A Hybrid Knowledge-Based and Collaborative Filtering Recommender System Model for Recommending Interventions to Improve Elderly Wellbeing

Abstract

A recommender system is an information filtering system that helps users select items that most match their preferences from a vast amount of information available. It has been widely applied in many domains such as e-commerce, healthcare, entertainment and so on. Currently, there are some efforts have been done for recommending interventions to improve elderly well-being in different aspects of successful ageing. However, the recommendations are focused on only a single aspect of successful ageing such as nutrition or health. There are many aspects of successful ageing that should be considered when given interventions to improve elderly wellbeing namely socialization, health, physical, cognitive, nutrition, spirituality and environment. This paper aims to propose a Hybrid Knowledge-Based and Collaborative Filtering (KBCF) recommendation model to recommend interventions based on multiaspects of successful ageing in order to improve the elderly wellbeing. The Knowledge-based (KB) recommendations are generated by consulting a knowledge base which is developed based on knowledge provided by the domain experts. The Collaborative Filtering (CF) approach is applied to find similar users based on elderly profiles generated from the result of assessments done to the elderly. The results of the KB recommendations and the CF recommendations are integrated and ranked to select the final recommendations. The result of experiments conducted using precision, recall and F1 measure shows that the proposed KBCF model outperforms the baseline models which are Basic Search, Collaborative Filtering and Knowledge-based. This result demonstrates the proposed KBCF model provides more accurate and meaningful intervention recommendations for improving the elderly wellbeing in multiaspects of successful ageing.

Comparison between Collaborative Filtering and Neural Collaborative Filtering in Music Recommendation System

“…Collaborative filtering approaches are used widely nowadays. Research by [5,[7][8][9][10][11] implemented CF and showed that CF resulted in high accuracy and suitable for recommendation systems. This approach has for about 2 thousand users, and the accuracy is 80-90% [9].…”

“…The principal in creating a neighbour-based CF recommendation system is to identify user similarity from their preferred items, then select top most similar k users [5,10,11]. Neighbour-based CF provides good recommendations.…”