Rationalizing Recommender Systems

This is the repository for the code written for my Master's Thesis. It trains and evaluates an interpretable Recommender System model as described in MThesis_Michael_Buehler.pdf.

All code is stored in the folder Source/. The folder Data/Preprocessed_Data/ stores the data sets (see below). The data from the experiments described in the written part can be found in the folder Results.

Installation instructions

Getting the Data

The code comes together with all necessary data sets to train the model on a small data set called Wine. The data sets are stored in the folder Data/Preprocessed_Data/Wine/.

The data sets for training the categories Kindle and Magazine Subscriptions can be downloaded here. Store the downloaded and unpacked folders in Data/Preprocessed_Data/.

Dependencies

Please make sure that the following libraries are successfully installed on your system for running the code in 'Source/Model/':

1. Theano >= 0.9
2. Python 2.7
3. Numpy
4. Pytorch >= 0.1.12 (to run the sentence embedding model)

To run the code to prepare the data sets (optional) you additionally need:

5. NLTK + the 'punkt' package from the NLTK Downloader

Running the Code

1. Add the folder Source/ to your python path. (E.g. run export PYTHONPATH=$PYTHONPATH:/path/to/RRS/Source in your terminal.)

2. Go to the folder Source/Model/FLIC/ and then run the command make. This will construct the file _flm.so which is needed for the FLIC model.

3. If you want to train the FLIC model (optional), go to the folder Source/Model/FLIC/. Then you can train the FLIC model by running python flic.py [OPTIONS]. We recommend that you have a look at the available options by running python flic.py -h.

4. If you want to train the Sentence Embeddings (optional), go to the folder Source/Model/Sentence_Embedding/. Then run python sentence_embeddings.py [OPTIONS]. We recommend that you have a look at the avaliable options by running python sentence_embeddings.py -h.

5. If you want to train the Recommender System model, go to the folder Source/Model/RRS/. There you can train and evaluate the data set Wine by running sh run_Wine.sh. Please have a look at the available options by running python rrs.py -h.
   After you have downloaded the data sets Kindle and Magazine Subscriptions (see above), you can train and evaluate the models by running sh run_Kindle.sh or sh run_Magazine.sh.