Report on ICCBR'99

July 27 - 30, Monastery Seeon, Munich, Germany.

In a relaxed summer mood at the lovely Seeon Monastery, the worldwide CBR community assembled for the ICCBR-99. A well filled program with up to fifteen talks a day expected the above hundred participants on a screened peninsula in a Bavarian lake near Chiemsee. Now and then, some allowed themselves a little break for swimming in the warm lake and escaping the enclosure of the cloister. The pieceful atmosphere of the buildings influenced the talks and discussions. Despite of the mostly professional presentations, the discussion parts and the coffee breaks have been used busily for exchanging inspirations and ideas.

Brigitte Bartsch-Spoerl and Wolfgang Wilke as chair persons guided us through a well attended industry day.

Ian Watson gave examples of some US-American applications (Lockheed, Compaq, and Broderbund) that showed how CBR can successfully be commercialised. He stated that until now case-based retrieval was useful without the other phases of the CBR cycle and that the future of CBR will be on-line. As one main success factor of some applications, Watson considered the fact that CBR supports unpopular jobs like calming down angry customers.

In further talks, commercial applications in the areas hotline support and e-commerce had been presented. Two issues mainly attracted my attention: - The introduction of CBR at Nuon, a Dutch energy provider, came along with business re-engineering of the whole customer service. So probably, they could create an optimal environment for the usage of a CBR system. - 'Let on the Net', an on-line market for the rental of houses and apartments, made Hooke & MacDonald to the market leaders in the new homes sector in Ireland. Joining case-based eBusiness greatly enhanced their corporate image.

In a concluding talk on CBR in practice, Brigitte recommended us not to forget some psychological aspects like recruiting the right team and keeping all persons on board.

One of the highlights of ICCBR-99 has been the workshop session. Four workshops were scheduled in parallel:

(1) WS1 considered business processes involved in knowledge acquisition, in maintenance, and organizational impacts on/of CBR systems. After the talks, we had animated discussions concerning the allegorical terms 'life-cycle' and accordingly 'death' of a case. It has not been clear how to determine useless or old knowledge and how to enable people to throw things away. We agreed that the usefulness of a case base/case depends on three factors: the goal of the CBR system, the context (e.g., whether it is the era of Win 3.11), and the competence/way of thinking of the user. In future, the CBR community should develop monitoring tools or other mechanisms for the detection of obstructing knowledge. In the second part of our discussion, we crystallized only one impact of CBR systems on organizations, namely reducing dependency on single experts. However, a company should keep its experts for acquiring new cases. (We described a case base as a black box for the users, which is transparent for the experts and shouldn't become completely white for the users. When the users know the case base inside out, it is useless.) So, the usual interactions between CBR and business processes of an organization are still unknown and many questions are open. An exception might be a strongly organised company, where the return on investment of a CBR project could be estimated.

(2) WS2 dealt with strategies for building and maintaining corporate memories. The participants worked out a five-step methodology: 1. set up/structure the corporate memory 2. populate the system (explicate tacit knowledge) 3. retrieve 4. utilize (integrate it into existing processes, give active support by the system) 5. maintain (defined review, multiple authorship, structured user feedback)

(3) WS3 handled the formalisation of adaptation. The contributions presented real applications with consistent proposals. In the spectrum of representation frameworks, adaptation algorithms (discrepancy reduction, adaptation operators, similarity paths), and user interaction, the participants need further research. They thereby agreed with most of the conclusions of the invited talk of Boi Faltings (see below).

(4) WS4 presented and discussed hybrid CBR systems. Different combinations with different degrees of coupling had been presented and discussed. So, the set of presented approaches should have been quite hybrid :-). As the main issue for advance in this area, the participants need methods for the transformation of knowledge.

Intelligent Systems in Electronic Commerce (Padraig Cunningham): New models of selling have been introduced by Padraig Cunningham. The most interesting aspect is the exploitation of available data on the user which is already done by PTV for generating personalised TV guides. The user has to specify some attribute values for solving the priming pump problem, i.e. the lack of observations. Automatic collaborative filtering which is a technique clustering users into virtual communities can be combined with case-based user profiling. The interaction with users can be smoothed by using historic profiles within a case retrieval net. As a succeeding problem, I suppose that many users will not trust this mechanism and switch it off for security scruples. Maybe, demonstrating transparency like explicitely allowing cookies will remedy this.

Case Adaptation - New Perspectives for CBR (Boi Faltings): In an intentionally provocative way, Boi Faltings stated that adaptation is the defining feature of CBR, because adaptation tools actually reason while retrieval could also be done by information retrieval tools. As expected, this met with opposition in the discussion. Boi recommended two essential concepts for adaptation: situated problem-solving and interactive intelligence, i.e. adaptation under user control and with users' help by adding knowledge. Additionally to the conclusions already mentioned in the section on Workshop 3, he postulated the integration of adaptation with existing software, e.g., word processors and drawing tools, even if this leads to conflicts with powerful representations.

Text-Based CBR (Kevin Ashley): Kevin Ashley outlined several textual CBR applications (FAQFinder, FAllQ, SPIRE, SMILE) and wondered how much of the knowledge acquisition could be done automatically by information extraction or machine learning. He has been expecting some chances in stable domains. Kevin Ashley concluded, that NLP is quite frustruating but sometimes helpful for CBR, that we should spend more time for evaluation, and that domain-specific and other knowledge is leveraging for TCBR which has hybrid connections to information retrieval, information extraction, and machine learning.

We heard a variety of very interesting, exciting talks on topics like autonomous information gathering before retrieval processes (Q. Yang) which might be influenced by ideas of active databases, or like case retention strategies presented by H. Munoz-Avila, on image processing etc. I felt, that maintenance and integration issues are most fashioned this year. All papers can be found in the proceedings (LNAI 1650 at Springer Verlag). The award for the best research paper got Steffi Brueninghaus and Kevin Ashley; the award for the best application paper won David Leake et al. Congratulations to the winners!

With a clever combination of the poster session and a buffet, everybody had leisure enough to inspect all posters and to talk to the authors. Especially the musicians in the community liked the case-based saxophon experiments of Arcos et al.

Remarkable are the opulent meals we enjoyed in the monastery. So, even the last presentation on Friday was well attended followed by the farewell lunch. A culinaric highlight has been the conference dinner with bavarian musicians performing a 'Schuhplattler' with the friendly support of the 'spoon-percussionists' M. Richter and K. Ashley. As a dessert, we got served the singing INRECA team under conduction of Michel Manago.

A chess tournament in the 'Klosterstueberl' with live music of some case-based reasoners gave us the possibility to try the Bavarian beers far into the night.

In the community meeting, the location of the next ICCBR has been announced: ICCBR 2001 will take place in Vancouver.

At least, I want to thank Christine Harms very much for the professional local organisation of ICCBR-99, as well as Ralph Bergmann, the conference chair, with the two program chairs Klaus-Dieter Althoff and Karl Branting for the three-and-a-half-day conference program. They did great work in preparing this satisfying and inspiring meeting.