Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

A Method and System for Optimizing Vertical Search Results based on User Feedback

IP.com Disclosure Number: IPCOM000220230D
Publication Date: 2012-Jul-26
Document File: 3 page(s) / 34K

Publishing Venue

The IP.com Prior Art Database

Related People

Sudarshan Lamkhede: INVENTOR [+5]

Abstract

A method and system for optimizing vertical search results based on user feedback is disclosed. The method and system utilizes the user feedback to identify Vertical Search Engines to be queried and to sort and present the results obtained from the queried VSEs.

This text was extracted from a Microsoft Word document.
This is the abbreviated version, containing approximately 35% of the total text.

A Method and System for Optimizing Vertical Search Results based on User Feedback

Abstract

A method and system for optimizing vertical search results based on user feedback is disclosed.  The method and system utilizes the user feedback to identify Vertical Search Engines to be queried and to sort and present the results obtained from the queried VSEs.

Description

Disclosed is a method and system for optimizing vertical search results based on user feedback.  The user feedback includes one or more of, implicit user feedback and explicit user feedback.  Examples of implicit user feedback include, but are not limited to, clicks, page views and mouse movements.  Examples of explicit user feedback include, but are not limited to, responses to questionnaires and surveys.  Given a search context, the method and system utilizes the user feedback to identify which Vertical Search Engines (VSEs) should be queried and to determine which results obtained from the VSEs should be presented.  In addition, the method and system utilizes the user feedback to determine how a Search Engine Results Page (SERP) should be composed.

Assuming K VSEs are to be federated in addition to a traditional Web Search Engine (WSE) and that a final result has to be presented in a list of size N in the descending order of relevance/usefulness, the objective of the method is to optimize the entire set of actions taken for the SERP.  In a scenario, the actions include actions performed by a federation module such as putting results at any of the N positions.

The method and system utilizes the user feedback to optimize the entire set of actions.  Initially, the method defines a user satisfaction metric quantitatively for collecting the user feedback.  Thereafter, a runtime system is instrumented to record search context, content, user activity and feedback.  Subsequently, over a predefined time period a small percentage of random queries are exposed to randomly taken actions from the set of all possible actions.  The possible set of actions is determined by the results returned by each of the VSEs, page layout and other business rules.  All such actions have a known, non-zero probability of being taken and this probability is either recorded or can be computed exactly using the logged data.  After enough data has been gathered machine learning models are trained to identify which VSEs should be queried, which results should be shown and how the SERP should be composed based on the features computed from the logs.  The models' responses are obtained at runtime for the incoming queries by computing and/or retrieving the same features.  These responses are then transformed into appropriate execution steps and are used to compose the final result set.  Thus, the information gathered through exploration is exploited at runtime to make optimal search federation decisions.

In a scenario, the method quantifies user satisfaction to obtain the user satisfaction metric for a...