king@rd1632.Dayton.NCR.COM (James King) (03/01/89)
The following is an abridged version of a "survey", or call for information, which has been sent over USmail and I am now sending this out to the net to hit anyone that has an interest and that isn't on the mailing list. ------------------------- cut here -------------------------- A Survey on Similarity Important Address for Return James A. King NCR Corporation WHQ-5E 1700 S. Patterson Dayton, OH 45479 EMAIL responses are welcome: j.a.king@dayton.ncr.com j.a.king%dayton.ncr.com@relay.cs.net With questions phone: (513)-445-1090 (days) (317)-478-5910 (nights) Any related papers, dissertations, notes, software, etc. are welcome. All information will be collected and reviewed with the purpose of producing a report on similarity. This survey is part of an effort to produce Case-Based Reasoning mechanisms which is under the direction of Robert Simpson, with consulting assistance from Dr. Edwina Rissland and Dr. Janet Kolodner. Please return this survey by March 10th if possible. Acknowledgements to: - Ray Bareiss, Vanderbilt University - and a multitude of respondents to the initial request for participation Introduction (body of text) excluded ... This survey is directed towards the following objectives: - To collect researcher's definitions and approaches for utilizing similarity metrics; - To collect example systems which have made an operational commitment to a particular similarity metric (or suite of similarity judgments); - To collect empirical results and judgments as to the effectiveness of these metrics; - To establish a space of similarity measurements which could be factored by domains, tasks, etc.; - To produce an informative survey report on metrics for similarity. - To report during a panel session at the DARPA sponsored Case-Based Reasoning Workshop to be held May 31 - June 2 in Pensacola, Florida. Thank you for your participation. ____________________________________________________________ Survey on Similarity General Information This survey consists of three parts: I. General Questions II. Survey Questions III. Optional "Open" Questions IV. Request for supporting information ____________________________________________________________ I. General Questions Name: Position: Organization: Address: Phone: EMAIL Address: 1. What is your primary research area? 2. Is similarity assessment an important aspect of your work? 3. In which domains are you applying (and assessing) similarity measurements? 4. What software have you designed, implemented, or directed to be built which involves similarity assessment? On which platforms, and with which languages, or tools, have these systems been constructed? II. Survey Questions 1. In your opinion what makes one case (i.e., an object, situation, or event) similar to another? In other words, what about a new case reminds you a past experience, object, sense, etc.? 2. What forms should measurements of similarity take (e.g., quantitative and qualitative)? How should they be processed? How should quantitative and qualitative methods be combined? 3. How should similarity be assessed when case descriptions are not uniform (i.e., when similar information is provided by non-identical features)? How important is this problem? 4. How much inferential effort is worthwhile for determining the equivalence of nonidentical features? How should a system determine the amount? 5. How should features be weighted with respect to importance during similarity assessment? 6. What commonalties and differences exist between case retrieval (i.e., recalling a potentially similar experience) and similarity assessment? Are these distinct processes? 7. Have you (formally or informally) compared different methods of similarity assessment (e.g., additive vs. multiplicative similarity functions)? III. Open Questions: 1. In which domains are reminding processes and measurements of similarity applicable? (In other words, in which domains is case-based reasoning applicable?) 2. What forms can verification of a reminding process take? 3. What role should statistics (e.g., Bayesian analysis) play in similarity assessment? 4. Are exemplars ground instances of categories defined in terms of observable features or can their features be abstracted? 5. What role should neural networks play in research on similarity assessment (e.g., does this model provide a more compelling explanation of the phenomena being discussed)? IV. Other Information: 1. Please list papers you have written which discuss similarity assessment. If possible, attach copies of these papers or abstracts. 2. Please suggest other researchers (names and addresses) who could provide opinions on this topic. 3. Please make general comments on the survey (e.g., which questions should we have asked?).