Browse Prior Art Database

Method and System for Extracting Useful Problem Resolution Information from Historical Problem Resolution Knowledge for Efficient Service Delivery

IP.com Disclosure Number: IPCOM000202361D
Publication Date: 2010-Dec-14
Document File: 3 page(s) / 138K

Publishing Venue

The IP.com Prior Art Database

Abstract

A method and system for extracting useful problem resolution information from historical problem resolution knowledge, such as, public discussion forum threads and tickets data for efficient service delivery is disclosed.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 45% of the total text.

Page 01 of 3

Method and System for Extracting Useful Problem Resolution Information from Historical Problem Resolution Knowledge for Efficient Service Delivery

Disclosed is a method and system for extracting useful problem resolution information from historical problem resolution knowledge, such as, public discussion forum threads and tickets data for efficient service delivery. The method and system formulates the extraction of useful problem resolution information as a sequence labeling problem. In order to label a portion of information in historical problem resolution knowledge, a learning model is developed based on Conditional Random Fields (CRF). The learning model is trained with manually labeled data. Subsequently, the learning model is used to correctly label new historical problem resolution data and extract the relevant problem resolution information. The method and system then summarizes the extracted useful problem resolution information to generate a clean problem description and a list of resolution steps to be followed to solve a problem. The method and system thus suggests a well defined list of resolution steps that are found automatically from a historical knowledge to solve the problem.

Figure 1 illustrates an exemplary public discussion forum thread that may be used for extracting useful problem resolution information.

(This page contains 00 pictures or other non-text object)

Figure 1

The extraction of useful problem resolution information is modeled as a sequence labeling problem. In other words, each portion of the historical problem resolution information needs to be identified as belonging to a predefined category such as problem, resolution, successful resolution, feedback from the problem originator etc. Accordingly, a label corresponding to the predefined category is assigned to each portion of the historical problem resolution information. The assignment is performed by a learning model based on Conditional Random Fields (CRF) which is trained on

1


Page 02 of 3

manually tagged information. The manually tagged information includes a training sequence (input data) x = x1, x2, x3 .. xn and corresponding set of tags (label sequence) y= y1, y2, y3 .. yn, which were manually provided. Subsequently, the learning model is used for assigning labels to new data.

For example, labels may be assigned by the learning model to annotations of discussion threads. Labels may indicate the actual problem being addressed in the discussion thread. For example, when a discussion thread includes text such as, "As soon as we added SSL encryption it did not work anymore," it may be inferred that the discussion thread talks about a problem description or a related error information. Accordingly, a label, "ProblemRelatedInfo" may be assigned to that particular sequence in the discussion thread. Similarly, a label, such as "ProblemEnd" may be assigned to a sequence based on detection of request for help that is usually present at the end of a...