DOI: 10.5176/978-981-08-6308-1_27
Authors: Gokhan Capan, Alper Kursat Uysal, Ozgur Yilmazel
Abstract:
Searching is very important in an e-commerce application, especially if its content is generated by users. A query recommendation system, which recommends new queries to a user after his initial query, plays a key role for shortening the search session by guiding the user to reach the product they need. Because the amount of potential queries is nearly infinite, query recommendation can be considered as an information filtering process, which may either be content-based filtering or collaborative filtering. In this paper we describe our hybrid method for performing scalable, fast responding query recommendation system, which is a combination of content-based and collaborative filtering techniques. We show how the method has been applied to a set of user log data collected from an e-commerce site by creating a model in two-months of query logs and evaluated the model using a separate set of query logs of one- month period. Finally, since (1) most of the query sessions are short, (2) the goal of the query recommendation system is keeping the session short, and (3) tradition evaluation metrics suffer from the lack of rated data; we introduce a new evaluation metric for query recommendation. The new metric returns closer results to human-based evaluation which is an advantage over existing evaluation measures. Results obtained from the evaluation show that the system decreases the search path, and allows users to reach products with fewer clicks.
