JP2022051305A - Information processing apparatus, system, learning device, information processing method and program - Google Patents

Abstract

To provide an information processing apparatus, a system, a learning device, an information processing method and a program that can improve accuracy of vector conversion and can suppress consumption of a memory.SOLUTION: An information processing apparatus comprises: an acquisition unit that acquires mail data which includes data of a plurality of items associated with an electronic mail; and an attribute classification unit that acquires a classification result indicating whether the electronic mail associated with the mail data is the electronic mail having a predetermined attribute by inputting data of the acquired plurality of items into a classifier which is learned such that when the data of the plurality of items is input, the classification result indicating whether the electronic mail associated with the mail data is the electronic mail having the predetermined attribute is output. And, the classifier includes: a child classifier for each item that output a feature quantity for each item when each of data of the plurality of items is input; and an integrated classifier that outputs the classification result when the feature quantity is input for each of the items.SELECTED DRAWING: Figure 1

Description

The present invention relates to an information processing device, a system, a learning device, an information processing method, and a program.

Technology for identifying unsolicited e-mails such as advertisements and promotions is known so that users can comfortably read e-mails. For example, Patent Document 1 discloses a technique for classifying an e-mail into normal e-mail or junk e-mail based on the header information of the e-mail. The present technology acquires a feature vector from the header information of an e-mail and compares the acquired feature vector with a classification rule to identify unsolicited e-mail.

Japanese Unexamined Patent Publication No. 2011-90442

The feature vector in the technique described in Patent Document 1 merely expresses meta information related to header information such as whether or not a sender is registered in an address book as a discrete value, and does not represent the content of the header information itself. .. Further, when expressing the content of the header information by a feature vector, it is conceivable that the header information is converted into a distributed representation by using a corpus that accumulates text examples. In that case, even though the header information has a plurality of items, it is converted into a distributed representation using a single corpus common to all the items. As a result, even for items with low conversion accuracy and small amount of information (for example, mime_type), when the one-hot vector is converted to the feature vector, the dimension of the vector becomes higher than necessary and the memory becomes excessive. It is assumed that it will be consumed in the future. As described above, in the conventional technique, resource consumption such as memory and CPU load may be high, or the classification accuracy of the classification target may not be sufficient.

The present invention has been made in consideration of such circumstances, and provides an information processing device, a system, a learning device, an information processing method, and a program capable of improving classification accuracy while suppressing resource consumption. That is one of the purposes.

In the information processing apparatus according to one aspect of the present invention, when the acquisition unit that acquires the data of a plurality of items related to e-mail and the data of the acquired plurality of items are input, the data of the plurality of items is input. By inputting to a classifier trained to output a classification result indicating whether or not the e-mail related to the mail data is an e-mail having a predetermined attribute, the e-mail related to the mail data has the predetermined attribute. It is equipped with an attribute classification unit that obtains a classification result indicating whether or not it is an e-mail, and the classifier outputs a feature amount for each item when each of the data of the plurality of items is input. It includes a device and an integrated classifier that outputs the classification result when the feature amount for each item is input.

According to one aspect of the present invention, the classifier includes a child classifier for each item that outputs a feature amount for each item when each of the data of a plurality of items is input, so that the classification accuracy is suppressed while suppressing resource consumption. Can be enhanced.

It is a block diagram which shows the example of the structure of the system 1 which concerns on embodiment of this invention. It is a figure for demonstrating the classification process by the attribute classification unit 130. It is a figure for demonstrating the generation process by a classifier generation unit 150. It is a sequence diagram which shows an example of the flow of the process performed by the terminal apparatus 10, the mail server 50, and the information processing apparatus 100. It is a figure for demonstrating an example of the distribution of the e-mail 53A by the terminal apparatus 10 and the input of feedback information. It is a block diagram which shows the example of the structure of the mail server 50A which concerns on the 1st modification. It is a figure for demonstrating an example of the push notification by a promotion notification part 55. It is a block diagram which shows the example of the structure of the terminal apparatus 10A which concerns on the 2nd modification.

Hereinafter, the information processing device, the system, the learning device, the information processing method, and the program according to the embodiment of the present invention will be described with reference to the drawings.

<Embodiment>
[System configuration]
FIG. 1 is a block diagram showing an example of the configuration of the system 1 according to the embodiment of the present invention. The system 1 includes, for example, a mail server 50 and an information processing device 100. The mail server 50 communicates with the terminal device 10 via the network NW. The network NW includes the Internet, LAN (Local Area Network), WAN (Wide Area Network), cellular network and the like.

The terminal device 10 is, for example, a computer device having a communication function such as a personal computer, a business computer, a smartphone, or a tablet terminal. The terminal device 10 operates an application program or a browser for sending and receiving e-mails, downloads an e-mail 53A addressed to itself from the mail server 50, or requests the mail server 50 to send the e-mail 53A to another mail server. Or something.

Further, the terminal device 10 receives the classification result indicating whether or not the e-mail 53A is a promotion mail from the mail server 50 together with the e-mail 53A, and receives the e-mail 53A according to the content of the classification result. Allocate to the main mail folder 30 or the promotion mail folder 40. Specifically, the terminal device 10 distributes the e-mail 53A to the main mail folder 30 when the classification result indicates the main mail, while the terminal device 10 distributes the e-mail to the main mail folder 30 when the classification result indicates the promotion mail. Allocate 53A to the promotion mail folder 40. Hereinafter, the classification result is also referred to as flag information.

The mail server 50 is, for example, a device that sorts and forwards e-mails 53A transmitted and received by the terminal device 10. The mail server 50 includes, for example, a communication unit 51, a transmission / reception control unit 52, and a storage unit 53. The communication unit 51 includes, for example, a communication interface such as a NIC, and communicates with another mail server, a terminal device 10, and an information processing device 100 (not shown).

The transmission / reception control unit 52 uses the communication unit 51 to send the e-mail 53A received from another mail server to the terminal device 10 under its jurisdiction, or to the mail server corresponding to the destination requested by the terminal device 10. Send e-mail 53A. The transmission / reception control unit 52 stores the transmitted / received e-mail 53A in the storage unit 53. The storage unit 53 is realized by a storage device such as an HDD, a flash memory, or a RAM (Random Access Memory).

The transmission / reception control unit 52 further transmits mail data 180A including data of a plurality of items related to the mail 53A received from another mail server to the information processing apparatus 100, and requests the classification of the mail 53A related to the mail data 180A. do. The transmission / reception control unit 52 receives from the information processing apparatus 100 the flag information indicating whether or not the e-mail 53A is a promotion e-mail, which is returned in response to the request. The transmission / reception control unit 52 combines the e-mail 53A and the flag information and transmits the e-mail to the terminal device 10. The mail data 180A includes, for example, a "display name" indicating the name of the sender of the e-mail 53A, a "subject" indicating the title, and a "mime_type" indicating the file format ("a plurality of items"). This is an example of "data"). The mail data 180A may be the e-mail 53A itself.

For example, the information processing apparatus 100 receives the mail data 180A related to the e-mail 53A addressed to the user from the mail server 50 in response to the request from the mail server 50, and receives the e-mail 53A related to the mail data 180A. For example, it is classified into either a main mail or a promotion mail (an example of "predetermined attribute"), and flag information indicating the classification result is transmitted to the mail server 50.

The information processing apparatus 100 includes, for example, a communication unit 110, a mail data acquisition unit 120, an attribute classification unit 130, a flag information transmission unit 140, a classifier generation unit 150, a feedback information reception unit 160, and a re-learning instruction. A unit 170 and a storage unit 180 are provided.

Each of the transmission / reception control unit 52, the mail data acquisition unit 120, the attribute classification unit 130, the flag information transmission unit 140, the classifier generation unit 150, the feedback information reception unit 160, and the relearning instruction unit 170 are, for example, CPUs (Central Processing Units). It is realized by a hardware processor such as Unit) executing a program (software). Some or all of these components are hardware (circuit parts) such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), GPU (Graphics Processing Unit). It may be realized by (including circuits), or it may be realized by the cooperation of software and hardware. The program may be stored in advance in a storage device (a storage device including a non-transient storage medium) such as an HDD (Hard Disk Drive) or a flash memory, or a removable storage such as a DVD or a CD-ROM. It is stored in a medium (non-transient storage medium) and may be installed by mounting the storage medium in a drive device.

The communication unit 110 includes, for example, a communication interface such as a NIC. The communication unit 110 receives the mail data 180A from the mail server 50. Further, the communication unit 110 transmits the flag information obtained by the attribute classification unit 130, which will be described later, to the mail server 50. The network in which the mail server 50 and the information processing apparatus 100 communicate with each other may be a part of the network NW.

[Inference stage]
Hereinafter, the processing of the inference stage using the classifier 180B will be described. The mail data acquisition unit 120 acquires the mail data 180A received by the communication unit 110 and stores it in the storage unit 180. The storage unit 180 is realized by a storage device such as an HDD, a flash memory, or a RAM.

By inputting the mail data 180A into the classifier 180B generated by the classifier generation unit 150 described later, the attribute classification unit 130 classifies whether or not the e-mail 53A related to the mail data 180A is a promotion mail. The result, i.e., the flag information is obtained.

Here, the details of the classification process using the classifier 180B by the attribute classification unit 130 will be described. FIG. 2 is a diagram for explaining the classification process by the attribute classification unit 130.

The attribute classification unit 130 inputs the display name, the subject, and mime_type from the mail data 180A to the classifier 180B. Here, each of the display name, the subject, and mime_type is input with the specified number of words as the specified length. As a specific example, 13 words are input as the display name, 44 words are input as the subject, and 1 word is input as the specified length for mime_type. The specified number of words for each item may be appropriately set so as not to lower the accuracy of the vector conversion. For example, the specified number of words may be set so as to accommodate a predetermined ratio (for example, 95%) or more of the plurality of mail data 180A stored in the storage unit 180 as teacher data. The word string of the data exceeding the set number of words may be truncated. Arbitrary numerical values may be set as a predetermined number of words and a predetermined ratio.

The attribute classification unit 130 converts the input display name, subject, and mime_type into a one-hot vector. The one-hot vector of a certain target data is a vector having 1 as a component when the word is included in the target data and 0 as a component when the word is not included in each of the words registered in advance. Since the display name and subject are considered to be expressed in natural language, the attribute classification unit 130 performs morphological analysis on the display name and subject, divides them into words, and then excludes elements such as numbers. Generates a one-hot vector based on.

The attribute classification unit 130 obtains a display name vector by inputting the one-hot vector of the display name obtained by the one-hot vectorization into the dictionary for the display name. Similarly, the attribute classification unit 130 obtains the subject vector by inputting the one-hot vector of the subject into the dictionary for the subject. Similarly, the attribute classification unit 130 obtains the mime_type vector by inputting the one-hot vector of mime_type into the dictionary for mime_type. Each of the display name dictionary, the subject dictionary, and the mime_type dictionary is an example of a child classifier for each item. Each of the display name vector, the subject vector, and the mime_type vector is an example of the feature amount for each item.

Here, the dictionary means a rule or algorithm that defines a conversion from a one-hot vector to a feature vector having a lower dimension than the one-hot vector and having continuous values within a predetermined range, and of each item. It is generated by machine learning from a corpus that collects examples. Such feature vectors include, for example, a distributed representation vector. By preparing a corpus and dictionary for each item instead of using a corpus and dictionary common to all items, it is possible to reduce the dimensions during vector conversion, reduce resource consumption, and further improve classification accuracy. can.

Here, a case where a plurality of target words are generated as a result of the morphological analysis will be described. In this case, a plurality of one-hot vectors are generated, and a plurality of display name vectors or subject vectors (hereinafter referred to as feature quantity vectors) corresponding thereto are generated. However, the attribute classification unit 130 may generate, for example, a plurality of feature quantity vectors. Is used as the input data to the classifier 180B. Instead of this, the attribute classification unit 130 may generate one one-hot vector in which the element corresponding to each of the plurality of target words is 1.

The attribute classification unit 130 inputs to a neural network (an example of an "integrated classifier") that integrates the display name vector, the subject vector, and the mime_type vector obtained by the conversion, and outputs the electronic data related to the mail data 180A as an output. Obtains flag information indicating whether or not the mail 53A is a promotion mail. Thereby, the attribute classification unit 130 can specify whether or not the e-mail 53A is a promotion mail.

The flag information transmission unit 140 uses the communication unit 110 to transmit the flag information obtained by the attribute classification unit 130 to the mail server 50.

[Learning stage]
Hereinafter, the processing of the learning stage of the classifier 180B will be described. In the process of the learning stage, the storage unit 180 stores a part or all of the mail data 180A as a label with flag information indicating whether or not the e-mail 53A related to the mail data 180A is a promotion mail. The label is information equivalent to the flag information, and for example, the result of classifying the e-mail 53A by a person is given as a label.

FIG. 3 is a diagram for explaining the generation process by the classifier generation unit 150. As shown in FIG. 3, the classifier generation unit 150 uses the mail data 180A stored in advance by the storage unit 180 as training data, and the label attached to the training data as training data, for example, using an error back propagation method. Then, machine learning is executed to generate a classifier 180B. The classifier generator 150 adjusts the parameters of the training model so that the result of inputting the training data to the training model approaches the teacher data. Then, when the parameter adjustment of the scheduled number of times is completed, the learning model at that time becomes the trained model, that is, the classifier 180B. The loss function used in the backpropagation method is any function, such as sum-of-squares error or cross entropy. The classifier generation unit 150 stores the generated classifier 180B in the storage unit 180.

In the present embodiment, the information processing apparatus 100 includes a classifier generation unit 150, and the storage unit 180 stores the classifier 180B generated by the classifier generation unit 150. However, as an alternative, the information processing apparatus 100 does not have these functional units, and there is another apparatus (learning apparatus) including a classifier generation unit 150 and a classifier 180B generated by the classifier generation unit 150. You may.

The feedback information receiving unit 160 receives feedback information (indicating correctness), which is the result of the user determining whether or not the classification result by the attribute classification unit 130 is correct, and receives the e-mail 53A related to the classification result. Accept from. The details of the feedback information will be described later with reference to FIG.

The re-learning instruction unit 170 causes the classifier generation unit 150 to execute re-learning or reinforcement learning using the feedback information based on the feedback information received by the feedback information receiving unit 160. The re-learning may be performed by the same method as the method in which the classifier generation unit 150 generates the classifier 180B (for example, the error back-propagation method), and any algorithm may be used for reinforcement learning.

[Processing sequence]
Next, with reference to FIG. 4, a series of processes executed by the terminal device 10, the mail server 50, and the information processing device 100 will be described. FIG. 4 is a sequence diagram showing an example of the flow of processing performed by the terminal device 10, the mail server 50, and the information processing device 100. In this figure, it is assumed that the entire e-mail 53A is transmitted from the mail server 50 to the information processing apparatus 100.

First, the mail server 50 receives the e-mail 53A sent to the user of the terminal device 10 from another mail server (S1). Next, the mail server 50 transmits the received e-mail 53A to the information processing apparatus 100 (S2).

When the information processing apparatus 100 receives the e-mail 53A from the mail server 50, the information processing apparatus 100 extracts the mail data 180A from the e-mail 53A (S3). Next, the information processing apparatus 100 uses the attribute classification unit 130 to input the mail data 180A into the classifier 180B, so that the flag information indicating whether or not the e-mail 53A related to the mail data 180A is a promotion mail. (S4). Next, the information processing apparatus 100 transmits the acquired flag information to the mail server 50 (S5).

When the mail server 50 receives the flag information, the mail server 50 together with the e-mail 53A and the flag information is transmitted to the terminal device 10 (S6). The transmission of the e-mail 53A in this step may be performed in response to a download request, an HTTP request, or the like from the terminal device 10.

When the terminal device 10 receives the e-mail 53A and the flag information, the terminal device 10 distributes and displays the e-mail 53A to the main mail folder 30 or the promotion mail folder 40 according to the content of the flag information (S7). Next, the terminal device 10 receives an input of feedback information indicating correctness for distribution from the user, and transmits the received feedback information to the information processing device 100 (S8).

When the information processing apparatus 100 receives the feedback information, the information processing apparatus 100 instructs the classifier generation unit 150 to relearn based on the feedback information (S9). This ends the process.

Next, the distribution of the e-mail 53A according to S7 and the input of the feedback information according to S8 will be described. FIG. 5 is a diagram for explaining an example of distribution of e-mail 53A and input of feedback information by the terminal device 10.

As shown in FIG. 5, the mail inbox of the terminal device 10 includes a main mail folder 30 and a promotion mail folder 40. When the flag information received from the mail server 50 indicates the main mail, the terminal device 10 distributes the e-mail 53A to the main mail folder 30. On the other hand, when the flag information indicates a promotion mail, the e-mail 53A is distributed to the promotion mail folder 40. The user of the terminal device 10 can browse the e-mail 53A of each folder and input feedback information according to the folder. Specifically, the user can select "promotion report" for the e-mail 53A stored in the main mail folder 30. On the other hand, for the e-mail 53A stored in the promotion mail folder 40, "report not promotion" can be selected. The feedback information input by the user is transmitted to the information processing apparatus 100 via the mail server 50, and is used for re-learning by the classifier generator 150.

According to the embodiment described above, in the information processing apparatus 100, when each of the data of a plurality of items is input, the child classifier for each item that outputs the feature amount for each item and the feature amount for each item are input. Then, using a classifier including an integrated classifier that outputs a classification result, flag information indicating whether or not the e-mail 53A is a promotion mail is acquired, and the terminal device 10 obtains flag information based on the flag information. Sort e-mail 53A. At that time, the child classifier for each item outputs the feature amount for each item by using the dictionary for each item instead of the dictionary common to each item. As a result, the information processing apparatus 100 can improve the classification accuracy while suppressing the resource consumption.

<First modification>
Next, the mail server of the first modification will be described with reference to FIG. The purpose of this modification is to perform effective promotion by sending a promotion remind notification to the user while executing the classification of the main mail and the promotion mail.

FIG. 6 is a block diagram showing an example of the configuration of the mail server 50A according to the first modification. In addition to the configuration shown in FIG. 1, the mail server 50A includes a promotion data collection unit 54, a promotion notification unit 55, a genre classification unit 56, and an open rate calculation unit 57.

The promotion data collecting unit 54 analyzes the text of the promotion mail sent to the terminal device 10, specifically, the e-mail 53A with the flag information indicating that it is a promotion mail, and the promotion data such as the campaign deadline. Collect 53B. The promotion data collecting unit 54 stores the collected data as promotion data 53B in the storage unit 53.

The promotion notification unit 55 refers to the promotion data 53B stored in the storage unit 53 at regular intervals (for example, every day), identifies the promotion data 53B whose campaign deadline is near, and the promotion deadline for the specified promotion data 53B. A reminder notification indicating that is near is transmitted to the terminal device 10 (first plan). At this time, the promotion notification unit 55 may send a reminder notification at any time when it is considered that the user can easily pay attention to the campaign, such as one day before the campaign deadline. Further, even if the promotion notification unit 55 sends a reminder notification only for the promotion data 53B of a genre having a high opening rate calculated by the opening rate calculation unit 57, which will be described later, instead of all the promotion data 53B whose deadline is near. Good (second plan). The promotion notification unit 55 may, for example, send a push notification or an email notification as a reminder notification. FIG. 7 is a diagram for explaining an example of push notification by the promotion notification unit 55. As shown in FIG. 7, the promotion notification unit 55 sends a push notification to the terminal device 10 for the promotion mail whose campaign deadline is near.

The genre classification unit 56 analyzes the text of the promotion mail sent to the terminal device 10, specifically, the e-mail 53A to which the flag information indicating that the mail is a promotion mail is attached, and uses a known clustering method or the like. Use to classify genres. The genre classification unit 56 stores the genre classified for each promotion mail in the storage unit 53 as additional information of the promotion data 53B. The genre to be classified may be any genre such as food, fashion, cosmetics, travel, and books.

The open rate calculation unit 57 divides the total number of received e-mails 53A by the number of opened e-mails 53A for each genre classified by the genre classification unit 56 and for each user. Is calculated. The open rate calculation unit 57 stores the calculated open rate in the storage unit 53 as additional information of the promotion data 53B.

According to the first modification described above, in the first plan, by sending a reminder notification regarding the promotion data 53B whose campaign deadline is near, effective promotion is performed while classifying the main mail and the promotion mail. It can be carried out. Further, according to the first modification, in the second proposal, the genre classification unit 56 and the open rate calculation unit 57 are used to send a reminder notification only for the genre that each user is highly interested in, thereby personalizing the method. More effective promotion can be done.

<Second modification>
In the embodiment or the first modification described above, the classification of the main mail and the promotion mail and the remind notification of the promotion are executed by the information processing apparatus 100 and the mail server 50 on the server side. Alternatively, these functions may be implemented by the terminal device 10 as an application.

FIG. 8 is a block diagram showing an example of the configuration of the terminal device 10A according to the second modification. As shown in FIG. 8, the terminal device 10A includes a communication unit 12 including a communication interface such as a NIC or a wireless communication device, a display unit 14 such as a display, an input unit 16 such as a touch panel, a keyboard, and a mouse, and mail data acquisition. A unit 18, an attribute classification unit 20, a promotion data collection unit 22, and a promotion notification unit 24 are provided.

The basic functions of the mail data acquisition unit 18, the attribute classification unit 20, the promotion data collection unit 22, and the promotion notification unit 24 are the mail data acquisition unit 120, the attribute classification unit 130, the promotion data collection unit 54, and the promotion notification unit. It is the same as the function of 55. The mail data acquisition unit 120 receives the mail data 180A from the mail server 50, whereas the mail data acquisition unit 18 receives the e-mail 53A from the mail server 50 and the mail data 180A from the e-mail 53A. Is obtained by extracting. Further, the promotion notification by the promotion notification unit 55 is realized by the cooperative operation of the mail server 50 and the terminal device 10, while the promotion notification by the promotion notification unit 24 is realized by the operation of only the terminal device 10.

According to the second modification described above, it is not necessary to implement the notification function on the server side, and the notification can be easily realized on the terminal side.

Although the embodiments for carrying out the present invention have been described above using the embodiments, the present invention is not limited to these embodiments, and various modifications and substitutions are made without departing from the gist of the present invention. Can be added.

10, 10A terminal device 12 communication unit 14 display unit 16 input unit 18 mail data acquisition unit 20 attribute classification unit 22 promotion data collection unit 24 promotion notification unit 30 main mail folder 40 promotion mail folder 50, 50A mail server 51 communication unit 52 transmission / reception Control unit 53 Storage unit 53A E-mail 53B Promotion data 54 Promotion data collection unit 55 Promotion notification unit 56 Genre classification unit 57 Opening rate calculation unit 100 Information processing device 110 Communication unit 120 Mail data acquisition unit 130 Attribute classification unit 140 Flag information transmission unit 150 Classifier generator 160 Feedback information reception unit 170 Re-learning instruction unit 180 Storage unit 180A Mail data 180B Classifier

Claims (13)

An acquisition unit that acquires email data that includes data for multiple items related to email,
Classification learned to output the classification result indicating whether or not the e-mail related to the e-mail data is an e-mail having a predetermined attribute when the data of the plurality of items acquired is input. It is provided with an attribute classification unit that obtains a classification result indicating whether or not the e-mail related to the e-mail data is an e-mail having the predetermined attribute by inputting to the device.
The classifier outputs a child classifier for each item when each of the data of the plurality of items is input, and outputs the classification result when the feature amount for each item is input. Including integrated classifier,
Information processing equipment. The item-by-item child classifier learns to use the population corresponding to each of the data of the plurality of items as learning data and output the feature amount for each item when each of the data of the plurality of items is input. Was done,
The information processing apparatus according to claim 1. The data of the plurality of items includes a part or all of the display name, subject, and type of e-mail of the e-mail.
The information processing apparatus according to claim 1 or 2. The predetermined attribute indicates an e-mail sent for the purpose of promoting a product, service, or business.
The information processing apparatus according to any one of claims 1 to 3. Generation to generate the classifier by machine learning using the data of the plurality of items as training data and the flag information indicating whether or not the e-mail related to the mail data is the e-mail of the predetermined attribute as the teacher data. With more parts,
The information processing apparatus according to any one of claims 1 to 4. A reception unit that receives feedback information indicating the correctness of the classification result by the attribute classification unit from the user who received the e-mail, and a reception unit.
An instruction unit that causes the generation unit to execute re-learning or reinforcement learning using the feedback information.
The information processing apparatus according to claim 5, further comprising. The information processing apparatus according to any one of claims 1 to 6 and the information processing apparatus.
A mail server that manages the transmission and reception of the e-mail and provides the mail data to the information processing device.
A system equipped with. The predetermined attribute indicates an e-mail sent for the purpose of promoting a product, service, or business.
The mail server is
A collection unit that collects data regarding the deadline of the promotion of e-mails classified by the attribute classification unit as having the predetermined attribute.
A notification unit that sends a notification regarding the deadline to the information terminal that received the e-mail, and
7. The system according to claim 7. The predetermined attribute indicates an e-mail sent for the purpose of promoting a product, service, or business.
The mail server is
A genre classification unit that classifies e-mail genres classified as having the predetermined attributes by the attribute classification unit, and
A calculation unit that calculates the open rate obtained by dividing the total number of received e-mails by the number of opened e-mails for each genre and each user.
A notification unit that sends a predetermined notification to an information terminal that has received the e-mail based on the classification result by the genre classification unit and the calculation result by the calculation unit.
7. The system according to claim 7. An acquisition unit that acquires email data that includes data for multiple items related to email,
When the data of the plurality of items is input, the classifier that outputs the classification result indicating whether or not the e-mail related to the mail data is an e-mail having a predetermined attribute is used as the learning data of the data of the plurality of items. It is provided with a generation unit generated by machine learning using flag information indicating whether or not the e-mail related to the e-mail data is an e-mail having the predetermined attribute as teacher data.
The classifier outputs a feature amount for each item when the contents of each of the plurality of items of the data are input, and a child classifier for each item, and when the feature amount for each item is input, the classification result is output. Including an integrated classifier to output,
Learning device. Information processing equipment
Get email data that contains data for multiple items related to email,
Classification learned to output the classification result indicating whether or not the e-mail related to the e-mail data is an e-mail having a predetermined attribute when the data of the plurality of items acquired is input. By inputting to the device, a classification result indicating whether or not the e-mail related to the mail data is an e-mail having the predetermined attribute is obtained.
The classifier outputs a child classifier for each item when each of the data of the plurality of items is input, and outputs the classification result when the feature amount for each item is input. Including integrated classifier,
Information processing method. On the computer
Get email data that includes data from multiple items related to email,
Classification learned to output the classification result indicating whether or not the e-mail related to the e-mail data is an e-mail having a predetermined attribute when the data of the plurality of items acquired is input. By inputting to the device, a classification result indicating whether or not the e-mail related to the mail data is an e-mail having the predetermined attribute is obtained.
The classifier outputs a child classifier for each item when each of the data of the plurality of items is input, and outputs the classification result when the feature amount for each item is input. Including integrated classifier,
program. The predetermined attribute indicates an e-mail sent for the purpose of promoting a product, service, or business.
In addition to the computer
The classifier classifies the e-mail as having the predetermined attribute, and causes the user to acquire data including at least the deadline of the opened e-mail.
To send a notification to the user based on the acquired data.
The program according to claim 12.

Images