State transition analysis system, method, and program

Abstract

To provide a system capable of automatically explaining relationship between time-series data of a plurality of factors.SOLUTION: A state transition analysis system comprises: an analysis part 5 which executes time-series analysis on time-series data of a plurality of factors based upon source information to mathematize transition of the respective factors and generate data on transition patterns of the respective factors; and a relation verification part 6 which generates data on relationship between the respective factors based upon knowledge related to factors stored in an information DB 8 and the transition patterns of the respective factors. The system further has an explanation part 7 which generates an explanation sentence for the relationship through natural language processing based upon the transition patterns and relationship of the respective factors in the form of the data and outputs it to an output device.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present invention relates to a state transition analysis and analysis system, and particularly to a state transition analysis and analysis system that analyzes or analyzes the state transition that can be grasped from the original information by a program corresponding to the original information.

Background Art

[0002] Conventionally, there is a system in which a fixed program (such as a predetermined data analysis program) corresponding to each text data is linked in advance, and a computer selects a fixed program corresponding to the input text data and executes predetermined data analysis processing or the like.

[0003] However, the analysis or analysis process that can be performed corresponding to each input data is fixed and lacks flexibility.

[0004] Also, in the field of data analysis processing, a person observes a plurality of time-series data (graph images) and verifies the relationship between each factor shown therein. However, as the number of related factors increases, the manual verification of the relationship becomes more complicated and involves a great deal of labor.

[0005] [[ID=二十三]] The applicant is unaware of prior art documents related to the system.

Summary of the Invention

Problems to be Solved by the Invention

[0006] An object of the present invention is to improve the disadvantages of the above conventional example, and particularly to enable execution of a more flexible analysis or analysis program than before for each original information in the analysis or analysis of state transitions that can be grasped from the original information.

[0007] Another object is to provide a system that can automatically explain the relationship of time-series data of multiple factors.

Means for Solving the Problems

[0008] To solve this problem, the present invention includes a component linkage information storage unit that stores component linkage information linking raw information that is the subject of state transition analysis or interpretation and program components that perform various state transition analysis or interpretation processes. It also includes a program configuration unit that refers to the component linkage information, selectively acquires a plurality of program components linked to the raw information, and connects the acquired program components to form an executable program. Furthermore, it includes a program execution unit that executes the executable program configured by the program configuration unit and outputs the execution result.

[0009] In addition, the program execution unit includes an analysis unit that performs time-series analysis on time-series data of multiple factors based on the source information, formulas the changes of each factor into mathematical equations, and converts the change patterns of each factor into data; and a relationship verification unit that converts the relationships between each factor into data based on the knowledge about the factors stored in the information DB and the change patterns of each factor. Furthermore, it has an explanation unit that generates an explanatory text of the relationships based on the digitized change patterns and relationships of each factor using natural language processing and outputs it to an output device. [Effects of the Invention]

[0010] According to the present invention, in the analysis of state transitions that can be grasped from raw information, a more flexible analysis or analysis program can be executed for each piece of raw information than in the conventional method.

[0011] Furthermore, it can automatically explain the relationships that exist between time-series data of multiple factors. [Brief explanation of the drawing]

[0012] [Figure 1] Figure 1 is a system configuration diagram showing one embodiment of the present invention. [Figure 2] Figure 2 is a system configuration diagram showing Embodiment 1 of the present invention. [Figure 3] Figure 3 is a graph showing the raw information for Example 1. [Figure 4] Figure 4 is a graph showing the raw information for Example 2. [Figure 5] Figure 5 is a diagram showing an example of the data stored in the information database shown in Figure 2. [Figure 6] Figure 6 is a diagram showing an example of the data stored in the analysis result storage unit of Figure 2. [Figure 7] Figure 7 is a diagram showing an example of the data stored in the standard phrase memory unit of Figure 2. [Modes for carrying out the invention]

[0013] One embodiment of the present invention will be described below with reference to Figure 1. The state transition analysis system shown in Figure 1 is a computer system and comprises an input device 1 such as a keyboard, an output device 2 such as a display device, a processing unit (processor) 3 such as a CPU, and a storage device 4 such as a hard disk drive.

[0014] The processing unit 3 comprises an analysis / analysis program configuration unit 31 and an analysis / analysis program execution unit 32. These units 31 and 32 are realized when the processing unit 3 executes a predetermined program.

[0015] The analysis / analysis program configuration unit 31 configures a program for analyzing or performing state transitions by selecting and connecting multiple program components according to the source information. The analysis / analysis program execution unit 32 then executes the program configured by the analysis / analysis program configuration unit 31 and performs state transition analysis or processing on the source information.

[0016] The storage device 4 comprises a source information storage unit 41, a component association information storage unit 42, and a program component storage unit 43. Each of the storage units 41, 42, and 43 is provided in the storage area of ​​the storage device 4. However, the program component storage unit 43 is not limited to the local storage device 4, but may be provided in an external storage device (not shown) on a computer network. In that case, the external program component storage unit 43 is used via a communication device such as a NIC.

[0017] The original information storage unit 41 stores the original information to be the target of analysis of state transition. The original information is, for example, data including elements representing state transition from a certain state to a subsequent state, graph images, other images, natural sentences, or the like. The original information is acquired from the input device 1 or a communication device (not shown) and stored in the original information storage unit 41. Further, the natural sentence as the original information may be generated by the state transition analysis / analysis system itself and stored in the original information storage unit 41.

[0018] The program component storage unit 43 stores a library of a plurality of program components that execute various analysis processes or analysis processes used for analysis of state transition. The program components are, for example, software components corresponding to an API (Application Programming Interface). Each program component performs, for example, the following processes.

[0019] · A process of analyzing gradients, differences, distributions, etc. · A process of finding state transitions using Fourier functions or mathematical models for frequency analysis · A process of performing correlation analysis after organizing time, location, etc. to find related factors · A process of expressing and outputting the results of analysis or analysis in a program, algorithm, or natural sentence · A process of finding correlations · A process of performing cause analysis · A process of performing extreme comparison · A process of comparing rates of change · A process of calculating the degree of deviation from a specific factor · A process of formulating state transitions · A process of creating a state transition model · A process of creating a natural sentence regarding state transition · A process of organizing the relevance of analyzed or analyzed items and elements (factors) · A process of structuring (for example, formulating) the organized relevance

[0020] The component linking information storage unit 42 stores component linking information that links source information with program components corresponding to that source information. Several forms of linking source information and program components are possible. For example, component linking information may be obtained by linking eigenvalues ​​that can be extracted from source information with eigenvalues ​​of each program component. Alternatively, component linking information may be obtained by linking eigenvalues ​​that can be recognized from source information according to the type of source information with eigenvalues ​​of program components. Eigenvalues ​​according to the type of source information (for example, data, graph images, other images, or natural language) may be obtained as output by a trained AI (artificial intelligence) model in response to source information input.

[0021] The component linking information may link the unique values ​​of multiple program components in the order in which those program components are executed. Alternatively, the linking of the unique values ​​(nodes) of multiple program components may be formed in a graph (network) structure, and weights may be assigned to the linking between these nodes according to the execution order of the program components. Such a data structure can be managed, for example, by a graph database. Furthermore, the linking of the unique values ​​of the source information to the unique values ​​of the program components, and the linking of the unique values ​​of multiple program components, may be performed manually by an operator using an input device 1 or the like.

[0022] Next, the operation of this embodiment will be described. First, the analysis program configuration unit 31 reads the original information from the original information storage unit 41 and obtains the unique values ​​extracted or recognized from the original information. Subsequently, it refers to the component association information storage unit 42 and obtains the unique values ​​of one or more program components associated with the unique values ​​of the original information, and, if there are multiple program components, information on their execution order. Subsequently, it selects and obtains the program component corresponding to the obtained unique values ​​of the one or more program components from the program component storage unit 43. If only one program component is obtained, it generates that program component as an executable program. On the other hand, if multiple program components are obtained, it generates an executable program by connecting the multiple program components in the order of execution.

[0023] The analysis program execution unit 32 executes the execution program generated by the analysis program configuration unit 31. For example, if a graph image is read as source information, and the program components associated with the graph image are for mathematical formulating, state transition model creation, and natural language creation, the execution of the program makes it possible to achieve flexible processing such as mathematically formulating the state transitions shown in the graph image, creating a state transition model, and then creating and outputting an explanatory text about the state transitions based on the state transition model.

[0024] Similarly, by linking program components for analysis or interpretation according to the source information, flexible processing becomes possible. For example, for a graph image as source information, the gradient, differences, and distribution of the graph can be analyzed, state transitions can be found using Fourier functions or mathematical models for frequency analysis, correlation analysis can be performed after organizing time, location, etc., and related factors can be identified. The results can then be output in the form of a program, algorithm, or natural language. [Examples]

[0025] As an example of implementation, we present a system that measures the changes in a graph showing the blood concentration of drug A and the changes in a graph showing CRP levels, and explains the relationship between the two.

[0026] [System Configuration] The configuration of this embodiment is shown in Figure 2. Parts identical to those in the above embodiment are denoted by the same reference numerals, and redundant explanations are omitted.

[0027] The graph explanation system shown in Figure 2 comprises a source information storage unit 41, a graph analysis unit 5, a graph relationship verification unit 6, a graph explanation unit 7, and an output device 2. It also includes an information database 8, an analysis result storage unit 9, and a standard phrase storage unit 11.

[0028] In this embodiment, the original information storage unit 41 stores time-series data and graph images. The time-series data includes, for a given subject, the daily values ​​of the blood concentration of drug A, the daily values ​​of the CRP (C-reactive protein) value, and the daily values ​​of the blood concentration of drug C. The graph images include, as a first image, an image (Figure 3) that overlays a line graph showing the daily trend of the blood concentration of drug A and a line graph showing the daily trend of CRP from the time-series data. As a second image, in addition to the data in Figure 3, an image (Figure 4) that further overlays a line graph showing the daily trend of the blood concentration of drug C from the time-series data.

[0029] The graph analysis unit 5 performs time series analysis on the time series data read from the original information storage unit 41 or the time series data extracted from the graph image, and converts the trends of drugs A and C and CRP values ​​into data.

[0030] The graph relationship verification unit 6 analyzes the relationships between drugs A and C and the changes in CRP values, which have been converted into data by the graph analysis unit 5, and converts these relationships, such as correlations and inverse correlations, into data. The graph relationship verification unit 6 also determines whether the converted relationships are consistent with known knowledge data about drugs A and C and CRP that are pre-registered in the information database 8.

[0031] The graph explanation unit 7 receives data representing the trends and relationships between drugs A and C and CRP values, which have been deemed consistent by the graph relationship verification unit 6, and generates and outputs natural language text incorporating this data using natural language processing.

[0032] [Software Configuration] In the prototype of this embodiment, we utilized open-source modules such as standard modules or extension modules for the Python programming language. Each of these modules (libraries) corresponds to a program component in the above embodiment. On the other hand, the code may be written not only in Python, but also in Julia, Node.js, C++, etc.

[0033] The graph analysis unit 5, graph relationship verification unit 6, and graph explanation unit 7 described above operate when the processor executes a program. As in the previous embodiment, this program is composed of the analysis / interpretation program configuration unit 31 and executed by the analysis / interpretation program execution unit 32.

[0034] [Operation of each part] In this example, the time-series data or graph image shown in Figure 3 is used as the source information. As described above, Figure 3 shows a superimposed line graph of the daily changes in the blood concentration of drug A and a line graph of the daily changes in CRP. The horizontal axis represents the number of days, with one division representing one day, and the vertical axis represents the blood concentration of drug A and the CRP value. For the CRP value, one division represents 1.0.

[0035] Information DB8 pre-stores information about drug A and information about CRP values. In this embodiment, as shown in Figure 5, information A for drug A records data indicating that the "effect" of administering drug A is "inflammation suppression." Information B for CRP records data indicating that "CRP" is an "inflammation value," that an "increase" in CRP indicates "inflammation," that a "decrease" in CRP indicates "inflammation suppression," and that the "reference range" for CRP is "0.3 or less." In addition, information C for drug C records data indicating that the "effects" of drug C are "inflammation suppression" and "pain suppression."

[0036] When the graph analysis unit 5 reads a graph image without real values ​​from the original information storage unit 41, it uses a library such as OpenCV to detect the contours of each graph contained in the image and obtains time-series data of each graph using wavelength analysis such as Fourier transform and wavelet transform.

[0037] The graph analysis unit 5 performs time series analysis on time series data read from the original information storage unit 41 or time series data obtained from images, and converts each time series data into a mathematical formula (model). Then, it determines the pattern of changes in drug A and CRP based on this formula. Libraries such as CatBoost, LightGBM, SymPy, NumPy, Pandas, and scikit-learn are used for the time series analysis.

[0038] In this embodiment, the graph analysis unit 5 is said to have obtained the following patterns. Specifically, the pattern for drug A is data showing that the "increase" occurs on "day 4", the "peak" occurs on "day 7", and the "administration" period is "5 days". The pattern for CRP is data showing that the "increase" occurs on "day 3", the "peak" occurs on "day 4", the "decrease rate" is "10%", the "convergence" occurs on "day 9", and the "convergence value" is "0.4". The graph analysis unit 5 stores these patterns in the analysis result storage unit 9.

[0039] Next, the graph relationship verification unit 6 uses the patterns stored in the analysis result storage unit 9 and the information A and B stored in the information DB 8 to determine the following relationships.

[0040] (1) Based on the above pattern and information B, the converged value of CRP 0.4 is "close" to the reference range of 0.3 or less (the difference is within a specified value).

[0041] (2) Based on information A and B, there is a causal relationship that the administration of "drug A" has an "anti-inflammatory effect" and therefore the "CRP value" "decreases".

[0042] (3) In comparison to (2) above, the fact that "drug A" is administered when "CRP" is "decreasing" (after the peak) in the above pattern is consistent. (Verification of causal relationship between correlation) When the graph relationship verification unit 6 determines that there are no inconsistencies, it passes the above pattern data and the verification result data representing the above relationship to the graph explanation unit 7.

[0043] The graph description unit 7 selects corresponding parameters to be embedded in a predefined text from the pattern data and verification result data received from the graph relationship verification unit 6, and automatically generates a predefined text for the time-series data or graph image of the original information using natural language processing. The predefined text storage unit 11 stores various predefined texts, some of which are variables, as shown in Figure 7, for example. In this embodiment, for example, the graph description unit 7 generates the following predefined text and outputs it to the output device 2. On the third day, CRP levels rose, so drug A was administered. After being administered for "5 days," its "anti-inflammatory effect" caused the "CRP value" to "close to the normal range" by the "9th day." The processing in this graph description section 7 uses libraries such as Markovify, N-grams Counter, MeCab, Bert, and GPT3.

[0044] According to this embodiment, it is possible to automatically generate and output explanatory text for time-series data and graph images. [Examples]

[0045] In this example, in addition to the data from Example 1, we will explain the processing when time-series data (graph) for drug C is also included, as shown in Figure 4. When the rate of decrease in CRP values ​​is higher than in Figure 3, after adding the trends for drug C to the graphs for drug A and CRP, an explanation of the graph will be output stating that the difference is due to the effect of drug C. The system configuration is the same as in Example 1.

[0046] As shown in Figure 5, the information DB8 contains data indicating that the "effects" of administering drug C are "inflammation suppression" and "pain suppression," as information C regarding drug C.

[0047] The graph analysis unit 5 performs time-series analysis on the time-series data of drug A, CRP, and drug C, and converts each graph into a mathematical formula. Then, it stores the data showing the patterns obtained from each formula in the analysis result storage unit 9.

[0048] In this embodiment, data exhibiting the following pattern will be stored. • The increase in "CRP" occurred on the "third day". The increase in "drug A" occurred on the "4th day". The peak of "CRP" is on "day 4". • A 10% decrease in "CRP" occurs on "day 5". • The increase in "drug C" occurred on the "5th day". • A 30% decrease in "CRP" occurs "after day 6". • The "convergence" of "CRP" is on "day 8". The convergence value of "CRP" should be "0.32".

[0049] The graph relationship verification unit 6 determines that there are no inconsistencies in the following relationships based on the patterns stored in the analysis result storage unit 9 and the knowledge information of drugs A and C and CRP stored in the information DB 8.

[0050] (1) Administration of "drug A" has an "anti-inflammatory effect" and the "CRP value" will "decrease" by "10%".

[0051] (2) Administration of "drug A" and "drug C" has an "anti-inflammatory effect" and the "CRP value" is "decreased" by "30%".

[0052] (3) The difference in "CRP value" and "decrease rate" is due to the administration of "drug C".

[0053] (4) Administration of "drug C" has an "anti-inflammatory effect" and causes "CRP levels" to "decrease".

[0054] When the graph relationship verification unit 6 determines that there are no inconsistencies, it passes the above pattern data and the verification result data representing the above relationship to the graph explanation unit 7.

[0055] The graph description unit 7 selects corresponding parameters to be embedded in a standard text from the pattern data and verification result data received from the graph relationship verification unit 6, and automatically generates a descriptive text for the original time-series data or graph image using natural language processing. In this embodiment, for example, the graph description unit 7 generates the following descriptive text and outputs it to the output device 2. On the third day, the CRP level increased, so drug A was administered. The CRP level decreased. When drug C was administered from day 5, the rate of decrease in CRP levels increased. Due to the anti-inflammatory effects of drugs A and C, the CRP level decreased to near the normal range on the 8th day.

[0056] As described above, this embodiment makes it possible to automatically generate explanatory text for the causal relationships between three or more time-series data. When the number of drug types or blood test items increases, the analysis of causal relationships becomes complex and difficult for humans, but this problem can be solved.

[0057] Herein, the scope of the present invention is limited to the scope of the invention described in the claims, and is not limited to the above embodiments and examples. While suitable for the medical field, such as pharmaceuticals, it can be widely used in other fields as well. [Explanation of Symbols]

[0058] 1 Input device 2 Output device 3 Processing Unit 4 Storage device 5. Graph Analysis Section 6. Graph Relationship Verification Section 7. Graph Explanation Section 8 Information DB 9 Analysis result storage section 11 Fixed phrase memory section 31 Analysis and Analysis Program Configuration Section 32 Analysis and Analysis Program Execution Unit 41 Original information storage unit 42. Component linking information storage unit 43 Program component storage unit

Claims

1. A component linkage information storage unit stores component linkage information that links the original information to be analyzed or processed for state transitions with program components that perform various analysis or processing of state transitions. A program configuration unit that refers to the aforementioned component linking information, selectively acquires a plurality of program components linked to the aforementioned original information, and connects the acquired program components to construct an executable program, The system comprises a program execution unit that executes the executable program configured by the program configuration unit and outputs the execution result, The program execution unit, An analysis unit performs time series analysis on time series data of multiple factors based on raw information, converts the changes of each factor into mathematical formulas, and converts the change patterns of each factor into data. A relationship verification unit that converts the relationships between the factors into data based on the knowledge of the factors stored in the information DB and the transition patterns of each factor, An explanatory unit generates an explanatory text of the relationship between each of the digitized factors using natural language processing and outputs it to an output device. A state transition analysis and interpretation system that operates the system.

2. Using component linking information that links the raw information to be analyzed or processed in state transitions with program components that execute various analysis or processing steps for state transitions, A program configuration step involves referencing the aforementioned component linking information to selectively acquire a plurality of program components linked to the aforementioned source information, and connecting the acquired program components to construct an executable program. A program execution step which executes the configured executable program and outputs the execution result, In this program execution step, The analysis step involves performing time series analysis on time series data of multiple factors based on the original information, formulating the changes of each factor into mathematical equations, and converting the change patterns of each factor into data. A relationship verification step that converts the relationship between each factor into data based on the knowledge of the factors stored in the information DB and the transition patterns of each factor, An explanation step in which, based on the transition patterns and relationships of each of the digitized factors, an explanatory text describing the relationship is generated by natural language processing and output to an output device, A state transition analysis method performed by a computer.

3. Using component linking information that links the raw information to be analyzed or processed in state transitions with program components that execute various analysis or processing steps for state transitions, A program configuration step involves referencing the aforementioned component linking information to selectively acquire a plurality of program components linked to the aforementioned source information, and connecting the acquired program components to construct an executable program. A program execution step which executes the configured executable program and outputs the execution result, In this program execution step, The analysis step involves performing time series analysis on time series data of multiple factors based on the original information, formulating the changes of each factor into mathematical equations, and converting the change patterns of each factor into data. A relationship verification step that converts the relationship between each factor into data based on the knowledge of the factors stored in the information DB and the transition patterns of each factor, An explanation step in which, based on the transition patterns and relationships of each of the digitized factors, an explanatory text describing the relationship is generated by natural language processing and output to an output device, A state transition analysis program that causes a computer to perform the following actions.

4. An analysis unit performs time series analysis on time series data of multiple factors based on raw information, converts the changes of each factor into mathematical formulas, and converts the change patterns of each factor into data. A relationship verification unit that converts the relationships between the factors into data based on the knowledge of the factors stored in the information DB and the transition patterns of each factor, An explanatory unit generates an explanatory text of the relationship between each of the digitized factors using natural language processing and outputs it to an output device. A state transition analysis and analysis system equipped with the following features.

5. The state transition analysis system according to claim 4, wherein the source information is a graph image showing the transitions of multiple factors, and the time-series data is extracted from the graph image.

6. The analysis step involves performing time series analysis on time series data of multiple factors based on the original information, formulating the changes of each factor into mathematical equations, and converting the change patterns of each factor into data. A relationship verification step that converts the relationship between each factor into data based on the knowledge of the factors stored in the information DB and the transition patterns of each factor, An explanation step in which, based on the transition patterns and relationships of each of the digitized factors, an explanatory text describing the relationship is generated by natural language processing and output to an output device, A state transition analysis method performed by a computer.

7. The state transition analysis method according to claim 6, wherein the source information is a graph image showing the transitions of multiple factors, and the time-series data is extracted from the graph image.

8. The analysis step involves performing time series analysis on time series data of multiple factors based on the original information, formulating the changes of each factor into mathematical equations, and converting the change patterns of each factor into data. A relationship verification step that converts the relationship between each factor into data based on the knowledge of the factors stored in the information DB and the transition patterns of each factor, An explanation step in which, based on the transition patterns and relationships of each of the digitized factors, an explanatory text describing the relationship is generated by natural language processing and output to an output device, A state transition analysis program that causes a computer to perform the following actions.

9. The state transition analysis program according to claim 8, wherein the source information is a graph image showing the transitions of multiple factors, and the time-series data is extracted from the graph image.

Images