Akinori Abe

Recently, researches on discovery science and knowledge discovery have been studied for various fields. Indeed, these studies deal with real and large data, however basically, they are types of learning that learn tendencies from the sets of data in the same or similar categories. Accordingly, they cannot foretell the events that are different from the trend. When we predict unknown events, it is quite unnatural to have a stochastic model assumption that is made only from the known data and ignores abnormal phenomena. Of course, it is important to make a stochastic model to predict a future feature, but it is more important to find factors that cannot be reasoned by a stochastic process and are quite critical to an unknown event. That is, finding exception rules and exception facts is also important to risk management or risk avoidance in a real world....
Chance Discovery is a research to discover chance!!! (For details pls refer to to the CD site.)
By Ohsawa, chance is defined as follows;

A chance (risk) is a new event/situation that can be conceived either as an opportunity or a risk.

Definition: Chance

1. Chance is a set of unknown hypotheses. Therefore, explanation of an observation is not influenced by it. Accordingly, a possible observation that should be explained cannot be explained. In this case, a hypotheses base or a knowledge base lacks necessary hypotheses. Therefore, it is necessary to generate missing hypotheses. Missing hypotheses are characterized as chance.
2. Chance itself is a set of known facts, but it is unknown how to use them to explain an observation. That is, a certain set of rules is missing. Accordingly, an observation cannot be explained by facts. Since rules are usually generated by inductive ways, rules that are different from the trend cannot be generated. In this case, rules are generated by abductive methods, so trends are not considered. Abductively generated rules are characterized as chance.

My standpoint is to apply Abductive Analogical Reasoning to Chance Discovery.

In the above definition, we defined two types of chance. The followings show two types of chance discovery by using a framework of AAR.

1. Type 1: When some of hypotheses are unknown. When some of necessary hypotheses are unknown (not found), the current observation cannot be explained. In a context of chance discovery, chance seems to be a set of unknown hypotheses. And this type of inference can be regarded as pure abduction.
2. Type 2: When some of rules are unknown. When some of necessary rules are unknown, even if we are aware of all of symptoms, the future observation cannot be explained (predicted). In this case, usually, reasons are shown afterwards. This type of inference can be regarded as AAR.

From psychological viewpoint

1. Chance Discovery in Multi-lingual Translation Site
   Recently, the data on web site has been thought of as meaningful data to analyze our society. It is used to see the trends in specific field. In addition, it can be used to predict the next trend or current hidden needs. This is because the data is natural (unintentional). We run the multi-lingual machine translation service site. In fact, in more than one year, we have collected huge mount of log data. They are user's access data, user's translation data (word, text, url, language pair etc.), and user's feedback comments. From these data, we can obtain a lot of information. We analyzed log data to find a certain chance.
   From the viewpoint of commercially maintaining machine translation site, chance will be new business model. On the other hands, from the viewpoint of sociology, chance will be results from the change of user's interests. They reflect the situation of society and if we use them correctly, we can guess novel or rare events before they explicitly appear in the world.

   (business) chance:

   • The users do not frequently search their interesting page by using search engine, but they check magazine or come from linked pages. Therefore, very famous site like {\tt www.asahi.com} and {\tt www.cnn.com} can live long. However, other infamous site should register not on search engine but on special site and special media like magazine or TV.
   • Language pair should be English-centered. Actually, translation services between Chinese and Japanese and between Korean and Japanese are required, if we have translation services between Chinese and English and between Korean and English, it will be better for the users in the world. This is because, recently, Asian countries are thought of as a significant place from the various viewpoint like economy, culture etc.

2. User's interests change
   Recently, the necessity to the assistance of creative work has been increased. Creative works are usually regarded as specialized skills. However, from the viewpoint of chance discovery, creative work can be thought of as chance. According to the definition of (chance), the creative work can be thought of as chance. This is because creativity means novel (not rare) and creative work comes from nothing or from something relative but whose relation was hidden.
   This research showed chance as creative work, and shows the process of chance discovery. This formalization comes from our experience. We work in abroad and does not bring enough references and there is not proper libraries around office and house. When he wrote paper, he naturally used an internet search engine like yahoo. Each search engine has its own feature. Therefore we must know their feature to obtain proper results. This task will be a sort of craftsman performance. Sometimes search engine returns different result from our intention. Usually, such wrong results are useless. However, in some cases, the wrong result leads us novel thinking or confirm our original target. We modeled this type of process as a chance discovery process.
   

Applications of Chance Discovery

1. Creativity support

   See above.
   In addition, we are taking into consideration an aspect of chance discovery to the Augmented Music Composition Support that supports music composition by showing differnence between the composition of experts and that of the users.

2. Chance Discovery in Nursing Informatics (NI)

   Logical view:
   Abst: We analyze the feature of chance in nursing risk management for better management. Recently, it has been recognized that medical risk management is very important both for hospitals and hospital patients. To reduce nursing accidents, examples of nursing accidents are usually collected for analysis. This allows us to obtain certain tendencies of nursing accidents and causality between environment and nursing accidents. Such knowledge can be useful in nursing risk management, but is not fully adequate. In addition, in a real situation, it is necessary to deal with a hidden relationship between an ignored event or factor and an accident. Thus, we analyze the feature of hidden factors in nursing accidents and propose a way to determine chance (= hidden or ignored factors) as abductive hypotheses.

   ***Nursing risk prediction as Chance Discovery***
   Risk management itself can be thought of as an application of Chance Discovery. A chance in nursing risk management can be defined as an ``ignored'' event, factor, environment, personal relationship or personal matter that has the possibility to cause a serious nursing accident or incident in the future.
   

   in the case of a slip, due to unexpected events, although an ideal or desired situation has not actually been achieved, we unconsciously convince ourselves that we have achieved such an ideal or desired situation. Thus because of (unintentional) discontinuation of tasks, we cannot keep our contexts. A chance exists at the hatched circle in the above figure.

   How to determine a chance?
   It is difficult to detect a critical point beforehand. However, it would be possible to abductively detect a critical point as a point where an inconsistency between an ideal result and his/her actual activity occurs. For example, if we use a hypothetical reasoning framework such as Theorist, we might be able to deal with the situation. In the case where we know all of the possible hypotheses and their ideal observations, we can detect malpractice. This is because if he/she selects a wrong hypothesis set, an ideal observation cannot be explained. A simple framework is shown below:
   

   Linguistic view:
   

3. Chance Discovery in Medical Informatics (MI)

   Regarding medical diagnosis support system, the problem of knowledge base incompleteness makes it necessary to combine an inference part based on abduction and a knowledge acquisition and learning part based on induction. We think it is necessary to do an automatic medical rule generation method from the viewpoint of induction and chance discovery, to do abduction for chance discovery, and to seek for possible methods including KeyGraph to achieve medical chance discovery.

4. Chance Discovery in Medical Informatics (MI)

   Regarding medical diagnosis support system, the problem of knowledge base incompleteness makes it necessary to combine an inference part based on abduction and a knowledge acquisition and learning part based on induction. We think it is necessary to do an automatic medical rule generation method from the viewpoint of induction and chance discovery, to do abduction for chance discovery, and to seek for possible methods including KeyGraph to achieve medical chance discovery.

   Our definitions of chance in MI:

   1. A phenomenon that is slightly different from the typical phenomenon.
      This can be regarded as a symptom in the gray zone. The knowledge is similar or close to the typical one. It might be regarded as an exception if we extend the definition of exception.
   2. A supporting phenomenon that is quite rare or novel, so we cannot think an existing or known knowledge works for a reason.
      In this case, the result (observation) is explicit, but we cannot find any explanations or reasons (hypotheses in abduction) for the result.
   3. A phenomenon that is so rare or novel that we have no knowledge to deal with it, therefore, we cannot infer any results.
      In this case, in spite of any symptoms, results are implicit. This is because we do not have any knowledge from which to diagnose from the symptoms.

   To do chance discovery in MI, we can use logical methods etc.

   • the role of induction: to detect the gray zone as well as to build a standardized knowledge base.
   • the role of abduction: to show (suggest) known events as chance by generating new or novel knowledge.
   • the role of KeyGraph: graphically outputs a relationship among elements in a data set: to find rare or hidden relationships.

   Currently, we have the following conclusions;
   Since disease is an outbreak with many factors, such as virus and environment, etc., it is quite difficult to find a chance from only medical data. Consequently, we need to build a medical knowledge base with consideration of the environment. Similarly, we need to make an inference with consideration of the environment. We think a relationship between medical data and the environment is hidden, so it is necessary to find such a hidden relationship by abduction and KeyGraph.

1. Abe A.: The role of abduction in Chance Discovery, Proc. of SCI2001, Vol. VIII, pp. 400-405 (2001)
2. Abe A.: User's interests change as Chance Discovery, Proc. of KES2002, pp. 1291-1295 (2002)
3. Abe A., Toong C. K., Nakamura M., Tsukada M., and Kotera H.: Finding Chance in Multi-lingual Translation Site, Proc. of AAAI Fall Symp. on Chance Discovery: The Discovery and Management of Chance Evnet (FS-02-01), pp. 22-27 (2002)
4. Abe A.: Is forecasting harder than it used to be?, panel discussion in AAAI Fall Symp. on Chance Discovery (2002) slide
5. Abe A.: The role of abduction in Chance Discovery, New Generation Computing, Vol.21, No.1, pp. 61-71 (2003)
6. Abe A., Kogure K. and Hagita N.: Discovery of Hidden Relations from Medical Data, Proc. of HCI2003 3rd. Int'l Workshop on Chance Discovery, pp. 37-43 (2003)
7. Abe A.: Abduction and Analogy in Chance Discovery, in Chance Discovery (Osawa Y. and McBurney P. eds.), Chap. 16, pp. 231-248, Springer (2003)
8. 阿部 明典: 発想的推論とチャンス発見, in チャンス発見の情報技術 (大澤 幸生 編), 第 8 章, 東京電機大学出版局 (2003)
9. Abe A., Berry R., M. Suzuki, and Hagita N.: Augmented Music Composition Support as Active Mining, Technical Report of IEICE, Vol. 103, No. 304, pp.59-64 (Technical Report of JSAI, SIG-KBS-A301, pp.59--64) (2003) paper
10. Abe A. and Oehlmann: From Data Mining to Interpersonal Communication for Scenario Development, Proc. of ECAI2004 Workshop on Chance Discovery, pp. 1--8 (2004)
11. Abe A., Kogure K. and Hagita N.: Determination of A Chance in Nursing Risk Management, Proc. of ECAI2004 Workshop on Chance Discovery, pp. 222-231 (2004)
12. Abe A., Kogure K. and Hagita N.: Nursing Risk Prediction as Chance Discovery, Proc. of KES2004 Vol. II, pp. 815--822 (2004).
13. Abe A. and Ohsawa Y. eds: Readings in Chance Discovery, International Series on Advanced Intelligence (2005)
14. Abe A., Toong C. K., Nakamura M., C. W. Lim, Tsukada M., and Kotera H.: Finding Chance in Multi-lingual Translation Site, in Readings in Chance Discovery (Abe A. and Ohsaw Y. eds.), Chap. 3, pp. 25--36 (2005)
15. Abe A.: Creativity as Chance Discovery, Proc. of JCIS05 1st Annual Workshop on Rough Sets and Chance Discovery, pp. 1782--1785 (2005)
16. Abe A., Naya F., Ozaku H.I., Kuwahara N., and Kogure K.: Scenario Violation in Nursing Activities, Proc. of the ICML05 4th Int'l Workshop on Chance Discovery, pp. 102--109 (2005)
17. Abe A., Naya F., Ozaku H.I., Sagara K., Kuwahara N., and Kogure K.: Risk Management by Focusing on Critical Words in Nurses' Conversations, Proc. of KES2005, Vol I, pp. 1167--1173 (2005)
18. Abe A., Ozaku H.I., Kuwahara N., and Kogure K.: Scenario-base Construction for Abductive Nursing Risk Management, Proc. of IPUM2006, pp. 206--213 (2006)
19. Abe A., Ozaku H.I., Kuwahara N., and Kogure K.: What Should be Abducible in Abductive Nursing Risk Management?, Proc. of KES2006 (LNAI4253), Vol III, pp. 22--29 (2006)
20. Abe A., Ozaku H.I., Kuwahara N., and Kogure K.: Relation between Abductive and Inductive Nursing Risk Managements, Proc. of RM2006 (JSAI2006), pp. 121--132 (2006)
21. Abe A. and Kogure K.: E-Nightingale: Crisis detection in nursing activities, in Chance Discoveries in Real World Decision Making (Ohsawa Y. and Tsumoto S. Eds.), Data-based Interaction of Human intelligence and Artificial Intelligence Series: Studies in Computational Intelligence, Chap. 15, pp. 357--371, Vol. 30, 2007, XIV, 404 (2006)
22. Abe A., Ozaku H.I., Kuwahara N., and Kogure K.: Cooperation between Abductive and Inductive Nursing Risk Management, Proc. of RM2006 (ICDM2006), pp. 705--708 (2006)
23. Abe A., Ozaku H.I., Kuwahara N., and Kogure K.: Scenario Violation in Nursing Activities --- Nursing Risk Management from the viewpoint of Chance Discovery, Soft Computing Journal, Springer, Vol. 11, No. 8, pp. 799--809 (2007)
24. Abe A., Ozaku H.I., Kuwahara N., and Kogure K.: Relation between Abductive and Inductive Types of Nursing Risk Management, Post-proc. of JSAI2006 (LNAI 4834), pp. 387--400 (2007)
25. Abe A., Ozaku I.H., Ohsawa Y., Sagara K., Kuwahara N., and Kogure K.: Communication error determination model for multiply layered situations, Proc. of RI2007 (JSAI2007), pp. 38--48 (2007)
26. Abe A., Ozaku H.I., Sagara K., Kuwahara N., and Kogure K.: Nursing Risk Management by Focusing on Critical Words or Phrases in Nurses' Conversations, Int'l J. of Knowledge-based and Intelligent Engineering Systems, Vol. 11, No. 5, pp. 281--289 (2007)
27. Abe A., Ohsawa Y., Ozaku H.I., Sagara K., Kuwahara N., and Kogure K.: Communication error determination model for multi-layered or chained situations, Proc. of PAKDD 2008 Working Notes of Workshops on Data Mining for Decision Making and Risk Management, pp. 305--316 (2008)
28. Abe A, Hagita N, Furutani M, Furutani Y, and Matsuoka R.: Categorized and Integrated Data Mining of Clinical Data, in Communications and Discoveries from Multidisciplinary Data (Iwata S., Ohsawa Y., Tsumoto S., Zhong N., Shi Y., Magnani L. eds.), Studies in Computational Intelligence, Vol. 123, pp. 315--330, Springer Verlag (2008)
29. Abe A., Hagita N., Furutani M., Furutani Y., and Matsuoka R.: Exceptions as Chance for Computational Chance Discovery, Proc. of KES2008 (LNAI5179), Vol. II, pp. 750--757, Springer Verlag (2008)
30. Abe A.: Cognitive Chance Discovery, in Universal Access in Human-Computer Interaction --- Addressing Diversity (Stephanidis eds.), UAHCI 2009, Held as Part of HCI International 2009, Proceedings, Part I (LNCS5614), pp. 315--323, Springer-Verlag (2009)
31. Abe A., Hagita N., Furutani M., Furutani Y., and Matsuoka R.: An Interface for Medical Diagnosis Support ---from the viewpoint of Chance Discovery, International Journal of Advanced Intelligence Paradigms, Vol. 2, No. 2/3, pp. 283--302 (2010)
32. Ohsawa Y., Abe A., and Nakamura J.: Chance Discovery as Analogy based Value Sensing, International Journal of Organizational and Collective Intelligence, Vol. 1, No. 1, pp. 44--57 (2010)
33. Abe A., Ohsawa Y., Ozaku I.H., Sagara K., Kuwahara N., Kogure K.: Communication Error Determination System for Multi-layered or Chained Situations, Fundamenta Informaticae, 98, pp. 123--142 (2010)
34. Abe A., Ohsawa Y., Kuwahara N., Ozaku I.H., Sagara K., Kogure K.: Scenario Violation as Gaps between Activity Patterns, New Mathematics and Natural Computation, Vol. 6, No. 2, pp. 193--208 (2010)
35. Abe A., Hagita N., Furutani M., Furutani Y., and Matsuoka R.: Categorized and Integrated Data Mining of Medical Data from the Viewpoint of Chance Discovery, Proc. of KES2010 (LNAI6278), Part III, pp. 307--314, Springer Verlag (2010) to appear
36. Abe A.: Abduction dealing with potential values, Proc. SMC2010, pp. 1279--1285 (2010)
37. Abe A.: Curation in Chance Discovery, Proc. ICDM2010 5th International Workshop on Chance Discovery, pp. 793--799 (2010)
38. Abe A., Hagita N., Furutani M., Furutani Y., and Matsuoka R.: An interactive interface for medical diagnosis support, in Sequence and Genome Analysis: Methods and Applications (Zhongming Zhao eds.), Chap. 17, pp. 289--305, iConcept Press (2011)
39. Abe A.: Curation and Communication in Chance Discovery, Proc. of IJCAI2011 6th International Workshop on Chance Discovery, pp. 3--8 (2011)
40. Abe A.: Relation between chance discovery and Black Swan awareness, Proc. of KES2011 (LNAI6882), Part II, pp.495--504, Springer Verlag (2011)