Browse Prior Art Database

Method and System for Providing a Two-Dimensional Click Model for Query Auto-Completion

IP.com Disclosure Number: IPCOM000240681D
Publication Date: 2015-Feb-18
Document File: 9 page(s) / 600K

Publishing Venue

The IP.com Prior Art Database

Related People

Yanen Li: INVENTOR [+6]

Abstract

A method and system is disclosed for providing a two-dimensional click model (TDCM) for Query Auto-Completion (QAC). Based on two key user behavior observations, the method and system provides the TDCM for modeling the QAC process. The TDCM includes of a horizontal component that captures a skipping behavior, a vertical component that depicts the vertical examination behavior, and a relevance model that reflects intrinsic relevance between a prefix and a suggested query.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 30% of the total text.

Method and System for Providing a Two-Dimensional Click Model for Query Auto-Completion

Abstract

A method and system is disclosed for providing a two-dimensional click model (TDCM) for Query Auto-Completion (QAC).  Based on two key user behavior observations, the method and system provides the TDCM for modeling the QAC process.  The TDCM includes of a horizontal component that captures a skipping behavior, a vertical component that depicts the vertical examination behavior, and a relevance model that reflects intrinsic relevance between a prefix and a suggested query.

Description

Recent Query Auto-Completion (QAC) algorithms incorporate global signals such as query count and personal features.  Currently, click models are trained on query logs with simulated user interaction.  The lack of real user interaction information prevents the click models from further improving the QAC performance.  In information retrieval such as web search, the click models are developed to infer a perceived relevance of the user by explaining position bias.  However, due to the difference between QAC and document retrieval, the click models can not apply to the QAC problem without significant modification.

Disclosed is a method and system for providing a two-dimensional click model (TDCM) for Query Auto-Completion (QAC).  The TDCM understands user behavior in a QAC process.  For example, the QAC process can be considered as a search for relevant queries to a prefix while the retrieval process is a search for relevant documents to a query.  Two basic assumptions for the QAC problem are, one is to address the click bias due to the skipping behavior and the other is to address the click bias on vertical positions.  In skipping bias assumption, a query cannot receive a click if the user did not stop and examine the suggested list of queries, regardless of the relevance of the query.  The skipping bias assumption explains the reason of no one click even when a relevant query is ranked top in the suggestion list, leading to all of the clicks concentrated on the final prefix.  In vertical position bias assumption, a query on higher rank tends to attract more clicks regardless of the relevance to the prefix.  Similar to the click modeling of the document retrieval, the vertical position bias assumption explains the reason of receiving fewer clicks for relevant queries if the relevant queries are ranked in lower positions.

Based on the two basic assumptions defined, the TDCM explains the observed clicks, which include a horizontal model (H model) that explains the skipping behavior, a vertical model (D model) that depicts the vertical examination behavior, and a relevance model (R model) that measures the intrinsic relevance between the prefix and a suggested query.  There are two key user behaviors, as shown in Fig. 1, which includes the skipping behavior and the vertical click position bias behaviors.  The two key user behaviors are prevalent and importan...