Semantic Knowledge Discovery for User Profiling for Location-Based Recommender Systems

Abstract

This paper introduces a purposed Location-based Recommender System (LBRS) that combines sentiment analysis and topic modelling techniques to improve user profiling for enhancing recommendations of Points of Interest (POIs). Using additional feature extraction, we built user profiles from a Foursquare dataset to evaluate our model and provide recommendations based on user opinions toward venues. Our combined model performed favourably against the baseline models, with an overall improved accuracy of 0.67. The limitations were the use of one dataset and that user profiles were constructed using predicted emotions extracted as features from review data with topic modelling, rather than literal user emotions. Nevertheless, this provides a step forward in user profile and emotion scoring, contributing further to the development of LBRS in the Tourism domain.

A Place Recommendation Approach Using Word Embeddings in Conceptual Spaces

“…Zhao et al [31] have combined semantic, textual, and location information to train a deep learning model with the aim of identifying appropriate jobs for users. Tao et al [32] have combined sentiment analysis and topic modeling techniques to enhance the recommendation of places of interest (POIs). Bafna et al [33] have used ontologies to compute semantic similarities between terms in news.…”