A hybrid context-aware approach for e-tourism package recommendation based on asymmetric similarity measurement and sequential pattern mining

Personalized travel package recommendation for tourists is an important task in the e-Tourism field. With the excessive growth of data over the internet, it has become difficult to find relevant information among all of the Points-of-Interest (PoIs) for tourists. Recommendation Systems (RSs) are the essential tools to tackle the information overload problem and suggesting relevant items to the users. But, due to differences in users' interests, dynamic contexts and sequential travel patterns in their behaviors, the existing recommender systems could not provide accurate recommendations to the users. Moreover, due to the lack of sufficient historical data about users in tourism area, cold start and data sparsity problems lead to more challenges for recommending precise and reliable suggestions. In this paper, a personalized travel package recommendation approach is proposed to generate dynamic recommendations to the users by taking into account the multidimensional data of time, location, users' implicit ratings, tourists' characteristics and their sequential travel movement patterns based on the geo-tagged photos. Therefore, we combine four different recommendation methods including, Collaborative Filtering (CF), Context-Awareness (CA), Demographic-Based (DB) and Sequential Pattern Mining (SPM) to ameliorate the quality of the suggestion results. Further, a dynamic contextual modelling approach is used to incorporate contexts information into the recommendation procedure to cope with the context pre/post filtering problems, and provide dynamic travel recommendation to the tourist based on his/her current contexts information. Furthermore, this approach extends the CF method by using multidimensional data and an asymmetric similarity scheme for score function to alleviate the limitation of CF and improve the accuracy of predicted recommendations. Moreover, the proposed approach recommendations will be optimized by applying the personalized PoIs to the sequential travel patterns and personalizes the travel package recommendations for each user based on his/her interests. The experimental results based on Flickr dataset illustrate the effectiveness and quality of the proposed approach over state-of-the-art approaches under cold-start and data sparsity conditions by contextualizing users' interests, utilizing demographic information, applying the asymmetric similarity scheme and using the sequential pattern mining.