JP6813432B2 - Document processing equipment, document processing methods and programs - Google Patents

Description

The present invention relates to a device, a method and a program for extracting phrases that appear unevenly in one of a set of documents.

Conventionally, in the field of natural language processing, a method for extracting words that appear unevenly in one of a set of documents has been known. For example, Non-Patent Document 1 is a method for extracting technical terms used exclusively for ordinary people from among technical terms, and numerically indicates a bias in the use of technical terms in a corpus for general public and a corpus for professionals. It describes a method of extracting technical terms that are biased toward the corpus for the general public as technical terms for the general public.

Ran Kim, Rika Kiryu, Asuko Kondo, Makio Tanaka, "Attempts to Extract'Technical Terms for the General Public'-Examples of Medical Terms-", Proceedings of the 2008 Spring Meeting of the Japanese Society, Japanese Society, May 2008 Month, pp.199-206

Conventionally, it has been generally difficult to quickly understand the contents of the examination viewpoint description document that describes the examination viewpoint, which is used in the examination work. In order to help the quick understanding of the examination viewpoint description document, it is conceivable to use the above-mentioned prior art to extract and list only the part of the document in which the examination viewpoint is described. It was not possible to extract accurately by the method using.

The present invention has been made in view of such circumstances, and an object of the present invention is to accurately extract rules from a rule description document.

In order to solve the above problems, the document processing apparatus according to the present invention includes a dependency structure of clauses constituting a rule-described document, a dependency analysis unit for extracting a dependency structure of clauses constituting a comparison target document, and a dependency analysis unit. For each of the interleaved clauses constituting the rule-described document, the dependency structure of the clause extracted from the rule-described document by the dependency analysis unit and the dependency structure of the clause extracted from the comparison target document are used. Of the dependency ratio calculation unit that calculates the likelihood ratio indicating the degree to which the dependency clause appears more biasedly in the rule description document than in the comparison target document, and the dependency clause that constitutes the rule description document. A predetermined number of related clauses in which the value of the likelihood ratio calculated by the likelihood ratio calculation unit is relatively large, or a related clause in which the value of the likelihood ratio calculated by the likelihood ratio calculation unit is larger than the threshold value. It is provided with a relational clause extraction unit for extracting.

In a preferred embodiment, the document processing apparatus is based on the dependency structure of the clause extracted from the rule description document by the dependency analysis unit for each of the dependency clauses extracted by the dependency clause extraction unit. , It is further provided with a relational element clause extraction unit for extracting the relational element clause.

In a more preferred embodiment, the above-mentioned document processing device uses the relational element clause extracted by the relational element clause extraction unit for each of the relational clauses extracted by the relational element clause extraction unit, and the word included in the relational element clause. It also has a section for classifying original clauses that classify based on the meaning of.

In a more preferred embodiment, the entwined clause extraction unit extracts the entangled clause representing the condition.

In a more preferred embodiment, the document processing apparatus analyzes the structure of the rule-described document and the structure of the comparison target document to form the comparison-target document corresponding to one sentence constituting the rule-described document. The document structure analysis unit for specifying the other sentences to be used is further provided, and the likelihood ratio calculation unit has the above-mentioned one for each of the interposition clauses constituting the one sentence. The likelihood ratio indicating the degree of appearance biased to the sentence is calculated, and the related clause extraction unit calculates the likelihood ratio of the related clauses constituting the one sentence, which is calculated by the likelihood ratio calculation unit. Extracts a predetermined number of related clauses having a relatively large value of, or a dependent clause having a likelihood ratio value calculated by the likelihood ratio calculation unit larger than the threshold value.

Further, the document processing method according to the present invention is a document processing method executed by a document processing apparatus, and has a dependency structure of clauses constituting a rule-described document and a dependency structure of clauses constituting a document to be compared. For each of the dependency analysis step to be extracted and the dependency clauses constituting the rule description document, the dependency structure of the clause extracted from the rule description document and the comparison target document were extracted from the dependency analysis step. Based on the dependency structure of the clause, the likelihood ratio calculation step for calculating the likelihood ratio indicating the degree to which the dependency clause appears more biasedly in the rule description document than in the comparison target document, and the rule description document Among the constituent clauses, the value of the likelihood ratio calculated by the likelihood ratio calculation step is relatively large, or the value of the likelihood ratio calculated by the predetermined number of clauses or the likelihood ratio calculation step. It includes a dependency clause extraction step for extracting a dependency clause in which is larger than the threshold.

Further, the program according to the present invention uses a computer as a dependency analysis unit for extracting a dependency structure of clauses constituting a rule description document, a clause dependency structure constituting a comparison target document, and the rule description document. For each of the constituent clauses, the dependency clause is based on the dependency structure of the clause extracted from the rule description document by the dependency analysis unit and the dependency structure of the clause extracted from the comparison target document. Of the likelihood ratio calculation unit that calculates the likelihood ratio indicating the degree to which is biased toward the rule description document rather than the comparison target document, and the related clauses that constitute the rule description document, the likelihood ratio calculation A predetermined number of involved clauses having a relatively large likelihood ratio value calculated by the unit, or a related clause having a likelihood ratio value calculated by the likelihood ratio calculation unit larger than the threshold value. It functions as a phrase extractor.

According to the present invention, rules can be accurately extracted from a rule description document.

It is a figure which shows an example of the structure of the document processing apparatus 1. It is a flow chart which shows an example of examination viewpoint extraction processing. It is a figure which shows an example of the table which is stored in the document storage part 103. It is a figure which shows an example of the result of the analysis by the dependency analysis unit 104. It is a figure which shows an example of the table which is stored in the dependency structure storage part 105. It is a figure which shows an example of the table which is stored in the examination viewpoint description document receiving section storage unit 107. It is a figure which shows an example of the table which is stored in the comparison target document receiving section storage unit 108. It is a figure which shows an example of the table which is stored in the receiving clause list storage unit 111. It is a figure which shows an example of the table stored in the clause list storage unit 113. It is a figure which shows an example of the table which is stored in the examination logic list storage part 116.

1. 1. Embodiment 1-1. Configuration FIG. 1 is a diagram showing an example of the configuration of the document processing apparatus 1 according to the embodiment of the present invention. The document processing device 1 is a computer including an arithmetic processing unit such as a CPU, a storage device such as an HDD, and a communication device such as a NIC, for extracting a viewpoint of examination from a document describing the viewpoint of examination. As shown in FIG. 1, the document processing device 1 includes a document input unit 101, a document structure analysis unit 102, a document storage unit 103, a dependency analysis unit 104, a dependency structure storage unit 105, and a receiving section extraction. Section 106, document receiving section storage section 107 for review viewpoint description, document receiving section storage section 108 for comparison, LLR (Log-Likelihood Ratio) calculation section 109, receiving section list creation section 110, receiving section list storage section It has functions of 111, a related section list creation unit 112, a related section list storage unit 113, a related section classification unit 114, an examination logic list creation unit 115, an examination logic list storage unit 116, and an output unit 117. Among these functions, the document storage unit 103, the dependency structure storage unit 105, the examination viewpoint description document receiver storage unit 107, the comparison target document receiver storage unit 108, and the receiver list storage unit 111 are involved. The section list storage unit 113 and the examination logic list storage unit 116 are realized by the storage device, and other functions are realized by the arithmetic processing unit executing the program stored in the storage device. The document processing device 1 may be composed of a plurality of server devices connected to each other by a communication line.

Among the functions provided in the document processing device 1, the document input unit 101 acquires the document data described in the examination viewpoint and the document data to be compared from the storage device or the communication device. Here, the examination viewpoint description document is a document in which the examination viewpoint is described. In other words, it is a runbook or rulebook. The examination viewpoint description document is mainly composed of a set of sentences consisting of a conditional clause and a main clause (for example, "B in the case of A"). As a specific example of the document describing the examination viewpoint, there is a notice of points to be noted. Here, the notice of notes is a document in which notes on implementation of the notified contents are described. Examples of notices include points to note regarding implementation due to the partial revision of the medical fee calculation method, points to note regarding the calculation of material prices for specified insurance medical materials, and the amount of expenses required for specific education and childcare. There are precautions for implementation due to the revision of the calculation standard and points for implementation due to the revision of the calculation standard for the amount of expenses required for designated disability welfare services. Next, the comparison target document is a document to be compared with the examination viewpoint description document. Since the comparison target document is a document that is referred to to extract only the examination viewpoint from the contents described in the examination viewpoint description document, the fields of the examination viewpoint description document and the description content overlap, and the creator is the same. Is preferable. In addition, it is preferable to have a document structure corresponding to the document describing the examination viewpoint. Specifically, it is preferable that the contents related to each other are described in the same order between the comparison target document and the examination viewpoint description document. A specific example of the document to be compared is a notification. When the notes on implementation due to the partial revision of the medical fee calculation method are entered as the examination viewpoint description document, the notification of the partial revision of the medical fee calculation method will be issued as the comparison target document. Entered. When the notes regarding the calculation of the material price of the specified insurance medical material are input as the examination viewpoint description document, the notification of the partial revision of the material price of the specified insurance medical material is input as the comparison target document. .. If the document describing the examination viewpoint contains notes on implementation due to the revision of the calculation standard for the amount of expenses required for specific education / childcare, etc., the document for comparison is the cost required for specific education / childcare, etc. A notification will be entered to revise a part of the amount calculation standard. If notes on implementation due to the revision of the calculation criteria for the amount of expenses required for designated disability welfare services, etc. are entered as a document describing the examination viewpoint, the costs required for designated disability welfare services, etc. A notice is entered to revise a part of the amount calculation standard.

The document structure analysis unit 102 analyzes the document structure of each of the examination viewpoint description document represented by the examination viewpoint description document data acquired by the document input unit 101 and the comparison target document represented by the comparison target document data. , Identify the sentences that make up the document to be compared, which correspond to each sentence that makes up the document that describes the examination viewpoint. At that time, the document structure analysis unit 102 identifies the correspondence between the sentences by using the numbers assigned to the headings of each sentence and the blank lines in the document as clues. Here, the heading is a heading of a division (part), chapter, section, paragraph, or the like. When the document structure analysis unit 102 identifies the corresponding sentences in the comparison target document for the sentences constituting the examination viewpoint description document, the document storage is associated with the corresponding sentence set and the identification information of the division to which those sentences belong. Stored in unit 103. If the document structure analysis unit 102 cannot specify the correspondence relationship between the examination viewpoint description document and the comparison target document on a sentence-by-sentence basis, the entire examination viewpoint description document and the comparison target document are combined with the corresponding sentence set. Is stored in the document storage unit 103.

The dependency analysis unit 104 divides the text of the document describing the examination viewpoint corresponding to each category stored in the document storage unit 103 and the text of the document to be compared into sentence units, performs the dependency analysis, and composes the sentence. Extract the dependency structure (in other words, the dependency tree) of the phrase to be used. At that time, for dependency analysis, for example, CaboCha (https://taku910.github.io/cabocha/) and KNP (http://nlp.ist.i.kyoto-u.ac.jp/index. You may use a well-known dependency analyzer such as php? KNP). When the dependency analysis unit 104 extracts the dependency structure of the clause from the text of the document describing the examination viewpoint corresponding to one division and the text of the document to be compared, the set of the extracted dependency structure and the identification of the one division are identified. The information is associated with the information and stored in the dependency structure storage unit 105.

The receiving section extraction unit 106 refers to the dependency structure of the clauses of the examination viewpoint description document and the comparison target document corresponding to each category stored in the dependency structure storage unit 105, and constitutes the dependency relationship. (Hereinafter, simply referred to as "reception clause") is extracted. The receiving clause extracted here is, in particular, the verb at the end of the sentence. When the receiving clause extraction unit 106 extracts the receiving clause by referring to the dependency structure of the clause of the examination viewpoint description document corresponding to one division, the receiving clause and the frequency at which the receiving clause appears in the one division (In other words, the number of occurrences) is stored in the examination viewpoint description document receiving section storage unit 107 in association with the identification information of the one category. On the other hand, when the receiving clause is extracted by referring to the dependency structure of the clause of the comparison target document corresponding to one division, the receiving clause and the frequency at which the receiving clause appears in the one division (in other words, the number of occurrences). Is stored in the comparison target document receiving section storage unit 108 in association with the identification information of the one category.

The LLR calculation unit 109 compares each receiving section stored in the examination viewpoint description document receiving section storage unit 107 with the classification of the examination viewpoint description document from which the receiving section is extracted with the same classification of the comparison target document. The logarithmic likelihood ratio, which indicates the degree of uneven appearance, is calculated using the formula of Equation 1 with reference to the document receiving section storage unit 107 described in the examination viewpoint and the document receiving section storage unit 108 to be compared.

In the formula of Equation 1, a represents the frequency with which the receiving clause from which the log-likelihood ratio is calculated appears in the classification of the examination viewpoint description document from which the receiving clause is extracted. b represents the frequency with which the receiving clause appears in the same category of the comparison target document. c represents the frequency with which the receiving clauses other than the receiving clauses appear in the same category of the examination viewpoint description document. d represents the frequency with which the receiving clauses other than the receiving clause appear in the same category of the comparison target document. N represents the number of documents (“2” in the case of this embodiment). Here, log is a common logarithm. When the LLR calculation unit 109 calculates the log-likelihood ratio for the receiving node, the calculated log-likelihood ratio is associated with the receiving node and stored in the examination viewpoint description document receiving section storage unit 107.

The receiving clause list creation unit 110 extracts a predetermined number of receiving clauses having a relatively large log-likelihood ratio value from the receiving clauses stored in the receiving clause storage unit 107 of the examination viewpoint description document. When a predetermined number of receiving clauses are extracted, the extracted receiving clauses, the identification information of the division from which the receiving clauses are extracted, and the log-likelihood ratio of the receiving clauses are associated with each other in the order of the log-likelihood ratio. It is stored in the list storage unit 111. The receiving clause list creation unit 110 is an example of the “relationship clause extraction unit” according to the present invention.

The dependency list creation unit 112 receives, for each reception clause stored in the reception list storage unit 111, among the dependency structures of the clauses of the examination viewpoint description document stored in the dependency structure storage unit 105. With reference to the dependency structure of the clause corresponding to the section from which the clause is extracted, the clause of the dependency that constitutes the dependency relationship with the clause (hereinafter, simply referred to as "dependent clause") is extracted. The relational clause extracted here is, in particular, a relational clause included in the same sentence as the receiving clause, and is a relational clause including an object directly related to the receiving clause. When the clause list creation unit 112 extracts the clause, the extracted clause, the receiving clause to which the clause is related, the identification information of the division from which the clause is extracted, and the clause appearing in the division. It is stored in the clause list storage unit 113 in association with the frequency of occurrence (in other words, the number of occurrences). The related clause list creation unit 112 is an example of the “related original clause extraction unit” according to the present invention.

The relational clause classification unit 114 is a relational clause extracted in the same division as the relational clause for each relational clause stored in the relational clause list storage unit 113, and is among the relational clauses having a common receiving clause. Classify based on the meaning of the words contained in the clause. At that time, the related clause classification unit 114 refers to a synonym dictionary (not shown) and assigns the same classification code to the related clause groups including synonyms with each other. Alternatively, the distance between the internodes is calculated using a well-known similarity calculation method, and the same classification code is assigned to the interlocking nodes whose calculated distance is within the threshold value. The classification code assigned to the related clause is stored in the related clause list storage unit 113 in association with the related clause.

The examination logic list creation unit 115 receives each of the receiving clauses stored in the dependency list storage unit 113 among the clauses of the examination viewpoint description document stored in the dependency structure storage unit 105. By referring to the dependency structure of the clause corresponding to the section in which the clause is extracted, we try to extract the dependency clause that constitutes the dependency relationship with the clause and represents the condition. The relational clause extracted here is, in particular, a relational clause included in the same sentence as the receiving clause, and is a relational clause directly related to the receiving clause. The examination logic list creation unit 115 uses keywords representing conditions such as "case", "time", and "time" as clues when extracting the clauses representing the conditions. When the examination logic list creation unit 115 extracts the relational clauses representing the conditions, the examination logic list creation unit 115 further extracts the relational clauses related to the extracted relational clauses and specifies them as the condition parts. In addition, a clause that is included in the same sentence as the clause that represents the extracted condition and that is directly related to the receiving clause stored in the clause list storage unit 113 and includes the object is extracted. Then, it is specified as a processing unit together with the receiving clause. If such a clause does not exist, the extraction may be omitted. When the condition unit and the processing unit are specified, they are stored in the examination logic list storage unit 116 in association with the identification information of the extracted categories. The examination logic list creation unit 115 is an example of the “relationship source clause extraction unit” according to the present invention.

The output unit 117 outputs the receiving clause list with reference to the receiving clause list storage unit 111. Further, the related clause list is output with reference to the related clause list storage unit 113. When outputting the related clause list, the output unit 117 groups and outputs a group of related clauses having the same classification code, which will be described later. Further, the output unit 117 outputs a list of examination logics with reference to the examination logic list storage unit 116. Here, the output refers to screen output of information, print output, or transmission of data representing information.

1-2. The document processing method executed by the operation document processing apparatus 1 will be described. Specifically, the examination viewpoint extraction process for extracting the examination viewpoint from the examination viewpoint description document will be described. FIG. 2 is a flow chart showing an example of the examination viewpoint extraction process. In the explanation of this operation example, the points to be noted in implementation due to the partial revision of the medical fee calculation method are input as the examination viewpoint description document, and the medical fee calculation method is partially revised as the comparison target document. It is assumed that a notification of the matter to be entered is input.

When the document input unit 101 acquires the examination viewpoint description document data and the comparison target document data, the document structure analysis unit 102 compares the acquired examination viewpoint description document with the examination viewpoint description document represented by the acquired examination viewpoint description document data. For each of the comparison target documents represented by the target document data, the document structure is analyzed, and the sentences constituting the comparison target document corresponding to each sentence constituting the examination viewpoint description document are specified (S1). When the corresponding sentence is specified in the comparison target document for each sentence constituting the examination viewpoint description document, the corresponding sentence set and the identification information of the category to which the sentence belongs are associated and stored in the document storage unit 103. FIG. 3 is a diagram showing an example of a table stored in the document storage unit 103. In the table shown in the figure, for example, it is a set of the text of the examination viewpoint description document and the text of the comparison target document in association with the classification identification information "Chapter 1", "Part 1" and "General rules". , Corresponding sets of sentences are stored.

When the correspondence between the texts of the examination viewpoint description document and the comparison target document is specified, the dependency analysis unit 104 compares the texts of the examination viewpoint description documents corresponding to each category stored in the document storage unit 103. The sentence of the target document is divided into sentence units, and the dependency analysis is performed to extract the dependency structure of the clauses constituting the sentence (S2). As an example, FIG. 4 shows a dependency structure that is output when the sentence "The initial examination fee is calculated when there is a medical practice called the first examination medically" is performed by the dependency analysis unit 104. It is a figure which shows an example. In the dependency structure shown in the figure, the first, fourth, seventh, tenth, 14, 17, 21 and 24 lines starting with the asterisk are the clause number and the clause number of the destination (in the case of no contact, "-1". The other lines are composed of the surface form of the word, the part of speech, the part of speech subclassification 1, the part of speech subclassification 2, the part of speech subclassification 3, the inflected type, the inflected form, the original form, the reading and the pronunciation. Asterisks in the other lines indicate that the information is not registered in the dictionary. The dependency structure shown in the figure shows, for example, that the phrase "medically" is related to the phrase "said". When the dependency analysis unit 104 extracts the dependency structure of the clause from the text of the document describing the examination viewpoint corresponding to each category and the text of the document to be compared, the set of the extracted dependency structure and the identification information of the category are obtained. It is stored in the dependency structure storage unit 105 in association with each other. FIG. 5 is a diagram showing an example of a table stored in the dependency structure storage unit 105. In the table shown in the figure, for example, the dependency structure of the clause of the examination viewpoint description document and the clause of the comparison target document are associated with the classification identification information "Chapter 1", "Part 1" and "General rules". A set of dependency structures is stored.

When the dependency structure is extracted for the text of the examination viewpoint description document corresponding to each category and the text of the document to be compared, the receiving clause extraction unit 106 corresponds to each category stored in the dependency structure storage unit 105. By referring to the dependency structure of the clauses of the document describing the examination viewpoint and the document to be compared, the receiver clauses constituting the dependency relationship are extracted (S3). When the receiving clause is extracted by referring to the dependency structure of the clause of the examination viewpoint description document corresponding to one classification, the receiving clause and the frequency at which the receiving clause appears in the one division are classified into the one classification. It is stored in the examination viewpoint description document receiving section storage unit 107 in association with the identification information of. FIG. 6 is a diagram showing an example of a table stored in the examination viewpoint description document receiving section storage unit 107. In the table shown in the figure, for example, in association with the classification identification information "Chapter 1", "Part 1" and "General rules", the receiving clause "state", the application frequency "48" and the LLR "80. 78 ”is stored. The LLR "80.78" is stored as a result of the execution of step S4 described later. On the other hand, when the receiving clause extraction unit 106 extracts the receiving clause by referring to the dependency structure of the clause of the comparison target document corresponding to one division, the receiving clause and the receiving clause appear in the one division. The frequency is stored in the comparison target document receiving section storage unit 108 in association with the identification information of the one category. FIG. 7 is a diagram showing an example of a table stored in the comparison target document receiving section storage unit 108. In the table shown in the figure, for example, the receiving clause "state" and the application frequency "8" are stored in association with the classification identification information "Chapter 1", "Part 1" and "general rule". ..

When the receiving clause is extracted from the examination viewpoint description document and the comparison target document, the LLR calculation unit 109 examines each receiving clause stored in the examination viewpoint description document receiving clause storage unit 107 from which the receiving clause is extracted. The logarithmic likelihood ratio, which indicates the degree of biased appearance of the comparison target document in the category of the viewpoint description document, is set to the examination viewpoint description document receiver storage unit 107 and the comparison target document receiver storage unit 108. It is calculated with reference to (S4). When the log-likelihood ratio is calculated for each receiving node, the calculated log-likelihood ratio is stored in the examination viewpoint description document receiving section storage unit 107 in association with the receiving node.

When the log-likelihood ratio is calculated for each receiving clause, the receiving clause list creating unit 110 has a relative value of the log-likelihood ratio among the receiving clauses stored in the receiving clause storage unit 107 of the document describing the examination viewpoint. A large predetermined number of receiving clauses are extracted (S5). When a predetermined number of receiving clauses are extracted, the extracted receiving clauses, the identification information of the division from which the receiving clauses are extracted, and the log-likelihood ratio of the receiving clauses are associated with each other in the order of the log-likelihood ratio. It is stored in the list storage unit 111. FIG. 8 is a diagram showing an example of a table stored in the receiving clause list storage unit 111. In the table shown in the figure, for example, the receiving clauses "implement" and LLR "326.54" are stored in association with the classification identification information "Chapter 1", "Part 1" and "general rules". There is.

When the receiving clause list is created, the dependency list creation unit 112 determines the clause of the examination viewpoint description document stored in the dependency structure storage unit 105 for each receiving clause stored in the receiving clause list storage unit 111. With reference to the dependency structure of the clause corresponding to the section in which the reception clause is extracted, the dependency clause constituting the dependency relationship with the reception clause is extracted (S6). When a related clause is extracted, the extracted related clause, the receiving clause to which the related clause is related, the identification information of the division from which the relevant clause is extracted, and the frequency at which the relevant clause appears in the relevant division are associated with each other. It is stored in the section list storage unit 113. FIG. 9 is a diagram showing an example of a table stored in the clause list storage unit 113. The table stored in the storage unit is created for each receiving node stored in the receiving section list storage unit 111. In the table shown in the figure, for example, the classification identification information "Chapter 1" and "Part 1" , And the relational clause "reason", the frequency of appearance "27", and the classification code "1" are stored in association with the "general rule". The classification code "1" is stored as a result of the execution of step S7 described later.

When the clause list is created, the clause classification unit 114 is a clause extracted in the same division as the clause for each clause stored in the clause list storage unit 113, and is a receiving clause. Is classified based on the meaning of the words included in the clauses in the common clauses (S7). When a classification code is assigned to each related clause, it is stored in the related clause list storage unit 113 in association with the related clause.

When each clause is classified, the examination logic list creation unit 115 describes the clause of the examination viewpoint description document stored in the dependency structure storage unit 105 for each receiving clause stored in the dependency list storage unit 113. Of the dependency structure of, the dependency structure of the clause corresponding to the extracted division is referred to, and the dependency clause that constitutes the dependency relationship with the reception clause of the Try to extract. When the clauses representing the conditions are extracted, the clauses related to the extracted clauses are further extracted and they are specified as the condition part. In addition, a clause that is included in the same sentence as the clause that represents the extracted condition and that is directly related to the receiving clause stored in the clause list storage unit 113 and includes the object is extracted. Then, it is specified as a processing unit together with the receiving clause (S8). When the condition unit and the processing unit are specified, they are stored in the examination logic list storage unit 116 in association with the identification information of the extracted categories. FIG. 10 is a diagram showing an example of a table stored in the examination logic list storage unit 116. In the table shown in the figure, for example, the condition part "in case of emergency" and the processing part "state the reason" are associated with the classification identification information "Chapter 1", "Part 1" and "General rules". Is stored.
The above is the description of the examination viewpoint extraction process.

In the document processing device 1 described above, the receiving clause list creating unit 110 creates a receiving clause list that appears specifically in the document describing the examination viewpoint. By referring to this list of receiving clauses, the reader who is going to read the document describing the examination viewpoint can know which clause in this document should be focused on to quickly understand the viewpoint of examination. In addition, in this document processing device 1, since the log-likelihood ratio is calculated by comparing the sentences related to each other between the document describing the examination viewpoint and the document to be compared, the log-likelihood ratio is calculated. Extraction accuracy is improved. Further, since the extraction is performed not in units of words but in units of phrases, it is possible to distinguish between "state" and "not describe", for example. In addition, since only the verbs that are the receiving clauses and the end of the sentences are extracted instead of all the clauses, the noise extraction is suppressed as compared with the case where they are not.

Further, in the document processing device 1, the related clause list creation unit 112 creates a list of related clauses related to the receiving clauses that specifically appear in the document describing the examination viewpoint. By referring to this list of related clauses, it becomes easier for readers who are going to read the examination viewpoint description document to understand the examination viewpoint. In addition, each related clause is classified by the related clause classification unit 114, and the related clauses having the same classification code are grouped and output, so that the outline of the examination viewpoint can be easily grasped.

Further, in the document processing apparatus 1, the examination logic list creation unit 115 extracts a list of examination logic including the condition unit and the processing unit. By referring to this list of examination logic, it is possible to grasp the outline of the examination viewpoint without reading the examination viewpoint description document.

2. 2. Modification Example The above embodiment may be modified as described below. In addition, one or more modified examples described below may be combined with each other.

2-1. Modification 1
The document structure analysis unit 102 may be omitted. In that case, the document storage unit 103 corresponds to a set of the entire examination viewpoint description document represented by the examination viewpoint description document data acquired by the document input unit 101 and the entire comparison target document represented by the comparison target document data. It is stored in the document storage unit 103 as a set of sentences to be written.

2-2. Modification 2
The above-mentioned document processing device 1 is supposed to process a Japanese document, but may be capable of processing a language other than Japanese. For example, when processing an English document, the dependency analysis unit 104 uses, for example, Maltparse (http://www.maltparser.org/) to determine the dependency structure of the phrases that make up the English document. It may be extracted.

2-3. Modification 3
The above-mentioned receiving clause extraction unit 106 extracts the receiving clause which is the verb at the end of the sentence, but the extracted receiving clause does not have to be limited to the receiving clause at the end of the sentence or the receiving clause of the verb.

2-4. Modification 4
In the above equation of Equation 1 for calculating the log-likelihood ratio, the common logarithm is taken, but the natural logarithm may be taken. Alternatively, it is not necessary to take the logarithm in the first place.

The formula for calculating the log-likelihood ratio is not limited to the above formula of Equation 1. The LLR calculation unit 109 may calculate the log-likelihood ratio using, for example, the equation of Equation 2 instead of the equation of Equation 1.

In the formula of Equation 2, a represents the frequency with which the receiving clause from which the log-likelihood ratio is calculated appears in the classification of the examination viewpoint description document from which the receiving clause is extracted. b represents the frequency with which the receiving clause appears in the same category of the comparison target document. c represents the frequency (in other words, the total number of receiving clauses) in which all receiving clauses appear in the same category of the examination viewpoint description document. d represents the frequency (in other words, the total number of receiving clauses) in which all receiving clauses appear in the same category of the comparison target document.

2-5. Modification 5
The above-mentioned receiving clause list creating unit 110 extracts the receiving clauses whose log-likelihood ratio value is smaller than the threshold value, instead of extracting a predetermined number of receiving clauses having a relatively large log-likelihood ratio value. May be good.

2-6. Modification 6
The above-mentioned clause list creation unit 112 extracts the clauses including the object directly related to the receiving clause, and the extracted clauses are the clauses and the objects directly related to the receiving clause. It is not necessary to limit the clause to the clause including.

2-7. Modification 7
The above-mentioned relational clause classification unit 114 is a relational clause extracted in the same division and classifies in the relational clauses having a common receiving clause, but is classified in the relational clauses extracted in different divisions. Or you may classify in the relational clauses that do not have a common receiving clause.

2-8. Modification 8
The program for realizing each function included in the document processing device 1 may be provided via a recording medium readable by the computer device. Here, the recording medium is, for example, a magnetic recording medium such as a magnetic tape or a magnetic disk, an optical recording medium such as an optical disk, an optical magnetic recording medium, a semiconductor memory, or the like. In addition, this program may be provided via a network such as the Internet.

2-9. Modification 9
The examination viewpoint description document is an example of the rule description document. Here, the rule description document is a document in which the rule is described. More specifically, for example, it is a document that describes the standard matters of human actions and administrative procedures.

1 ... Document processing device, 101 ... Document input unit, 102 ... Document structure analysis unit, 103 ... Document storage unit, 104 ... Dependency analysis unit, 105 ... Dependency structure storage unit, 106 ... Receiving clause extraction unit, 107 ... Examination Perspective description document receiving section storage unit, 108 ... Comparison target document receiving section storage unit, 109 ... LLR calculation unit, 110 ... receiving section list creating unit, 111 ... receiving section list storage unit, 112 ... related clause list creating unit, 113 ... Dependent section list storage unit, 114 ... Dependent section classification unit, 115 ... Examination logic list creation unit, 116 ... Examination logic list storage unit, 117 ... Output unit

Claims (7)

The dependency structure of the clauses that make up the rule description document, the dependency analysis unit that extracts the dependency structure of the clauses that make up the document to be compared, and
For each of the interleaved clauses constituting the rule description document, the dependency structure of the clause extracted from the rule description document by the dependency analysis unit and the dependency structure of the clause extracted from the comparison target document are used. Therefore, a likelihood ratio calculation unit that calculates a likelihood ratio indicating the degree to which the related clause appears more biasedly in the rule description document than in the comparison target document,
Among the related clauses constituting the rule description document, a predetermined number of related clauses having a relatively large likelihood ratio value calculated by the likelihood ratio calculation unit or the likelihood ratio calculation unit calculated the likelihood ratio. A document processing device including a dependency clause extraction unit that extracts a dependency clause whose likelihood ratio value is larger than a threshold value. For each of the dependency clauses extracted by the dependency clause extraction unit, the dependency clause extraction is performed by extracting the dependency clauses based on the dependency structure of the clauses extracted from the rule description document by the dependency analysis unit. The document processing apparatus according to claim 1, further comprising a unit. For each of the related clauses extracted by the related clause extraction unit, the related original clauses extracted by the related original clause extraction unit are classified based on the meanings of the words included in the related original clauses. The document processing apparatus according to claim 2, further comprising. The document processing apparatus according to claim 2, wherein the entwined clause extraction unit extracts the entangled clause representing a condition. A document structure analysis unit that analyzes the structure of the rule description document and the structure of the comparison target document and identifies another sentence constituting the comparison target document corresponding to one sentence constituting the rule description document. Further prepare
The likelihood ratio calculation unit calculates a likelihood ratio indicating the degree to which the related clauses appear more biased toward the one sentence than the other sentences for each of the related clauses constituting the one sentence. ,
The dependency clause extraction unit is a predetermined number of dependency clauses or the likelihood in which the value of the likelihood ratio calculated by the likelihood ratio calculation unit is relatively large among the dependency clauses constituting the one sentence. The document processing apparatus according to any one of claims 1 to 4, wherein a dependency clause in which the value of the likelihood ratio calculated by the degree ratio calculation unit is larger than the threshold value is extracted. A document processing method executed by a document processing device.
The dependency structure of the clauses that make up the rule description document, the dependency analysis step that extracts the dependency structure of the clauses that make up the document to be compared, and
For each of the interleaved clauses constituting the rule description document, the dependency structure of the clause extracted from the rule description document by the dependency analysis step and the dependency structure of the clause extracted from the comparison target document are used. Therefore, the likelihood ratio calculation step of calculating the likelihood ratio indicating the degree to which the related clause appears more biasedly in the rule description document than in the comparison target document, and
Among the related clauses constituting the rule description document, a predetermined number of related clauses having a relatively large likelihood ratio value calculated by the likelihood ratio calculation step or the likelihood ratio calculation step calculated the likelihood ratio. A document processing method including a dependent clause extraction step for extracting a related clause whose likelihood ratio value is larger than the threshold. Computer,
The dependency structure of the clauses that make up the rule description document, the dependency analysis unit that extracts the dependency structure of the clauses that make up the document to be compared, and
For each of the interleaved clauses constituting the rule description document, the dependency structure of the clause extracted from the rule description document by the dependency analysis unit and the dependency structure of the clause extracted from the comparison target document are used. Therefore, a likelihood ratio calculation unit that calculates a likelihood ratio indicating the degree to which the related clause appears more biasedly in the rule description document than in the comparison target document,
Of the related clauses constituting the rule description document, a predetermined number of related clauses having a relatively large likelihood ratio value calculated by the likelihood ratio calculation unit or the likelihood ratio calculation unit calculated the likelihood ratio. A program that functions as a dependency clause extraction unit that extracts a dependency clause whose likelihood ratio value is larger than the threshold value.

Images