Studies in Computational Intelligence 423 Editor-in-Chief Prof.JanuszKacprzyk SystemsResearchInstitute PolishAcademyofSciences ul.Newelska6 01-447Warsaw Poland E-mail:[email protected] Forfurthervolumes: http://www.springer.com/series/7092 Yukio Ohsawa and Akinori Abe (Eds.) Advances in Chance Discovery Extended Selection from International Workshops ABC Editors Prof.Dr.YukioOhsawa AkinoriAbe TheUniversityofTokyo FacultyofLetters Bunkyo-ku ChibaUniversity Japan Chiba Japan ISSN1860-949X e-ISSN1860-9503 ISBN978-3-642-30113-1 e-ISBN978-3-642-30114-8 DOI10.1007/978-3-642-30114-8 SpringerHeidelbergNewYorkDordrechtLondon LibraryofCongressControlNumber:2012937234 (cid:2)c Springer-VerlagBerlinHeidelberg2013 Thisworkissubjecttocopyright.AllrightsarereservedbythePublisher,whetherthewholeorpartofthe materialisconcerned,specificallytherightsoftranslation,reprinting,reuseofillustrations,recitation,broad- casting,reproductiononmicrofilmsorinanyotherphysicalway,andtransmissionorinformationstorage andretrieval,electronicadaptation,computersoftware,orbysimilarordissimilarmethodologynowknown orhereafterdeveloped.Exemptedfromthislegalreservationarebriefexcerptsinconnectionwithreviews orscholarly analysis ormaterial suppliedspecifically forthepurposeofbeingentered andexecuted ona computersystem,forexclusive usebythepurchaser ofthework.Duplication ofthis publication orparts thereofispermittedonlyundertheprovisionsoftheCopyrightLawofthePublisher’slocation,initscur- rentversion,andpermissionforusemustalways beobtained fromSpringer. Permissionsforusemaybe obtainedthroughRightsLinkattheCopyrightClearanceCenter.Violationsareliabletoprosecutionunder therespectiveCopyrightLaw. Theuseofgeneraldescriptivenames,registerednames,trademarks,servicemarks,etc.inthispublication doesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfromtherelevant protectivelawsandregulationsandthereforefreeforgeneraluse. Whiletheadviceandinformationinthisbookarebelievedtobetrueandaccurateatthedateofpublication, neither the authors northe editors nor the publisher can accept any legal responsibility for any errors or omissionsthatmaybemade.Thepublishermakesnowarranty,expressorimplied,withrespecttothematerial containedherein. Printedonacid-freepaper SpringerispartofSpringerScience+BusinessMedia(www.springer.com) Preface Yukio Ohsawa and Akinori Abe The beginning of year 2000, we started to study methods for discovering events/situations that significantly affect decision making. The triggers to this movement and its aftershocks occurred with the initiator Ohsawa’s movements from/to schools in engineering to business sciences since around in April 1999. When he was studying methods for visualizing data on earthquakes with his original KeyGraph(cid:2) (See chapters in [Ohsawa and McBurney, 2003] for some use cases), where high-risk faults of earthquakes were shown by red nodes rep- resentingrarelyquakingfaults affected by frequently quakingactive faults visu- alized by black nodes, the business students who work at daytime who learned rare but significant events may get highlighted by KeyGraph suggested such a graph may be useful for aiding decision making in business. As a result, they started to collaborate with companies to visualize data on businesses, to detect noteworthy events which might be rare but useful for planning new actions and decisions. Such events or situations are sometimes called opportunities, and other times crisis. With realizing this kind of suc- cesses (also see Fig. 1, they started series of workshops and special sessions on what they called Chance Discovery (IECON2000, KES 2000, AAAI2001, ECAI2004, etc) — second editor Abe joined these activities, and now is lead- ing the world-wide network of researchers on chance discovery. Through these active efforts, the participants’ vague dream converged to a unique definition of a chance, i.e., an event which has a significant impact on human’s deci- sion making. Thus a chance provides an uncertain opportunity/risk for human. And, chance discovery is the discovery of a chance, emphasized in contrast to discovery by chance. Because the essential aspect of a chance is that it can be the seed of new and significant benefits/loss for human, the sense of hu- man(s) about benefit/loss is a significant factor for chance discovery as well as machine’s power to analyze and visualizing data for aiding humans’ talent. VI Preface Fig. 1. A successful case in the classical chance discovery: The clusters of frequently ordered items are bridged via a rare item, from which textile marketers concluded to propose a new selling scenario: Make a jacket for business people with the new corduroy,towhich they can change from suitsto go out for dinnerafter workingtime [Ohsawa and Usui, 2005]. As wellasworkshopsonchancediscovery,they publishedbooksandissuesfrom journals, characterizing chance discovery with studies on: (1) Humanfactorsincognitions,communications,andthoughtsforcatchingup with a chance and understanding its significance for decision making, (2) Process for externalizing candidates of chances and discovering chances on objective evidences — reinforcing (1) by data in other words, and (3) Method and tools mining/visualizing data that aid in chance discovery — accelerating and sustaining the process of (2) in other words. In other words, the design of interaction among human(s), computers, and their environment came to the core interests in studies for realizing chance dis- covery. The papers contributed to chance discovery so far have been by re- searchersfromvariousdomainssuchasartificialintelligence,creativitysupport, economics, business administration, risk management, operation research, lin- guistics, mathematics, physics, psychology etc showing approaches relevant to classes (1), (2), and (3). For the present book, we selected authors of outstanding papers which were presented on chance discovery in relevant workshops, conferences, and sym- posia — mainly from workshops in the International Joint Conference on Arti- ficialIntelligence (IJCAI 2011),IEEEInternationalConference onData Mining (ICDM2011),IEEEInternationalConferenceonSystems,Man,andCybernetics Preface VII (SMC2010). Collaboratorswho utilized or extended studies in chance discovery forcontributingtootherbutrelevantconferences,suchasofmulti-agentsandju- ristic communications (AAMAS2011, JURISN2011) were also invited. As guest editors, we are happy to find three chapters for class (1), five for (2), and seven for (3) in this book. Readers will find new visions for the decision making of human, from this harmonious combination of contributions. In the first part, we show contributions to the aspect of Cognition and Com- municationtowardChanceDiscovery.Abe’sintroductionoftheconceptcuration in Chapter 1 encourages us to discuss the meaning of chance discovery from a different aspect coming from the exhibition of artwork in a gallery. Borrowing ideasfromcasesofdatacurationandcurationofartwork,hesuggestsadirection to curation for aiding chance discovery, and leads to the concept to the role of communicationinchancediscovery.InChapter2,byBardoneandMagnani,the bullshitting phenomenathatisasignificantkindofchancefaking,ishighlighted and explained with respect to one’s carelessness about frames, evidences, and simplicity that one should stay aware of for taking real advantage of notewor- thy chances.The messagessuggestedin this paper areinstructive andusefulfor people who seek chances — we learn from this paper that we need evidential data and also concrete interests in explaining the data in the form of a simple logic under a suitable context, whether or not data mining tools are provided. Terai and Miwa links chance discovery to their original studies on insight in Chapter 3, that means to release humans from impasse, so that noticing the value of a new event gets enabled. They argue that in insight problem solv- ing tasks, a prepared mind precedes the “aha” experience, as in the process to chance discovery. In order to preparing mind for chance discoveries, they point out interdisciplinary studies between cognitive researches and techniques that encourage chance discovering.Also introducing an original study attempting to revealthepreparedprocessofinsightproblemsolvingusingeyemovementdata, they point out possible approaches from cognitive science to chance discovery. All in all, the first part focuses on how we can open human’s mind out to the value of external information. We have been endeavoring to reinforce such cognitive forces toward chance discovery, by developing technical aspects of human-machine and human-human interactions with developing techniques of data visualization — as in the case KeyGraph has been used for discovery of an opportunistic item in the market as shown in Fig. 1. In other words, data visualizationplaystheroleastheenvironmentforchancecuration.Thatis,how to present data on facts as evidences for viewers to notice chances — this is the concept Chapter 1 introduced in this book and the second part embodies by presentingfivecontributions.And,thisisthereasonwhywegivethetitle“Data Visualization as Chance Curation” to the second part. The second part starts from Chapter 4, where Sarlin writes an approach to chancediscoveryinfinance,withself-organizingmapsfordiscoveringimbalances in financialnetworks.Here we find the diffusionof financialcrisisfrom/tocoun- tries and financial state transitions are easy to see on the visualized maps, so that early signs of changes shall be caught as chances of people in businesses VIII Preface including investments andthe government.InChapter 5,Sunayamaet alshows a recursive method for clustering and visualizing textual data, so that human can detect useful information from the atlas of documents. As chance discovery favors utility sometimes more than accuracy, their method which does not al- ways present accurate clusters but provide useful information is expected to be a promising tool for chance discovery. Another feature of the second part is that communication is introduced as a building block of chance discovery. In Chapter 6, Nitta presents a method for analyzing discussions,coupling logicalanalysis of the discussion subject and wordstatisticalanalysis—atemporalwordclusteringmethod—ofthelogtext of discussion. We can appreciate this work as a way to integrate chance discov- ery and logical modeling of the dynamics in the target domain. By integrating thesetwomethods,thediscussionmoderationskillsareanalyzedtofindreusable knowledge for other discussion. Furthermore,nonverbalinformation can be tar- geted by extending Nitta’s method. In Chapter 7, Lin et al. visualizes graphs representing the social network linked to innovation and its diffusion. By thus visualized evidences, we can understand the dynamics of innovation and detect essential events which, and early adopters who, may have caused or may cause thegrowthofindustry.ThisapproachissucceedingtheessentialfeatureofKey- Graph (Fig. 1), but their visualization will be found easier to see in interpret. WangandOhsawa,inChapter8,presentsiChanceaWeb-basedenvironmentfor innovative communication with chance curation. Here, the gaming principle of InnovatorsMarketplace(cid:2) (thisisOhsawa’strademarkinUS:InnovatorsMarket Game(cid:2) is a trademark in Japan) has been implemented on the Web and re- inforced by Wang’s style of communication where inventors evaluate consumers (aswellasviceversa)andhisnewvisualizationtechniqueGallaxy.Weareshow- ing how chance discovery can be really realized by the collective intelligence of humans and machines. Let us look at the first and the second parts above from another aspect: While humans perceive and understand chances and plan scenarios of actions and events in the future, how are data processed? As in Fig. 2, we have been modelingtheprocessofchancediscoveryasadoublespiral—humans’deepening ofinterestsinchancesanddataprocessing(collectionandvisualizationaccording to humans) interest. We can say the first part of this book has been dedicated to the humans’ spiral, where collaborators communicate with curating chances, reframing, and reaching insights to revise interests to adopt to the changing real world. And, the second part extends our view to include technical aspects where machine spiralare involved.Here we notice — the viewpoints of human(s) are revised in steppingupthespiralsothatthedatainhandandcollectedfromexternalworld areviewedandinterpretedina new direction.Thatis,in the case ofFig.1,two clusters of textile items, which had been respectively regarded as materials for casualclothesandfor businesssuits,came to be interpretedasone meta-cluster created by combining the two existing clusters, by introducing a new viewpoint tosatisfy businesspeople totakeeasyclothesafter theirworkingtime. Inshort, Preface IX Fig. 2. Double Helix (DH) process of chance discovery (also see [Ohsawa and McBurney,2003] — not only chapter 1 but others are directly or indirectly based on this process). the thing which occurred here and is expected by following the DH process is the synthesis of evidences, via the analysis and visualization of data. Thethirdpartofthisbookmainlyhighlightstechniquesfordatamining,that may seem to essentially focus on the analytical aspect of data processing. How- ever, we position these work as computational and logical cutting edges for not only dataanalysisbut alsodata synthesis.InChapter 9,Furuhata,Mizuta,and So shows a method for tracking concept drift, for forecasting different types of suddenunexpectedchangesandto beadaptivetothesechanges.Theirforecast- ing method based on the stable evaluator and the reactive evaluator are good at dealing with consecutive concept drifts. An application of this method is Fi- nance, as tested in this paper to financial data in US including the late-2000s recessions.InChapter10,Zhang,Leung,Pang,andTangproposesamulti-agent system for finding satisfactory services.As this work is dedicated to the general problem to match increasing providers and recipients efficiently, reader in any service domain is expected to project this work to his/her own field e.g., how can all patients find the best doctor for the health problem of each? In Chapter 11, Pogorelc and Gams focuses join health problem: Their technique enables to discovercluesformedicaldecisionmaking—donothingorcallforhelp—from data on daily movements of elderly people. Although their approach once pulls us back to the traditional view point of data mining, that is classification accu- racy, we notice there are domains where accuracy plays a critical role in quick and suitable decision making. It is thus important to consider the use scenario X Preface where data are finally synthesized into decision making, in order to choose the viewpoint in data analysis. Thoughts about computation techniques sometimes give birth to ideas about frameworks not only for artificial but also for natural (human’s) intelligence. In Chapter 12, Vladimir Rybakov proposes a framework of temporal logics for dealing with chance discovery. In chance discovery, the causality of events be- fore and after a chance is uncertain, and only partially observable — the causal logics behind them cannot be completely possible to explain and are hierarchi- cally structured from local sequences and global scenarios. Rybakov proposes a decidable logic of which the satisfiability is solvable, where operations of the linear temporal logic are combined with introducing operators corresponding to uncertainty temporally local and global discoveries. This will be not only a basic of computation but a starting point to discuss in what logical framework we should discuss chance discovery.In Chapter 13, Ogaardand March focus on pilotbehaviorstakenfromaircrafttelemetry datacollectedbyamobileground- basedsense-and-avoidsystemforUnmannedAircraftSystem(UAS)withsensors and customized visualization software. In order to estimate the current risk of midair collision, they developed probabilistic models for the behavior of pilots of manned aircraft. Complex subpaths were discovered from the data using an antcolony algorithm,andprobabilistic models were mined from those subpaths by extending existing algorithms. Signs of failures may be detected objectively by the method, and we may also expect a metacognitive effect i.e., pilots may find mental subpaths that are latent in their own cognitive process which may lead to unexpected risky behaviors. In Chapter 14, Hidenao Abe and Shusaku Tsumoto presents a method for characterizing and clustering a large number of documents, considering the features of temporal changes in the terms appeared inadocumentcollection.Theyexperimentallyobtainedessentialfeaturesinthe temporalbehaviorsoffourconferenceseriesondatamining,fromcorresponding published document collections. An extracted change may turn out to be a chance, if applied to documents of one’s interesting domain. Finally in Chapter 15, Yoshiaki Okubo, Makoto Haraguchi, and Takeshi Nakajima presents a method for finding an indicative pattern, that is a rare but noteworthy pattern — an itemset consisting of sev- eral general items but has a small degree of correlation. Then they discuss the relationshipofindicativepatternsandchancepatterns,thatistheiroriginalab- stractionofKeyGraph?whichhadbeencreatedbyOhsawa.Achancepatternis anindicativepatternsupportedbyapairofmorefrequentbasepatternsBLand BR. Such a pattern is expected to imply some hidden relationship between BL andBR,asinthe wayKeyGraph(cid:2) visualizedbridgesbetweenfrequentpatterns as in Fig. 1. They conclude that it might be required to take a causal relation- ship betweena chance pattern anda base pattern into accountso thata chance pattern can actually work as a valuable trigger for chance discovery. Insummary,wecanpointoutthreemajorprogressesinchancediscoverysince ourlastbookonchancediscovery[Ohsawa and Tsumoto, 2006]:First,basedon our common sense that chance discovery is the multidisciplinary research field Preface XI based on at least two established domains (cognitive science and data mining with visualization), we are involving fresh colleagues from application domains and basic technologies and sciences. Application domains include health care, aircraftcontrol,energyplant,managementoftechnologiesandfinances,product designs/innovations,marketing,etc, andbasic technologiesandsciences include sensor technologies, medical sciences, etc. Second, both cognitive scientists and computational scientists who interacted with us deepened and extended their fruits — they are too deep and wide to follow up in this narrow space. Third, time came to be considered and introduced explicitly as a significant variable ruling the causality from backgroundsituations to chances and from chances to its impacts onevents and actions of humans in the future. Readers may urge us tolistthe fourth,fifth,sixth,... progressesandwemaybe able todoso,but let us stop here. Aseditorsweexpectthereaderswillinvolvecolleagues,livingintherealworld of human’s decisionmaking,to have them join our future activities — although wedonotaimtobealargecommunitybutratherseektostaycompactenoughto continuemeaningful synthesisofthoughts viacurationsof chancesandintimate communications. May 2012 Yukio Ohsawa Tokyo Akinori Abe References [Ohsawa and McBurney,2003] Ohsawa, Y., McBurney, P. (eds.): Chance Discovery. Springer (2003) [Ohsawa and Usui, 2005] Ohsawa,Y.,Usui,M.:WorkshopwithTouchableKeyGraph Activating Textile Market. In: Abe, A., Ohsawa, Y. (eds.) Readings in Chance Discovery, Advanced Knowledge International (2005); note: The experiments for this paper were conducted by2001 [Ohsawa and Tsumoto, 2006] Ohsawa, Y., Tsumoto, S. (eds.): Chance Discoveries in Real World Decision Making, Data-based Interaction of Human Intelligence and Artificial Intelligence. SCI, vol. 30. Springer(2006)