Research Information System
2022 World Congress on Engineering, WCE 2022, London, England, 6 - 08 July 2022, vol.2244, pp.64-69
© 2022 Newswood Limited. All rights reserved.Successfully predicting rarely occurring events in large systems can be extremely valuable in various scenarios, such as fraud detection, quality control, sales prediction, etc. In tourism, predicting connections between hotels via their similarity scores can form the basis of a hotel recommendation engine. In this work, we propose a link prediction framework for such an application. This framework first extracts a hotel-to-hotel network from hotel-customer raw data sets. Then, it applies various link prediction approaches. Besides employing well-known node similarity metrics such as Adamic Adar, Jaccard Coefficient, and Preferential Attachment, we also contribute to developing their weighted versions. These six metrics are executed in the basic supervised task of link prediction. The results are evaluated by precision and AUC. In our experiments, we used two novel data sets from the tourism sector: SeturTech and Otelpuan. The results demonstrate that the proposed weighted Adamic Adar returns the most accurate link predictions.