Applying a fuzzy aggregator to segment user profiles based on their opinions

In a recommender platform context, we propose a user profiling method based on the ratings and attributes of individuals using the system in their search for an item. This user profiling method allows recommender systems to assist users with finding items suited to their personal preferences. Preferences are derived from topic analysis on the user's previous reviews which he introduced on the platform after experiencing an event and integrated with user provided attributes. We propose a new fuzzy machine learning approach based on a measure of similarity between users considering categorical variables. The similarity measure uses a fuzzy aggregation operator. This proposed method differs from previous similarity measures in three ways. First, the method considers the frequency distribution of each possible value of an attribute within each cluster rather than the frequency of a value in the entire dataset. Second, the method does not view a match simply by a match or mismatch rather it compares the similarity of the frequency distribution of attribute values between two clusters. Third, the method applies a parameter of tolerance which may or may not favor the frequently matched attribute values in a cluster. The model is implemented in a real case dataset of restaurants from the Yelp platform.