A Review of Explanation and Explanation in Case-Based Reasoning

@inproceedings{citeulike:2347388, author = {Doyle, Donal and Tsymbal, Alexey and Cunningham, Padraig }, citeulike-article-id = {2347388}, comment = {Review * Different types of users - Experts - prefer trace "how" (Mao and Benbasat, 2000) - prefer low-level explanations (Ramberg, 1996; Wolverton, 1995) - only use explanations for anomalous situations - Novice - prefer justification "why" - prefer higher-level explanations + background info + low-level explanations (Ramberg, 1996; Wolverton, 1995) - greater need for explanations * user modeling difficult - so people normally use Rules based on stereotypes * Toulmin's Argument Structure (see Fig 2) (Ehninger \& Brockriede, 1960) - data, warrant, backing, qualifier, rebuttal, claim * Swartout \& Moore (1993) aspects of good explanation 1)fidelity 2) understandability 3) sufficiency 4) low construction overhead 5) efficiency * Existing explanation methods - Viewpoints - different rules presented to different users, depending on their roles (e.g. claimant, adjucator) - Satisficing Conclusion-Substantiating Explanation - leave out details that - users cannot understand - would not add to users level of belief - expertise variant - uses a belief threshold point system - each explanation increments belief by certain points - increment may be too small, so leave out - threshold may not be reached, so need to find optimal inclusion of explanations - Simulation Explanation - can handle "blackbox" systems, by simulating user - modules: UI, Simulation Model, User Model, Explanation Model, Natural Language Generator * Current cased - MoCAS (Pews \& West, 1993) (see fig. 6) - multi-level explanations -> Abstractions - Fact abstraction: {fact1,...factN} \~{} fact - Rule abstraction: {rule1 + fact + rule2} \~{} rule - others ... * Usefulness of Explanations - see Cunningham 2003}, keywords = {case-based, explanation, reasoning, review}, posted-at = {2008-02-07 01:44:51}, priority = {0}, title = {A Review of Explanation and Explanation in Case-Based Reasoning}, year = {2003} }

TY - CONF ID - citeulike:2347388 L3 - citeulike-article-id:2347388 TI - A Review of Explanation and Explanation in Case-Based Reasoning KW - case-based KW - explanation KW - reasoning KW - review N1 - Review * Different types of users - Experts - prefer trace "how" (Mao and Benbasat, 2000) - prefer low-level explanations (Ramberg, 1996; Wolverton, 1995) - only use explanations for anomalous situations - Novice - prefer justification "why" - prefer higher-level explanations + background info + low-level explanations (Ramberg, 1996; Wolverton, 1995) - greater need for explanations * user modeling difficult - so people normally use Rules based on stereotypes * Toulmin's Argument Structure (see Fig 2) (Ehninger & Brockriede, 1960) - data, warrant, backing, qualifier, rebuttal, claim * Swartout & Moore (1993) aspects of good explanation 1)fidelity 2) understandability 3) sufficiency 4) low construction overhead 5) efficiency * Existing explanation methods - Viewpoints - different rules presented to different users, depending on their roles (e.g. claimant, adjucator) - Satisficing Conclusion-Substantiating Explanation - leave out details that - users cannot understand - would not add to users level of belief - expertise variant - uses a belief threshold point system - each explanation increments belief by certain points - increment may be too small, so leave out - threshold may not be reached, so need to find optimal inclusion of explanations - Simulation Explanation - can handle "blackbox" systems, by simulating user - modules: UI, Simulation Model, User Model, Explanation Model, Natural Language Generator * Current cased - MoCAS (Pews & West, 1993) (see fig. 6) - multi-level explanations -> Abstractions - Fact abstraction: {fact1,...factN} ~ fact - Rule abstraction: {rule1 + fact + rule2} ~ rule - others ... * Usefulness of Explanations - see Cunningham 2003 AU - Doyle, Donal AU - Tsymbal, Alexey AU - Cunningham, Padraig PY - 2003/// ER -