Abstract
Search systems have played an essential role in improving user experience and information accessibility on the web, allowing users to express their information needs (provided as search queries) and serving users with the results that satisfy those needs. However, a user's search task can be complex and may not be expressed using a single search query, requiring the user to write several queries to fulfill all the aspects of his or her needs. In such scenarios, an intelligent search system would be beneficial to iden-tify and understand the original search task issued by a user and then suggest several search tasks (in a form of key-phrases or short topics) related to the original search task. Aiming to tackle this limitation, this paper proposes a framework for applying several unsupervised learning approaches, including topic modeling and log mining. The results of applying these approaches to large user session data show that, indeed, these approaches would be applicable in search suggestion and task recommendation, reaching a significant improvement over a strong baseline.(c) 2021 The Authors. Published by Elsevier B.V. on behalf of King Saud University. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).