JP7129585B1 - SNS Post Classification Providing System, SNS Post Classification Providing Method and Program - Google Patents

Abstract

An object of the present invention is to make it possible to accurately grasp what kind of evaluation is made on an SNS. SOLUTION: In the SNS post classification provision for classifying SNS posts, a plurality of first posts including a specific first character string are collected from the SNS, and each of the plurality of collected first posts is classified into a predetermined determines whether each category belongs to a positive, neutral, or negative evaluation, learns by associating each of the collected first posts with the evaluation for each of the predetermined categories determined for each, and obtains a learning result a second post including the specific first character string newly posted from the SNS; and based on the generated learned model, the collected second post The two posts are classified according to which of the positive, neutral, and negative evaluations they belong to for each of the predetermined categories, and the classification results are provided. [Selection drawing] Fig. 1

Description

Article 30, Paragraph 2 of the Patent Act applies Published on the press release site of PR TIMES Co., Ltd. on July 7, 2021

The present invention relates to a technique effective for classifying posts on SNS (Social Networking Service).

In recent years, technology related to SNS has attracted attention.
For example, in Patent Document 1, in order to collect preference information of users who use Internet services such as SNS, a posted image is analyzed to identify an object or object type in the posted image, and comments from viewers are analyzed. Based on this, a technology is provided for estimating a viewer's preference tendency.
In addition, Patent Literature 2 provides a technology for classifying unknown SNS clients by generating a classification model for classifying SNS clients that output posted data by purpose based on posted data on the SNS. ing.

Japanese Patent Application Laid-Open No. 2021-092931 JP 2016-006583 A

Posting on SNS is popular, and in particular, positive or negative reactions on SNS can have a large impact on business sales profits.
In order to accurately grasp what kind of evaluation is done on SNS, for example, in the case of a restaurant, for each category such as customer service, taste, price, interior, exterior, advertising expression, etc., It is important to classify and grasp whether a positive post is made, a neutral post is made, or a negative post is made.
Therefore, there is a demand for a technique for classifying SNS posts into positive, neutral, and negative evaluations for each category and providing the evaluations.
However, with the techniques described in Patent Literatures 1 and 2, it was not possible to classify SNS posts into positive, neutral, and negative evaluations for each category and provide them.
Therefore, the present inventor focused on a mechanism for classifying SNS posts into positive, neutral, and negative evaluations for each category and providing the evaluations.

The present invention classifies posts on SNS into positive, neutral, and negative evaluations for each category and provides them, so that it is possible to accurately grasp what kind of evaluation is being done on SNS. It is an object of the present invention to provide an SNS post classification providing system, an SNS post classification providing method, and a program.

The present invention is an SNS post classification providing system for classifying SNS posts,
a first collection unit that collects a plurality of first posts containing a specific first character string from the SNS;
a decision unit that decides which evaluation of positive, neutral, and negative each of the plurality of collected first posts belongs to for each predetermined category;
an extraction unit that further extracts a third post containing a second character string different from the first character string from the collected first posts;
a learning unit that learns by associating each of the collected first posts with the evaluation determined for each of the predetermined categories;
a generation unit that generates a trained model based on the learning result;
a second collecting unit that collects, from the SNS, second posts containing the specific first character string that have been newly posted;
a classification unit that classifies the collected second posts into positive, neutral, or negative evaluations for each of the predetermined categories, based on the generated learned model;
a provider that provides classification results;
with
When the learning unit learns by associating each of the collected first posts with the evaluation for each of the predetermined categories determined for each, the evaluation for each of the predetermined categories determined for the extracted third post. is weighted higher than the weight of the evaluation for each of the predetermined categories determined for the first post,
To provide an SNS post classification providing system.

According to the present invention, an SNS post classification providing system for classifying SNS posts collects a plurality of first posts containing a specific first character string from the SNS, and each of the plurality of collected first posts determines which evaluation of positive, neutral, and negative each predetermined category belongs to, and from the collected first posts, a third post including a second character string different from the first character string is extracted , each of the collected first posts is associated with the evaluation for each of the predetermined categories determined for each and learned, a learned model is generated based on the learning result, and a new model is generated from the SNS collects second posts including the specific first character string posted to , and determines whether the collected second posts are positive, neutral, or negative for each of the predetermined categories based on the generated trained model classifying which evaluation belongs to, providing classification results , and learning by associating each of the collected first posts with the evaluation for each of the predetermined categories determined for each, the extracted third Learning is performed by making the weight of the evaluation for each of the predetermined categories determined to be posted higher than the weight of the evaluation for each of the predetermined categories determined for the first post.

Although the present invention is in the category of systems, the same actions and effects can be achieved with methods and programs.

ADVANTAGE OF THE INVENTION According to this invention, it becomes possible to grasp exactly what kind of evaluation is made by SNS.

1 is a diagram illustrating an overview of an SNS post classification providing system 1; FIG. 1 is a diagram showing a functional configuration of an SNS post classification providing system 1; FIG. 2 is a diagram showing a flowchart of learning processing executed by the SNS post classification providing system 1. FIG. FIG. 7 is a diagram showing a flowchart of period setting processing executed by the SNS post classification providing system 1; FIG. 5 is a diagram showing a flowchart of condition setting processing executed by the SNS post classification providing system 1; FIG. 4 is a diagram showing a flowchart of a classification result providing process executed by the SNS post classification providing system 1; FIG. 11 is a diagram schematically showing an example of a classification result summary 40; FIG. 11 is a diagram schematically showing an example of a classification result summary 50; FIG. 4 is a diagram showing a flowchart of first notification processing executed by the SNS post classification providing system 1; 4 is a diagram schematically showing an example of a first alert 60; FIG. FIG. 10 is a diagram showing a flowchart of a second notification process executed by the SNS post classification providing system 1; It is a figure which showed an example of the 2nd alert 70 typically. FIG. 4 is a diagram showing a flowchart of re-learning processing executed by the SNS post classification providing system 1. FIG.

EMBODIMENT OF THE INVENTION Hereinafter, with reference to an accompanying drawing, the form (henceforth, embodiment) for implementing this invention is demonstrated in detail. In subsequent figures, the same numbers or symbols are attached to the same elements throughout the description of the embodiments.

[Basic Concept/Basic Configuration]
FIG. 1 is a diagram for explaining an outline of an SNS post classification providing system 1. As shown in FIG. The SNS post classification providing system 1 is a system including at least a computer 10 for classifying SNS posts.
In this embodiment, the SNS post classification providing system 1 is a system in which a computer 10, an SNS computer 20 managed by an SNS operator, and a user terminal 30 managed by a user are connected so as to be capable of data communication.

In this embodiment, as a premise, one or more third parties other than the user classify posted content posted to one or more SNSs.

An overview of processing steps when the SNS post classification providing system 1 classifies posts posted on SNS and provides classification results will be described with reference to FIG.

The computer 10 collects a plurality of first posts containing a specific first character string from SNS (step S1).
The first character string is, for example, the name of a company, product, service, event, brand, etc. specified by the user.
The computer 10 accesses the SNS and collects posts including a specific first character string among posts posted to the SNS by third parties as first posts.

The computer 10 determines which of the positive, neutral, and negative evaluations each of the collected first posts belongs to for each predetermined category (step S2).
Categories are, for example, products, services, places, advertisements, access methods, uses, quality, etc., and are subject to evaluation. , interiors, exteriors, advertising expressions, and others.
In addition, positive means that the content of the post is positive, neutral means that the content of the post is neither positive nor negative, and negative means that the content of the post is positive. means that it has negative content.
The computer 10 assigns each first post collected from users and system administrators positive, neutral, and negative for each predetermined category, for example, customer service, taste, price, interior decoration, exterior decoration, and advertising expression. Receives the input of which evaluation belongs to. The computer 10 determines the received evaluation as the evaluation for each category in the first post. Also, the computer 10 determines an evaluation for each category according to the posted content of the first post.
The category may be automatically sorted according to the content of the first post, or may be sorted by the user or system administrator. Also, one first post may be sorted into one category, or may be sorted into a plurality of categories.

The computer 10 learns by associating each of the collected first posts with the evaluation for each predetermined category determined for each (step S3), and based on the learning result, generates a trained model (step S4).
The computer 10, for example, uses the determined evaluation for each category as teacher data, executes supervised learning in which the content of each of the collected first posts is associated with the determined evaluation for each category, and obtains a learning result. Generate a trained model based on

The computer 10 collects newly posted second posts containing a specific first character string from the SNS (step S5).
The computer 10 accesses the SNS and collects, as second posts, posts containing a specific first character string among new posts posted by a third party on the SNS.

Based on the generated learned model, the computer 10 classifies the collected second posts into positive, neutral, and negative evaluations for each predetermined category (step S6).
The computer 10 refers to the generated learned model, and classifies, for each predetermined category, whether the second post belongs to which evaluation of positive, neutral, and negative, based on the posted content of the collected second post. do.

The computer 10 provides classification results (step S7).
The computer 10 provides the user terminal 30 with the classification result of the second post.

According to such an SNS post classification providing system 1, it is possible to accurately grasp what kind of evaluation is made on the SNS.

[Function configuration]
Based on FIG. 2, the functional configuration of the SNS post classification providing system 1 will be described.
The SNS post classification providing system 1 includes at least a computer 10. The computer 10 communicates data with an SNS computer 20 managed by an SNS operator, a user terminal 30 managed by a user, and a network 9 such as a public line network. Connected as possible.
In addition to the computer 10, the SNS post classification providing system 1 may include an SNS computer 20, a user terminal 30, other terminals and devices. In this case, the SNS post classification providing system 1 executes each process, which will be described later, by any one or a combination of included computers, terminals, devices, and the like.

The computer 10 is a computer, a personal computer, or the like having a server function for predicting the occurrence of complaints.
The computer 10 may be realized by, for example, one computer, or may be realized by a plurality of computers like a cloud computer. A cloud computer in this specification is a computer that uses any computer in a scalable manner to perform a specific function, or includes multiple functional modules to realize a certain system, and uses the functions in a free combination. It can be anything.

The computer 10 includes a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory), etc. as a control unit, and a communication unit to communicate with other terminals, devices, etc. It includes a device for enabling communication, a first collection unit 11 for collecting first posts, a second collection unit 12 for collecting second posts, a providing unit 13 for providing classification results, and the like.
The computer 10 also includes a data storage unit such as a hard disk, a semiconductor memory, a storage medium, or a memory card as a storage unit.
Further, the computer 10 includes, as a processing unit, various devices that execute various processes, a determination unit 14 that determines the evaluation of the first post for each category, and a learning unit that learns by associating the first post with the evaluation for each category. 15, a generation unit 16 that generates a trained model based on the learning result, a classification unit 17 that classifies the evaluation of the second post for each category, and the like.

In the computer 10, the control unit reads a predetermined program to cooperate with the communication unit to perform a first collection module, a period reception module, a condition reception module, a second collection module, a provision module, a first notification module, a first 2 Realize a notification module and a correction reception module.
Further, in the computer 10, the control unit reads a predetermined program to realize a learned model storage module, a period storage module, and a condition storage module in cooperation with the storage unit.
Further, in the computer 10, the control unit reads a predetermined program, and cooperates with the processing unit to perform a determination module, an extraction module, a weighting module, a learning module, a generation module, a period setting module, a condition setting module, and a classification module. , a summary module, a first condition determination module, a first alert creation module, a second condition determination module, and a second alert creation module.

The SNS computer 20 is server software or a server computer having a server function managed by an SNS operator. The SNS computer 20 may execute processing necessary for providing general SNS services, and detailed description thereof will be omitted.

The user terminal 30 is a terminal device such as a mobile terminal such as a mobile phone, a smartphone, a tablet terminal, or a personal computer managed by the user, and includes a CPU, GPU, RAM, ROM, etc. as a terminal control unit, It includes devices and the like for enabling data communication with the computer 10 and the SNS computer 20, and various devices and the like for executing data input/output.

Each process executed by the SNS post classification providing system 1 will be described below together with the process executed by each module described above.

[Learning processing executed by computer 10]
Learning processing executed by the computer 10 will be described with reference to FIG. This figure is a diagram showing a flowchart of the learning process executed by the computer 10 . This learning process includes the above-described collecting process of the first post (step S1), the process of determining the evaluation of the first post for each category (step S2), and the learning process of learning by associating the first post with the evaluation for each category. (Step S3) and details of the process of generating a trained model based on the learning result (Step S4).

The first collection module collects a plurality of first posts containing a specific first character string from SNS (step S10).
The first character string is, for example, the name of a company, product, service, event, brand, etc. specified by the user.
The first collection module collects, as first posts, posts including a specific first character string among posted contents posted to one or more SNSs by one or more third parties different from the user.
The first collection module accesses one or a plurality of SNSs, performs character recognition, image analysis, etc. on the content posted on the SNS by a third party, and detects posts containing a specific first character string. . The first collection module collects this detected post as the first post.
Note that the number of first posts collected by the first collection module is not particularly limited, but may be any number necessary for use in the process described later. Further, the first posts collected by the first collection module may be only posts that satisfy predetermined conditions such as a predetermined period, a predetermined timing, a predetermined SNS, etc. among the posts containing the detected specific first character string. .
Also, the first collection module can be configured to exclude posts for advertising purposes from the collected first posts. In this configuration, the first collection module excludes a post if the post contains a specific character string indicating that the post is for the purpose of advertisement, publicity, promotion, PR, or the like.

The determination module determines whether each of the collected first posts belongs to a positive, neutral, or negative evaluation for each predetermined category (step S11).
Categories are, for example, products, services, places, advertisements, access methods, uses, quality, etc., and are subject to evaluation. , interiors, exteriors, advertising expressions, and others. This category may be set in advance, or may be set each time such as by input or generation at the time of determination of evaluation. Also, the category may be automatically sorted according to the content of the first post, or may be sorted by the user or system administrator. Also, the number of categories in one first post is not limited to one, and may be plural.
Also, positive means that the content of the first post is positive, neutral means that the content of the first post is neither positive nor negative, and negative means that the content of the first post is neither positive nor negative. , means that the content of the first post is negative.
If one first post is in one category, for example, if there is only customer service in the above example, the determination module determines which evaluation of positive, neutral, and negative this customer service belongs to, If the first post is in multiple categories, for example, if it is customer service, taste, price, interior, exterior, advertising expression, etc. in the above example, determine which evaluation of positive, neutral, and negative the customer service belongs to Then, it determines whether the charge belongs to positive, neutral, or negative evaluation, and similarly determines whether the charge belongs to positive, neutral, or negative evaluation, respectively.
A determination module determines a category-by-category rating for each of the collected first posts.
The determination module receives input from the system administrator as to whether each first post belongs to which evaluation of positive, neutral, or negative for each category, and converts the received evaluation to each of the first posts. , to determine the evaluation for each category. In this case, the determination module accepts selective input and direct input of categories and evaluations belonging to each category for the first post via the user terminal 30, physical device or virtual device connected to the computer 10. , determines the received evaluation to be the evaluation for each category of the first post.
Also, the determination module determines a category and to which evaluation of positive, neutral, and negative each category belongs according to the posted content of each of the first posts. In this case, the determination module performs character recognition, image analysis, or the like on the first post to recognize the content of the first post. The determination module refers to a registered database in which posted content, the category of this posted content, and the evaluation for each category are associated and registered, and determines the category and the evaluation for each category according to the posted content of each of the first posts. decide.
Note that the method by which the determination module determines the evaluation for each category for each of the first posts is not limited to the example described above.

In the above description, the determination module is described as determining which of the three grades of positive, neutral, and negative evaluation each first post belongs to for each predetermined category. However, this evaluation could be classified in more detail. Specifically, the positive may consist of a strong positive and a weak positive, and the negative may consist of a strong negative and a weak negative. Strong positives are more positive than weak positives, and strong negatives are more negative than weak negatives.
The determination module also has a configuration for determining which of the five levels of evaluation, strong positive, weak positive, neutral, weak negative, and strong negative, each first post belongs to for each predetermined category. It is possible.
It should be noted that it is possible to configure the evaluation to be classified in more detail than that described above. In addition, the evaluation is not limited to character strings such as positive, neutral, and negative, and may be intended to be positive, neither positive nor negative, or negative. Good luck.

The extraction module further extracts a third post including a specific second character string different from the first character string from the collected first posts (step S12).
The second character string is a character string different from the first character string described above, for example, a character string that limits the first character string in more detail.
The extraction module extracts, from the collected first posts, first posts containing a specific second character string as third posts.
The extraction module performs character recognition, image analysis, or the like on the first post to recognize the content of the first post. The extraction module extracts, as the third post, the first post including the second character string in the recognized post content.
The number of third posts extracted by the extraction module is not particularly limited. Further, the third posts extracted by the extraction module may be only first posts including the second character string that satisfy predetermined conditions such as a predetermined period, a predetermined timing, and a predetermined SNS.

The weighting module weights the evaluation for each predetermined category determined for the extracted third post (step S13).
The weighting module weights the evaluation for each predetermined category determined for the third post more heavily than the evaluation for each predetermined category determined for the first post. That is, the weighting module gives the third post more importance than the first post.

The learning module learns by associating each of the collected first posts with the evaluation determined for each predetermined category (step S14).
Learning methods include, for example, machine learning such as supervised learning, unsupervised learning, and reinforcement learning, and deep learning such as convolutional neural networks, recurrent neural networks, and long/short-term memory. In this embodiment, the learning module performs machine learning by supervised learning as a learning method.
The learning module uses the determined evaluation for each category of the first post as training data, and performs supervised learning in which the posted content of each of the collected first posts is associated with the determined evaluation for each category. Here, the learning module performs learning according to the weighting for the third post. The learning module executes in a normal state regarding supervised learning that associates each of the first posts excluding the third posts with the evaluation for each category determined for each, and performs each of the third posts and each Supervised learning associated with the determined evaluation for each category is weighted and executed.
Note that the learning method executed by the learning module is not limited to the example described above.

The generation module generates a learned model based on the learning result (step S15).
The generation module generates, for example, a regression model or a classification model based on learning results as a trained model. The method by which the generation module generates the learned model based on the learning result may be any existing method, and detailed description thereof will be omitted.

The learned model storage module stores the generated learned model (step S16).

The above is the learning process.
The computer 10 uses the learned model generated by the learning process described above to execute the process described later.
The computer 10 can also be configured to execute the processing described later without storing the trained model in the learning processing. In this case, the generated trained model can be used as it is in the processing described later.

[Period Setting Process Executed by Computer 10]
Based on FIG. 4, the period setting process executed by the computer 10 will be described. This figure is a diagram showing a flowchart of the period setting process executed by the computer 10 . This period setting process is a process related to the second post collection process (step S5) described above, and can be executed before or after the learning process described above.

The period reception module receives input of a period for collecting the second posts (step S20).
Periods are, for example, time intervals, date intervals, time and date intervals.
The period acceptance module outputs to the user terminal 30 a predetermined input screen for accepting the input of the period. The user terminal 30 receives this input screen and displays it on its own display unit. The period acceptance module accepts input of a period through a predetermined input screen displayed on the user terminal 30 .
The user terminal 30 accepts direct input or selective input of the period. This may accept inputs for one period or may accept inputs for a plurality of periods.
The user terminal 30 transmits the accepted period to the computer 10 .
By receiving this period, the period acceptance module accepts input of the period for collecting the second posts.

The period setting module sets a period for collecting second posts (step S21).
The period setting module sets the received period to the period for collecting the second posts.

The period storage module stores the set period for collecting the second posts (step S22).

The above is the period setting process.
The computer 10 uses the period set by the period setting process described above to execute the process described later.
Note that this period setting process does not necessarily have to be executed. When the SNS post classification providing system 1 does not execute this period setting process, the SNS post classification providing system 1 starts and ends the process described later at an arbitrary timing.
Also, the period for collecting the second posts set by this period setting process can be changed by the user via the user terminal 30 . In this case, by executing the same processing as each of the above-described processing, an input for changing the period for collecting the second post is accepted, and the received period is set as the period for collecting the second post, and stored. 2 The period for collecting posts should be updated to the period set this time.

[Condition Setting Process Executed by Computer 10]
The condition setting process executed by the computer 10 will be described with reference to FIG. This figure is a diagram showing a flowchart of the condition setting process executed by the computer 10 . This condition setting process can be executed before or after the learning process described above.

The condition reception module receives input of a predetermined condition for alert notification (step S30).
The predetermined conditions relate to, for example, a period and the number of posts, and for example, the number of posts collected within 24 hours is 50 or more. That is, this predetermined condition is for judging whether there is a rapid increase in the number of posts in a short period of time. In addition to the period and the number of posts, the prescribed conditions include other conditions such as whether or not the content of the collected posts is posted on a prescribed website, such as being compiled on a prescribed website. May be included.
The condition reception module outputs to the user terminal 30 a predetermined input screen for receiving input of predetermined conditions. The user terminal 30 receives this input screen and displays it on its own display unit. The condition acceptance module accepts an input of a prescribed condition through a prescribed input screen displayed on the user terminal 30 .
The user terminal 30 accepts direct input or selective input of predetermined conditions. This may accept input of one condition, or may accept input of a plurality of conditions.
The user terminal 30 transmits the received predetermined condition to the computer 10 .
By receiving this predetermined condition, the condition reception module receives the input of the predetermined condition for alert notification.

The condition setting module sets a predetermined condition for alert notification (step S31).
The condition setting module sets the received predetermined condition as a condition for notifying an alert, which will be described later.

The condition storage module stores a predetermined condition for notifying the set alert (step S32).

The above is the condition setting process.
The computer 10 uses the predetermined conditions set by the condition setting process described above to execute the processes described later.
Note that this condition setting process does not necessarily have to be executed. When the SNS post classification providing system 1 does not execute this condition setting process, the SNS post classification providing system 1 executes a process described later under a preset condition.
Also, the predetermined condition set by this condition setting process can be changed by the user via the user terminal 30 . In this case, by executing the same processing as each processing described above, an input for changing the predetermined condition is received, the received predetermined condition is set as a predetermined condition for notifying an alert, and the predetermined condition is stored. should be updated to the predetermined conditions set this time.

[Classification Result Providing Processing Executed by Computer 10]
Based on FIG. 6, the classification result providing process executed by the computer 10 will be described. This figure is a diagram showing a flow chart of the classification result providing process executed by the computer 10 . This classification result provision process is the details of the second post collection process (step S5), the second post evaluation classification process for each category (step S6), and the classification result provision process (step S7).
This classification result providing process is a process executed during the period set by the period setting process described above. Further, this classification result providing process is a process that is executed at an arbitrary timing when the period setting process described above is not executed.

The second collection module collects second posts containing a specific first character string newly posted from the SNS (step S40).
The first character string in the process of step S40 is the same as the first character string in the process of step S10 described above.
The second collection module collects, as a second post, posts including a specific first character string among post contents newly posted to one or more SNSs by one or more third parties different from the user. do.
The second collection module accesses one or more SNSs, performs character recognition, image analysis, etc. on the contents of posts newly posted on SNSs by third parties, and collects posts containing a specific first character string. To detect. The second collection module collects this detected post as a second post.
The number of second posts collected by the second collection module is not particularly limited. Further, the second posts collected by the second collection module may be only posts that satisfy predetermined conditions such as a predetermined period, predetermined timing, predetermined SNS, etc., among the posts containing the detected specific first character string. .
The second collection module can also be configured to exclude posts for advertising purposes from the collected second posts. In this configuration, the second collection module excludes a post if it contains a specific character string indicating that the post is for the purpose of advertising such as advertisement, publicity, promotion, PR, or the like.

Based on the learned model, the classification module classifies the collected second posts into positive, neutral, and negative evaluations for each predetermined category (step S41).
The categories and evaluations in the process of step S41 are the same as those in the process of step S11 described above.
The classification module uses the learned model generated by the learning process described above to classify the collected second posts into positive, neutral, and negative evaluations for each predetermined category. The classification module classifies the evaluation for each category by referring to the learned model and based on the collected post content of the second post.
The classification module compares the post content of the second post collected this time with the post content of the first post in the learned model, and identifies the post content of the second post that matches or is similar to the post content of the second post. . The classification module compares the posted content of the first post with the posted content of the second post to determine the degree of matching of the posted content. The classification module identifies the first post in the trained model that most matches the posted content of the second post collected this time. If the classification module cannot identify the first post in the trained model that matches the posted content of the second post collected this time, the classification module identifies the first post in the trained model that is most similar to the posted content of the second post. do.
The classification module classifies the category and per-category rating associated with the identified first post as the category and per-category rating of the second post.
The classification module also associates the collected second posts with the classified categories and evaluations for each category.
In addition, if the classification module cannot identify the first post in the trained model that matches and approximates the posted content of the second post collected this time, the category of this second post and the evaluation for each category cannot be classified. . At this time, the classification module preliminarily sets a category and an evaluation for each category with respect to the second post whose category and evaluation for each category cannot be classified, and classifies this second post into the set category and category. A configuration for classifying into each evaluation is also possible.
In addition, the computer 10 generates an alert associated with the category and the unclassifiable evaluation for each category for the second post for which the category and the evaluation for each category are unclassifiable. A configuration for notifying the administrator is also possible. At this time, the computer 10 performs a re-learning process, which will be described later, to accept the category of the unclassifiable second post and the correction of the evaluation for each category. Even if there is, it is possible to classify the category and the evaluation for each category.

The summary module summarizes the classification results in chronological order (step S42).
The summarizing module summarizes, in chronological order, evaluations for each predetermined category classified with respect to the plurality of second posts. Here, the summarizing module may summarize the classification results for each category, or may collectively summarize the classification results for all categories. The summary module creates a visual summary of the classification results in chronological order, such as a table or graph (see FIG. 7).

[Summary of classification results]
Based on FIG. 7, the classification result summary created by the summary module will be described. This figure is a diagram schematically showing an example of a classification result summary created by the summary module.
The summary module creates a classification result summary 40 in which a first summary column 41 summarizing the contents of the classification results and a second summary column 42 summarizing the second posts in chronological order are arranged.
The first summary column 41 summarizes the name of the user, the first character string, the collection period of the second post, the category of the second post, the classification result, and the number of collected second posts. The first summary column 41 includes the user name, company name, first character string, project name, second post collection period, search period, second post category, and others as classification results. , the level is shown as the number of second posts collected, and the number of posts is shown, respectively. The company name indicates the name of the user. The project shows the first string. The search period indicates the period set by the period setting process described above, the date and time when the collection of the second posts was started, and the date and time when the collection of the second posts was finished. The level shows the positive, neutral, or negative classification results for the collected second posts as a whole. The number of posts indicates the total number of second posts collected.
The second summary column 42 summarizes the classification results, information about the second post, the number of followers of the poster of the second post, the number of citations of the second post by a person other than the poster, and the contents of the second post. It is. As a classification result, the level is information about the second post, the post information is the number of followers of the person who posted the second post, and the follower is the number of citations of the second post by a person other than the person who posted the second post. The number, the content of the second post, and the posted text are shown, respectively. The second summary column 42 summarizes the second posts in chronological order, and arranges the collected second posts in ascending order of posting date and time. At the level, the result of classification of whether the second post is positive, neutral, or negative is indicated by icons and characters. The post information indicates the name of the SNS where the second post was posted, the date and time of posting, and the account name of the poster. Followers indicates the number of followers of the person who posted the second post. The number of reposts indicates the number of posts in which posters other than the poster of the second post quoted the second post. The posted text indicates a copy of the posted content of the second post.
The summary module also creates a classification result summary with similar content for other categories.
In addition, although the second summary column 42 is shown as a table in the figure, the second summary column 42 may be arranged as a graph or the like.

In the above description, the summarizing module summarizes the classification results in chronological order, but it is also possible for the summarizing module to summarize the classification results regardless of the chronological order.
This case will be explained.
The summary module summarizes evaluations for each category classified with respect to a plurality of second posts. Here, the summarizing module may summarize the classification results for each category, or may collectively summarize the classification results for each category. The summary module creates a summary of the classification results by visualizing them in a table, graph, or the like (see FIG. 8).

[Another example of summary of classification results]
Another example of classification result summary created by the summary module will be described with reference to FIG. This figure schematically shows another example of classification result summary created by the summary module.
The summary module creates a classification result summary 50 in which evaluations classified for each category are summarized as a table. In the figure, the categories are customer service, price, taste, advertising expression, and others.
The summarizing module sums the evaluations for each category classified into each of the second posts for each category and for each evaluation, and summarizes the summation results as a table.
In the classification result summary 50, "-" is the number of second posts classified as negative, "±" is the number of second posts classified as neutral, and "+" is the second post classified as positive. Each is shown as the number of posts of posts. The classification result summary 50 shows that, in the customer service category, the number of second posts classified as negative is 5, the number of second posts classified as neutral is 10, and the number of second posts classified as positive. is 35, and similarly for other categories, the number of second posts classified as negative, the number of second posts classified as neutral, and the second posts classified as positive The number of posts for each post is indicated.
In addition, although the classification result summary 50 is shown as a table in the figure, the classification result summary 50 may be created as a graph or the like.

Returning to FIG. 6, the continuation of the classification result providing process will be described.
The provision module provides the classification result (step S42).
The provision module provides the created classification result summary (classification result summary 40 and classification result summary 50) to the user. The providing module transmits this classification result summary to the user terminal 30 .
The user terminal 30 receives this classification result summary and displays it on its own display unit.
The provision module provides the classification result by displaying the classification result summary on the user terminal 30 .

The above is the classification result providing process.
According to such an SNS post classification providing system 1, it is possible to accurately grasp what kind of evaluation is made on the SNS.

[First notification process executed by computer 10]
The first notification process executed by the computer 10 will be described with reference to FIG. This figure is a diagram showing a flowchart of the first notification process executed by the computer 10 . This first notification process is a process performed during execution of the above-described classification result providing process.

The first condition determination module determines whether the number of second posts classified as negative satisfies a predetermined condition (step S50).
The predetermined condition is a condition set by the condition setting process described above.
When the first condition determination module determines that the number of posts of the second posts classified as negative does not satisfy the predetermined condition (step S50 NO), the first condition determination module determines that the second posts are currently classified as negative. The computer 10 determines that the number of second posts has not increased sharply, that is, that the content related to the first character string has not gone up in flames, and the computer 10 terminates the first notification process.
In the present embodiment, the term "flaming" refers to a state in which the number of second posts classified as negative has increased rapidly. In addition to this state, other conditions such as a state in which the second post is posted on a predetermined website may be included.

On the other hand, if the first condition determination module determines that the number of second posts classified as negative satisfies the predetermined condition (step S50 YES), the first condition determination module currently classifies as negative. The number of posted second posts is rapidly increasing, that is, it is determined that the content related to the first character string is on fire, and the first alert creating module creates a first alert (step S51).
The first alert notifies that content related to the first character string is on fire.
A first alert creation module creates a first alert using a summary of classification results (see FIG. 10).

[1st alert]
Based on FIG. 10, the first alert created by the first alert creation module will be described. The same figure is a figure which shows typically an example of the 1st alert which a 1st alert preparation module produces.
The first alert creation module creates a first alert 60 in which a message 61 pointing out the occurrence of a fire and a classification result summary 62 summarizing the classification results for each category of the second post are arranged at predetermined positions.
The message 61 is a character string pointing out that a flame is occurring, and in the figure, "Flame has been detected." Note that the character string of this message 61 is not limited to the example described above, and its content and arrangement can be changed as appropriate as long as it points out the occurrence of a fire.
The classification result summary 62 is similar to the classification result summary 50 described above. Note that the classification result summary 62 is not limited to the above-described example, and the content and arrangement thereof can be changed as appropriate as long as the classification results can be used to grasp the occurrence of fire.
It should be noted that the first alert can also be configured to add a content that has been changed to make it more conspicuous in order to make it clearer that a fire is occurring. For example, bolding a message, adding highlights, adding icons, resizing and recoloring text and icons.
Further, the content included in the first alert is not limited to the above example, and may include other content, or may include only either the message 61 or the classification result summary 62 .

The first notification module notifies the created first alert (step S52).
The first notification module transmits the created first alert 60 to the user terminal 30 .
The user terminal 30 receives this first alert 60 and displays it on its own display.
The first notification module notifies the created first alert by displaying the first alert 60 on the user terminal 30 .

The above is the first notification process.
According to such an SNS post classification providing system 1, it becomes easy for the user to understand that a problem has occurred.

[Second Notification Processing Executed by Computer 10]
The second notification process executed by the computer 10 will be described with reference to FIG. 11 . This figure is a diagram showing a flowchart of the second notification process executed by the computer 10 . This second notification process is a process performed during execution of the above-described classification result providing process.

The second condition determination module determines whether the number of second posts classified as positive satisfies a predetermined condition (step S60).
The predetermined condition is a condition set by the condition setting process described above.
If the second condition determination module determines that the number of second posts classified as positive does not satisfy the predetermined condition (step S60 NO), the second condition determination module determines that the number of posts classified as positive at this time The computer 10 determines that the number of second posts has not increased rapidly, that is, that there has been no buzz regarding the content related to the first character string, and the computer 10 terminates the second notification process.
Buzzing in the present embodiment means a state in which the number of positively classified second posts increases sharply. In addition to this state, other conditions such as a state in which the second post is posted on a predetermined website may be included.

On the other hand, if the second condition determination module determines that the number of second posts classified as positive satisfies the predetermined condition (step S60 YES), the second condition determination module currently classifies the second posts as positive. It is determined that the number of posted second posts is rapidly increasing, that is, there is buzz about the content of the first character string, and the second alert creating module creates a second alert (step S61).
The second alert notifies that the content related to the first character string is buzzing.
A second alert creation module creates a second alert using the summarized classification results (see FIG. 12).

[Second alert]
The second alert created by the second alert creation module will be described with reference to FIG. The same figure is a figure which shows typically an example of the 2nd alert which a 2nd alert preparation module produces.
The second alert creation module creates a second alert 70 in which a message 71 pointing out the occurrence of buzz and a classification result summary 72 summarizing the classification results for each category of the second post are arranged at predetermined positions.
A message 71 is a character string pointing out that a buzz is occurring, and in the figure, "Buzz is detected." is shown. Note that the character string of this message 71 is not limited to the example described above, and its content and arrangement can be changed as appropriate as long as it points out the occurrence of buzz.
The classification result summary 72 is similar to the classification result summary 50 described above. Note that the classification result summary 72 is not limited to the above-described example, and the contents and arrangement thereof can be changed as appropriate as long as it is possible to grasp the occurrence of buzz regarding the classification results.
It should be noted that the second alert can also be configured to add content with a changed attention level to make it more obvious that buzz is occurring. For example, bolding a message, adding highlights, adding icons, resizing and recoloring text and icons.
Further, the content included in the second alert is not limited to the above example, and may include other content, or may include only either the message 71 or the classification result summary 72 .

The second notification module notifies the created second alert (step S62).
The second notification module transmits the created second alert 70 to the user terminal 30 .
The user terminal 30 receives this second alert 70 and displays it on its own display.
The second notification module notifies the created second alert by displaying the second alert 70 on the user terminal 30 .

The above is the second notification process.
According to such an SNS post classification providing system 1, it becomes easy for the user to grasp the popularity.

[Re-learning process executed by computer 10]
The relearning process executed by the computer 10 will be described with reference to FIG. 13 . This figure is a diagram showing a flowchart of the relearning process executed by the computer 10 . This re-learning process is a process that is performed after the above-described learning process, and is a process that is performed at an arbitrary timing.

The correction acceptance module accepts corrections to the evaluation for each predetermined category in the classified second posts (step S70).
The user terminal 30 allows the user terminal 30 to input the second post in the provided classification result and the category of the second post that was determined to be unclassifiable in the above-described classification result providing process and the correction of the evaluation for each category on a predetermined input screen. accept.
The user terminal 30 transmits the second post, the corrected category and the evaluation for each category to the computer 10 .
The correction receiving module receives the second post, the corrected category and the evaluation for each category, and receives correction of the evaluation for each predetermined category in the classified second post.

The learning module associates the collected second post with the corrected evaluation for each category and re-learns (step S71).
The learning module uses the corrected evaluation for each category as teacher data, and performs supervised learning again in association with the posted content of the second post in which the evaluation for each category is corrected and the corrected evaluation for each category.
Note that the relearning method executed by the learning module is not limited to the example described above.

The computer 10 may be configured to re-execute the processes of steps S12 and S13 described above when the corrected second post includes a specific second character string. In this case, the learning module performs learning according to the weighting for the second post containing the second character string.

The generation module updates the learned model based on the learning result (step S72).
The generation module updates the learned model based on the learning result by an existing method, similar to the process of step S15 described above.

The learned model storage module stores the updated learned model (step S73).

The above is the relearning process.
When the computer 10 newly collects the second post in the classification result providing process described above, the computer 10 uses the learned model updated by the relearning process to perform the process of step S42 described above.

In addition, the learning module performs supervised learning again by associating the second post and the evaluation for each category with the evaluation for each category as teacher data for the second post for which the correction of the evaluation for each category is not accepted. A configuration is also possible. Here, the computer 10 may be configured to re-execute the processes of steps S12 and S13 described above when the second post for which correction has not been accepted is a post containing a specific second character string. In this case, the learning module performs learning according to the weighting for the second post containing the second character string.
By doing so, the SNS post classification providing system 1 can further improve the accuracy of classification of the evaluation of the second post for each category.

Each of the processes described above is described as separate processes, but the computer 10 can also be configured to execute a combination of some or all of the processes described above. Further, the computer 10 can be configured to execute the processing even at timings other than the timings described in each processing.

The means and functions described above are realized by a computer (including a CPU, an information processing device, and various terminals) reading and executing a predetermined program. The program may be provided, for example, from a computer via a network (SaaS: software as a service) or provided as a cloud service. Also, the program may be provided in a form recorded on a computer-readable recording medium. In this case, the computer reads the program from the recording medium, transfers it to an internal recording device or an external recording device, records it, and executes it. Alternatively, the program may be recorded in advance in a recording device (recording medium) and provided from the recording device to the computer via a communication line.

Although the embodiments of the present invention have been described above, the present invention is not limited to these embodiments described above. Moreover, the effects described in the embodiments of the present invention are merely enumerations of the most suitable 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.

(1) An SNS post classification providing system for classifying SNS posts,
a first collection unit (e.g., first collection unit 11, first collection module) that collects a plurality of first posts containing a specific first character string from the SNS;
a determination unit (e.g., determination unit 14, determination module) that determines whether each of the plurality of collected first posts belongs to a positive, neutral, or negative evaluation for each predetermined category;
a learning unit (for example, a learning unit 15, a learning module) that learns by associating each of the collected first posts with the evaluation for each of the predetermined categories determined for each;
a generation unit (eg, generation unit 16, generation module) that generates a trained model based on the learning result;
a second collection unit (for example, a second collection unit 12, a second collection module) that collects a second post containing the specific first character string newly posted from the SNS;
A classification unit (e.g., a classification unit 16, a classification module) that classifies whether the collected second posts belong to positive, neutral, or negative evaluations for each of the predetermined categories, based on the generated learned model. )When,
a providing unit (for example, providing unit 13, providing module) that provides classification results;
SNS post classification providing system.

According to the invention of (1), it is possible to accurately grasp what kind of evaluation is made on the SNS.

(2) the positive comprises a strong positive and a weak positive;
The negative consists of a strong negative and a weak negative,
The SNS post classification providing system according to (1).

According to the invention of (2), it is possible to classify even the degree of positivity and the degree of negativity.

(3) an extraction unit (e.g., an extraction module) that further extracts a third post containing a second character string different from the first character string from the collected first posts;
further comprising
When the learning unit learns by associating each of the collected first posts with the evaluation for each of the predetermined categories determined for each, the evaluation for each of the predetermined categories determined for the extracted third post. is weighted higher than the weight of the evaluation for each of the predetermined categories determined for the first post,
The SNS post classification providing system according to (1).

According to the invention of (3), it is possible to improve the classification accuracy by weighting the weight of the evaluation of the post containing the specific character string for learning.

(4) a summarizing unit (for example, a summarizing module) that summarizes the classification results in chronological order;
further comprising
The providing unit provides the summarized result as the classification result,
The SNS post classification providing system according to (1).

According to the invention of (4), it becomes easy to judge whether it is heading in a good direction or a bad direction.

(5) when the second post classified as negative satisfies a predetermined condition, a first notification unit (for example, a first notification module) that notifies to that effect;
The SNS post classification providing system according to (1), further comprising:

According to the invention of (5), it becomes easy for the user to understand that a problem has occurred.

(6) when the second post classified as positive satisfies a predetermined condition, a second notification unit (for example, a second notification module) that notifies to that effect;
The SNS post classification providing system according to (1), further comprising:

According to the invention of (6), it becomes easy for the user to grasp the popularity.

(7) a period acceptance unit (for example, a period acceptance module) that accepts setting of a period for collecting the second posts;
further comprising
The second collection unit collects the second post only during the period of acceptance.
The SNS post classification providing system according to (1).

According to the invention of (7), only the evaluations are collected during the period desired by the user, and it becomes possible to more accurately grasp what kind of evaluations are made on the SNS.

(8) the classification unit classifies one of the second posts into each of the plurality of categories according to which evaluation of positive, neutral, and negative;
The SNS post classification providing system according to (1).

According to the invention of (8), it is possible to accurately grasp what kind of evaluation is made on the SNS.

(9) A computer-executed SNS post classification providing method for classifying SNS posts,
a step of collecting a plurality of first posts containing a specific first character string from the SNS (for example, step S10);
a step of determining which evaluation of positive, neutral, and negative each of the plurality of collected first posts belongs to for each predetermined category (for example, step S11);
a step of learning by associating each of the collected first posts with the evaluation for each of the predetermined categories determined for each (for example, step S14);
a step of generating a trained model based on the learning result (for example, step S15);
a step of collecting newly posted second posts containing the specific first character string from the SNS (for example, step S40);
a step of classifying whether the collected second posts belong to positive, neutral, or negative evaluations for each of the predetermined categories based on the generated learned model (for example, step S41);
providing a classification result (e.g., step S43);
SNS post classification providing method.

(10) A computer that classifies SNS posts,
a step of collecting a plurality of first posts containing a specific first character string from the SNS (for example, step S10);
a step of determining which evaluation of positive, neutral, and negative each of the plurality of collected first posts belongs to for each predetermined category (for example, step S11);
a step of learning by associating each of the collected first posts with the evaluation for each of the predetermined categories determined for each (for example, step S14);
a step of generating a trained model based on the learning result (for example, step S15);
a step of collecting newly posted second posts containing the specific first character string from the SNS (for example, step S40);
a step of classifying whether the collected second posts belong to positive, neutral, or negative evaluation for each of the predetermined categories based on the generated learned model (for example, step S41);
providing a classification result (e.g. step S43);
A computer readable program for executing

1 SNS post classification providing system 9 network 10 computer 20 SNS computer 30 user terminal 40 classification result summary 41 first summary column 42 second summary column 50 classification result summary 60 first alert 61 message 62 classification result summary 70 second alert 71 message 72 Summary of classification results

Claims (9)

An SNS post classification providing system for classifying SNS posts,
a first collection unit that collects a plurality of first posts containing a specific first character string from the SNS;
a decision unit that decides which evaluation of positive, neutral, and negative each of the plurality of collected first posts belongs to for each predetermined category;
an extraction unit that further extracts a third post containing a second character string different from the first character string from the collected first posts;
a learning unit that learns by associating each of the collected first posts with the evaluation determined for each of the predetermined categories;
a generation unit that generates a trained model based on the learning result;
a second collecting unit that collects, from the SNS, second posts containing the specific first character string that have been newly posted;
a classification unit that classifies the collected second posts into positive, neutral, or negative evaluations for each of the predetermined categories, based on the generated learned model;
a provider that provides classification results;
with
When the learning unit learns by associating each of the collected first posts with the evaluation for each of the predetermined categories determined for each, the evaluation for each of the predetermined categories determined for the extracted third post. is weighted higher than the weight of the evaluation for each of the predetermined categories determined for the first post,
SNS posting classification providing system. The positive consists of a strong positive and a weak positive,
The negative consists of a strong negative and a weak negative,
The SNS post classification providing system according to claim 1. a summarizing unit for summarizing the classification results in chronological order;
further comprising
The providing unit provides the summarized result as the classification result,
The SNS post classification providing system according to claim 1. a first notification unit that notifies when the second post classified as negative satisfies a predetermined condition;
The SNS post classification providing system according to claim 1, further comprising: a second notification unit that notifies when the second post classified as positive satisfies a predetermined condition;
The SNS post classification providing system according to claim 1, further comprising: a period reception unit that receives setting of a period for collecting the second post;
further comprising
The second collection unit collects the second post only during the period of acceptance.
The SNS post classification providing system according to claim 1. The classification unit classifies the one second post into each of the plurality of categories according to which evaluation of positive, neutral, and negative it belongs,
The SNS post classification providing system according to claim 1. A computer-executed SNS post classification providing method for classifying SNS posts, comprising:
collecting a plurality of first posts containing a specific first character string from the SNS;
a step of determining to which evaluation of positive, neutral, and negative each of the plurality of collected first posts belongs to for each predetermined category;
a step of extracting a third post including a second character string different from the first character string from the collected first posts;
a step of learning by associating each of the collected first posts with the evaluation determined for each of the predetermined categories;
generating a trained model based on the learning results;
Collecting newly posted second posts containing the specific first character string from the SNS;
a step of classifying whether the collected second posts belong to positive, neutral, or negative evaluations for each of the predetermined categories, based on the generated learned model;
providing classification results;
When learning by associating each of the collected first posts with the evaluation for each of the predetermined categories determined for each, the weight of the evaluation for each of the predetermined categories determined for the extracted third post is a step of learning by making the weight of the evaluation for each of the predetermined categories determined as the first post heavier than the weight of the evaluation;
SNS post classification providing method. A computer that classifies SNS posts,
collecting a plurality of first posts containing a specific first character string from the SNS;
determining whether each of the plurality of collected first posts belongs to a positive, neutral, or negative evaluation for each predetermined category;
further extracting a third post containing a second character string different from the first character string from the collected first posts;
learning by associating each of the collected first posts with the evaluation for each of the predetermined categories determined for each;
generating a trained model based on the learning results;
a step of collecting newly posted second posts containing the specific first character string from the SNS;
classifying whether the collected second posts belong to positive, neutral, or negative evaluations for each of the predetermined categories, based on the generated learned model;
providing classification results;
When learning by associating each of the collected first posts with the evaluation for each of the predetermined categories determined for each, the weight of the evaluation for each of the predetermined categories determined for the extracted third post is a step of learning with a heavier weight than the weight of the evaluation for each of the predetermined categories determined as the first post;
A computer readable program for executing

Images