Browse Prior Art Database

A Method and System for Improving Search Assist Relevance using a Two-Dimensional Click Model

IP.com Disclosure Number: IPCOM000249790D
Publication Date: 2017-Apr-05
Document File: 4 page(s) / 85K

Publishing Venue

The IP.com Prior Art Database

Related People

Anlei Dong: INVENTOR [+6]

Abstract

A method and system is disclosed for improving search assist relevance using a two-dimensional click model. The two-dimensional click model extracts a user's click behavior while the user interacts with a search assist system and utilizes the user's click behavior to improve the search assist relevance.

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 45% of the total text.

A Method and System for Improving Search Assist Relevance using a Two-Dimensional Click Model

Abstract

A method and system is disclosed for improving search assist relevance using a two-dimensional click model.  The two-dimensional click model extracts a user’s click behavior while the user interacts with a search assist system and utilizes the user’s click behavior to improve the search assist relevance.

Description

Disclosed is a method and system for improving search assist relevance using a two-dimensional click model.  The two-dimensional click model extracts a user’s click behavior while the user interacts with a search assist system and utilizes the user’s click behavior to improve the search assist relevance.

In an embodiment, the method and system collects a high-resolution query auto-completion (QAC) query log that records each and every keystroke entered by the user in a QAC session.  The keystrokes recorded in each QAC session may include, but not limited to, a final submitted query, keystrokes entered by a user, a timestamp of a keystroke, top 10 suggested queries to a prefix, an anonymous user ID and the like.  The QAC session also records previous queries submitted by the user to combine the high-resolution QAC query log with traditional query log for enabling many new researches in the QAC session.  Here, the traditional query log can be the user’s demographics and the user’s query history.  Thus, all the lists of suggested queries are utilized to improve the relevance of ranking the QAC session.

Further, in response to the high-resolution QAC query log collected from the user, the method and system leverages the QAC query log for modeling user behavior in the QAC session.  The user behavior modeling in the QAC session can be verified by using a horizontal skipping bias assumption and a vertical position bias assumption for predicting relevant search queries in the QAC session.

The horizontal skipping bias assumption is used to identify the queries skipped by the user while selecting a list of suggested queries because of fast typing speed, too deep to look up the query and the like.  If the skipped query is found to be most relevant to the prefix of the search query entered by the user and listed in top 3 relevant queries, then the user is stopped from clicking final prefix unless the user examines the skipped query.  Thus, the user cannot skip from missing the relevant query suggested in the QAC session.

Similarly, the vertical position bias assumption is used in the QAC session for analyzing a same set of QAC sessions and computes clicks distributed according to the queries positioned in a final suggestion list and a final prefix length.  The positioning of clicks distributed can vary from desktop to mobile.  The vertical position bias suggests the method and system to improve the relevance queries estimated if the queries are clicked on lower ranks.

In another embodiment, the method and system includ...