JP6621385B2 - Text analysis system and text analysis method - Google Patents

Description

  The present invention relates to a text analysis system.

  Various text information is accumulated in society, and it is required to analyze and utilize them. For example, it is necessary to analyze the causal relationship between events accumulated as text information.

  There are the following prior arts as background art of this technical field. Patent Document 1 (Japanese Patent Application Laid-Open No. 2008-203964) discloses a word that constitutes an event with respect to a causal relationship between a plurality of different events extracted from a natural language sentence that is a document described in a natural language. A group of events that are partly the same and have a common cause or effect event are selected as clustering targets by the cluster target selection unit, and have a data structure in which only common causal relationships are integrated in all clustering target events The causal relationship graph is stored in the causal relationship storage unit, the complexity of the structure of the causal relationship graph is digitized as a clustering score, and the event group selected by the cluster target selection unit so that the clustering score is minimized is the event cluster evaluation A causal relationship analysis apparatus that performs clustering in each section is disclosed.

JP 2008-203964 A

  The causal relationship analysis apparatus described in Patent Literature 1 extracts a causal relationship between events based on predefined patterns and causal relationships. However, since sentences written by humans have infinite patterns and causal relationships, it is difficult to pre-define all patterns and causal relationships. If they are not pre-defined, it is difficult to analyze text, and causal relationships. Extraction becomes difficult.

  An object of the present invention is to analyze text information without defining a pattern.

  A typical example of the invention disclosed in the present application is as follows. That is, the text analysis system includes a processor that executes a program and a storage device that is accessed by the processor, and the text information analyzed by the text analysis system includes item names defined in each of a plurality of columns. A plurality of line data including character or numeric data corresponding to the item name, and the text analysis system abstracts the data of the designated column using the data of other columns for each line. A language processing unit for extracting keywords from the abstracted data, an aggregating unit for aggregating rows with similar data according to the keyword similarity in the specified column, and representative data of a target row for statistical processing A summation unit that calculates values, and the abstraction unit generates a materialization process result obtained by replacing the data in the designated column with data in another column of the row. .

  According to one aspect of the present invention, text information can be analyzed without defining a pattern. Problems, configurations, and effects other than those described above will become apparent from the description of the following embodiments.

It is a figure which shows the structure of the text analysis system of the Example of this invention. It is a block diagram which shows the physical structure of the server of a present Example. It is a flowchart of the text analysis process of a present Example. It is a flowchart which shows the detail of an abstraction process. It is a flowchart which shows the detail of a data aggregation process. FIG. 6A is a diagram illustrating an example of the failure information A, and FIG. 6B is a diagram illustrating an example of the failure information A1. FIG. 7A is a diagram illustrating an example of the failure information A2. FIG. 8A is a diagram illustrating an example of the aggregation processing intermediate result A3 ′, and FIG. 8B is a diagram illustrating an example of the aggregation processing result A3. It is a figure which shows a data aggregation process. FIG. 10A is a diagram illustrating an example of the total result A4, and FIG. 10B is a diagram illustrating an example of the materialization processing result A5. It is a flowchart which shows the detail of a pattern extraction process. It is a figure which shows an example of a maintenance operation business pattern display screen. It is a figure which shows an example of a maintenance operation business pattern display screen.

<Example 1>
FIG. 1 is a diagram showing a configuration of a text analysis system according to an embodiment of the present invention.

  The text analysis system of this embodiment is configured by a server 101. The server 101 is connected to one or more computers 102 via the network 105. The physical configuration of the server 101 will be described later with reference to FIG.

  The computer 102 is a computer used by the user 100, and includes a processor (CPU) that executes a program, a storage device that stores the program and data, and a communication interface. The computer 102 is connected to an output device such as a display 103 and an input device 104 (keyboard, mouse, touch panel, etc.), and provides a user interface for the user 100.

  The server 101 includes a data abstraction unit 110, a natural language processing unit 111, a data aggregation unit 112, a data aggregation unit 113, and a pattern extraction unit 114. The processing executed by each unit will be described with reference to FIG.

  The server 101 has a failure information database 120 and a maintenance operation work pattern database 121.

  The failure information database 120 is a database that stores original data (failure information) from which the text analysis system of the present embodiment analyzes text information and extracts relationships between items included in the text information. This will be described with reference to FIG. The maintenance operation work pattern database 121 is a database for storing the result (materialization process result A5) analyzed by the text analysis system of this embodiment, and the details will be described with reference to FIG.

  FIG. 2 is a block diagram showing the physical configuration of the server 101 that constitutes the text analysis system of this embodiment.

  The server 101 of this embodiment is configured by a computer having a processor (CPU) 1, a memory 2, an auxiliary storage device 3, and a communication interface 4.

  The processor 1 executes a program stored in the memory 2. The memory 2 includes a ROM that is a nonvolatile storage device and a RAM that is a volatile storage device. The ROM stores an immutable program (for example, BIOS). The RAM is a high-speed and volatile storage device such as a dynamic random access memory (DRAM), and temporarily stores a program executed by the processor 1 and data used when the program is executed.

  The auxiliary storage device 3 is configured by a large-capacity and non-volatile storage device such as a magnetic storage device (HDD) or a flash memory (SSD), for example, and stores a program executed by the processor 1 and data used when the program is executed. Store. That is, the program is read from the auxiliary storage device 3, loaded into the memory 2, and executed by the processor 1.

  The communication interface 4 is a network interface device that controls communication with other devices (such as the computer 102) according to a predetermined protocol.

  The server 101 may have an input interface 5 and an output interface 8. The input interface 5 is an interface to which a keyboard 6 and a mouse 7 are connected and receives an input from an administrator. The output interface 8 is an interface to which a display device 9 or a printer is connected, and outputs the state of the server 101 and the execution result of the program in a format that can be viewed by an administrator.

  The program executed by the processor 1 is provided to the server 101 via a removable medium (CD-ROM, flash memory, etc.) or a network, and is stored in the nonvolatile auxiliary storage device 3 that is a non-temporary storage medium. For this reason, the server 101 may have an interface for reading data from a removable medium.

  The server 101 is a computer system configured on a plurality of computers that are physically configured on one computer or logically or physically, and is a virtual computer constructed on a plurality of physical computer resources. It may work on. Further, the program executed on the server 101 may operate in a separate thread on the same computer.

  In the server 101, all or a part of the functional blocks implemented by the program may be configured by a physical integrated circuit (for example, Field-Programmable Gate Array).

  FIG. 3 is a flowchart of the text analysis process of this embodiment.

First, the data abstraction unit 110 acquires the failure information A (FIG. 6A) from the failure information database 120, executes an abstraction process on the designated columns C _{0 to} C _N , and acquires the failure information A1 (FIG. B)) is generated and temporarily written in the storage device (S301). Abstraction process, the data for the specified column C ₀ -C _N, as compared with the data of the other columns in the row, abstracts using item names of the columns data matches. Details of the abstraction processing will be described later with reference to FIG. Note that the designated column is data on which a natural sentence is analyzed by the text analysis processing of the present embodiment, and is data including a causal relationship to be analyzed.

Next, the natural language processing unit 111 executes the natural language processing to specify columns _C 0 -C _N of the failure information A1 (FIG. 6 (B)), to extract the keyword, failure information A2 (FIG. 7 (A )) Is generated and temporarily written in the storage device (S302). For natural language processing, a known technique can be used, and a word is extracted from a natural sentence included in the failure information A1. For example, using a morphological analysis technique, a word is extracted from a natural sentence, and a predetermined part of speech (for example, a noun or a verb) is extracted as a keyword.

Thereafter, data aggregation unit 112 executes the fuzzy aggregation processing failure information A2 to data of the specified column _C 0 -C _N in (FIG. 7 (A)) in a predetermined order, aggregation processing result A3 (FIG. 8 (B)) is generated and temporarily written in the storage device (S303). Ambiguity aggregation process compares the similarity of the data in the specified columns C ₀ -C _N, data aggregates row similar. Details of the fuzzy aggregation processing will be described later with reference to FIG.

Next, the data aggregation unit 113, aggregates processing result A3 running (Fig. 8 (B)) statistical processing to specified target sequence _V 0 ~V _M of aggregation result A4 to (Fig. 10 (A)) Generate and temporarily write to the storage device (S304). The target column V ₀ ~V _M statistical process, mainly, the number is recorded. In the statistical processing, the numerical value of the target column is statistically processed to calculate data representing the row (representative value). In the statistical processing, for example, an average value, a total value, a minimum value, a maximum value, a variance value, a standard deviation value, a mode value, a median value, or the like may be calculated, or a probability distribution (histogram) may be created. In the present embodiment, the average of the values included in the work man-hours and the stop time of the aggregation processing result A3 (FIG. 8B) is calculated.

Thereafter, the data abstraction unit 110, counting result A4 is running a specific process in the designated column _C 0 -C _N in (FIG. 10 (A)), embodied processing result A5 (FIG. 10 (B)) Is written into the maintenance operation pattern database 121 (S305). For example, item names that are included in the specified column C ₀ -C _N because strings that are abstracted by the abstraction process, and returns to the original value. Specifically, when “device name” is included in the phenomenon column, the device name data in the row includes a plurality (or one or more) device names aggregated by the aggregation process. Therefore, the “device name” in the phenomenon column is replaced with “aaa0ZZZ00 ## ccc1XXX02” in the device name column. In addition, “failed part” in the cause column is replaced with “part 001 ## part 008” in the failed part column. Further, when “failed part” is included in the phenomenon column, the data of the failed part in the row includes a plurality of (or one or more) device names that are aggregated by the aggregation process. For this reason, referring to the metadata that records which line of data matches and is abstracted into the item name, “failed part” in the phenomenon column is replaced with “part 008” in the failed part column.

  Finally, the pattern extraction unit 114 extracts maintenance operation work patterns from the materialization processing result A5, ranks them, and recommends them to the user 100 (S306). Details of the pattern extraction processing will be described later with reference to FIG.

  FIG. 4 is a flowchart showing details of the abstraction processing S301.

First, the data abstraction unit 110 defines the number of rows of the failure information A as R and the number of columns M, and defines them as designated columns C ₀ ,..., C _n , ..., C _N (0 ≦ n ≦ N). The number of designated columns is N + 1 columns (S401). Further, the parameter r for controlling the row and the parameter n for controlling the designated column are initially set to 0 (S402).

  Then, the data abstraction unit 110 determines whether the parameter n is N or less (S403). If the parameter n is greater than N (NO in S403), the processing is terminated because all the specified column data has been abstracted. On the other hand, if the parameter n is N or less (YES in S403), it is determined whether the parameter r is smaller than the maximum number R of rows (S404).

As a result, if the parameter r is equal to or greater than the maximum number of rows R (NO in S404), the abstraction processing for the designated column currently being processed is completed, so 1 is added to the parameter n (S412), and the process goes to step S403. Return and abstract the data of the next specified column. On the other hand, if the parameter r is smaller than the maximum number of rows R (YES in S404), the data D _rCn of the r row and C _n column of the failure information A is acquired (S405). Further, the parameter s for controlling the row is initialized to 0, and the parameter m for controlling the column is initialized to 0 (S406).

  The data abstraction unit 110 determines whether the column parameter m is smaller than the maximum number M of columns (S407). As a result, if the column parameter m is equal to or greater than the maximum number M of columns (NO in S407), the comparison with the data of all columns in the currently processed row is completed, so 1 is added to the parameter r ( S411), the process returns to step S404, and the next row of data is abstracted.

  On the other hand, if the column parameter m is smaller than the maximum number M of columns (YES in S407), it is determined whether the row parameter s is smaller than the maximum number R of rows (S408). As a result, if the row parameter s is greater than or equal to the maximum number R of rows (NO in S408), the processing for the column is completed, so 1 is added to the parameter m (S414), and the process returns to step S407. Process column data.

On the other hand, if the parameter s of the row is smaller than the maximum number R of rows (YES in S408), the data D _sm of the s row and m-th column of the failure information A is acquired (S409). Then, in the acquired _DrCn , the character string that matches D _sm is replaced under the following conditions. First, if there is a character string in which the row parameter s is equal to the row parameter r, that is, the character string D _{sm in} the other column of the row is equal to the character string that matches D _sm in the D _rCn , (S410). For example, “aaa0ZZZ00” in the phenomenon column on the first line of the failure information A (FIG. 6A) is the same as the device name data on the corresponding line, and is thus replaced with “device name” which is the item name.

In addition, if there is a character string in which the row parameter s is different from the row parameter r, that is, there is a character string equal to the character string D _sm of another row, the character string that matches D _sm in D _rCn is the item name in the m-th column. Replace with '. For example, since “component 008” in the phenomenon column on the second line of the failure information A (FIG. 6A) is the same as the data on the failed component on the third line, the item name “failed component '” is displayed. replace.

  In this way, “′” records whether the data is replaced in accordance with the data in the corresponding row or is replaced in accordance with the data in another row. At this time, it is preferable to have data indicating which row data matches and is abstracted into the item name.

  Thereafter, 1 is added to the parameter s of the line (413), and the process returns to step S408 to process the next line.

  As described above, the abstraction process abstracts the data in the designated column by replacing the word in the data in the designated column with the item name of the column having the same word in the same row. Data can be abstracted without setting As mentioned above. In the abstraction processing of the present embodiment, data can be abstracted without using dictionary data, but dictionary data may be used together, and the accuracy of abstraction can be improved.

  Further, in the abstraction process, metadata for enabling the abstracted character string to be distinguished may be recorded. For example, the start position (number of characters) of the abstracted character string may be recorded and referred to when the abstracted character string is returned in the materialization process S305. Further, information for identifying the replaced item name may be associated with the abstracted character string.

  FIG. 5 is a flowchart showing details of the data aggregation processing S303.

First, the data aggregation processing S303 defines the failure information A2 as the number of rows R, the number of columns M, the designated columns C ₀ ,..., C _n , ..., C _N (0 ≦ n ≦ N), and the similarity threshold is defined as T. To do. Also, L is set to an empty list (S501). In addition, a parameter r for controlling the row and a parameter n for controlling the designated column are initially set to 0 (S502).

  Then, the data aggregation processing S303 determines whether the parameter n is N or less (S503). If the parameter n is larger than N (NO in S503), the processing is terminated because the data of all the designated columns are collected. On the other hand, if the parameter n is N or less (YES in S503), it is determined whether the parameter r is smaller than the maximum number R of rows (S504).

As a result, if the parameter r is equal to or greater than the maximum number R of rows R (NO in S504), the data aggregation processing S303 has completed the data aggregation processing for the specified column currently being processed, so that the list L temporarily created Combine included lines into one line. At this time, the value of the column other than the specified column _{C n} is connect by delimiters according to the specified character string (e.g., ##) (S512). Note that even if the same values are directly connected by a delimiter, only one identical value may be left and the others may be deleted.

Then, 1 is added to the parameter n (S513), and the process returns to step S503 to aggregate the data of the next designated column. On the other hand, if the parameter r is smaller than the maximum number R of rows (YES in S504), r rows of failure information A2, acquires data _{D RCN} of _{C n} (S505). Further, the line parameter k is initialized to r + 1 (S506). This is because in the data aggregation processing, it is determined whether the rows can be aggregated, so there is no need to determine again the combination for which the determination has already been completed, so it is only necessary to compare the rows after the currently processed r row. .

  Then, the data aggregation processing S303 determines whether the row parameter k is smaller than the maximum number R of columns (S507). As a result, if the column parameter k is equal to or greater than the maximum number R of columns (NO in S507), the comparison between the data in the currently processed row and the data in the other row is completed, and 1 is added to the parameter r. Then (S511), the process returns to step S504, and the data of the next row is collected.

On the other hand, if the parameter k of rows is less than the maximum number R of rows (YES at S507), k rows of failure information A2, acquires data _{D KCN} of _{C n-th} column (S508). Then, the degree of similarity between the acquired _DrCn and _DkCn is calculated, and if the calculated degree of similarity is larger than the threshold T, {r, k} is added to the list L (S509).

  For calculating the similarity, a known cosine similarity can be used. FIG. 9A shows an example in which the similarity is calculated for the phenomenon column of the failure information A2 (FIG. 7A). The similarity between “device name, abnormality, occurrence” on the first line of failure information A2 and “device, failure, present” on the second line is 0.00, and “device, failure, present” on the second line. And “Device Name, Device, Abnormal, Occurrence” on the third line is 0.33, “Device Name, Device, Abnormal, Occurrence” on the third line and “Device Name, Abnormal” on the first line The similarity to “occurrence” is 0.75. For example, when the threshold value T is set to 0.7, it is determined that the first and third lines are similar in the phenomenon column.

  Further, as shown in FIG. 9B, a specific example of data aggregation using similarity is “device” in the phenomenon column of the first line (failure ID is F0) of the failure information A2 (FIG. 7A). “Name, Abnormality, Occurrence” is similar to “Device Name, Device, Abnormality, Occurrence” in the phenomenon column of the third row (Failure ID is F2). To integrate. On the other hand, as shown in FIG. 9C, “device name, abnormality, occurrence” in the phenomenon column of the first line (fault ID is F0) of the failure information A2 (FIG. 7A) is the second line ( Since the failure ID is not similar to “device, failure, present” in the phenomenon column of F1), the data of failure ID F0 and the data of F1 are not integrated.

  Next, with reference to FIG. 8, a process of aggregating data of a plurality of designated columns of the failure information A2 and generating an aggregation processing result A3 will be described.

First, aggregate data phenomena column is the first of the indicated column C _0. As described above, since the data in the phenomenon column is similar to the first line (fault ID is F0) and the third line (fault ID is F2) of the fault information A2, the data with the fault ID F0 and F2 Integrate with other data. At this time, the values of the columns other than the phenomenon column are connected by ## that is a delimiter by a designated character string. FIG. 8A shows an intermediate result A3 ′ obtained by collecting the data in the phenomenon column.

Next, aggregate data causes column is specified column C _1. Since there is no data whose similarity exceeds the threshold T in the cause column, rows are not aggregated depending on the cause column.

Next, aggregate data for work content field is specified column C _2. In the work content column, there are two “failed parts, replacement” lines, and two “progress and observation” lines. Therefore, as shown in FIG. 8 (B), to classify the row aggregated by the specified columns C ₀ into two groups. The cause column from the failure ID column remains unchanged. Note that the similarity threshold T may be changed for each designated column.

  Returning to FIG. 1 is added to the parameter k of the row (510), and the process returns to step S507 to process the next column.

  As described above, in the data aggregation process, data having the same tendency can be collected and shaped in an easy-to-understand manner. Further, since data is aggregated even when the data does not completely match, data can be aggregated even for items for which input candidates are not limited. For example, data can be aggregated even for items that the operator inputs in natural sentences.

  Next, using FIG. 6 to FIG. 10, the change in data from the failure information A stored in the failure information database 120 to the instantiation processing result A5 stored in the maintenance operation pattern database 121 is shown. explain. In the following drawings, italics indicate data replaced by the abstraction processing S301, and underline indicates data changed by the processing step.

  FIG. 6A is a diagram illustrating an example of failure information A stored in the failure information database 120. The failure information A includes data of failure ID, occurrence date / time, device name, device classification, failure part, phenomenon, cause, work content, work man-hour, and stop time. The failure information A is input by, for example, an operator who has performed work for the failure.

  The failure ID is identification information for uniquely identifying data stored in the failure information database 120. The occurrence date and time is the date and time when the failure occurred. The device name is the name of the device in which the failure has occurred. The device classification is a classification of a device in which the failure has occurred. The failed part is the model number or name of the failed part. The phenomenon is the phenomenon of the failure, and the cause is the cause of the failure. The work content is the work content for the failure. The work man-hour is the man-hour for performing work for the failure. The stop time is a time when the apparatus is stopped due to the failure.

  Note that the phenomenon, cause, and work content are defined as a designated column, work man-hours, and stop time as target columns. In addition to the items shown in the figure, the date and time of occurrence, the location (region) responder, etc. may be defined in the designated column to be subjected to text analysis.

  The fault information A1 shown as an example in FIG. 6B is generated from the fault information A by the abstraction process S301. Of the data in the designated column (phenomenon, cause, work content), the same data as the data of other items is replaced with the item name of the data.

  The failure information A2 shown as an example in FIG. 7A is generated from the failure information A1 by the natural language processing S302. Data (natural sentences) in a designated string (phenomenon, cause, work content) is broken down into words by natural language processing.

  The aggregation processing result A3 shown as an example in FIG. 8B is generated from the failure information A2 by the data aggregation processing S303. If the data in the specified column is similar, the row is aggregated. Specifically, at least one of the phenomenon, the cause, and the work content is collected on the same line, and the relationship between items is analyzed. FIG. 8A shows an intermediate result A3 ′ for deriving the aggregation processing result A3.

  An aggregation result A4 shown as an example in FIG. 10A is generated from the aggregation processing result A3 by the aggregation processing S304. Data of the target column (number of work items, stop time) is statistically processed and updated with an average value.

  The materialization process result A5 shown in FIG. 10B is generated from the totalization result A4 by the materialization process S305 and stored in the maintenance operation work pattern database 121. The data replaced in the abstraction processing S301 is replaced with the original data (device name, part name, etc.).

  FIG. 11 is a flowchart showing details of the pattern extraction processing S306.

  First, designation of the type of index that the user 100 focuses on is input to the computer 102. The pattern extraction unit 114 acquires an index designated by the user 100 (S1101). At this time, the user 100 may specify the number of data to be output and the arrangement order (ascending order or descending order) of data in addition to the type of index.

  Then, the pattern extraction unit 114 extracts a predetermined number of maintenance operation business patterns from the maintenance operation business pattern database 121 in ascending or descending order of the designated index values (S1102).

  Finally, the pattern extraction unit 114 generates data for visualizing the maintenance operation business pattern extracted from the maintenance operation business pattern database 121 in a table format or a graph format, and transmits the data to the computer 102 (S1103). The computer 102 displays the data transmitted from the server 101 (pattern extraction unit 114) on the display 103.

  12 and 13 are diagrams illustrating an example of the maintenance operation work pattern display screen.

  FIG. 12 shows an example of a maintenance operation work pattern display screen when the user 100 designates the work man-hour as the index type. As shown in the figure, at the top of the screen, three maintenance operation work patterns are displayed in descending order of work man-hours. At the bottom of the screen, a pie chart and a bar chart showing the work man-hour values are displayed. For example, the user 100 may consider the replacement time of the HDD from the data in the first row. Further, the server 101 and work contents (replacement of HDD) may be recommended.

  FIG. 13 shows an example of a maintenance operation work pattern display screen when the user 100 designates a stop time as an index type. As shown in the figure, at the top of the screen, three maintenance operation work patterns are displayed in the order of longer stop time. A pie chart and a bar chart showing the value of the stop time are displayed at the bottom of the screen.

  The text analysis method according to the present invention has been described above by taking failure information as an example, but it can also be applied to the analysis of other text information.

  As described above, the text analysis system according to the embodiment of the present invention is abstracted with the data abstraction unit 110 that abstracts the data of the designated column using the data of other columns for each row in the table. A natural language processing unit 111 that extracts keywords from the collected data, a data aggregation unit 112 that aggregates rows with similar data according to the similarity of the keywords in the specified column, and a representative value of the data of the target row for statistical processing The data abstraction unit 110 generates the materialization processing result A5 in which the data of the designated column is replaced with the data of the other column of the row, so that the event can be performed without defining in advance. The relationship between them can be extracted. In addition, if data having the same tendency is scattered, it is difficult to understand the tendency, but data having the same tendency can be collected.

  In addition, for the index value input by the user, a predetermined number of data is extracted from the materialization processing result A5 in ascending or descending order of the representative value, and the extraction unit presenting the extracted value to the user is extracted. it can.

  Further, the data abstraction unit 110 compares the data in each designated column with the data in the other columns of the row, and replaces the word in the data in each designated column with the item name of the column having the same word in the same row. So it can absorb the swing of proper nouns and expressions.

  The data abstraction unit 110 replaces the item name included in the data of the specified column with the data of the item name in the row, so that the data is abstracted without defining in advance. It can be embodied.

  Further, the data aggregating unit 112 compares the data of each designated column with the data of other rows of the designated column, and collects similar rows of data when the degree of similarity of the data is greater than a predetermined threshold. Data can be aggregated even for items whose natural sentence expression is shaking or whose input candidates are not limited.

  The present invention is not limited to the above-described embodiments, and includes various modifications and equivalent configurations within the scope of the appended claims. For example, the above-described embodiments have been described in detail for easy understanding of the present invention, and the present invention is not necessarily limited to those having all the configurations described. A part of the configuration of one embodiment may be replaced with the configuration of another embodiment. Moreover, you may add the structure of another Example to the structure of a certain Example. In addition, for a part of the configuration of each embodiment, another configuration may be added, deleted, or replaced.

  In addition, each of the above-described configurations, functions, processing units, processing means, etc. may be realized in hardware by designing a part or all of them, for example, with an integrated circuit, and the processor realizes each function. It may be realized by software by interpreting and executing the program to be executed.

  Information such as programs, tables, and files that realize each function can be stored in a storage device such as a memory, a hard disk, and an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, and a DVD.

  Further, the control lines and the information lines are those that are considered necessary for the explanation, and not all the control lines and the information lines that are necessary for the mounting are shown. In practice, it can be considered that almost all the components are connected to each other.

101 Server 102 Computer 110 Data Abstraction Unit 111 Natural Language Processing Unit 112 Data Aggregation Unit 113 Data Aggregation Unit 114 Pattern Extraction Unit 120 Failure Information Database 121 Maintenance Operation Business Pattern Database

Claims (10)

A text analysis system,
A processor that executes a program; and a storage device that is accessed by the processor;
The text information analyzed by the text analysis system includes an item name defined in each of a plurality of columns, and a plurality of line data including character or numerical data corresponding to the item name,
The text analysis system includes:
For each row, an abstraction unit that abstracts the data of the designated column using data of other columns,
A language processing unit for extracting keywords from the abstracted data;
An aggregation unit that aggregates rows with similar data according to the similarity of keywords in the specified column;
A totaling unit that calculates the representative value of the data of the target row for statistical processing,
The abstraction unit generates a materialization process result by replacing the data in the designated column with data in another column of the row. The text analysis system according to claim 1,
A text analysis system comprising: an extracting unit that extracts a predetermined number of data from the materialization processing result in ascending or descending order of the representative value for an index input by a user and presents it to the user. The text analysis system according to claim 1,
The abstraction unit is:
Compare the data in each specified column with the data in other columns in the row,
A text analysis system that abstracts data in a specified column by replacing words in the data in each specified column with item names of columns having the same word in the same row. The text analysis system according to claim 3,
The abstraction unit replaces the item name included in the data of the designated column with the data of the item name in the row and concretizes the text analysis system. The text analysis system according to claim 1,
The aggregation unit is
Compare the data in each specified column with the data in other rows of the specified column,
A text analysis system that aggregates rows with similar data when the similarity of the data is greater than a predetermined threshold. A text analysis method executed by a computer,
The computer includes a processor that executes a program, and a storage device that is accessed by the processor.
The text information analyzed by the computer includes a plurality of item names and a plurality of line data including character or numerical data corresponding to the item names,
The text analysis method includes:
The processor, for each row, using the data of the other column to abstract the data of the designated column;
The processor includes a language processing step of extracting a keyword from the abstracted data;
The processor includes an aggregation step of aggregating rows with similar data according to the similarity of keywords in the designated column;
The processor includes a totaling step for calculating a representative value of data of a target row for statistical processing,
A text analysis method comprising: an instantiation step of generating an instantiation processing result obtained by replacing the data in the designated column with data in another column of the row. The text analysis method according to claim 6, comprising:
The processor includes an extraction step of extracting a predetermined number of data from the materialization process result in ascending or descending order of the representative value for the index input by the user and presenting the result to the user. Method. The text analysis method according to claim 6, comprising:
In the abstraction step,
The processor compares the data in each designated column with the data in the other columns of the row,
The text analysis method, wherein the processor abstracts the data of the designated column by replacing words in the data of the designated column with item names of columns having the same word in the same row. The text analysis method according to claim 8, comprising:
In the abstraction step, the processor is configured to replace the item name included in the data of the designated column with the data of the item name in the row and to make it concrete. The text analysis method according to claim 6, comprising:
In the aggregation step,
The processor compares the data of each designated column with the data of other rows of the designated column,
The text analysis method according to claim 1, wherein when the similarity of data is greater than a predetermined threshold, the processor aggregates lines having similar data.

Images