Development of mobile recommender system based on images analysis using with neural networks

In this study we focus on the problem of user interests’ classification in mobile product recommender systems. We propose a two-stage procedure. At first, the image features are learned by fine-tuning a convolutional neural net-work, e.g., MobileNet. In the second stage, we use learnable pooling techniques such as a neural aggregation network and context gating in order to compute a weighted average of image features. As a result, we can capture the relationships between the images of products purchased by the same user. We provide an experimental study with the Amazon Fashion dataset that shows that our approach achieves an F1-measure of 0.58 for 15 recommendations, which is much higher when compared to 0.29 F1-measure classification of traditional averaging of the feature vector. Moreover, a mobile recommender system application was developed that utilizes the proposed approach as its core algorithm.