JP5442401B2 - Behavior information extraction system and extraction method - Google Patents

Description

  The present invention relates to a behavior information extraction system and an extraction method for accessing information on a consumer's purchased products and other behavior patterns by accessing a website such as a blog or SNS existing on the Internet.

  Currently, various product advertisements are broadcast on TV and radio programs, for example. However, there is no effective way to investigate whether the product advertisement really affects the consumer behavior pattern and leads to the purchase of the product. For example, even if a product questionnaire is conducted, it takes time and data There is a limit. Therefore, even if a program producer wants to produce a new program of a type similar to the program currently being broadcast, he / she does not know what kind of company or organization can make it interesting. In addition, general companies and organizations have not been able to predict with high accuracy what kind of television program or the like can be used to improve sales.

  These problems are not limited to product advertisements using broadcasting media such as televisions, but also occur in product advertisements using signboards installed in various facilities, product advertisements placed on travel brochures, concert flyers, etc. there were. If it is possible to analyze an enormous number of articles disclosed on various websites on the Internet for this problem, for example, information and data indicating the relationship between the product purchased by the consumer and the product advertisement are extracted. It will be possible to do this.

  However, consumer articles such as blogs and SNSs that describe everyday events of consumers are generally written by people who are not accustomed to writing and are spoken. For this reason, the conventional method of performing keyword extraction and dependency analysis cannot accurately extract behavior patterns such as consumer purchases. In addition, there is a problem that it is not possible to obtain information on specific actions other than those specified in the text.

  On the other hand, an apparatus and a method for analyzing relatively ambiguous sentences such as colloquial styles and extracting the opinions and intentions claimed by the person who wrote the sentences have been proposed. For example, as disclosed in Patent Document 1, a model of an opinion structure is set in advance, and an input text sentence is applied to the model, so that an object, a focus of evaluation on the object, an object There is an opinion extraction device that detects partial texts that contain four elements: the nature of the evaluation, positive / negative / neutral evaluation. In particular, this opinion extraction device uses an estimation means for estimating the meaning of a partial text sentence, even if the corresponding partial text sentence is ambiguously expressed when extracting three elements excluding the object among the above four elements. , The person's opinions can be extracted accurately.

JP 2004-157841 A

  As described above, information and data indicating the relevance of consumer purchase behavior and other consumer behavior to advertisements such as products are accurately obtained from articles conventionally disclosed on various websites on the Internet. There was no way to extract.

  Moreover, although the opinion extraction apparatus or method of patent document 1 can extract how the creator of a text sentence is evaluating the property of a target object, for example, a target object is "book" "novel" "When expressed in different hierarchical concepts such as" reason novel "or specific product names, the object itself cannot be accurately specified, and each object cannot be classified by a predetermined product category. Therefore, it is not suitable for a purpose of obtaining statistical information indicating the relevance between the consumer's product purchase behavior and the advertisement of the product.

  The present invention has been made in view of the above-described background art, and searches for articles disclosed on various websites on the Internet, and various types of information related to consumer behavior when accessing an advertising medium or the like of a specific product. It is an object of the present invention to provide a behavior information extraction system and an extraction method that can be accurately extracted in a form that can cope with various statistical processes.

The present invention accesses a website on the Internet that is constituted by a computer system and discloses consumer articles in which daily events are written by consumers, and performs a search based on a search keyword entered by a search user. A behavior information extraction system that performs processing, extracts information related to consumer behavior that matches the search keyword from the collected consumer articles, and outputs the information, and accesses the website to retrieve the consumer articles. An information collecting device having a consumer article collecting means for collecting; an input means for a search user to input a search keyword; and an output means for outputting information on consumer behavior obtained as a result of the search processing in a predetermined format. Analyzing sentences for each consumer article collected by the input / output terminal and the consumer article collection means, Search means for extracting a consumer article including a search keyword as a related article, and a narrowing means for extracting, from the related article, an article having a low ratio of including an advertising phrase in the sentence of the article as an analysis target article An analysis target article extraction means provided and an action which is a noun indicating an action target of consumer behavior from a sentence portion representing behavior information when the consumer performs some action by analyzing a sentence for each analysis target article Action object / attachment information extraction means for extracting object information and attached information that is a noun indicating a matter related to the action object, a noun indicating the action object of the action, and an action object related to the noun and is the action target dictionary formed by registering in association with each other attributes defined in a hierarchical structure nouns, nouns indicating an action target of the action, the noun indicating an associative concepts association from the noun A concept dictionary that is registered in association with attributes defined in a hierarchical structure and the action target dictionary or the concept dictionary for each analysis target article, and extracted by the action target / attached information extraction unit By extracting and combining attributes corresponding to the action target information and the attached information, specific information that specifically identifies the action target and action corresponding to the action target information is generated, and the series of processing contents described above Is an action information extraction system comprising action target information specifying means for sending to the output means of the input / output terminal.

  The action target information specifying means refers to an action dictionary in which an action name, a noun indicating the action, and a noun indicating an action related to the noun are registered in association with attributes defined in the hierarchical structure. The action target information specifying means refers to the action dictionary and selects a noun that is an action target in the action target dictionary.

  Furthermore, a behavior pattern template in which a plurality of template sentences representing behavior patterns when a consumer acts is set, and the behavior target / attached information extraction unit performs the behavior pattern on the analysis target article text. A template that refers to any template text is extracted with reference to a template, and nouns corresponding to the action target information and attached information are extracted from a specific portion of the extracted text.

  The invention also analyzes product text for each article to be analyzed, product information that is a noun indicating the product itself purchased by the consumer from the text part representing the behavior when the consumer purchases the product, and the product Product / attachment information extraction means for extracting attached information that is a noun indicating a matter related to the product, the name of the product on the market, the noun indicating the product, and the noun indicating the product associated with the noun Are defined as attributes defined in the hierarchical structure, a product dictionary that is registered in association with attributes defined in the hierarchical structure, a noun indicating a product concept, and a noun indicating a product concept associated with the noun. Corresponding to the product information and the accessory information extracted by the product / attachment information extraction means by referring to the concept dictionary registered in association with each other and the concept dictionary or the product dictionary for each analysis target article Product information specifying means for generating specific product information that specifically specifies the product corresponding to the product information by extracting and combining the characteristics, and sending the series of processing contents to the output means of the input / output terminal It is an action information extraction system provided with.

  A spam word dictionary in which spam words that are words and characters for advertisement are defined is defined, and the narrowing down means of the analysis target article extracting means includes the total number of words of the article and the spam word for each related article. An occupancy ratio that is a ratio to the number of corresponding words is calculated, and related articles whose occupancy ratio is equal to or less than a reference value are extracted as analysis target articles.

  In the narrowing-down means, a relatively small reference value is set in accordance with the total number of words of the related article when the total number of words is relatively small.

  In addition, a purchase pattern template in which a plurality of template texts representing behavior patterns when a consumer purchases a product is set is provided, and the product / attached information extraction unit is configured to purchase the purchase text for the analysis target article text. By referring to a pattern template, a sentence applicable to any of the template sentences is extracted, and nouns corresponding to the product information and the attached information are extracted from a specific portion of the extracted sentence.

  The product information specifying means refers to the product dictionary, and when a product that matches the product information of the analysis target article is registered as a product name attribute, the product information specifying unit is registered in association with the product name attribute. The specific product information is generated by extracting and combining other attribute groups, and when the product information that matches the product information is not registered in the product name attribute of the product dictionary, refer to the concept dictionary, The specific merchandise information is generated by extracting and combining attributes matching the merchandise information and the attached information and other attribute groups registered in association with the attributes.

The concept dictionary is an associative concept dictionary in which nouns indicating matters related to various products and nouns indicating matters related to products associated with the nouns are registered in association with attributes defined in the hierarchical structure. A product classification concept dictionary in which nouns indicating various products and nouns indicating products associated with the nouns are registered in association with attributes defined in the hierarchical structure, and the product information specifying The means refers to the associative concept dictionary, extracts attributes that match the attached information of the analysis target article and other attributes registered in association with the attributes as attached attribute groups, and refers to the product classification concept dictionary Then, an attribute that matches the product information of the article to be analyzed and other attributes registered in association with the attribute are extracted as a product attribute group, and the attached attribute group and the product attribute group are extracted. When there is an attribute that exists throughout, the attribute is extracted as a product specific attribute group, the specific product information is generated by combining the product specific attributes, and is common to the attached attribute group and the product attribute group when there is no attributes present, it is an intact the specific product information the product information.

In addition, the information collection device accesses a mail order site on the Internet, analyzes the hierarchical structure of each mail order site, and analyzes the lowest layer page for analyzing the bottom page of the hierarchical structure. And a mail order information collecting means comprising a product dictionary registration means for registering information analyzed by the site structure analysis means and the lowest layer page analysis means in the product dictionary, and the site structure analysis means comprises the hierarchy By acquiring and analyzing the HTML data displayed in each menu part, the hierarchical category and the product category assigned to each hierarchy are recognized, and the lowermost layer page analysis means adds to the body part of the lowermost page. The product name of the product disclosed by extracting and analyzing HTML data of the displayed product list is extracted, and the product dictionary Recording means, wherein the product category associated with the trade name of trade name and site structure analyzing means for the lowermost page analysis means has extracted recognizes, and registered in association with each other in the attributes defined in the product dictionary, When there is a new product category that is not defined as an attribute of the product dictionary, the behavior information extraction system registers the new product category in the new attribute after defining the new attribute in the product dictionary. .

In addition, the present invention accesses a Web site on the Internet where a consumer article in which a daily life event is written by a consumer is disclosed, performs a search process based on a search keyword input by a search user, and performs a search. A behavior information extracting method for extracting information related to consumer behavior that matches the search keyword from the consumer articles, wherein the search user inputs a search keyword, so that the consumer article on the Web site A search step for extracting a consumer article including the search keyword from among the related articles, and a narrowing down of extracting the articles having a low proportion of advertisement terms from the related articles as sentences to be analyzed Steps and sentences that analyze the sentence for each of the articles to be analyzed and represent behavior information when the consumer takes some action An action object / attachment information extracting step for extracting action object information, which is a noun indicating an action object of consumer behavior, and attached information, which is a noun indicating a matter related to the action object, from the portion; By extracting and combining the action target information extracted in the attached information extraction step and the attribute corresponding to the attached information, the action target corresponding to the action target information and the action are specifically identified, and the specific action target information is obtained. A behavior information extraction method comprising a behavior target information identification step to be generated.

  The action target information specifying step refers to an action dictionary in which an action name, a noun indicating the action, and a noun indicating an action related to the noun are registered in association with attributes defined in the hierarchical structure. The noun that is the action target in the action target dictionary is selected.

  The action target / attached information extraction step refers to an action pattern template in which a plurality of template sentences representing action patterns when a consumer behaves is set, A sentence applicable to the template sentence is extracted, and nouns corresponding to the action target information and the attached information are extracted from a specific portion of the extracted sentence.

  The invention also analyzes product text for each article to be analyzed, product information that is a noun indicating the product itself purchased by the consumer from the text part representing the behavior when the consumer purchases the product, and the product The product / attachment information extraction step for extracting adjunct information, which is a noun indicating matters related to the product, the noun indicating the product concept, and the noun indicating the product concept associated with the noun are defined in a hierarchical structure An attribute in which a concept dictionary that is registered in association with attributes, a product name of a commercially available product, a noun indicating the product and a noun indicating a product concept associated with the noun are defined in a hierarchical structure To extract and combine the product information extracted in the product / attachment information extraction step and the attribute corresponding to the accessory information. What is behavioral information extraction method and a commodity information specifying step of specifically identifying the product corresponding to the product information.

  The narrowing-down step refers to a spam word dictionary in which spam words that are words and characters for narrowing advertisement are defined, and for each related article, the total number of words of the article and the number of words corresponding to the spam word, Occupancy ratio, which is the ratio of, and related articles whose occupancy ratio is below a reference value are extracted as analysis target articles.

  In the narrowing-down step, when the total number of words is small, determination is performed using a relatively small reference value according to the total number of words in the related article.

  The product / attachment information extraction step refers to a template text representing a purchase pattern when a consumer purchases a product, extracts a text that applies to any of the template texts from the analysis target article, and extracts The noun corresponding to the product information and the attached information is extracted from a specific part of the sentence.

Further, the product information specifying step refers to the product dictionary, and when a product that matches the product information of the analysis target article is registered as a product name attribute, the product information specifying step is registered in association with the product name attribute. Specific product information that specifically identifies the product corresponding to the product information is generated by extracting and combining the other attribute groups, and the product name attribute in the product dictionary matches the product information When not registered in the concept dictionary, the specific product information is obtained by extracting and combining the attribute matching the product information and the attached information and other attribute groups registered in association with the attribute with reference to the concept dictionary Is generated.

In addition, the concept dictionary is an associative concept in which nouns indicating matters relating to various products and nouns indicating matters relating to products associated with the nouns are registered in association with attributes defined in the hierarchical structure. The product includes a dictionary and a product classification concept dictionary in which nouns indicating various products and nouns indicating products associated with the nouns are registered in association with attributes defined in a hierarchical structure. The information specifying step refers to the associative concept dictionary, extracts attributes corresponding to the attached information of the analysis target article and other attributes registered in association with the attributes as attached attribute groups, and the product classification concept dictionary , The attribute that matches the product information of the article to be analyzed and other attributes registered in association with the attribute are extracted as a product attribute group, and the attached attribute group and the product When there is an attribute that exists in common with the sex group, the attribute is extracted as a product specific attribute group, the specific product information is generated by combining the product specific attributes, and the attached attribute group and the product attribute when there is no attribute that is present commonly in the group are those as the specific product information the product information.

A site structure analyzing step of accessing a mail order site on the Internet and analyzing a hierarchical structure of each mail order site; a bottom layer page analyzing step of analyzing a bottom layer page of the hierarchical structure; and the site structure A product dictionary registration step for registering the information analyzed in the analysis step and the lowermost layer page analysis step in the product dictionary, and comprising a mail order information collection step performed in advance before the search step, The structural analysis step recognizes the hierarchical structure and the product category assigned to each tier by acquiring and analyzing the HTML data displayed in the menu portion for each tier, and the bottom layer page analyzing step includes: Acquire HTML data of a list of products displayed in the body part of the lowermost page and The product was extracted trade name of the product disclosed by analysis, the product dictionary registering step associated with the product name the product name and site structure analysis step of the lowermost page analysis step is extracted recognizes Categories are registered in association with attributes defined in the product dictionary, and when there is a new product category that is not defined as an attribute of the product dictionary, after defining a new attribute in the product dictionary, This is a behavior information extraction method for registering the new product category in the new attribute.

  The behavior information extraction system and the extraction method according to the present invention automatically acquire information on purchasing behavior and other behaviors of consumers who have accessed an advertising medium such as a specific product such as a viewer of a television program. It can be extracted with high accuracy. Then, by analyzing the statistical information obtained by the behavior information extraction system and the extraction method of the present invention, for example, the creator of the advertising medium can easily know companies and industries suitable as sponsors. In addition, it is very convenient for a sponsoring company or the like to easily know an advertising medium having a large advertising effect on consumers.

  In particular, by analyzing consumer purchases using product and concept dictionaries that define product features in a hierarchical manner, individual information extracted for each article can be converted into, for example, “books”, “novels”, “inferences”. It can be classified by an arbitrary hierarchical concept such as “novel”. Therefore, statistical information suitable for the purpose of the search user can be obtained freely by aggregating individual information in various modes.

  In addition, since information about specific products on the market is registered in the product dictionary with appropriate attributes, even if the product information extracted from the consumer article is the product name, it is easy to purchase products. Can be specified.

  In addition, information on new products that increase daily can be obtained periodically from Internet mail order sites and automatically registered in product dictionaries, etc., which is sufficient for changes in market trends and trends. Surveys corresponding to can be conducted.

It is a block diagram which shows the structure of one Embodiment of the action information extraction system of this invention. It is a flowchart explaining operation | movement of this embodiment. It is a figure explaining the detection keyword used in this embodiment, and the related article group extracted. It is a flowchart explaining step 12 of FIG. It is a figure explaining the operation | movement of the spam word dictionary of this embodiment, and the narrowing-down means. It is a figure explaining the purchase pattern template of this embodiment, the extracted attached information, and merchandise information. It is a flowchart explaining step S14 of FIG. It is a figure explaining operation | movement of the goods dictionary and goods information specific | specification means of this embodiment. It is a figure explaining operation | movement of the associative concept dictionary of this embodiment, and a merchandise information identification means. It is a figure explaining operation | movement of the goods classification concept dictionary and goods information specific | specification means of this embodiment. It is a figure explaining the operation | movement in which the goods information specific | specification means of this embodiment produces | generates specific goods information. It is a figure which shows an example of the statistical information which the output means of this embodiment outputs. It is a block diagram which shows the modification of an information collection device. It is a flowchart explaining the operation | movement in which the information collection apparatus of this modification registers new goods in a goods dictionary. It is a flowchart explaining step S21 of FIG. It is a flowchart explaining step S22 of FIG. It is a flowchart explaining step S23 of FIG. It is a block diagram which shows the structure of other embodiment of the action information extraction system of this invention.

  Hereinafter, an embodiment of the behavior information extraction system of the present invention will be described with reference to the drawings. The behavior information extraction system of this embodiment extracts information about the behavior of purchasing a product by a consumer. The purchased product information extraction system 10 of this embodiment is configured by a computer system, and is a system that extracts information about a product purchased by a general consumer who has watched a specific television program or a product (purchased product) that is about to be purchased. is there.

  As shown in FIG. 1, the purchased product information extraction system 10 searches for various websites disclosed on the Internet and collects document information about a predetermined type of website for which information is collected. An information collecting device 16 having consumer article collecting means 12 for collecting articles is provided. Further, based on the input / output terminal 18 operated by the search user and the search keyword input by the search user, an analysis target for extracting an article to be analyzed from the consumer articles collected by the consumer article collection means 12 Article extraction means 20 is provided. Further, a product / attached information extracting means 22 for extracting information for specifying a purchased product from the text of a specific analysis target article, and a product for specifying a purchased product based on the information extracted by the product / attached information extracting means 22 Information specifying means 24 is provided. Furthermore, a spam word dictionary 26, a purchase pattern template 28, a product dictionary 30, and a concept dictionary 32 that are used when each of the above-described means performs predetermined processing are provided.

  The consumer article collection means 12 collects consumer articles in which consumers disclose daily events around them. The collection of consumer articles may be performed at the timing when the search user operates the input / output terminal 18, but in order to shorten the search user's search time, the system administrator periodically collects the information in advance to collect the information. It is preferable to accumulate in

  The input / output terminal 18 includes an input unit 18a for a search user to input a search keyword, and an output unit 18b for outputting information about a purchased product as a search result. Here, the search keyword is a specific television program name.

  The analysis target article extraction means 20 accesses the consumer article collection means 12 and the search means 20a for extracting the related articles corresponding to the search keyword input to the input means 18a, and the seller seems to have written from the related articles. A narrowing means 20b for extracting analysis target articles excluding articles is provided. When the narrowing-down means 20b narrows down articles, the spam word dictionary 26 in which the words used by the merchant of the product as the words for advertisement are defined as spam words is referred to. Then, an article that is subject to analysis is extracted by excluding articles that have a high proportion of advertising terms in the entire related article text and extracting only articles that the consumer seems to have written.

  The product / attached information extracting unit 22 applies each analysis target article to a purchase pattern template 28 in which a plurality of template sentences representing behavior patterns when a consumer purchases a product is set, and the article / attachment information extracting unit 22 applies to each template sentence. The corresponding sentence is extracted, and nouns corresponding to the product information and the attached information are extracted from the specific part of the extracted sentence.

  The product dictionary 30 associates product names of various commercially available products, nouns indicating the products, and nouns indicating products associated with the nouns with attributes defined in a tree-like hierarchical structure. Registered database.

  The concept dictionary 32 includes an associative concept dictionary 32a and a merchandise classification dictionary 32b. The associative concept dictionary 32a is a database in which nouns indicating matters related to various products and nouns indicating matters related to products associated with the noun are registered in association with attributes defined in a tree-like hierarchical structure. It is. The product classification concept dictionary 32b is a database in which nouns indicating various products and nouns indicating products associated with the nouns are registered in association with attributes defined in a tree-like hierarchical structure. .

  The merchandise information specifying means 24 refers to the merchandise dictionary 30 and the concept dictionary 32 based on the merchandise information and the accessory information extracted by the merchandise / attached information extracting means 22, extracts and combines attributes related to the merchandise information, and the merchandise Specific product information that specifically identifies the information is generated. The analysis target article extracting unit 20, the product / attachment information extracting unit 22, and the product information specifying unit 24 serve to send information indicating a series of processing contents to the output unit 18b of the input / output terminal 18.

  Next, the operation | movement and the extraction method in which the purchased merchandise information extraction system 10 acquires the information regarding purchased merchandise are demonstrated based on the flowchart of FIG. First, the search user inputs a specific television program name as a search keyword into the input means 18a (step S10).

  When the TV program name is input, the search means 20a accesses the consumer article collection means 12, searches the sentence of the consumer article individually in full text, and relates the consumer article including the TV program name as a character string. An article is extracted (step S11). For example, when the search user inputs the television program name “Leisurely Travel Diary”, articles 1, 2, and 4 are extracted as related articles from a large number of consumer articles as shown in FIG. Similarly, when the TV program name “Live & Live” is input, Article 3 is extracted as a related article.

  Next, the narrowing down means 20b extracts an analysis target article that is considered to have been written by the consumer from related articles (step S12). Hereinafter, the detailed process of step S12 is demonstrated using the flowchart of FIG.

  First, the sentence of the related article is divided into words by morphological analysis (step S121). Then, for each related article, the total word number α and the spam word number β, which is the number of words corresponding to the spam word defined in the spam word dictionary 26, are acquired (step S122). In the spam word dictionary 26, as shown in FIG. 5, spam words such as “free”, “sales”, and “topic”, which are generally used as advertising term phrases, are registered in advance. The definition of the spam word does not need to be strict, and here, the system administrator sets it based on an empirical rule or the like. For example, in the case of article 1, since the total number of words α = 13 and no spam word is included therein, the number of spam words β = 0. On the other hand, in the case of article 4, since the total number of words α = 20 and the spam words “topic”, “noisy”, and “bulletin board” are included therein, the number of spam words β = 3.

  Next, a reference value γ for determining the level of spam word occupancy (β / α) is selected (step S123). Here, the reference value γ is selected to be a constant value of 10%. In the case of a sentence with a small total word number α, there is a tendency that it is determined that the occupancy (β / α) is high with only a small number of spam words β, so when the total word number α is small, It is preferable to adjust so that a relatively small value is automatically selected as the reference value γ.

  Next, the spam word occupancy (β / α) is calculated and the reference value γ is compared (step S124). When the spam word occupancy (β / α) is lower than the reference value γ, It is determined that the related article is useful because it seems to be an article written by a consumer, and is handled as an analysis target article (step S125). On the other hand, if the spam word occupancy (β / α) is higher than the reference value γ, it is determined that the relevant article is not useful because it seems to be an article written by the seller, Exclude (step S126).

  Next, returning to the flowchart of FIG. 2, the analysis target article extracted in step S <b> 12 is sent to the product / attached information extraction means 22 for analysis. Upon receiving the analysis target article, the product / attachment information extraction unit 22 applies the sentence of the analysis target article to the purchase pattern template 28 and extracts the product information and the accessory information (step 13). In the purchase pattern template 28, as shown in FIG. 6, a consumer purchases a product such as “Bought △△△ for XXX” or “Go to XXX to buy △△△”. A plurality of template sentences indicating the action pattern of the hour are registered. Then, the sentence of the analysis target article is applied to the template sentence, the noun corresponding to “ΔΔΔ” is extracted as product information, and the noun corresponding to “XXX” is extracted as attached information. For example, in the case of article 1, the sentence “Bought“ 100 famous castles in Japan ”for travel” is applied to the template sentence T1, and product information “100 selected castles in Japan” and attached information “travel” are extracted. To do. Similarly, in the case of article 3, the sentence “go to buy a keyboard to the BBB musical instrument store” applies to the template sentence T3, and product information “keyboard” and attached information “BBB musical instrument store” are extracted.

  Next, the product information and attached information extracted for each analysis target article are sent to the product information specifying means 24 and analyzed. Upon receiving the product information and the attached information, the product information specifying unit 24 generates specific product information that specifically specifies the product information (step S14). Hereinafter, the detailed processing of step S14 will be described using the flowchart of FIG.

  First, the product dictionary 30 is referenced to check whether the product information of the analysis target article is registered as an attribute (product name) (step S141). For example, as shown in FIG. 8, the product dictionary 30 includes a product name “Bakumatsu and I”, a noun indicating the product and nouns indicating associated products, “book”, “essay”, “history”, “geography”. Are registered in association with attributes defined in a tree-like hierarchical structure. Similarly, for the product name “100 famous castles in Japan”, the attributes “book”, “guidebook”, and “travel” are registered in association with each other.

  For example, in the case of article 1, since the product information is “100 selections of Japanese famous castles”, it is registered in the attribute (product name) of the product dictionary 30. Therefore, “book”, “guidebook”, and “travel” are extracted as the product identification attributes for “100 famous castles in Japan” (step S142). Then, by combining these product specifying attributes, the product information “100 famous castles in Japan” of the article 1 is specified as a travel guidebook, and specific product information “travel guidebook” is generated (step S143). ). On the other hand, in the case of article 2, since the product information is “guidebook” and is not registered in the attribute (product name) of the product dictionary 30, the process proceeds to step S144.

  When the product information is not registered in the product dictionary 30 as the attribute (product name) as in the article 2, the association concept dictionary 32a is referred to check whether the attribute information is registered in any attribute. If so, other attributes associated with the attribute are also extracted as attached attributes (step S144). For example, as shown in FIG. 9, the association concept dictionary 32 a is a noun indicating a matter related to a product “travel” and a noun indicating a matter related to a product “destination” “location” “location” “Japan”. Are registered in association with attributes defined in a tree-like hierarchical structure. Similarly, there is also an association of attributes “travel”, “transportation means”, and “airplane”.

  For example, in the case of article 2, since the attached information is “Kyushu”, the attached attributes “travel”, “destination”, “place name”, and “Kyushu” are extracted. On the other hand, in the case of article 3, since the attached information is “BBB musical instrument store”, it is not registered as an attribute in the associative concept dictionary 32a, and there is no attached attribute.

  Next, with reference to the merchandise classification concept dictionary 32b, it is checked whether or not merchandise information is registered in any attribute. If registered, other attributes associated with the attribute are also extracted as merchandise attributes. (Step S145). For example, as shown in FIG. 10, the product classification concept dictionary 32b includes a noun indicating a product “book” and a noun indicating a product associated with “book” such as “guidebook” and “travel”. Are registered in association with attributes defined in a hierarchical structure. Similarly, there is a correspondence of “book”, “novel”, and “inference”.

  For example, in the case of article 2, since the product information is “guide book”, product attributes “book”, “guide book”, “travel”, “television program”, “medicine”, and “medical / welfare” are extracted. On the other hand, in the case of article 3, since the product information is “keyboard”, it is not registered as an attribute in the product classification concept dictionary 32b, and there is no product attribute.

  Next, the attached attribute extracted in step S144 is compared with the product attribute extracted in step S145 to check whether there is an attribute that is common to both attributes (step S146). For example, in the case of article 2, as shown in FIG. 11A, since the attribute “travel” is common, “travel” is extracted as a product specifying attribute (step S147). Then, the product information “guide book” is specified as a travel guide book by the product specification attribute “travel”, and the specific product information “travel guide book” is generated (step S143). On the other hand, in the case of article 3, since there are no product attribute and attached attribute, the process proceeds to step S148.

  When there is no attribute that is common to both the attached attribute and the product attribute as in the article 3, as shown in FIG. 11B, the product information “keyboard” is directly used as the specific product information “keyboard”. To do. In other words, the specific product information “keyboard” includes a keyboard which is an electronic musical instrument and a keyboard for a personal computer, which will be treated as similar information without distinction thereafter. If the system administrator determines that this handling is not preferable, the definition can be easily improved by correcting the definition of the attribute related to the “keyboard” in the concept dictionary 32.

  Next, returning to the flowchart of FIG. 2, the analysis contents for each consumer article performed in steps S11 to S14 are sent from the product information specifying means 24 to the output means 18b, and the output means 18b converts the analysis contents into a predetermined format. Are arranged and output (step S15). For example, when the search user wants to know the purchased product of the viewer of the TV program “Leisurely Travel Diary”, the output means 18b corresponds to “Leisurely Travel Diary” as shown in FIGS. Statistical information obtained by aggregating the number of articles to be analyzed and specific product information acquired from the articles can be output. The output format can be freely changed by the search user operating the input / output terminal 18.

  As described above, the purchased product information extraction system 10 and the extraction method exclude the noise information that is not information from a personal site such as a commercial website by the spam word dictionary 26, and further, consumers By using a purchase pattern template 28 and various dictionaries specially designed for searching for products purchased by customers, information on products purchased by consumers who have watched TV programs is automatically extracted with high accuracy. be able to.

  Moreover, in order to analyze the consumer's purchased products using the product dictionary 30 and the concept dictionary 32 defined by hierarchizing product characteristics, individual information for each consumer article is extracted in a systematically organized form. can do. Therefore, for example, when a plurality of pieces of individual information related to a book are aggregated, they can be classified by an arbitrary hierarchical concept such as “book”, “novel”, “reasonable novel”, or “detective novel reasoning novel”. Thereby, the output means 18b can total the individual information regarding the obtained purchased product in various modes, and can freely output statistical information suitable for the purpose of the search user.

  Moreover, since the information of the specific product marketed is given the appropriate attribute and registered in the product dictionary 30, even if the product information extracted from the consumer article is a specific product name, The purchased product can be easily identified.

  Then, by analyzing the statistical information obtained by the purchased product information extraction system 10 and the extraction method, producers of product advertisement media such as TV programs can easily know companies and industries suitable as sponsors. it can. In addition, a sponsor company can easily know a product advertisement medium having a large advertising effect on consumers.

  Next, an information collection device 40, which is a modification of the information collection device 16 of the purchase product information extraction system 10 described above, will be described with reference to FIGS. In addition to the consumer article collection means 12 that collects consumer articles on the Internet, the information collection device 40 accesses the Internet mail order site, collects information on commercially available products, and newly registers in the product dictionary 30. A mail order information collecting device 42 is provided.

  The mail order information collecting device 42 includes a site structure analyzing unit 42a for analyzing the hierarchical structure of each mail order site, a lowermost layer page analyzing unit 42b for analyzing the lowermost page of the hierarchical structure, and a site structure analyzing unit 42a. The product dictionary registration unit 42c registers the information analyzed by the lowest layer page analysis unit 42b in the product dictionary 30.

  Next, the operation of the mail order information collecting apparatus 42 will be described based on the flowchart of FIG. First, the site hierarchical structure analyzing means 42a accesses the mail order site (step S20). Then, the site structure is analyzed focusing on the display of the menu portion, and the product category assigned to each hierarchy is recognized (step S21). Hereinafter, the detailed processing of step S21 will be described using the flowchart of FIG.

  First, HTML (Hyper Text Markup Language) data displayed in the menu part is acquired (step S211). Next, in the HTML data, a character string surrounded by a predetermined tag is extracted as a product category candidate (step S212). For example, a noun that is a character string surrounded by <div> tags, <li> tags, or the like may be used as a product category candidate. And the character string which deleted the character string which shows the size and color of goods from the character string of a goods category candidate is extracted, and it recognizes with the information of the said hierarchical structure as a goods category of the said hierarchy (step S213).

  Next, returning to the flowchart of FIG. 14, the bottom layer page analyzing means 42b analyzes the bottom layer page of the hierarchical structure recognized in step S21, and extracts the product names of the sold products (step S22). Hereinafter, the detailed processing of step S22 will be described using the flowchart of FIG.

  First, the HTML data of the product list displayed in the body part of the lowest page is acquired (step S221). Next, a noun that is a character string surrounded by a predetermined tag is extracted from the HTML data as a linked product image. For example, a character string surrounded by an <image> tag or an <a> tag for setting a link to another page may be used as a linked product image. Next, a character string in which the preceding and following character strings are highlighted is selected from the character strings of the linked product image (step S223). For example, a character string enclosed by <h1> tag, <font> tag, or the like is applicable. And the noun which is a character string which deleted the character string which shows the size, model number, etc. of a product from the character string of the product image with a link is extracted and recognized as a product name (step S224).

  Next, returning to the flowchart of FIG. 14, the merchandise dictionary registration means 42c registers the new merchandise name and merchandise category recognized in steps S21 and S22 in the merchandise dictionary 30 (step S23). Hereinafter, the detailed processing of step S23 will be described using the flowchart of FIG.

  First, the extracted new product name is registered in the attribute (product name) of the product dictionary 30 shown in FIG. 8 (step S231). Next, it is checked whether the product category associated with the product name is defined in another attribute of the product dictionary 30 (step S232). If defined, the product category is associated with the product name and registered in the corresponding attribute (step S233). On the other hand, if not defined, an attribute corresponding to the product dictionary 30 is newly defined (step S234), a product category is associated with the product name, and registered in the newly defined attribute (step 235).

  In this way, the purchased product information collection device 10 provided with the information collection device 40 periodically obtains information about new product information that increases daily from the Internet mail order website, and the product dictionary 30 automatically. It is possible to conduct surveys that fully respond to changes in market trends and trends.

  The behavior information extraction system and the extraction method of the present invention are not limited to the above-described embodiment. For example, in addition to purchasing a product, an action of going somewhere to see some action target for the purpose of sightseeing or appreciation. It is also possible to extract actions such as going on a trip with family and friends. Hereinafter, the behavior information extraction system 50, which is an embodiment including other behaviors in addition to the purchase operation described above, will be described based on the block diagram of FIG. Here, the same components as those in the above embodiment are denoted by the same reference numerals, and description thereof is omitted.

  The behavior information extraction system 50 of this embodiment includes the configuration of the purchased product information extraction system 10 of the above embodiment, and extracts various behavior information including product purchase behavior from the text of a specific analysis target article. To do. The behavior information extraction system 50 extracts behavior target / attached information for extracting information for specifying a behavior target in behavior information extraction (in the present invention, a person or an abstract thing is included in addition to an object). Means 52 and action target information specifying means 54 for specifying the action target based on the information extracted by the action target / attached information extracting means 52 are provided.

  The action target / attached information extracting means 52 includes the product / attached information extracting means 22 and is provided so as to be able to refer to the action pattern template 56 including the purchase pattern template 28 as in the above embodiment. The behavior pattern template 56 includes a template of a pattern that recognizes some object and acts, such as “I went to to see” or “I went to”. Then, the action target / attached information extraction unit 52 applies each analysis target article to the action pattern template 56 in which a plurality of template sentences representing the action patterns when the consumer acts are set, and any of the template sentences Sentences that fall under are extracted, and nouns corresponding to action target information and attached information are extracted from specific parts of the extracted sentences.

  The action target information specifying unit 54 includes the product information specifying unit 24, and refers to various dictionaries as in the above embodiment. The dictionary referred to by the action target information specifying unit 54 includes an action target dictionary 58 including the product dictionary 30 in addition to the product dictionary 30 of the above embodiment. Similar to the product dictionary 30, the action target dictionary 58 has a tree-like hierarchical structure in which names of various recognized objects that can be action objects, nouns indicating the recognized objects, and nouns associated with the nouns. This is a database registered in association with the attributes defined in.

  The concept dictionary 32 includes an associative concept dictionary 32a and an action target classification concept dictionary 32c including a product classification concept dictionary 32b. The line animal classification concept dictionary 32c is a database in which nouns indicating various recognized objects and nouns indicating things associated with the nouns are registered in association with attributes defined in a tree-like hierarchical structure. .

  Further, the action target information specifying means 54 is provided with an action dictionary 60 in which nouns indicating various actions are recorded so as to be referred to. The behavior dictionary 60 is configured by registering an action name, a noun indicating the action, and a noun indicating an action related to the noun in association with attributes defined in the hierarchical structure.

  In the analysis target article extraction unit 20 of this embodiment, as in the above embodiment, the narrowing unit 20b defines a word or phrase used as a promotion word by a merchant that handles products or other action targets as a spam word. By referring to the spam word dictionary 26, articles having a high ratio of advertising terms in the entire related article text are excluded, and only the articles that the consumer seems to have written are extracted by the action target / attached information extraction means 52. , Subject to analysis.

  Then, the action target information specifying unit 54 refers to the action dictionary 60, the action target dictionary 58, and the concept dictionary 32 based on the action target information extracted by the action target / attached information extraction unit 52, the attached information, and the like. The attribute related information is extracted and combined, and the specific action target information that specifically specifies the action target information and the action is generated. Furthermore, the above-mentioned analysis target article extracting means 20, action target / attached information extracting means 52, and action target information specifying means 54 send information indicating a series of processing contents to the output means 18b of the input / output terminal 18. To do. Thereby, in addition to the purchase of merchandise, consumer behavior patterns can be extracted and analyzed for various other consumer behaviors.

  The behavior information extraction system and the extraction method of the present invention provide information on the behavior of consumers who have accessed various advertising media, such as viewers of television and radio programs, subscribers of magazines, users of various facilities, and the like. In the extraction system or extraction method, the action dictionary, the action target dictionary, the concept dictionary, and the like may be provided with a plurality of dictionaries having different contents suitable for the respective uses.

  Further, the merchandise dictionary registration means of the mail order information collection device converts the character string of the merchandise category obtained by analyzing the mail order site using the thesaurus dictionary, and the converted character string is already attributed to the merchandise dictionary. May be registered in the product dictionary instead of the product category. In addition, the hierarchical structure of the product classification concept dictionary and the like does not necessarily match the hierarchical structure of the attributes of the product dictionary, but the same configuration is used for the other attributes except the product dictionary attribute (product name). Also good. By doing so, the merchandise classification concept dictionary and the merchandise dictionary are updated while maintaining the same performance, so the quality and performance of the purchased merchandise information extraction and the like are stabilized, and maintenance by the system administrator is facilitated.

DESCRIPTION OF SYMBOLS 10 Purchase goods information extraction system 12 Consumer article collection means 16 Information collection apparatus 18 Input / output terminal 18a Input means 18b Output means 20 Analysis object article extraction means 20a Search means 20b Narrowing means 22 Goods / attached information extraction means 24 Product information specification Means 26 Spam word dictionary 28 Purchase pattern template 30 Product dictionary 32 Concept dictionary 32a Association concept dictionary 32b Product classification concept dictionary 40 Information collection means 42 Mail order information collection device 42a Site hierarchy structure analysis means 42b Bottom layer page analysis means 42c Product dictionary registration Means 52 Action target / attached information extraction means 54 Action target information specifying means 56 Action pattern template

Claims (20)

It is configured by a computer system, accesses a website on the Internet where a consumer article in which daily events are written by a consumer is disclosed, performs a search process based on a search keyword input by a search user, In the behavior information extraction system that extracts and outputs information on consumer behavior that matches the search keyword from the collected consumer articles,
An information collection device having consumer article collection means for accessing the website and collecting the consumer articles;
An input / output terminal having an input means for a search user to input a search keyword, and an output means for outputting information relating to consumer behavior obtained as a result of the search processing in a predetermined format;
A sentence is analyzed for each consumer article collected by the consumer article collection means, and a search means for extracting a consumer article containing the search keyword in the sentence as a related article; Analytical article extraction means provided with a narrowing means for extracting as articles to be analyzed those articles that contain a low proportion of advertising terms in the text,
Analyzing sentences for each of the analysis target articles, from the sentence part representing the behavior information when the consumer performs some action, the action target information that is a noun indicating the action target of the consumer action and the action target Action object / attachment information extracting means for extracting attached information that is a noun indicating a matter;
An action object dictionary in which a noun indicating an action object of the action and a noun indicating an action object related to the noun are registered in association with attributes defined in a hierarchical structure;
A concept dictionary in which a noun indicating an action target of the action and a noun indicating an associative concept associated with the noun are registered in association with attributes defined in a hierarchical structure;
By referring to the action target dictionary or the concept dictionary for each analysis target article and extracting and combining the action target information extracted by the action target / attached information extraction unit and the attribute corresponding to the attached information, Action target information specifying means for generating specific information that specifically specifies the action target and action corresponding to the action target information and sending the above-described series of processing contents to the output means of the input / output terminal An action information extraction system characterized by   The action target information specifying means refers to an action dictionary in which an action name, a noun indicating the action, and a noun indicating an action related to the noun are registered in association with attributes defined in the hierarchical structure. 2. The behavior information extraction system according to claim 1, wherein the behavior target information specifying unit is configured to select a noun that is a behavior target in the behavior target dictionary with reference to the behavior dictionary. It has an action pattern template with multiple template sentences that represent the behavior patterns when consumers act,
The action target / attached information extraction unit extracts a sentence that applies to any of the template sentences with reference to the action pattern template with respect to the sentence of the analysis target article, and extracts the sentence from a specific portion of the extracted sentence. The behavior information extraction system according to claim 1, wherein nouns corresponding to the behavior target information and attached information are extracted. Analyzing the sentence for each analysis target article, from the sentence part representing the behavior when the consumer purchases the product, product information which is a noun indicating the product itself purchased by the consumer, and matters related to the product Product / attachment information extracting means for extracting attached information, which is a noun to indicate;
A product dictionary in which a product name of a commercially available product, a noun indicating the product and a noun indicating a product associated with the noun are registered in association with attributes defined in the hierarchical structure;
A concept dictionary in which a noun indicating a product concept and a noun indicating a product concept associated with the noun are registered in association with attributes defined in the hierarchical structure;
Refer to the concept dictionary or the product dictionary for each article to be analyzed, and correspond to the product information by extracting and combining the product information extracted by the product / attached information extraction unit and the attribute corresponding to the attached information The behavior information extraction system according to claim 1, further comprising: product information specifying means for generating specific product information that specifically specifies a product to be sent and sending the series of processing contents to the output means of the input / output terminal. . It has a spam word dictionary with defined spam words, which are promotional words and characters,
The narrowing down means of the analysis target article extracting means calculates an occupancy ratio, which is a ratio between the total number of words of the sentence of the article and the number of words corresponding to spam words for each related article, and the occupancy ratio is a reference The behavior information extraction system according to claim 1, wherein a related article having a value less than or equal to a value is extracted as an analysis target article.   The behavior information extraction system according to claim 1 or 4, wherein the narrowing means is set with a relatively small reference value when the total number of words is relatively small, according to the total number of words of the related article. . It has a purchase pattern template with multiple template sentences representing behavior patterns when consumers purchase products,
The product / attachment information extracting means extracts a sentence that matches one of the template sentences with reference to the purchase pattern template with respect to the sentence of the analysis target article, and extracts the commodity from a specific portion of the extracted sentence. The behavior information extraction system according to claim 4, wherein a noun corresponding to the information and the attached information is extracted. The product information specifying means includes
When the product dictionary matches the product information of the analysis target article with reference to the product dictionary and is registered as a product name attribute, another attribute group registered in association with the product name attribute is extracted. The specific product information is generated by combining
When the product information that matches the product information is not registered in the product name attribute of the product dictionary, the product dictionary and the attribute that matches the product information and the attached information are registered with reference to the concept dictionary. The behavior information extraction system according to claim 4, wherein the specific product information is generated by extracting and combining other attribute groups. The concept dictionary is
An associative concept dictionary in which nouns indicating matters related to various products and nouns indicating matters related to products associated with the noun are registered in association with attributes defined in the hierarchical structure;
Composed of a product classification concept dictionary in which nouns indicating various products and nouns indicating products associated with the noun are registered in association with attributes defined in the hierarchical structure,
The product information specifying means includes
Referencing the associative concept dictionary, extracting attributes that match the attached information of the analysis target article and other attributes registered in association with the attributes as an attached attribute group,
Refer to the product classification concept dictionary, extract the attribute that matches the product information of the analysis target article and other attributes registered in association with the attribute as a product attribute group,
When there is an attribute that exists in common in the attached attribute group and the product attribute group, the attribute is extracted as a product specific attribute group, and the specific product information is generated by combining the product specific attributes,
Wherein when no attributes present in common in the attached attribute group and the product attribute group behavior information extraction system according to claim 4 or 8, wherein the intact the specific product information the product information. The information collecting apparatus accesses a mail order site on the Internet and analyzes a site structure analyzing means for analyzing the hierarchical structure of each mail order site, and a lowermost layer page analyzing means for analyzing a lowermost page of the hierarchical structure. The mail order information collecting means comprising the product dictionary registration means for registering the information analyzed by the site structure analysis means and the lowest layer page analysis means in the product dictionary,
The site structure analysis means recognizes the product category assigned to each hierarchical structure and each hierarchy by acquiring and analyzing the HTML data displayed in the menu portion for each hierarchy,
The bottom layer page analysis means extracts the product name of the product disclosed by acquiring and analyzing the HTML data of the product list displayed in the body part of the page of the bottom layer,
The product dictionary registration unit associates the product name extracted by the bottom layer page analysis unit and the product category related to the product name recognized by the site structure analysis unit with the attributes defined in the product dictionary. 5. When there is a new product category that is registered and is not defined as an attribute of the product dictionary, the new product category is registered in the new attribute after the new attribute is defined in the product dictionary. Or the action information extraction system of 8 description. By accessing a website on the Internet where a consumer article containing daily events written by the consumer is disclosed, a search process is performed based on the search keyword entered by the search user, and the searched consumer article In the behavior information extraction method for extracting information on consumer behavior that matches the search keyword from the inside,
A search step of extracting a consumer article including the search keyword as a related article from consumer articles on the website by the search user inputting a search keyword;
From the related articles, a narrowing step of extracting, as an analysis target article, a thing with a low ratio of including an advertising phrase in the sentence of the article;
Analyzing sentences for each of the analysis target articles, from the sentence part representing the behavior information when the consumer performs some action, the action target information that is a noun indicating the action target of the consumer action and the action target An action object / attachment information extraction step for extracting ancillary information that is a noun indicating a matter;
By specifically extracting and combining the action target information extracted in the action target / attached information extraction step and the attribute corresponding to the attached information, the action target and the action corresponding to the action target information are specifically identified and specified. An action information extraction method comprising: an action target information specifying step for generating action target information. The action target information specifying step refers to an action dictionary in which an action name, a noun indicating the action, and a noun indicating an action related to the noun are registered in association with attributes defined in the hierarchical structure. The action object in the action object dictionary in which the noun indicating the action object of the action and the noun indicating the action object related to the noun are registered in association with each other with the attributes defined in the hierarchical structure. The behavior information extracting method according to claim 11, wherein a noun is selected.   The action target / attached information extraction step refers to an action pattern template in which a plurality of template sentences representing action patterns when a consumer behaves is set, The behavior information extraction method according to claim 11, wherein a sentence applicable to the template sentence is extracted, and nouns corresponding to the action target information and the attached information are extracted from a specific portion of the extracted sentence. Analyzing the sentence for each analysis target article, from the sentence part representing the behavior when the consumer purchases the product, product information which is a noun indicating the product itself purchased by the consumer, and matters related to the product A product / attachment information extraction step for extracting attached information that is a noun to be indicated;
A concept dictionary in which a noun indicating a product concept and a noun indicating a product concept associated with the noun are registered in association with attributes defined in the hierarchical structure, and a product name of a commercially available product A product dictionary in which a noun indicating the product and a noun indicating a product concept associated with the noun are registered in association with attributes defined in a hierarchical structure, and the product / attachment information extracting step The behavior information according to claim 11, further comprising: a product information specifying step for specifically specifying a product corresponding to the product information by extracting and combining the attributes corresponding to the product information extracted in step 1 and the attached information. Extraction method.   The narrowing-down step refers to a spam word dictionary in which spam words that are words and characters for narrowing advertisement are defined, and for each related article, the total number of words of the article and the number of words corresponding to the spam word, 15. The behavior information extraction method according to claim 11 or 14, wherein an occupancy ratio that is a ratio of the occupancy ratio is calculated, and related articles whose occupancy ratio is a reference value or less are extracted as analysis target articles.   15. The behavior information extraction method according to claim 11 or 14, wherein the narrowing-down step performs determination using a relatively small reference value when the total number of words is small according to the total number of words of the related article.   The product / attachment information extraction step refers to a template text representing a purchase pattern when a consumer purchases a product, extracts a text that applies to any of the template texts from the analysis target article, and extracts The behavior information extraction method according to claim 14, wherein nouns corresponding to the product information and the attached information are extracted from a specific part of the sentence. The product information specifying step includes:
When the product dictionary matches the product information of the analysis target article with reference to the product dictionary and is registered as a product name attribute, another attribute group registered in association with the product name attribute is extracted. To generate specific product information that specifically identifies the product that corresponds to the product information,
When the product information that matches the product information is not registered in the product name attribute of the product dictionary, the product dictionary and the attribute that matches the product information and the attached information are registered with reference to the concept dictionary. The behavior information extraction method according to claim 14, wherein the specific product information is generated by extracting and combining other attribute groups. The concept dictionary is
An associative concept dictionary in which nouns indicating matters related to various products and nouns indicating matters related to products associated with the noun are registered in association with attributes defined in the hierarchical structure;
Composed of a product classification concept dictionary in which nouns indicating various products and nouns indicating products associated with the noun are registered in association with attributes defined in the hierarchical structure,
The product information specifying step includes:
Referencing the associative concept dictionary, extracting attributes that match the attached information of the analysis target article and other attributes registered in association with the attributes as an attached attribute group,
Refer to the product classification concept dictionary, extract the attribute that matches the product information of the analysis target article and other attributes registered in association with the attribute as a product attribute group,
When there is an attribute that exists in common in the attached attribute group and the product attribute group, the attribute is extracted as a product specific attribute group, and the specific product information is generated by combining the product specific attributes,
Wherein when no attributes present in common in the attached attribute group and the product attribute group, behavioral information extraction method according to claim 14 or 18 wherein the directly the specific product information the product information. A site structure analysis step for accessing a mail order site on the Internet and analyzing the hierarchical structure of each mail order site, a bottom layer page analysis step for analyzing the bottom layer page of the hierarchical structure, and the site structure analysis step And a product dictionary registration step for registering the information analyzed in the lowermost layer page analysis step in the product dictionary, comprising a mail order information collection step performed in advance before performing the search step,
The site structure analysis step recognizes the hierarchical structure and the product category assigned to each level by acquiring and analyzing the HTML data displayed in the menu part for each level,
The bottom layer page analysis step extracts the product name of the product disclosed by acquiring and analyzing the HTML data of the product list displayed in the body part of the page of the bottom layer,
The product dictionary registration step associates the product name extracted in the lowest layer page analysis step and the product category related to the product name recognized in the site structure analysis step with attributes defined in the product dictionary. 15. When there is a new product category that is registered and not defined as an attribute of the product dictionary, the new product category is registered in the new attribute after the new attribute is defined in the product dictionary. Or the action information extraction method of 18 .

Images