In English

Recommending social platform content using deep learning

A reccurrent neural network model as an alternative to existing recommender systems

Maxim Goretskyy ; Alexander Håkansson ; Jesper Jaxing ; Jonatan Almén ; Axel Olivecrona
Göteborg : Chalmers tekniska högskola, 2017. 57 s.
[Examensarbete för kandidatexamen]

In this thesis a model based on artificial neural networks, seeing how they have found successes in a variety of fields in recent years, is proposed as an alternative to existing recommender systems. A recurrent neural network model is built for recommending content to users on the social forum Reddit based on the titles of posts. The model is compared against Facebook’s fastText classifier and a model based on N-grams. The model proposed is performing close to, but not as well as either fastText or the N-grams based model. The model does not show any real advantages in its current state but a lot of potential improvements are proposed.

Nyckelord: Artificial Neural Networks, ANN, Recurrent Neural Networks, RNN,Recommender Systems, Automatic Recommendations, Text Classification

Publikationen registrerades 2017-09-06. Den ändrades senast 2017-09-06

CPL ID: 251698

Detta är en tjänst från Chalmers bibliotek