Aspect-Aware Latent Factor Model (ALFM)

This is our implementation for the paper:

Zhiyong Cheng, Ying Ding, Lei Zhu, Mohan Kankanhalli. Aspect-Aware Latent Factor Model: Rating Prediction with Ratings and Reviews. In Proceedings of WWW '18, Lyon, France, April 23-27, 2018.

Our model is worked in two steps: in the first step, a topic model is used to extract the topic representation of aspects; in the second step, the resutls of the topic model are integrated into an aspect-aware latent factor model to estimate users and items latent factors as well as factor weights.

Please cite our WWW'18 paper if you use our codes. Thanks!

Author: Dr. Zhiyong Cheng (https://sites.google.com/view/zycheng)

Codes

Our code is in java:

"topicmodel" package: the implementation of the topic model: -- please run "tuningAspectNumberandTopicNumber.java" to get the results by setting data path correctly

"alfm" package: the implementation of the aspect-aware latent factor model -- please run “topicFactorTuning.java" to get the results.

Notice that alfm is implemented based on "LibRec". We already include the necessary 'jar' packages in the 'lib' fold

Examples:

In the "data" fold, we put the "Music Instruments" dataset to show the data format used in the codes.

Data format: "userIndex \t\t itemIndex \t\t rating \t\t reviews". Sentences in reviews are seperated by "||".

The model can be tested on this dataset by running "tuningAspectNumberandTopicNumber.java" frist and then “topicFactorTuning.java"

The topic results (based on 5 aspects and 5 topics) are saved in "model/topicmodel/"

Results of "alfm" are saved into "model/alfm/"