Conversational CaseBased Reasoning Shruti Bhandari Overview Concept Rule

Conversational Case-Based Reasoning Shruti Bhandari

Overview • • • Concept Rule Based Systems Representation Problem Solving Process Challenges Applications

Case Based Reasoning • • • Direct Reuse of prior knowledge Cases and Case Base Retrieve Reuse Revise Retain

CBR 3 Approaches to CBR • Textual Approach • Structural Approach • Conversational Approach

Textual Approach • • Cases recorded as free text Large collection of documents Easy Case Acquisition Keyword Matching Syntactic Retrieval Complexity Example: Frequently Asked Questions

Structural Approach • Case Represented according to vocabulary • Assigned values to predefined attributes • Partially filled query description • Example: Sales Support for Electronic Devices

CCBR • • • Pioneered by Inference Corporation Interactive Problem Assessment Incremental Approach Solutions available during conversation A-priori knowledge not required

CCBR (contd) • • • Customer/Agent Conversations List of questions No Domain Model No Structure Domains of high volume of simple problems • Example: Call Center for Printer Problems

CCBR vs Rule Based System • • • Problem solving method Learning from experience Complexity of systems Scaling Cost

Case Representation • Problem Cp=Cd+Cqa – Description Cd – Specification Cqa • Solution Cs={Ca 1, Ca 2…}

Steps in CCBR • Input of problem description Qd • Computation of similarity s(Q, C) • Display of solutions of top ranked cases, Ds and unanswered questions, Dq • Selection by user • Re-computation of similarity • Successful/Unsuccessful Termination

Generic CCBR problem solving process

Generic Algorithm

Component Reuse using CCBR • • Component based Software Development Component Retrieval Different Methods for Retrieval Assumptions

Conversational Component Retrieval Module (CCRM)

Parts of CCRM • • • Knowledge Base New Case Generating Module Knowledge Intensive CBR module Component Displaying module Question Generating and Ranking Module Question Displaying Module

Challenges • Case Authoring • Dialog Inferencing • Expanded Applicability

Case Authoring • Art of designing good libraries • Design guidelines • 3 phase revision of cases

Dialogue Inferencing • Lack of Intelligence • Challenges – Input Size – Comprehensibility – Maintenance

Example of Dialogue Inferencing

Na. Co. DAE • • • To address the challenges HICAP Text Processing Question Ranking Case Ranking

Text Processing • Problem description and User Input are canonicalized • Nouns identified • Similarity Calculation

Question Ranking • Frequency Calculation • No need of information gain

Case Ranking • Similarity Calculation score(Q, C) = same(Qqa, Cqa)-diff(Qqa, Cqa) |Cqa| • Bias Control

Applications • Maintenance and Repair of Complex systems like aircrafts, trucks etc • CREEK • Na. Co. DAE • HICAP • Expert Clerk

Final Remarks • CCBR – interactive and incremental approach • Main components of CCBR • Future of CCBR • Our Project

References • Conversational Case-Based Reasoning – David Aha, Leonard Breslow, Hector Munoz-Avila, Sept 1999 • Experience Management – Ralph Bergmann • Conversational Case-Based Software Reasoning in Reuse – Mingyang Gu

THANK YOU!!!