SAME: Semantic Aware ModeEls

SAME: Semantic Aware ModeEls is a framework with different traditional recommendation models and a semantic-aware content-based recommendation model that exploits textual features of items obtained from the Linked Open Data and deep learning transformers like BERT (Bidirectional Encoder Representations from Transformers).

Description

In this project, we have included the following traditional recommendation models:

• A random recommendation model that predicts a random rating based on the distribution of the training set, which is assumed to be normal. This model was provided by Surprise library.

• A traditional content-based recommendation model based on Vector Space Model (VSM) to represent the textual information of items, by using TF-IDF weights. This model was implemented by using Sklearn library.

• A semantic-aware content-based recommendation model, based on BERT classifier able to train and test any text information with its related labels.

• An alternative of the deep content-based recommendation model proposed by Musto, called deepCBRS, that uses a binary classifier based on Bidirectional Recurrent Neuronal Networks (BRNNs). Ir order to use 5 classes, we adapted the source code of the original model. In addition, we do not use embedding models to represent the textual information of the items.

C. Musto, T. Franza, G. Semeraro, M. de Gemmis, and P. Lops, “Deep content-based recommender systems exploiting Recurrent Neural Networks and Linked Open Data,” in 26th Conference on User Modeling, Adaptation and Personalization (UMAP). ACM, July 2018, pp. 239–244.

Requirements

The libraries used in this project with its respective versions can be seen in requirements.txt.

Usage

1. To use the BERT recommendation model, prepare the task and the input correctly. Some examples are available in data_processors.py.
2. Create JSON configuration file for each recommender.
3. Executable scripts are provided under the /experiments directory. Specifically, recommendation scripts are available in /experiments/models/recommendation package:
   • bert_recommender.py
   • random_recommender.py
   • content_based_recommender.py
   • deepcbrs_recommender.py

Configuration files

The 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 configuration file for the bert_classifier.py file is composed by the following fields:

{
	"bert_model" : "bert-base-uncased",
	"task_name" : "cbrs",
	"do_lower_case" : true,
	"train_batch_size" : 5,
	"test_batch_size" : 1,
	"gradient_accumulation_steps" : 1,
	"num_train_epochs" : 7,
	"learning_rate" : 0.000001,
	"warmup_proportion" : 0.1,
	"max_seq_length" : 512,
	"local_rank" : -1,
	"no_cuda" : false,
	"path" : "~/Semantic_aware_models/"
}

License

Open source license: If you are creating an open source application under a license compatible with the GNU GPL license v3 you may use SAME under its terms and conditions.

Contributors

• María del Carmen Rodríguez Hernández - [email protected]
• Sergio Sabroso Lasa - [email protected]
• Rosa María Montañés Salas - [email protected]
• Rafael del Hoyo Alonso - [email protected]
• Sergio Ilarri - [email protected]

Reference paper

@inproceedings{9316466,
  author={María del Carmen Rodríguez-Hernández and Rafael del-Hoyo-Alonso and Sergio Ilarri and Rosa María Montañés-Salas and Sergio Sabroso-Lasa},
  booktitle={17th ACS/IEEE International Conference on Computer Systems and Applications (AICCSA)}, 
  title={An Experimental Evaluation of Content-based Recommendation Systems: Can Linked Data and BERT Help?}, 
  year={2020},
  publisher={IEEE Computer Society},
  location={Antalya, Turkey},  
  pages={1-8},
  doi={10.1109/AICCSA50499.2020.9316466},
  issn={2161-5330},
  month={November}  
}