JP2021022264A - Text data analysis system, text data analysis method, and fault response recommend system - Google Patents

Abstract

To supply a user with information that contributes to efficiently solving a problem and improve the efficiency of solving a problem irrespective of experience.SOLUTION: A text data analysis system 100 for processing the text data of a plurality of documents includes an analysis server 120 comprising: an item extraction unit for extracting a "notification" and a "confirmation" item previously defined as text data in a document having a specific meaning from the text data of the plurality of documents; an item text classification unit for classifying the plurality of text data extracted as belonging to the same item into a plurality of categories; and a report analysis information creation unit for associating the categories into which the extracted text data was classified with the document from which the text data was extracted.SELECTED DRAWING: Figure 1

Description

The present invention relates to a technique for analyzing text data.

In social activities, documents with written reports and explanations are created. In modern times, many documents are stored in storage devices as text data. It is expected that a lot of knowledge will be obtained by analyzing the accumulated text data.

Data mining for text strings is also called text mining. Text mining is a text data analysis method that extracts useful information by dividing data consisting of ordinary sentences into words and phrases and analyzing the frequency of their appearance, the correlation of co-appearance, the tendency of appearance, and the time series. Is. In recent years, analysis using artificial intelligence has also been carried out.

For example, Patent Document 1 discloses a system for analyzing text information without defining a pattern.

JP-A-2018-18254

A typical document that has been converted into text data is a report. The report is a document that describes the background and results of the work, and is used daily by many companies and government offices.

There are various types of reports. For example, in what is called a "maintenance report" or "fault report", information that contributes to efficient problem solving can be obtained by analyzing these text data. By giving it to workers who perform maintenance or breakdowns, it is expected to improve the efficiency and accuracy of workers' work when problems such as breakdowns occur, regardless of the degree of experience of the workers.

However, since the text data of the report is created only for the purpose of "reporting", there is a problem that it is difficult to divert it to other purposes.

For example, a "maintenance report" or a "fault report" is often described in chronological order because it is necessary to accurately report the situation of the worker at the failure site. Therefore, it is difficult to identify and extract "information that contributes to efficient problem solving" from the text data in which this work is described in chronological order, and by extension, workers when problems such as breakdowns occur. There was a problem that it was difficult to improve the efficiency and accuracy of the work.

A preferred aspect of the present invention is a text data analysis system that processes text data of a plurality of documents. In this system, it is extracted as the same item as the item extraction unit that extracts the items of "notification" and "confirmation" that are defined in advance as text data having a specific meaning in the document from the text data of multiple documents. It is provided with an item classification unit that classifies a plurality of text data into a plurality of categories, and a report analysis information creation unit that associates the categories in which the extracted text data is classified with the document from which the text data is extracted. ..

Another preferred aspect of the present invention is a text data analysis method that processes text data of a plurality of documents. In this method, items defined in advance as text data having a specific meaning in a document are extracted from the text data of a plurality of documents, and the plurality of text data extracted as the same item are classified into a plurality of categories. Then, the category in which the extracted text data is classified is associated with the document in which the text data is extracted.

Another preferred aspect of the present invention is a failure response recommendation system that processes text data of a plurality of documents including contents related to a failure and proposes a response to the failure, from the text data of the plurality of documents in the document. The item extraction unit that extracts the items of "passive information" and "active information" that are defined in advance as text data with a specific meaning of, and multiple text data extracted as the same item are classified into multiple categories. Based on the analysis result of the report analysis information creation unit, it is provided with an item classification unit and a report analysis information creation unit that associates the categories in which the extracted text data is classified with the document from which the text data is extracted. It is a failure response recommendation system that outputs failure response recommendations.

It is possible to give the user information that contributes to efficient problem solving and improve the efficiency of problem solving regardless of experience.

The block diagram of the text data analysis system of an Example. Overall flow diagram of the system of the embodiment. Detailed flow chart of process S2 for extraction and classification of causes. The figure which shows the data structure example of the report DB112 with a cause category. The detailed flow chart of the process S3 of the extraction of other items. The figure which shows the data structure example of the report DB113 with an item tag. A detailed flow chart of the grouping process S4 for each of the other items. The figure which shows the data structure example of the report DB115 with item tag classification. A detailed flow chart of the cause estimation model creation process S5. The figure which shows the data structure example of the report analysis information DB116. The detailed flow chart of the operation process S6 of the cause estimation model. The conceptual diagram which shows the example of the cause estimation model 134 composed of a decision tree. The flow diagram which shows the example of another whole flow using the system of FIG. The figure which shows the example which made the data of the report analysis information DB116 into a list format. The flow diagram which shows the example of another whole flow using the system of FIG.

The embodiment will be described in detail with reference to the drawings. However, the present invention is not construed as being limited to the description of the embodiments shown below. It is easily understood by those skilled in the art that a specific configuration thereof can be changed without departing from the idea or purpose of the present invention.

In the configuration of the invention described below, the same reference numerals may be used in common among different drawings for the same parts or parts having similar functions, and duplicate description may be omitted.

When there are a plurality of elements having the same or similar functions, they may be described by adding different subscripts to the same code. However, if it is not necessary to distinguish between a plurality of elements, the subscript may be omitted for explanation.

Notations such as "first", "second", and "third" in the present specification and the like are attached to identify the components, and do not necessarily limit the number, order, or contents thereof. is not it. In addition, numbers for identifying components are used for each context, and numbers used in one context do not always indicate the same composition in other contexts. Further, it does not prevent the component identified by a certain number from having the function of the component identified by another number.

The position, size, shape, range, etc. of each configuration shown in the drawings and the like may not represent the actual position, size, shape, range, etc. in order to facilitate understanding of the invention. Therefore, the present invention is not necessarily limited to the position, size, shape, range, etc. disclosed in the drawings and the like.

The publications, patents and patent applications cited herein form part of the description herein.

Components represented in the singular form herein shall include the plural form unless explicitly stated in the context.

<1. System configuration>
FIG. 1 shows a constituent block of the text data analysis system of the embodiment. The system of this embodiment is basically composed of an information processing device including an input unit, an output unit, a control unit, and a storage unit. Examples of information processing devices are terminals such as tablets and servers.

The text data analysis system 100 of this embodiment includes a data storage server 110, an analysis server 120, an analysis model storage server 130, and a user terminal 140, each of which is composed of an information processing device. These are connected to each other by a wired or wireless network, and software resources for information processing can be used with each other.

The data storage server 110 is a server that mainly manages data. In this example, the report database (DB) 111, the report DB 112 with the cause category, the report DB 113 with the item tag, the item classification list DB 114, the report DB 115 with the item tag classification, and the report analysis information DB 116 are stored. The report database (DB) 111 stores the text data of the above-mentioned report.

The analysis server 120 is a server that mainly analyzes data, and also manages learning of an analysis model. In this example, the model generation unit 121, the cause category extraction unit 122, the item extraction unit 123, the item text classification unit 124, the report analysis information creation unit 125, and the analysis unit 126 exist as functional blocks.

In this embodiment, functions such as calculation and control are realized in cooperation with other hardware by executing a program stored in the storage device of each information processing device by the processing device. Will be done. A program executed by a computer or the like, its function, or a means for realizing the function may be referred to as a "function", a "means", a "part", a "unit", a "module", or the like. In FIG. 1, the functions of an information processing device such as a server are shown as functional blocks, and the well-known input unit, output unit, control unit, and storage unit are omitted.

The analysis model storage server 130 stores various analysis models used by the analysis server 120. In this example, the cause extraction model 131, the item extraction model 132, the item grouping unit 133, and the cause estimation model 134.

The user terminal 140 is a personal computer or a mobile terminal that the user directly operates. The control unit 141, which is a function of the user terminal, sends an instruction to the data storage server 110, the analysis server 120, and the analysis model storage server 130 via the network, and these hardware, software, and data can be used. ..
The details of the components listed above will be described in detail in subsequent examples.

The text data processed in this embodiment may be input by the user from the keyboard, may be converted into text by voice recognition, or may be automatically collected by a chatbot or the like. That is, the data of various DBs described in this embodiment regardless of the collection method can basically be cross-referenced by the document number. Therefore, it is possible to synthesize these even if there are a plurality of DBs, and conversely, it is also possible to divide a single DB into a plurality of DBs.

The above configuration is not limited to the configuration shown in FIG. 1, and may be configured by a single computer, or any part of the input device, output device, processing device, and storage device is connected by a network. It may be composed of other computers. These servers are not limited to physical servers, but may be in the form of cloud computing.

In this embodiment, the same function as the function configured by the software can be realized by the hardware such as FPGA (Field Programmable Gate Array) and ASIC (Application Specific Integrated Circuit). Such aspects are also included in the scope of the present invention.

<2. Overall flow>
FIG. 2 shows the overall flow of this system. Only the outline will be described here, and the details will be described in a later section.

In the process S1, the user accesses the report database (DB) 111 by the user terminal 140 and selects (a set of) the target report data (the text data of the report is simply referred to as "report" below. "). In addition, the extraction items to be extracted from the report are defined. The control unit 141 has, for example, a browser function, and the user can browse the data of the data storage server 110 and download the data.

As the target report, for example, it is desirable to select a report corresponding to a specific purpose and application such as "Maintenance report of equipment A". As the extraction items, for example, the user can freely define the items appearing in the “maintenance report of equipment A”.

In this embodiment, the information notified to the worker when a failure occurs, in other words, the information about the failure received by the worker is "notified", and the cause and countermeasure of the notified event are examined. We perform confirmation to collect necessary information for the purpose, "confirm (content)" the information about the content of the event obtained as a result of the confirmation, and "cause" the information about the cause derived by the worker based on the result of the confirmation. , Information on the measures planned and implemented based on the cause is defined as "countermeasures".

In some maintenance reports and failure reports, the work contents are described in chronological order or in free text format as described above, so the above "notification", "confirmation", "countermeasure", and "cause" Each item may be described sporadically, the same item may appear multiple times, or the content that spans multiple items may be text, making it difficult to extract.

However, the above four items are almost all the contents described in the above maintenance report and the like, and by analyzing these, the work efficiency of the worker can be dramatically improved. Is found for the first time, and this embodiment realizes this.

Both "notification" and "confirmation" are information obtained about the event, but "notification" is information obtained passively from the viewpoint of the subject who created the report, and "confirmation" is actively obtained. Information. For example, a "notification" is a failure report from a customer, that is, the content of an issue, and is often the first event in chronological order. In addition, "confirmation" is the content of the event recognized to solve the problem, and is passive information. Actions taken by on-site workers for recognition are sometimes called "instructions (contents)", but this is active information. In addition, "countermeasures" are actions taken by on-site workers to solve the notified problems, and this is active information. A "cause" is an event that would not have caused a problem without it, which is passive information.

In this way, each information can be roughly divided into two types, passive information (passive information) and active information (active information), and may include the passive information and the active information. , Consistent with the concept of failure response recommendation, which is to "recommend countermeasures (active events) when an event called failure (passive event) occurs". Therefore, by classifying and organizing the relationship between the passive information and the active information included in the report, it is possible to make an accurate recommendation when a failure or the like occurs.

In this embodiment, the items of "notification", "confirmation", "countermeasure", and "cause" described above will be described, but the present invention is not limited to this, and the number of items may be further increased. For example, items such as "report", "work content", "instruction", and "others" may be defined according to the purpose and character of the target report.

In the process S2, the item of "cause" is extracted from the target report and the "cause" is classified. The classification is user-defined. The "cause" item is often stated as the conclusion of the report, and is also the item that the user wants to know most. The item of "cause" can be extracted and classified by using, for example, DNN (Deep Neural Network). The classified "cause" category is added as a tag (cause label) to the text data of the report.

In the process S3, items other than the “cause” are extracted from the target report. In this embodiment, the items of "notification", "confirmation", and "countermeasure" are applicable. Extraction of items can be performed using, for example, DNN. As a result of the process S3, a set of text data extracted from the report is formed for each item.

In the process S4, the text data of each item extracted in the process S3 is grouped by similar items. Then, an ID (group name) that identifies the group of the items included in each report is added as a tag.

In process S5, a cause estimation model is created with the group name as the explanatory variable and the cause label as the objective variable.

In process S6, the cause estimation model is operated.

Although the above processes S1 to S6 are a series of processes, they can also be executed as independent processes. In that case, the processes S1 to S6 can be executed by different systems.

<3. Extraction and classification of causes (processing S2)>
The process following the process S1 will be described below. In the following description, only the target report data selected in the process S1 will be described. Data that has not been selected may also be stored in the report DB 111, but these are not mentioned for the sake of explanation.

FIG. 3 is a detailed flow chart of the cause extraction and classification process S2. For the sake of explanation, the processing performed by the user operating the user terminal 140 and the analysis server 120, the processing performed by the analysis model of the analysis model storage server 130, and the text data stored in the data storage server 110 are described separately (). The same applies below).

In the process S201, the user examines the target report selected in the process S1 and determines the category of the probable cause. Therefore, for example, the control unit 141 calls the corresponding report from the report DB 111, displays it on the user terminal 140, and examines the contents. The definition, content and type of the cause category are arbitrary, and can be arbitrarily determined by the user according to the target report.

It is desirable that the definition of the cause category be decided by an experienced user. For example, "electrical system failure", "mechanical system failure", "human error" and the like. Of course, the cause may be further subdivided. The determined cause category is stored in the cause category extraction unit 122 of the analysis server 120.

In process S202, the user extracts a report as teacher data. A part of the set of reports to be targeted selected in S1 may be arbitrarily selected. It may be randomly selected.

In the process S203, the user browses the reports as teacher data on the user terminal 140, and assigns a cause category to each report according to the definition defined in the process S201. The report with the cause category is stored in the report DB 112 with the cause category of the data storage server 110.

FIG. 4 is an example of the data structure of the report DB 112 with cause category. The text data 402 and the cause category 403 correspond to the document number 401 of the report. The data structure of the report DB 111 excludes the cause category 403 from the configuration of FIG.

Returning to FIG. 3, in the process S204, the model generation unit 121 inputs the pair of the text data 402 and the cause category 403 as the teacher data 404 into the cause extraction model 131.

In the process S205, the cause extraction model 131 before learning of the analysis model storage server 130 takes in the teacher data under the control of the model generation unit 121. The cause estimation model is not particularly limited, but is, for example, Bayesian DNN.

In process S206, the cause extraction model 131 performs supervised learning under the control of the model generation unit 121. A known method may be used as the learning method.

In the process S207, the text data of the report other than the teacher data is input to the learned cause extraction model 131 under the control of the cause category extraction unit 122. Although the cause category 403 is not attached to the report other than the teacher data, the cause extraction model 131 properly learned in the process S206 outputs the cause category 403 in response to the input of the text data 402 of the report. ..

In the process S208, the cause category extraction unit 122 transmits the output of the classification result to the user terminal 140. In process S209, the user checks the contents as necessary to determine the cause category of the report other than the teacher data.

By the above processing, the cause category is allotted to the report of the target data selected by the user in the process S1 by the user or the cause category extraction unit 122. This data format is similar to that shown in FIG. Under the control of the control unit 141, the user stores the report to which the cause category is assigned in the report DB 112 with the cause category of the data storage server.

<3. Extraction of other items (processing S3)>
FIG. 5 is a detailed flow chart of the process S3 for extracting other items.

In process S301, the user extracts a report as teacher data from the report database 111. This may be arbitrarily selected from the set of target data selected in S1. It may be different from the report selected in the process S202.

From the text data of the report 3010 as a teacher stored in the data storage server 110, the work contents 3011 described in chronological order are sent to the user terminal.

In the process S302, the user operates the user terminal 140 to tag an item for each line of the work content 3011 described in chronological order. Items to be tagged follow the definition made in process S1. In this example, they are "notification", "confirmation", and "countermeasure". The tagging unit does not have to be line by line as long as it separates the text of the report according to a predetermined rule. It may be for each time stamp, for each paragraph, or for each sentence separated by a frame. It may be from punctuation to punctuation. The tagged report may be stored in the data storage server 110 as the item-tagged report DB 113.

In the process S303, under the control of the model generation unit 121, the report tagged with the item is input to the item extraction model 132 of the analysis model storage server 130 as teacher data.

FIG. 6 is an example of the report tagged in the process S302, and is an example of the data structure of the item-tagged report DB 113. In this example, "notification", "confirmation", and "countermeasure" are tagged for each line of line number 601. A pair of text data 602 and item tag 603 for each line is input to the item extraction model 132 as teacher data 604.

In the process S304, the item extraction model 132 takes in the teacher data under the control of the model generation unit 121. At this time, if there is the accumulated classification data 3012, it can be similarly taken in as teacher data.

In the process S305, the item extraction model is supervised learning or relearned under the control of the model generation unit 121. The performance of the item extraction model 132 can be improved by repeating learning as necessary. The training of the analysis model is the same as that of the process S206.

In the process S306, the item extraction unit 123 inputs the text data 602 of the report other than the teacher data into the trained item extraction model 132, for example, line by line. The item extraction model 132 appropriately trained in the process S305 outputs the item tag 603 for each line of the report.

In the process S307, each line of the report with the output item tag is sent to the user terminal 140.

In the process S308, the control unit 141 of the user terminal 140 lists each line of the report with the item tag by item. In this example, the text on each line of the report is classified into three categories, "notification", "confirmation", and "countermeasure", and listed.

By the above processing, the item classification list 3013 in which each line of the text data of the past report data is associated with a predetermined item is created. The item classification list may be stored in the data storage server 110 as the item classification list DB 114.

<4. Grouping of other items (processing S4)>
FIG. 7 is a detailed flow chart of the grouping process S4 for each of the other items. In the process S401, the item grouping unit 133 of the analysis model storage server 130 classifies the texts associated with the same item in the item classification list DB 114 created by the process S3, and groups similar sentences together.

Known techniques can be applied to the text classification method. For example, the text is converted to a vector, and the data is automatically grouped without an external standard (unsupervised) by a cluster analysis method.

In the process S402, the item grouping unit 133 assigns a group name (ID) to the grouped data. The group name may be automatically assigned, or the user may operate the control unit 141, see the content of the text on the user terminal 140, and assign an appropriate name. Then, in the process S403, the group name of each item is tagged in the report, and the report linked with the items is stored in the data storage server 110 as the report DB 115 with the item tag classification.

By the above processing, the contents of each item entered in free expression in each report are categorized by similar items, and the fluctuation of the description is absorbed.

For example, even if the description of "notification" that expresses a door failure is made with different notations such as "door does not open", "door is broken", "opening / closing abnormality", these are notified in the same category. Can be treated as being.

FIG. 8 is a table diagram showing an example of the data structure of the report DB 115 with item tag classification. The line number 601, the text data 602 of the line, the item tag 603 of the line, and the item classification 801 of the item are associated with the document number 401 of the report.

<5. Cause estimation model creation (processing S5)>
FIG. 9 is a detailed flow chart of the cause estimation model creation process S5.

In the process S501, the report analysis information creation unit 125 combines the data of the report DB 115 with the item tag classification and the data of the report DB 112 with the cause category to generate the report analysis information data. That is, the cause category 403 of the report DB 112 with the cause category of FIG. 4 is added for each document number of the data of the report DB 115 with the item tag classification of FIG. 8, and this is stored in the data storage server 110 to analyze the report. Information DB 116. The data in the report analysis information DB can be used as the cause estimation model teacher data, but it is also useful as described later.

FIG. 10 is a diagram showing a data structure (data structure of the report analysis information DB 116) generated as a result of the process S501. As described above, the cause category 403 is added to the data of the item tag classification report DB 115. Here, the pair of the item classification 801 and the cause category 403 can be used as the cause estimation model teacher data 1001.

In the process S502, the control unit 141 of the user terminal instructs the model generation unit 121 to input the cause estimation model teacher data 1001 into the cause estimation model 134.

In the process S503, the cause estimation model 134 takes in the item classification 801 of the cause estimation model teacher data 1001 as an explanatory variable and the cause category 403 as an objective variable.

In the process S504, the cause estimation model 134 learns under the control of the model generation unit 121.

In the process S505, for the cause estimation model 134 that has been learned by the teacher data, the report without the cause category 403 of the item tag classification report DB 115 is used as the learning data (training data), and the cause is caused. Make an estimate. That is, the item classification 801 is input, and the cause estimation model 134 is trained to output the correct cause category 403.

In the process S506, the output cause category 403 is confirmed, the estimation result is evaluated, and the accuracy of the cause estimation model 134 is confirmed. If the accuracy is insufficient, the cause estimation model 134 is retrained. If the accuracy is sufficient, the cause estimation model 134 is completed and put into actual operation.

The cause estimation model 134 to be used may be selected from known supervised learning models such as decision trees, random forests, and support vector machines (SVMs). A known method may be used as the learning method.

<6. Operation of cause estimation model (processing S6)>
FIG. 11 is a detailed flow chart of the operation process S6 of the cause estimation model.

It is assumed that some kind of failure occurs in the process S601. The content of the failure is input as a trigger in the process S602 as the content of the "notification". The notification is input to the cause estimation model 134.

In the process S603, when the cause estimation model 134 is composed of a decision tree, the cause estimation model 134 in which the "notification" is input outputs an instruction corresponding to the branching "confirmation" item.

In response to the output confirmation item instruction, the on-site worker or the like confirms the situation and responds in the process S604. In process S605, the answer is input to the cause estimation model 134 as the "confirmation" content.
In process S606, the cause estimation model 134 outputs the estimated cause.

FIG. 12 is an example of a cause estimation model 134 composed of a decision tree. This decision tree inputs "notification", "confirmation", "cause", and "countermeasure", but the operation required to perform "confirmation" can be defined as "instruction". Now, when an event of notification category 1 that "the motor does not operate" is input, an instruction "check the warning light" corresponding to the branch of whether or not the warning light of the cause estimation model 134 is lit is given to the site ( Process S603). As can be seen in the example of FIG. 12, the items "notification" and "confirmation" are important because they often correspond to the (especially early) branching of the decision tree.

In response to the instruction, "warning light is on (confirmation category 1)" or "warning light is not on (confirmation category 2)" is input to the cause estimation model 134 (process S604), and the cause estimation model The estimation is advanced by the execution of (Process S605).

When the cause is finally estimated, the estimated cause is output (processes S606 to S607), the fact confirmation based on the output result is instructed to the site (process S608), and countermeasures are taken (process S609). For example, if the cause of a blown fuse is presumed (FIG. 12, cause category 1), a countermeasure of replacing the fuse is taken (FIG. 12, countermeasure category 1).

FIG. 13 shows an example of another overall flow using the system of FIG. 1 for the purpose of explaining Example 2.

In FIG. 1 of the first embodiment, the system creates and operates a cause estimation model (processes S5 and S6) using the data obtained in the processes S1 to S4. However, the data obtained in the processes S1 to S4 may be processed by the report analysis information creation process S100 and directly shown to the user.

FIG. 14 shows a report analysis information creation process S100 performed by the report analysis information creation unit 125, and the data of the report analysis information DB 116 shown in FIG. 10 is transformed into the form of the list 1400. The combination of categories for each item that appears in the report is shown, allowing the user to know which confirmation or cause category is relevant to each notification category.

The user can display the list 1400 created from the data of the report analysis information DB 116 on the user terminal 140 by the control unit 141.

FIG. 15 shows an example of another overall flow using the system of FIG. 1 for the purpose of explaining Example 3.

In FIG. 1 of the first embodiment, a process S2 separate from the extraction process S3 of the categories of other items is performed in order to extract the category of the cause item. However, the category extraction of the cause item may be performed in the same manner as other items.

In the item extraction process S3-2 of the third embodiment, the item of "cause" is extracted and categorized in the same manner as "notification", "confirmation", and "countermeasure". There is a merit that the processing is simplified as compared with the first embodiment, but there is a demerit that the extraction method peculiar to the "cause" cannot be used.

As described above, in this embodiment, even if the report has no fixed format or expression, the contents of the report are formulated and organized by extracting and categorizing the items, and the user can organize the contents. It can be made easier to understand. In addition, it can be easily used as teacher data for machine learning of artificial intelligence.

Text data analysis system 100, data storage server 110, analysis server 120, analysis model storage server 130, user terminal 140, report database DB 111, report DB 112 with cause category, report DB 113 with item tag, item classification list DB 114, item Report DB 115 with tag classification, report analysis information DB 116, model generation unit 121, cause category extraction unit 122, item extraction unit 123, item text classification unit 124, report analysis information creation unit 125, analysis unit 126

Claims (15)

A text data analysis system that processes text data from multiple documents.
An item extraction unit that extracts "notification" and "confirmation" items that are defined in advance as text data having a specific meaning in the document from the text data of the plurality of documents.
An item classification unit that classifies multiple text data extracted as the same item into multiple categories,
The report analysis information creation department that associates the categories in which the extracted text data is classified with the documents in which the text data is extracted, and
A text data analysis system equipped with. The item classification unit
Multiple text data extracted as the same item are classified into multiple categories by grouping similar ones.
The text data analysis system according to claim 1. The item extraction unit
As the above items, "notification" that describes passively obtained information and "confirmation" that describes actively obtained information are extracted.
The text data analysis system according to claim 1. The item extraction unit
As the item, the "cause" of the event that is the content of the "notification" is extracted, the "notification" and the "confirmation" are extracted by the first algorithm, and the "cause" is extracted by the second algorithm.
The text data analysis system according to claim 3. The item extraction unit
As the above item, "countermeasures" for eliminating the event that is the content of "notification" are extracted.
The text data analysis system according to claim 3. The report analysis information creation department
Generates information indicating that the category in which the text data extracted as "Notification", "Confirmation" and "Cause" items is classified is associated with the extracted document.
The text data analysis system according to claim 4. The report analysis information creation department
Learn the analysis model using the category in which the text data extracted as the "notification" and "confirmation" items is classified as the explanatory variable and the category in which the text data extracted as the "cause" item is classified as the objective variable. Generate teacher data for
The text data analysis system according to claim 4. A text data analysis method that processes text data from multiple documents.
From the text data of the plurality of documents, items defined in advance as text data having a specific meaning in the document are extracted.
Multiple text data extracted as the same item is classified into multiple categories,
The category in which the extracted text data is classified is associated with the document in which the text data is extracted.
Text data analysis method. Multiple text data extracted as the same item are classified into multiple categories by grouping similar ones.
The text data analysis method according to claim 8. As the above items, "notification" that describes passively obtained information and "confirmation" that describes actively obtained information are extracted.
The text data analysis method according to claim 9. Correspond the category in which the text data extracted as the "notification" and "confirmation" items is classified to the document in which the text data is extracted.
The text data analysis method according to claim 10. From the text data of the plurality of documents, text data having the meaning of "cause" of the event that is the content of "notification" in the document is extracted according to a predefined classification.
The category in which the text data extracted as the "cause" item is classified is associated with the document in which the text data is extracted.
The text data analysis method according to claim 11. Information indicating that the category in which the text data extracted as an item is classified is associated with the extracted document is displayed to the user.
The text data analysis method according to claim 12. The analysis model is learned by using the information in which the text data extracted as items is classified and the information associated with the document from which the text data is extracted as teacher data.
Analyze new text data using the learned analysis model,
The text data analysis method according to claim 12. It is a failure response recommendation system that processes text data of multiple documents including the contents related to the failure and proposes a response to the failure.
An item extraction unit that extracts items of passive information and active information defined in advance as text data having a specific meaning in the document from the text data of the plurality of documents.
An item classification unit that classifies multiple text data extracted as the same item into multiple categories,
It is equipped with a report analysis information creation unit that associates the categories in which the extracted text data is classified with the documents in which the text data is extracted.
Outputs failure response recommendations based on the analysis results of the report analysis information creation unit.
Failure response recommendation system.

Images