JP2015118466A - Information processing apparatus and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make it easy to reach the conditions and conclusion of a useful branch from a decision tree.SOLUTION: Creation means 102 creates, from data acquired by acquisition means 101, a decision tree showing data on a second item of an object, which is predicted when data on a first item of the object satisfies the conditions. Extraction means 103 extracts, from the decision tree created by the creation means 102, the rule that when the data on the first item satisfies the branch conditions the data on the second item is obtained. Calculation means 104 calculates, from the data acquired from the acquisition means 101, the degree of significance indicating the degree of significance of the rule extracted by the extraction means 103. Output means 105 outputs list data showing a list of the decision tree created by the creation means 102, the rule extracted by the extraction means 103, and the value calculated by the calculation means 104.

Description

  The present invention relates to a technique for supporting utilization of data using a decision tree.

  There are technologies to support the utilization of data using decision trees. For example, Patent Document 1 defines a plurality of types of medical images, branching conditions (for example, pixel values in pixel units of medical images), and presence / absence or types of lesion candidates determined based on these conditions. A technique for assisting a doctor to grasp a lesion candidate from a medical image using the determined decision tree is described.

JP 2009-61170 A

In the technique of Patent Document 1, a decision tree in which a branch condition useful for deriving a conclusion (in this technique, the presence or absence of a lesion candidate or its type) is used in advance is used. If the conclusion is not known, a decision tree may be generated from the data, and useful branch conditions and conclusions may be found and used. In that case, it is difficult to determine which conditions and conclusions are useful even if only a decision tree showing only the conditions and conclusions described in Patent Document 1 is viewed.
Accordingly, an object of the present invention is to make it easier to find useful branch conditions and conclusions from a decision tree.

  The present invention is predicted when the data of the first item of the target satisfies the condition from the acquisition unit that acquires the data of the plurality of items regarding each of the plurality of targets and the data acquired by the acquisition unit A generation unit that generates a decision tree representing data of the second item of interest, and a decision tree generated by the generation unit when the data of the first item satisfies the condition Extraction means for extracting the rule that the data of the matter is obtained, and calculation means for calculating the significance indicating the degree of significance of the rule extracted by the extraction means from the data acquired by the acquisition means; , An output for outputting list data indicating a list of the decision tree generated by the generation unit, the rules extracted by the extraction unit, and the significance calculated by the calculation unit An information processing apparatus and a stage.

  Further, the acquisition means acquires data of the plurality of items at a first time and a second time after the first time, and the generation means acquires the first time acquired by the acquisition means. The decision trees are generated from the data of the plurality of items at the first and second periods, respectively, the extraction unit extracts the rules for each of the first and second periods, and the output unit includes: Of the generated decision tree or the extracted rules, image data representing the maintained portion that has been maintained from the first time to the second time and the changed portion that has changed with images of different modes The list data may be output.

Further, the output means may output the image data representing an image of the aspect in which the change portion is more easily recognized than the maintenance portion.
Further, the output means, when there are a plurality of data of the second matter in the decision tree, represents the data differently, and satisfies the common condition in the first and second periods When there is common second item data, the image data representing the same second item data as well as the same distance from the root of the decision tree may be output.

In addition, when there are a plurality of the rules, the output means represents the rules different from each other, and the second means common when the common conditions are satisfied in the first and second periods. In the case where there is a common rule that data of the item can be obtained, the image data represented by adding a common code to the common rule may be output.
In addition, when there are a plurality of the common rules, the output means determines the first common rule and the second common rule at any one of the first time and the second time. Also, image data that is separated by the same number of rows may be output.
The calculating means may calculate the extracted p-value (probability value) of the rule as the significance.

  The present invention provides an acquisition step of acquiring a plurality of items of data related to each of a plurality of targets in a computer, and when the data of the first item of the target satisfies a condition from the data acquired in the acquisition step A generation step of generating a decision tree representing data of the second item of interest to be predicted, and a decision tree generated in the generation step, when the data of the first item satisfies the condition An extraction step for extracting the rule that the data of the second matter is obtained, and a significance level representing the degree of significance of the rule extracted in the extraction step is calculated from the data acquired in the acquisition step. A calculation step; a decision tree generated in the generation step; a rule extracted in the extraction step; It provides a program for executing an output step for outputting a list data indicating a list of significant degrees calculated at step out.

  According to the present invention, it is possible to easily find useful branch conditions and conclusions from a decision tree.

1 shows an overall configuration of an information providing system 1 according to a first embodiment. The figure which shows the hardware constitutions of the information processing apparatus 10 The figure which shows an example of the function structure which the information processing apparatus 10 implement | achieves Diagram showing an example of the generated decision tree Diagram showing an example of extracted rules Figure showing an example of calculated significance Diagram for explaining how to calculate the chi-square value Diagram showing an example of a list of displayed decision trees, rules, and significance levels Sequence diagram showing an example of the operation of each device in the acquisition process Sequence diagram showing an example of operation of each device in output processing The figure which shows the example of the image displayed when list request | requirement operation is performed The figure which shows an example of the image data output The figure which shows another example of the image data output

[1] First Embodiment [1-1] Overall Configuration and Hardware Configuration FIG. 1 is a diagram showing an overall configuration of an information providing system 1 according to the first embodiment of the present invention. In the present embodiment, the information providing system 1 is a system for comprehensively providing medical information to medical personnel such as doctors and nurses. The information providing system 1 includes an information processing device 10, a network 2, and a plurality of server devices 4 and a plurality of client devices 5 provided in each medical facility 3. The network 2 is a system that mediates exchange of data between devices, and is, for example, a LAN (Local Area Network) provided in the medical facility 3 and a WAN (Wide Area Network) that connects these LANs. The network 2 may include the Internet or a wireless communication line.

  The server device 4 is a computer that executes a function of a system (hereinafter referred to as “medical system”) used for medical work. The medical system is, for example, an electronic medical record system, an automatic inspection apparatus connection system, a bacteria inspection / navigation system, or the like. The server device 4 is connected to the network 2 and exchanges data related to the medical system via the network 2.

  The client device 5 is a computer that stores and executes a program for medical personnel to use the medical system, and is, for example, a personal computer or a tablet terminal. The client device 5 is connected to the network 2 by wire or wireless (in the example of FIG. 1, by wire), and exchanges data via the network 2. The client device 5 includes display means for displaying an image, and displays an image related to the medical system. In addition, the medical staff operates the client device 5 and inputs data related to the medical system. In addition, the client device 5 stores and executes a browser program to display the above-described medical information on the display unit. Thereby, the information regarding medical care is provided to the medical staff. The browser program may also serve as a program for using the medical system.

  The information processing apparatus 10 is a computer having a function of processing information. The information processing apparatus 10 is connected to the network 2 and collects medical data from a plurality of server apparatuses 4 via the network 2. The information processing apparatus 10 provides a data analysis support service that supports analysis of collected data to the user of the client apparatus 5, that is, a medical worker.

  FIG. 2 is a diagram illustrating a hardware configuration of the information processing apparatus 10. The information processing apparatus 10 includes a control unit 11, a storage unit 12, and a communication unit 13. The control unit 11 includes a central processing unit (CPU), a read only memory (ROM), a random access memory (RAM), and a real time clock. The CPU controls the operations of the storage unit 12 and the communication unit 13 by executing a program stored in the ROM or the storage unit 12 using the RAM as a work area. The real time clock has a function of calculating the current date and time. The storage unit 12 includes a hard disk or the like, and stores data, programs, and the like used by the control unit 11 for control. The communication unit 13 includes a communication circuit for performing communication via the network 2.

[1-2] Functional Configuration The information processing apparatus 10 performs an analysis support process that is a process for supporting data analysis performed by a medical worker based on the hardware configuration described above. The storage unit 12 stores a program for performing analysis support processing. The control unit 11 executes the program to control each device shown in FIG.
FIG. 3 is a diagram illustrating an example of a functional configuration realized by the information processing apparatus 10. The information processing apparatus 10 includes an acquisition unit 101, a generation unit 102, an extraction unit 103, a calculation unit 104, and an output unit 105.

[1-2-1] Acquisition Unit The acquisition unit 101 is an example of a unit that acquires data on a plurality of items (hereinafter referred to as “related data”) regarding each of a plurality of objects. In the present embodiment, the target is a patient, and the related data is data related to the patient stored in the above-described medical system. Specifically, the patient ID (Identification) for identifying the patient, the patient's age, symptom, examination result, number of operations, and the like. As the related data, for example, data submitted to the secretariat of Nosocomial Infections Surveillance (JANIS) may be used.

  The related data is stored in the server apparatus 4 shown in FIG. In this embodiment, each server device 4 stores the server device 4 at, for example, a predetermined time interval (for example, every month) or a predetermined date and time (for example, the first and the 15th of every month at midnight). Related data is transmitted to the information processing apparatus 10. The acquisition unit 101 acquires the related data transmitted in this way. The acquisition unit 101 may transmit request data for requesting related data to each server device 4 and acquire the related data transmitted in response to the request data. The acquisition unit 101 stores the acquired related data so that the generation unit 102 and the calculation unit 104 can refer to the acquired related data.

[1-2-2] Generating Unit The generating unit 102 predicts the data of the second item of the target predicted from the data acquired by the acquiring unit 101 when the data of the target first item satisfies the condition. Is an example of a means for generating a decision tree representing The data of the first item is also called an explanatory factor, for example, and the data of the second item is also called an objective factor, for example. In the present embodiment, the generation unit 102 sets the item called “outcome” representing the survival and death of the patient as the second item, “age classification”, “degree of burn”, “presence / absence of airway burn”, “sepsis” The decision tree is generated with five items of “presence / absence of” and “number of operations” as the first items. The age classification is data in which the patient's age is classified as less than 10 years old, 10 years old or older and 40 years old or less, 41 years old or older and less than 70 years old, 70 years old or older. The data is represented by 30 or less, 31 or more and 59 or less, or 60 or more. The generation unit 102 refers to the related data stored by the acquisition unit 101, and determines a decision tree from the data of the first and second items using a known algorithm such as an ID3 (Iterative Dichotomiser 3) algorithm. Generate.

  FIG. 4 is a diagram illustrating an example of the generated decision tree. In this example, two branches branch from the root of the decision tree (also referred to as “root”) under the condition of “sepsis”. As described above, the condition used in the decision tree represents a condition for branching, and is hereinafter referred to as a “branch condition”. From the branch “sepsis = none”, three branches are shown that branch under the branch condition “degree of burn”. In the first branch, it is shown that the data of the second item “survival” can be obtained when “the degree of burn = 10 or less”. The “survival” in this case is referred to as target second data, and is also referred to as “leaf” in the decision tree. The numerical value of “20-> 18” shown after the leaves indicates that there are 20 patients who satisfy the branching conditions of “sepsis = none” and “degree of burn = 10 or less”, of which 18 are This indicates that data of the second item “survival” has been acquired. That is, it represents the probability that the target data of the second item is obtained when the data of the first item satisfies the branch condition. Similarly, it is shown that the data of the second item “survival” was obtained with a probability of “14−> 14” even when “the degree of burn = 11-30 (11 to 30)”. Has been.

  In the case of “degree of burn = 31-59”, three branches that further branch under the branch condition “age classification” are shown. The first branch shows that “survival” was obtained with a probability of “3-> 3” in the case of “age classification 10-40 (10 to 40 years old)”. Subsequently, in the second branch, when “age classification 41-70” is indicated, it is indicated that “death” was obtained with a probability of “2-> 2”, and when “age classification is less than 10” It is shown that “survival” was obtained with a probability of “1-> 1”. In addition, in the branch “sepsis = present”, two branches that branch under the branch condition “presence / absence of airway burn” are shown. In the case of “airway burn = none”, it is shown that “survival” was obtained with a probability of “6-> 5”, and in the case of “airway burn = present”, “death” was “6- It is shown that it was obtained with a probability of> 5 ”. The generation unit 102 supplies decision tree data indicating the generated decision tree to the extraction unit 103 and the output unit 105.

[1-2-3] Extraction Unit The extraction unit 103 determines from the decision tree generated by the generation unit 102 that the data of the second item is obtained when the data of the first item satisfies the branch condition. It is an example of a means for extracting. For example, the extracting unit 103 determines the rule of an if-then rule composed of an if part representing a branch condition and a then part representing a conclusion obtained when the if part is true, that is, when the branch condition is satisfied. The data is extracted from the decision tree indicated by the decision tree data supplied from the generation unit 102.

  FIG. 5 is a diagram illustrating an example of the extracted rules. In this example, the extraction unit 103 extracts an if-then rule in which the top leaf of the decision tree shown in FIG. 4 is the then part and the branch connecting the leaf and the root is the if part. Specifically, the rule is extracted that the conclusion “then survival” is obtained when “if sepsis = none & degree of burn = 10 or less”. The extraction means 103 extracts the rules for the other leaves shown in FIG. The extraction unit 103 supplies rule data indicating the extracted rule to the calculation unit 104 and the output unit 105.

[1-2-4] Calculation Unit The calculation unit 104 is an example of a unit that calculates, from the data acquired by the acquisition unit 101, a significance level that represents the degree of significance of the rules extracted by the extraction unit 103. . The fact that a rule is significant means that it is unlikely that the rule stood by chance and that the rule is considered meaningful. In other words, the significance level is a degree indicating a low possibility that the rule is accidentally established, and in other words, a degree indicating a high possibility that the rule is meaningful. The calculation unit 104 refers to the related data stored in the acquisition unit 101, and from the data of the first and second items, the significance of each rule indicated by the rule data supplied from the extraction unit 103 Is calculated.

  FIG. 6 is a diagram illustrating an example of the calculated significance level. In the example of FIG. 6, the calculation unit 104 calculates “probability”, “unexpectedness”, and “p value” as significance levels. The calculation means 104 calculates the value of 18 ÷ 20 = 0.9 (90%) as “probability” in the case of the leaf “20−> 18” described above. In other words, the calculation means 104 is the number of objects for which the data of the first item satisfies the branching conditions (in this example, the branching conditions of “sepsis = none” and “degree of burn = 10 or less”), of which the target A value obtained by dividing the number of objects for which data of the second item (in this example, “survival”) is obtained is calculated as the significance (in this case, “probability”).

  Further, the calculation unit 104 calculates “unexpectedness” indicating the degree of unexpected occurrence as the significance level. For example, the probability that this is expected to be 1 when rolling a dice (this is called “expected probability”) is 1/6. If 1 actually comes out of 6 times when a dice is shaken, the probability (this is called “actual occurrence probability”) is 1/6. The ratio of these probabilities is called unexpectedness, and in this example, unexpectedness = probability (1/6) ÷ expected probability (1/6) = 1.00. The unexpectedness is greater than 1.00, indicating that something unexpected is happening. For example, if 1 comes out 5 times out of 6 times, unexpectedness = probability of occurrence (5/6) ÷ expected probability (1/6) = 5.00, which is more unexpected than the previous example. Represents that. In the example of FIG. 6, the calculation unit 104 calculates the unexpectedness as 234.852. The greater the unexpectedness, the less likely that the event happens by chance, and the greater the correlation between the “if” part and the “then” part in that event.

Further, the calculation means 104 calculates a p-value (probability-value) used as a significance level when determining whether to reject a hypothesis that the rule is not correct (so-called null hypothesis). The calculation means 104, in this embodiment, calculates the p value using the chi-square value X ² ₀ .
FIG. 7 is a diagram for explaining a method of calculating the chi-square value. In FIG. 7, for a certain rule, the number of subjects whose “if part is true” and “then part is true” is “a”, “if part is true” and “then part is not true”. Is “b”, “if part is not true” and “then part is true”, the number of cases is “c”, “if part is not true” and “then part is not true”. ”Is shown in a 2 × 2 cross table. In this table, the total of each row and column (such as “a + b”) and the total (“a + b + c + d”) are shown. For example, the calculation unit 104 calculates the chi-square value X ² ₀ using the following equation (1).

The calculating means 104 calculates the p value from the chi-square value X ² ₀ calculated in this way using a predetermined calculation formula. In the example of FIG. 6, the calculation unit 104 calculates a p value of 0.18180. Note that the calculation unit 104 is not limited to this, and may calculate the p-value by another known method. The calculation means 104 supplies significance data indicating the significance, probability, and p-value calculated in this way to the output means 105.

[1-2-5] Output Unit The output unit 105 is a list showing a list of the decision tree generated by the generation unit 102, the rules extracted by the extraction unit 103, and the values calculated by the calculation unit 104. It is an example of a means for outputting data. For example, when receiving the request data requesting the list data transmitted from the client device 5, the output unit 105 generates list data based on the decision tree data, the rule data, and the significance data supplied from each unit. The generated list data is output to the client device 5. The client device 5 displays the list indicated by the output list data on the display unit by using a browser function, for example.

  FIG. 8 is a diagram illustrating an example of a list of displayed decision trees, rules, and significance levels. In this example, the character string “data analysis support service” is displayed in the title bar A1 of the browser, the decision tree column A3 in which the decision tree is displayed below the menu bar A2, and the rules in this embodiment. An if-then rule column A4 displaying a certain if-then rule and the significance calculated for each rule is displayed. The decision tree column A3 and the if-the rule column A4 are displayed side by side.

  In FIG. 8, the decision tree having seven leaves shown in FIG. 4 is displayed in the decision tree column A3, and seven rules corresponding to each leaf are displayed in the if-then rule column A4. In the if-then rule column A4, there are also a character string “outcome” as an objective factor, a character string such as “age classification” as an explanatory factor, and related data used for generating the decision tree. The number of related objects is displayed. As described above, when there are a plurality of pieces of data of the second item in the decision tree, the output unit 105 outputs list data representing these data in different sequences (in this case, different rows). Further, when there are a plurality of rules, the output unit 105 outputs list data representing the rules in different sequences (similarly different rows).

[1-3] Operation The information processing apparatus 10 performs the analysis support process described above based on the above configuration. The analysis support process includes an acquisition process for acquiring related data and an output process for outputting list data generated based on the acquired related data. Below, the operation | movement which each apparatus with which the information provision system 1 is provided in an acquisition process and an output process demonstrates with reference to FIG.9 and FIG.10, respectively.

  FIG. 9 is a sequence diagram illustrating an example of the operation of each device in the acquisition process. In FIG. 9, two of the plurality of server devices 4 are shown, but the server devices 4 other than these perform the same operations. First, for example, when an operation for inputting information related to a medical system is performed by a user of the medical system, the server device 4 receives the input operation (step S11). The information input here is patient vital data, for example. The server device 4 stores related data according to the accepted input operation (step S12), and transmits the related data to the information processing device 10 at a predetermined time interval or a predetermined time (step S13). .

  Another server device 4 executes a predetermined process of the medical system, for example (step S21). Next, the server device 4 stores the result of the process executed in step S21 as related data (step S22). The related data stored here is, for example, a detection result of an infectious disease pathogen. Subsequently, the server apparatus 4 transmits related data to the information processing apparatus 10 in the same manner as in step S13 described above (step S23). In steps S13 and S23, the information processing apparatus 10 receives the received related data, and acquires the received related data (step S31). Steps S13, S23, and S31 are operations performed by the acquisition unit 101.

  FIG. 10 is a sequence diagram illustrating an example of the operation of each device in the output process. In FIG. 10, one of the plurality of client devices 5 is shown, but the other client devices 5 perform the same operation. The output process is started when the user of the information providing system 1 performs an operation (hereinafter referred to as “list request operation”) for requesting a list of decision trees, rules, and significances from the client device 5. . Details of the list request operation will be described with reference to FIG.

  FIG. 11 is a diagram illustrating an example of an image displayed on the client device when a list request operation is performed. In FIG. 11A, an image for selecting related data used for generating a decision tree is displayed. In this example, a list image B1 that is a list of related data that can be selected, a selection data image B2 that indicates the selected related data, and related data are selected from the list image B1, or related data is selected from the selected data image B2. An operator image B3 for canceling the selection is displayed. In FIG.11 (b), the image for selecting the relevant data made into a target factor from the selected relevant data is displayed. In this example, six related data of “presence / absence of sepsis”, “degree of burn”, “presence / absence of airway burn”, “age classification”, “number of operations” and “outcome” are selected. A selected data image B2 in which “Outcome” is checked and an operator image B4 including a character string “Determine” are displayed.

  When the user performs an operation of selecting the operator image B4 in the state shown in FIG. 11B, the client device 5 accepts this operation as a list request operation (step S41), and the objective factor (the second matter) Data) is “outcome” data, and explanatory factors (data of the first item) are “sepsis presence / absence”, “degree of burn”, “presence / absence of airway burn”, “age classification” and “number of operations” Request data for requesting a list of decision trees to be generated and corresponding rules and significances is transmitted to the information processing apparatus 10 (step S42).

  When the information processing apparatus 10 receives the request data from the client apparatus 5 in step S42, the information processing apparatus 10 generates a decision tree using the related data of the objective factor and the explanatory factor indicated by the request data (step S43). Step S43 is an operation performed by the generation unit 102. Next, the information processing apparatus 10 extracts rules from the generated decision tree (step S44), and calculates the significance for each extracted rule (step S45). Step S44 is an operation performed by the extraction unit 103, and step S45 is an operation performed by the calculation unit 104. Subsequently, the information processing apparatus 10 generates list data using the generated decision tree, the extracted rules, and the calculated significance (step S46), and outputs the generated list data to the client apparatus 5. (Step S47). Steps S46 and S47 are operations performed by the output means 105. Upon receiving the list data in step S47, the client device 5 displays a list indicated by the received list data (step S48).

[1-4] Effects of First Embodiment In this embodiment, list data indicating a list in which decision trees, rules, and significance levels are collected is output. The decision tree shows the overall picture of how the explanatory factor branches from the root and leads to the leaf, that is, the target factor. In the decision tree, for example, an explanatory factor closer to the root tends to have a larger correlation with the target factor. In the example of FIG. 4, the correlation with the “outcome” as the objective factor is largest in “presence / absence of sepsis”, followed by “degree of burn” and “presence / absence of airway burn”, and then “age classification” Continuing, “the number of operations” has the smallest correlation. Thus, the decision tree is useful for grasping the magnitude of the correlation between the explanatory factor and the objective factor.

  Further, in the example of FIG. 5, the rule is that the relationship between each leaf (conclusion) and the branch (branch condition) leading to it in the decision tree is as follows: explanatory factor (if part) and objective factor (then part) In a simple manner, the significance level indicates how statistically the conclusion obtained in each rule is. By looking at these three pieces of information together, the user can compare the magnitude of the correlation between each explanatory factor and the objective factor and the degree of statistical significance (significance) of each conclusion. The usefulness of branch conditions and conclusions can be determined. According to the present embodiment, it is possible to make it easier to find useful branch conditions and conclusions from a decision tree than when no list data is output.

  In the present embodiment, the p value is used as the significance. For example, when the p-value is less than 0.05, it is common to reject the null hypothesis described above, that is, the hypothesis that the extracted rule is not correct, but that means that the correct rule is hardly extracted. If so, less than 0.1 may be rejected, or less than 0.01 may be rejected if the significance is judged more strictly. Thus, according to the present embodiment, it is possible to easily determine how accurately the null hypothesis is rejected as compared with the case where the p value is not used.

[2] Second Embodiment Hereinafter, a second embodiment of the present invention will be described focusing on differences from the first embodiment. In the first embodiment, the relationship between the data of the first and second items and the time is not particularly mentioned, but in the second embodiment, the related data at two different times is used. The related data at each time is related data generated at the time. The length of time used in the present embodiment varies depending on the related data used. For example, if there is the presence or absence of sepsis or the degree of burn used in the example of FIG. In addition, if the data is related to infectious diseases, the situation changes frequently, so a period of several days to several weeks or one month is used.

[2-1] Configuration of the Second Embodiment In the present embodiment, the acquisition unit 101 acquires data of a plurality of items in the first period and the second period after the first period. The first and second periods are, for example, last month and last month. When the acquisition unit 101 acquires related data, for example, by storing date / time data indicating the acquired date / time together, it is indicated at what time the acquired related data is acquired. So that The generation unit 102 generates a decision tree from the data of a plurality of items at the first and second times acquired by the acquisition unit 101. The generation unit 102 generates the decision tree in this way by using the stored related data group for each period. The extraction unit 103 extracts the rules for each of the first and second periods.

  The output means 105 is a portion that is maintained from the first time to the second time (hereinafter referred to as “maintain”) of at least one of the generated decision tree and the extracted rule (both in the present embodiment). Image data representing the image of the aspect different from the “part” and the changed part (hereinafter referred to as “change part”) is output as list data. In the present embodiment, the output means 105 outputs image data representing an image in a form in which the changed portion is more easily recognized than the maintained portion.

[2-2] Overview of Second Embodiment FIG. 12 is a diagram illustrating an example of output image data. In this example, a decision tree column A5 displaying a decision tree generated using related data in the previous month and an if-then rule column A6 displaying rules extracted from the decision tree are displayed in the upper half. On the lower side, a decision tree column A7 displaying the decision tree generated using the related data in the previous month and an if-then rule column A8 displaying the rules extracted from the decision tree are displayed. The decision trees, rules, and significance levels are underlined in different portions, that is, changed portions on the upper side (last month) and the lower side (last month). As described above, the output unit 105 outputs image data representing an image in a mode in which the underlined change portion is more easily recognized than the maintenance portion that is not underlined. As a result, compared with the case where such image data is not output, the changed portion is more easily recognized than the maintained portion, and the user can grasp the changed portion that has changed from the first period to the second period. It can be easy.

  In the example of FIG. 12, the number of leaves is the same as 6 in the last month and the month before, but the branching conditions are partially different. For example, the decision tree of last month shows the branching conditions of “sepsis = none”, “degree of burn = 31-59” and “age classification = less than 10”, but it is shown in the decision tree of last month. Instead, the branching conditions “sepsis = none” and “degree of burn = 11-30”, which were not shown in the decision tree of the previous month, are shown. In this example, the branch condition rows that are represented only in one of these decision trees are represented by the same row in the other decision tree.

  For this reason, for example, the leaf “survival” that satisfies the branch condition “sepsis = yes” and “airway burn = none”, that is, the objective factor is the root (“*** decision tree: last month” (Or the month before)) The 9th row is counted from the row where the character string “***” is represented). As described above, when there is a common objective factor that satisfies the common branch condition in the first and second periods, the output unit 105 has the same distance from the root of the decision tree and the common objective factor. Output the image data to represent. By outputting the image data indicating such a decision tree, it is generated using the related data in the first and second periods, respectively, as compared with the case where the common objective factors are expressed in different distances from the root. This makes it easier to compare the decision trees.

  Moreover, in the example of FIG. 12, the common code | symbol is attached | subjected and represented about the rule in which a branch condition and an objective factor are common. For example, a branching condition of “sepsis = none” and “degree of burn = 10 or less” and a rule represented by a leaf of “survival” are marked with “A”, and “sepsis = none” and “burn” The rule represented by the branch condition of “degree of degree = 11-30” and the leaf of “survival” is labeled “B”, and “sepsis = none”, “degree of burn = 31-59” and “ A branching condition “age classification = 10−40” and a rule represented by a leaf “survival” are labeled with “C”. As described above, when there is a common rule that the common objective factor can be obtained when the common branch condition is satisfied in the first and second periods, the output unit 105 determines the common rule. Image data represented by a common reference numeral is output. By outputting image data indicating such a rule, the rule and significance in the first and second periods can be compared with the case where these common rules are not assigned a common sign. It can be made easier to compare.

  In addition, in the if-then rule column A8, the rule B with the sign “B” and its significance are not displayed, but the corresponding lines are displayed with a space. For this reason, the rule A with the sign “A” and the rule C with the sign “C” are separated by the same number of rows in both the last month and the last month. become. Specifically, rule B and its significance are represented by two lines, and the top and bottom are vacant one by one, so rule A and C are 4 lines apart, which is the sum of those lines in both months It is expressed. As described above, when there are a plurality of rules common to the first and second periods, the output unit 105 determines that the first common rule (Rule A in the above example) and the second common rule. The image data expressing the thing (same rule C) is separated by the same number of rows (four lines in the above example) in both the first and second periods. Thereby, it is possible to make it easier to compare the rules and the significance levels in the first and second periods as compared with the case where the rules having the common signs are separated from each other by the number of different sequences.

[2-3] Another Example of Second Embodiment Note that the output unit 105 may output image data indicating an image having a mode different from that shown in FIG. For example, the output unit 105 may output image data representing an image in a mode in which the maintenance part is more easily recognized than the change part.
FIG. 13 is a diagram illustrating another example of output image data. In this example, contrary to the example of FIG. 12, the maintenance part is underlined and the change part is not underlined. As a result, compared to the case where image data indicating an image as shown in FIG. 12 is not output, the maintenance part is more easily recognized than the change part, and the user changes from the first period to the second period. It is possible to make it easier to grasp the maintenance parts that are not. As described above, the output unit 105 only needs to output image data representing the maintenance part and the change part with images of different modes. Thereby, compared with the case where such image data is not output, it is possible to make it easier to grasp which of the maintenance part and the change part the user has focused on.

  In the example shown in FIGS. 12 and 13, the output unit 105 makes it easier to recognize one of the changed portion and the maintained portion than the other by drawing an underline. However, the present invention is not limited to this. For example, the character may be easily recognized by enlarging the character or changing the color or thickness of the character. Moreover, you may make it easy to recognize by blinking a character or making it animation. In short, the output unit 105 may make it easier to recognize one than the other by changing the modes of the change part and the maintenance part.

  Further, the acquisition unit 101 may acquire not only related data at two periods as described above but also related data at three or more periods. In that case, the generation unit 102 generates a decision tree at each acquired time, and the extraction unit 103 extracts a rule from each of the generated decision trees. Further, the calculation unit 104 calculates the significance for each extracted rule. The output unit 105 outputs, for example, image data representing a maintenance part and a change part at different times before and after in different images. In this case, the output unit 105 performs output using the first and second timings as successive timings. Further, the output unit 105 may output image data representing the maintenance part and the change part when the latest time and the previous time are compared with different images. In this case, the output unit 105 performs output with the latest time as the second time and the previous time as the first time. As the time for comparison increases in this way, for example, it becomes easier to grasp the tendency between what is maintained and what is changed.

[3] Modified Examples Each of the above-described embodiments is merely an example of the embodiment of the present invention, and may be modified as follows. Moreover, you may implement each embodiment mentioned above and each modification shown below, combining each as needed.

[3-1] Target and related data In each of the above embodiments, data related to a patient is used as related data for the patient, but the present invention is not limited to this. For example, it is good also as the date and time, boarding distance, route | route, the weather at the time of boarding, etc. which board relevant data for the passenger of a transportation system. In addition, the related data may be sales, sales area, sales price, environment around the store, etc. with the target as a product. In short, any target related data may be used as long as the data represents the significance sufficient to generate a decision tree.

[3-2] Extraction Unit Although the extraction unit 103 has extracted the rules as many as the number of leaves of the decision tree in each of the above embodiments, the present invention is not limited to this. For example, if there is a branching condition in which one of the branching conditions is the other necessary condition, the extracting unit 103 may simplify them. For example, data of the first item of “under 10 years old” and “under 20 years old” is used as the age classification, and “survival” is performed under the branching conditions of “sepsis = none” and “under 10 years old” as the branching conditions. When the rule X that obtains the conclusion is extracted, and the rule Y that obtains the conclusion of “survival” under the branch condition of “sepsis = none” and “under 20 years old” is extracted, the branch condition of the rule X is determined This is a necessary condition for the branch condition of Y. In this case, the extraction unit 103 may extract only one rule that is simplified by combining the rule X with the rule Y.

  Note that the extraction unit 103 determines the simplified rule as described above only when the number of objects satisfying the branch condition is the same in the rules X and Y and the conclusion “survival” is obtained with the same probability. You may extract things. This means that the numerical values “◯-> □” shown in FIG. Further, in this case, the extraction unit 103 may combine not only the rule X into the rule Y but also the rule Y into the rule X.

[3-3] Calculation unit The calculation unit 104 may calculate significance other than the above. For example, the calculating unit 104 may calculate the chi-square value X ² ₀ used for calculating the p value as the significance level. In this case, if the chi-square value X ² ₀ is larger than the value (= 3.84) of the chi-square distribution table having a degree of freedom of 1 and a rejection value of 0.05 (= 3.84), the null hypothesis is rejected, so that it is statistically significant. It can be said that there is a relevance. In addition, any value may be calculated as the significance as long as it represents the level of significance of the rules extracted by the extraction unit 103.

[3-4] Location Image Output Destination The output unit 105 outputs the list data to the client device 5 in each embodiment described above. However, the present invention is not limited to this. If provided, the list data may be output to the display device. Further, when the user uses an electronic mail, the output unit 105 may output list data for the address of the electronic mail. Even if the list data is output to any destination, it is preferable that the user can browse the list indicated by the output list data.

[3-5] Character String Direction In the example of FIG. 8 or the like, the output unit 105 outputs the list data in which the decision tree, the rule, and the significance are shown in horizontal writing, but not limited thereto, it is shown in vertical writing. May be output. The output unit 105 may output each branch and each leaf of the decision tree using an image, or may output list data that more visually represents how each branch branches.

[3-6] Category of Invention In addition to the information processing apparatus and the information providing system including the information processing apparatus, the present invention can also be understood as an information processing method for realizing processing performed by the information processing apparatus. is there. The processing here is, for example, the analysis support processing (acquisition processing and output processing) shown in FIGS. 9 and 10. The present invention can also be understood as a program for causing a computer such as an information processing apparatus to function as each unit shown in FIG. This program can be provided in the form of a recording medium such as an optical disk in which it is stored, or provided in the form of being downloaded to a computer via a network such as the Internet, and making it available for installation. It may be done.

DESCRIPTION OF SYMBOLS 1 ... Information provision system, 2 ... Network, 4 ... Server apparatus, 5 ... Client apparatus, 10 ... Information processing apparatus, 11 ... Control part, 12 ... Memory | storage part, 13 ... Communication part, 101 ... Acquisition means, 102 ... Generation means , 103 ... extraction means, 104 ... calculation means, 105 ... output means.

Claims (8)

An acquisition means for acquiring data of a plurality of items related to each of a plurality of objects;
Generating means for generating, from the data acquired by the acquiring means, a decision tree representing data of the second item of the target predicted when the data of the first item of the target satisfies a condition;
Extraction means for extracting a rule that the data of the second item is obtained when the data of the first item satisfies the condition from the decision tree generated by the generation unit;
A calculating means for calculating a significance level representing the degree of significance of the rules extracted by the extracting means from the data acquired by the acquiring means;
An information processing apparatus comprising: a decision tree generated by the generation unit; an output unit that outputs list data indicating a list of the rules extracted by the extraction unit and the significance calculated by the calculation unit. The acquisition means acquires data of the plurality of items in a first period and a second period after the first period,
The generating means generates the decision trees from the data of the plurality of matters at the first and second times acquired by the acquiring means,
The extraction means extracts the rules for each of the first and second periods;
The output means has an image of a mode in which a maintenance portion maintained from the first time to the second time and a changed changed portion of the generated decision tree or the extracted rules are different. The information processing apparatus according to claim 1, wherein image data represented by is output as the list data. The information processing apparatus according to claim 2, wherein the output unit outputs the image data representing an image of the aspect in which the change portion is more easily recognized than the maintenance portion. When there are a plurality of data of the second matter in the decision tree, the output means represents the data differently, and in the first and second periods, the output means 4. The information according to claim 2, wherein when there is data of the second item, the image data representing the data of the common second item having the same distance from the root of the decision tree is output. 5. Processing equipment. When there are a plurality of rules, the output means represents the rules different from each other, and the second item is common when the common conditions are satisfied in the first and second periods. 5. When there is a common rule that data of the above can be obtained, the image data that represents the common rule with a common code is output. 5. Information processing device. In the case where there are a plurality of the common rules, the output means has the same first common rule and the second common rule in both the first time period and the second time period. The information processing apparatus according to claim 5, wherein image data that is separated by the number of lines is output. The information processing apparatus according to any one of claims 1 to 6, wherein the calculation unit calculates a p value (probability value) of the extracted rule as the significance. On the computer,
An acquisition step of acquiring data of a plurality of items related to each of a plurality of objects;
A generation step for generating a decision tree representing data of the second item of the target predicted when the data of the first item of the target satisfies a condition from the data acquired in the acquisition step;
An extraction step for extracting from the decision tree generated in the generation step a rule that the data of the second item is obtained when the data of the first item satisfies the condition;
A calculation step for calculating a significance level representing the degree of significance of the rules extracted in the extraction step from the data acquired in the acquisition step;
An output step means for outputting list data indicating a list of the decision tree generated in the generation step, the rules extracted in the extraction step, and the significance calculated in the calculation step; program.

Images