JP2011238019A - Category processing device and method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To appropriately and efficiently improve a category structure which is used in a determination of a category etc.SOLUTION: Related terms or categories corresponding to a given sentence are determined. Category structure which is information associating one or more related terms for each category is stored in category storage means. The category structure stored in the category storage means is improved by changing the category structure using principle of genetic algorithm based on propriety of the category determined by a determination means with respect to a search result based on a web search request containing the related terms.

Description

  The present invention relates to the delivery of web advertisements.

  In recent years, by implementing natural language processing such as morphological analysis on a computer, there are related words (such as “feature words”) that are characteristic words in sentences such as web pages. And a technique for determining a related field (called “category”) is widely used. As an example of using this type of technology, select an advertisement based on the relevance between a keyword such as a category specified in advance by the advertiser for each advertisement and a related word or category determined for the web page viewed by the user. There are fields to display.

  For example, in Patent Document 1 by the present applicant, information (referred to as “category structure” in the present application) in which related word groups called keywords are associated with each category is prepared in advance. Then, the related word group determined based on the web page is compared with the related word group in the category structure, the category corresponding to the web page is determined based on the degree of association between the two, and the advertisement of the category is determined. Etc. are displayed on the web page. Further, in Patent Document 1, in order to expand the category structure, an example is shown in which a web search is performed using related words already in the category, and related words that frequently appear in the search result are further added to the related words of the category. Yes.

JP 2009-266204 A

  However, in the conventional technology as described above, the category structure can be improved in terms of expansion of related words, but not all other related words that frequently appear in the related word search results are appropriate for the category. It was a challenge to maintain the appropriateness of However, manually performing the operation of adding and removing related words while judging such appropriateness is heavy, and it is not practical to perform it for a large number of categories.

  In view of the above problems, an object of the present invention is to appropriately and efficiently improve a category structure used for determining a category or the like.

  Based on the above object, the category processing apparatus according to one aspect (1) of the present invention includes a determination unit that determines a related word or category corresponding to a given sentence, and one or more related words for each category. With respect to the category storage means storing the category structure as the associated information and the category structure stored in the category storage means, the determination means for the search result based on the web search request including the related word Improvement means for improving by changing the combination of the related words using the principle of the genetic algorithm based on the suitability of the determined category.

  Another aspect (7) of the present invention is a category storage that captures the above aspect from the viewpoint of a method and stores a category structure that is information in which one or two or more related words are associated with each category. A category processing method executed by a computer having means for determining a related word or category corresponding to a given sentence, and the computer storing the category structure stored in the category storage means An improvement step for improving the search result based on the web search request including the related word based on the suitability of the category determined in the determination step by changing using the principle of a genetic algorithm. And

  A category processing program according to another aspect (8) of the present invention captures the above aspect from the viewpoint of a computer program, and is a category structure that is information in which one or more related words are associated with each category. Is a computer program for controlling a computer having category storage means for storing a computer, which causes a computer to determine a related word or category corresponding to a given sentence, and is stored in the category storage means. The category structure is improved by changing the category result based on the suitability of the determined category with respect to a search result based on a web search request including the related word.

  As described above, the principle of the genetic algorithm that changes the category structure used for advertisement selection etc. while selecting the combination of the related words according to whether the web search result by the related words is determined as the correct category. By using it, the category structure can be improved appropriately and efficiently.

  According to another aspect (2) of the present invention, in any of the above aspects, the improvement means uses at least one of crossover or mutation and natural selection as a search point for a new prototype individual with a category structure. It uses a genetic algorithm that generates and arrives at an appropriate solution by multipoint search.

  In this way, by applying a genetic algorithm to the present invention, it is possible to reach a practical solution or an optimal solution without grasping or elucidating the condition of individual suitability, and thus it is easy to improve the category structure appropriately and efficiently It becomes.

  According to another aspect (3) of the present invention, in any one of the above aspects, the improvement means stores individual storage means for storing one or more prototype individuals for the category structure of a category, and storage in the individual storage means. Crossing means for creating a new prototype individual as a descendant by combining a plurality of prototype individuals being ancestors, or a category or relationship included in the prototype individual stored in the individual storage means Mutation that changes the prototype individual by at least one of the process of acquiring a new related word based on the word and adding it to the related word of the prototype individual, or the process of deleting a part of the related word contained in the prototype individual Search execution means for performing a web search using related terms contained in the prototype stored in the individual storage means, and at least one of the means, and the web search Sample generation means for generating a sample character string by concatenating a plurality of character strings based on each web page included in the search results, and causing the determination means to determine a category for the generated sample character string, and the determined category , Based on the difference with the category of the prototype individual on which the sample character string is based, a natural selection means for natural selection by deleting the prototype individual or the prototype individual that is an ancestor or descendant of the prototype individual, It is characterized by having.

  In this way, the prototype of the category structure is changed by crossover, mutation, etc., and the sample character string based on the web search result using the related words of the prototype is left in the prototype, and the others are naturally deceived. As a result, it is possible to utilize the principle of the genetic algorithm that an appropriate category structure that matches the category is selected and survived by the survival of the appropriate person, and the category structure can be improved appropriately and efficiently.

  According to another aspect (4) of the present invention, in any one of the above aspects, the mutation means performs a web search using a category or a related word in a prototype of a category structure, and a web page included in the search result It is characterized in that a new related word is acquired from a character string based on.

  In this way, a web search is performed using category names and related words included in the prototype of the category structure, and new related words are efficiently acquired based on the search results to cause mutations. The improvement of the category structure by the evolution of individuals can be effectively promoted.

  According to another aspect (5) of the present invention, in any one of the above aspects, the end of the operation of the improvement means is determined based on a predetermined end condition, and is stored in the individual storage means at that time. It has an end confirmation means for storing the prototype individual as a new category structure in the category storage means.

  In this way, the improvement process based on the principle of the genetic algorithm is terminated under a predetermined termination condition, and the category structure is determined and stored, so that it is not necessary to manually determine the termination and perform processing, and the category structure can be changed. It can be improved appropriately and efficiently.

  The category structure improving apparatus according to another aspect (6) of the present invention includes a determination unit that determines a related word or category corresponding to a given sentence, and information in which one or more related words are associated with each category. A category storage device having a category storage means for storing a category structure, and a category structure improving device for a category etc. acquisition device, wherein the related terms are assigned to the category structure stored in the category storage means. Improvement means for improving by changing the search result based on the included web search request by using the principle of the genetic algorithm based on the suitability of the category determined by the determination means.

  It should be noted that a category (method for the apparatus, program for the method, etc.) different from each of the above-described modes and more specific modes described below are also included in the present invention. For different categories, “means” shall be appropriately read as “step”.

  According to the present invention, the category structure used for determining a category or the like can be improved appropriately and efficiently.

The functional block diagram which shows the structure of embodiment of this invention. The figure which illustrates the information (data) used by embodiment of this invention. The flowchart which shows the process sequence in embodiment of this invention. The conceptual diagram which illustrates crossing regarding the genetic algorithm in embodiment of this invention. The functional block diagram which shows the structure of 2nd Embodiment of this invention.

  Next, modes for carrying out the present invention (referred to as “embodiments”) will be described with reference to the drawings. It should be noted that assumptions common to those already described in the background art and problems are omitted as appropriate.

[1. Constitution〕
FIG. 1 is a diagram showing a configuration of the present embodiment. In the present embodiment, a category processing apparatus 1 (hereinafter referred to as “this apparatus 1” or “this apparatus”) that determines a related word or category corresponding to a given sentence. Application example). The apparatus 1 includes a general computer configuration, for example, an arithmetic control unit 6 such as a CPU, a storage device 7 such as an external storage device (HDD or the like) or a main memory, and a communication network N (Internet, mobile phone network, Communication means 8 (LAN adapter etc.) etc. with a company network etc. are provided. The other devices 3 and 4, the web server W, and the terminal T are the same although not shown.

  In each device, a predetermined computer program (not shown) stored (installed) in advance in the storage device 7 controls the arithmetic control unit 6 so that the elements (11, 12, 21,. 22, 31, 32, 41, 42, etc.). Among these elements, the information storage means can be realized in the storage device 7 in any format such as various files and databases (also referred to as “DB”), variables such as arrays, various stacks and registers, and system setting values.

  Among such storage means, the category storage means 22 is a means for storing a category structure that is information in which one or two or more related words are associated with each category. As exemplified in 1), representative words such as category names and their synonyms may be included in addition to related words.

  Further, the advertisement distribution device 3 determines whether the web server W distributes to the terminal T based on the determination result of the category or the like by the category processing device 1 and the category structure stored in the category storage unit 22. A server device that distributes advertisements. The advertisement distribution device 3 includes an advertisement storage unit 31 that stores advertisements, an acquisition unit 32 that acquires a category or a related word as a reference for selecting an advertisement for a web page, and an advertisement processing unit that selects and distributes an advertisement. 33.

  Among these, the advertisement data stored in the advertisement storage unit 31 includes, for example, a bid keyword specified by the advertiser for each advertisement such as a character string identified by the advertisement ID, as illustrated in FIG. Among the advertisements that match the bid keyword including the bid price and the category or characteristic word representing the content of the web page, the advertisement with the higher bid price is preferentially delivered to the web page.

  Each means other than the storage means is a processing means for realizing and executing the following information processing functions and operations.

[2. (Advertising effect)
In the present embodiment configured as described above, the advertisement distribution process based on the determination of the category or the like is performed as follows in the normal time except when the category structure improvement process according to the present invention is performed. First, the advertisement processing means 33 of the advertisement distribution apparatus 3 receives character string data (also simply referred to as “text”) such as HTML data representing the contents of a web page to be displayed by incorporating an advertisement from the web server W or the terminal T that is a cooperation destination. ) Is sent to the category processing device 1.

  The determination means 21 of the category processing apparatus 1 includes morphological analysis, tf · idf (feature words and important words in sentences, tf (Term Frequency: word appearance frequency) and idf (Inverse Document Frequency)). Using a language processing technology such as an algorithm to extract by one index) or syntactic analysis, determine related words or categories corresponding to a given sentence, and represent character strings, word IDs, and categories representing these related words The determination result is passed to the acquisition means 32 of the advertisement distribution apparatus 3 in the form of a character string such as a word, a word ID, or a category ID. Note that the determination means 21 is set in a state where appropriate determination is made in advance by machine learning or other learning algorithms using correct answer data.

  The acquisition unit 32 of the advertisement distribution device 3 refers to the category structure stored in the category storage unit 22 with respect to the related word or category determined by the determination unit 21 as described above, thereby corresponding category or related word. To get. This is because the category and other related words included in the category structure to which the determined related word belongs, and other related words included in the category structure of the determined category are designated as bid keywords. Means that the advertisement that is present may be displayed on the web page.

  In this case, for example, an advertisement having a related keyword “pasta” included in the same category structure as the category “Italian” as a bid keyword may be displayed on the web page determined as the category “Italian”. Therefore, the matching accuracy between the web page and the advertisement can be improved by improving the category structure as follows.

[3. (Improved category structure)
[3-1. Overview of improvement)
As an outline, the improvement means 10 uses the genetic algorithm based on the suitability of the category determined by the determination means 21 for the search result based on the web search request including the related words for the category structure stored in the category storage means 22. The category structure is improved by changing the combination of the related terms using the principle of More specifically, the improvement means 10 generates a new prototype individual with a category structure as a search point using at least one of crossover or mutation and natural selection, and reaches an appropriate solution by multipoint search. The genetic algorithm is executed by the following processing procedure illustrated in the flowchart of FIG.

  That is, the improvement means 10 has an individual storage means 11 for storing one or more prototype individuals for the category structure of the category. First, the category structure to be improved is transferred from the category storage means 22 to the individual storage means. 11 is copied (step S10). Further, the improvement means 10 has at least one of the crossover means 12 or the mutation means 13 as means for changing or evolving a prototype individual starting from the category structure copied as described above. Shall. Then, a new prototype individual is created by crossover and mutation by the crossover means 12 and the mutation means 13 (step S11).

[3-2. (Crossover)
Among these, the crossover means 12 uses a plurality of prototype individuals stored in the individual storage means 11 as ancestors, and creates a new prototype individual as a descendant by combining the respective partial elements. For example, as illustrated in FIG. 4A, for a certain category X, a prototype individual X1 in which related words A1, A2, A3, A4 are combined and related words B1, B2, B3, B4 of the same category X Assume that the combined prototype individual X2 exists as a parent.

  In this case, an example in which the crossing means 12 creates offspring is a combination of the first half of the prototype individual X1 (related words A1 and A2) and the second half of the prototype individual X2 (related words B3 and B4) to form the prototype individual X12. Also, the second half of the prototype individual X1 (related words A3 and A4) and the first half of the prototype individual X2 (related words B1 and B2) are combined to form the prototype individual X21. Of course, the number of related words is free and may differ for each prototype, the number of parents that can be extracted and combined can be 3 or more, and the number, location, number, etc. of related words to be separated when extracting the portion Is also free.

[3-3. mutation〕
Moreover, the mutation means 13 implement | achieves a mutation by changing a part of trial manufacture individual | organization | working at the appropriate timing of periodic or irregular. Specifically, the mutation means 13 acquires a new related word based on a category or related word included in the prototype stored in the individual storage means 11 and adds it to the related word of the prototype individual, Alternatively, the prototype individual is changed by at least one of processing for deleting a part of the related words included in the prototype individual.

  An example of a mode in which the mutation means 13 acquires a new related word is to use web search. In this case, a web search is performed using a category (a representative word such as a category name) or a related word in a prototype individual of the category structure, and the search result is received. From the character string based on the web page included in the search result, A new related word is acquired by morphological analysis and Japanese processing such as tf / idf.

  In order to use the web search in this way, a search server device 4 that is the same as or similar to a so-called search engine (search site) is provided to the user by using an API (Application Programming Interface / Application Program Interface), etc. Access such as transmitting a search request according to a web browser is performed, and web page data such as an HTML file transmitted in response to the search request is received as a search result. As a search keyword in this web search, that is, a search query, categories and related words may be targeted one by one, or for example, a logical operation using a plurality of related words may be used. In the search server device 4, the search processing means 42 searches the web page group corresponding to the search query based on so-called index data prepared in advance in the search related data storage means 41 and transmits it to the access source as a search result page.

  In addition, as a “character string based on a web page” used for acquiring related terms, for example, an explanatory text displayed for each web page hit in a search result, an extracted sentence such as a snippet, and the contents of a linked web page And so on. In addition to adding or deleting related words, for example, replacing related words with synonyms, words of higher-level concepts or lower-level concepts, etc., or character types expressing related words as kanji, hiragana, katakana, Mutation of the prototype individual may be realized by changing between character types such as alphabet.

[3-4. Natural selection〕
New prototypes created or changed by crossover or mutation as described above are subject to natural selection as follows. That is, first, the search execution unit 14 targets at least a newly created or changed prototype individual stored in the individual storage unit 11 and uses related words included in the prototype individual. The search is performed by accessing the search server device 4 with an API or the like (step S12). This web search is performed for each related word for all or a part of related words included in the category structure that is a target prototype individual, and the search query for each related word may be a single related word or a plurality of related words. Related words, related words, and categories (actually representative words such as category names) may be combined by a logical operation.

  Subsequently, the sample generation unit 15 includes a character string based on each web page included in the search result of the web search performed by the search execution unit 14 (for example, an extracted sentence such as an explanatory sentence accompanying a list of hit web pages, a snippet, or a link) A sample character string is generated by concatenating a plurality of web page contents and the like (step S13). As a simple example based on FIG. 2 (1), for the category “Ramen”, a character string included in the search result page of “Tonkotsu”, a character string included in the search result page of “Churshu”, and “trigger” The character strings included in the search result page are connected together to form a sample character string.

  Then, the natural selection means 16 causes the determination means 21 to determine the category for the generated sample character string (step S14), and the difference between the determined category and the category of the prototype individual that is the basis of the sample character string is determined. Based on the above, by deleting the prototype individual or the prototype individual that becomes the ancestor or descendant of the prototype individual, the natural selection of the prototype individual is realized (step S15).

  Details such as how to generate the sample string, the significance of the category “difference”, and the criteria for deletion can be freely defined. For example, in order to leave a relatively excellent one among multiple prototypes, For example, it is possible to generate sample character strings by selecting related words included in the prototype individual in various combinations, and to delete other prototype individuals whose sample strings are judged to be in the correct category. .

[3-5. (Judgment of termination)
While repeating the improvement process as described above, the end determination means 17 determines the end of the operation of the improvement means 10 based on a predetermined end condition (step S16), and the individual storage means at that time 11 is stored in the category storage means 22 as a new category structure (step S17).

  Although the content of the end condition is free, for example, for the target category structure, the number of related words increases to a predetermined number or more, and sample characters generated by selecting related words included in the prototype individual with various combination patterns Conditions such that the categories determined by the determining means 21 for the column are all correct or correct at a predetermined ratio or more are conceivable. In addition, regarding the determination of the end condition, a human may be partially or wholly involved, for example, by presenting a status value to the operator and requesting the determination.

[4. effect〕
In the present embodiment, as described above, the category structure used for advertisement selection or the like is changed while selecting the combination of the related words according to whether the web search result based on the related words is determined as the correct category. By using the principle of the genetic algorithm, the category structure can be improved appropriately and efficiently.

  In particular, in this embodiment, by applying a genetic algorithm to the present invention, it is possible to reach a practical solution or an optimal solution without grasping or elucidating conditions of suitability of individuals, so that the category structure is improved appropriately and efficiently. It becomes easy.

  Further, in this embodiment, the prototype individual of the category structure is changed by crossover or mutation, and the sample character string based on the web search result using the related words of the prototype individual is left as the prototype individual that is determined as the correct category. By natural selection, it is possible to utilize the principle of the genetic algorithm that an appropriate category structure that matches the category is selected and survived by the survival of the right person, and the category structure can be improved appropriately and efficiently.

  Furthermore, in this embodiment, a web search is performed using category names and related words included in the prototype of the category structure, and new related words are efficiently acquired based on the search results to cause mutations. The improvement of the category structure due to the evolution of the prototype can be effectively promoted.

  In addition, in the present embodiment, the end determination means 17 ends the improvement process based on the principle of the genetic algorithm under a predetermined end condition, and determines and stores the category structure, so that the end determination and processing are manually performed. Thus, the category structure can be improved appropriately and efficiently.

[5. Second Embodiment]
In the above embodiment (also referred to as the first embodiment), the determination means 21 such as related words, the category storage means 22 that stores the category structure, and the category structure improvement means 10 are included in the present invention. Although the example which the category processing apparatus 1 has was shown (FIG. 1), the present invention does not need to include the determination unit and the category storage unit as components. That is, when there is an existing category acquisition device including a determination unit and a category storage unit, the category structure improvement device for improving the category structure by combining it with the acquisition unit etc. It can be grasped as one mode.

  Such a second embodiment is shown in the block diagram of FIG. In the second embodiment, a category storing a category structure that is information that associates one or two or more related words for each category with a determination unit 21 that determines a related word or category corresponding to a given sentence. A category structure improving apparatus 111 for a category etc. acquiring apparatus 2 having a storage means 22 and having an improving means 10. This improvement means 10 uses the principle of the genetic algorithm based on the suitability of the category determined by the determination means 21 for the search result based on the web search request including the related words for the category structure stored in the category storage means 22. The details are omitted because they are the same as the improvement means in the first embodiment.

  According to the second embodiment, since the category structure can be improved by applying the present invention to an existing category etc. acquisition device, the existing system resources can be used effectively.

[6. Other embodiments]
In addition, said each embodiment is only an illustration, and this invention includes what is illustrated below and other embodiment other than that. For example, each of the above embodiments exemplifies advertisement distribution as an application field, but the category structure improvement technique according to the present invention can be applied to any field such as automatic category classification of news articles and product search in electronic commerce. Good. Moreover, the specific contents of the genetic algorithm are free, and the combined use of crossover and mutation is not essential.

  Furthermore, each element such as means is not limited to the arithmetic control unit of the computer, and may be realized by other information processing mechanisms such as an electronic circuit based on wired logic, etc., and each configuration diagram, data diagram, and flowchart diagram These are merely examples, and the presence / absence of each element, its order, specific contents, and the like can be changed as appropriate. For example, each device may be realized by using a plurality of devices such as servers, and a configuration in which each storage unit is realized by a separate server device or system is also common. Depending on the function, the configuration can be flexibly changed, for example, by calling an external platform or the like with an API (application program interface) or network computing (so-called cloud or the like).

DESCRIPTION OF SYMBOLS 1 Category processing apparatus 2 Category etc. acquisition apparatus 3 Advertisement delivery apparatus 4 Search server apparatus 6 Arithmetic control part 7 Storage apparatus 8 Communication means 10 Improvement means 111 Category structure improvement apparatus 11 Individual storage means 12 Crossover means 13 Mutation means 14 Search execution means 15 Sample generation means 16 Natural selection means 17 Termination confirmation means 21 Determination means 22 Category storage means 31 Advertisement storage means 32 Acquisition means 33 Advertisement processing means 41 Search related data storage means 42 Search processing means A1, A2, A3, A4, B1, B2, B3, B4 Related terms N Communication network T Terminal W Web server X Category X1, X2, X12, X21 Prototype

Claims (8)

A determination means for determining a related word or category corresponding to a given sentence;
For each category, category storage means for storing a category structure that is information in which one or more related words are associated with each other;
The category structure stored in the category storage means is changed using the principle of a genetic algorithm based on the suitability of the category determined by the determination means for the search result based on the web search request including the related word. Improvement means to improve by,
A category processing apparatus comprising:   The improvement means uses a genetic algorithm that generates a new prototype individual with a category structure as a search point using at least one of crossover or mutation and natural selection and reaches an appropriate solution by multipoint search. The category processing apparatus according to claim 1. The improvement means includes
Individual storage means for storing one or more prototypes for the category structure of a category;
Crossover means for creating a new prototype individual as a descendant by combining a plurality of prototype individuals stored in the individual storage means and combining the respective subelements, or a prototype individual stored in the individual storage means Prototype by at least one of the process of acquiring a new related word based on the category or related word included in and adding it to the related word of the prototype individual, or the process of deleting a part of the related word contained in the prototype individual At least one of mutation means for changing the individual,
Search execution means for performing a web search using related terms included in the prototype individual stored in the individual storage means;
Sample generation means for generating a sample character string by concatenating a plurality of character strings based on each web page included in the search result of the web search;
Based on the difference between the determined category and the category of the prototype individual that is the basis of the sample character string, the prototype individual or the prototype of the sample character string is generated. A natural selection method for natural selection by deleting a prototype individual that is an ancestor or descendant of the individual,
The category processing apparatus according to claim 1, wherein the category processing apparatus includes: The mutation means performs a web search using a category or a related word in a prototype individual having a category structure, and acquires a new related word from a character string based on a web page included in the search result. Item 4. The category processing device according to item 3. End of determining the end of the operation of the improvement means based on a predetermined end condition and storing the prototype individual stored in the individual storage means at that time in the category storage means as a new category structure The category processing apparatus according to claim 1, further comprising a determination unit. A determination means for determining a related word or category corresponding to a given sentence;
For each category, category storage means for storing a category structure that is information in which one or more related words are associated with each other;
A category structure improving device for a category etc. acquiring device, comprising:
The category structure stored in the category storage means is changed using the principle of a genetic algorithm based on the suitability of the category determined by the determination means for the search result based on the web search request including the related word. Improvement means to improve by,
A category structure improving apparatus comprising: A category processing method executed by a computer having category storage means for storing a category structure that is information in which one or more related words are associated with each category,
A determination step in which a computer determines a related word or category corresponding to a given sentence;
The computer uses the principle of the genetic algorithm based on the suitability of the category determined in the determination step for the search result based on the web search request including the related word for the category structure stored in the category storage means. Improvement steps to improve by changing
The category processing method characterized by including. A computer program for controlling a computer having category storage means for storing a category structure, which is information in which one or more related words are associated with each category,
Let the computer determine the related words or categories corresponding to the given sentence,
The computer changes the category structure stored in the category storage means based on the suitability of the determined category with respect to the search result based on the web search request including the related word using the principle of the genetic algorithm. A category processing program characterized by being improved by

Images