JP2013200794A - Device, method, and program for attribute extraction - Google Patents

Abstract

PROBLEM TO BE SOLVED: To extract an appropriate range of attribute information from text data and prevent erroneous extraction of the attribute information.SOLUTION: An attribute extraction device 1 has a language analysis means 11 that executes language analysis of text data including a user profile and adds a semantic attribute to an indeclinable head included in the text data, a head semantic attribute extraction pattern storage means 11 that stores therein at least one semantic attribute pattern and corresponding attribute classification, and a head semantic attribute extraction means 11 that, if the semantic attribute added to the indeclinable head matches any one of the semantic attribute pattern stored in the head semantic attribute extraction pattern storage means, extracts a noun phrase of an articulation including the indeclinable head as attribute information of the attribute classification corresponding to the matched semantic attribute pattern.

Description

  The present invention relates to an attribute extraction apparatus, an attribute extraction method, and an attribute extraction program that analyze text data and extract user attribute information.

  Analyzing text data distributed on the Internet by morphological analysis, etc., and extracting words and phrases (Non-Patent Document 1), and attributes describing attributes corresponding to the notation of the words for the extracted words It is compared with the dictionary and classified.

“Basic Japanese Analysis Technology as a Base for Knowledge Extraction from Text”, NTT Technical Journal 2008.6, p20-23

  If the extracted word matches the word registered in the attribute dictionary, attributes such as occupation can be determined. However, when the word registered in the attribute dictionary is a part of a certain attribute, the entire extracted word cannot be extracted as attribute information. For example, when “engineer (profession)” is registered in the attribute dictionary, even if the word extracted from the text is “network engineer”, only “engineer” can be extracted as the occupation.

  Also, if the attribute is not represented in the sentence structure, it may not be extracted as an appropriate attribute or may be erroneously extracted. For example, when “engineer (profession)” is registered in the attribute dictionary, even if the word extracted from the text is “organized by an infrastructure engineer study group”, “engineer” is erroneously extracted as a profession.

  In order to extract the appropriate range of attribute information and to prevent erroneous extraction, it is possible to create an attribute dictionary that registers a wide range of notations and words, but for creating and maintaining an attribute dictionary that covers a wide range of notations, A very large load is required.

  The present invention has been made in view of the above circumstances, and an object of the present invention is to extract an appropriate range of attribute information from text data and to prevent attribute information from being erroneously extracted, an attribute extraction method, And providing an attribute extraction program.

  In order to achieve the above object, the present invention is an attribute extraction device, which linguistically analyzes text data including a user profile and assigns a semantic attribute to a syntactic main word included in the text data, and at least A main meaning semantic attribute extraction pattern storage means storing one semantic attribute pattern and a corresponding attribute classification, and a semantic attribute given to the syntactic main character is stored in the main meaning meaning attribute extraction pattern storage means. And the meaning attribute extraction means for extracting the noun phrase of the phrase including the syntactic main word as attribute information of the attribute classification corresponding to the matching semantic attribute pattern.

  The present invention is an attribute extraction method performed by an attribute extraction apparatus, which performs language analysis on text data including a user profile, and assigns semantic attributes to a syntactic headword included in the text data; If the semantic attribute assigned to the main word matches one of the semantic attribute patterns stored in the main word semantic attribute extraction pattern storage means, the noun phrase of the phrase containing the body language main word corresponds to the matching semantic attribute pattern A main meaning attribute extracting step for extracting as attribute information of the attribute classification to be performed.

  The present invention relates to an attribute extraction program executed by an attribute extraction device, which performs language analysis on text data including a user profile and assigns a semantic attribute to a syntactic subject included in the text data. Means, at least one semantic attribute pattern, and a main meaning semantic attribute extraction pattern storage means in which a corresponding attribute classification is stored, and a semantic attribute assigned to the syntactic main character is stored in the main meaning semantic attribute extraction pattern storage means. If it matches any of the stored semantic attribute patterns, it functions as a main meaning attribute extraction means that extracts the noun phrase of the phrase containing the body part as the attribute information of the attribute classification corresponding to the matched semantic attribute pattern This is an attribute extraction program.

  According to the present invention, it is possible to provide an attribute extraction device, an attribute extraction method, and an attribute extraction program that extract attribute information in an appropriate range from text data and prevent erroneous extraction of attribute information.

It is a block diagram which shows the structure of the attribute extraction apparatus which concerns on embodiment of this invention. It is a block diagram which shows the structure of an attribute extraction part. It is a figure which shows the specific example of a process of an attribute extraction part. It is a figure which shows an example of a head meaning attribute extraction pattern memory | storage part. It is a figure which shows an example of a part of speech and description extraction pattern memory | storage part. It is a figure which shows an example of a specific expression extraction pattern memory | storage part. It is a figure which shows an example of word-of-mouth data with an attribute and total data. It is a figure which shows an example of an analysis result.

  Hereinafter, embodiments of the present invention will be described.

  FIG. 1 is a configuration diagram showing a configuration of an attribute extraction device 1 according to an embodiment of the present invention. The attribute extraction apparatus 1 according to the present embodiment is an apparatus that inputs a large amount of text data to be disclosed on a network, and analyzes and classifies data using user attribute information extracted from each text data.

  In the present embodiment, as text data, a profile text 21 in which a user profile is described and a review text 22 such as a tweet, a review, and an impression input by the user of the profile text 21 are paired and the attribute extraction device 1. Is input. In this embodiment, the review text 22 acquired from the network and the profile text 22 of the user who has input the review text 22 are input to the attribute extraction apparatus 1 as a pair.

  In addition, as input text data (profile text 21 and word-of-mouth text 22), for example, Twitter, blog, Facebook, etc. can be considered.

  The attribute extraction apparatus 1 according to the present embodiment includes an attribute extraction unit 11, a word-of-mouth extraction unit 12, a word-of-mouth information storage unit 13 with attributes, and an analysis unit 14.

  The attribute extraction unit 11 analyzes the input profile text 21 and extracts attribute information. The attribute information includes occupation, hobby / preference, age / age, location, and the like.

  FIG. 2 is a block diagram illustrating a configuration of the attribute extraction unit 11. The attribute extraction unit 11 shown in the figure includes a basic language analysis unit 101, a subject meaning attribute extraction unit 102, a part of speech / notation extraction unit 103, a specific expression extraction unit 104, a subject part meaning attribute extraction pattern storage unit 105, a part of speech / A notation extraction pattern storage unit 106 and a specific expression extraction pattern storage unit 107 are provided.

  When the profile text 21 is input, the basic language analysis unit 101 performs basic language analysis such as morphological analysis, specific expression extraction, and dependency analysis on the profile text 21. Morphological analysis divides input text into words and adds information such as parts of speech to each word. Specific expression extraction specifies specific expression data such as a person name, place name, and organization name from a word string divided by morphological analysis, and assigns a specific expression type (person name, place name, organization name, etc.) to the specified specific expression data. . Dependency analysis analyzes which clause modifies (related) which clause.

  In addition, the basic language analysis unit 101 of this embodiment gives a semantic attribute to each word. The semantic attribute is a predetermined classification system. For example, a Japanese vocabulary system (ISBN4-00-130101-6 C3881 Iwanami Shoten) may be used. In addition, the syntactic headwords used in the following explanation are those that are closest to the end of the phrase in the phrase (parts of speech are nouns, noun suffixes, numerals, classifiers, etc.).

  The subject meaning attribute extraction unit 102 extracts attribute information from the profile text 21 using the subject meaning attribute extraction pattern storage unit 105. That is, the main meaning semantic attribute extraction unit 102 includes the main body main character when the semantic attribute given to the main character system main character matches any semantic attribute pattern stored in the main character meaning attribute extraction pattern storage unit 105. The noun phrase of the phrase is extracted as attribute information of the attribute classification corresponding to the matched semantic attribute pattern. The main meaning attribute extraction pattern storage unit 105 stores at least one semantic attribute pattern and a corresponding attribute classification (attribute type, attribute class).

  The part-of-speech / notation extraction unit 103 refers to the part-of-speech / notation extraction pattern storage unit 106 and extracts attribute information from the profile text 21 based on the part of speech / word notation. That is, the part-of-speech / notation extraction unit 103 sets the attribute classification corresponding to the matched part-of-speech notation pattern to data that matches any part-of-speech notation pattern stored in the part-of-speech / notation extraction pattern storage unit 106. Extract as attribute information. The part-of-speech / notation extraction pattern storage unit 106 stores at least one part-of-speech notation pattern, which is a combination of part-of-speech and notation, and a corresponding attribute classification.

  The specific expression extraction unit 104 extracts attribute information from the profile text 21 using the specific expression extraction pattern storage unit 107. That is, the unique expression extraction unit 104 matches the specific expression data when the specific expression type given to the specific expression data matches one of the specific expression patterns stored in the specific expression extraction pattern storage unit 107. It is extracted as attribute information of the attribute classification corresponding to the specific expression pattern. The specific expression extraction pattern storage unit 107 stores at least one specific expression pattern and a corresponding attribute classification.

  The review extraction unit 12 analyzes the review text 22 input as a pair with the profile text 21 and extracts the review information. For example, the word-of-mouth communication extraction unit 12 extracts, as word-of-mouth information, evaluation attributes (objects), evaluation expressions (expressions described in the profile text 22 (cool, refreshing, etc.)), polarities (favorable or unpopular), etc. To do.

  The attributed word-of-mouth information storage unit 13 stores the attribute information extracted by the attribute extraction unit 11 and the word-of-mouth information extracted by the word-of-mouth extraction unit 12 in association (in pairs). However, nothing is stored in the attributed word-of-mouth information storage unit 13 for the word-of-mouth text 22 from which no keyword has been extracted.

  The analysis unit 14 analyzes the attribute information and the review information stored in the attributed review information storage unit 13. That is, the analysis unit 14 analyzes the word-of-mouth information extracted from the word-of-mouth text 22 using at least one attribute information extracted from the profile text 21. The analysis unit 14 illustrated in FIG. 1 includes an analysis target attribute analysis unit 15, a competition comparison analysis unit 16, and an attribute-specific trend analysis unit 17.

  The analysis target attribute analysis unit 15 analyzes user attribute information for a certain analysis target keyword and displays the analysis result. For example, the attribute information of each profile text 21 corresponding to each review text 22 including a certain analysis target keyword is analyzed / aggregated (for example, occupation class aggregation, hobby preference aggregation, gender aggregation, etc.), and each is represented by a pie chart, a bar chart, etc. indicate.

  The competition comparison attribute analysis unit 16 analyzes user attribute information for a plurality of analysis target keywords, and compares and displays the analysis results. For example, the attribute information of each profile text 21 corresponding to each review text 22 including each analysis target keyword is analyzed and aggregated (for example, occupation class aggregation, hobby preference aggregation, gender aggregation, etc.), and a pie chart, bar graph, etc. indicate.

  The attribute-specific trend analysis unit 17 displays the analysis target keyword aggregation result for a certain user attribute. For example, the ranking of the analysis target keywords aggregated only by users having user attributes of gender = “female” and hobbies / preferences = “music” is performed.

  As the attribute extraction apparatus 1 described above, for example, a general-purpose computer system including a CPU, a memory, an external storage device such as an HDD, an input device, and an output device can be used. In this computer system, each function of the attribute extraction device 1 is realized by the CPU executing a program for the attribute extraction device 1 loaded on the memory. Further, the program for the attribute extraction apparatus 1 can be stored in a computer-readable recording medium such as a hard disk, a flexible disk, a CD-ROM, an MO, or a DVD-ROM, or can be distributed via a network.

  Next, the processing of this embodiment will be described.

  FIG. 3 shows an attribute extraction process of the attribute extraction unit 11.

  In the example shown in the figure, it is assumed that “I am a river engineer. I like foreign cars. Kenji Miyazawa has read all” 301 is input to the attribute extraction unit 11 as the profile text.

  The basic language analysis unit 101 performs basic language analysis on the input profile text 301 to output analysis result data 302 (S11). That is, the basic language analyzing unit 101 performs morphological analysis, divides the word into units divided by “/” or “//”, and gives a part of speech (not shown) to each word. Further, in this embodiment, when performing morphological analysis, a semantic attribute (a classification representing the meaning of the word) is assigned to all words. In the example shown in the drawing, a semantic attribute of “professional technical position” is given to the physicist main word “engineer”. “Engineer”, which is a synonymous headword, is the most backward noun (or pronoun) in the “river / engineer / wo” clause. Note that semantic attributes are also given to other words, but they are omitted here.

  Further, the basic language analysis unit 101 extracts “Kenji Miyazawa” as a specific expression of a person name by performing a specific expression extraction, and “(person name (specific expression))” (specific expression type) is included in the extracted specific expression data. ) Is given. Further, the basic language analysis unit 101 performs dependency analysis to detect a phrase of a word delimited by “/” by morphological analysis, and sets “//” at a position that becomes a phrase boundary line. The basic language analysis unit 101 outputs the analysis result data 302 by performing such language analysis.

  The main meaning attribute extraction unit 102, the part of speech / notation extraction unit 103, and the specific expression extraction unit 104 extract attribute information from the analysis result data 302 of the profile text, respectively (S21).

  The subject meaning attribute extraction unit 102 extracts attribute information using the subject meaning attribute extraction pattern storage unit 105. Specifically, the subject meaning attribute extraction unit 102 collates the semantic attributes assigned to the subject with each semantic attribute pattern of the subject meaning attribute extraction pattern storage unit 105 for each body language subject, For the syntactic main word that matches the semantic attribute pattern, a noun phrase included in the same phrase as the syntactic main word is extracted as attribute information. In the analysis result data 302 shown in FIG. 3, the syntactic main words are “engineer”, “like”, “kenji”, and “all”.

  FIG. 4 is a diagram illustrating an example of the main meaning attribute extraction pattern storage unit 105. The main character meaning attribute extraction pattern storage unit 105 shown in the figure stores attribute types and attribute classes (attribute classifications) and semantic attribute patterns in association with each other. The subject meaning attribute extraction unit 102 collates each meaning attribute assigned to each of “engineer”, “like”, “kenji”, and “all” with the meaning attribute pattern of the subject meaning attribute extraction pattern storage unit 105. In this case, it is assumed that the semantic attribute (technical technical position) of “engineer” matches the semantic attribute pattern (professional technical position or technical position) of the main word meaning attribute extraction pattern storage unit 105, and the others do not match. The main meaning attribute extraction unit 102 extracts “river engineer” of the noun phrase included in the same phrase as the matched “engineer” as attribute information (attribute value). Further, the main meaning attribute extraction unit 102 uses the attribute classification (profession) and the attribute class (special technical job) of the matched data (record) in the main meaning attribute extraction pattern storage unit 105 as the attribute classification of the extracted “river engineer”. Is identified. Then, the main meaning attribute extraction unit 102 outputs the extracted attribute information (“river engineer”) as attribute information together with the identified attribute classification (“attribute type (profession), attribute class (professional technical position)”). To do.

  The part-of-speech / notation extraction unit 103 uses the part-of-speech / notation extraction pattern storage unit 106 to extract attribute information. The part-of-speech / notation extraction unit 103 extracts an expression or a part of the expression that matches any of the part-of-speech / notation extraction patterns stored in the part-of-speech / notation extraction pattern storage unit 106.

  FIG. 5 is a diagram illustrating an example of the part of speech / notation extraction pattern storage unit 106. The part-of-speech / notation extraction pattern storage unit 106 shown in the figure stores attribute types, attribute classes, and part-of-speech / notation patterns in association with each other. In the illustrated part-of-speech / notation extraction pattern storage unit 106, an example of extracting age, hobbies / preferences, and sex is given. That is, the part-of-speech / notation extraction unit 103, when there is a description in the analysis result data 302 that matches “part-of-speech = noun” and “notation = like” set in the part-of-speech / notation pattern, And the noun part is extracted as attribute information. In the example shown in FIG. 3, since “like a car” matches the part of speech / notation pattern, “car” is extracted as attribute information (attribute value). The part-of-speech / notation extraction unit 103 identifies the attribute type (hobby / preference) and attribute class (others) of the matched part-of-speech / notation pattern in the part-of-speech / notation extraction pattern storage unit 106 as the attribute classification of the “external vehicle” extracted. Then, the part-of-speech / notation extraction unit 103 outputs the extracted attribute information together with the identified attribute classification as attribute information.

  In addition, the part of speech / notation extraction unit 103 includes, in the analysis result data 302, a description that matches “part of speech = numerical”, “part of speech = classifier, notation = age or age” set in the part of speech / notation pattern. , It is determined that the attribute information is age, and the number (for example, 24) is extracted as attribute information (attribute value). The part-of-speech / notation extraction unit 103 also determines the attribute type (age) and attribute class (10's / 20's / 30's) of the matched part of speech / notation pattern storage unit 106 as the attribute classification of the extracted attribute information. Identify the cost. For the attribute class, the corresponding age is set from the attribute information (for example, in the case of 24, the 20s are set). Then, the part-of-speech / notation extraction unit 103 outputs the extracted attribute information together with the identified attribute classification as attribute information. Note that the attribute information of gender is extracted in the same manner using the part of speech / notation extraction pattern storage unit 106.

  The specific expression extraction unit 104 uses the specific expression extraction pattern storage unit 107 to extract attribute information. Specifically, the specific expression extraction unit 104 compares the specific expression type of the data with the specific expression pattern of the specific expression extraction pattern storage unit 107 for the specific expression data extracted as the specific expression in the analysis result data 302. If there is a match, the specific expression data is extracted as attribute information.

  In the example shown in FIG. 6, the specific expression data in the analysis result data 302 is “Kenji Miyazawa” and the type is “person name”. Therefore, the specific expression extraction unit 104 extracts “Kenji Miyazawa” as attribute information (attribute value), and the attribute classification (hobby / preference) of the corresponding specific expression pattern in the specific expression extraction pattern storage unit 107 as the attribute classification, Specify the attribute class (person name). Then, the unique expression extraction unit 104 outputs the extracted attribute information as attribute information together with the specified attribute classification. In addition, the attribute information of the attribute class “artificial” (product name, book name, etc.) with the attribute type “hobby / preference”, and the attribute information of the attribute class “location” with the attribute type “location” Extraction is similarly performed using the extraction pattern storage unit 107.

  FIG. 7A is a diagram illustrating an example of attributed review data stored in the attributed review information storage unit 13. The review data with attributes shown in the figure are the basic review data (review ID, outgoing date, review text), basic profile data (user ID, profile text), and review extraction results (analysis keyword, evaluation attribute / evaluation expression / polarity) And the profile extraction result (attribute type / attribute class / attribute value of each extracted attribute information) are stored in association with each other. The review extraction result is the review information shown in FIG. 1, and the profile extraction result is the attribute information shown in FIG.

  When the word-of-mouth data shown in the figure “Newly released XX-01 is cool!” Is input, the word-of-mouth extraction unit 12 performs language analysis on the data, so that the evaluation attribute (none), the evaluation expression (cool), The polarity (popular) is extracted as the evaluation / review for the analysis keyword (XX-01).

FIG. 7B is a data example in which the analysis unit 14 analyzes and summarizes the attributed word-of-mouth data in FIG. 7A. The analysis unit 14 performs a desired analysis using such aggregated data and outputs an analysis result.
FIG. 8 shows an example of the analysis result of the competition comparison attribute analyzed by the competition comparison attribute analysis unit 16. In the illustrated example, attribute information (profession) of each profile text 21 corresponding to each review text 22 including a plurality of analysis target keywords (each convenience store name, each department store name) is analyzed and tabulated, and a bar graph is displayed for each analysis keyword. it's shown. From this graph, it can be seen that, for example, the distribution tendency of occupations differs between department stores and convenience stores. Specifically, it is possible to understand that the convenience store mainly has students and the department store has specialized technical positions (however, department store A tends to be a convenience store). By such comparative analysis, analysis such as measurement of event / campaign effect and estimation of target layer can be easily performed.

  FIG. 8 shows an example of the analysis result of the competitive comparison attribute, but one bar graph of FIG. 8 shows an example of the analysis result of the analysis target analyzed by the analysis target attribute analysis unit 15. Moreover, based on the analysis result (not shown) of the trend analysis unit 17 by attribute, it analyzes and aggregates what is currently being analyzed (keywords to be analyzed) with the specified attribute information (and other conditions), and develops a new product. Etc. For example, it is possible to analyze keyword rankings that many “students” tweet in the middle of the night, and keyword rankings that tweet “music lovers” in “female” living in “Tokyo”.

  In the present embodiment described above, attribute information such as occupation, hobbies, and preferences is extracted from an appropriate range from profile text (user profile such as Twitter and blog) described in a free format, and an error in attribute information is detected. Extraction can be prevented. Specifically, even attribute information (attribute values) that are not registered in the main meaning attribute extraction pattern storage unit 105, the part-of-speech / notation extraction pattern storage unit 106, and the unique expression extraction pattern storage unit 107 are appropriately extracted. be able to. Moreover, erroneous extraction can be prevented by assigning an appropriate attribute classification to the extracted attribute information (attribute value).

  In addition, in this embodiment, attribute information can be automatically extracted from a profile text written in a free format in a wide range of expressions, so that target information can be obtained without conducting a questionnaire for hobby / preference investigations. Can be collected. Further, even if there is no specific entry for each category of occupation, hobby, and preference in the profile text (user profile field, etc.), desired attribute information can be extracted.

  In the present embodiment, the word-of-mouth extraction unit 12 analyzes the word-of-mouth text 22 and extracts evaluation attributes, evaluation expressions, and polarities as evaluation / review for the analysis keyword, and analyzes the extracted word-of-mouth information using the attribute information. This makes it possible to analyze finer positives and negatives.

  In addition, this invention is not limited to the said embodiment, Many deformation | transformation are possible within the range of the summary.

DESCRIPTION OF SYMBOLS 1: Attribute extraction apparatus 11: Attribute extraction part 101: Basic language analysis part 102: Main part meaning attribute extraction part 103: Part of speech / notation extraction part 104: Specific expression extraction part 105: Main part meaning attribute extraction pattern storage part 106: Part of speech / notation Extraction pattern storage unit 107: Specific expression extraction pattern storage unit 12: Review extraction unit 13: Attributed review information storage unit 14: Analysis unit 15: Analysis target attribute analysis unit 16: Competitive comparison attribute analysis unit 17: Trend analysis unit by attribute 21: Profile text 22: Review text

Claims (10)

An attribute extraction device,
Linguistic analysis of text data including a user profile, and language analysis means for assigning a semantic attribute to the syntactic subject included in the text data;
A main meaning attribute extraction pattern storage unit storing at least one semantic attribute pattern and a corresponding attribute classification;
If the semantic attribute assigned to the syntactic main character matches any semantic attribute pattern stored in the main character semantic attribute extraction pattern storage means, the noun phrase of the phrase containing the syntactic main character is matched. An attribute extraction device comprising: a main meaning attribute extraction means for extracting as attribute information of an attribute classification corresponding to an attribute pattern. The attribute extraction device according to claim 1,
The language analysis means divides the text data into words, gives parts of speech to each word,
A part-of-speech notation extracting means for extracting attribute information from the text data based on the word part-of-speech and the word notation;
Part-of-speech notation extraction pattern storage means storing at least one part-of-speech notation pattern, which is a combination of part of speech and notation, and a corresponding attribute classification;
The part-of-speech notation extraction unit extracts data or a part of the data that matches any of the part-of-speech notation patterns stored in the part-of-speech notation extraction pattern storage unit as attribute information of an attribute classification corresponding to the matched part-of-speech notation pattern An attribute extraction device characterized by The attribute extraction device according to claim 1,
The language analysis means identifies specific expression data included in the text data, assigns a specific expression type to the specific expression data,
A unique expression extraction pattern storage means storing at least one unique expression type pattern and a corresponding attribute classification;
When the unique expression type given to the specific expression data matches any of the specific expression type patterns stored in the specific expression extraction pattern storage unit, the specific expression data corresponds to the matched specific expression type. An attribute extraction device characterized by further comprising specific expression extraction means for extracting the attribute classification as attribute information. The attribute extraction device according to any one of claims 1 to 3,
The attribute extraction apparatus characterized in that the attribute information includes at least one of occupation, age, sex, and location. The attribute extraction device according to claim 1,
The attribute extraction apparatus characterized in that the main meaning attribute extraction means extracts occupation attribute information. The attribute extraction device according to claim 3,
The part-of-speech notation extraction unit extracts at least one attribute information of age, hobby, and preference. The attribute extraction device according to any one of claims 1 to 6,
The text data is data published on a network and is input by a user of a user profile included in the text data and associated with input text data published on the network. Attribute extraction device. The attribute extraction device according to claim 7,
An attribute extraction apparatus characterized by further comprising an analysis means for analyzing the input text data using at least one attribute information extracted from the text data. An attribute extraction method performed by an attribute extraction device,
Linguistic analysis of text data including a user profile, and assigning semantic attributes to the syntactic subject included in the text data;
When the semantic attribute assigned to the syntactic main character matches any semantic attribute pattern stored in the main character semantic attribute extraction pattern storage means, the semantic attribute that matches the noun phrase of the phrase including the syntactic main character An attribute extraction method characterized by performing a main meaning attribute extraction step for extracting as attribute information of an attribute classification corresponding to a pattern. An attribute extraction program executed by the attribute extraction device,
Attribute extraction device
Linguistic analysis of text data including a user profile, and linguistic analysis means for assigning a semantic attribute to a syntactic subject included in the text data;
A main meaning attribute extraction pattern storing means storing at least one semantic attribute pattern and a corresponding attribute classification; and
If the semantic attribute assigned to the syntactic main character matches any semantic attribute pattern stored in the main character semantic attribute extraction pattern storage means, the noun phrase of the phrase containing the syntactic main character is matched. Main meaning attribute extracting means for extracting as attribute information of attribute classification corresponding to attribute pattern,
Attribute extraction program to function as

Images