An image recommendation technique based on fuzzy inference system

Nowadays, social network users are facing with a large number of images with the fast development in the social image sharing media. In this situation, finding the images which can satisfy user preferences is more important. Since users' liking behavior has not a certain manner, it can be modeled by fuzzy logic because fuzzy logic is a decision-making system based on degrees of truth and deals with vague and imprecise information. Therefore, in the present study, we proposed a fuzzy inference system based on user preferences, consisting of a set of rules according to distinguishing features for each user. Also together with fuzzy inference system, this method is proposed based on feature extraction and feature selection in order to predict the images which are liked by different users. In the proposed method, the features from user preferred images are extracted. Then, feature selection is used to determine the most discriminant features for each user, and accordingly fuzzy inference system is employed to learn user preferences. We used fuzzy clustering to generate soft clusters and fuzzify user preferences. Then, this system is used to predict images that should be recommended to the user. Experimental results on datasets from Flickr and Pinterest indicated that the proposed method can significantly improve Recall@k and Precision@k for personalized image recommendation compared to some related works.