JP2014174753A - Information processing unit, information processing method, and information processing program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an information processing unit configured such that in a case where a click prediction model is learned using a difference vector, a characteristic vector is generated so that demographic information and so on that are common to each click can be used for the formation of a click prediction model.SOLUTION: Based on log data that is generated each time when an advertisement in a Web page on which a plurality of advertisements are displayed is clicked and that relates to combinations of advertisements clicked and advertisements not clicked, characteristic vectors for selections of advertisements to be displayed are generated. Using the log data, a click percentage in a combination with an element composing the characteristic vector of the advertisement is calculated for each attribute relating to a person who clicked the advertisement. A click percentage with respect to the target advertisement of the characteristic vector and corresponding to a specific element composing the characteristic vector is extracted, and this click percentage is added as an element of the characteristic vector.

Description

  The present invention relates to an information processing apparatus that generates a feature vector for selecting an advertisement to be displayed based on log data relating to an advertisement click.

  In an advertisement on a web page, an advertisement is selected and displayed corresponding to a user. For example, in order to measure how easily a certain advertisement is clicked according to the browsed web page and the user's interest, a model that predicts the click rate using the click log of the advertisement as learning data uses machine learning. Has been built. In this click prediction model, a difference between feature vectors of a clicked advertisement and an advertisement not clicked is taken within one request, and a prediction model is constructed using the difference vector as learning data. By doing in this way, the information that the advertisement was chosen from a plurality of advertisements can be learned, and a highly accurate prediction model can be created.

  As a technology related to this, for example, in Patent Document 1, it is an object to improve the numerical update frequency for necessary advertisements without significantly increasing the overall load, and the log data of the CTR calculation target is improved. Among them, apart from the time interval for calculating the CTR of all advertisements (for example, half a day), the CTR is calculated only for an advertisement whose display count satisfies a predetermined condition at a shorter second time interval (for example, 30 minutes). Thus, it is disclosed that the CTR update frequency for a necessary advertisement can be improved as compared with the conventional technique without significantly increasing the overall load.

JP 2011-192224 A

However, because the difference between advertisements in one request is taken, demographic information such as gender and age, and context information such as day of the week, time zone, and location are negated, and it cannot be used as click prediction information There's a problem. In other words, because the demographic information or context information is the same for the clicked advertisement and the advertisement that was not clicked, even if it is adopted as an element of the feature vector, it can be canceled out, and there is no information about the element. turn into. For this reason, for example, a click prediction model is not created using information such as “an advertisement easy for men to click”.
The present invention has been made in the background of such a background art, and when learning a click prediction model with a difference vector, a click rate reflecting demographic information or context information common to each click. It is an object of the present invention to provide an information processing apparatus, an information processing method, and an information processing program that generate feature vectors so that they can be used to create a click prediction model.

The gist of the present invention for achieving the object lies in the inventions of the following items.
(1) An advertisement to be displayed based on log data that is generated each time an advertisement is clicked on a web page on which a plurality of advertisements are displayed, and that is based on log data relating to a combination of an advertisement that has been clicked and an advertisement that has not been clicked. Clicks in combination with feature vector generation means for generating a feature vector for selecting and for each attribute relating to the person who clicked on the advertisement using log data, with the elements constituting the feature vector of the advertisement A calculating means for calculating a rate, an extracting means for extracting a click rate corresponding to a predetermined element constituting the feature vector, the click rate for the advertisement targeted by the feature vector, and extracted by the extracting means The click rate is added as an element of the feature vector generated by the feature vector generating means An information processing apparatus characterized by comprising pressurizing means.

  (2) using the log data, further comprising first learning means for generating a model relating to an advertisement and an attribute by learning, wherein the calculating means uses the model learned by the first learning means, The information processing apparatus according to (1), wherein a click rate is calculated.

  (3) A collection unit that collects log data related to an advertisement click, and a difference vector that generates a difference vector between a clicked advertisement and a feature vector of an advertisement that has not been clicked when the collection unit collects log data. A generating unit, a second learning unit that learns a criterion for selecting an advertisement to be displayed using the difference vector generated by the difference vector generating unit, and a reference learned by the second learning unit. Therefore, the information processing apparatus according to (1) or (2), further including selection means for selecting an advertisement to be displayed.

  (4) An information processing method performed by an information processing apparatus including a feature vector generation unit, a calculation unit, an extraction unit, and an addition unit, wherein the feature vector generation unit includes a web page displaying a plurality of advertisements. Generating a feature vector for selecting an advertisement to be displayed based on log data generated each time an advertisement is clicked, the log data relating to a combination of a clicked advertisement and an advertisement not clicked; The log data is used to calculate, for each attribute related to the person who clicked on the advertisement, the click rate in combination with the elements constituting the feature vector of the advertisement. An attribute corresponding to a predetermined element constituting the feature vector, and the additional hand Is an information processing method characterized by adding a click rate extracted by said extracting means as elements of the feature vectors generated by the feature vector generation unit.

  (5) Based on log data relating to a combination of a clicked advertisement and an advertisement that has not been clicked, log data that is generated each time an advertisement is clicked on a web page on which a plurality of advertisements are displayed, Feature vector generation means for generating a feature vector for selecting an advertisement to be displayed, and elements constituting the feature vector of the advertisement for each attribute related to a person who clicked on the advertisement using log data A calculation means for calculating a click rate in a combination; an extraction means for extracting an attribute corresponding to a predetermined element constituting the feature vector, the extraction ratio being a click rate for the advertisement targeted by the feature vector; The extracted click rate is used as an element of the feature vector generated by the feature vector generating means. An information processing program to function as addition means for adding.

  According to the information processing apparatus, the information processing method, and the information processing program according to the present invention, when learning a click prediction model using a difference vector, a click rate reflecting demographic information or context information that is common to each click Can be used to generate a feature vector so that it can be used to create a click prediction model.

It is a conceptual module block diagram about the structural example of this Embodiment. It is a flowchart which shows the process example by this Embodiment. It is explanatory drawing which shows the data structure example of a click log table. It is explanatory drawing which shows the example of a data structure of a feature vector table. It is explanatory drawing which shows the example of a data structure of a user attribute table. It is explanatory drawing which shows the example of a data structure of an advertisement attribute table. It is explanatory drawing which shows the example of a data structure of the attribute / word click rate table. It is explanatory drawing which shows the data structure example of a feature vector. It is a flowchart which shows the process example by this Embodiment. It is a block diagram which shows the hardware structural example of the computer which implement | achieves this Embodiment.

Hereinafter, an example of a preferred embodiment for realizing the present invention will be described with reference to the drawings.
FIG. 1 shows a conceptual module configuration diagram of a configuration example of the present embodiment.
The module generally refers to components such as software (computer program) and hardware that can be logically separated. Therefore, the module in the present embodiment indicates not only a module in a computer program but also a module in a hardware configuration. Therefore, the present embodiment is a computer program for causing these modules to function (a program for causing a computer to execute each procedure, a program for causing a computer to function as each means, and a function for each computer. This also serves as an explanation of the program and system and method for realizing the above. However, for the sake of explanation, the words “store”, “store”, and equivalents thereof are used. However, when the embodiment is a computer program, these words are stored in a storage device or stored in memory. It is the control to be stored in the device. Modules may correspond to functions one-to-one, but in mounting, one module may be configured by one program, or a plurality of modules may be configured by one program, and conversely, one module May be composed of a plurality of programs. The plurality of modules may be executed by one computer, or one module may be executed by a plurality of computers in a distributed or parallel environment. Note that one module may include other modules. Hereinafter, “connection” is used not only for physical connection but also for logical connection (data exchange, instruction, reference relationship between data, etc.). “Predetermined” means that the process is determined before the target process, and not only before the process according to this embodiment starts but also after the process according to this embodiment starts. In addition, if it is before the target processing, it is used in accordance with the situation / state at that time or with the intention to be decided according to the situation / state up to that point. When there are a plurality of “predetermined values”, they may be different values, or two or more values (of course, including all values) may be the same. In addition, the description having the meaning of “do B when it is A” is used in the meaning of “determine whether or not it is A and do B when it is judged as A”. However, the case where it is not necessary to determine whether or not A is excluded.
In addition, the system or device is configured by connecting a plurality of computers, hardware, devices, and the like by communication means such as a network (including one-to-one correspondence communication connection), etc., and one computer, hardware, device. The case where it implement | achieves by etc. is also included. “Apparatus” and “system” are used as synonymous terms. Of course, the “system” does not include a social “mechanism” (social system) that is an artificial arrangement.
In addition, when performing a plurality of processes in each module or in each module, the target information is read from the storage device for each process, and the processing result is written to the storage device after performing the processing. is there. Therefore, description of reading from the storage device before processing and writing to the storage device after processing may be omitted. Here, the storage device may include a hard disk, a RAM (Random Access Memory), an external storage medium, a storage device via a communication line, a register in a CPU (Central Processing Unit), and the like.

The information processing apparatus according to the present embodiment is log data generated for each click of an advertisement in a web page on which a plurality of advertisements are displayed, and is a log relating to a combination of a clicked advertisement and an advertisement not clicked Based on the data, a feature vector for selecting an advertisement to be displayed is generated. As shown in the example of FIG. 1, a log storage module 110, a feature vector generation module 120, and a click prediction model generation module 150 are provided. Have. The feature vector generation module 120 or the click prediction model generation module 150 may perform processing every time a click occurs (log data is generated), or may perform processing every predetermined period.
Here, an outline of processing performed by the information processing apparatus according to the present embodiment will be described using an example.
(1) There is a web page on which three advertisements A, B, and C are posted, and a click log of user A (attribute: male) generated by a request for the web page is acquired. For example, log data “not clicked” for the advertisement A, “clicked” for the advertisement B, and “not clicked” for the advertisement C will be generated.
(2) Then, an attribute: male click rate is added as an element of the feature vector of advertisements A, B, and C. That is, the click rates of advertisements A, B, and C are extracted from a model generated from a click log only for men, and added as elements of respective feature vectors. Therefore, for example, when the information including the words included in the advertisements A, B, and C is applied to the model, the click rate is calculated.
(3) Next, as teacher data, for example, advertisement B-advertisement A (positive data), advertisement B-advertisement C (positive data), advertisement A-advertisement B (negative data), advertisement C-advertisement B (negative data) ) Is generated. That is, even if the difference vector is generated, the elements reflecting the demographic information or the context information are not canceled out.
(4) The ranking learning is performed from the difference vector in (3).

The log storage module 110 is connected to the feature vector generation module 120 and the click prediction model generation module 150. The log storage module 110 is a log indicating that an advertisement in a web page is displayed or clicked by operating a PC or a mobile terminal (for example, a mobile phone, a smartphone, or the like) owned by the user on the Internet. I remember the data. For example, the click log table 300 is stored. FIG. 3 is an explanatory diagram showing an example of the data structure of the click log table 300. The click log table 300 includes a click date / time column 310, an advertisement ID column 320, a user ID column 330, a URL column 340, and a location column 350, and information indicating that an advertisement displayed on a web page has been clicked ( Hereinafter, it is also referred to as a click log). The click date / time column 310 stores the date / time when the advertisement was clicked (click date / time). The advertisement ID column 320 stores an advertisement ID (IDentification) that uniquely identifies the clicked advertisement in the present embodiment. The user ID column 330 stores a user ID that uniquely identifies the user who clicked the advertisement in the present embodiment. The URL column 340 stores a URL (Uniform Resource Locator) indicating a web page on which an advertisement is displayed. The location column 350 stores the location where the click was made (for example, latitude and longitude). This location data is measured by GPS (Global Positioning System) or the position of a base station in a mobile phone network.
In addition to the click log, information (hereinafter also referred to as an impression log) indicating that an advertisement is displayed on a web page (hereinafter referred to as an impression) may be stored. For example, in the impression log, the date and time when an advertisement was displayed on a Web page (that is, the date and time when the Web page was transmitted from the Web server; hereinafter also referred to as distribution date and time), the URL indicating the Web page, and the URL were displayed. An advertisement ID indicating an advertisement and a user ID indicating a user who has browsed the Web page are included. One impression log may be registered for each advertisement displayed. That is, for example, when three advertisements are displayed on one Web page, three impression logs are registered.
In particular, the log storage module 110 stores log data generated each time a click occurs in one request. That is, the displayed advertisement (impression log), the clicked advertisement in the advertisement, and the advertisement not clicked are stored as log data. That is, the combination of the clicked advertisement and the advertisement not clicked is stored as log data. Thereby, the difference vector generation module 154 generates a difference vector.
Note that the log data used by the click prediction model generation module 150 includes log data of combinations of clicked advertisements and advertisements that were not clicked. The log data used by the feature vector generation module 120 does not necessarily include log data of a combination of clicked advertisements and advertisements that have not been clicked. However, the log data of the combination of the clicked advertisement and the advertisement not clicked is not excluded. That is, the log data used by the click prediction model generation module 150 and the log data used by the feature vector generation module 120 may be different log data or the same, or one of them includes the other. It may be log data.

The feature vector generation module 120 includes a feature vector generation processing module 122, a click rate calculation module 124, a click rate extraction module 126, and a feature vector element addition module 128. The feature vector generation module 120 is connected to the log storage module 110.
The feature vector generation processing module 122 generates a feature vector for selecting an advertisement to be displayed based on the log data regarding the click of the advertisement stored in the log storage module 110. The feature vector generated here is, for example, the feature vector table 400. FIG. 4 is an explanatory diagram showing an example of the data structure of the feature vector table 400. The feature vector table 400 has an advertisement column 410, a click rate column 420, a word column 430, a category column 440, a page similarity column 450, an action similarity column 460, etc. in the column direction, and an advertisement A column in the row direction. 482, an advertisement B column 484, an advertisement C column 486, an advertisement D column 488, and the like. That is, it represents a vector having elements such as a click rate, a word, a category, a page similarity, and a behavioral similarity for each advertisement. The advertisement column 410 stores an advertisement name (may be an advertisement ID). The click rate column 420 stores a CTR (Click Through Rate) for the advertisement. CTR is the rate at which an advertisement is clicked and is the number of clicks / impression (the number of advertisements displayed / distributed). The word column 430 stores words used for the advertisement or keywords assigned to the advertisement. Examples of words used in the advertisement include a title and a word string included in the detailed description. The category column 440 stores the classification name to which the advertisement belongs. The page similarity column 450 stores the similarity between the advertisement and the web page containing the advertisement. The behavior similarity column 460 stores the similarity between the advertisement and the user's web browsing behavior. For example, the behavioral similarity is an index indicating that if there are many automobile-related web pages as the user's web browsing behavior, the advertisement is similar if the advertisement is related to the automobile. It shows that it is not more similar than a car-related advertisement.

  The click rate calculation module 124 uses the log data stored in the log storage module 110, and for each attribute related to the user who clicked on the advertisement, clicks in combination with the elements constituting the feature vector of the advertisement Calculate the rate. The attributes related to the user here include, for example, gender as demographic information, age (may be age), nationality, address, etc., day of week, time zone, place when clicked as context information Etc. For example, the user attribute table 500 is used as an attribute as demographic information about the user. FIG. 5 is an explanatory diagram showing an example of the data structure of the user attribute table 500. The user attribute table 500 includes a user ID column 510, a gender column 520, an age column 530, a nationality column 540, an address column 550, and an interest category column 560. The user ID column 510 stores a user ID. The sex column 520 stores the gender of the user indicated by the user ID. The age column 530 stores the age of the user indicated by the user ID. The nationality column 540 stores the nationality of the user indicated by the user ID. The address column 550 stores the user's address indicated by the user ID. The interest category column 560 stores a category (there are movies, sports, etc. as hobbies, and baseball, soccer, etc. as sports) that the user indicated by the user ID is interested in. Further, the attributes as context information may be extracted from log data including the date and time when the user clicks, the location, and the like.

  For example, the advertisement attribute table 600 is used to manage the words used in the advertisement or the keywords assigned to the advertisement. FIG. 6 is an explanatory diagram showing an example of the data structure of the advertisement attribute table 600. The advertisement attribute table 600 has an advertisement ID column 610, a category column 620, and a word column 630. The advertisement ID column 610 stores an advertisement ID. The category column 620 stores a category (there are movies, sports, etc., and baseball, soccer, etc. as sports) including the advertisement indicated by the advertisement ID. This may be a keyword assigned to the advertisement. The word column 630 stores words used in the advertisement. This word may be extracted by performing morphological analysis on the targeted advertisement.

The click rate calculation module 124 creates an attribute / word click rate table 700 as a processing result. FIG. 7 is an explanatory diagram showing an example of the data structure of the attribute / word click rate table 700. The attribute / word click rate table 700 has a word 1 column 710, a word 2 column 720, a word 3 column 730, etc. in the column direction, and an attribute 1 column 782, an attribute 2 column 784, and an attribute 3 column in the row direction. 786 and the like. The intersecting cells store the click rate. That is, the attribute / word click rate table 700 includes a click rate in a combination of a word (a word used in the advertisement or a keyword assigned to the advertisement) and an attribute (gender, age, etc. as a user attribute). Is remembered. Thereby, the click rate corresponding to the combination of the user attribute and the word included in the advertisement can be extracted. For example, when the clicked user is male and in their 30s, the probability of responding to an advertisement including the word “car” (click rate), the probability of responding to an advertisement including the word “cosmetics” (click rate), etc. are extracted. Will be able to.
The click rate calculation module 124 calculates the click rate for each attribute from the feature vector of the advertisement.

  Further, the element used by the click rate calculation module 124 is not particularly limited as long as it is an element constituting a feature vector. For example, a word used in an advertisement or a keyword attached to the advertisement (word field 430 of the feature vector table 400) may be used. Moreover, you may make it produce | generate the model regarding an advertisement and an attribute by learning using log data. For example, using a model learned from log data using the words used in the advertisement or the keywords attached to the advertisement, the click rate, which is a value indicating how easy the advertisement is clicked, is calculated. May be. For example, the click rate when a male sees the advertisement is calculated by learning only the log data of the male. As a learning method, for example, a neural network may be used.

  The click rate extraction module 126 extracts a click rate corresponding to a predetermined element constituting the feature vector, which is a click rate for the advertisement targeted by the feature vector. The “click rate for an advertisement targeted by a feature vector” is the click rate of an advertisement that is a learning target by the learning module 156. The click rate is calculated by the click rate calculation module 124. As the predetermined element, for example, the aforementioned word or the like may be used. The click rate is extracted from the attribute / word click rate table 700 using the attribute of the clicked user and the word included in the click target advertisement. For example, if the user is male and in his 30s and clicks on a car advertisement (which includes the word “car”), the attribute is male and the word is “car” from the attribute / word click rate table 700. ”Is extracted, and the click rate in the combination of the attribute“ 30s ”and the word“ automobile ”is extracted.

The feature vector element addition module 128 adds the click rate extracted by the click rate extraction module 126 as an element of the feature vector generated by the feature vector generation processing module 122. For example, the feature vector 800 is generated. FIG. 8 is an explanatory diagram showing an example of the data structure of the feature vector 800. The feature vector 800 has an advertisement column 810, a click rate column 820, a word column 830, a category column 840, a click rate A column 850, a click rate B column 860, etc. in the column direction, and an advertisement A column 882 in the row direction. , An advertisement B column 884, an advertisement C column 886, an advertisement D column 888, and the like. That is, a click rate A column 850 and a click rate B column 860 are added to the above-described feature vector table 400 illustrated in FIG. The click rate extracted by the click rate extraction module 126 is obtained by adding two or more click rates. As a result, the click rate reflecting demographic information or context information is included in the feature vector.
In the above example, if the advertisement A is related to a car as an element of a feature vector when a male, 30-year-old user clicks on a car advertisement, the click rate A column 850 indicates that the user is a male. The click rate when the car advertisement is clicked and the click rate B column 860 are added with the click rate when the user is in his 30s and clicks the car advertisement. In addition, as an element of the feature vector when a woman or a user in her 50s clicks on an insurance advertisement, if the advertisement B is insurance-related, the click rate A column 850 indicates that the user is a woman and the insurance The click rate when the advertisement is clicked and the click rate B column 860 are added with the click rate when the user is in his 50s and clicks on the insurance advertisement.
As a result, a feature vector including the click rate reflecting the user attribute as an element is generated. In addition, in the other feature vector when generating the difference vector, the click rate A column 850 and the click rate B column 860 in the advertisement not clicked are the male user in the word included in the advertisement not clicked. A click rate and a click rate of a user in their 30s are added respectively, and the situation that the element in the difference vector is always 0 is avoided.

The click prediction model generation module 150 includes a log collection module 152, a difference vector generation module 154, a learning module 156, and an output module 158. The click prediction model generation module 150 is connected to the log storage module 110.
The log collection module 152 collects log data regarding advertisement clicks from the log storage module 110. The log data collected by the log collection module 152 may be the same as the log data used by the feature vector generation module 120, or different log data may be collected. The log data collected here includes at least log data generated by one click in order to generate a difference vector, and includes a combination of a clicked advertisement and an advertisement not clicked.
When log data is collected by the log collection module 152, the difference vector generation module 154 generates a difference vector between feature vectors of clicked advertisements and advertisements not clicked. Specifically, a difference vector composed of differences between the elements of the two feature vectors is generated. Note that the clicked advertisement and the advertisement not clicked here are advertisements on the same web page, and in the log data of one click, there is one clicked advertisement, which is on the web page. The other advertisements are advertisements that are not clicked. The difference vector, for example, as correct data (positive data), subtracts each corresponding element of the clicked advertisement feature vector from each element of the clicked advertisement feature vector (for example, behavioral similarity). The value obtained by subtracting each element of the clicked advertisement feature vector from each element of the advertisement feature vector not clicked may be used as negative data. Learning is performed using the correct answer data and the negative data.
The learning module 156 learns a criterion for selecting an advertisement to be displayed (ranking learning) using the difference vector generated by the difference vector generation module 154. The learning here is so-called machine learning, which generates a reference model for selecting an advertisement to be displayed and is supervised learning. However, any learning using the above-described difference vector may be used. Specifically, it is learning using correct data and negative data as teacher data, and may be learned using, for example, a neural network.
The output module 158 selects advertisements to display according to the criteria learned by the learning module 156. The selection here includes ranking (ranking).

FIG. 2 is a flowchart illustrating a processing example according to the present exemplary embodiment (feature vector generation module 120).
In step S202, the feature vector generation processing module 122 generates a feature vector using the log data in the log storage module 110.
In step S204, the click rate calculation module 124 calculates the click rate in combination with words for each attribute using the log data in the log storage module 110.
In step S206, the click rate extraction module 126 extracts the click rate corresponding to the element (word) in the feature vector and the attribute of the clicked user.
In step S208, the feature vector element addition module 128 generates a feature vector including the click rate as an element.

FIG. 9 is a flowchart illustrating a processing example according to the present exemplary embodiment (click prediction model generation module 150).
In step S902, the log collection module 152 collects log data from the log storage module 110.
In step S904, the difference vector generation module 154 generates a feature vector difference vector for each click.
In step S906, the learning module 156 performs a ranking learning process.
In step S908, the output module 158 outputs a click rate prediction model. Thereafter, an advertisement to be displayed on the web page is selected using the prediction model.

  Note that the hardware configuration of the computer on which the program according to the present embodiment is executed is a general computer, specifically, a personal computer, a computer that can be a server, or the like, as illustrated in FIG. That is, as a specific example, the CPU 1001 is used as a processing unit (arithmetic unit), and the RAM 1002, the ROM 1003, and the HD 1004 are used as storage devices. For example, a hard disk may be used as the HD 1004. Programs such as the feature vector generation processing module 122, the click rate calculation module 124, the click rate extraction module 126, the feature vector element addition module 128, the log collection module 152, the difference vector generation module 154, the learning module 156, and the output module 158 are executed. A CPU 1001, a RAM 1002 for storing programs and data thereof, a ROM 1003 storing a program for starting up the computer, an HD 1004 as an auxiliary storage device, and a user operation for a keyboard, a mouse, a touch panel, etc. Communication for receiving data based on the receiving device 1006, an output device 1005 such as a CRT or a liquid crystal display, and a communication network such as a network interface card Line interface 1007, and, and a bus 1008 for exchanging data by connecting them. A plurality of these computers may be connected to each other via a network.

Among the above-described embodiments, the computer program is a computer program that reads the computer program, which is software, in the hardware configuration system, and the software and hardware resources cooperate with each other. Is realized.
Note that the hardware configuration illustrated in FIG. 10 illustrates one configuration example, and the present embodiment is not limited to the configuration illustrated in FIG. 10, and is a configuration capable of executing the modules described in the present embodiment. I just need it. For example, some modules may be configured by dedicated hardware (for example, ASIC), and some modules may be in an external system and connected via a communication line. A plurality of systems shown in FIG. 5 may be connected to each other via communication lines so as to cooperate with each other.

The program described above may be provided by being stored in a recording medium, or the program may be provided by communication means. In that case, for example, the above-described program may be regarded as an invention of a “computer-readable recording medium recording the program”.
The “computer-readable recording medium on which a program is recorded” refers to a computer-readable recording medium on which a program is recorded, which is used for program installation, execution, program distribution, and the like.
The recording medium is, for example, a digital versatile disc (DVD), which is a standard established by the DVD Forum, such as “DVD-R, DVD-RW, DVD-RAM,” and DVD + RW. Standard “DVD + R, DVD + RW, etc.”, compact disc (CD), read-only memory (CD-ROM), CD recordable (CD-R), CD rewritable (CD-RW), Blu-ray disc ( Blu-ray Disc (registered trademark), magneto-optical disk (MO), flexible disk (FD), magnetic tape, hard disk, read-only memory (ROM), electrically erasable and rewritable read-only memory (EEPROM (registered trademark)) )), Flash memory, Random access memory (RAM) SD (Secure Digital) memory card and the like.
The program or a part of the program may be recorded on the recording medium for storage or distribution. Also, by communication, for example, a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a wired network used for the Internet, an intranet, an extranet, etc., or wireless communication It may be transmitted using a transmission medium such as a network or a combination of these, or may be carried on a carrier wave.
Furthermore, the program may be a part of another program, or may be recorded on a recording medium together with a separate program. Moreover, it may be divided and recorded on a plurality of recording media. Further, it may be recorded in any manner as long as it can be restored, such as compression or encryption.

DESCRIPTION OF SYMBOLS 110 ... Log storage module 120 ... Feature vector generation module 122 ... Feature vector generation processing module 124 ... Click rate calculation module 126 ... Click rate extraction module 128 ... Feature vector element addition module 150 ... Click prediction model generation module 152 ... Log collection module 154 ... Difference vector generation module 156 ... Learning module 158 ... Output module

Claims (5)

Select the advertisement to display based on the log data that is generated for each click of the advertisement in the web page where multiple advertisements are displayed, and is related to the combination of the clicked advertisement and the advertisement not clicked. Feature vector generating means for generating a feature vector for
Calculating means for calculating a click rate in combination with an element constituting the feature vector of the advertisement for each attribute related to a person who clicked on the advertisement using log data;
Extraction means for extracting a click rate corresponding to a predetermined element constituting the feature vector, the click rate for the advertisement targeted by the feature vector;
An information processing apparatus comprising: an adding unit that adds the click rate extracted by the extracting unit as an element of the feature vector generated by the feature vector generating unit. First learning means for generating a model relating to an advertisement and an attribute by learning using the log data;
The information processing apparatus according to claim 1, wherein the calculation unit calculates a click rate using the model learned by the first learning unit. A means of collecting log data for ad clicks,
When log data is collected by the collecting means, a difference vector generating means for generating a difference vector between the clicked advertisement and the advertisement vector not clicked;
Second learning means for learning a criterion for selecting an advertisement to be displayed using the difference vector generated by the difference vector generating means;
The information processing apparatus according to claim 1, further comprising a selection unit that selects an advertisement to be displayed according to the criterion learned by the second learning unit. An information processing method performed by an information processing apparatus including a feature vector generation unit, a calculation unit, an extraction unit, and an addition unit,
The feature vector generation means is log data generated for each click of an advertisement in a web page on which a plurality of advertisements are displayed, and is based on log data relating to a combination of an advertisement clicked and an advertisement not clicked , Generate feature vectors for selecting ads to display,
The calculation means uses the log data to calculate a click rate in combination with an element constituting a feature vector of the advertisement for each attribute related to a person who clicked on the advertisement.
The extraction means is a click rate for the advertisement targeted by the feature vector, and extracts an attribute corresponding to a predetermined element constituting the feature vector,
The information processing method, wherein the adding means adds the click rate extracted by the extracting means as an element of the feature vector generated by the feature vector generating means. Computer
Select the advertisement to display based on the log data that is generated for each click of the advertisement in the web page where multiple advertisements are displayed, and is related to the combination of the clicked advertisement and the advertisement not clicked. Feature vector generating means for generating a feature vector for
Calculating means for calculating a click rate in combination with an element constituting the feature vector of the advertisement for each attribute related to a person who clicked on the advertisement using log data;
Extraction means for extracting an attribute corresponding to a predetermined element constituting the feature vector, the click rate for the advertisement targeted by the feature vector;
An information processing program for causing a click rate extracted by the extracting unit to function as an adding unit that adds the click rate as an element of the feature vector generated by the feature vector generating unit.

Images