User Emotion and Behaviour as Contextual Parameters with Contextual Pre-Filtering and Contextual Modeling in Travel Recommendation

Abstract

Emotion-based recommendation is widely used in many recommendation domains and has been discussed in context-aware recommendation. Adapting emotion is mainly followed by the notion of the different role of emotion in the process of recommendation as users interact with the system. Ample researches have been suggested in the domains of music and movie recommendation but very few studies are found on travel recommendation. Thus, the proposed study used emotion and user behavior as contexts and used pre-filtering and contextual modeling approaches to find the effectiveness of travel destination recommendations. In the filtering approach, the system recommendation was implemented in contextual pre-filtering paradigm and used the contextual information to select the most relevant item and user data for generating recommendations by using item-item collaborative filtering. Top five destinations were generated as recommendations to trace the effectiveness of the recommendation by using Loglikelihood similarity and Simple Weighted Average predictive rating calculation algorithm. In contextual modeling, CANDECOMP/PARAFAC(CP) Tensor Factorization model was adapted, which used ratings from M users for N items under Q types of contexts as a threedimensional tensor and generated the top five recommendations for each context. As the contexts, both emotion and user behavior details were incorporated in recommendation engine for the comparison. A new corpus with emotion context for place recommendation was developed by using Semantic Analysis techniques due to the lack of properly recorded dataset and the derived dataset used in the implementation. In the process of deriving emotion tags, we used the text reviews collected from TripAdvisor and defined an emotion tag for each selected destination based on lexicon-based semantic classification. Both approaches with contexts outperformed with the selected contextual parameters and results of tensor factorization approach with user emotion and user behavior proved higher effectivity in tourist destination recommendation compared to other approaches (Mean Average Precision = 81.59 %). Our study focused on a challenging field, such as tourist destination recommendation while selecting the emotion and user behavior as contextual parameters and the selected contextual parameters proved user satisfactions towards the recommendation generated by the system.