JP2010170211A - Important sentence extraction program and device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To extract an important sentence from a document even when an important word corresponding to a field of the document is absent inside the document. <P>SOLUTION: An extraction target document input part 31 inputs an extraction object document imparted with a category according to operation of a user. An important word determination part 32 reads a first important word corresponding to the category imparted to the extraction target document and a second important word corresponding to the category related to the category from a hierarchical important word storage part 22. The important word determination part 32 determines the first and second important words as first and second extracting important words when the first and second important words are included in the extraction target document. An important word extraction part 33 extracts a sentence including at least one of the first and second extracting important words as the important sentence from the extraction target document. An important sentence output part 35 outputs the important sentences in order of an importance level calculated based on the first or second extracting important word included in the important sentence. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to an important sentence extraction program and an important sentence extraction apparatus for extracting an important sentence in a document composed of a plurality of sentences from the document.

  In recent years, an important sentence extraction apparatus that extracts an important sentence that appropriately represents the contents of a document composed of a plurality of sentences from the document is known. Thus, extracting an important sentence from a document is necessary for grasping the contents of the document without reading the whole sentence of the document.

  As a technique for extracting an important sentence from a document as described above, a technique (hereinafter referred to as a prior art) that can easily extract an important sentence in a document with high accuracy is disclosed (for example, a patent). Reference 1). According to this prior art, for example, an important expression table (important word dictionary) describing important expressions (important words) corresponding to the field of a document is prepared, and sentences including the important expressions are extracted as important sentences. .

Japanese Patent Laid-Open No. 11-272686

  However, in the above-described prior art, when an important word corresponding to the field of the document does not exist in the document, an important sentence cannot be extracted from the document.

  In the prior art, since the structure of the important word dictionary does not represent the relationship between the important words (that is, the hierarchical structure), the important words related to the important words corresponding to the field of the document are extracted from the important word dictionary. It is difficult to find.

  Accordingly, an object of the present invention is to provide an important sentence extraction program and an important sentence extraction apparatus capable of extracting an important sentence from the document even when the important word corresponding to the field of the document does not exist in the document. Is to provide.

  According to one aspect of the present invention, important word storage means for storing important words corresponding to each of categories into which a document is classified, wherein the categories related to the categories are represented in a hierarchical structure. An important sentence extraction apparatus comprising an external storage device having a word storage means and a computer using the external storage device, wherein the important sentence extraction program is executed by the computer, and the computer is operated by a user A step of inputting an extraction target document composed of a sentence including a plurality of words and having a category to which the extraction target document is classified, and the input extraction target document Reading the first important word corresponding to the category assigned to the important word storage means, and the read first important word is A step of determining whether or not the input target document includes the first important word, and if it is determined that the first important word is included in the extraction target document, A step of determining as an important word for extraction, a step of reading a second important word corresponding to a category associated with the category assigned to the input extraction target document from the important word storage means, and the reading Determining whether the second important word is included in the input extraction target document, and if it is determined that the second important word is included in the extraction target document, A key word as a second key word for extraction, and a sentence including at least one of the first key word for extraction and the key word for second extraction as a key sentence, Input extraction Extracting from an elephant document, and calculating the importance of the important sentence based on the determined first important word for extraction or second important word for extraction included in the extracted important sentence And an important sentence extraction program for executing the extracted important sentences in the order of the calculated importance.

  According to the present invention, it is possible to extract an important sentence from a document even if the important word corresponding to the field of the document does not exist in the document.

The block diagram which shows the hardware constitutions of the important sentence extraction apparatus which concerns on the 1st Embodiment of this invention. The block diagram which mainly shows the function structure of the important sentence extraction apparatus 30 shown in FIG. The figure which shows an example of the data structure of the hierarchy important word storage part 22 shown in FIG. The flowchart which shows the process sequence of the important sentence extraction process of the important sentence extraction apparatus 30 which concerns on this embodiment. The figure which shows an example of the extraction object document input by the extraction object document input part 31. FIG. The figure which shows an example when the important sentence whose score is 5 or more is output. The block diagram which mainly shows a function structure of the important sentence extraction apparatus which concerns on the 2nd Embodiment of this invention. The figure which shows an example of the data structure of the important sentence storage part 24 shown in FIG. The flowchart which shows the process sequence of the document classification | category process of the important sentence extraction apparatus 40 concerning this embodiment. The figure which shows an example of the classification object document input by the classification object document input part 41. FIG. The figure which shows an example at the time of a classification result being output by the classification result output part 44. FIG.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[First Embodiment]
First, a first embodiment of the present invention will be described with reference to FIGS. FIG. 1 is a block diagram showing a hardware configuration of the important sentence extracting apparatus according to the present embodiment. As shown in FIG. 1, the computer 10 is connected to an external storage device 20 such as a hard disk drive (HDD). The external storage device 20 stores a program 21 executed by the computer 10. The computer 10 and the external storage device 20 constitute an important sentence extraction device 30.

  In the present embodiment, the important sentence extraction device 30 has a function of extracting, from the document, an important sentence that appropriately represents the content of a document composed of a plurality of sentences, for example.

  FIG. 2 is a block diagram mainly showing a functional configuration of the important sentence extraction device 30 shown in FIG. As illustrated in FIG. 2, the important sentence extraction device 30 includes an extraction target document input unit 31, an important word determination unit 32, an important sentence extraction unit 33, a score calculation unit 34, and an important sentence output unit 35. In the present embodiment, these units 31 to 35 are realized by the computer 10 shown in FIG. 1 executing the program (important sentence extraction program) 21 stored in the external storage device 20. This program 21 can be stored in advance in a computer-readable storage medium and distributed. The program 21 may be downloaded to the computer 10 via, for example, a network.

  The important sentence extraction device 30 includes a hierarchical important word storage unit (hierarchical important word dictionary) 22, a synonym dictionary 23, and an important sentence storage unit 24. In the present embodiment, the hierarchical important word storage unit 22, the synonym dictionary 23, and the important sentence storage unit 24 are stored in, for example, the external storage device 20.

  The hierarchical important word storage unit 22 stores (stores) important words corresponding to each of categories into which a document including a plurality of words (character strings) is classified in association with the category. This important word is determined in advance as a word that appropriately represents the contents of the corresponding category, for example. Further, in the hierarchical important word storage unit 22, the relationship between the categories (corresponding important words) into which the document is classified is expressed in a hierarchical structure. That is, in the hierarchical important word storage unit 22, the categories related to the categories into which the documents are classified are represented in a hierarchical structure.

  If the document to be classified is, for example, a patent document (patent document), FI (File Index assigned by the Japan Patent Office) is used as the category in the hierarchical keyword storage unit 22 and corresponds to the category. As an important word, an explanatory text corresponding to the FI (defined by the Japan Patent Office, hereinafter referred to as “FI explanatory text”) is used.

  Note that the contents of the hierarchical important word storage unit 22 are constructed in advance by hand, for example. Further, for example, the hierarchical important word storage unit 22 is created by statistically determining the important words corresponding to the category based on the appearance frequency of words (character strings) included in the document classified into the category. It does not matter.

  The extraction target document input unit 31 inputs a document from which an important sentence is extracted (hereinafter referred to as an extraction target document). The extraction target document input unit 31 inputs an extraction target document in accordance with a user operation. This extraction target document is composed of a plurality of sentences including a plurality of words. The extraction target document is given a category in which the extraction target document is classified.

  As described above, the extraction target document input by the extraction target document input unit 31 includes, for example, a patent document. A patent document is an example of an extraction target document, and the extraction target document may be text data.

  In the synonym dictionary 23, for example, synonyms of important words corresponding to each of the categories stored in the hierarchical important word storage unit 22 are stored. In the synonym dictionary 23, for example, when an important word “storage medium” is stored in the hierarchical important word storage unit 22, “recording device” and “recording member” are synonymous with the important word “storage medium”. Is stored.

  The important word determination unit 32 reads the important word (first important word) corresponding to the category assigned to the extraction target document input by the extraction target document input unit 31 from the hierarchical important word storage unit 22. Hereinafter, an important word corresponding to a category assigned to an extraction target document input by the extraction target document input unit 31 is referred to as a corresponding important word.

  The important word determination unit 32 determines whether or not the read corresponding important word is included in the extraction target document input by the extraction target document input unit 31. When it is determined that the corresponding important word is included in the extraction target document, the important word determination unit 32 determines the corresponding important word as an extraction important word (hereinafter referred to as a first extraction important word). .

  The important word determination unit 32 includes the corresponding important word in the extraction target document when the synonym of the corresponding important word stored in the synonym dictionary 23 is included in the extraction target document. It is determined that In this case, in the subsequent processing, synonyms of the corresponding important words included in the extraction target document are handled in the same manner as the corresponding important words.

  The important word determination unit 32 has an important word corresponding to a category associated with the category (the category corresponding to the read corresponding important word) assigned to the extraction target document input by the extraction target document input unit 31. 2 key words) is read from the hierarchical key word storage unit 22. Hereinafter, an important word corresponding to a category related to a category assigned to the extraction target document input by the extraction target document input unit 31 is referred to as a neighboring important word (a neighboring important word of the corresponding important word).

  Here, the neighboring important words include, for example, first to third important words. The first important word is an important word corresponding to a category located (positioned) in the same hierarchy as the category assigned to the extraction target document in the hierarchical important word storage unit 22. The second important word is an important word corresponding to a category positioned at the upper or lower level of the category assigned to the extraction target document in the hierarchical important word storage unit 22. Further, the third important word is an important word corresponding to a category located in the upper or lower order of the category assigned to the extraction target document in the hierarchical important word storage unit 22. The neighboring important words may be, for example, only the first and second important words among the first to third important words described above, or important words other than the first to third important words. May be included.

  The important word determination unit 32 determines whether or not the read nearby important word is included in the extraction target document input by the extraction target document input unit 31. When it is determined that the neighboring important word is included in the extraction target document, the important word determining unit 32 determines the neighboring important word as an extracting important word (hereinafter referred to as a second extracting important word). .

  In addition, the important word determination unit 32 includes the neighboring important word in the extraction target document when the synonym of the neighboring important word stored in the synonym dictionary 23 is included in the extraction target document. It is determined that In this case, in the subsequent processing, synonyms of the neighborhood important words included in the extraction target document are handled in the same manner as the neighborhood important words.

  The important sentence extraction unit 33 is input by the extraction target document input unit 31 with a sentence including the extraction important word (first and second extraction important words) determined by the important word determination unit 32 as an important sentence. Extract from the extraction target document. The important sentence extraction unit 33 extracts a sentence including at least one of the extraction important words determined by the important word determination unit 32 as an important sentence. Details of the important sentence extraction processing by the important sentence extraction unit 33 will be described later.

  The score calculation unit 34 calculates the score (importance) of each important sentence extracted by the important sentence extraction unit 33 based on the extraction important words included in the important sentence. The score calculation unit 34 determines whether the important word included in the important sentence extracted by the important sentence extraction unit 33 is the first extraction important word (corresponding important word) or the second extraction important word (neighboring important word). ) To calculate a score. Details of the important sentence score calculation processing by the score calculation unit 34 will be described later.

  The score calculation unit 34 stores the important sentence extracted by the important sentence extraction unit 33 and the calculated score of the important sentence in the important sentence storage unit 24 in association with each other. At this time, the score calculation unit 34 associates the combination of the important sentence and the score of the important sentence with the category assigned to the extraction target document (that is, the document from which the important sentence is extracted). 24.

  The important sentence output unit 35 outputs the important sentences extracted by the important sentence extraction unit 33 in the order of the scores calculated by the score calculation unit 34. Specifically, the important sentence output unit 35 presents (displays) the important sentences to the user in the order of scores.

  FIG. 3 shows an example of the data structure of the hierarchical keyword storage unit 22 shown in FIG. As shown in FIG. 3, the hierarchical key word storage unit 22 shows a hierarchical structure of relationships between categories into which documents are classified. The hierarchical important word storage unit 22 stores important words (important words related to the category) corresponding to each category into which the document is classified.

  In the example illustrated in FIG. 3, a patent document (patent document) is assumed as a document from which an important sentence is extracted by the important sentence extraction device 30. That is, in the hierarchical important word storage unit 22, FI is used as a category in which documents are classified, and FI explanatory text is used as an important word corresponding to the category. This FI is, for example, a number used as a search key for patent documents.

  Specifically, referring to FIG. 3, the hierarchical key word storage unit 22 stores the FI explanation “inkjet” as a key word corresponding to the FI “B41J 3/04 101” corresponding to the category into which the document is classified, for example. Is stored.

  The hierarchical key word storage unit 22 stores the FI explanation “color” as the key word corresponding to the FI “B41J 3/04 101A”.

  The hierarchical important word storage unit 22 stores the FI explanation “ink mist” as an important word corresponding to the FI “B41J 3/04 101B”.

  The hierarchical key word storage unit 22 stores the FI explanation “storage medium” as the key word corresponding to the FI “B41J 3/04 101Y”.

  In the hierarchical important word storage unit 22, the FI explanation “ink supply device” is stored as an important word corresponding to the FI “B41J 3/04 102”.

  In the hierarchical important word storage unit 22, the FI explanation “head cleaning” is stored as an important word corresponding to the FI “B41J 3/04 102H”.

  As shown in FIG. 3, in the hierarchical important word storage unit 22, each FI is expressed in a hierarchical structure. In the example illustrated in FIG. 3, the FIs “B41J 3/04 101A”, “B41J 3/04 101B”, and “B41J 3/04 101Y” are positioned in the lower hierarchy of the FI “B41J 3/04 101”. . That is, the FIs “B41J 3/04 101A”, “B41J 3/04 101B”, and “B41J 3/04 101Y” are positioned in the same hierarchy.

  The FI “B41J 3/04 102” is positioned in the same hierarchy as the FI “B41J 3/04 101”.

  Furthermore, FI “B41J 3/04 102H” is positioned in a lower hierarchy of FI “B41J 3/04 102”.

  In this way, in the hierarchical important word storage unit 22, the categories (FI) represent FIs that are related to each other (upper, lower, or positioned in the same hierarchy) in a hierarchical structure.

  Next, with reference to the flowchart of FIG. 4, the process sequence of the important sentence extraction process of the important sentence extraction apparatus 30 which concerns on this embodiment is demonstrated. Note that the hierarchical important word storage unit 22 included in the important sentence extraction device 30 has the data structure shown in FIG.

  First, the extraction target document input unit 31 inputs a document (extraction target document) composed of a plurality of sentences including a plurality of words in accordance with a user operation (step S1). For example, the user can designate an extraction target document by operating the important sentence extraction device 30. The extraction target document input unit 31 inputs an extraction target document to which a category (FI) to which the extraction target document is classified is assigned.

  Here, FIG. 5 shows an example of the extraction target document input by the extraction target document input unit 31. In the example shown in FIG. 5, the extraction target document 100 is a patent document. The category assigned to the extraction target document 100 is FI “B41J 3/04 101B”. In the following description, it is assumed that the extraction target document 100 shown in FIG.

  Next, the important word determination unit 32 acquires the extraction target document 100 input by the extraction target document input unit 31. The important word determination unit 32 reads an important word (corresponding important word) corresponding to the category assigned to the acquired extraction target document 100 from the hierarchical important word storage unit 22 (step S2).

  Here, the process of step S2 will be specifically described with reference to FIGS. 3 and 5 described above. The category assigned to the extraction target document 100 shown in FIG. 5 is FI “B41J 3/04 101B”. Therefore, according to the hierarchical important word storage unit 22 shown in FIG. 3, the important word determination unit 32 reads “ink mist” from the hierarchical important word storage unit 22 as the corresponding important word.

  The keyword determining unit 32 determines whether or not the read corresponding keyword is present (included) in the acquired extraction target document 100 (step S3). Since the extraction target document 100 shown in FIG. 5 includes “ink mist” that is the corresponding important word, the important word determination unit 32 determines that the corresponding important word exists in the extraction target document 100 (in step S3). YES)

  When it is determined that the corresponding important word is present in the extraction target document 100, the important word determination unit 32 selects the corresponding important word (here, “ink mist”) as the extraction important word (first extraction important word). ) (Step S4).

  Next, the important word determination unit 32 reads the important words (neighboring important words) corresponding to the categories associated with the categories assigned to the acquired extraction target document 100 from the hierarchical important word storage unit 22 (step) S5).

  Here, it is assumed that the first to third important words described above are included in the neighboring important words. As described above, the first important word is an important word corresponding to a category positioned in the same hierarchy as the category assigned to the extraction target document 100. The second important word is an important word corresponding to a category positioned at the upper level (hierarchy) or lower (hierarchy) of the category assigned to the extraction target document 100. In addition, the third important word is an important word corresponding to a category positioned at a higher level (hierarchy) or a lower lower level (hierarchy) of the category assigned to the extraction target document 100.

  In this case, specifically, using FIG. 3 described above, the hierarchical keyword storage unit 22 is positioned in the same hierarchy as the FI “B41J 3/04 101B” that is a category assigned to the extraction target document 100. As categories, FI “B41J 3/04 101A” and “B41J 3/04 101Y” exist. Therefore, the important word determination unit 32 stores “color” and “storage medium” corresponding to the FI “B41J 3/04 101A” and “B41J 3/04 101Y” as hierarchical important words as neighboring important words of the corresponding important words. Read from the unit 22. The “color” and “storage medium” are neighboring important words corresponding to the first important word described above.

  Further, in the hierarchy important word storage unit 22, FI “B41J 3/04 101” is set as a category positioned in the upper hierarchy of FI “B41J 3/04 101B” which is a category assigned to the extraction target document 100. Exists. Therefore, the important word determination unit 32 reads “inkjet” corresponding to the FI “B41J 3/04 101” from the hierarchical important word storage unit 22 as a neighboring important word of the corresponding important word. This “inkjet” is a neighboring important word corresponding to the second important word described above.

  The important word determination unit 32 determines whether or not the read corresponding important word exists (is included) in the acquired extraction target document 100 (step S6). Since the extraction target document 100 shown in FIG. 5 includes “color” and “inkjet” which are neighboring important words, the important word determination unit 32 determines that the neighboring important words are present in the extraction target document 100 ( YES in step S6).

  Note that the extraction target document 100 shown in FIG. 5 does not include the “storage medium” that is a nearby important word. However, assuming that “recording device” and “recording member” are stored in the synonym dictionary 23 as synonyms of the nearby important word “storage medium” as described above, the important word determination unit 32 performs the “recording”. It is determined whether or not “device” and “recording member” are included in the extraction target document 100. In this case, since “recording device” and “recording member”, which are synonyms of the neighborhood important word “storage medium”, are included in the extraction target document 100, it is determined that the neighborhood important word exists in the extraction target document 100. Is done. In this case, the “recording device” and “recording member” existing in the extraction target document 100 are handled in the same manner as the “storage medium” that is a neighboring important word in the subsequent processing.

  When it is determined that the neighboring important word is present in the extraction target document 100, the important word determination unit 32 extracts the neighboring important word that is determined to be present in the extraction target document 100 as the extraction important word (second extraction word). Important word) (step S7). In this case, the important word determination unit 32 determines the adjacent important words “color”, “recording apparatus”, “recording member”, and “inkjet” as extraction important words. The “storage medium” is a nearby important word, but is not determined as an extraction important word because it is not included in the extraction target document 100.

  Next, the important sentence extraction unit 33 extracts, as an important sentence, a sentence including the extraction important word (at least one of the first and second extraction important words) determined by the important word determination unit 32. Extracted from the document 100 (step S8). The important sentence extraction unit 33 performs syntax analysis of the extraction target document 100 as necessary in the important sentence extraction process.

  Here, the important sentence extraction processing by the important sentence extraction unit 33 will be described in detail. The important sentence extraction unit 33 extracts a portion (part) corresponding to an extraction condition (extraction rule) described below in the extraction target document 100 as an important sentence.

  The extraction conditions described above include the first and second extraction conditions. According to the first extraction condition, if there is a part that starts with an extraction key word and includes one verb related to the key word for extraction, the part from the key word for extraction to the verb is the key sentence. To do. According to the second extraction condition, if there is one verb immediately before the extraction important word, the important words from the verb related to the extraction important word to the extraction important word are important. A sentence.

  As described above, when the extraction important words determined by the important word determination unit 33 are “ink mist”, “color”, “recording device”, “recording member”, and “inkjet”, the extraction target shown in FIG. The important sentence extracted from the document 100 will be specifically described.

  First, the case where the key word for extraction is “ink mist” will be described. From the extraction target document 100 shown in FIG. 5, “Prevent failure due to ink mist” is extracted as an important sentence according to the first extraction condition described above. . In the extraction target document 100, “Prevents defects due to ink mist” is stated. However, in an important sentence extracted according to the first extraction condition, the verb at the end of the important sentence is corrected to an end form.

  The case where the key word for extraction is “color” will be described. The key sentence is “color filter capable of performing preliminary ink ejection” from the extraction target document 100 shown in FIG. Extracted as Further, “provide color filter manufacturing apparatus” is extracted from the extraction target document 100 as an important sentence according to the first extraction condition described above.

  The case where the key word for extraction is “recording device” will be described. “Inkjet recording device for recording” is extracted as an important sentence from the extraction target document 100 shown in FIG. 5 according to the second extraction condition described above. In the extraction target document 100, there are two descriptions of “inkjet recording apparatus for recording”, but one sentence is extracted as an important sentence.

  The case where the key word for extraction is “recording member” will be described. It is important that “ink is ejected toward the recording member by the inkjet head” from the extraction target document 100 shown in FIG. 5 according to the first extraction condition described above. Extracted as a sentence. The important sentences extracted here include “inkjet” which is an important word for extraction. In this case, “toward” is also a verb, but by performing syntax analysis, the verb related to “inkjet”, which is the key word for extraction, is “discharge”, and thus the above important sentence is extracted.

  The case where the key word for extraction is “inkjet” will be described. According to the first extraction condition described above from the extraction target document 100 shown in FIG. 5, “ink is ejected toward the recording member by the inkjet head” and second The “ink jet recording apparatus that performs recording” is extracted as an important sentence according to the extraction conditions. Further, according to the first extraction condition described above from the extraction target document 100, “cover the nozzle surface of the inkjet head 120a”, “relatively move the inkjet head 120a and the cap member 31a”, and “ink from the inkjet head 120a”. “Discharge” is extracted as an important sentence.

  As described above, the important sentence extraction unit 33, from the extraction target document 100, “discharges ink toward the recording member by the inkjet head”, “inkjet recording apparatus that performs recording”, “prevents defects due to ink mist”, “Providing a color filter manufacturing device”, “Covering nozzle surface of inkjet head 120a”, “Moving inkjet head 120a and cap member 31a relative to each other” And “to eject ink from the inkjet head 120a” are extracted as important sentences.

  Next, the score calculation part 34 calculates the score for every important sentence extracted by the important sentence extraction part 33 (step S9). For each important sentence extracted by the important sentence extraction unit 33, the score calculation unit 34 determines a score predetermined for the extraction important word included in the important sentence (hereinafter simply referred to as the extraction important word score). The score of the important sentence is calculated based on the notation.

  The score of the keyword for extraction is determined based on the positional relationship in the hierarchical structure of the hierarchical keyword storage unit 22 between the category corresponding to the keyword for extraction and the category assigned to the extraction target document 100. For example, among the extraction important words, the first extraction important word (corresponding important word) has a high score, and the second extraction important word (neighboring important word) has a positional relationship in the hierarchical structure ( That is, the score is determined in accordance with the distance to the category assigned to the extraction target document 100.

  In the following description, the category assigned to the extraction target document 100 is the target category, the category positioned in the upper hierarchy of the target category is positioned in the upper hierarchy category, and the lower hierarchy of the target category. A category that is positioned in the same hierarchy as the target category is called a lower hierarchy category.

  Here, it is assumed that a score “6” is defined for the first extraction important word (corresponding important word) corresponding to the target category. Of the second extraction important words (neighboring important words), the second extraction important word corresponding to the upper hierarchy category and the lower hierarchy category has a score of “3”, and the second extraction important word corresponds to the same hierarchy category. It is assumed that a score “2” is defined for the extraction important word.

  Although not present in the hierarchical keyword storage unit 22 shown in FIG. 3, for example, the score “1” is included in the second extraction important word corresponding to the category positioned in the higher hierarchy of the upper hierarchy category. It has been established. A similar score is set for the second extraction important word corresponding to the category positioned in the lower hierarchy of the lower hierarchy category.

  The score calculation unit 34 calculates the score of the important sentence by adding the scores of the extraction important words included in the important sentence extracted by the important sentence extraction unit 33. That is, the score of the important sentence corresponds to “6 × number of corresponding important words (first extraction important words) + 3 × higher (or lower) hierarchical category” according to the above-described example of the important words for extraction. Number of neighboring important words (second extracting important words) + 2 × number of neighboring important words (second extracting important words) corresponding to the same hierarchical category (+ 1 × higher (or lower) hierarchical category ( Or the number of neighboring important words (second extraction important words) corresponding to the category positioned in the lower hierarchy)).

  Here, the score calculation process of each important sentence extracted by the above-described important sentence extraction unit 33 will be specifically described. Note that “ink mist”, which is an extraction important word determined by the important word determination unit 32, is a corresponding important word (first extraction important word), and is “color”, “recording device”, and “recording member”. Is a neighboring important word (second extracting important word) corresponding to the same hierarchical category, and “inkjet” is a neighboring important word (second extracting important word) corresponding to the upper hierarchical category.

  The important sentences extracted by the important sentence extraction unit 33 are, as described above, “injecting ink toward the recording member by the ink jet head”, “ink jet recording apparatus that performs recording”, and “defect caused by ink mist”. "Prevent", "Color filter capable of preliminary ink ejection", "Provide manufacturing apparatus for color filter", "Cover nozzle surface of inkjet head 120a", "Inkjet head 120a and cap member 31a relative to each other" And “move ink from the inkjet head 120a”.

  When the important sentence is “Ejecting ink toward the recording member by the ink jet head”, the important sentence corresponds to “inkjet” which is a neighboring important word corresponding to the upper hierarchy category and the same hierarchy category. One “recording member” which is a neighboring important word is included. Therefore, the score of this important sentence is 3 × 1 + 2 × 1 = 5 by the above-described score formula.

  When the important sentence is “an inkjet recording apparatus that performs recording”, the important sentence includes “inkjet” that is a neighboring important word corresponding to the upper hierarchical category and “neighboring important words that correspond to the same hierarchical category” One “recording device” is included. Therefore, the score of this important sentence is 3 × 1 + 2 × 1 = 5 by the above-described score formula.

  When the important sentence is “Preventing failure due to ink mist”, the important sentence contains one “ink mist” which is a corresponding important word. Therefore, the score of this important sentence is 6 × 1 = 6 by the above-described score formula.

  When the important sentence is “a color filter capable of performing preliminary ink ejection”, the important sentence includes one “color” which is a neighboring important word corresponding to the same hierarchical category. Therefore, the score of this important sentence is 2 × 1 = 2 by the above-described score formula.

  When the important sentence is “provide a color filter manufacturing apparatus”, the important sentence includes one “color” which is a neighboring important word corresponding to the same hierarchical category. Therefore, the score of this important sentence is 2 × 1 = 2 by the above-described score formula.

  When the important sentence is “covers the nozzle surface of the inkjet head 120a”, the important sentence includes one “inkjet” which is a neighboring important word corresponding to the upper hierarchical category. Therefore, the score of this important sentence is 3 × 1 = 3 by the above-described score formula.

  When the important sentence is “relatively move the inkjet head 120a and the cap member 31a”, the important sentence includes one “inkjet” which is a neighboring important word corresponding to the upper hierarchy category. ing. Therefore, the score of this important sentence is 3 × 1 = 3 by the above-described score formula.

  When the important sentence is “discharge ink from the inkjet head 120a”, the important sentence includes one “inkjet” that is a neighboring important word corresponding to the upper hierarchical category. Therefore, the score of this important sentence is 3 × 1 = 3 by the above-described score formula.

  Even when a single important sentence includes a plurality of the same important words (corresponding important words or neighboring important words), the number is assumed to be one in the score calculation process.

  When the score for each important sentence is calculated as described above, the score calculation unit 34 includes the category (here, FI “B41J 3/04 101B”) assigned to the extraction target document 100, and the important sentence extraction unit 33. The important sentence extracted by the above and the score of the important sentence are stored in the important sentence storage unit 24 in association with each other.

  That is, the important sentence storage unit 24 stores a combination of an important sentence extracted from a document classified into the category by category and a score of the important sentence each time the above-described important sentence extraction process is performed. Is done.

  Next, the important sentence output unit 35 outputs the important sentences extracted by the important sentence extraction unit 33 in the order of the scores calculated by the score calculation unit 34 (for example, display to the user) (step S10). The important sentence output unit 35 outputs the important sentences extracted by the important sentence extraction unit 33 in, for example, descending order of the score values of the important sentences.

  Further, the important sentence output unit 35 may output only the important sentence having a score of 5 or more, for example, without outputting all the important sentences extracted by the important sentence extracting unit 33.

  FIG. 6 shows an example when an important sentence with a score of 5 or more is output (displayed). In the example shown in FIG. 6, among the important sentences extracted by the important sentence extraction unit 33, the important sentence “Prevents defects due to ink mist (score 6)” and “Eject ink toward the recording member by the inkjet head. “Score 5” ”and“ Inkjet recording apparatus for recording (score 5) ”are output in descending order of score.

  On the other hand, when it is determined in step S3 described above that the corresponding important word does not exist in the extraction target document 100, the process of step S5 is executed. That is, when the corresponding important word does not exist in the extraction target document 100, the corresponding important word is not determined as the extraction important word.

  Further, when it is determined in step S6 described above that there is no neighboring important word in the extraction target document 100, the processing is terminated. Even when it is determined in step S6 that the neighboring important word does not exist in the extraction target document 100, if it is determined in step S3 that the corresponding important word exists in the extraction target document 100, The important sentence including the corresponding important word is extracted from the extraction target document 100, and the important sentence is output. That is, when all of the corresponding important words and the neighboring important words are not present in the extraction target document 100, the process is terminated without outputting the important sentence.

  As described above, in the present embodiment, important words (corresponding important words) corresponding to the category assigned to the extraction target document 100 and categories related to the category (upper hierarchy category, lower hierarchy category, same hierarchy category, etc.) ) Is determined as an extraction important word, and a sentence including the extraction important word is determined as an important sentence from the extraction target document 100. Extract.

  Thereby, in this embodiment, even if the corresponding important word corresponding to the category assigned to the extraction target document 100 does not exist in the extraction target document 100, the relation to the category corresponding to the corresponding important word is related. An important sentence can be extracted using a neighboring important word corresponding to a certain category.

  In the present embodiment, as described above, the use of neighboring important words makes it possible to extract important sentences compared to the case where important words or the like of a category not related to the category into which the extraction target document 100 is classified are used. And more important sentences can be extracted as compared with the case where important sentences are extracted using only corresponding important words.

  In the present embodiment, it is possible to extract an important sentence corresponding to the granularity of the description of the content of the extraction target document 100. In other words, in the present embodiment, when the description granularity of the content of the extraction target document 100 is coarse, the neighboring important words corresponding to the upper hierarchy category, while when the description granularity is fine, the neighboring important words corresponding to the lower hierarchy category. By using, it is possible to extract an important sentence according to the contents of the extraction target document 100. Further, in the present embodiment, since the scores differ depending on the neighboring important words corresponding to the upper and lower hierarchical categories or the neighboring important words corresponding to the same hierarchical category, the importance score can be expressed in detail.

  Although the present embodiment has been described on the assumption that an important sentence is extracted from a patent document using FI and FI explanation, this embodiment can also be applied to, for example, a mail group stored in a mail server. is there. In this case, if the mail group is stored in a directory structure and a keyword is assigned to each directory, an important sentence can be extracted from the mail group in each directory using the assigned keyword. .

[Second Embodiment]
Next, a second embodiment of the present invention will be described with reference to FIG. FIG. 7 is a block diagram mainly showing a functional configuration of the important sentence extracting apparatus according to the present embodiment. The same parts as those in FIG. 2 described above are denoted by the same reference numerals, and detailed description thereof is omitted. Here, parts different from FIG. 2 will be mainly described.

  The hardware configuration of the important sentence extraction apparatus according to the present embodiment is the same as that of the first embodiment described above, and will be described with reference to FIG. 1 as appropriate.

  In the present embodiment, in addition to the important sentence extraction process in the first embodiment described above, a process for classifying documents (document classification process) using the important sentence stored in the important sentence storage unit 24 is performed. However, this is different from the first embodiment described above. In other words, in this embodiment, the important sentence extraction process in the first embodiment described above is repeated, so that the important sentence storage unit 24 accumulates a set of important sentences and score of the important sentences for each category. It is assumed that.

  As illustrated in FIG. 7, the important sentence extraction device 40 includes a classification target document input unit 41, a concept search unit 42, a category classification unit 43, and a classification result output unit 44. In the present embodiment, these units 41 to 44 are realized by the computer 10 shown in FIG. 1 executing the program 21 stored in the external storage device 20.

  The important sentence extraction device 40 includes a classification result storage unit 25. In the present embodiment, the classification result storage unit 25 is stored in, for example, the external storage device 20.

  The classification target document input unit 41 inputs a document to be classified (hereinafter referred to as a classification target document). The classification target document input unit 41 inputs a classification target document in accordance with a user operation. This classification target document is composed of a plurality of sentences including a plurality of words. Unlike the extraction target document in the first embodiment described above, no category is assigned to this classification target document.

  The classification target document input by the classification target document input unit 41 includes, for example, a patent document (patent document). A patent document is an example of a classification target document, and the classification target document may be text data.

  The concept search unit 42 performs a concept search (natural sentence search) by using the important sentence stored in the important sentence storage unit 24 as a search key and the classification target document input by the classification target document input unit 41 as a search target. Thus, it is determined whether or not the classification target document input by the classification target document input unit 41 matches the important sentence stored in the important sentence storage unit 24. The concept search unit 42 performs a concept search for all the important sentences stored in the important sentence storage unit 24.

  Based on the concept search result, the category classification unit 43 matches the classification target document (search target) input by the classification target document input unit 41 with the important sentence (search key) stored in the important sentence storage unit 24. It is determined whether or not.

  The category classification unit 43 is an important sentence associated with a category stored in the important sentence storage unit 24, and is an important sentence determined to match the classification target document input by the classification target document input unit 41. Is calculated for each category stored in the important sentence storage unit 24 as the score (confidence) of the category. That is, the category classification unit 43 calculates a score for each category stored in the important sentence storage unit 24 based on the concept search results for all the important sentences stored in the important sentence storage unit 24.

  The classification result output unit 44 outputs the category stored in the important sentence storage unit 24 and the calculated score of the category. At this time, the classification result output unit 44 outputs each of the categories stored in the important sentence storage unit 24 in the order of the score of the category calculated by the category classification unit 43.

  The classification result output unit 44 classifies the classification target document input by the classification target document input unit 41 into categories according to the user's operation based on the output category and the score of the category, and the classification result is displayed. Stored in the classification result storage unit 25.

  That is, the user can specify a category to which the classification target document input by the classification target document input unit 41 should be classified with reference to the category output by the classification result output unit 44 and the score of the category. . That is, the classification target document is classified into a category designated by the user among the categories output by the classification result output unit 44.

  FIG. 8 shows an example of the data structure of the important sentence storage unit 24 shown in FIG. As shown in FIG. 8, the important sentence storage unit 24 stores category names, important sentences, and score values in association with each other. The important sentence indicates an important sentence extracted from a document classified in the associated category. The score value indicates the score of the associated important sentence.

  The important sentence storage unit 24 stores a plurality of sets of important sentences and scores for each category.

  In the example illustrated in FIG. 8, the important sentence storage unit 24 stores a set of the important sentence 1 “print color image” and the score value “6” in association with the category name “B41J 3/04 101A”. Yes. The important sentence storage unit 24 stores a set of an important sentence 2 “execute color printing method” and a score value “6” in association with the category name “B41J 3/04 101A”.

  The important sentence storage unit 24 stores a set of an important sentence 1 “Ink mist occurs” and a score value “6” in association with the category name “B41J 3/04 101B”. Further, the important sentence storage unit 24 stores a set of an important sentence 2 “having an inkjet recording apparatus” and a score value “5” in association with the category name “B41J 3/04 101B”.

  Next, a processing procedure of document classification processing of the important sentence extraction device 40 according to the present embodiment will be described with reference to the flowchart of FIG. In the important sentence storage unit 24, the important sentences extracted by the process described in the first embodiment and the scores of the important sentences are accumulated for each category.

  First, the classification target document input unit 41 inputs a document (classification target document) composed of a plurality of sentences including a plurality of words in accordance with a user operation (step S21). For example, the user can designate a classification target document by operating the important sentence extraction device 40. The classification target document input by the classification target document input unit 41 is not given a category.

  Here, FIG. 10 shows an example of the classification target document input by the classification target document input unit 41. In the example shown in FIG. 10, the classification target document 200 is a patent document. Further, unlike the extraction target document 100 shown in FIG. 5 described above, no category is assigned to the classification target document 200. In the following description, it is assumed that the classification target document 200 shown in FIG.

  Next, the concept search unit 42 acquires one important sentence stored in the important sentence storage unit 24 (step S22).

  The concept search unit 42 performs a concept search (natural sentence search) on the classification target document 200 input by the classification target document input unit 41 using the acquired important sentence (step S23). In this case, the concept search unit 42 performs a concept search using an important sentence as a search key and the classification target document 200 as a search target. The concept search is a search method for searching for information having contents close to the sentence using, for example, a sentence having a content to be searched as a search key.

  The category classification unit 43 determines whether or not the classification target document 200 input by the classification target document input unit 41 matches the acquired important sentence based on the concept search result by the concept search unit 42 (step S24). ). The search result in the concept search is represented by the degree (%) that the search target (classification target document 200) matches the search key (important sentence). The category classification unit 43 determines that the classification target document 200 matches the important sentence when the degree of matching of the search target in the concept search result is a certain value (for example, 70%) or more.

  When it is determined that the classification target document 200 matches the important sentence (YES in step S24), the category classification unit 43 stores the important sentence in association with the important sentence (important sentence used as the search key in step S23). The score (score of the important sentence) stored in the unit 24 is acquired from the important sentence storage unit 24.

  The category classification unit 43 adds the score of the acquired important sentence to the score of the category associated with the important sentence acquired by the concept search unit 42 in the important sentence storage unit 24 (step S25).

  On the other hand, when it is determined that the classification target document 200 does not match the important sentence (NO in step S24), a process in step S26 described later is executed.

  Note that the processes of steps S22 to S25 described above are executed for all important sentences stored in the important sentence storage unit 24. Accordingly, the category classification unit 43 is associated with an important sentence that is associated with a category stored in the important sentence storage unit 24 and is determined to match the classification target document 200. The total value of the scores (important sentence scores) is calculated as the score of the category. Moreover, the category classification | category part 43 scores about all the categories stored in the important sentence storage part 24 (score according to category) by performing the process of above-mentioned step S22-step S25 about all the important sentences. Is calculated.

  The category score calculation process will be specifically described with reference to FIG. For example, the classification target document 200 matches the important sentence 1 “print color image” and the important sentence 2 “execute color printing method” associated with the category name “B41J 3/04 101A” shown in FIG. Assume a case. In this case, the score of category name “B41J 3/04 101A” is score value “6” associated with important sentence 1 “print color image” and important sentence 2 “execute color printing method”. The sum of the associated score values “6”, that is, “12”.

  Further, for example, the classification target document 200 does not match the important sentence 1 “ink mist is generated” associated with the category name “B41J 3/04 101B” shown in FIG. It is assumed that it matches “having”. In this case, the score of the category name “B41J 3/04 101B” is the score value “5” associated with the important sentence 2 “having the inkjet recording apparatus”.

  That is, when the above-described steps S22 to S25 are executed for all important sentences stored in the important sentence storage unit 24, scores of all categories are calculated.

  When the process of step S25 described above is executed, it is determined whether or not the processes of steps S22 to S25 described above have been executed for all the important sentences stored in the important sentence storage unit 24 (step S26). . If it is determined that the processes in steps S22 to S25 are not executed for all important sentences, the process returns to step S22 and is repeated.

  On the other hand, when it is determined that the processing of steps S22 to S25 has been executed for all important sentences (YES in step S26), that is, when the scores for each category stored in the important sentence storage unit 24 are calculated. The classification result output unit 44 outputs the category and the score of the category. At this time, the classification result output unit 44 outputs the categories stored in the important sentence storage unit 24 (that is, category candidates into which the classification target document is classified) in the order of score for each category (step S27).

  Here, FIG. 11 shows an example when category candidates are output (displayed) by the classification result output unit 44. In the example illustrated in FIG. 11, for example, category names “B41J 3/04 101A” and “B41J 3/04 101A” are output in the order of the scores together with the score (confidence) of the category. That is, the classification result output unit 44 outputs a plurality of category candidates to which the classification target document 200 should be classified in the order of scores (confidence levels).

  When category candidates are output as shown in FIG. 11 described above, the user may designate a category into which the classification target document 200 should be classified with reference to the category name (category candidate) and the score of the category. it can. That is, even if the score of the category is low, if the user thinks that the category is appropriate, the classification target document 200 can be classified into the category.

  The classification result output unit 44 stores the category into which the classification target document 200 is classified in the classification result storage unit 25 as a classification result in accordance with the above-described user operation (designation) (step S28).

  As described above, in the present embodiment, when a classification target document to which no category is assigned is input, the classification is performed using the important sentence for each category stored in the important sentence storage unit 24 and the score of the important sentence. The total score of important sentences that match the target document is calculated as the score (confidence) of the category. Thereby, in this embodiment, it becomes possible to classify | categorize the said classification target document into the category which should classify a classification target document based on the score of the calculated category. At this time, by outputting a score for each category, the user can specify a category in which the classification target document should be classified by referring not only to the category having the highest score but also to the second and lower categories. .

  In the present embodiment, the calculated score for each category is output, and the classification target document is classified into the category specified by the user referring to the score for each category. A configuration may be adopted in which the classification target documents are automatically classified based on the score. In this case, the above-described category classification unit 43 classifies the classification target document into, for example, a category having the highest score based on the calculated category score (confidence).

  Further, in the present embodiment, it has been described that when a document to be classified is input, the important sentence storage unit 24 already stores the important sentence and the score of the important sentence for each category. When a sufficient amount of important sentences and the score of the important sentences are not accumulated in the storage unit 24, for example, the diagram in the first embodiment described above for a large number of documents accumulated in the document DB or the like. After the important sentence storage unit 24 is constructed by repeatedly executing the process shown in FIG. 4, the classification process for the classification target document may be performed.

  Note that the present invention is not limited to the above-described embodiments as they are, and can be embodied by modifying constituent elements without departing from the scope of the invention in the implementation stage. Further, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the above embodiments. For example, some components may be deleted from all the components shown in each embodiment. Furthermore, you may combine the component covering different embodiment suitably.

  DESCRIPTION OF SYMBOLS 10 ... Computer, 20 ... External storage device, 22 ... Hierarchical important word storage part (important word storage means), 23 ... Synonym dictionary, 24 ... Important sentence storage part, 25 ... Classification result storage part, 30, 40 ... Important sentence Extraction device, 31 ... extraction target document input unit, 32 ... important word determination unit, 33 ... important sentence extraction unit, 34 ... score calculation unit, 35 ... important sentence output unit, 41 ... classification target document input unit, 42 ... concept search Part, 43 ... category classification part, 44 ... classification result output part.

Claims (5)

An important word storage means for storing important words corresponding to each of the categories into which the document is classified, and an external storage device having important word storage means in which categories related to the categories are represented in a hierarchical structure; In an important sentence extraction device composed of a computer using the external storage device, an important sentence extraction program executed by the computer,
In the computer,
Inputting an extraction target document composed of a sentence including a plurality of words according to a user's operation, to which a category to which the extraction target document is classified is assigned;
Reading a first important word corresponding to a category assigned to the input extraction target document from the important word storage means;
Determining whether the read first important word is included in the input extraction target document;
When it is determined that the first important word is included in the extraction target document, the first important word is determined as a first extraction important word;
Reading a second important word corresponding to a category associated with a category assigned to the input extraction target document from the important word storage means;
Determining whether the read second important word is included in the input extraction target document;
When it is determined that the second important word is included in the extraction target document, the second important word is determined as a second extracting important word;
Extracting from the input extraction target document a sentence including at least one of the determined first and second extraction important words as an important sentence;
Calculating the importance of the important sentence based on the determined first extraction important word or second extracted important word included in the extracted important sentence;
An important sentence extraction program for executing the step of outputting the extracted important sentences in the order of the calculated importance. In the computer,
Storing the category assigned to the input extraction target document, the extracted important sentence and the calculated importance of the important sentence in association with each other in an important sentence storage unit;
Inputting a classification target document composed of sentences including a plurality of words according to a user operation;
Determining whether the input classification target document matches an important sentence stored in the important sentence storage means;
An important sentence associated with a category stored in the important sentence storage means, and a total value of the importance levels of the important sentences associated with the important sentence determined to match the classification target document For each category stored in the important sentence storage means as the certainty of the category,
Classifying the input classification target document based on the certainty of the calculated category;
The important sentence extraction program according to claim 1, further comprising the step of storing the classification result in a classification result storage means. The step of classifying the classification target document includes:
Outputting the category stored in the important sentence storage means and the certainty factor of the calculated category;
3. The important sentence according to claim 2, further comprising: classifying the input classification target document into the category according to the user's operation based on the output category and the certainty factor of the category. Extraction program. The important word corresponding to each of the categories stored in the important word storage means includes a positional relationship in a hierarchical structure between the category corresponding to the important word and the category assigned to the input extraction target document. Based on your score,
In the step of calculating the importance of the important sentence, a predetermined score is added to the determined first extraction important word or the second extraction important word included in the important sentence. The important sentence extraction program according to claim 1, wherein the importance degree of the important sentence is calculated. Important word storage means for storing important words corresponding to each of categories into which a document is classified, and important word storage means in which categories related to the category are represented in a hierarchical structure;
Input means for inputting an extraction target document composed of a sentence including a plurality of words according to a user's operation, to which a category to which the extraction target document is classified is assigned;
First reading means for reading a first important word corresponding to a category assigned to the input extraction target document from the important word storage means;
First determination means for determining whether or not the read first important word is included in the input extraction target document;
First determination means for determining the first important word as a first extraction important word when it is determined that the first important word is included in the extraction target document;
Second reading means for reading a second important word corresponding to a category associated with a category assigned to the input extraction target document from the important word storage means;
Second determination means for determining whether the second important word read by the second reading means is included in the input extraction target document;
A second determination for determining the second important word as a second extracting important word when the second determining means determines that the second important word is included in the extraction target document; Means,
The sentence including at least one of the first extracting important word determined by the first determining means and the second extracting important word determined by the second determining means is input as the important sentence. Extracting means for extracting from the extraction target document;
Based on the first extraction important word determined by the first determination means or the second extraction important word determined by the second determination means included in the extracted important sentence, the important sentence A calculation means for calculating the importance of
An important sentence extraction apparatus comprising: output means for outputting the extracted important sentences in the order of the calculated importance.

Images