_bpr LAMBDA_BPR] [--num_BPRpairs NUM_BPRPAIRS]
[--batch_size BATCH_SIZE] [--learning_rate lr] [--num_run NUM_RUN]
[--num_iter_user NUM_ITER_USER] [--num_iter_item NUM_ITER_ITEM]
[--random_seed]
The results will be stored in ./results/
In our papers, we used two widely used benchmark datasets collected from Amazon and yelp. The files in data folder are examples of preprocessed dataset with extracted feature opinion scrores. The data format is:
user_id, item_id, rating, [list of feature opinions]
Example:
1, 0, 4, 1 1 2 1
user_id = 1, item_id = 0, rating = 4, rating for feature 1 = 1, rating for feature 2 = 1.
If you use this code to produce results for your scientific publication, please refer to our SIGIR 2019 paper:
@inproceedings{,
title={The FacT: Taming Latent Factor Models for Explainability with Factorization Trees},
author={Tao, Yiyi and Jia, Yiling and Wang, Nan and Wang, Hongning},
booktitle={The 42nd International ACM SIGIR Conference on Research \& Development in Information Retrieval},
year={2019},
organization={ACM}
}
