A Multi-information Fusion Model for Shop Recommendation Based on Deep Learning

摘要

With the development of the e-commerce, people have become accustomed to posting their shopping experiences on websites and making consumption decisions based on reviews given by other consumers. Recommendation for shops has gradually evolved with the rise of review websites. Traditional recommendation methods are usually based on rating matrix. As abundant information is readily accessible, information of the context is gradually used by the recommendation algorithms. This paper proposes a fusion recommendation model RCFM, an integration of the Recurrent Neural Network (RNN), the Convolutional Neural Network (CNN) and the Factorization Machine (FM), which utilized the semantic information of the reviews, the feature information of the shop images and the attribute information of the customer and shop. RNN and CNN undertake tasks of text semantic extraction and image feature extraction, respectively. Feature-fusion vector is then passed into the Factorization Machine to generate the prediction rating. Finally RCFM is trained and tested with real data sets. Results show that RCFM significantly outperforms traditional baseline methods.