چكيده به لاتين
Offering products and services related to customer needs have always been the main challenge of recommender systems. Studies indicated that existing tourism recommender systems, according to the good dominant logic, offering products and services based on the users’ functional needs. while based on the service dominant logic, value proposition is co-created in intraction with customers. Based on the reviewed studies, although previous systems have utilized social network analysis and sentimental analysis method to analyzed implicit feedback of users; but the process relationship between the user behavior and recommender system is less discussed. Therefore, in this dissertation, considering process nature of users’ click stream as implicit feedback in tourism platform, we propose a novel collaborating filtering recommender system.
To this end, at first, the data of Expedia tourism platform were collected and preprocessed, then the data enrichment was performed using two data enrichment methods, index-based encoding and frequency-based encoding algorithm. Then, focused on predictive process monitoring technique, ensemble tree-based classification methods, random forest and gradient boosting have been developed using enriched data. Finally, the necessary evaluations were performed, the accuracy of the resulting model is significantly better compared to the based models, and lead to 37% improvement in the accuracy of the predicted model. The results indicate that combining the two approaches of predictive process analysis and recommending systems can provide better and more accurate suggestions for the performance of predictive model development.
