DIPT: Diversified Personalized Transformer for QAC systems

Today, with the explosion of information on the web, search engines play a more prominent role in serving users' information needs. Query auto-completion (QAC) is one of the most crucial aspects of search engines that helps users formulate relevant and precise queries based on their information needs. A QAC system generates a list of query candidates according to the user's provided prefix and then updates it with each new keystroke. The existing methods mostly focus on personalizing query candidates to make them revolve around the user's interests and past interactions. Due to the limited number of suggested queries, mere personalization can result in pushing redundant suggestions into the list as well as the exclusion of effective ones. In this paper, we address the diversification task in QAC systems, presenting a novel method called DIPT, a Diversified Personalized Transformer for QAC systems. The proposed method diversifies the suggested queries to include the potential future interests of users in addition to their past interests and interactions. Experimental results on the AOL standard dataset demonstrate the advantage DIPT over state-of-the-art personalized QAC systems in terms of the MRR(Mean Reciprocal Rank) criterion.