Recommendations on Streaming Data: E-Tourism Event Stream Processing Recommender System

Abstract

The Association for Computing Machinery ACM recommendation systems challenge (ACM RecSys) [1] released an e-tourism dataset for the first time in 2019. Challenge shared hotel booking sessions from trivago website asking to rank the hotels list for the users. Better ranking should achieve higher click out rate. In this context, Trivago dataset is very important for e-tourism recommendation systems domain research and industry as well. In this paper, description for dataset characteristics and proposal for a session-based recommender system in addition to a comparison of several baseline algorithms trained on the data. The developed model is personalized session-based recommender taking into consideration user search preferences. Technically, paper compare between six different models vary from learning to rank, nearest neighbor and popularity approaches and compared results with two benchmark accuracy. Taking into consideration the ability to deploy model into production environments and the accuracy evaluation based on mean reciprocal rate as per challenge guidelines. Our winning experiment is using one learning to rank model achieving 0.64 mean reciprocal rate compared to 37 model achieving 0.68 by ACM challenge winning team [2].