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 Analyzing Topic Transition based on a Search Behavior Driven Using a Markov Model

IP.com Disclosure Number: IPCOM000249789D
Publication Date: 2017-Apr-05
Document File: 5 page(s) / 581K

Publishing Venue

The IP.com Prior Art Database

Related People

Hongbo Deng: INVENTOR [+5]

Abstract

A method and system is disclosed for analyzing topic transition based on a user's search behavior driven using a markov model. The markov model is used to simultaneously learn multiple topic transition rules and store search behaviors of users. Further, a specific search behavior pattern can be identified and associated with a distinct topic transition rule.

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 32% of the total text.

A Method and System for Analyzing Topic Transition based on a Search Behavior Driven Using a Markov Model

Abstract

A method and system is disclosed for analyzing topic transition based on a user’s search behavior driven using a markov model.  The markov model is used to simultaneously learn multiple topic transition rules and store search behaviors of users.  Further, a specific search behavior pattern can be identified and associated with a distinct topic transition rule.

Description

Disclosed is a method and system for analyzing topic transition based on a user’s search behavior driven using a markov model.  The markov model is used to simultaneously learn multiple topic transition rules and store search behaviors of users.  Further, a specific search behavior pattern can be identified and associated with a distinct topic transition rule.  The topic transition detection can be learnt from a single query sequence or multiple query sequences.

Fig. 1 illustrates a method and system for analyzing topic transition based on the user’s search behavior driven using a markov model.

Figure 1

As illustrated in Fig. 1, each query is pre-labeled with certain pre-defined topics and a query sequence is segmented into a series of query fragments based on the topic associated for each query.  Each query fragment is a sequence of successive queries sharing the same topic.  The search behavior of the user is indicated in an nth query fragment with a feature vector dn of size M, and the topic of this query fragment is denoted as ln.  Further, the method and system models the search behavior on the present query fragment and the topic of the present query fragment to determine the topic of the next query fragment.

For example, if k search factors lie in all query fragments, each search factor k determines a distinct probability matrix δk of the topic transitions.  Each query fragment belongs to a single search factor and the markov model suggests the labels of all the query fragments as

Here, the query fragments can be selected using the search factor k associated with a feature vector ωk , and a topic transition matrix δk.

Each query fragment n is assigned a randomly selected search factor membership Yn.  Similarly, each query fragment n can selects a real search factor feature vector of the query fragment dn ∼ Gaussian(ωYn,σ)2.  Further, each query fragment can select a label of the next query fragment ln+1 ∼ Multinomial (δYn,ln)3.

According to the method and system illustrated in Fig. 1, the topic transition probability graph δ varies based upon the search factor of the current query fragment, instead of being invariant for all query fragments.  Similarly, topics that users may search in future are jointly determined by both the current search topics of the users and search factors of the current searching behaviors.

For example, a Ph.D. student is searching sports news for having fun. If the time length of the search query fragm...