Code for the paper "Ask Me Any Rating: A Content-based Recommender System based on Recurrent Neural Networks".

In this work we propose Ask Me Any Rating (AMAR), a novel content-based recommender system based on deep neural networks which is able to produce top-N recommendations leveraging user and item embeddings which are learnt from textual information describing the items.

A comprehensive experimental evaluation conducted on state of-the-art datasets such as MovieLens 1M and DBbook showed a significant improvement over all the baselines taken into account.

• Lua
• Torch
  • nn
  • rnn
  • cudnn
  • cunn
  • cutorch
  • optim
  • pl
  • cjson
  • lfs

1. Retrieve item descriptions and save them in ".txt" files. Each file should be named as the item identifier in the used dataset (e.g., item 1 has a description file named 1.txt)
2. Create JSON configuration file for the training file that you want to use and start the training using Torch specifying the configuration filename with the -config parameter
3. Evaluate the trained model using the run_amar_experiments.lua program specifying the configuration filename with the -config parameter

Training and evaluation configuration files are in JSON format and are composed by specific fields. They are used in order to modify model parameters and to specify the supplementary files used to train the models or to evaluate the models.

For instance, the training configuration file for the train_amar_rnn.lua file is composed by the following fields:

• items: path of item descriptions
• genres: filename of item genres (optional)
• models_mapping: dictionary which associates training sets to models
• rnn_unit: RNN unit identifier used in rnn package
• optim_method: optimization method identifier used in optim package
• training_params: parameters of the optimization method
• batch_size: number of training examples in a batch
• num_epochs: number of training epochs
• save_after: save model each save_after epochs

We report in the following code snippet a real configuration script that we used for our model:

{
  "items": "../datasets/ml1m/content/simple",
  "models_mapping": {
    "../datasets/ml1m/ratings/u1.base": "../resources/ml1m/amar/rnn/base/u1.model",
    "../datasets/ml1m/ratings/u2.base": "../resources/ml1m/amar/rnn/base/u2.model",
    "../datasets/ml1m/ratings/u3.base": "../resources/ml1m/amar/rnn/base/u3.model",
    "../datasets/ml1m/ratings/u4.base": "../resources/ml1m/amar/rnn/base/u4.model",
    "../datasets/ml1m/ratings/u5.base": "../resources/ml1m/amar/rnn/base/u5.model"
  },
  "optim_method": "rmsprop",
  "training_params": {
    "learningRate": 1e-3,
    "alpha": 0.9
  },
  "batch_size": 32,
  "num_epochs": 20,
  "save_after": 5
}

In addition, the evaluation configuration file for the run_amar_experiments.lua file is in JSON format and is composed by the following fields:

• items: path of items descriptions
• genres: filename of items genres
• models_mapping: dictionary which associates test files to models
• predictions: generated predictions filename
• batch_size: number of examples in a batch
• topn: list of cutoff values

We report in the following code snippet a real configuration script that we used for our model:

{
  "items": "../datasets/ml1m/content/simple",
  "models_mapping": {
    "../datasets/ml1m/ratings/u1.test": "../resources/ml1m/amar/rnn/base/u1.model",
    "../datasets/ml1m/ratings/u2.test": "../resources/ml1m/amar/rnn/base/u2.model",
    "../datasets/ml1m/ratings/u3.test": "../resources/ml1m/amar/rnn/base/u3.model",
    "../datasets/ml1m/ratings/u4.test": "../resources/ml1m/amar/rnn/base/u4.model",
    "../datasets/ml1m/ratings/u5.test": "../resources/ml1m/amar/rnn/base/u5.model"
  },
  "predictions": "../experiments/results/ml1m/amar/rnn/base/predictions_%d_%d.txt",
  "batch_size": 32,
  "topn": [10]
}

All the following authors have equally contributed to this project (listed in alphabetical order by surname):

• Claudio Greco (github)
• Alessandro Suglia (github)