Multi-objective Consensus Clustering Framework for Flight Search Recommendation

Abstract

In order to provide personalized recommendations for travel search queries to online customers, an appropriate segmentation of customers is required using information from the search query. Clustering ensemble approaches have been proposed to address well-known issues of classical clustering methods that each relies on a different theoretical model and can thus identify in the data space only clusters corresponding to this model, and ensemble methods aggregate diverse clustering solutions from dissimilar algorithmic configurations to generate more robust consensus clusters corresponding to agreements between initial clusters. We put forward a new clustering ensemble multi-objective optimization-based framework developed to improve personalized recommendations generated by the flight search engine of the company Amadeus. This framework optimizes diversity in the clustering ensemble search space and finds an appropriate number of clusters automatically without requiring any user input. Experimental results compare the efficacy of this method with other existing approaches on Amadeus customer flight search data in terms of the adjusted Rand index and a business metric defined and used by the company.