Patent Application
20150324482
The present invention relates to a method or generating a network graph constituted by vertices and edges or nodes and links, as well as a decision-making support system, and specifically to a method of creating a network graph or a scenario map using big data suitable for the decision-making support system.
With a rapid growth of telecommunications such as the Internet, social media, Sensor Networks, mobile phones, and the like, there is an increasingly active movement of using the big data generated therefrom for decision-making by means of statistical analysis and data mining.
For example, in a chance discovery method, an event series based on a certain context is referred to as a scenario, a significant event or situation triggering the scenario to transit is regarded as a chance, and decision-making is considered as selection of a scenario at the chance. As a way of presenting a scenario, a scenario map is used such as a network graph and a potential map that visualizes a frequency and a co-occurrence degree of the event, and KeyGraph and KeyBird are known as a tool thereof.
Nonpatent Literature 1 discloses a scenario map that has occurred from the past to the present using “Polaris” as an integrated data miner for chance discovery. Moreover, Nonpatent Literature 2 discloses future prediction based on a history analysis of a scenario as a chance discovery method, and Nonpatent Literature 3 discloses visualization of a hidden event by a data crystallization method.
Patent Literature 1 discloses a subject and a configuration of communication by visualising communication data of participants of a computer-based collaboration with a network graph using the chance discovery method, thereby supporting the collaboration.
Patent Literature 2 proposes, in knowledge extraction from a text database, clarifying relations and differences among associated data and extracting useful knowledge not merely by presenting associated data bat by extracting associative networks in a predetermined co-occurrence relation from the database and integrating a synonym.
Patent literature 3 proposes eliminating ambiguity in parsing by learning a hierarchical relation of concepts between words and a co-occurrence degree of the words and the concepts as a network graph structure that is a concept hierarchy tree, thereby optimizing .Language processing in a language processing system such as a machine translation.
Patent Literature 1; U.S. Patent Publication no. 2005/0276479
Patent Literature 2: Japanese Patent Laid-open No. H06-168129
Patent Literature 3: Japanese Patent Laid-open No. H03-305608
Nonpatent Literature 1: Okazaki, N. and Ohsawa, Y., “Polaris: An Integrated Data Miner for Chance Discovery”, in Workshop of Chance Discovery and Its Management fin conjunction with International Human Computer interaction Conference (HCI2003)), pp. 27-30, Crete, Greece (2003)
Nonpatent Literature 2: Ohsawa, Y., “KeyGraph as Risk Explorer from Earthquake Sequence”, Journal of Contingencies and Crisis Management (Blackwell) Vol. 10, No. 3, pp. 119-1.28 (2002)
Nonpatent Literature 3: Ohsawa, Y., “Data Crystallization: A Project Beyond Chance Discovery for Discovering Unobservable Events”, IEEE international Conference on Granular Computing, Beijing (2005), Vol. 1, pp. 51-56
According to Simon's decision-making theory, unprogrammed decision-making regarding a complex systems such as a social system or an economic system is restricted by a bounded rationality of information acquiring capability and information processing capability and cannot correctly predict a future. Thus, there is a need for executing the decision-making based on the satisfaction principle from among alternatives that satisfy a certain target level.
According to Luhmann's social systems theory, the social system is an autopoiesis system based on a chain of communication consisting of information, message, and understanding, and the chance discovery is, according to Ohsawa's decision-making technique, a double helix process consisting of concern, understanding, proposal, and action based on an interaction between a computer and a human that constitutes the society.
Combining both, the decision-making support system can be regarded as a double helix autopoiesis system based on coordination between a human and a computer, and the computer needs to provide a service that satisfies the satisfaction principle within a bounded rationality to an uncertain future while adapting to change of humans and environment.
With an upcoming decision-making support system, it will be important to comprehensively present various scenarios that may not be recognized by a human with limited information, capability, and time for the uncertain future within a bounded rationality.
However, the conventional scenario map described in the above prior art documents can merely present an event from the past to the present and analyze and metamorphose the scenario therefrom.
For example, Nonpatent Literature 2 allows for estimating an event that will occur in the future by comparing histories of the scenario maps from the past to the present, Patent Literature 1 allows for visualizing the network graph of communication. Patent Literature 2 allows for extracting an associated co-occurrence network, and Patent Literature 3 allows for learning the concept hierarchy network. However, none of the above presents the future scenario itself.
It is a subject of the present invention to provide a network graph generation method of creating various scenarios that may occur in the future, and a decision-making support system supporting decision-making by presenting various network graphs that satisfy the satisfaction, principle for the uncertain future.
A typical example of the present invention is as follows. A network: graph generation method using a decision-making support system, wherein the decision-making support system includes a condition input reception function, a data acquisition function, a graph generation function, a simulation function, and a database; the method comprising: receives an input of a network graph generation condition; acquires data about a certain context based on the generation condition input thereto and accumulates the data in the database; generates a first network graph at a first time from the past to the present corresponding to the generation condition based on the acquired data about the certain context; generates a second network graph at a second time from the past to the present that differs from the first time corresponding to the generation condition based on the acquired data about the certain context; and generates a third network graph at a virtual third time based on the first network graph and the second network graph by simulation corresponding to the generation condition.
The present invention provides an effect of supporting the user to make a satisfactory decision for the uncertain future by creating various network graphs at new times, namely scenario maps, and presenting them to the user that is a decision-making entity.
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
The present invention provides a decision-making support service that satisfies the satisfaction principle for the uncertain future to achieve an autopoietic decision-making support system in cooperation between a human and a computer by using big data and presenting various scenarios that may occur in the future as a bundle of possibilities to a human that is a decision-making entity.
According to a typical embodiment of the present invention, a decision-making support system includes a means for generating first and second network graphs constituted by vertices and edges or nodes and links at several time points from the past, to the present, generating a third network graph of another virtual time based thereon, and presenting them as scenario snaps. More preferably, it sets the virtual time in the future ahead of the present and presents a future scenario map.
The means for presenting the first and second network graphs from the past to the present as well as the newly generated third network graph for the future, graphically displays various network graphs according to time specification made by a time slider or selection of the generation method.
The means for generating the network graph for the future generates various third network graphs for the future by, for example, developing the first and second network graphs into the future based on the change (difference) from the past to the present, or by growing, deriving, alternating, or disturbing them.
With a client-server system for generating the network graph, a client inputs a data acquisition condition and a simulation condition, a server generates the first and second network graphs from the past to the present based on the data and newly generates the third network graph of the virtual time using a simulation, and also displays the first to third network graphs on the client.
Hereinafter, embodiments of the network graph generation method according to the present invention will be described in detail with reference to drawings.
The multiple servers 100-10n constitute a distributed processing system using the server 100 as a master and the servers 101-10n as workers, and include multiple processors 110-11n, multiple memories 120-12n, and multiple network interfaces 180-18n, respectively. The memories 120-12n are equipped with multiple programs 130-13n for having a computer (processor) implement various functions respectively. That is, they include data acquisition programs 140-14n for having the computer implement the data acquisition function, multiple network graph generation programs 150-15n for implementing the graph generation function, and multiple simulation programs 160-16n for implementing the simulation function, on multiple distributed processing platforms 170-17n, respectively.
It should be noted that various types of simulation programs are installed on each of the multiple servers 100-10n to enable multiple types of simulations based on different prediction methods and presentation of various network graphs of the virtual third time.
The client 20 includes a processor 21, a memory 22, a network interface 28, and a display 29. The memory 22 is equipped with multiple programs 23 for having the computer (processor) implements various functions. That is, it includes a data acquisition condition input unit 24 and a simulation condition input unit 25 for having the computer implement the generation condition input, reception function, and a network-graph display condition input unit 26 and a network-graph display unit 27 for having the computer implement the display condition input reception function, as an interface of a user terminal 50.
The client 20 can make the user terminal 50 and the decision-making support system 1 interactively cooperate with each other by having the user terminal 50 input the data acquisition condition, input the simulation condition or select the simulation method, and input the network-graph display condition. The network-graph display unit 27 displays the generation result of the network graph on the screen of the display 29.
The network 30 connects the multiple servers 100-10n to the multiple network interfaces 180-18n, 28 of the client 20 to constitute the client-server system. The database 40 stores therein the data from the past to the present and supplies the data required for the decision-making to the multiple servers 100-10n via the network 30.
At Step 203, the server (master) 100 receives the data acquisition condition 21 and the multiple simulation conditions 25 from the client 20, and develops them into the distributed processing platforms 170-17n of the multiple servers 100-10n.
At Step 204, the multiple servers (workers) 101-10n execute data acquisitions 140-14n to match the data acquisition condition 24 from the database 40. The database 40 includes various data such as a text, an image, a video, and sensor data depending on an object of the decision-making, and those data are systematized in the form of the “context” or “content” contained therein. As a data acquisition method, it is also possible to make use of a web search engine and social media. For example, the multiple servers (workers) can access an external web search engine via the network and collets the data matching the data acquisition condition. In this manner, the data about a certain context from the past to the present can be acquired based on the input acquisition condition 24.
At Step 205, the multiple servers (workers; 101-10n execute the network graph generation at a first time t1 in the past based on the acquired data 140-14n (first graph generation) and the network graph generation at a second time t2 (present or near present) from the past to the present (second graph generation; 150-15n.
The network graph as the first time t1 in the past indicates a real history or the fact actually occurred at the first time t1 in the past generated as a scenario map or a network graph from the past to the present. Moreover, the network graph at the second time t2 can also be created, for example, by a technique of generating a scenario map based on the fact and the history occurred from the past to the second time t2.
As a means for generating the network graph at the first time t1 and the second time t2, the techniques described in Nonpatent Literature 1 and Nonpatent Literature 2 may also be used.
As described in detail in the following embodiment, the first time t1 and the second time t2 respectively include one or multiple time points such as a first time period (t11-t1n and a second time period (t21-t2n), respectively. By using these multiple data pieces of the first time t1 and the second time t2, the accuracy of the simulation can be improved and various network graphs can be generated that satisfy the satisfaction principle at the virtual third time or in the third time period (t31-t3n).
At Step 206, the servers (workers) 101-10n execute the simulations 160-16n that match the simulation condition 25 based on the acquired data 140-14n, and at Step 207, the servers (workers) 101-10n execute the network graph generation (third graph generation) 150-15n at the virtual third time t3 (optional past or future time) not included in the acquired data 140-14n from the simulation execution results 160-16n.
As multiple types of simulation methods based on different prediction methods for generating the network graph at the time t3 in she future (or optional past), for example, a method of performing a statistical prediction using an autoregression model or a moving-average model based on a time-series change in the frequency and the co-occurrence degree in the data from the past to the present (historical drift, growth), a method of adding analogical data and associated data to the initial data acquisition condition (phylogeny, derivation), a method of alternating a data co-occurring pair with data having a higher co-occurrence degree (genetic mutation, heterogeness), and a method of causing a critical state of the data using a sandpile model and an earthquake model in the track of complex system approaches to the natural world or societies (selection, disturbance) are useful. By combining multiple types of simulations based on these different prediction methods, various possibilities for the future can be presented.
At Step 208, the server (master) 100 integrates the network graph generation results 150-15n at the Step 205 and Step 207 from the servers (workers) 101-10n, and at Step 209, the server (master) 100 transmits the network graph generation results (first to third graph generation results; 150-15n to the client 20.
At Step 210, the client 20 receives the network graph generation results 150-15n, and in response to sensing it, at Step 211, the user inputs the network-graph display condition 26 from the client 20 through the terminal 50.
At Step 212, the client 20 executes the network graph display 27 on the display 23 according to the display condition 26, and presents the network graph generation results (first to third graph generation results) 150-15n at the first time t1 and the second time t2 in the past and the virtual third time t3 not included therein, to the user 50 as the scenario map.
At Step 213, if it is necessary for the user 50 to change the network graph display condition 26, the process returns to Step 211, or if not necessary it proceeds to Step 214.
At Step 214, if the user is satisfied with the presented result of the network-graph display result 27, namely the scenario map, as a choice for the decision-making, the process proceeds to the next Step 215 to be terminated, or if the user is not satisfied, the process returns to Step 202 to perform the network graph generation (first to third graph generations) again.
In this manner, she first network graph at the first time from the past to the present and the second network graph at the second time different from the first time are generated based on the acquired data and the generation condition, and furthermore, based on the first network, graph and the second network graph, the simulation corresponding to the generation condition is executed to generate the third network graph at the virtual third time.
An exemplary display screen 60 in
At Step 211, when the user specifies a time (black portion) on the time slider 62 via the terminal 50, the network graph generation results (first to third graph generation results) 150-15n corresponding to the time are extracted and the network graph display 27 is executed at Step 212.
In the exemplary display screen 60, regarding the given context, the left screen displays a network graph 71 (first graph generation result) at the first time t1 in the past specified by the time slider 62, the center screen displays a network graph 72 (second graph generation result) at the second time t2 specified by the time slider 62, and the right screen displays a network graph 73 (third graph generation result) at the third time t3 specified by the time slider 62.
The network-graph display unit 27 outputs and displays the network graph generated result on the display screen 60 of the display 29. The network graphs 71-73 on the exemplary display screen 60 in
According to the network graph generation method described in the first embodiment, by the decision-making support system 1 presenting the new network graph 73 at the virtual time (third time t3) along with the network graphs 71, 72 at the first time t3 from the past to the present and the second time t2 on the terminal 50 of the user as the scenario map for the given context, there can be an increased range of choices for the decision-making for the uncertain future restricted by the bounded rationality and improved effect of supporting the user's concern and understanding. That is, by creating the various network graphs or scenario maps at the virtual time (third time) and presenting them to the user, there is an effect of supporting the satisfactory decision-making for the uncertain future.
Although the network graphs 71-73, namely the scenario maps, are generated with the data frequency depicted as the vertex and the co-occurrence degree as the edge in the first embodiment, they may be depicted as a potential map or a mind map.
Although the decision-making support system 1 includes the distributed processing platforms 170-17n to use the big data, the simulation programs 160-16n operating thereon need to generate the network graph constituted by a huge amount of vertices and edges, and therefore a multi-agent simulation and an asynchronous parallel computation with an actor model are suitable. Moreover, although the client-server system is constituted to have the client 20 serve as a user interface and hove the servers 100-10n perform a computation for the network graph generation, it is also possible to have multiple clients perform a distributed processing and the invention is not limited to the system configuration described in the first embodiment.
The first embodiment can achieve the decision-making support system allowing for an interactive cooperation between a human and a computer using the big data by introducing the client-server system.
Moreover, by introducing the time slider 62 as a method of the network-graph display condition input 26, it is possible to continuously comprehend the transition of the network graphs 71-73 spanning from the past to the present and then to the future, thereby deepening an insight for the future. By presenting the scenario maps panoramically or locally changing not only the time but also the time range (time period) with the time slider 62 or by displaying the scenario maps as a movie by automatically forwarding the time, new concern can be more easily induced in the decision-making. That is, by presenting the various scenario maps as alternative choices using the big data, the degree of freedom for the decision-making is increased and more opportunities for the chance discovery can be provided, and displaying the scenario maps according to the time slider or the choices supports the concern and the understanding of the human that is the decision-making entity.
As a second embodiment, an example is given that is more concrete than the first embodiment using text data as an object of the “context,” thereby showing a method of developing a network graph into the future.
A flowchart 300 starts from Step 301 of inputting a condition setting.
At Step 302 in
After Step 303, the process diverges into four simulations of growth, derivation, alternation, and disturbance depending on the condition of the future scenario set by the user.
At Step 304 (growth) of the future scenario, the frequency and the co-occurrence degree for toe future are estimated by simulation based on the time-series analysis of the frequency and the co-occurrence degree from the past to the present (Step 305), and the simulation is repeated until the termination condition is satisfied. When the termination condition is satisfied (YES at Step 306), the network graph of the context scenario map from the past to the future (third graph=growth) is generated (Step 307), the network graph is displayed according to the display condition (Step 308), arid the process proceeds to Step 309 to be terminated. The prediction technique based on the simulation may be selected from the regression analysis method, the moving-average method, the exponential, smoothing method, and the like, and the periodicity and the causal effect may be taken into account.
At Step 310 (derivation), a word with a high frequency or a high co-occurrence degree is added to search words for a re-search (Step 314), the frequency and the co-occurrence degree as a result of the re-search are calculated by the morphological analysis (Step 315), and the process returns to Step 314 depending on the simulation condition to repeat the re-search. When the simulation is terminated (YES at Step 316), the network graph taking the repetition of the re-search as a time evolution for the future (third graph-derivation) is generated (Step 317), the graph is displayed (Step 313), and the process is terminated at Step 309.
At Step 320 (alternation), the re-search is performed using a co-occurring pair of words (Step 324), the two words constituting the co-occurring pair are alternated with a highly co-occurring word other than the words(Step 325), and the process returns to Step 324 according to the simulation condition to repeat the research. When the simulation is terminated (YES at Step 326), the network graph using the repetition of the alternation at Step 321 as the time evolution for the future (third graph=alternation) is generated (Step 327), the graph is displayed (Step 328), and the process is terminated at Step 309.
At Step 330 (disturbance), the frequency of the word, is accumulated randomly or stochastically (Step 334). If the frequency of the word exceeds a threshold according to a predetermined rule, the frequency is distributed to its co-occurring word depending on the co-occurrence degree (Step 335), and the process returns to Step 334 according to the simulation condition to repeat the accumulation. Although this method follows the sandpile avalanche model in complex systems, the earthquake model or the like may be otherwise used. When the simulation is terminated (YES at Step 336), the network graph using the repetition of the accumulation as the time evolution for the future is generated (Step 337), the graph (third graph=disturbance) is displayed (Step 338), and the process is terminated at Step 309.
A network graph 420 at the step shown in
In the network graph 420 at the step shown in
Next,
According to the second embodiment, by creating various network graphs or scenario maps at new time points, there is an effect of performing more satisfactory decision-making for the uncertain future. Moreover, by presenting various scenario maps as alternative choices, the degree of freedom for the decision-making is increased and more opportunities for the chance discovery can be provided, and displaying the scenario maps according to the time slider or the choices supports the concern and the understanding of the human that is the decision-making entity. Furthermore, the decision-making support system can be achieved that allows for the interactive cooperation between a human and a computer using the big data.
Especially according to the network graph generation method for the future scenario shown in the second embodiment, based on the scenario maps from the past to the present, by historically developing the data along the trend or the periodicity at Steps 304-308 (growth), systematically differentiating the data at Steps 310-318 (derivation), genetically alternating generations at Steps 320-328 (alternation), and causing the natural selection at Steps 330-338 (disturbance), it is possible to present various scenario maps that may occur in the future as network graphs (growth, derivation, alternation, disturbance), which are useful for the decision-making support service and context-aware service.
Although the network graphs are generated based on the analogy of ecosystem in the second embodiment, another approach such as a pattern language or a game theory may be introduced to the network graph generation. Moreover, although the explanation is given taking an example of the text data as the object of the “context,” the graph generation method according to the second embodiment or based on a similar simulation can be extended to other time-series data of stock prices, distribution, traffic, earthquakes, and the like, design pattern data of a city, a building, software, and the like, and network data of a social medium, a community, an enterprise organization, and the like.
A third embodiment of the present invention describes another example of a display screen of a network graph 500 generated by the processing according to the first and second embodiments and displayed by the display 29.
Network graphs 510 are multiple first graphs generated based on the history data from the past to the present, network graphs 511 are multiple second graphs generated based on the history data from the past to the present, and network graphs 521-524 are multiple future scenario graphs (third graphs) generated based on the history data or the network graphs 510, 511, which are diverged variedly depending on the possibilities that may occur in the future.
Multiple network graphs 512, 513 are the graphs (third graphs) of the past that could have occurred, which are generated based on the multiple history data 510, 511 from the past to the present or going back from the present situation, and network graphs 531-533 are the graphs (third graphs) spanning from the past that could have occurred to the future that can possibly occur, which are generated based on the third graphs 512, 513.
Although the time axis 501 in the third, embodiment indicates the flow of the time from the past to the future and the network graphs 510, 511 are displayed along the time axis 501 of the absolute time, the network graphs 512, 513, 521-524, 531-533 may be displayed along the time axis 501 of the absolute time or the relative time depending on the graph generation method.
The third embodiment provides the similar effects to the first and second embodiments.
Especially, according so the third embodiment, by generating the network graphs 510-513, 521-524, 531-533 according to the data acquisition condition and the simulation condition and displaying them on the screen of the display 29 according to the graph display condition, it is possible to visualize various future scenarios to contribute to the chance discovery and the decision-making.
A fourth embodiment of the present invention describes another example of the display screen of the network graph displayed on the display 29 of the client in the first embodiment.
Displayed on the display 601 in
To change the default setting, it suffices to select an option from the menu bar 620. A start date (year-month-day), an end date (year-month-day), and an interval date (year-month-day) are input to a pull-up menu 621 for the search condition, checkboxes of unification of the letter type, unification of the synonyms, an unnecessary word filter, and a user specification are selected as a processing of the searched text data in a pull-up menu 622 for the processing condition, and checkboxes of the growth, the derivation, the alternation, the disturbance, and the user specification are selected as the simulation condition in a pull-up menu 623 for the future scenario.
Displayed on a network-graph display unit 630 of the display 601 in
The network graph 631 is displayed according to the specification of the time from the past to the present and to the future by the time slider 640, according to the specification of the playback, step forward, fast forward, reverse playback, step backward, rewind, stop, or pause by the action buttons 641, and according to the checkboxes of the growth, derivation, alternation, disturbance, and user specification selected by the future scenario selection unit 642.
The fourth embodiment also provides the similar effects to the first to third embodiments.
Especially, by interactively entering the data acquisition condition, the simulation condition, and the graph display condition from the client terminal 600 through tire display 601 as described in the fourth embodiment and thereby associating the search in the scenario map with the decision-making with the client or the human and the computer cooperating with each other, it is possible to achieve the autopoiesis system developing into the future.
Although a graphic user interface of a tablet terminal or a mobile terminal is assumed as the client terminal 600 described in the fourth embodiment, other human-computer interaction may be used such as a nonverbal interface based on audio and gestures, a multi-user interface for cooperative activities, and a virtual reality interface.
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/JP2012/080945 | 11/29/2012 | WO | 00 |
