JP2009145968A - Advertisement distribution device, advertisement distribution method, advertisement distribution program and advertisement bidding method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To estimate an attribute profile of a user from an expression history of the user, and to provide advertisement information based on the estimated attribute profile of the user. <P>SOLUTION: This advertisement distribution device 10 equipped with an advertisement DB 62 is provided with a learning part 51 for mechanically learning a reference profile based on an expression history as information to be transmitted from a user terminal and the known attribute information of a user; an estimation part 52 for estimating the attribute profile of the user based on the reference profile; an advertisement securing part 53 for securing advertisements stored in an advertisement DB 62 based on the attribute profile of the user estimated by the estimation part 52; and an advertisement distribution part 54 for distributing the advertisements reserved by the advertisement reserving part 53 to the user. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to an advertisement distribution device, an advertisement distribution method, an advertisement distribution program, and an advertisement bidding method that distribute advertisements based on attribute profiles estimated from text or the like.

  2. Description of the Related Art Conventionally, a technique is known in which a computer terminal and a server that distributes advertisement information are connected to a network, and advertisement information that is optimal for the needs of a user who operates the computer terminal is transmitted only to a specific computer terminal. For example, a technique is known in which information related to a product previously purchased by a user is managed and advertisement information of a similar product is distributed to the user's computer terminal.

  For example, in Japanese Patent Laid-Open No. 2004-228688, the user's interest is high, the customer profile, the purchase history / frequency, the access history / frequency, the participation history / frequency of events, etc., the response to the e-mail notification regarding the product group and the related product group. , And a method for providing information, advertisements, and services in accordance with a response to a coupon grant or the like.

Moreover, in patent document 2, the advertisement data which should be delivered to the said user terminal is extracted from the advertisement data memorize | stored beforehand with reference to the user profile corresponding to a user terminal, and a network is connected to a user terminal in the extracted advertisement data. A method of distributing via the Internet is known.
Japanese Patent Laid-Open No. 2001-175761 Japanese Patent Laid-Open No. 2002-236846

  However, Patent Literature 1 and Patent Literature 2 have a problem that when advertising information is distributed to a user terminal, the advertising information cannot be distributed unless some action is taken from the distributor side.

  Moreover, in patent document 2, in order to deliver advertisement information, the distributor side needs to acquire the profile to which a user belongs beforehand. However, the profile to which the user belongs may be known for all profiles necessary for distributing advertisement information, or may be partially or completely unknown. Furthermore, from the viewpoint of privacy information protection, it is not preferable that the distributor side accumulates the profile to which the user, which is the user's personal information, more than necessary.

  In view of this, the present invention provides information transmitted by a user who is a target for distributing advertisement information, for example, a query for search, contents of a blog or homepage, writing in a comment field of a blog (hereinafter referred to as expression history). ) To estimate the profile to which the user belongs, and to provide advertisement information based on the estimated profile to which the user belongs.

  The present invention provides the following solutions.

  (1) In an advertisement distribution device including an advertisement information database, a learning unit that machine-learns a reference profile based on an expression history that is information transmitted from a user terminal of an information service and user-known attribute information; An estimation unit that estimates an attribute profile of the user based on the reference profile; an advertisement allocation unit that allocates an advertisement stored in the advertisement information database based on the attribute profile of the user estimated by the estimation unit; An advertisement distribution apparatus comprising: a distribution unit that distributes the advertisement allocated by the advertisement allocation unit to the user.

  According to the configuration of (1), in the advertisement distribution apparatus including the advertisement information database, the apparatus calculates the reference profile based on the expression history that is information transmitted from the user terminal and the user's known attribute information. Machine learning, estimating an attribute profile of the user based on the reference profile, assigning an advertisement stored in the advertisement information database based on the estimated attribute profile of the user, and sending the allocated advertisement to the user To deliver.

  In this way, based on the user's known attribute information and the expression history, the user's attribute profile is estimated using the reference profile, and the advertisement information is provided based on the estimated user's attribute profile. it can. Therefore, it is possible to improve the accuracy of distributing advertisement information. Furthermore, there is no need to take action from the distributor side, and the advertisement distribution apparatus can automatically allocate advertisement information and distribute it to the user.

  (2) The learning unit acquires the expression history, an analysis unit that analyzes morphological analysis of the expression history that is text information acquired by the acquisition unit, and the morpheme analyzed by the analysis unit A classification unit for classifying based on the attribute profile of the user, and a generation unit for generating a reference profile based on the morpheme classified by the classification unit, wherein the estimation unit obtains a representation history by the user An acquisition unit; and a user expression history analysis unit that performs a morphological analysis on the expression history of the user acquired by the user expression history acquisition unit, and uses the reference profile generated by the generation unit to use the attribute profile of the user And the attribute profile of the user is estimated based on the morpheme analyzed by the user expression history analysis unit The advertisement distribution device according to the above feature the Rukoto (1).

  According to the configuration of (2), in the advertisement distribution device according to (1), the learning unit acquires the expression history, performs morphological analysis on the expression history that is the acquired text information, and is analyzed The morpheme is classified based on the attribute information, a reference profile is generated based on the classified morpheme, the estimation unit acquires the expression history by the user, and performs the morphological analysis on the acquired expression history by the user The user's attribute profile is estimated based on a part of the user's attribute profile and the analyzed morpheme.

  Thus, by classifying the morphemes of the expression history, the accuracy of the reference profile can be improved in the learning unit, and the accuracy of estimating the attribute profile of the user can be improved in the estimation unit.

  (3) The classification unit classifies the morpheme by calculating a probability that the morpheme analyzed by the analysis unit belongs to the attribute information using a naive Bayes method, and the reference profile generated by the generation unit Is the probability that the classified morpheme belongs to a predetermined attribute profile, and the estimation unit determines the probability that the morpheme classified by the classification unit belongs to the predetermined attribute profile as a reference profile generated by the generation unit. The advertisement distribution device according to (2), wherein the attribute profile of the user is estimated.

  According to the configuration of (3), in the advertisement distribution device according to (2), the classification unit calculates a probability that the morpheme analyzed by the analysis unit belongs to the attribute information using a naive Bayes method. The reference profile generated by the generation unit is a probability that the classified morpheme belongs to a predetermined attribute profile, and the estimation unit includes the reference profile generated by the generation unit, The attribute profile of the user is estimated using the probability that the morpheme classified by the classification unit belongs to a predetermined attribute profile.

  In this way, by classifying the morphemes using the Naive Bayes method, the learning unit can perform stable and high-speed learning, and the estimation unit has an accuracy of estimating the user attribute profile. Can be improved.

  (4) The classification unit classifies the morpheme by determining whether the morpheme analyzed by the analysis unit belongs to the attribute information by using a support vector machine, and is generated by the generation unit The reference profile is an identification plane defined by a support vector, and the estimation unit estimates the attribute profile of the user using the identification plane as a reference profile generated by the generation unit. The advertisement distribution device according to (2).

  According to the configuration of (4), in the advertisement delivery device according to (2), the classification unit uses a support vector machine to determine whether the morpheme analyzed by the analysis unit belongs to the attribute information. The morpheme is classified by discrimination, and the reference profile generated by the generation unit is an identification plane defined by a support vector, and the estimation unit uses the identification profile as a reference profile generated by the generation unit. The attribute profile of the user is estimated using the surface.

  In this way, by classifying morphemes using the support vector machine, the learning unit can perform efficient learning, and the estimation unit improves the accuracy of estimating the user's attribute profile. be able to.

  (5) The classification unit classifies the morpheme by detecting an expression in which the morpheme analyzed by the analysis unit meets a predetermined condition, and the estimation unit generates a reference profile generated by the generation unit. The advertisement distribution apparatus according to (2), wherein the attribute profile of the user is estimated using a result of detecting an expression in which the morpheme meets a predetermined condition.

  According to the configuration of (5), in the advertisement distribution device according to (2), the classification unit detects the morpheme by detecting an expression in which the morpheme analyzed by the analysis unit matches a predetermined condition. The estimation unit estimates the attribute profile of the user using a result of detecting an expression in which the morpheme meets a predetermined condition as the reference profile.

  In this way, by classifying morphemes by detecting expressions that match a predetermined condition, the learning unit can perform efficient learning, and the estimation unit can easily perform user attributes. The profile can be estimated.

  (6) A learning step for machine learning of a reference profile based on an expression history that is information transmitted from a user terminal and user's known attribute information, and an estimation for estimating the user's attribute profile based on the reference profile An advertisement allocation step of allocating an advertisement stored in an advertisement information database based on the estimated attribute profile of the user, and a distribution step of distributing the allocated advertisement to the user Delivery method.

  According to the configuration of (6), it is possible to provide an invention that exhibits the same effects as the device of (1) as a method invention.

  (7) A learning step for machine learning of a reference profile based on an expression history that is information transmitted from a user terminal and user's known attribute information, and an estimation for estimating the user's attribute profile based on the reference profile An advertisement for causing a computer to execute a step, an advertisement assigning step for assigning an advertisement stored in an advertisement information database based on the estimated attribute profile of the user, and a distribution step for delivering the assigned advertisement to the user Delivery program.

  According to the configuration of (7), it is possible to provide an invention having the same operational effects as the device of (1) as a computer program.

  (8) A learning step for machine learning of a reference profile based on an expression history that is information transmitted from a user terminal and a user's known attribute information, and an estimation step for estimating a user's attribute profile based on the reference profile And setting a bid frame for advertisement information based on the estimated attribute profile of the user.

  According to the configuration of (8), the reference profile is machine-learned based on the expression history that is information transmitted from the user terminal and the known attribute information of the user, and the attribute profile of the user is estimated based on the reference profile Then, a bid frame for the advertisement information is set based on the estimated attribute profile of the user.

  In this way, it is possible to estimate the user profile using the reference profile based on the known attribute information and the expression history, and to set a bid frame for the advertisement information based on the estimated user attribute profile. it can. Therefore, since the set bid frame is highly likely to belong to a user to whom the advertisement information is distributed, the advertisement frame can be sold at a higher price.

  (9) In an advertisement distribution apparatus that includes an advertisement information database and uses information on a Web page that provides a question answering service, an acquisition unit that acquires a target that is a topic from the Web page, and an analysis unit that performs a morphological analysis on the target An estimation unit that estimates an attribute profile of the user of the Web page from the morpheme analyzed by the analysis unit, a determination unit that determines whether the user has experience with the target based on the attribute profile, When the result determined by the determination unit is experienced, an advertisement for experience is assigned to the object from the advertisement information database, and when there is no experience, the object is retrieved from the advertisement information database. An ad allocation unit that allocates an advertisement for those who have no experience to the user, and an advertisement allocated by the advertisement allocation unit. The advertisement distribution device and a distribution unit to be delivered to The.

  According to the configuration of (9), in an advertisement distribution apparatus that includes an advertisement information database and uses information on a Web page that provides a question answering service, the apparatus acquires a target that is a topic from the Web page, and The target is subjected to morphological analysis, the attribute profile of the user of the Web page is estimated from the analyzed morpheme, the presence or absence of the user's experience with the target is determined based on the attribute profile, and the determination unit determines If the result is experienced, the advertisement for the subject is assigned to the target from the advertisement information database. If the result is not experienced, the subject is not experienced for the subject from the advertisement information database. The advertisement is allocated, and the advertisement allocated by the advertisement allocation unit is distributed to the user.

  Thus, based on the estimated user attribute profile, it is possible to distribute advertisements according to the presence / absence of experience in the subject that is the user's topic in order to determine the presence / absence of experience in the subject that is the topic. .

  (10) The estimation unit estimates an attribute profile by calculating a probability that the morpheme analyzed by the analysis unit belongs to a known attribute profile stored in the profile database using a naive Bayes method. (9) The advertisement distribution apparatus according to (9).

  According to the configuration of (10), in the advertisement distribution device according to (9), the estimation unit calculates the probability that the morpheme analyzed by the analysis unit belongs to a known attribute profile stored in the profile database. The attribute profile is estimated by calculating using the naive Bayes method.

  Thus, the accuracy of estimating the attribute profile can be improved by estimating the user's attribute profile using the naive Bayes method.

  (11) The estimation unit estimates an attribute profile by determining whether the morpheme analyzed by the analysis unit belongs to a known attribute profile stored in the profile database using a support vector machine. (9) The advertisement distribution device according to (9).

  According to the configuration of (11), in the advertisement distribution device according to (9), the estimation unit determines whether the morpheme analyzed by the analysis unit belongs to a known attribute profile stored in the profile database. The attribute profile is estimated by discriminating these using a support vector machine.

  Thus, the accuracy of estimating the attribute profile can be improved by estimating the user's attribute profile using the support vector machine.

  (12) The advertisement distribution according to (9), wherein the estimation unit estimates an attribute profile by detecting an expression that meets a predetermined condition from the morphemes analyzed by the analysis unit. apparatus.

  According to the configuration of (12), in the advertisement distribution device according to (9), the estimation unit detects an expression that matches a predetermined condition from morphemes analyzed by the analysis unit. Estimate the profile.

  In this way, by detecting an expression that meets a predetermined condition, it is possible to improve the accuracy of estimating the attribute profile by estimating the user's attribute profile.

  (13) An acquisition step of acquiring a subject as a topic from a Web page that provides a question answering service, an analysis step of analyzing the morpheme of the target, and an estimation step of estimating an attribute profile of the user of the Web page from the analyzed morpheme And a determination step for determining whether or not the user has experience in the target based on the attribute profile, and if the determined result is experienced, the user is experienced from the advertisement information database to the target. If there is no experience, an advertisement distribution step comprising: an advertisement allocation step of distributing an advertisement for inexperience to the target from the advertisement information database; and a distribution step of distributing the allocated advertisement to the user Method.

  According to the structure of (13), the invention which exhibits the same effect as the apparatus of (9) can be provided as a method invention.

  (14) An acquisition step of acquiring a subject as a topic from a Web page that provides a question answering service, an analysis step of analyzing the morpheme of the target, and an estimation step of estimating an attribute profile of the user of the Web page from the morpheme A determination step for determining whether or not the user has experience with the target based on the attribute profile; and if the determined result is experienced, the advertisement information database is directed to the target with experience. If an advertisement is assigned and the user has no experience, the computer executes an advertisement assigning step for assigning an advertisement with no experience to the target from the advertisement information database, and a distribution step for delivering the assigned advertisement to the user. Advertising distribution program for.

  According to the configuration of (14), it is possible to provide an invention having the same operational effects as the device of (9) as a computer program.

  ADVANTAGE OF THE INVENTION According to this invention, a user's attribute profile can be estimated from a user's expression log | history, and advertisement information can be provided based on the estimated user's attribute profile. Therefore, it is possible to improve the accuracy of distributing advertisement information. Furthermore, there is no need to take action from the distributor side, and the advertisement distribution apparatus can automatically allocate advertisement information and distribute it to the user.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[Overview of the entire system]
FIG. 1 is a block diagram showing the overall configuration of a computer system 1 and the functional configuration of an advertisement distribution apparatus 10 according to an example of a preferred embodiment of the present invention. The advertisement distribution device 10 and the user terminal 20 are connected via a communication line 30. A plurality of Web servers 40 including a WebDB 41 having data on the Web site exist on the computer system 1 and are connected to the advertisement distribution apparatus 10 via the communication line 30. The advertisement distribution apparatus 10 may be configured to include a Web server, a DB server, and an application server as necessary, and may be configured with one server or different servers. The user terminal 20 is represented by a PC, a mobile, a PDA (Personal Digital Assistant) terminal or the like, and a web page or the like stored in the advertisement distribution device 10 or the web server 40 via the communication line 30 Any content can be used as long as the content can be referred to.

  The advertisement distribution apparatus 10 adds advertisement data to the page requested to be transmitted from the user terminal 20 and transmits the advertisement data to the requesting user terminal 20 together with the page requested to be transmitted. The control unit 50 and the storage unit 60.

  The control unit 50 controls the entire advertisement distribution apparatus 10 and includes a learning unit 51, an estimation unit 52, an advertisement allocation unit 53, and an advertisement distribution unit 54.

  The storage unit 60 stores data necessary for processing of the advertisement distribution apparatus 10 and includes a profile DB 61 and an advertisement DB 62. In the present embodiment, the profile DB 61 and the advertisement DB 62 are built in the advertisement distribution apparatus 10, but the present invention is not limited to this. The advertisement distribution apparatus 10 may be connected, and the advertisement distribution apparatus 10 may be connected from the outside via a network.

[First embodiment]
A first embodiment according to the present invention will be described below. FIG. 2 is a table showing a part of known user attribute information stored in the profile DB 61 in a table. The user attribute information includes, for example, the user ID, gender, age, occupation, and the like, and all elements of each attribute information are known. The known user attribute information is used to generate a reference profile (hereinafter referred to as a reference profile) to be described later.

  FIG. 3 is a diagram illustrating an example of an expression history (blog) transmitted by a user whose attribute information is known in FIG. A blog is a word that omits a web log, and is a website centered on a dated article page. Sites that can link to each article, can be commented on each article, and have new articles always displayed at the top are often called blogs. The blog shown in FIG. 3 is transmitted by a user having AAA user ID whose attribute information is known. The expression history transmitted by the user whose attribute information is known is used for generating a reference profile, which will be described later, together with the above-described known user attribute information.

  Here, the reference profile generated by the learning unit 51 will be described. The reference profile is generated based on the attribute information of the known user and the expression history transmitted by the user whose attribute information is known, and is used for the estimation of the attribute profile by the estimation unit described later. The learning unit acquires text data of a specific Web site, divides the acquired text data into morphemes by performing morphological analysis. Then, whether each morpheme belongs to a predetermined attribute profile is classified. Note that a morpheme is a word, phrase, characteristic wording, characteristic grammatical expression, etc., and not only a content word used in normal document classification, but also a sentence ending expression (for example, “is”, “ ”, Etc.”, impression verbs (eg, “Oh”, “Yata”, etc.), greetings (eg, “Ohayo”, “Sayonara”, etc.) are also used as morphemes.

  For example, you may classify | categorize using a naive Bayes system. Here, a method for estimating the attribute profile of the user who created the blog article using the naive Bayes method will be described.

First, the post article of the blog is (D1, D2... Dn), and the average language model of the user having the label l is θl. Here, the label l is an element constituting the attribute profile, for example, gender, age, occupation, and the like. The probability that the user u has the label l is P (l | u) = P (D1, D2,..., Dn | _θl ) / _ΣlεLP (D1, D2,..., Dn | θl ). However, L does not have an overlapping area, and one is always a set of labels to be given. (For example, 0-9 years old male, 10-19 years old male, 20-20 years old male, 30-39 years old male, 40-49 years old male, over 50 years old male, 0-9 years old female 10-19 years old, female, 20-20 years old, female, 30-39 years old, female, 40-49 years old, female, over 50 years old, female.) In addition, rather than considering the label separately for age and gender, Considering the above, the performance of estimating the attribute profile is improved.

In the Naive Bayes system, it is assumed that each article appears as an independent event, and P (D1, D2,..., Dn | θl) = Π ⁿ _{k = 1} P (Dk | θl).

In P (Dk | θl), it is assumed that the occurrence probability P (fm | θl) of individual morphemes (f1, f2,. P (Dk | θl) = Π ^m _{j = 1} P (fj | θl).

As described above, when the probability that the user u has the label l is modeled by the naive Bayes method, P (l | u) = Π ⁿ _{k = 1Π} ^m _{j = 1} P (fj | _θl ) / Σ _{lε L} Π ⁿ _{k = 1} Π ^m _{j = 1} P (fj | θl).

In the learning of the reference profile, the probability P (fk | θl) is learned from a user whose attribute information is known. P (fk | θl) = { Σ u∈Ul freq (fk, u) +1} / {Σ u∈Ul Σ m k = 1 freq (fk, u) + N}. Here, Ul is a set of users having the label l, N is the number of all vocabularies, and freq (fk, u) is the number of times the user u has expressed the morpheme fk.

  FIG. 4 is a flowchart showing a flow in which the control unit 50 performs machine learning of a reference profile based on the attribute information of a known user and the expression history transmitted by a user whose attribute profile is known.

  First, in step S <b> 1, the advertisement distribution apparatus 10 connects to the Web server 40 via the communication line 30, and the learning unit 51 receives text data of a specific Web site stored in the WebDB 41 of the Web server 40. get.

  Next, in step S2, the learning unit 51 performs morphological analysis of the text data acquired in step S1, and divides the text data into each morpheme.

  In step S3, the control unit 50 reads the known attribute information stored in the profile DB 61, and the learning unit 51 calculates the probability that each morpheme analyzed in step S2 belongs to a predetermined attribute profile. Here, the naive Bayes method described above is used to calculate the probability that each morpheme belongs to a predetermined attribute profile.

  In step S4, the learning unit 51 performs machine learning using the probability calculated in step S4 as a reference profile, and ends this process.

  In this way, the known attribute information and the expression history of the user having the attribute information are acquired, the acquired expression history is morphologically analyzed, and the naive Bayes method is used based on the analyzed morpheme and the known attribute information. The analyzed morpheme is classified by calculating the probability. Then, the machine profile learns the probability that each classified morpheme belongs to a predetermined attribute profile, thereby generating a reference profile. Such processing enables stable and high-speed machine learning.

  FIG. 5 is a diagram showing the reference profile calculated by the machine learning flow of FIG. 4 as a table. FIG. 5A shows the probability that each morpheme belongs to an attribute profile classified by gender and age. Further, in FIG. 5B, the probability that each morpheme belongs to the attribute profile classified by occupation.

  FIG. 6 is a table showing attribute profiles of users who are partly known and stored in the profile DB 61. The user attribute profile includes, for example, a user ID, gender, age, occupation, and the like. However, the elements of each attribute profile include unknown elements. In the example of FIG. 4, only the user ID is known, but other elements are unknown. Therefore, the estimation unit 52 estimates the user's attribute profile using the reference profile.

  FIG. 7 is a diagram showing an example of an expression history (blog) transmitted by a user whose attribute profile is partially known in FIG. The blog in FIG. 7 is transmitted by a user having an EEE user ID whose attribute profile is partially known. For the expression history transmitted by the user whose attribute profile is partially known, the estimation unit 52 estimates the attribute profile of the user using the reference profile together with the attribute profile of the partially known user.

  FIG. 8 shows a flow in which the advertisement distribution apparatus 10 estimates a user attribute profile using a machine-learned reference profile, allocates an advertisement based on the estimated user attribute profile, and distributes the allocated advertisement. It is a flowchart.

  First, in step S <b> 11, the advertisement distribution apparatus 10 connects to the Web server 40 via the communication line 30, and the estimation unit 52 uses a partially known attribute profile stored in the WebDB 41 of the Web server 40. The text data of the expression history (blog) by the possessed user is acquired.

  In step S12, the estimation unit 52 performs morphological analysis of the text data acquired in step S11, and divides the text data into each morpheme.

  In step S13, the estimation unit 52 reads the user's attribute profile that is partly stored in the profile DB 61, and uses the reference profile together with the morpheme analyzed in step S12, so that the part is already known. Estimate the attribute profile to which the user belongs.

  In step S <b> 14, the advertisement allocation unit 53 allocates advertisement information corresponding to the user attribute profile estimated by the estimation unit 52 from the advertisement DB 62.

  In step S15, the advertisement distribution unit 54 distributes the advertisement information allocated by the advertisement allocation unit 53 to the user whose estimation unit 52 has estimated the attribute profile in step S13.

  In this way, the expression history of a user who has a partly known attribute profile is subjected to morphological analysis, and the attribute profile of each morpheme analyzed using the reference profile is estimated. Then, advertisement information corresponding to the estimated attribute profile is allocated from the advertisement DB, and the allocated advertisement is distributed to the user. By such processing, the user's attribute profile can be estimated based on the partially known attribute profile and the expression history. For this reason, it is not necessary to take action from the distributor side, and the advertisement distribution apparatus can automatically allocate advertisement information and distribute it to the user.

[Second Embodiment]
Next, a second embodiment according to the present invention will be described below. FIG. 9 is a table showing a part of known user attribute information stored in the profile DB 61 in a table. The user profile is composed of, for example, the user's gender, age, occupation, etc., and all the elements of the user attribute information are known. The known user attribute information is used to generate a reference profile to be described later.

  Next, FIG. 10 is a diagram showing an example of an expression history (Web page) written by a user whose user attribute information is known in FIG. The Web page shown in FIG. 10 is intended to provide a question answering service that allows users to ask and answer questions. For example, when a user writes a question sentence about information he / she wants to know, the written question sentence is displayed on the Web page. Another user who viewed the written question sentence writes the answer to the question, and the third party who viewed the question sentence and the answer sentence evaluates whether the question sentence and the answer sentence were useful information for the third party. Is a service. In the example of FIG. 10, a user with a BBB user ID answers a question sentence asked by a user with an AAA user ID, and a third party user sends a question sentence and an answer sentence to a third party. The result of evaluating whether it was useful information is displayed. The expression history transmitted by the user whose attribute information is known is used for generating a reference profile, which will be described later, together with the above-described known user attribute information.

  Here, the reference profile generated by the learning unit 51 will be described. The reference profile is generated based on the attribute information of the known user and the expression history transmitted by the user whose attribute information is known, and is used for the estimation of the attribute profile by the estimation unit 52 described later. The learning unit 51 acquires text data of a specific Web site, divides the acquired text data into morphemes by performing morphological analysis. Then, whether each morpheme belongs to a predetermined attribute profile is classified.

  For example, you may classify | categorize using a support vector machine (SVM: Support Vector Machine). A support vector machine is a learning system that uses a hypothesis space of a linear function in a high-dimensional feature space. As a result of learning, the reference profile includes an identification surface for identifying a set of data that is correct as being a predetermined attribute profile and a set of data that is incorrect.

  FIG. 11 is a diagram showing an example in which a reference profile is created by inputting a large number of morphemes and statistically processing these morphemes by a support vector machine. In FIG. 11, the hypothesis space is shown in two dimensions, but the coordinate axes of the vertical axis and the horizontal axis may correspond to each morpheme as described above. That is, there may be as many coordinate axes as the number of morphemes, and the hypothesis space may be a space having the number of morpheme dimensions. In this way, the support vector machine places a large number of morphemes in the hypothesis space. And a morpheme group is comprised from arrangement | positioning of many morphemes, and an identification surface is produced | generated. A reference profile is an identification surface defined by a support vector.

  A hypothesis space is generated for each attribute profile. FIG. 11 is a hypothetical space indicating whether or not the mobile phone of company S is owned. In this hypothesis space, a reference profile is generated by generating an identification plane. This identification surface is a standard for distinguishing whether it belongs to the attribute profile that owns the mobile phone of company S is correct or not (not correct).

  Next, a process in which the learning unit 51 generates a reference profile based on the above-described known attribute information and a user's expression history having the known attribute information will be described.

  FIG. 12 is a flowchart illustrating a flow in which the control unit 50 performs machine learning of the reference profile based on the known user attribute information and the expression history transmitted by the user whose attribute information is known.

  First, in step S <b> 21, the advertisement distribution apparatus 10 connects to the Web server 40 via the communication line 30, and the learning unit 51 receives text data of a specific Web page stored in the WebDB 41 of the Web server 40. get.

  Next, in step S22, the learning unit 51 performs morphological analysis of the text data acquired in step S21, and divides the text data into each morpheme.

  In step S23, the control unit 50 reads the known attribute information stored in the profile DB 61, and the learning unit 51 determines whether or not the morpheme analyzed in step S22 belongs to a predetermined attribute profile. Use to classify.

  In step S24, the learning unit 51 performs machine learning on the identification plane defined by the support vector classified in step S23 as a reference profile, and ends this process.

  As described above, the expression history by the user having the known attribute information is morphologically analyzed, and using the support vector machine, the attribute profile to which each analyzed morpheme belongs is classified. Then, machine learning is performed using the identification surface into which each morpheme is classified as a reference profile. By such processing, the reference profile is machine-learned, so that stable and high-speed machine learning is possible.

  FIG. 13 is a table showing the attribute profiles of users who are partly stored in the profile DB 61 in a table. The user attribute profile includes a user ID, sex, age, occupation, and the like. However, the elements of each attribute profile include elements that are unknown. In the example of FIG. 13, only the user ID is known, but other elements are unknown. For this reason, the advertisement distributor cannot distribute the advertisement based on the attribute profile of the user. Therefore, based on the expression history of the user who has a partially known profile, the user's attribute profile is estimated using the above-described reference profile.

  FIG. 14 is a diagram illustrating an example of an expression history (Web page) transmitted by a user whose attribute profile is partially known in FIG. The Web page of FIG. 14 is intended to provide a question answering service that allows users to ask and answer questions as in the Web page of FIG. 10 described above. For the expression history transmitted by the user whose attribute profile is partially known, the estimation unit 52 estimates the attribute profile of the user using the reference profile together with the attribute profile of the partially known user.

  FIG. 15 is a flowchart illustrating a flow in which an advertisement distribution apparatus estimates a user attribute profile using a machine-learned reference profile, allocates an advertisement based on the estimated user attribute profile, and distributes the allocated advertisement. It is.

  First, in step S31, the advertisement distribution apparatus 10 is connected to the Web server 40 via the communication line 30, and the estimation unit 52 of the control unit 50 is stored in the WebDB 41 of the Web server 40. In the Web page providing the answer service, the text data of the expression history (Web page) by the user who has a partially known attribute profile is acquired.

  In step S32, the estimation unit 52 performs morphological analysis of the text data acquired in step S31, and divides the text data into morphemes.

  In step S <b> 33, the estimation unit 52 finds a product that is a topic on the Web page that provides the question answering service from the analyzed morphemes.

  In step S34, the estimation unit 52 of the control unit 50 reads out a user attribute profile that is partly stored in the profile DB 61, and is generated by the learning unit based on the morpheme analyzed in step S32. An attribute profile to which a partially known user belongs is estimated using an identification plane defined by a support vector that is a reference profile. For example, as shown in FIG. 16, it is estimated that the correct answer case set owns the mobile phone of company S, while the incorrect answer case set is assumed not to own the mobile phone of company S. .

In step S35, the control unit 50 determines from the estimation result in step S34 whether the user has purchased the product that is the topic and is skilled. If the user has purchased the product as a topic and is skilled (Yes), the process proceeds to step S36. On the other hand, when the user has not purchased or is not skilled at the topic item (No), the process proceeds to step S38.

  In step S <b> 36, the control unit 50 stores in the attribute profile of the questioner in the profile DB 61 that the product as the topic has been purchased / skilled.

  In step S37, since the user has purchased / skilled the topic product, the advertisement requesting unit 53 assumes that the advertisement desired by the user is a top similar product / related product / consumable product of the topic product. From the advertisement DB 62, the top similar product / related product / consumable product advertisement of the topic product is allocated, and the process proceeds to step S40. Here, the top similar product refers to a product similar to the product classified as the top of the topic product (for example, the top classified product of the employment information for university students is a product for university students, and the top similar product This applies to products for graduate students.)

  In step S <b> 38, the control unit 50 stores in the attribute profile of the questioner in the profile DB 61 that the topic product is not purchased or is unskilled.

  In step S39, the advertisement allocation unit 53 determines that the advertisement desired by the user is a similar product to the topic product, since the user has not purchased the product that is the topic and is not skilled, and the product that becomes the topic from the advertisement DB 62. The advertisement of the similar product is allocated, and the process proceeds to step S40.

  In step S40, the advertisement distribution unit 54 distributes the advertisement allocated by the advertisement allocation unit 53 to the user who is the questioner.

  In this way, the expression history of a user who has a partly known attribute profile is subjected to morphological analysis, and the attribute profile of each morpheme analyzed using the reference profile is estimated. Then, advertisement information corresponding to the estimated attribute profile is allocated from the advertisement DB, and the allocated advertisement is distributed to the user. By such processing, the user's attribute profile can be estimated based on the partially known attribute profile and the expression history. For this reason, it is not necessary to take action from the distributor side, and the advertisement distribution apparatus can automatically allocate advertisement information and distribute it to the user. Furthermore, by classifying morphemes using a support vector machine, it is possible to improve the accuracy with which the estimation unit estimates a user's attribute profile.

  In the second embodiment, the questioner's attribute profile is estimated based on the questioner's expression history in the question answering service. However, the present invention is not limited to this. The attribute profile of the user may be estimated based on the expression history of the respondent in the question answering service, the evaluator who evaluated the question text, and the questioner who viewed the answer text. In this case, for example, the text data of the answer sentence to the question sentence is acquired, the text data is morphologically analyzed, the product newly mentioned in the answer sentence is found from the analyzed morpheme, and the advertisement of the discovered product is displayed. You may distribute to an answerer, an evaluator, and a questioner at the time of answer reading. Thereby, the advertisement of the product newly mentioned by the answer sentence to a question can be shown to a user.

  When the evaluator evaluates the question sentence and the answer sentence, the text data of the question sentence and the answer sentence is acquired, the obtained text data is subjected to morphological analysis, and the question sentence and the answer are analyzed from the analyzed morpheme. When a product that becomes a topic in a sentence is found, and the evaluator evaluates the found product as a useful question and answer, it is stored in the evaluator's attribute profile that the discovered product is unpurchased and unskilled. If it is judged that the question and the answer are not useful, it is stored in the evaluator's attribute profile that the found product has been purchased / skilled. If the attribute profile of the discovered product is not purchased or not skilled, the top similar product / related product / consumable product advertisement of the topic product is allocated from the advertisement DB 62, and the allocated advertisement is sent to the evaluator. To deliver. On the other hand, if the attribute profile for the found product has been purchased and is skilled, an advertisement for a similar product of the product as a topic is allocated from the advertisement DB 62, and the allocated advertisement is distributed to the evaluator. Thereby, the advertisement according to evaluation with respect to the evaluator's question text and answer text can be distributed.

  In the above-described second embodiment, the support vector machine is used for classification, but the present invention is not limited to this. It is also possible to classify using the result of detecting an expression that matches a predetermined condition from the expression history. For example, in the question answering service shown in FIG. 10 and FIG. 14, by detecting a specific expression such as “I bought ...” from the question sentence, the learning unit 51 sets the reference profile. The generation and estimation unit 52 can estimate an individual attribute profile using the learned reference profile. Thereby, a user's attribute profile can be simply estimated from the expression history.

  In the first and second embodiments described above, the method for distributing advertisement information using the estimated user attribute profile has been described. However, the advertisement is bid using the estimated user attribute profile. It is also possible to provide a method.

  The advertising bidding method performs machine learning of a reference profile based on a user's expression history and a known attribute profile, estimates a user's attribute profile based on the learned reference profile, and performs advertisement based on the estimated user's attribute profile. Set up bids for information. By performing such a process, the set bid frame is likely to belong to a user to whom the advertisement information is to be distributed, so that the advertisement frame can be sold at a higher price.

[Hardware Configuration of Advertisement Distribution Device 10]
FIG. 17 is a diagram illustrating a hardware configuration of the advertisement distribution apparatus 10 according to an example of the preferred embodiment of the present invention. The advertisement distribution apparatus 10 includes a CPU (Central Processing Unit) 110 (a CPU 120 including a CPU 120 in a multiprocessor configuration) that configures a control unit 50 including a learning unit 51, an estimation unit 52, an advertisement allocation unit 53, and an advertisement distribution unit 54. Bus line 105, communication I / F 140, main memory 150, BIOS (Basic Input Output System) 160, USB port 190, I / O controller 170, keyboard and mouse 180, and other input means and display devices 122.

  Storage means such as a tape drive 172, a hard disk 174, an optical disk drive 176, and a semiconductor memory 178 can be connected to the I / O controller 170.

  The BIOS 160 stores a boot program executed by the CPU 110 when the advertisement distribution apparatus 10 is started, a program depending on the hardware of the advertisement distribution apparatus 10, and the like.

  The hard disk 174 constituting the storage unit 60 including the profile DB 61 and the advertisement DB 62 stores various programs for functioning as the advertisement distribution apparatus 10 and programs for executing the functions of the present invention, and various databases as necessary. Can be configured.

  As the optical disk drive 176, for example, a DVD-ROM drive, a CD-ROM drive, a DVD-RAM drive, or a CD-RAM drive can be used. In this case, the optical disk 177 corresponding to each drive is used. A program or data may be read from the optical disk 177 by the optical disk drive 176 and provided to the main memory 150 or the hard disk 174 via the I / O controller 170. Similarly, the tape medium 171 corresponding to the tape drive 172 can be used mainly for backup.

  The program provided to the advertisement distribution apparatus 10 is provided by being stored in a recording medium such as the hard disk 174, the optical disk 177, or a memory card. This program may be installed in the advertisement delivery apparatus 10 and executed by being read from the recording medium via the I / O controller 170 or downloaded via the communication I / F 140.

  The aforementioned program may be stored in an internal or external storage medium. Here, in addition to the hard disk 174, the optical disk 177, or the memory card, a magneto-optical recording medium such as an MD, or a tape medium 171 can be used as a storage medium constituting the storage unit 60. Further, a storage device such as a hard disk 174 or an optical disk library provided in a server system connected to a dedicated communication line or the Internet may be used as a recording medium, and the program may be provided to the advertisement distribution apparatus 10 via the communication line. .

  Here, the display device 122 displays a screen for accepting data input to the administrator of the advertisement distribution device 10 or displays a calculation processing result screen by the advertisement distribution device 10. CRT) and display devices such as a liquid crystal display (LCD).

  Here, the input means accepts input by the administrator of the advertisement distribution apparatus 10 and may be configured by a keyboard and a mouse 180 or the like.

  The communication I / F 140 is a network adapter for enabling the advertisement distribution apparatus 10 to be connected to a terminal via a dedicated network or a public network. The communication I / F 140 may include a modem, a cable modem, and an Ethernet (registered trademark) adapter.

  In the above example, the advertisement distribution apparatus 10 has been mainly described. However, the functions described above can be realized by installing a program in a computer and causing the computer to operate as the advertisement distribution apparatus 10. Therefore, the functions realized by the advertisement distribution apparatus 10 described as an embodiment in the present invention are executed by executing the above-described method by the computer, or by introducing the above-described program into the computer and executing it. Is also feasible.

[Hardware Configuration of Web Server 40]
The Web server 40 can also be realized with the same configuration as that of the advertisement distribution apparatus 10 described above.

[Hardware Configuration of User Terminal 20]
The user terminal 20 can also be realized with a configuration similar to that of the above-described advertisement distribution apparatus 10. Further, in the above-described example, an example realized by a so-called computer has been described. However, as long as the principle of the present invention is applicable, it may be realized by various terminals such as a mobile terminal and a PDA terminal.

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

It is a block diagram showing the whole structure of a computer system, and the function structure of an advertisement delivery apparatus. It is the figure which showed the known attribute information in the table. It is a figure which shows the expression log | history (blog) of a user with known attribute information. It is the figure which showed the attribute profile partially known in the table. It is a figure which shows the expression history (blog) of the user who has a partly known attribute profile. It is a flowchart which shows the learning process of a reference | standard profile. It is the figure which showed the reference | standard profile in the table. It is a flowchart which shows the estimation process of an attribute profile, an advertisement allocation process, and an advertisement delivery process. It is the figure which showed the known attribute information in the table. It is a figure which shows the expression history (Web page) of a user with known attribute information. It is the figure which showed the learning result of the reference | standard profile. It is a flowchart which shows the learning process of a reference | standard profile. It is the figure which showed the attribute profile partially known in the table. It is a figure which shows the expression log | history (Web page) of the user who has a partly known attribute profile. It is a flowchart which shows the estimation process of an attribute profile, an advertisement allocation process, and an advertisement delivery process. It is the figure which showed the estimation result of the attribute profile. It is a figure which shows hardware constitutions, such as advertisement delivery apparatus.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Computer system 10 Advertising distribution apparatus 20 Terminal 30 Communication line 40 Web server 41 WebDB
DESCRIPTION OF SYMBOLS 50 Control part 51 Learning part 52 Estimation part 53 Advertisement allocation part 54 Advertisement distribution part 60 Storage part 61 Profile DB
62 Advertising DB
105 Bus line 110 CPU1
120 CPU2
122 Display device 140 Communication I / F
150 main memory 160 BIOS
170 I / O Controller 171 Tape Media 172 Tape Drive 174 Hard Disk 176 Optical Disk Drive 177 Optical Disk 178 Semiconductor Memory 180 Keyboard / Mouse 190 USB Port

Claims (14)

In an advertisement distribution apparatus provided with an advertisement information database,
A learning unit that machine-learns the reference profile based on the expression history that is information transmitted from the user terminal and the known attribute information of the user,
An estimation unit for estimating a user attribute profile based on the reference profile;
An ad allocation unit that allocates an advertisement stored in the advertisement information database based on the attribute profile of the user estimated by the estimation unit;
An advertisement distribution apparatus comprising: a distribution unit that distributes the advertisement allocated by the advertisement allocation unit to the user. The learning unit
An acquisition unit for acquiring the expression history;
An analysis unit that performs a morphological analysis on an expression history that is text information acquired by the acquisition unit;
A classification unit that classifies morphemes analyzed by the analysis unit based on the attribute information;
A generation unit that generates the reference profile based on morphemes classified by the classification unit;
The estimation unit includes
A user expression history acquisition unit for acquiring an expression history by the user;
A user expression history analysis unit that performs a morphological analysis of the expression history by the user acquired by the user expression history acquisition unit, using a reference profile generated by the generation unit, a part of the attribute profile of the user, The advertisement distribution apparatus according to claim 1, wherein the attribute profile of the user is estimated based on the morpheme analyzed by the user expression history analysis unit. The classification unit includes:
The morpheme analyzed by the analysis unit classifies the morpheme by calculating a probability that the morpheme belongs to the attribute information by using a naive Bayes method, and the reference profile generated by the generation unit has a predetermined morpheme Is the probability of belonging to the attribute profile of
The estimation unit includes
The attribute profile of the user is estimated using a probability that the morpheme classified by the classification unit belongs to a predetermined attribute profile as the reference profile generated by the generation unit. Advertising distribution device. The classification unit includes:
The morpheme analyzed by the analysis unit is classified by using a support vector machine to determine whether the morpheme belongs to the attribute information, and the reference profile generated by the generation unit is defined by the support vector. Identification surface
The estimation unit includes
The advertisement distribution apparatus according to claim 2, wherein the attribute profile of the user is estimated using the identification plane as a reference profile generated by the generation unit. The classification unit includes:
The morpheme analyzed by the analysis unit classifies the morpheme by detecting an expression that meets a predetermined condition,
The estimation unit includes
The advertisement distribution according to claim 2, wherein the attribute profile of the user is estimated using a result of detecting an expression in which the morpheme meets a predetermined condition as a reference profile generated by the generation unit. apparatus. A learning step for machine learning of a reference profile based on an expression history that is information transmitted from a user terminal and user's known attribute information;
An estimating step of estimating an attribute profile of the user based on the reference profile;
An ad assignment step for assigning an advertisement stored in an advertisement information database based on the estimated attribute profile of the user;
A distribution step of distributing the allocated advertisement to the user. A learning step for machine learning of a reference profile based on an expression history that is information transmitted from a user terminal and user's known attribute information;
An estimating step of estimating an attribute profile of the user based on the reference profile;
An ad assignment step for assigning an advertisement stored in an advertisement information database based on the estimated attribute profile of the user;
An advertisement distribution program for causing a computer to execute a distribution step of distributing an allocated advertisement to the user. A learning step for machine learning of a reference profile based on an expression history that is information transmitted from a user terminal and user's known attribute information;
An estimating step of estimating an attribute profile of the user based on the reference profile;
An advertisement bidding method comprising: setting a bid frame for advertisement information based on the estimated attribute profile of the user. In an advertisement distribution apparatus that includes an advertisement information database and uses information on a Web page that provides a question answering service,
An acquisition unit for acquiring a subject as a topic from the Web page;
An analysis unit for morphological analysis of the object;
An estimation unit for estimating an attribute profile of the user of the Web page from the morpheme analyzed by the analysis unit;
A determination unit that determines whether or not the user has experience with the target based on the attribute profile;
If the result of the determination by the determination unit is experienced, an advertisement for experience is assigned to the target from the advertisement information database, and if there is no experience, the advertisement information database An ad servicing section that targets ads with no experience to the target,
An advertisement distribution apparatus comprising: a distribution unit that distributes the advertisement allocated by the advertisement allocation unit to the user. The estimation unit includes
The advertisement distribution apparatus according to claim 9, wherein the attribute profile is estimated by calculating a probability that the morpheme analyzed by the analysis unit belongs to the attribute profile using a naive Bayes method. The estimation unit includes
The advertisement distribution apparatus according to claim 9, wherein the attribute profile is estimated by determining, using a support vector machine, whether or not the morpheme analyzed by the analysis unit belongs to the attribute profile. The estimation unit includes
The advertisement distribution apparatus according to claim 9, wherein the attribute profile is estimated by detecting an expression that matches a predetermined condition from morphemes analyzed by the analysis unit. An acquisition step of acquiring a subject as a topic from a web page that provides a question answering service;
An analysis step for morphological analysis of the object;
An estimation step of estimating an attribute profile of the user of the Web page from the analyzed morpheme;
A determination step of determining whether or not the user has experience with the target based on the attribute profile;
If the determined result is experienced, the advertisement information database is assigned an advertisement for experience to the target. If the result is not experienced, the advertisement information database is directed to the target for no experience. An ad allocation step to allocate ads,
A distribution step of distributing the allocated advertisement to the user. An acquisition step of acquiring a subject as a topic from a web page that provides a question answering service;
An analysis step for morphological analysis of the object;
An estimation step of estimating an attribute profile of the user of the Web page from the analyzed morpheme;
A determination step of determining whether or not the user has experience with the target based on the attribute profile;
If the determined result is experienced, the advertisement information database is assigned an advertisement for experience to the target. An ad allocation step to allocate ads,
An advertisement distribution program for causing a computer to execute a distribution step of distributing an allocated advertisement to the user.

Images