JP5085584B2 - Article feature word extraction device, article feature word extraction method, and program - Google Patents

Description

  The present invention relates to an article feature word extraction device, an article feature word extraction method, and a program for extracting feature words for a title corresponding to an input article.

  Conventionally, users obtain various information from the Internet on a network. For example, news articles are accumulated as a large amount of information on the Internet. Articles are generally classified into categories, and a relatively short document represented by a title is provided so that the contents of the article can be understood. By attaching categories and titles to the articles, the user can efficiently obtain the desired articles. However, the work of creating the title of a news article that is sent out every day is complicated. Therefore, an apparatus for assisting these operations is disclosed (for example, Patent Document 1).

JP 2000-29882 A

  For example, the device described in Patent Document 1 performs linguistic analysis on a target document and a document in the same field, and based on the result, determines an important word, an example word, and the like and creates a summary sentence. The important words shown in such a process are those having a high appearance frequency among documents in the same field. Therefore, in an article, since a specific expression such as a place name or a person name indicating the feature of the content has a low appearance frequency, the specific expression is not likely to be an important word.

  An object of the present invention is to provide an article feature word extraction device, an article feature word extraction method, and a program that extract feature words for a title corresponding to an input article based on a specific expression.

  The present inventor has found a method of extracting a specific expression as a feature word that matches a similar article title by generalizing the specific expression, and has completed the present invention. Specifically, the present invention provides the following.

(1) In an article feature word extraction device that extracts feature words for a title corresponding to an input article,
An article input means for receiving an input article with a category;
A similar article similar to the input article whose category matches the input article received by the article input unit is extracted from the past article storage unit that stores the title in association with the past article. Similar article extraction means,
Using a specific expression storage unit for storing a specific expression, specific expression extraction means for extracting the specific expression from the input article and the title of the similar article;
A specific expression generalizing means for generalizing the specific expression extracted by the specific expression extracting means and applying a generalized specific expression to the title of the input article and the similar article;
Feature word extraction means for extracting a feature word from the title of the input article and the similar article after application of the generalized specific expression by the specific expression generalization means;
Article feature word extraction device.

  According to such a configuration of the present invention, the feature word is extracted by generalizing each unique expression included in the title of a similar article that is a past article in the same category as the input article, thereby extracting the feature word. It is possible to extract a unique expression such as a place name or a person name indicating. Here, generalization refers to, for example, tagging or keyword replacement. In this way, by generalizing different unique expressions depending on articles, it is possible to grasp the same in the superordinate concept and to extract more precise similarities. In the following description of the present specification, the “title” includes a subtitle and a summary.

(2) Temporary title output means for creating and outputting a provisional title related to the input article by replacing the feature word of the title of the similar article extracted by the feature word extraction means with the feature word of the corresponding input article Comprising
The article feature word extraction device according to (1).

  According to such a configuration of the present invention, the provisional title of the input article is created by replacing the feature word of the title of the similar article with the feature word of the corresponding input article, and the created temporary title is output. Creation of input article titles can be semi-automated, and temporary titles can be output. Therefore, the efficiency of title creation by the user can be improved.

(3) The feature word extraction unit extracts, as the feature word, a word, a compound word, and a phrase that match the input article and the title of the similar article.
The article feature word extraction device according to (1) or (2).

  According to such a configuration of the present invention, words, compound words, and phrases that match the input article and the titles of similar articles are extracted, so that words, compound words, and phrases common to articles can be extracted.

(4) The specific expression storage unit stores a pattern of the specific expression extracted using machine learning.
The specific expression extraction unit extracts the specific expressions from the input article and the title of the similar article based on the specific expression pattern,
The article feature word extraction device according to any one of (1) to (3).

  According to such a configuration of the present invention, a pattern of a specific expression is stored in advance, and by using it, it can be easily determined whether or not it is a specific expression, and a specific expression can be extracted.

(5) The similar article extracting means extracts a plurality of the similar articles by changing the number of extraction targets.
The article feature word extraction device according to any one of (1) to (4).

  According to such a configuration of the present invention, the number of articles extracted as similar articles can be changed to an arbitrary number. Therefore, the feature word to be extracted can be changed by increasing or decreasing the number of similar articles and changing the number of similar articles to be compared.

(6) comprising an article output means for outputting the input article in which the feature words extracted by the feature word extraction means can be identified;
The article feature word extraction device according to any one of (1) to (5).

  According to such a configuration of the present invention, since the input article is output with the feature word being identifiable, the feature word in the input article is underlined or bolded, for example, Feature words can be output in a manner that is easy for the user to understand.

(7) The feature word extraction unit extracts, as the feature word, the generalized specific expression that matches the input article and the title of the similar article,
The article output means outputs the input article by specializing the generalized specific expression extracted as the feature word into the specific expression;
The article feature word extraction device according to (6).

  According to such a configuration of the present invention, a generalized specific expression that matches the input article and the title of the similar article is extracted, and the specific expression can be identified and output. By comparing with the titles, words, compound words, and phrases that are familiar to the titles can be extracted as feature words. Further, since the extracted generalized specific expression is returned to the original specific expression and output so as to be identifiable, the specific expression can be output in an easily understandable manner.

(8) Title input means for receiving input of a title related to the input article created by the user;
Article storage means for storing the title and the input article in association with each other in the past article storage unit;
Comprising
The article feature word extraction device according to any one of (1) to (7).

  According to such a configuration of the present invention, an article with a title created by the output of the apparatus can be included in a past article. Since past articles are stored in association with titles given by the user, the output input article can be a similar article from which feature words are extracted in other input articles.

(9) An article feature word extraction method for extracting feature words for a title corresponding to an input article by a computer,
A past article storing step of storing past articles classified into categories and their titles in association with each other;
A proper expression storage step for storing the specific expression;
An article input step for accepting an input article with the category;
A similar article extraction step for extracting a similar article similar in category to the input article received in the article input step from the past article stored in the past article storage step;
A specific expression extraction step for extracting each of the specific expressions stored in the specific expression storage step from the input article and the title of the similar article;
A specific expression generalizing step of generalizing the specific expression extracted by the specific expression extracting step, and applying a generalized specific expression to the title of the input article and the similar article;
A feature word extraction step of extracting a feature word from the title of the input article and the similar article after applying the generalized specific expression by the specific expression generalization step;
Article feature word extraction method including

  (10) A program for causing a computer to execute the steps of the article feature word extraction method according to (9).

  According to the present invention, feature words for titles can be efficiently extracted based on similar articles that are past articles in the same category as the input article. In particular, by generalizing the specific expressions, the specific expressions are extracted as feature words that match the titles of similar articles, so that specific expressions that are important in the title can be extracted as feature words.

It is a figure which shows the function structure of the article feature word extraction apparatus which concerns on this embodiment. It is a figure which shows the hardware constitutions of the article feature word extraction apparatus which concerns on this embodiment. It is a figure which shows the example of the process by the article feature word extraction apparatus which concerns on this embodiment. It is a flowchart of the main process which concerns on this embodiment. It is a flowchart of the similar article extraction process which concerns on this embodiment. It is a figure which shows the extraction and generalization method of the specific expression which concerns on this embodiment. It is a figure which shows the specific example which concerns on this embodiment.

  Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings. This is merely an example, and the technical scope of the present invention is not limited to this.

(Embodiment)
[Functional configuration of article feature word extraction device 1]
FIG. 1 is a diagram illustrating a functional configuration of an article feature word extraction device 1 according to the present embodiment. The article feature word extraction device 1 is a device that extracts a feature word for a title related to an input article and outputs a feature word that can be identified. One of the categories classified in advance is associated with the input article. “For title” means that the content of the input article can be expressed in a straightforward manner, for example, in the creation of a title, a subtitle, and a summary. A feature word is a word, compound word, or phrase in which a feature about an input article appears.

  The article feature word extraction device 1 includes an input unit 3, an output unit 5, a control unit 10, and a storage unit 20. The input unit 3 is, for example, a keyboard or a mouse, and is a device that inputs an input article to be processed by the article feature word extraction device 1. The input unit 3 may be a communication unit that receives articles input from a terminal (not shown) or a document server connected via a communication line (not shown). The output unit 5 is, for example, a display or the like, and is a device that outputs an input article after extracting feature words for titles. The output unit 5 may be a communication unit that transmits the processed input article to a terminal connected via a communication line.

  The control unit 10 includes an article input unit 11, a similar article extraction unit 12, a specific expression extraction unit 13, a specific expression generalization unit 14, a feature word extraction unit 15, an article output unit 16, a temporary title output unit 17, and a title input unit 18. The article storage means 19 is provided. The storage unit 20 includes a past article storage unit 22 that stores past articles, and a specific expression storage unit 24 that stores patterns of specific expressions. Specific expressions (named entities) refer to, for example, proper nouns such as place names, personal names, and organization names, dates, time expressions, monetary expressions, and quantity expressions. Details of each function will be described later.

  The article feature word extraction device 1 is not limited in the number of hardware, and may be configured by one or a plurality of hardware as necessary. Further, when the article feature word extraction device 1 is configured by a plurality of hardware, for example, each hardware may be connected via a communication line. The functions of this embodiment may be realized by using a separate server for each function described above and linking the servers by transmitting and receiving signals between the servers.

[Hardware configuration diagram of article feature word extraction device 1]
FIG. 2 is a diagram illustrating a hardware configuration of the article feature word extraction device 1 according to the present embodiment. The processing apparatus in which the present invention is implemented may be a standard one, and an example of the configuration is shown below.

  The article feature word extraction device 1 includes a CPU (Central Processing Unit) 1010 (a plurality of CPUs such as a CPU 1012 may be added in a multiprocessor configuration), a bus line 1005, a communication I / F (I / F). F: Interface) 1040, main memory 1050, BIOS (Basic Input Output System) 1060, display device 1022, I / O controller 1070, input device 1100 such as a keyboard and mouse, hard disk 1074, optical disk drive 1076, and semiconductor memory 1078 Prepare. The hard disk 1074, the optical disk drive 1076, and the semiconductor memory 1078 are collectively referred to as the storage unit 20.

  The control unit 10 is a part that controls the article feature word extraction device 1 in an integrated manner. By appropriately reading and executing various programs stored in the hard disk 1074, the control unit 10 cooperates with the hardware described above, and Various functions are realized.

  The communication I / F 1040 is a network adapter when the article feature word extraction device 1 transmits and receives information to and from other devices via a communication line. The communication I / F 1040 may include a modem, a cable modem, and an Ethernet (registered trademark) adapter.

  The BIOS 1060 records a boot program executed by the CPU 1010 when the article feature word extraction apparatus 1 is started, a program depending on the hardware of the article feature word extraction apparatus 1, and the like.

  The display device 1022 includes a display device such as a cathode ray tube display device (CRT) or a liquid crystal display device (LCD).

  The I / O controller 1070 can be connected to a storage unit 20 that is a storage device such as a hard disk 1074, an optical disk drive 1076, and a semiconductor memory 1078.

  The input device 1100 accepts input by the administrator of the article feature word extraction device 1.

  The hard disk 1074 stores various programs for causing the hardware to function as the article feature word extraction device 1, a program for executing the functions of the present invention, the above-described past article storage unit 22, the specific expression storage unit 24, and the like. Note that the article feature word extraction device 1 can also use a hard disk (not shown) separately provided as an external storage device.

  As the optical disk drive 1076, for example, a DVD-ROM drive, a CD-ROM drive, a DVD-RAM drive, or a BD (Blu-ray (registered trademark) Disc) drive can be used. In this case, the optical disk 1077 corresponding to each drive is used. A program or data may be read from the optical disk 1077 by the optical disk drive 1076 and provided to the main memory 1050 or the hard disk 1074 via the I / O controller 1070.

  The computer referred to in the present invention refers to an information processing device including a storage device, a control unit, and the like, and the article feature word extraction device 1 includes an information processing device including the storage unit 20, the control unit 10, and the like. This information processing apparatus is included in the concept of the computer of the present invention.

[Specific Example 1]
FIG. 3 is a diagram illustrating an example of processing by the article feature word extraction device 1 according to the present embodiment. FIG. 3A shows the text 30 of the input article. The “warming” that is the category of the text 30 is designated in advance by the user. The category is a topic for classifying articles, for example, politics, society, economy, sports, etc., and the input articles are classified by category. The category may be in a hierarchical format such as a large item, a medium item, and a small item. In this case, it is desirable that the category specified for the input article is a minimum unit such as a small item.

  Next, the control unit 10 extracts the title and text of a similar article similar to the text 30 from the past article storage unit 22. The past article storage unit 22 is included in the storage unit 20. As shown in FIG. 3B, the past article storage unit 22 has items of a category 22a, a title 22b, and a body 22c, and one record (row) represents one article.

  FIG. 3 (2) shows similar articles extracted from the past article storage unit 22. Similar articles are extracted from data in the same category as the input article stored in the category 22 a of the past article storage unit 22. In this example, the same “warming” category as the input article is extracted. The similar article is an article to be compared for extracting feature words from the input article. The extracted similar article includes a title 31 and a body 32. The title 31 is acquired from the title 22b of the past article storage unit 22, and the text 32 is acquired from the text 22c of the past article storage unit 22, respectively.

  Next, the control unit 10 extracts and generalizes each specific expression from the body 30 of the input article and the title 31 of the similar article. FIG. 3 (3) shows a specific expression included in the text 30 and the title 31, and a generalized specific expression corresponding to the specific expression. The specific expression extracted from the text 30 of the input article is a place name “France” and a date “23 days”, and is stored in the specific expression unit 33. The generalization unit 34 stores a generalized specific expression obtained by generalizing the specific expression stored in the specific expression unit 33. The generalized expression is performed based on the specific expression storage unit 24. Similarly, the specific expression extracted from the title 31 of the similar article is stored in the specific expression unit 35. The generalization unit 36 stores a generalized specific expression obtained by generalizing the specific expression stored in the specific expression unit 35. These unique representation units 33 and 35 and generalization units 34 and 36 may be temporarily stored in the storage unit 20.

  Next, feature words are extracted in a state where the specific expressions are generalized. The feature word includes the generalized specific expression of the body 30 of the input article that matches the generalized specific expression included in the title 31 of the similar article. In this example, the feature words that are generalized specific expressions are “place name” and “date”. The feature words include words, compound words, and phrases excluding the generalized specific expression of the title 31 of the similar article, which are included in the text 30. In this example, the characteristic words are “Arctic Ocean”, “Sea Ice Area”, “Decrease”, “Announcement”. Here, “sea ice area” is a compound word including the word “ice”.

  FIG. 3 (4) shows the input article body 37 and the input article provisional title 38 in which feature words can be identified. Here, the feature words extracted by the generalized specific expression may be output as a feature word list. However, as shown in the figure, the feature words may be output so as to be identifiable while being included in the body 37 of the input article. Good. The text 37 of the input article may be underlined with the feature word as in this example, or may be colored or highlighted to be output in a state that can be identified at a glance to the user. Is desirable. Further, the temporary title 38 in the figure is obtained by replacing the part extracted as the feature word in the title 31 of the similar article in FIG. 3B with the one extracted as the feature word of the input article. Here, when there are a plurality of replacements, the temporary title 38 may be displayed as a plurality of title candidates. This replacement can be performed by associating the input article with the title of the similar article (particularly, associating the same specific expression type). At this time, the title 31 and the temporary title 38 of the similar article may be displayed side by side for the user. Further, it is desirable that the user can edit the temporary title 38 created in this way as necessary.

[flowchart]
Next, the flow of processing described in FIG. 3 will be described. FIG. 4 is a flowchart of the main process according to the present embodiment. FIG. 5 is a flowchart of similar article extraction processing according to the present embodiment. FIG. 6 is a diagram showing a method for extracting and generalizing a specific expression according to the present embodiment. Hereinafter, a description will be given with reference to FIGS.

  In FIG. 4, S <b> 1: the control unit 10 (article input unit 11) receives an input article with a category from the input unit 3. The input article input from the input unit 3 of the article feature word extraction device 1 may be temporarily stored in the storage unit 20.

  S2: The control unit 10 (similar article extraction means 12) performs a similar article extraction process for extracting an article (similar article) similar to the input article from the past article storage unit 22.

  Here, the similar article extraction processing will be described with reference to FIG.

  In FIG. 5, S11: the similar article extracting unit 12 extracts articles of the same category as the category of the input article from the past article storage unit 22.

  S12: The similar article extraction means 12 performs morphological analysis on the input article, and the extracted body text and title of the same category.

  S13: The similar article extracting means 12 uses TF (Term Frequency) / IDF (Inverse Document Frequency), which is an index based on the appearance frequency of words, for each word included in the body of each article by morphological analysis. Weighting is performed. Thereby, although it does not appear so much in other articles, weighting of words frequently used in the article increases.

  S14: The similar article extracting means 12 extracts words included in the input article and having higher weights as feature words. A plurality of feature words may be extracted.

  S15: The similar article extraction means 12 performs an OR search on similar articles of the same category from which the extracted feature words are extracted, and extracts n articles in order from the highest score. The number (n) of articles to be extracted may be determined in advance, or may be input from the input unit 3 by the user. Moreover, the control part 10 may determine the number of cases at random.

  In this way, the feature words to be extracted can be changed by changing the top n items used as similar articles.

  Returning to FIG. 4, S3: The control unit 10 (the specific expression extraction unit 13) extracts the specific expression from the input article and the title of the extracted similar article.

  S4: The control unit 10 (specific expression generalization means 14) generalizes each specific expression included in the input article and the title of the similar article based on the learning data in the specific expression storage unit 24. Here, the title means a headline whose content is limited to a predetermined maximum number of characters, and the maximum number of characters is, for example, 13 characters.

  Here, extraction and generalization of specific expressions will be described based on the example of FIG.

  FIG. 6A is an example of learning source data 40 stored in the specific expression storage unit 24. The learning source data 40 is obtained by performing morphological analysis on a document, adding a notation 41, part of speech 42, and utilization 43, and adding a correct tag to the tag 44. “ORG” of the tag 44 is a generalized unique expression representing an organization, and “PER” is a personal name. Further, “I−” of the tag 44 is a symbol that means that it follows “B−”. “O” of the tag 44 is a symbol indicating “other” which is not a unique expression.

  In addition, the specific expression storage unit 24 prepares a feature 50 for determining a specific expression. The feature 50 is one example, and a plurality of features 50 may be prepared. The learning data 51 is obtained by using the feature 50 for “Naohiro” shown in the row 45 of the learning source data 40. The feature 50 is performed for all the morphemes, that is, all the rows of the learning source data 40. Then, these are stored as learning data in the specific expression storage unit 24. By doing so, when an unknown document is input, it can be used to determine which word or the like is a specific expression.

  When the target data 60 of the document shown in FIG. 6B is input, the result data 62 uses the feature 50 for “Tetsu” in the row 61. From the result data 62, the two words “Mr.” and “ha” before and after the learning data 51 are common, so it can be inferred that “Tetsu” is attached with “PER”, which is a tag of the personal name. Therefore, “Tetsu” in line 61 can be made a unique expression by attaching a personal name tag “PER”.

  Returning to FIG. 4, S5: The control unit 10 (feature word extraction means 15) extracts feature words included in both the input article and the title of the similar article. Specifically, the control unit 10 extracts a generalized specific expression included in the input article and the title of the similar article, and extracts words, compound words, and phrases that match the input article and the title of the similar article. .

  S6: The control unit 10 (article output means 16) outputs the input article in which the feature word can be identified to the output unit 5. At that time, the control unit 10 returns (specializes) the generalized specific expression to the specific expression before generalization, and underlines the feature word. By doing so, the user can identify the feature word from the output input article, which is effective for, for example, oblique reading of the article.

  S7: The control unit 10 (temporary title output means 17) replaces the feature word of the title of the similar article with the feature word of the input article, creates a temporary title related to the input article, and outputs it to the output unit 5. At that time, the control unit 10 may return (specify) the generalized specific expression to the specific expression before generalization, and underline the feature word. As described above, since the title of the input article can be semi-automated by the control unit 10, the efficiency of title creation by the user can be improved.

  S8: The control unit 10 (title input means 18) receives the title of the input article created by the user from the input unit 3.

  S9: The control unit 10 (article accumulation means 19) stores the input article and its title in the past article storage unit 22 in association with each other.

  As described above, the article feature word extraction apparatus 1 can efficiently extract feature words for a title based on the title of a similar article that is a past article in the same category as the input article. In particular, by generalizing the unique expressions, the article feature word extraction device 1 extracts the unique expressions as feature words that match the titles of similar articles, so that the unique expressions that are important in the title are extracted as feature words. Can do.

[Specific Example 2]
Next, a specific example based on an actual article is shown. FIG. 7 is a diagram illustrating a specific example according to the present embodiment.

  FIG. 7A shows the text 70 of the input article. The category of the input article is “entertainment”. Note that the text 70 is not underlined or the like when the input article is first given.

  FIG. 7B shows three articles extracted from the articles stored in the past article storage unit 22 and whose category is “entertainment” and three titles similar to the input article. Similar articles can be extracted, for example, by the method shown in FIG. Here, there are three pairs: a body 71 and title 72 pair, a body 73 and title 74 pair, and a body 75 and title 76 pair. The texts 71, 73, and 75 are all similar to the text 70.

  FIG. 7 (3) is a generalization of the unique expression of the title. The titles 72, 74, and 76 commonly include a talent name, which is generalized with a tag <PER> representing the personal name. FIG. 7 (4) is a generalization of the specific expression of the text 70 of the input article. The talent name included in the text 70 is generalized with a tag <PER> representing the person name. These can be determined to represent personal names from the patterns stored as the learning data shown in FIG.

  In FIG. 7 (5), characteristic words of the text 70 and the titles 72, 74, 76 are extracted and reflected in the text 70. By extracting feature words by comparing the generalized expressions in a generalized state, it is possible to pick up specific expressions suitable for the title.

  Since the text 70 is changed to a state in which the user can easily identify the feature word and displayed on the output unit 5, the user extracts the feature word from the text 70, for example, “Mr. You can easily add a title that clearly shows the content of the article, “A fracture of the ribs”.

  As mentioned above, although embodiment of this invention was described, this invention is not restricted to embodiment mentioned above. The effects described in the embodiments of the present invention are only the most preferable effects resulting from the present invention, and the effects of the present invention are limited to those described in the embodiments of the present invention. is not.

(Deformation)
In the present embodiment, the feature article is underlined and the input article is output. However, if the user can identify the feature article, for example, the feature word is displayed differently from other words, compound words, and phrases. May be output. As a display mode, it is conceivable to use different colors and fonts. In this way, by outputting the feature word in a display mode different from other words, compound words, and phrases, the user can grasp the feature word so that it can be identified, which is effective for, for example, oblique reading of articles. Also, by changing the top n items used as similar articles, the feature words to be extracted can also be changed, so by selecting the appropriate number (n), multiple feature words for oblique reading are displayed. It is possible to make it.

  In this embodiment, in order to extract similar articles, the body of the article is extracted by performing morphological analysis and weighting using TF / IDF. However, if an article similar to the input article can be output, Other methods may be used.

  In the present embodiment, an example is shown in which a temporary title in which the feature word of the similar article title is replaced with the feature word of the input article is generated and output together with the input article in which the feature word is emphasized, but is not limited thereto. . For example, by outputting the input article and the title of a similar article and connecting the feature word of the input article that can be identified with the feature word of the title of the similar article that is also identifiable by a line, each correspondence You may make it show the place to do.

DESCRIPTION OF SYMBOLS 1 Article feature word extraction apparatus 10 Control part 11 Article input means 12 Similar article extraction means 13 Specific expression extraction means 14 Specific expression generalization means 15 Feature word extraction means 16 Article output means 17 Temporary title output means 18 Title input means 19 Article accumulation | storage Means 20 Storage unit 22 Past article storage unit 24 Specific expression storage unit

Claims (9)

In an article feature word extraction device that extracts feature words for a title corresponding to an input article,
An article input means for receiving an input article with a category;
A similar article similar to the input article whose category matches the input article received by the article input unit is extracted from the past article storage unit that stores the title in association with the past article. Similar article extraction means,
Using a specific expression storage unit for storing a specific expression, specific expression extraction means for extracting the specific expression from the input article and the title of the similar article;
A specific expression generalizing means for generalizing the specific expression extracted by the specific expression extracting means and applying a generalized specific expression to the title of the input article and the similar article;
In the title of the input article and the similar article after the application of the generalized specific expression by the specific expression generalization means , a word, a compound word, and a phrase that match the input article and the title of the similar article A feature word extracting unit that extracts the generalized specific expression that matches the input article and the title of the similar article as a feature word, as a feature word;
Article feature word extraction device. Replacing the feature word of the title of the similar article extracted by the feature word extraction means with the corresponding feature word of the input article, and provisional title output means for creating and outputting a temporary title related to the input article,
The article feature word extraction device according to claim 1. The specific expression storage unit stores the pattern of the specific expression extracted using machine learning,
The specific expression extraction unit extracts the specific expressions from the input article and the title of the similar article based on the specific expression pattern,
The article feature word extraction device according to claim 1 or 2 . The similar article extraction means extracts a plurality of the similar articles by changing the number of extraction targets.
The article feature word extraction device according to any one of claims 1 to 3 . Comprising an article output means for outputting the input article in which the feature words extracted by the feature word extraction means can be identified;
The article feature word extraction device according to any one of claims 1 to 4 . Before SL Article output unit outputs the input article is specialized in the named entities the generalization named entities extracted as the feature words,
The article feature word extraction device according to claim 5 . Title input means for receiving input of a title related to the input article created by the user;
Article storage means for storing the title and the input article in association with each other in the past article storage unit;
Comprising
The article feature word extraction device according to any one of claims 1 to 6 . An article feature word extraction method for extracting feature words for a title corresponding to an input article by a computer,
A past article storing step of storing past articles classified into categories and their titles in association with each other;
A proper expression storage step for storing the specific expression;
An article input step for accepting an input article with the category;
A similar article extraction step for extracting a similar article similar in category to the input article received in the article input step from the past article stored in the past article storage step;
A specific expression extraction step for extracting each of the specific expressions stored in the specific expression storage step from the input article and the title of the similar article;
A specific expression generalizing step of generalizing the specific expression extracted by the specific expression extracting step, and applying a generalized specific expression to the title of the input article and the similar article;
In the title of the input article and the similar article after the application of the generalized specific expression in the specific expression generalization step , a word, a compound word, and a phrase that match the input article and the title of the similar article A feature word extracting step of extracting the generalized specific expression that matches the input article and the title of the similar article as a feature word , and extracting it as a feature word;
Article feature word extraction method including A program for causing a computer to execute the steps of the article feature word extraction method according to claim 8 .

Images