Recommender System Incorporating User Personality Profile through Analysis of Written Reviews

In this work we directly incorporate user personality profiles into the task of matrix factorization for predicting user ratings. Unlike previous work using personality in recommender systems, we use only the presence of written reviews by users. Other work that incorporates text directly into the recommendation framework focuses primarily on insights into products/categories, potentially disregarding important traits about the reviewers themselves. By using the reviews to determine the users’ personalities directly, we can acquire key insights into understanding a user’s taste. Our ability to create the personality profile is based on a supervised model trained on the MyPersonality dataset. Leveraging a set of linguistics features, we are able to create a predictive model for all Big 5 personality dimensions and apply it to the task of predicting personality dimensions for users in a different dataset. We use Kernelized Probabilistic Matrix Factorization to integrate the personality profile of the users as side-information. Lastly, we show the empirical effectiveness of using the MyPersonality dataset for predicting user ratings. Our results show that combining the personality model’s raw linguistic features with the predicted personality scores provides the best performance. Furthermore, the personality scores alone outperform a dimensionality reduction of the linguistics features.