JP2022117019A - Information analysis program, information analysis method, and information analysis device - Google Patents

Abstract

To analyze the correspondence of keywords corresponding to technical words or the like by analyzing a causal relation of sentences after extracting unique expressions.SOLUTION: An information analysis program is a program that causes a processor to execute: a text information acquisition step of acquiring text information; a unique expression extraction step of extracting unique expressions from the text information and giving tags; a dependency extracting step of analyzing dependencies from the text information and extracting dependencies in which one or both of a dependency source or a dependency destination are unique expressions; a co-occurrence degree calculation step of calculating degree of co-occurrence for combinations of dependencies in the dependencies extracted from the text information in the dependency extraction step; and a co-occurrence extraction step of extracting a combination of dependencies based on the degree of co-occurrence.SELECTED DRAWING: Figure 2

Description

The present disclosure relates to an information analysis program, an information analysis method, and an information analysis apparatus for analyzing relationships between named entities from text data.

With recent advances in natural post-processing technology and machine learning technology, methods for automatically analyzing text data by computers have been proposed. For example, if it is possible to analyze the causal relationship of text data, it will be possible to efficiently systematize knowledge by analyzing the main points of contention of the literature from among a large amount of literature data.

For example, Patent Literature 1 proposes a method for broadly analyzing causal relationships using natural language processing.

JP-A-8-221415

However, the method disclosed in Patent Document 1 merely analyzes the causal relationship, and does not mention the application to other corresponding relationships such as purpose and means or method and effect.

For example, in the case of analyzing documents related to specialized technology, not only the analysis of causal relationships such as cause and effect, but also the relationship between a certain event and the analysis method of experiments, tests, inspections, grounds, etc. to support the event. If it can be analyzed, it becomes possible to extract more essential elements when systematizing the literature.

Therefore, in the present disclosure, when analyzing the causal relationship, the named entity is extracted and then the dependency is extracted, thereby extracting the dependency including the keyword. Furthermore, in the present disclosure, by extracting co-occurrence relationships of dependencies, more accurate causal relationships can be analyzed. The purpose is to analyze the relationship between words.

A program to be executed by a computer having a processor and memory. The program provides the processor with a text information acquisition step of acquiring text information, a named entity extraction step of extracting a named entity from the text information and adding a tag, analyzing a dependency from the text information, A dependency extracting step of extracting a dependency in which one or both of the dependency destinations are unique expressions, and calculating the co-occurrence degree for a combination of dependencies in the dependency extracted in the dependency extracting step from the text information. A program for executing a co-occurrence degree calculation step and a co-occurrence extraction step of extracting a combination of dependencies based on the co-occurrence degree.

According to the present disclosure, using natural language processing, it is possible to accurately analyze causality even for text data including proper nouns, technical terms, and the like. In addition, it is possible to analyze not only the causal relationship but also the relationship between various words desired by the user, such as the relationship between a certain event and its analysis method.

1 is a diagram showing the overall configuration of a server according to Embodiment 1 of the present disclosure; FIG. 1 is a diagram illustrating an example of a server configuration according to Embodiment 1 of the present disclosure; FIG. It is a figure which shows the specific example of named-entity extraction. It is a figure which shows the specific example which analyzes a dependency. FIG. 10 is a diagram showing a specific example of dependency analysis using information on named entities; It is a figure which shows the specific example of the sentence containing a technical term. It is a figure which shows the specific example of named-entity extraction. It is a figure which shows the specific example which analyzes a dependency. It is a figure which shows the specific example finally extracted as a dependency. 4 is a flow chart showing an example of the flow of processing according to the first embodiment in the information analysis device; FIG. 11 is a chart showing an example of a configuration of an information analysis device according to Embodiment 2 of the present disclosure; FIG. It is a figure which shows the specific example which displayed the modification in the display part. It is a figure which shows the specific example which paid its attention to the hierarchical structure and displayed the dependency on the display part. 10 is a flow chart showing an example of the flow of processing in the information analysis device according to the second embodiment; It is a figure which shows the specific example of the information-analysis apparatus via a network.

Hereinafter, an information analysis device, an information analysis method, and an information analysis program according to embodiments of the present disclosure will be described with reference to the drawings. It should be noted that the embodiments described below do not unduly limit the content of the present disclosure described in the claims. Moreover, not all the configurations described in the embodiments are essential constituent elements of the present disclosure. Further, in all the drawings for explaining the embodiments, common constituent elements are given the same reference numerals, and repeated explanations are omitted.

<Embodiment 1>
In the first embodiment, after acquiring and analyzing text information and extracting named entities, the dependencies and co-occurrence relationships of named entities are analyzed, and the degree of co-occurrence is calculated for each named entity.

<Overall composition>
FIG. 1 is a diagram showing the overall configuration of a server 10 of this embodiment. Server 10 is a general-purpose computer. The server 10 is implemented by, for example, a stationary PC (Personal Computer), a laptop PC, or the like. Also, the server 10 may be a portable computer such as a smart phone or a tablet terminal.

As shown in FIG. 1, the server 10 includes a processor 11, a memory 12, a storage 13, a communication IF 14, and an input/output IF 15.

The processor 11 is hardware for executing an instruction set described in a program, and is composed of an arithmetic unit, registers, peripheral circuits, and the like. The memory 12 is for temporarily storing programs, data processed by the programs, etc., and is realized by a volatile memory such as a DRAM (Dynamic Random Access Memory). The storage 13 is a storage device for storing data, and is implemented by, for example, flash memory or HDD (Hard Disc Drive). The communication IF 14 is an interface for transmitting and receiving signals so that the server 10 communicates with an external device. The input/output IF 15 functions as an interface with an input device for receiving input from the user and an output device for presenting information to the user.

<Server configuration>
The configuration of the server 10 will be described in detail below. As shown in FIG. 2 , the server 10 according to this embodiment includes a communication section 101 , a storage section 102 and a control section 103 . The communication unit 101 performs processing for the server 10 to communicate with other devices. The communication unit 101 performs transmission processing on the signal generated by the control unit 103 and transmits the signal to the outside. The communication unit 101 performs reception processing on a signal received from the outside and outputs the signal to the control unit 103 .

The storage unit 102 is configured by, for example, a flash memory or the like, and stores data and programs used by the server 10 .

The control unit 103 is implemented by the processor 11 reading a program stored in the storage unit 102 and executing instructions included in the program. The control unit 103 controls operations of the server 10 . Specifically, for example, the control unit 103 functions as a text information acquisition unit 1031, a named entity extraction unit 1032, a dependency extraction unit 1033, a co-occurrence calculation unit 1034, and a co-occurrence extraction unit 1035. Demonstrate.

A functional configuration of the control unit 103 of the server 10 will be described below.

The text information acquisition unit 1031 acquires text data to be analyzed. The text data to be analyzed is assumed to be text data consisting of a plurality of sentences. Further, the text information acquisition unit 1031 does not necessarily acquire the text data itself, and may acquire the image data of the document and convert it into text by OCR (Optical Character Recognition) or the like.

The text information acquisition unit 1031 is connected via the input/output interface 15 to computer-readable recording such as USB (Universal Serial Bus) memory, CD-ROM (Compact Disc-Read Only Memory), DVD (Digital Versatile Disc), and HDD. Text data may be obtained from the medium.

The text information acquisition unit 1031 may acquire text data from the network via the communication IF 15 including a NIC (Network Interface Card controller) for performing wired or wireless communication.

The named entity extracting unit 1032 extracts named entities from the text data acquired by the text information acquiring unit 1031 and attaches tags to them.

A named entity is a key element in extracting information from data. The named entity includes proper nouns such as personal names, place names, organization names, and product names, numerical expressions such as quantities, amounts, and percentages, and time expressions such as dates and times. A named entity may also include a particular substance, structure, function, processing method, analytical method, terminology, and the like. Note that the range of specific expressions may be defined in advance from all expressions.

A tag indicates identification information for classifying a named entity. For example, tags representing proper nouns such as "person's name" and "place name", and tags representing numerical expressions such as "quantity", "date" and "time" are prepared and attached to specific expressions. good too.

In addition, for the purpose of analyzing technical documents and documents in the natural science field, tags such as event tags that indicate "events" and analysis tags that indicate "analysis methods" for events may be prepared. good. Prepare a plurality of tags in advance.

Furthermore, tags may be defined with a hierarchical structure.

For example, in professional literature such as papers, especially in the field of natural science, there are “analytical methods” such as experiments, tests, inspections, and grounds that support certain “events” such as hypotheses, states, and results. is assumed to be shown. It is assumed that the "event" is indicated in the form of "the object is the state", such as "A is B". In addition, it is assumed that the "analytical method" is indicated in the form of "results obtained by tests, etc.", such as "A is B" in order to support the event.

Based on the above, for example, tags may be defined with a hierarchical structure such as "event: target", "event: state", "analysis: test method", "analysis: test result", etc. . In this case, "event" and "analysis" are defined as superordinate concepts, and "object" and "test method" are defined as tags having a hierarchical structure as subordinate concepts.

As a method of extracting named entities, the named entity extraction unit 1032 may prepare a dictionary representing named entities in advance, use the dictionary to extract named entities from an input sentence, and tag them. .

The named entity extraction unit 1032 may use a commonly known method such as NER (Named Entity Recognition) as a named entity extraction method. Also, the named entity extraction unit 1032 may extract named entities after creating a model for named entity extraction by machine learning using learning data.

In a general NER, proper nouns, numerical expressions, and the like are the named entities to be extracted. Therefore, if you want to extract expressions other than this as named entities, such as events and analysis methods, prepare training data for the expressions you want to extract, and then create a model for extracting named entities. to create

For example, the named entity extraction unit 1032 prepares learning data with predefined tags attached to text data prepared in advance, and uses these data to prepare a model for machine learning for named entity extraction. Keep After morphologically analyzing the text data to be analyzed acquired by the text information acquiring unit 1031, the NER method is used using the created model. As a result, the named entity extracting unit 1032 extracts the named entity and attaches one of the predefined tags to the extracted named entity.

The named entity extracting unit 1032 extracts and tags named entities, so that the text data acquired by the text information acquiring unit 1031 is tagged with words that are highly likely to have a particular meaning. It is possible to extract

The dependency extraction unit 1033 extracts dependencies. For this, a commonly used dependency analyzer may be used.

Dependency refers to a relationship between a pair of two words that has a relationship between a subject and a predicate, a modifier and a modified word, etc. when syntactic analysis is performed. In the case of a subject and a predicate, the subject is the source of the dependency and the predicate is the destination of the dependency. Thus, a dependency consists of a dependency source and dependency target pair.

A general dependency analyzer extracts relationships between subjects and predicates and relationships between modifiers and modified words as dependencies. The dependency extracting unit 1033 may extract both or either one of them as a dependency.

In addition, as a method of dependency analysis that is generally used, there is also a method of analyzing clue expressions such as "because of" and "because of" and extracting dependencies by focusing on the keywords before and after it. The dependency extracting unit 1033 may perform extraction using clue expressions and/or using subject/predicate relationships and modifier/modified word relationships.

Also, in the case of general dependency extraction, when W1, W2, W1W2 (for example, W1 is "precision", W2 is "machine", W1W2 is "precision machine", etc.) are named entities, W1- W2 may be extracted as a dependency. At this time, if a noun phrase is extracted as a dependency, the noun phrase may be searched further to analyze the dependency destination, and a dependency such as W1-W3 may be extracted. The term "noun phrase" as used herein refers to a phrase consisting of a plurality of words and having the same function as a single noun as a whole.

The dependency extracting unit 1033 analyzes the dependency of the text data acquired by the text information acquiring unit 1031, and then extracts the dependency including the named entity tagged by the named entity extracting unit 1032. FIG. At this time, it is more desirable to extract only dependencies in which both the dependency source and the dependency destination are named entities.

By extracting dependencies in the dependency extracting unit 1033, it is possible to extract relationships between words having some meaning, such as cause and effect, purpose and means, event and analysis method, and the like.

Moreover, by extracting only dependencies that include named entities, especially dependencies in which both the source and destination of the dependency are named entities, it is possible to extract particularly important expressions in the text.

For example, when all of W1, W2, W3, and W4 are named entities, the dependency extracting unit 1033 takes the first word W1 as the source of the entity. When there is a dependency between the first dependency W1-W3 and the second dependency W2-W4 whose dependency is the second word W2, and the compound word W1W2 is also a named entity , W1W2-W3, and W1W2-W4, the compound word W1W2 is used as the dependency source, and the first dependency target word W3 and the second dependency target word W4 It may be extended by adding dependencies assuming that there are dependencies. A compound word is a word formed by combining two or more originally independent words to have a new meaning and function as a single word.

Note that the expansion of the dependency in the dependency extracting unit 1033 is not limited to compound words of two words, and compound words of three or more words may be expanded.

As for the dependency in the named entity, by expanding the dependency of the named entity into which it is segmented in the compound term, it is possible to appropriately grasp the content to be expressed in the text.

In addition, the dependency extracting unit 1033 can generate a higher-level dependency by extending the dependency using the named entity.

The co-occurrence degree calculation unit 1034 analyzes the co-occurrence relation in which the dependency appears in the sentence for each dependency extracted by the dependency extraction unit 1033, and extracts the text data acquired by the text information acquisition unit 1031. The number of sentences in which any of the dependencies appears and the number of sentences in which the co-occurrence of the dependencies is analyzed, and the ratio is calculated as the degree of co-occurrence. For this, a generally used technique such as the Jaccard coefficient may be used.

For example, when the dependency extraction unit 1033 extracts dependencies W1-W2 and W3-W4, the co-occurrence calculation unit 1034 extracts W1-W2 or The ratio of the number of sentences including W3-W4 dependencies and the sentences including W1-W2 and W3-W4 dependencies is calculated as the degree of co-occurrence.

The co-occurrence calculation unit 1034 calculates the degree of co-occurrence of the dependencies, thereby making it possible to analyze combinations of dependencies with a high degree of co-occurrence.

The co-occurrence extraction unit 1035 extracts combinations of dependencies based on the degree of co-occurrence of dependencies.

For example, the co-occurrence extraction unit 1035 may extract a combination of dependencies having a degree of co-occurrence exceeding a threshold with respect to the degree of co-occurrence for any combination of dependencies calculated by the degree-of-co-occurrence calculation unit 1034. good.

For example, the co-occurrence extraction unit 1035 may extract a combination of dependencies Ki and Kj where Vij>T, where T is the threshold and Vij is the degree of co-occurrence of dependencies Ki and Kj.

For example, the co-occurrence extraction unit 1035 extracts the degree of co-occurrence for an arbitrary combination of dependencies calculated by the co-occurrence degree calculation unit 1034, starting from the selected dependency, starting with the highest degree of co-occurrence. The numbers specified in order, that is, the top N items may be extracted.

For example, if Vij is the degree of co-occurrence of dependencies Ki and Kj, the top N Kj may be extracted from Ki as a starting point in descending order of Vij.

For example, the co-occurrence extracting unit 1035 focuses on the degree of co-occurrence with a specific tag including hierarchical relationships with respect to the degree of co-occurrence with respect to any combination of dependencies calculated by the co-occurrence calculating unit 1034. may be extracted.

The co-occurrence extraction unit 1035 extracts the top N co-occurrence relationships that have a degree of co-occurrence exceeding a threshold value or that have a certain dependency as a starting point, thereby extracting combinations of co-occurrence relationships that are particularly meaningful in text data. can be extracted, and it is possible to extract relationships that are particularly meaningful and important from a large amount of text data.

(Specific example of processing in named entity extraction unit 1032)
A specific example of named entity extraction by the named entity extraction unit 1032 is shown below. For example, as shown in Figure 3, there is a sentence "On June 29, 2006, Prime Minister Koizumi visited Washington and had a summit meeting with President Bush." , 'place name', 'quantity', 'date', and 'time'.

The named entity extraction unit 1032 first performs morphological analysis and decomposes into strings of morphemes. In addition, the named entity extraction unit 1032 prepares a learning model for extracting predefined tags of "person's name", "place name", "quantity", "date", and "time". Then, the named entity extraction unit 1032 tags the sentences decomposed into morpheme sequences.

In the example of FIG. 3, "June 29, 2006" represents a date, "Koizumi" represents a person's name, "Washington" represents a place name, and "Bush" represents a person's name.

As shown in FIG. 3, the named entity extraction unit 1032 performs tagging after extracting words to be named named entities. This makes it possible to extract and classify words corresponding to tags whose unique expressions are defined in advance.

(Specific example of processing of the dependency extraction unit 1033)
A specific example of dependency extraction by the dependency extraction unit 1033 will be shown. As shown in FIG. 4, for example, assume that there are two sentences, "System shutdown has occurred" and "Possibility of power failure is suspected." The dependency extraction unit 1033 extracts dependencies. At this time, if the dependencies are analyzed using a general dependency analyzer, "system - stop", "stop - occur", "power supply - failure", "power supply - possibility", "abnormality - possibility" , “anomaly-guessed”, and “possible-guessed” dependencies can be extracted.

If there is a noun phrase, the dependency extracting unit 1033 extracts the destination of the previous dependency, skipping the dependency of "power supply-abnormality" and extracting only "abnormality-possibility".

The dependency extracting unit 1033 extracts only the tagged dependency as a named entity by the named entity extracting unit 1032 .

For example, when there are two types of tags "performance" and "state" as specific expressions, the sentences in FIG. 4 are as shown in FIG. Tags such as status, "power supply"=performance, "abnormality"=status, "possibility"=status, and "speculation"=status are attached. Also, the compound word "power failure" can be tagged as "power failure" = state.

When the dependency extracting unit 1033 extracts only the dependency in which both the words of the dependency are tagged, both the source and destination of the dependency are tagged as unique expressions. -Stop", "Stop-Occurrence", "Power Source-Possibility", "Abnormality-Possibility", "Abnormality-Inference", and "Possibility-Inference" are extracted.

Furthermore, when there are relationships of W1-W3 and W2-W4 and W1W2 is a named entity, the dependency extracting unit 1033 expands the relationships to W1W2-W3 and W1W2-W4. Here, since there is a relationship of W1=“power supply” and W2=“abnormality”, “power supply abnormality-possible” and “power supply abnormality-guess” are extracted as dependencies.

(Specific processing in co-occurrence calculation unit 1034)
The co-occurrence calculation unit 1034 calculates the co-occurrence degree for the dependencies extracted by the dependency extraction unit 1033 . For example, assume that the dependency extracting unit 1033 has extracted W1-W2 and W3-W4 dependencies. At this time, the total number of sentences including either or both of W1-W2 and W3-W4 dependencies is 100, and the total number of sentences including both W1-W2 and W3-W4 dependencies is 100. , the degree of co-occurrence is (number of sentences including W1-W2 or/and W3-W4 dependencies)/(number of sentences including W1-W2 and W3-W4 dependencies) = Calculated as 0.1.

(Specific processing in co-occurrence extraction unit 1035)
The co-occurrence extraction unit 1035 extracts a combination of dependencies having a degree of co-occurrence exceeding a threshold with respect to the co-occurrence degree for any combination of dependencies calculated by the co-occurrence calculation unit 1034 .

For example, when there are dependencies K1, K2, and K3, the degree of co-occurrence calculator 1034 determines that the degree of co-occurrence of K1 and K2 is 0.5, that of K1 and K3 is 0.1, and that of K2 and K3 is 0.2. After the calculation and setting the threshold as 0.3, the co-occurrence extracting unit 1035 extracts the co-occurrence relation of K1-K2 having a degree of co-occurrence exceeding the threshold.

(Modification 1:)
Hereinafter, a modified example in which not only the co-occurrence of dependencies but also the co-occurrence of dependencies and words or the co-occurrence of words is extracted will be described.

The co-occurrence calculation unit 1034 calculates, as a degree of co-occurrence, the rate of co-occurrence of the dependencies extracted by the dependency extraction unit 1033. The degree of co-occurrence with the word extracted as the named entity in 1032 and/or the degree of co-occurrence between the words extracted as named entity by the named entity extraction unit 1032 is calculated.

The co-occurrence extracting unit 1035 extracts the degree of co-occurrence exceeding a threshold for any combination of dependencies, a combination of a dependency and a word, and/or a combination of words calculated by the co-occurrence calculation unit 1034. Dependency combinations with origins, dependency-word combinations, and/or word-word combinations are extracted. At this time, different threshold values may be set for combinations of dependencies, combinations of dependencies and words, and combinations of words.

When extracting combinations of dependencies, combinations of dependencies and words, and/or combinations of words, the co-occurrence extraction unit 1035 extracts the tags given by the named entity extraction unit 1032 to those given specific tags. A limited co-occurrence relationship may be extracted.

By extracting co-occurrence relationships as described above, it is possible to extract not only co-occurrence relationships of dependencies but also co-occurrences of dependencies and words and/or words, and use them for analysis.

(Modification 2: Analysis example of specialized literature)
A modified example for analyzing specialized documents will be shown below. In technical literature, especially in the field of science and technology, phenomena such as hypotheses, conclusions, assertions, states, performances, properties, etc., and analytical methods (experimental methods, inspection methods) to support them in literature such as papers. is often shown. Therefore, by extracting events and analytical methods, important points can be extracted within the text. Therefore, two types of tags, "event" and "analysis", are prepared as tags that particularly indicate specific expressions. Assuming that subclassifications of "event" include, for example, "case" and "site", and subclassifications of "analysis" include "test item" and "test result", "event: case", Prepare tags for "event: site", "event: state", "analysis: inspection item", and "analysis: inspection result" and explain the following.

The text information acquisition unit 1031 acquires, for example, papers in the field of science and technology as text data. As shown in FIG. 6, it includes an example sentence of "yellowing occurred in the bulbar conjunctiva. The possibility of jaundice was suspected, and abnormal bilirubin was confirmed by a blood test." Specific processing is shown as what can be done.

The named entity extraction unit 1032 acquires named entities and tags them. For example, the sentence shown in FIG. 6 is, as shown in FIG. Jaundice" = Event: Case, "Symptom" = Event: State, "Possibility" = Event: State, "Inference" = Event: State, "Blood" = Event, Site, "Test" = Analysis: Test item, " Bilirubin” = analysis: test item, “abnormality” = analysis: test result, “confirmation” = event: state, and the named entity is extracted and tagged.

Furthermore, the named entity extracting unit 1032 tags compound words as follows: “bulbar conjunctiva” = event: site, “jaundice symptom” = event: case, “blood test” = analysis: inspection item. . FIG. 8 shows a specific example in which compound words are also tagged.

The dependency extraction unit 1033 extracts dependencies. Analyzing the dependency including the named entity for the sentence from which the named entity is extracted as shown in FIG. 7 results in the left side of FIG. That is, "eye-conjunctiva", "eye-yellowing", "conjunctiva-yellowing", "conjunctiva-occurrence", "yellowing-occurrence", "occurrence-occurred", "jaundice-symptom", "jaundice- Possible", "Symptom-Possible", "Symptom-Inferred", "Possible-Inferred", "Blood-Test", "Blood-Abnormal", "Test-Abnormal", "Test-Confirm", "Bilirubin -abnormal", "bilirubin-confirmed", "abnormal-confirmed", and "confirmed-had" are extracted as dependencies.

If there is a noun phrase, the dependency extracting unit 1033 searches for and extracts further dependencies, so that "eyeball-conjunctiva", "jaundice-symptoms", and "blood-examination" are not extracted. Dependencies such as those on the right side of FIG. 9 are extracted.

Further, the dependency extracting unit 1033 extracts only dependencies tagged as named entities by the named entity extracting unit 1032 for both the dependency source and the dependency destination. That is, "eyeball-conjunctiva", "eyeball-yellowing", "conjunctiva-yellowing", "conjunctiva-occurrence", "yellowing-occurrence", and "jaundice-symptom" circled on the right side of FIG. , "jaundice - probable", "symptom - probable", "symptom - probable", "possible - presumed", "blood - test", "blood - abnormal", "test - abnormal", "test - confirmed , ``bilirubin-abnormal'', ``bilirubin-confirmed'', and ``abnormal-confirmed'' are extracted as dependencies.

The dependency extracting unit 1033 expands the relationships of W1-W3 and W2-W4 to W1W2-W3 and W1W2-W4 when W1W2 is a named entity. For example, "ocular-yellowing", "conjunctiva-yellowing", "conjunctiva-occurrence" becomes "bulbar conjunctiva-yellowing", "bulbar conjunctiva-occurrence", and "jaundice-probable", "symptom-probable", "symptom - guess" becomes "jaundice-probable" and "jaundice-speculation", and "blood-abnormal", "test-abnormal" and "test-confirm" become "blood test-abnormal" and "blood test-confirm". That is, the dependency extracting unit 1033 extracts the dependency shown on the lower right side of FIG.

The co-occurrence degree calculation unit 1034 calculates the co-occurrence degree using a generally used technique such as the Jaccard coefficient. In this modification, the degree of co-occurrence is calculated particularly from two viewpoints, one regarding events and one regarding analysis.

First, from the viewpoint of events, the degree of co-occurrence is extracted for a dependency whose source is an event and whose destination is also an event. That is, when there is a dependency W1-W2 and a dependency W3-W4 related to an event, the number of sentences containing both W1-W2 and W3-W4 is The degree of co-occurrence is calculated by dividing (dividing) by the number of sentences with uke.

Next, from the viewpoint of analysis, the degree of co-occurrence may be calculated for a combination of a word with analysis as a tag, in addition to a dependency with an event as a tag for both the dependency source and the dependency receiver. In other words, when there is a dependency with an event tag of W1-W2 and a word with an analysis tag of B1, the number of sentences containing both the dependency of W1-W2 and the sentence containing B1 is Calculate the degree of co-occurrence by dividing (dividing) by the number of sentences that contain the words of Received and/or B1.

For example, if the text has dependencies on events W1-W2 and W3-W4 and words with analysis tags B1 and B2, (W1-W2, B1) (W1-W2, B2) ( The degree of co-occurrence is calculated for any combination of W3-W4, B1) (W3-W4, B2).

The co-occurrence extraction unit 1035 extracts combinations whose degree of co-occurrence exceeds a threshold in each of the event viewpoint and analysis viewpoint.

If the text is tagged with a hierarchical structure, the co-occurrence extracting unit 1035 may extract the text by focusing on the degree of co-occurrence with a specific tag including the hierarchical relationship.

(Processing flow)
FIG. 10 shows the flow of processing of the information analysis device according to Embodiment 1 of the present disclosure.

The text information acquisition unit 1031 acquires text data to be analyzed (step S1031).

The named entity extraction unit 1032 extracts named entities from the text data acquired by the text information acquisition unit 1031, and adds tags (step S1032). Note that the type of tag may be determined in advance.

The dependency extracting unit 1033 analyzes the dependency of the text data from which the named entity has been extracted (step S1033).

When extracting a dependency, if a noun phrase is extracted as a dependency, the sentence may be further searched to extract the dependency. Alternatively, the named entity extracting unit 1032 may extract only dependencies in which a word tagged as a named entity is a dependency source and a dependency destination.

The co-occurrence degree calculation unit 1034 calculates the degree of co-occurrence for the extracted dependency words (step S1034).

The co-occurrence extraction unit 1035 extracts a combination of dependencies having a degree of co-occurrence exceeding the threshold (step S1035).

(effect)
According to this embodiment, as a premise for dependency analysis, named entities are extracted, and only the dependencies related to named entities are extracted. It becomes possible to extract dependencies.

In addition, by calculating dependencies and co-occurrence degrees and extracting dependencies with a high degree of co-occurrence, it is possible to extract causal relationships, And/or it becomes possible to analyze the relationship between the event and the analysis method.

<Embodiment 2>
Embodiment 2 further includes a function of visually displaying to the user the analysis performed on the acquired text information.

<Configuration of information analysis device>
FIG. 11 shows the configuration of the server 20 in this embodiment. The control unit 203 of the server 20 includes a text information acquisition unit 1031 , a named entity extraction unit 1032 , a dependency extraction unit 1033 , a co-occurrence calculation unit 1034 , a co-occurrence extraction unit 1035 and a display unit 2036 . The text information acquisition unit 1031 , named entity extraction unit 1032 , dependency extraction unit 1033 , co-occurrence calculation unit 1034 , and co-occurrence extraction unit 1035 are the same as the control unit 103 of the server 10 .

The display unit 2036 visually expresses and displays the combination of dependencies analyzed by the co-occurrence extraction unit 1035 to the user. Specifically, after displaying the word of the source of the dependency and the word of the destination of the dependency, the dependencies may be connected by a line and represented in the form of a network diagram.

The display unit 2036 may visually represent and display to the user the dependency relationships of the words tagged with the event tags and the combinations of the words tagged with the analysis tags.

The display unit 2036 may display a combination of dependencies including the hierarchical structure when a tag having a text hierarchical structure is attached.

The display unit 2036 may express the degree of co-occurrence by changing the line thickness or color of the network diagram. In addition, the display unit 2036 may also display the degree of co-occurrence.

(Concrete example)
FIG. 12 shows a specific example of a visual representation of the dependency displayed by the display unit 2036. As shown in FIG. The display unit 2036 visually expresses and displays the dependencies analyzed by the dependency extraction unit 1033 using, for example, a network diagram representation. In the example of FIG. 12, for events, combinations of dependencies such as "system-stop", "power supply-possibility", and "power supply-abnormality", and these dependencies, "visual confirmation of parts", "capacitor", "cooling device""","Voltagemeasurement", and "Tester" that have a degree of co-occurrence exceeding the threshold are connected by a line and displayed. In the example of FIG. 12, the dependency relation of the event is displayed with a white background, and the words of the analysis method are displayed with a black background.

When the co-occurrence extracting unit 1035 extracts the co-occurrence relationship in consideration of the hierarchical structure of the tags, the display unit 2036 may display the hierarchical structure of the tags.

For example, with regard to analysis method tags, there are tags for "analysis: cause identification method" attached to "visual confirmation of parts" and "voltage measurement", and there are tags for "analysis: object" and tags for "capacitor" and " cooling device", a tag of "analysis: measuring equipment", and a tag of "tester". Here, when the co-occurrence relationship is extracted by focusing on "analysis: cause identification method", the co-occurrence relationship in the event and the relationship between "analysis: cause identification method" are extracted and illustrated.

FIG. 13 shows a specific example of visual expression when co-occurrence relations are extracted by paying attention to "analysis: method of specifying cause" for the dependency displayed by the display unit 2036. In FIG. Compared to FIG. 12, only the co-occurrence relationship of interest is extracted, making it easier to grasp the relationship.

Note that FIG. 13 focuses on "analysis: cause identification method", so only co-occurrence relationships between dependencies and words tagged with such tags are extracted. On the other hand, since "capacitor", "cooling device", and "tester" have a co-occurrence relationship with the words "visual confirmation of parts" or "voltage measurement", the co-occurrence relationship is extracted by connecting them. , are shown.

The display unit 2036 may display the co-occurrence degree calculated by the co-occurrence degree calculation unit 1034 together with the network diagram.

In this way, by using visual expressions, it becomes possible for the user to visually understand not only the relationships between dependencies but also the relationships between important words.

(Processing flow)
FIG. 14 shows the flow of processing of the information analysis device according to the second embodiment of the present disclosure.

The text information acquisition unit 1031 acquires text data to be analyzed (step S1031).

The named entity extraction unit 1032 extracts named entities from the text data acquired by the text information acquisition unit 1031, and adds tags (step S1032). Note that the type of tag may be determined in advance.

The dependency extraction unit 1033 extracts dependencies from the text data from which the named entity has been extracted (step S1033).

When extracting a dependency, if a noun phrase is extracted as a dependency, the sentence may be further searched to extract the dependency. Alternatively, the named entity extracting unit 1032 may extract only dependencies in which a word tagged as a named entity is a dependency source and a dependency destination.

The co-occurrence degree calculation unit 1034 calculates the degree of co-occurrence for the extracted dependency words (step S1034).

The co-occurrence extraction unit 1035 extracts a combination of dependencies having a degree of co-occurrence exceeding the threshold (step S1035).

The display unit 2036 displays the dependency to the user using a visual representation (step S2036). At this time, the display unit 2036 may also display the degree of co-occurrence and change the display method according to the degree of co-occurrence.

(effect)
According to the present embodiment, displaying extracted dependencies in consideration of the degree of co-occurrence helps the user to accurately grasp important words.

Especially in specialized fields and science and technology literature, there are many technical terms, so it has been difficult to extract important or meaningful words using the natural language processing methods that have been generally used so far. It is possible to express it after extracting it to

Although the description of the embodiment is finished above, the above embodiment is merely an example. Therefore, the specific configurations and processing contents of the servers 10 and 20 are not limited to those described in the above embodiment.

The information analysis apparatus according to the present disclosure may not only be implemented on a stand-alone computer, but may also operate as a server computer, for example.

For example, as shown in FIG. 15, a server 10 and a user terminal 30 operated by a user are connected via a network, the user terminal 30 is operated to cause the server 10 to analyze the document, and the result is transmitted to the user terminal. 30 may be displayed.

Also, the information analysis apparatus according to the present disclosure may realize its functions by executing a program by a computer, for example, without using the above apparatus. The program for realizing the function of the information analysis device can be read from a computer such as USB (Universal Serial Bus) memory, CD-ROM (Compact Disc-Read Only Memory), DVD (Digital Versatile Disc), HDD (Hard Disc Drive), etc. It may be stored in a possible recording medium, or may be downloaded to a computer via a network.

Although the preferred embodiments of the present disclosure have been described above, the present disclosure is not limited to such specific embodiments, and the present disclosure includes the invention described in the claims and their equivalents. be Further, the configurations of the apparatuses described in the above embodiments and modified examples can be appropriately combined as long as there is no technical contradiction.

Reference Signs List 10, 20 Server 11 Processor 12 Memory 13 Storage 14 Communication IF 15 Input/output IF 30 User terminal 101 Communication unit 102 Storage unit 103 Control unit 1031 1032 Named entity extraction unit 1033 Dependency extraction unit 1034 Co-occurrence degree calculation unit 1035 Co-occurrence extraction unit 2036 Display unit 211 Specific examples displayed by the display unit

Claims (16)

A program for causing a computer having a processor to perform processing, the processor comprising:
a text information acquisition step for acquiring text information;
a named entity extraction step of extracting a named entity from the text information and attaching a tag;
a dependency extracting step of analyzing dependencies from the text information and extracting dependencies in which one or both of the dependency source and the dependency receiver are unique expressions;
a co-occurrence degree calculating step of calculating a co-occurrence degree for a combination of dependencies in the dependencies extracted in the dependency extracting step from the text information;
a co-occurrence extracting step of extracting a combination of dependencies based on the degree of co-occurrence. 2. The program according to claim 1, wherein said tags include tags having a hierarchical structure. 3. The program according to claim 1, wherein said unique expression and said tag include at least an event tag representing an event and an analysis tag representing an analysis method for the event. 4. The named entity extraction step extracts a named entity from the text information using a model for entity extraction using machine learning, and attaches the tag to the text information. The program described in Section. 5. The program according to any one of claims 1 to 4, wherein the dependency extraction step analyzes a relationship between a modifier and a modified word as a dependency. 6. The program according to claim 5, wherein said dependency extracting step analyzes a word analyzed by a predetermined clue expression as a dependency in addition to the relationship between modifiers and modified words. 7. The dependency extracting step according to any one of claims 1 to 6, wherein when a noun phrase is extracted as a dependency, the noun phrase is further searched to extract a dependency destination. program. In the dependency extraction step, there are a first dependency having a first word as a dependency and a second dependency having a second word as a dependency, and the first word and the second word are included. When a word constitutes a compound word, add each of the dependencies whose dependency is the compound word and whose dependencies are the first and second dependencies. A program according to any one of claims 1 to 7, extended. 9. The co-occurrence degree calculating step according to any one of claims 1 to 8, wherein the degree of co-occurrence is calculated for combinations of dependencies, combinations of dependencies and words, and/or combinations of words and words. program. 10. The program according to any one of claims 1 to 9, wherein said co-occurrence extracting step extracts a combination of dependencies whose degree of co-occurrence exceeds a threshold. 11. The program according to any one of claims 1 to 10, wherein said co-occurrence extracting step extracts a specified number of combinations of dependencies in descending order of degree of co-occurrence with the selected dependency. . 12. The tag according to any one of claims 1 to 11, wherein the tags include tags having a hierarchical structure, and the co-occurrence extracting step extracts a combination of dependencies according to the degree of co-occurrence with the selected tag. The program described in Section. 13. The program according to any one of claims 1 to 12, further comprising a display step of displaying combinations of dependencies extracted by said co-occurrence extraction step. 14. The program according to claim 13, wherein said displaying step displays combinations of dependencies together with said degrees of co-occurrence. A method for causing a computer comprising a processor to perform processing, the method comprising:
a text information acquisition step for acquiring text information;
a named entity extraction step of extracting a named entity from the text information and attaching a tag;
a dependency extracting step of analyzing dependencies from the text information and extracting dependencies in which one or both of the dependency source and the dependency receiver are unique expressions;
a co-occurrence degree calculating step of calculating a co-occurrence degree for a combination of dependencies in the dependencies extracted in the dependency extracting step from the text information;
a co-occurrence extraction step of extracting a combination of dependencies based on the degree of co-occurrence. An information analysis device comprising a control unit, wherein the control unit
a text information acquisition step for acquiring text information;
a named entity extraction step of extracting a named entity from the text information and attaching a tag;
a dependency extracting step of analyzing dependencies from the text information and extracting dependencies in which one or both of the dependency source and the dependency receiver are unique expressions;
a co-occurrence degree calculating step of calculating a co-occurrence degree for a combination of dependencies in the dependencies extracted in the dependency extracting step from the text information;
and a co-occurrence extracting step of extracting a combination of dependencies based on the degree of co-occurrence.

Images