Learning user reformulation behavior for query auto-completion

Abstract

It is crucial for query auto-completion to accurately predict what a user is typing. Given a query prefix and its context (e.g., previous queries), conventional context-aware approaches often produce relevant queries to the context. The purpose of this paper is to investigate the feasibility of exploiting the context to learn user reformulation behavior for boosting prediction performance. We first conduct an in-depth analysis of how the users reformulate their queries. Based on the analysis, we propose a supervised approach to query auto-completion, where three kinds of reformulationrelated features are considered, including term-level, querylevel and session-level features. These features carefully capture how the users change preceding queries along the query sessions. Extensive experiments have been conducted on the large-scale query log of a commercial search engine. The experimental results demonstrate a significant improvement over 4 competitive baselines.

Selectively Personalizing Query Auto-Completion

“…One-query and no-click sessions are both excluded in our experiments as not enough search context is available. In addition, we follow previous QAC work and adapt a commonly used evaluation methodology in QAC [3,8,13] by only keeping cases where the final submitted query is included in the top N query completions returned by the MPC approach, For comparison, the following baselines are selected: (1) the most popular completion (MPC) method, which ranks query candidates by their frequency [1]; (2) a personalized QAC approach based on session context with a fixed tradeoff λ = 0.5 in (2), denoted as P-QAC [3].…”

“…Most previous work on query auto-completion [1,3,8,13] only considers the typed prefix for generating a list of query completions, ignoring the potential signal hidden in the typed prefix for personalization. However, the typed prefix normally reveals a strong clue for inferring a user's personal query activity, such as query expansion and query repetition, etc.…”

“…In general, personalized lists of query completions produced by a generic QAC ranking approach are obtained by distinguishing individuals and their contexts, either based on their search history [1,3,8] or on their profiles [13]. Such personalization strategies have been proven to be effective as evidenced by a series of personalized query auto-completion approaches [1,3,8,11,13].…”

Large-scale Generative Query Autocompletion

“…Di Santo et al [11] collate a series of other QAC models that are dened in the literature. A variety of other approaches have been since considered, including time-sensitive QAC [5,7,30,32] and contextual and demographic-based QAC [2,19,22,23,25,29]. For evaluation, Mean Reciprocal Rank (MRR) [8] has become the de facto measure [6].…”

Analyzing the Usefulness of the User’s Browser History for Generating Query Suggestions

“…Another paper where click through data was analysed and used for ordering the suggestions is [14], where they demonstrate that the higher a suggestion is present in a suggestions list, tends to attract more click. In [6] Jiang et al are reformulating the query by analysing how users previously reformulate their queries then adding words in the query and define a set of features which were applied using the LambdaMart [12] learning to rank algorithm. Others [13] have tried to apply probabilistic models, like Markov Process to predict what user's query will be immediately after he starts typing.…”