JP2017111479A - Advertisement text selection device and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To select advertisement texts about a same advertisement object, according to an individual person.SOLUTION: An advertisement text analysis part 22 of an advertisement distribution device 2 creates an advertisement text vector which indicates a feature of each word included in advertisement text data, about each of pieces of advertisement text data corresponding to an advertisement object. A user preference analysis part 24 creates a user vector which indicates a feature of each word included in a text related to an electronic content used by the user. An advertisement text extraction part 253 for every cluster performs clustering of the advertisement text vectors, then extracts the advertisement text data in which the advertisement text vector similar to a center vector is acquired, for every cluster. A user matching part 255 selects the center vector similar to the user vector, out of the center vectors of each of the plural clusters, and then selects the advertisement text data extracted for the selected center vector, as the advertisement text data which matches the user.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an advertisement text selection device and a program.

Until now, in the sales promotion process of introducing products to customers, the target of advertisements has been narrowed down to those who are likely to be interested in the product based on the target group considered during product development, purchase history, and the like.
As a technology for narrowing down the destination of advertisements, there is a system that selects and bundles advertisement sheets for new products based on attributes such as gender and age of customers when delivering purchased products (for example, , See Patent Document 1). Further, there is a system for displaying an advertisement related to an input search word on the web (see, for example, Patent Document 2). There is also a technique for displaying an advertisement that matches a document selected by an individual such as a bookmark (see, for example, Patent Document 3). In addition, there is a technology that preferentially distributes an advertisement that is presumed to be highly interested by the user based on the history of advertisements distributed to the user and the history of advertisements browsed by the user (see, for example, Patent Document 4). . In addition, there is a technique for selecting an advertisement to be distributed to a user based on a user's expression history such as a posted article such as a blog or a user's known attribute history (see, for example, Patent Document 5).

JP 2006-155420 A JP 2012-79021 A JP 2004-118716 A JP 2007-286832 A JP 2009-145968 A

The above-described conventional technology presents products and offers advertisements suitable for customers. Therefore, there are cases where there are few opportunities to advertise products for which customers are limited or products that are considered to have few customers. Moreover, even if the advertisement is personalized, only the advertisement of the product suitable for the person is provided, and there is a product that is not advertised at all by the person.
In addition, the number of online stores that sell products on the Internet has increased, and now that preferences have been personalized, it is no longer possible to determine who matches the product by attributes such as existing age and gender. ing. As a result, traditional advertising strategies that advertise products and promote purchases targeting specific audiences are fading, and new buyers have been introduced beyond those previously thought to be potential buyers. It is considered necessary to have an advertisement for developing the company. For this purpose, it is required to effectively advertise products to be advertised to not only a specific layer but also a wide range of people.

  The present invention has been made in view of such circumstances, and provides an advertisement text selection apparatus and program that can select an advertisement text for the same advertisement target in accordance with an individual.

One aspect of the present invention is an advertisement sentence analysis unit that generates an advertisement sentence vector representing a feature of a word included in the advertisement sentence data for each of a plurality of advertisement sentence data corresponding to advertisement objects, and electronic content used by a user A user preference analysis unit that generates a user vector that represents a feature of a word included in the text, and a cluster of the advertisement sentence vectors generated from each of the plurality of advertisement sentence data, and a plurality of clusters obtained by clustering For each of the plurality of ad text data, a cluster-specific ad text extraction unit that extracts ad text data from which an ad text vector similar to the center vector of the cluster is obtained; Among them, similar to the user vector generated by the user preference analysis unit A user matching unit that selects a heart vector and selects the advertising text data extracted by the cluster-specific advertising text extraction unit for the selected center vector as advertising text data that matches the user. This is an advertisement text selection device.
According to this invention, the advertising text selection device generates an advertising text vector for each of a plurality of advertising text data corresponding to the advertising target. The ad text vector represents the characteristics of the words included in the ad text data. Further, the advertisement text selection device generates a user vector from text related to electronic content used by the user. The user vector represents a feature of a word included in text related to electronic content. The advertisement sentence selection device clusters advertisement sentence vectors, and selects an advertisement sentence vector similar to the center vector of each cluster. The advertisement text selection device selects a center vector similar to the user vector, and selects advertisement text data from which an advertisement vector similar to the selected center vector is obtained as advertisement text data that matches the user.
Thereby, the advertising text selection device can select the advertising text of the cut according to the personal preference from a plurality of advertising texts prepared in advance for the same advertising target.

One aspect of the present invention is the above-described advertising sentence selection device, wherein the advertising sentence vector is a vector obtained by adding distributed expression vectors of words included in the advertising sentence data, and the user vector is a user Is a vector obtained by adding the distributed expression vectors of the respective words included in the text relating to the electronic content used.
According to the present invention, the advertising sentence selection device generates an advertising sentence vector by adding the distributed expression vectors of the respective words included in the advertising sentence data, and distributes each of the words included in the text relating to the electronic content used by the user. The user vector is generated by adding the expression vectors.
As a result, the advertising text selection device generates an advertising text vector including a plurality of concepts included in the advertising text and a user vector including a plurality of concepts included in the text related to the electronic content used by the user. Used to select the combined ad text. Therefore, it is possible to select an advertisement sentence that widely matches the user's preference.

One aspect of the present invention is the above-described advertising sentence selection device, in which the user preference analysis unit adds the distributed expression vectors of the words included in the text related to the electronic content that the user gave a negative evaluation. A user negative vector is generated, and the user matching unit selects a central vector similar to a vector obtained by subtracting the user negative vector from the user vector among the central vectors of the plurality of clusters. Features.
According to the present invention, the advertising sentence selection device generates a user negative vector that represents a feature of a word included in text negatively evaluated by the user. The advertising text selection device selects a vector similar to the vector obtained by subtracting the user negative vector from the user vector from among the center vectors of the clusters obtained by clustering the advertising text data. The advertisement sentence selection device selects advertisement data of an advertisement sentence vector similar to the selected center vector as advertisement data suitable for the user's preference.
Thereby, the advertising sentence selection device can select an advertising sentence that includes a concept that the user is interested in and that is not included in the concept that the user does not show interest.

  According to one aspect of the present invention, a computer uses an advertisement sentence analysis unit that generates an advertisement sentence vector representing a feature of a word included in the advertisement sentence data for each of a plurality of advertisement sentence data corresponding to an advertisement target, and a user uses the computer User preference analysis means for generating user vectors representing the characteristics of words included in the text relating to the electronic content, and a plurality of the advertisement sentence vectors generated from each of the plurality of advertisement sentence data, obtained by clustering For each of the clusters, a cluster-specific ad text extracting means for extracting ad text data from which a text vector similar to the center vector of the cluster is obtained from among the plurality of ad text data, Of the center vector, the user preference analysis means generates the A user matching means for selecting a center vector similar to a user vector and selecting the advertisement text data extracted by the cluster-based advertisement text extraction means for the selected center vector as advertisement text data matching the user; It is a program for functioning as an advertisement text selection device provided.

  ADVANTAGE OF THE INVENTION According to this invention, the advertisement text about the same advertisement object can be selected according to an individual.

It is a functional block diagram which shows the structure of the advertisement delivery system by one Embodiment of this invention. It is a figure which shows the example of the advertisement text data group by the embodiment. It is a flowchart which shows the advertisement text vector production | generation process in the advertisement delivery apparatus by the embodiment. It is a flowchart which shows the user vector production | generation process in the advertisement delivery apparatus by the embodiment. It is a flowchart which shows the advertising sentence extraction process according to cluster in the advertisement delivery apparatus by the embodiment. It is a figure which shows the example of the advertisement text data classified by cluster by the embodiment. It is a flowchart which shows the user matching process in the advertisement delivery apparatus by the embodiment.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
The advertisement distribution system of the present embodiment aims to advertise a certain advertisement target to various people, not to a specific layer of people, and to use / purchase the advertisement target. In the conventional advertisement distribution system, advertisements are provided only to customers whose advertising targets are likely to meet their preferences. In other words, it was an approach of recommending an advertising target suitable for an individual from a plurality of advertising targets. On the other hand, the advertisement distribution system of the present embodiment personally advertises advertisements such as PR phrases that draw appeal from various aspects of the advertisement object in order to have various people use and purchase the advertisement object. The advertisement using the selected advertisement text is provided. The advertisement target can be arbitrary, but examples include products, services, broadcast programs, sightseeing spots, and companies.

  For example, it is assumed that the advertisement target is a car. The advertisement distribution system according to the present embodiment is such that A who likes camping has a rich seat arrangement and realizes a comfortable sleep. For B who has a small child, “it is easy to put a stroller without pillars.” Select different ad text for each customer. In this way, it is possible to grasp the attractiveness of a product from various perspectives, and to provide an advertisement with an advertisement text from a different viewpoint depending on the other party even for the same product. Below, the case where an advertising object is goods is explained as an example.

  FIG. 1 is a functional block diagram showing a configuration of an advertisement distribution system 1 according to an embodiment of the present invention. In the figure, only functional blocks related to the present embodiment are extracted and shown. The advertisement distribution system 1 includes an advertisement distribution device 2 and a user terminal 3.

  The advertisement distribution device 2 is an example of an advertisement text selection device. The advertisement distribution device 2 is realized by one or a plurality of computer devices. When the advertisement distribution device 2 is realized by a plurality of computer devices, which functional unit is realized by which computer device can be arbitrary. One functional unit may be realized by a plurality of computer devices. As shown in the figure, the advertisement distribution device 2 includes an ad text storage unit 21, an ad text analysis unit 22, a user history storage unit 23, a user preference analysis unit 24, an ad text personalization processing unit 25, And an advertisement distribution unit 26.

The advertisement text storage unit 21 stores an advertisement text data group. The ad text data group includes a plurality of ad text data for the same advertising target. Each advertisement text data is an advertisement text described in one sentence or a plurality of sentences. The advertisement text is a phrase such as a PR text about the product to be advertised.
The advertising text analysis unit 22 generates an advertising text vector of each advertising text data included in the advertising text data group. The ad text vector represents the characteristics of words included in the ad text indicated by the ad text data.

  The user history storage unit 23 stores a user ID and user history information in association with each other. The user ID is identification information that identifies a user. A terminal ID that is identification information for specifying the user terminal 3 can also be used as the user ID. The user history information indicates text data related to electronic content used by the user specified by the user ID by the user terminal 3. For example, the user history information is text data included in a web page provided on a website accessed by the user terminal 3. Alternatively, the user history information may be a search term transmitted from the user terminal 3 for use in a search on a website. Further, the user history information may be text data that is added to electronic content such as an electronic book or a moving image used by the user terminal 3 and describes the electronic content. The user history information may indicate information on a storage location of electronic content used by the user terminal 3, such as a URL (Universal Resource Locator) of a website.

  User history information can be collected by any conventional technique. When the electronic content is a website, for example, information on the address of a website accessed in the past is read from the web browser of the user terminal 3 using a cookie or the like. A cookie is a mechanism that allows a website provider to temporarily write and store data on the computer of a user who has accessed the website through a web browser. In many cases, the date and time is recorded. When the user visits the same website again, a service is provided as in the previous continuation based on the identification of the user stored in the user terminal 3 or the like. In addition, the user ID transmitted when the user terminal 3 accesses the website is collected by a computer server that provides the website or a computer server that provides an advertisement to be pasted on a web page provided by the website. May be. The user history information is generated based on the correspondence between the user ID collected in this way and the address of the website accessed by the user.

  The user preference analysis unit 24 reads user history information from the user history storage unit 23 and generates a user vector from text data indicated by the user history information. The user vector represents the characteristics of words included in the text data indicated by the user history information. When the user history information includes information on the storage location of electronic content, the electronic content is read from the storage location, and text data is acquired from the read electronic content.

  The ad text personalization processing unit 25 selects ad text data suitable for the user. The ad text personalization processing unit 25 includes an ad text vector group storage unit 251, a user vector storage unit 252, a cluster-specific ad text extraction unit 253, a cluster-specific ad text storage unit 254, and a user matching unit 255. .

The ad text vector group storage unit 251 stores an ad text vector group. The ad text vector group includes each ad text vector generated by the ad text analysis unit 22 for each of a plurality of ad text data for the same advertising target.
The user vector storage unit 252 stores the user vector generated by the user preference analysis unit 24.
The cluster-specific ad text extraction unit 253 clusters the ad text vectors obtained from the respective ad text data for the same advertising target. The cluster-specific ad text extraction unit 253 selects a predetermined number of ad text data from which an ad text vector similar to the center vector of the cluster is obtained for each cluster.
The cluster-specific ad text storage unit 254 stores the center vector of each cluster and the ad text data selected by the cluster-specific ad text extraction unit 253 for the center vector in association with each other.
The user matching unit 255 selects a center vector similar to the user vector among the center vectors of each cluster stored in the cluster-specific advertisement text storage unit 254. The user matching unit 255 reads the ad text data associated with the selected center vector from the cluster-specific ad text storage unit 254, and sets it as ad text data that matches the user.

  The advertisement distribution unit 26 distributes advertisement data using the advertisement text data selected by the advertisement text personalization processing unit 25 to suit the user to the user terminal 3.

  The user terminal 3 is a user's computer terminal. For example, the user terminal 3 is a terminal that can access the Internet, such as a personal computer, a tablet terminal, or a smartphone. The user terminal 3 displays the electronic content on a display included in the user terminal by an application corresponding to the electronic content. For example, the user terminal 3 displays a web page with a web browser.

FIG. 2 is a diagram illustrating an example of an advertisement text data group.
As shown in the figure, the advertising text data group is information in which an advertising target ID that is information for specifying an advertising target is associated with a plurality of advertising text data of the advertising target. The ad text data group may be, for example, document data describing contents that can be used as ad text data. As such document data, a product planning document, a document in which a specification that is a feature of the product is written, or the like can be used. A company that manufactures and sells products is thought to have a large number of documents containing product phrases for the purpose of product planning and publicity. Therefore, such a document is used as an advertisement text data group. Moreover, you may set manually the information which shows that it is advertisement text data with respect to the description location which can be used as advertisement text data among the contents contained in the document data used as an advertisement text data group. Also, a part of the description of the document data may be manually corrected to generate an advertisement text data group. Note that the advertisement text data preferably has contents adapted to the individual life rather than the product specifications. For example, if the advertising target is a car, rather than “loading capacity **”, it conveys the convenience of traveling to a family or group to those who are hobbies that it is the loading capacity. “A stroller can be placed as it is” that conveys the convenience of being a pillarless to parents with small children who use a stroller can be appealed to by the buyer rather than “being a pillarless”. Furthermore, the accuracy of selection of advertisement text data by the advertisement text personalization processing unit 25 is also improved.

Next, the operation of the advertisement distribution device 2 will be described.
FIG. 3 is a flowchart showing an advertisement text vector generation process in the advertisement distribution apparatus 2. The advertisement distribution device 2 executes the advertisement sentence vector generation process shown in the figure for each advertisement sentence data group.
The advertising text analysis unit 22 reads the advertising text data group stored in the advertising text storage unit 21 (step S110). The ad text analysis unit 22 selects and reads ad text data for which no ad text vector has yet been generated from the read ad text data group (step S120).

  The advertising text analysis unit 22 generates an advertising text vector of the read advertising text data. Therefore, first, the advertising sentence analysis unit 22 performs morphological analysis on the advertising sentence data read in step S120 and decomposes it into words (step S130). Existing techniques can be used for morphological analysis. For example, Mecab (see, for example, “https://code.google.com/p/mecab/”) is used as an existing morphological analysis technique. The ad text analysis unit 22 selects and extracts words that are not yet processed in order from the first word of the ad text data (step S140). The advertising sentence analysis unit 22 may extract all words subjected to morphological analysis, or may extract only words of a predetermined part of speech such as nouns.

  The advertising sentence analysis part 22 produces | generates the word vector of the word taken out in step S130 (step S150). Existing techniques can be used to generate word vectors. Here, word2vec (see, for example, “https://code.google.com/p/word2vec/”) is used as a word vector generation technique. The word vector generated by word2vec is a distributed expression vector. This distributed expression vector is an intermediate layer of a neural network that learns co-occurring words as input and output, respectively. In addition, this distributed expression vector is different from the vector of the word frequency like bag-of-words, and the semantic addition and subtraction such as "king"-"man" + "woman" = "queen" it can. The generation of word vectors requires text data used for learning. Usually, it is thought that the number of words included in the advertisement text is small, but the semantic concept included in the word vector can be widened by learning the word vector using a large amount of learning data of various topics. . Note that the word vector of each word may be learned in advance and stored in the database, and the advertising sentence analysis unit 22 may read the word vector of the word extracted in step S130 from this database.

  The advertising sentence analysis unit 22 determines whether or not the word currently being processed is the last word included in the read advertising sentence data (step S160). If the advertising sentence analysis unit 22 determines that it is not the last word (step S160: NO), it repeats the processing from step S140.

  If the advertising sentence analysis unit 22 determines that it is the last word (step S160: YES), it generates an advertising sentence vector from the word vector (step S170). Specifically, the advertising sentence analysis unit 22 calculates a vector obtained by adding all the word vectors generated for each word included in the advertising sentence data read in step S120. The advertisement text analysis unit 22 normalizes the calculated vector by dividing it by the number of word vectors added for the calculation of the vector, thereby obtaining an advertisement text vector. The advertisement text vector calculated in this way is the sum of the meanings of the words included in the advertisement text data. For example, it is possible to add to the advertising text data the mechanical concept of the pillarless car and the parenting concept of the stroller.

  The advertising text analysis unit 22 determines whether all the advertising text data included in the advertising text data group has been selected (step S180). When it is determined that there is unselected advertisement text data (step S180: NO), the advertisement text analysis unit 22 repeats the processing from step S120. On the other hand, the advertising sentence analysis part 22 performs the process of step S190, when it is judged that all the advertising sentence data were selected (step S180: YES). In other words, the ad text analysis unit 22 performs an ad text personalization process on an ad text vector group in which an advertising target ID and ad text data included in the ad text data group are associated with an ad text vector generated for each ad text data. It outputs to the part 25 (step S190). The ad text vector group storage unit 251 of the ad text personalization processing unit 25 stores an ad text vector group.

FIG. 4 is a flowchart showing user vector generation processing in the advertisement distribution apparatus 2. The advertisement distribution device 2 executes the user vector generation process shown in FIG.
The user preference analysis unit 24 reads user history information stored in the user history storage unit 23 (step S210). The user preference analysis unit 24 performs morphological analysis on the text data acquired from the read user history information and breaks it down into words (step S220). Existing techniques such as Mecab can be used for morphological analysis. The user preference analysis unit 24 selects and extracts words that are not yet processed in order from the first word of the text data (step S230). The user preference analysis unit 24 may extract all words subjected to morphological analysis, or may extract only words of a predetermined part of speech such as nouns.

  The user preference analysis unit 24 acquires the word vector of the word extracted in step S230 in the same manner as in step S150 in FIG. 3 (step S240). The user preference analysis unit 24 determines whether or not the word currently being processed is the last word included in the text data acquired from the user history information (step S250). If the user preference analysis unit 24 determines that it is not the last word (step S250: NO), it repeats the processing from step S230.

  When determining that the user preference analysis unit 24 is the last word (step S250: YES), the user preference analysis unit 24 generates a user vector from the word vector, similarly to step S170 in FIG. 3 (step S260). That is, the user preference analysis unit 24 calculates a vector obtained by adding all the word vectors generated by word2vec for each word included in the text data acquired from the user history information. The user preference analysis unit 24 normalizes the calculated vector by dividing it by the number of word vectors added for the calculation of the vector. The user preference analysis unit 24 outputs the user ID associated with the read user history information and the generated user vector to the advertisement text personalization processing unit 25 (step S270). The user vector storage unit 252 of the advertisement text personalization processing unit 25 stores the user ID and the user vector input from the user preference analysis unit 24 in association with each other.

FIG. 5 is a flowchart showing cluster-specific advertisement text extraction processing in the advertisement distribution apparatus 2. The advertisement distribution device 2 executes the cluster-specific advertisement text extraction process shown in FIG.
The cluster-specific advertisement text extraction unit 253 reads the advertisement text vector group from the advertisement text vector group storage unit 251 (step S310). The cluster-specific ad text extraction unit 253 reads out and clusters a plurality of ad text vectors associated with the same advertisement target ID from the ad text vector group, and arbitrarily N (N is an integer of 1 or more) clusters ( It is divided into groups (step S320). An existing algorithm can be used for clustering. For example, a k-means method (for example, see “https://en.wikipedia.org/wiki/K%E5%B9%B3%E5%9D%87%E6%B3%95”) is used as a clustering algorithm. . The number N of clusters is given in advance according to the number of advertisement cuts assumed by the advertisement provider. Note that when the algorithm for determining up to the number of clusters is used for clustering, it is not necessary to give the number of clusters N.

  The cluster-specific advertisement text extraction unit 253 obtains the center vector of each cluster (step S330). The center vector is a vector whose elements are the average values of the elements of the vectors belonging to the cluster. Subsequently, the cluster-specific advertisement text extraction unit 253 selects n (n is an integer of 1 or more) advertisement text vectors that are similar to the center vector of each cluster. Therefore, the cluster-specific advertisement text extraction unit 253 calculates the similarity between the center vector of each cluster and each advertisement text vector. The cluster-specific advertisement text extraction unit 253 selects n advertisement text vectors having the highest similarity for each center vector of each cluster. For example, a cosine similarity can be used as the similarity that quantitatively represents the degree of similarity between two vectors. Since the degree of similarity is calculated based on the angle between vectors, vector normalization is not necessarily required when using cosine similarity. The advertisement text vector to be selected may not be an advertisement text vector belonging to the same cluster as the center vector, but may be limited to an advertisement text vector belonging to the same cluster. The cluster-specific ad text extraction unit 253 generates cluster-specific ad text data in which the center vector of each cluster is associated with the ad text data obtained from each of the n ad text vectors selected for the vector ( Step S340). The cluster-specific ad text extraction unit 253 writes the generated cluster-specific ad text data in the cluster-specific ad text storage unit 254.

  FIG. 6 is a diagram illustrating an example of cluster-specific advertisement data. The cluster-specific ad text data shown in the figure is data in which a cluster number, a center vector of a cluster specified by the cluster number, and n pieces of ad text data selected for the center vector are associated with each other. In the ad text-1 to the ad text-n, ad text data arranged in descending order of similarity between the ad text vector and the center vector is set.

FIG. 7 is a flowchart showing user matching processing in the advertisement distribution apparatus 2. In the user matching process, the advertisement distribution device 2 selects advertisement data that matches the advertisement distribution destination user.
The user matching unit 255 receives, from the advertisement distribution unit 26, the user ID of the advertisement distribution destination user and the advertisement target information. The advertisement distribution destination user is, for example, a user who has accessed a web page or a user who has registered an e-mail address in advance in order to distribute information. The user matching unit 255 reads the user vector associated with the user ID from the user vector storage unit 252, and reads the cluster-specific advertisement text data corresponding to the advertisement target advertisement ID from the cluster-specific advertisement text storage unit 254. The user matching unit 255 selects K (K is an integer of 1 or more) center vectors similar to the user vector among the center vectors included in the cluster-specific advertisement data (step S410). Therefore, the user matching unit 255 calculates the similarity between each center vector and the user vector. For example, the cosine similarity can be used as the calculated similarity. The user matching unit 255 selects k center vectors having the highest similarity to the user vector. In addition, the user matching unit 255 may select a center vector whose similarity with the user vector is higher than a predetermined value.

  The user matching unit 255 outputs k pieces (k is an integer of 1 or more) of ad text data among the ad text data associated with the selected center vector to the advertisement distribution unit 26 (step S420). For example, the user matching unit 255 may select k pieces of advertisement text data in an order in which the advertisement sentence vector is similar to the center vector, or may randomly select k pieces. Further, the user matching unit 255 stores the advertisement text data distributed to each user, and among the plurality of advertisement text data associated with the selected center vector, May be preferentially selected from the ad text data that has not been distributed to the user or has been delivered to the user the least number of times. When a plurality of center vectors are selected, the number of ad text data to be selected may be increased as the center vector and the user vector are closer.

  The advertisement distribution unit 26 distributes the advertisement data generated using the advertisement text data input from the user matching unit 255 to the user terminal 3 of the user of the advertisement distribution destination (step S430). For example, the advertisement distribution unit 26 embeds the selected advertisement text data in advertisement data generated in advance excluding the advertisement text data, and generates advertisement data to be distributed. The advertisement distribution unit 26 distributes the generated advertisement data to the user terminal 3. The advertisement data may be, for example, a banner advertisement embedded in a web page or an e-mail delivered to a user address registered in advance.

  Since it is considered that the words included in the advertisement text data are few, the advertisement text vector may include an analytical noise that deviates from the cluster center. Therefore, the user vector and the ad text vector are not directly compared with each other so that the ad text data of the ad text vector with the reliability close to the center vector is selected instead of the ad text vector that may cause noise. Select ad text data that matches users based on the comparison. Also, the processing load for comparison can be reduced.

  According to the advertisement distribution system 1 described above, an advertisement text tailored to an individual is automatically extracted from advertisement texts that respectively insist on multifaceted attractiveness of a single product, and the attractiveness of the product is determined by the advertisement text. Can appeal. In today's competitive environment, products with many PR points (sales points) are often offered. By registering in advance an advertisement text that describes each selling point, the advertisement distribution system 1 selects an advertisement text that suits individual tastes, embeds it in an advertisement, and distributes it to any customer. Product PR tailored to the partner can be made. Therefore, customers other than the assumed target layer can be cultivated, and as a result, the market can be expanded. Also, for buyers, products that have been overlooked because they were not related to them so far are actually useful for their lives, and new discoveries can be obtained. Further, not only for products but also for other advertising targets, PR can be performed for people of various layers according to the people.

  When the advertising target is a program, a program guide such as EPG (Electronic Program Guide) can be used as the advertisement text data group. When the advertisement text data group is an EPG, a program explanation sentence (a sentence explaining the outline of the program or the details of the program) is used as the advertisement sentence data. In addition, a list of performers can be used for advertisement text data. Alternatively, subtitles such as closed captions of programs can be used as the advertisement text data group. Whether the EPG or the closed caption is used as the advertisement text data group, information indicating that it is advertisement text data may be manually set for a portion that can be used as the advertisement text data.

  Further, the advertisement text data group may be document data in which a reputation related to the advertisement target is described. For such document data, reviews, reviews, free opinions obtained through questionnaires, and the like can be used.

  Also, it is assumed that the advertisement distribution apparatus 2 can collect text data related to electronic content in which a negative (negative) evaluation that the user does not like or is not interested is input. For example, the advertisement distribution device 2 collects information on websites for which the user has input a negative evaluation from the user terminal 3. In this case, the user preference analysis unit 24 of the advertisement distribution device 2 creates a user negative vector using text data included in a website evaluated as negative by the user. The user negative vector is generated by the same process as that for generating the user vector from the user history information. That is, the user preference analysis unit 24 adds the word vectors acquired for each word included in the text data of the website evaluated as negative, and divides by the number of the added word vectors to generate a user negative vector. . The user preference analyzing unit 24 subtracts the user negative vector from the user vector generated based on the user history information of the user (subtracted by each element) as a user vector to be output to the ad text personalization processing unit 25. .

Further, the advertisement distribution apparatus 2 of the present embodiment can also be applied to a conventional personalized advertisement in which a product that suits a customer is selected and advertised.
For each product, the advertisement distribution device 2 adds all the advertisement text vectors generated for the advertisement text of the product (sums each element) and calculates a product vector. The advertisement distribution device 2 specifies a product having a product vector close to the user vector, and selects advertisement text data for the specified product as in the above-described embodiment. The advertisement distribution device 2 generates advertisement data using the selected advertisement text data and distributes it to the user terminal 3.

  According to the embodiment described above, the advertisement distribution system selects an advertisement sentence with a cut according to personal preference from a plurality of advertisement sentences prepared in advance for the same advertisement target, and selects the selected advertisement sentence. The used advertisement can be distributed.

  Note that the advertisement distribution apparatus 2 and the user terminal 3 described above have a computer system therein. The operation process of the advertisement distribution apparatus 2 and the user terminal 3 is stored in a computer-readable recording medium in the form of a program, and the above-described processing is performed by the computer system reading and executing this program. . The computer system here includes a CPU, various memories, an OS, and hardware such as peripheral devices.

Further, the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used.
The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Furthermore, the “computer-readable recording medium” dynamically holds a program for a short time like a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line. In this case, a volatile memory in a computer system serving as a server or a client in that case, and a program that holds a program for a certain period of time are also included. The program may be a program for realizing a part of the functions described above, and may be a program capable of realizing the functions described above in combination with a program already recorded in a computer system.

DESCRIPTION OF SYMBOLS 1 Advertisement delivery system 2 Advertisement delivery apparatus 3 User terminal 21 Advertisement sentence storage part 22 Advertisement sentence analysis part 23 User history storage part 24 User preference analysis part 25 Advertisement sentence personalization processing part 26 Advertisement distribution part 251 Advertisement sentence vector group storage part 252 User vector storage unit 253 Cluster-specific ad text extraction unit 254 Cluster-specific ad text storage unit 255 User matching unit

Claims (4)

For each of a plurality of ad text data corresponding to the advertising target, an ad text analysis unit that generates an ad text vector representing a feature of a word included in the ad text data;
A user preference analysis unit that generates a user vector representing the characteristics of words included in text related to electronic content used by the user;
The advertisement text vectors generated from each of the plurality of advertisement text data are clustered, and for each of the plurality of clusters obtained by clustering, an advertisement similar to the central vector of the cluster among the plurality of advertisement text data A cluster-specific ad text extractor for extracting ad text data from which a sentence vector was obtained;
The advertisement that the center vector similar to the user vector generated by the user preference analysis unit is selected from the center vectors of each of the plurality of clusters, and the cluster-specific advertisement sentence extraction unit extracts the selected center vector. A user matching unit that selects sentence data as ad sentence data matching the user;
An advertising sentence selection device comprising: The ad text vector is a vector obtained by adding the distributed expression vectors of the respective words included in the ad text data,
The user vector is a vector obtained by adding the distributed expression vectors of the respective words included in the text relating to the electronic content used by the user.
The advertising sentence selection device according to claim 1, wherein: The user preference analysis unit generates a user negative vector by adding the distributed expression vectors of each word included in the text related to the electronic content that the user gave a negative evaluation,
The user matching unit selects a center vector similar to a vector obtained by subtracting the user negative vector from the user vector among center vectors of the plurality of clusters.
The advertising sentence selection device according to claim 2, wherein: Computer
For each of a plurality of ad text data corresponding to the advertising target, an ad text analysis means for generating an ad text vector representing a feature of a word included in the ad text data,
A user preference analysis means for generating a user vector representing the characteristics of words included in text related to electronic content used by a user;
The advertisement text vectors generated from each of the plurality of advertisement text data are clustered, and for each of the plurality of clusters obtained by clustering, an advertisement similar to the central vector of the cluster among the plurality of advertisement text data Cluster-specific ad text extraction means for extracting ad text data from which a sentence vector was obtained;
The advertisement that the center vector similar to the user vector generated by the user preference analysis unit is selected from the center vectors of each of the plurality of clusters, and the advertisement text extraction unit classified by cluster extracts the selected center vector. User matching means for selecting sentence data as ad sentence data matching the user;
The program for functioning as an advertising-text selection apparatus provided with.

Images