Machine Learning Techniques for Personality-Based Multimedia Recommender Systems

An individual’s personality can be defined as the set of attributes, feelings, and ways of thinking that, in combination, make her unique [2]. The most widely used model to represent it is the five-factor model [3], a multifactorial approach so named for considering the people’s personality consisting of five traits: Openness (to experience), Conscientiousness, Extraversion, Agreeableness, and Neuroticism (the acronym OCEAN is often adopted). The reason why such a model is so widespread also lies in its suitability for being quantitatively measured (i.e., numerical values for each of the factors). Recommender systems help online users find relevant content by suggesting items (i.e., products and services) of potential interest to them [5]. Numerous studies (see, for instance, [1,4,7]) have shown that taking into account the target user’s personality in the recommendation process can provide significant benefits in terms of performance. Our research activities focused on the study of the correlation between (a) visual features of high and low level extracted from images [6] and video [8] and (b) users’ psychometric traits, in order to produce personalized suggestions for mitigating the multimedia information overload. In this talk, we will present our experience and learned lessons in harnessing machine learning techniques (support vector machines, convolutional neural networks, regression techniques, etc.) for this task.