Recommending High Utility Queries via Query-Reformulation Graph

Abstract

Query recommendation is an essential part of modern search engine which aims at helping users find useful information. Existing query recommendation methods all focus on recommending similar queries to the users. However, the main problem of these similarity-based approaches is that even some very similar queries may return few or even no useful search results, while other less similar queries may return more useful search results, especially when the initial query does not reflect user's search intent correctly. Therefore, we propose recommending high utility queries, that is, useful queries with more relevant documents, rather than similar ones. In this paper, we first construct a query-reformulation graph that consists of query nodes, satisfactory document nodes, and interruption node. Then, we apply an absorbing random walk on the query-reformulation graph and model the document utility with the transition probability from initial query to the satisfactory document. At last, we propagate the document utilities back to queries and rank candidate queries with their utilities for recommendation. Extensive experiments were conducted on real query logs, and the experimental results have shown that our method significantly outperformed the state-of-the-art methods in recommending high utility queries.

Modeling the impact of out-of-schema questions in task-oriented dialog systems

“…To address the limitations of the Frequency-based Selection, mainly that the position of a question is not considered (meaning how far it is asked from the success state), we model the user behavior when interacting with a chatbot system. Previous studies have shown the effectiveness of Markov Chains in modeling and explaining the user query behavior (Jansen et al 2009), reformulating queries during interactions with systems via absorbing random walk for both web and virtual assistants (Wang et al 2015;Ponnusamy et al 2020), and modeling query utility (Zhu et al 2012). These approaches have been proven to be highly scalable and interpretable due to intuitive definitions of transition probabilities.…”

Feedback-Based Self-Learning in Large-Scale Conversational AI Agents

“…Studies have shown that Markov processes can be used to explain the user web query behavior (Jansen, Booth, and Spink 2005), and Markov chains have since been used successfully for web query reformulation via absorbing random walk (Wang, Huang, and Wu 2015), and modeling query utility (Xiaofei Zhu 2012). We here present a new method for query reformulation using Markov chain that is both highly scalable and interpretable due to intuitive definitions of transition probabilities.…”