A novel multidimensional clustering based dynamic destination recommender system employing optimization techniques

Abstract

As technology continues to advance through intelligent systems built specifically for travelers, the newlinedesire for personal, contextual and meaningful destination recommendations continues to grow. newlineThe conventional Destination Recommendation Systems (DRS), which typically employ a newlinecollaborative, content-based or hybrid algorithmic approach, have many deficiencies. DRS can be newlinetoo reliant on only one dimension of user interest; they restrict user preference representations to newlinethe fixed form; they suffer from poor information density because of limited numbers of newlinerecommendations; they cannot handle situations where data is scarce or where a traveller is not newlinefamiliar with a specific destination (e.g., moving to a new city); and they lack sufficient contextual newlinedata to support the delivery of personalized destination-specific recommendations. To overcome newlinethese drawbacks, this research presents a new Dynamic Destination Recommendations System newline(DDRS) that implements Weighted K-Means Clustering and two innovative Optimisation newlinetechniques: Artificial Bee Colony (ABC) Optimisation Algorithm and Particle Swarm newlineOptimization (PSO) to create accurate, adaptive, multidimensional recommendation services. newlineThe DDRS has compiled a comprehensive set of multi-dimensional data on Indian Travel Places, newlineincluding the geographical coordinates and all types of travel attributes and significance ratings for newlineeach of these places. The users also provide feedback concerning the importance of the various newlinetravel attributes based upon their own personal characteristics. Dynamic Weighting of Feature newlineProperties is a unique system that allows the user to indicate the comparative importance of such newlinetravel attributes as culture, history, nature, adventure, and popularity regarding each of the Indian newlinetravel places the user is interested in visiting. This method of dynamic weighting of feature newlineproperties results in a variation of the K-Means clustering algorithm by enhancing both how newlinecentroids are constructed, and how clusters are establis