Improving Hotel Search Autocomplete in Online Travel Agent (OTA) Mobile Apps Using Elasticsearch and Learning to Rank

The use of smart search autocomplete for accommodation discovery within Online Travel Agent (OTA) platforms is essential, as it significantly influences the user experience when utilizing these services. This paper presents a method to improve search autocomplete for accommodation discovery in OTA mobile apps using a Learning to Rank (LTR) model. The LTR model is implemented as the core module in Elasticsearch's Query Auto Completion (QAC) service. The proposed method uses a two-step process for generating autocomplete results, involving candidate filtering and final result generation based on the LTR prediction score. This study demonstrates the effectiveness of the proposed method in enhancing the average click-through position by 7.74% for queries with two to four characters, resulting in a 6.46 % increase in the number of items clicked in the first position and a reduction of the average time required for users to click on autocomplete results by 33 %. The findings that we conclude in this extensive study could be used to inform and provoke the work of researchers and engineers in the field of search autocomplete by providing insights on how to improve user experiences through the implementation of LTR techniques in Elasticsearch.