A Report to the Club of Rome. Pergamon. Press, 1998. [2] Fusako Kusunoki, Masanori Sugimoto, and Hi- romichi Hashizume. Discovering how other pupils think ...
A Comparative Study of Approaches to Chance Discovery (Preliminary Report) Helmut Prendinger and Mitsuru Ishizuka Department of Information and Communication Engineering School of Engineering, University of Tokyo 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan E-mail: {helmut,ishizuka}@miv.t.u-tokyo.ac.jp Abstract: This paper gives an initial evaluation of approaches to chance discovery and management, a new research field invented by Yukio Ohsawa. Our aims in this preliminary report are very modest. First, we briefly describe research work done with the intent to detect chances (i.e., opportunities and risks) in various application areas such as the web, natural disasters, or histories of bankruptcy. Second, we raise some methodological questions about the feasibility of chance discovery as a research field.
1
Introduction
Despite its infancy as a research field, chance discovery has already attracted considerable interest from researchers of various disciplines, including webrelated research, finance, and simulation of natural disasters. This new research field received a strong impulse from a workshop held in Brighton in 2000, organized by Yukio Ohsawa [6]. The aim of this paper is to give an assessment of the achievements of chance discovery, taking the Brighton workshop as the state-of-the-art. Chance discovery is mainly motivated from the practical, application-oriented side, which is clearly reflected in the contributions to the workshop. However, this field also gives rise to a series of more foundational questions, which seem highly interesting from a methodological view of science. For instance, chance discovery may shed new light on the question what prediction means in the presence of human initiative. Although we believe that applications should remain the prime focus of chance discovery, we will try to account for the more methodological problems as well. The rest of the paper is organized as follows. After giving a definition of chance discovery, relevant research work is summarized in Section 3. Next, some methodological questions are described. In Section 5, the considered research works are evaluated against the raised questions. Finally, we conclude the paper.
2
What is Chance Discovery?
Chance discovery aims at finding phenomena (or features) that will have a high impact in the future. Those phenomena are called chances whereby two complementary readings of ‘chance’ are explicitly intended: chance can mean an opportunity and it can mean a risk (e.g., Prendinger and Ishizuka [8]). Opportunities refer to the possibility of bringing about desirable effects, whereas risks refer to possible threats. Obviously, there is a strong interest to detect chances, as they can mean an advantage in the competitive business world or the avoidance of serious damage of goods. Chance management refers to activities following the discovery of chances, the ‘catching’ of chances, e.g., taking an opportunity to bring about desirable effects.
3
Application Areas
In this section, we will summarize some major chance discovery research works, grouped by their respective application area.
3.1
The Web as a Goldmine
A majority of the contributed papers consider the world-wide web as a profitable pool for discovering chances. This application is very natural since the
web is commonly considered as a prime candidate for testing advanced information technology. Terano and Murakami [12] motivate a collaborative filtering approach to provide users with chances. This approach is based on genetics-based machine learning techniques and allows to cluster similar users based on their interest profiles. Here, users may profit (i.e., catch chances) from web pages other users rated as interesting. The work of Matsumura et al. [3] aims to discover promising new topics on the web. Their ‘Expected Activation’ algorithm is based on the idea that chances can be found in web pages that are referred to (cited) by a small number of communities of authorized (i.e., respected) web pages. The extracted pages are called ‘chance pages’. By contrast, frequently cited pages are called ‘trend pages’. The merit of chance pages is to identify possible new trends at a very early stage. The following systems are in the tradition of enhanced web search. The system of Sunayama and Yachida [11] supports web search by suggesting keywords related to a user’s interest. Tsuda et al. [13] propose to built up a page index to discover users’ interests.
on real data from refugees of the South-Hyogo earthquake in 1995. Rather than discovering the place of a future earthquake, the authors aim to develop a risk management tool. As a computer-based support system, it allows for ‘context-shifting’ of the user’s mind to prepare the user for protection activities in case of a disaster.
3.2
4
Avoidance of Bankruptcy
Financial business is a rewarding application field for the area of chance discovery. In particular, significant attention is paid to the problem of detecting signs in a company’s history that eventually led to bankruptcy. Those signs are the risks that have significant impact on the future development of a company. Shirata [10] uses two data mining techniques (CART and Stepwise) to detect crucial factors for bankruptcy of Japanese companies. Most importantly, she investigates the ‘turning point of companies’, i.e., the time point which initiates a period leading to bankruptcy. Besides financial data, she also considers human decisions in her model. Similarly, Ogawa [5] motivates an extreme validity risk factor called IIE (Information of Liability and Equity) to identify signs that indicate future bankruptcy of a company. He shows that strong indicators of a company’s decline are already evident several terms before the bankruptcy happens.
3.3
Simulating Natural Disasters
One of the original applications of chance discovery is the identification (prediction) of the place of future earthquakes (Ohsawa and Yachida [7]). Recently, Nara and Ohsawa [4] expanded this work and developed a model for simulating natural disasters, based
3.4
Improving Learning
There exists a growing need for educational tools that facilitate productive learning. Kusunoki et al. [2] investigate opportunities inherent in a collaborative learning approach. They describe a system that allows to learn about urban planning and environmental problems in a game-playing (‘edutainment’) style. They report about empirical studies showing that the use of their system supports externalization of students’ thinking, active participation, and interactions among students. Although the paper does not directly address the chance discovery theme, their approach seems to be a good starting point to think about the pedagogical implications of the chance discovery field.
Methodological Problems
In this section, we take the viewpoint of methodology of science, and formulate some basic questions about chance discovery. Problem 4.1 (Explanation vs. Prediction) Assume as given a theory that explains why a particular phenomenon turned out to be a chance (opportunity or risk), as observed by its high (positive or negative) impact. Given a similar phenomenon, can we employ this theory to predict high impact under comparable circumstances? Of course, the notions of similar and comparable warrant further explication. In order to clarify the problem, consider the case of unstable or chaotic systems that support explanations without predictive value. Assume an ideal ball exactly on top of another ideal ball. Here, we cannot predict in which direction the ball will roll down, but after it rolled down, we can explain it by an unmeasureably small disturbance in the direction in which the ball rolled down (Schurz [9]). Thus, the ‘explanation vs. prediction’ problem raises the fundamental question about which systems support the predictive use of chance discovery results. Straightforward answers seem to be ruled out by the
fact that human initiative is essential to take opportunities or avoid risks, and the complexity of systems such as the web or financial markets. Problem 4.2 (Injection Problem) Assume a chance has been discovered. What kind of supporting measures (in case of opportunities), or preventive measures (in case of risks) should be taken? This problem is related to the management of discovered chances. It also covers the problem whether humans are aware of chances in the first place.
5
Taken ... and Missed Opportunities
Most of the research works on chance discovery take a ‘causal’ notion of chance (causal is understood in the sense of ‘giving reasons for’ rather than physical interpretations). For instance, in order to explain bankruptcy, an index is identified and proposed as a strong indicator for the eventual bankruptcy of a company (Shirata [10], Ogawa [5]). However, as the authors point out, their risk factors do not clearly reflect the role of human interference. An interesting problem would be to investigate managers’ wrong decisions in relation to the accounting point of view. Similarly, Matsumura et al. [3] detect certain web pages, i.e., chance pages, that might hint at a promising new topics. Those pages might be seen as the starting point of growing interest in certain topics. Although this seems very reasonable, we might imagine cases where those pages do not initiate future trends, as the reasons for upcoming trends are typically much more complex. A common problem of the mentioned approaches might be that both the economical market and the web are open system with a prohibitively large number of possible influences which implies high sensitivity to small external disturbances. This fact does not necessarily mean that those approaches cannot be successful, but (i) that it might turn out to be very difficult to justify their success scientifically, and (ii) that it is hard to define under the conditions under which the success can be reproduced. Regarding the injection problem, the work of Nara and Ohsawa [4] seems very promising as it gives an answer to the question how humans should deal with chances (in their case, risks). This paper is clearly on the chance management side of the spectrum. When reviewing the papers from the chance discovery workshop for the purpose of writing this report, we missed one aspect, which seems to be inti-
mately connected to this field: a description of human values in relation to catching chances, i.e., taking opportunities and avoiding risks. Most obviously, there are no opportunities or risks per se, they are only given with respect to certain values and goals of humans. To give a drastic example, the detection of a future earthquake is not only a high risk for people living in a particular region, it is also an opportunity for certain non-governmental organizations to take advantage of the chaos following the earthquake. On the other hand, we might even consider human values as the starting point for discovering chances, and take them as guidelines to bring about desirable results or avoid undesirable ones. A notion relevant in this respect is ‘anticipation’, described as follows: “[...] anticipation is not limited to simply encouraging desirable trends and averting potentially catastrophic ones: it is also the ‘inventing’ or creating of new alternatives where none existed before.” (see Botkin et al. [1, p. 25]). Anticipation explicitly focuses on the creation of possible and desirable futures and plans to install those futures. As such, it puts a clear stress on the crucial role of human initiative as a trigger for chances.
6
Conclusion
In this report, we gave a preliminary evaluation of approaches to chance discovery (and management). We described some of the main approaches and showed that they are already successfully applied to areas as diverse as the world-wide web and simulations of earthquakes. Taking a more methodological viewpoint, we described some of the problems that chance discovery poses as a scientific enterprise.
References [1] J. W. Botkin, M. Elmandjra, and M. Malitza. No Limits To Learning. Bridging The Human Gap. A Report to the Club of Rome. Pergamon Press, 1998. [2] Fusako Kusunoki, Masanori Sugimoto, and Hiromichi Hashizume. Discovering how other pupils think by collaborative learning in a classroom. In Ohsawa [6], pages 667–670. [3] Naohiro Matsumura, Mitsuru Ishizuka, and Yukio Ohsawa. Discovering promising new topics on the WWW. In Ohsawa [6], pages 804–807. [4] Yumiko Nara and Yukio Ohsawa. Tools for shifting human context into disasters. a case-based
guideline for computer-aided earthquake proofs. In Ohsawa [6], pages 655–658. [5] Shingo Ogawa. Building of trust evaluation model based on the failure prediction. In Ohsawa [6], pages 659–662. [6] Yukio Ohsawa, editor. Workshop on Chance Discovery and Management. In conjunction with the Forth International Conference on Knowledge-based Intelligent Engineering Systems and Allied Technologies (KES’2000), Brighton, UK, 2000. IEEE, Inc. [7] Yukio Ohsawa and Masahiko Yachida. Discover risky active faults by intexing an earthquake sequence. In Proceedings International Conference on Discovery Science (DS’99), 1999. [8] Helmut Prendinger and Mitsuru Ishizuka. Methodological considerations on chance discovery. In Proceedings IEEE International Conference on Industrial Electronics, Control and Instrumentation (IECON-00). Special Session on Chance Discovery., 2000.
[9] Gerhard Schurz. Scientific explanation: A critical survey. Foundations of Science, 3:429–465, 1995. [10] Cindy Yoshiko Shirata. Peculiar behavior of Japanese bankrupt firms: Discovered by AIbased data mining technique. In Ohsawa [6], pages 663–666. [11] Wataru Sunayama and Masahiko Yachida. An aiding system for internet surfings by associations: Get one through chances. In Ohsawa [6], pages 808–811. [12] Takao Terano and Eiji Murakami. Finding users’ latent interests for recommendation by learning classifier systems. In Ohsawa [6], pages 651–654. [13] Kazuhiko Tsuda, Osamu Yamagata, and Michitada Morisue. Shopping-chances in web-pages discovered from user’s access logs. In Ohsawa [6], pages 800–803.
