Flaming potential determination device, flaming potential determination method, and program

Abstract

This flammability determination device comprises: a first determination unit which determines whether or not there is a prescribed deviation of the frequency of posting to a first SNS by a first user, who has made a certain post to the first SNS, from the average of the frequencies of posting by all users of the first SNS; an estimation unit which, if there is the prescribed deviation, estimates an experience value indicating the first user's level of experience in using the Internet, on the basis of the likelihood that the first user will use a second SNS different from the first SNS; and a second determination unit which, if the experience value is less than a first threshold value, determines whether or not the content of said certain post contains a negative expression, and if the content of the certain post contains the negative expression, adds the content of the certain post to a set of post content that is likely to suffer flaming. Thus the flammability determination device makes it possible to collect post content that is highly likely to suffer flaming due to carelessness.

Description

[Technical field]

[0001] The present invention relates to an inflaming potential determination device, an inflaming potential determination method, and a program. [Background technology]

[0002] Conventionally, prediction of flaming has been performed on social networking services (SNS) and bulletin boards, etc. Conventionally, a system for preventing flaming has been constructed by collecting frequently occurring words and abusive expressions in flaming in advance and learning the frequently occurring words and abusive expressions in flaming using machine learning.

[0003] In addition, with a view to building a predictive model for inflammatory tweets, research is being conducted to extract the topics of comments and examine and predict how public opinion of those topics will change. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Flaming prediction based on harmful expressions using distributed representations, Interactive Information Access and Visualization Mining Study Group-SIG-AM (2017) [Non-Patent Document 2] Analysis of Flaming in CGM and Its Applications, Journal of the Japanese Society for Artificial Intelligence, Vol. 30, No. 1 [Non-Patent Document 3] Proposal of a list combination method for preventing and deterring pseudo-crimes on Twitter, Information Processing Society of Japan Research Report Groupware and Network Services (GN) 2012 Summary of the Invention [Problem to be solved by the invention]

[0005] Conventionally, past flaming tweets are collected from all over Twitter (registered trademark), and simple features such as surface information of words and co-occurrence relationships of adjacent words are learned to predict flaming, but the accuracy is not high. On Twitter (registered trademark), various expressions such as monologue and posts to friends are mixed, and verification results have shown that only about 70% of flaming tweets can be detected.

[0006] There are various types of flame wars, including those caused by the poster's own carelessness, those caused intentionally by self-staged posts, and stealth marketing. If one were to try to predict flame wars by covering all of these, it would be difficult to improve the accuracy of the prediction. Among these, flame wars caused by the poster's own carelessness can cause serious damage, including harassment from Internet users, information about the poster being spread online, and the poster's personal information being identified, placing a heavy mental burden that the poster never expected. In order to prevent such damage from spreading, it is necessary to suppress flame wars. To do this, it is necessary to collect the content of posts that are likely to cause flame wars due to carelessness.

[0007] The present invention has been made in consideration of the above points, and has an object to make it possible to collect posted content that is likely to cause a flame war due to carelessness. [Means for solving the problem]

[0008] In order to solve the above problem, the flame war potential determination device has a first determination unit that determines whether or not the posting frequency to a first SNS by a first user who has made a post to the first SNS has a predetermined deviation from the average posting frequency of all users who use the first SNS, an estimation unit that, if the predetermined deviation exists, estimates an experience value indicating the first user's level of experience in using the Internet based on the possibility of using a second SNS other than the first SNS, and a second determination unit that, if the experience value is less than a first threshold, determines whether or not the content of the certain post contains a negative expression, and if the negative expression is included, adds the content of the certain post to a set of post contents that have the potential to cause a flame war. Effect of the Invention

[0009] It is possible to collect posted content that is likely to cause an uproar due to carelessness. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is a diagram illustrating an example of a system configuration according to an embodiment of the present invention. [Diagram 2] 1 is a diagram illustrating an example of a hardware configuration of a flaming potential determination device 10 according to an embodiment of the present invention. [Diagram 3] 1 is a diagram illustrating an example of a functional configuration of a flaming potential determination device 10 according to an embodiment of the present invention. [Figure 4] 4 is a diagram illustrating an example of the configuration of a post related information storage unit 16. FIG. [Diagram 5] 11 is a flowchart illustrating an example of a processing procedure executed by a flame candidate determination unit 12. [Figure 6] FIG. 2 is a diagram illustrating an example of the configuration of a flame candidate storage unit 17. [Figure 7] 1 is a diagram illustrating an example of the configuration of a different SNS information storage unit 18. FIG. [Figure 8] 11 is a flowchart illustrating an example of a processing procedure executed by an experience value estimation unit 14. [Figure 9] FIG. 2 is a diagram illustrating an example of the configuration of a corpus storage unit 19. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Possible factors that can cause a careless post (such as a tweet) on a social networking service (SNS) to offend people and spark a flame war are lack of attention and insufficient care. In this embodiment, a user (poster) who is considered to be both inexperienced and lacking in attention is identified, and the contents of posts such as tweets that are likely to spark a flame war due to the user's own carelessness (hereinafter referred to as "flaming post contents") are stored in a corpus. Flaming potential is determined from the content of the post and the language used in the replies. A flame war refers to, for example, a large amount of criticism being directed at information that has been sent out.

[0012] In this embodiment, a user who posts relatively infrequently is identified as an inexperienced user, and a user who posts relatively frequently is identified as an attentive user. When posting carefully, the user carefully examines the content and takes time to post, but when the user's attention is low, the user tends to post continuously without examining the content, so the frequency of posting is noted as a lack of attention.

[0013] In this embodiment, for a user identified as being inexperienced or lacking in attention, an experience value regarding Internet use is calculated based on the amount of time that has elapsed since the creation of the account related to the post, the possibility of using other SNSs, etc.

[0014] In this embodiment, furthermore, based on the experience level of a user who has been identified as lacking in experience or attention, a determination is made as to the possibility that the content posted by that user will cause a controversy.

[0015] The posted content refers to any expression using natural language, symbols, etc. However, the posted content may also include images, sounds, etc.

[0016] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. Fig. 1 is a diagram showing an example of a system configuration in an embodiment of the present invention. In Fig. 1, a flame war potential determination device 10 is connected to an SNS server 20 and another SNS server 30, etc., via a network such as the Internet.

[0017] The flame-warping potential determination device 10 is one or more computers that extract (specify) flame-warping post contents from the contents posted to the SNS server 20, and generate a corpus that is a collection of the extracted (specified) flame-warping post contents.

[0018] The SNS server 20 is one or more computers that provide SNSs (hereinafter, referred to as “target SNSs”) that are targets for extracting flaming post contents by the flaming potential determination device 10.

[0019] The other SNS server 30 is one or more computers that provide SNSs (hereinafter, referred to as "other SNSs") other than the target SNS. Note that the other SNSs do not have to be one specific SNS, and may be a collection of multiple SNSs.

[0020] 2 is a diagram showing an example of a hardware configuration of the inflammability determination device 10 according to an embodiment of the present invention. The inflammability determination device 10 in FIG. 2 includes a drive device 100, an auxiliary storage device 102, a memory device 103, a processor 104, and an interface device 105, which are all connected to each other via a bus B.

[0021] A program for implementing the processing in the flame potential determination device 10 is provided by a recording medium 101 such as a CD-ROM. When the recording medium 101 storing the program is set in the drive device 100, the program is installed from the recording medium 101 via the drive device 100 into the auxiliary storage device 102. However, the program does not necessarily have to be installed from the recording medium 101, but may be downloaded from another computer via a network. The auxiliary storage device 102 stores the installed program as well as necessary files, data, and the like.

[0022] The memory device 103 reads out and stores the program from the auxiliary storage device 102 when an instruction to start the program is received. The processor 104 is a CPU or a GPU (Graphics Processing Unit), or a CPU and a GPU, and executes functions related to the inflaming potential determination device 10 according to the program stored in the memory device 103. The interface device 105 is used as an interface for connecting to a network.

[0023] 3 is a diagram showing an example of a functional configuration of the inflaming potential determination device 10 according to the embodiment of the present invention. As shown in FIG. 3, the inflaming potential determination device 10 includes a post related information collection unit 11, a inflaming candidate determination unit 12, an other SNS information acquisition unit 13, an experience value estimation unit 14, and an inflaming potential determination unit 15. Each of these units is realized by a process in which one or more programs installed in the inflaming potential determination device 10 are executed by a processor 104. The inflaming potential determination device 10 also uses a post related information storage unit 16, a inflaming candidate storage unit 17, an other SNS information storage unit 18, and a corpus storage unit 19. Each of these storage units can be realized by using, for example, an auxiliary storage unit 102, or a storage device that can be connected to the inflaming potential determination device 10 via a network.

[0024] The processing contents executed by each unit will be described below.

[0025] [Posting Related Information Collection Division 11] Each time a post content (hereinafter referred to as the “target post content”) is posted on the target SNS after a predetermined time (e.g., 12 hours) has passed, the post related information collection unit 11 receives information (hereinafter referred to as the “post related information”) related to the post (hereinafter referred to as the “target post”) as input, and outputs (stores) the input post related information to the post related information storage unit 16.

[0026] Post-related information includes values ​​for items such as "post content," "user name," "profile text," "post date and time," "reply," "number of replies," "number of likes," "number of reposts," "number of blocks," "account creation date and time," "number of likes given by the poster," "cumulative number of posts," "average number of posts per day by the poster," and "average number of posts per day by all users."

[0027] The "posted content" of the target post is the target posted content. The "user name" of the target post is the username of the user (hereinafter referred to as the "target user") who posted the target posted content. The "profile text" of the target post is a text (character string) indicating the profile of the target user. The "posted date and time" of the target post is the date and time when the target posted content was posted to the target SNS. The "reply" of the target post is one or more posts as a reaction (response) to the target posted content. The "number of replies" of the target post is the number of posts included in the "replies" to the target post. The "number of likes" of the target post is the number of "likes" given to the target posted content by users other than the target user who read the target post (the number of positive responses). The "number of reposts" of the target post is the number of posts (e.g., retweets) by the target user with the same content as the target posted content. The "number of blocks" of the target post is the number of accounts of other users blocked by the target user on the target SNS at the time of the target posting. The "account creation date and time" for a target post is the date and time when the target user created an account for the target SNS. The "number of likes given by the posting user" for a target post is the number of "likes" given by the target user to posts by other users from the account creation date and time to the time of the target post (i.e., the period during which the target user used the target SNS). The "cumulative number of posts" for a target post is the cumulative number of posts by the target user to the target SNS up to the time of the target post. The "average number of posts per day by the posting user" for a target post is a value indicating the posting frequency of the target user, and is the average number of posts per day by the target user up to the time of the target post (or up to the day before the post in question). The "average number of posts per day by all users" for a target post is a value indicating the average posting frequency of all users of the target SNS, and is the average number of posts per day by all users up to the time of the target post (or up to the day before the post in question).

[0028] In addition, if it is difficult to directly input statistical information such as the "cumulative number of posts," the "average number of posts per day by the posting user," and the "average number of posts per day by all users" from the SNS server 20, the post-related information collecting unit 11 may obtain these values ​​by searching the SNS server 20. Specifically, the "cumulative number of posts" can be obtained by searching all past posts of the target user (up to the time of the target post or the day before the target post) and counting the number of posts found. The "average number of posts per day by the posting user" can be obtained by dividing the cumulative number of posts of the target user by the number of days elapsed from the date and time of the target user's account creation to the time of the target post or the day before the target post. The "average number of posts per day by all users" can be obtained by counting the number of all posts registered in the SNS server 20 and dividing the number of posts by the number of days elapsed from the start of posting to the SNS server 20 to the time of the target post or the day before the target post. However, the "average number of posts per day by all users" may be calculated for a predetermined period, such as the past year.

[0029] Fig. 4 is a diagram showing an example of the configuration of the post related information storage unit 16. As shown in Fig. 4, the post related information storage unit 16 stores, for each piece of post related information output by the post related information collection unit 11, the values ​​of each item constituting the post related information.

[0030] [Flame candidate determination section 12] The flame candidate determination unit 12 receives the post related information stored in the post related information storage unit 16 as input and determines the possibility that the user who posted the post related information is lacking in experience or attention on the target SNS. This is because posts by users who may be lacking in experience or attention are relatively more likely to cause a flame. If the flame candidate determination unit 12 determines that the user is lacking in experience or attention, it outputs (stores) the post related information (part of the information) to the flame candidate storage unit 17; otherwise, it does not output (store) the post related information to the flame candidate storage unit 17.

[0031] FIG. 5 is a flowchart for explaining an example of a processing procedure executed by the flaming candidate determination unit 12. The processing procedure of FIG. 5 may be executed following the processing by the post related information collection unit 11, or may be executed at regular intervals (for example, every 12 hours). In the former case, the post related information to be processed in FIG. 5 is the post related information last output by the post related information collection unit 11 to the post related information storage unit 16. In the latter case, the post related information to be processed in FIG. 5 is the post related information stored in the post related information storage unit 16 in the most recent regular time. When there are multiple pieces of applicable post related information, the processing of FIG. 5 may be executed for each of the multiple pieces of post related information. In the following, one "post related information" to be processed is referred to as "target post related information".

[0032] In step S101, the flame candidate determination unit 12 acquires the latest "average number of posts per day by posting user" and the latest "average number of posts per day by all users" for each user from the posting related information storage unit 16 (FIG. 4). Here, the latest "average number of posts per day by posting user" for a certain user refers to the "average number of posts per day by posting user" included in the latest posting related information including the user's username. This (the "average number of posts per day by posting user") is acquired for each user. In addition, the latest "average number of posts per day by all users" refers to the "average number of posts per day by all users" included in one piece of posting related information last stored in the posting related information storage unit 16, and the value is one. Hereinafter, the latest "average number of posts per day by all users" is referred to as "μ". Note that μ corresponds to the average of the latest "average number of posts per day by posting user" for each user.

[0033] Next, the flame candidate determination unit 12 calculates the standard deviation σ of μ for each user's latest “average number of posts per day by posting user” (S102).

[0034] Next, the flame candidate determination unit 12 determines whether the absolute value of the difference between the "average number of posts per day by the posting user" of the target posting related information and μ is 1σ or more (S103). That is, it is determined whether the "average number of posts per day by the posting user" is smaller than 1σ with respect to μ, or is 1σ or more larger than μ. Such a determination corresponds to a determination of whether the "average number of posts per day by the posting user" of the target posting related information has a predetermined deviation with respect to the "average number of posts per day by all users". In addition, such a determination means a determination of whether the user related to the target posting related information (hereinafter referred to as the "target user") is inexperienced or inattentive. That is, in this embodiment, if the "average number of posts per day by the posting user" is smaller than 1σ with respect to μ, it is determined that the target user is inexperienced with respect to the target SNS. On the other hand, if the "average number of posts per day by the posting user" is larger than μ by 1σ or more, it is determined that the target user is inattentive with respect to the target SNS. Although an example has been described in which the threshold value for the difference is 1σ, a value other than 1σ may be used as the threshold value.

[0035] If the absolute value of the difference between the "average number of posts per day by the posting user" of the target post related information and μ is less than 1σ (No in S103), the processes in and after step S104 are not executed for the target post related information.

[0036] If the absolute value of the difference between the "average number of posts per day by the posting user" of the target post related information and μ is 1σ or more (Yes in S103), the flame candidate determination unit 12 determines whether the "number of likes" of the target post related information is equal to or more than a threshold value α (S104). The threshold value α is a value that is set in advance as a value representing a reaction to a flame post, and is, for example, 100. If the "number of likes" is less than the threshold value α (No in S104), the processing from step S105 onwards is not performed on the target post related information.

[0037] If the "number of likes" is 100 or more (Yes in S104), the flame candidate determination unit 12 determines whether the "number of replies" of the target post-related information is equal to or greater than a threshold value β (S105). The threshold value β is a value that is set in advance as a value representing a reaction to a flame post, and is, for example, 100. If the "number of replies" is less than the threshold value β (No in S105), step S106 is not executed for the target post-related information. If the "number of replies" is equal to or greater than the threshold value β (Yes in S105), the flame candidate determination unit 12 stores, among the information constituting the target post-related information, information necessary for determining the possibility of a flame in the flame candidate storage unit 17 (S106).

[0038] Fig. 6 is a diagram showing a configuration example of the flame candidate storage unit 17. As shown in Fig. 6, the flame candidate storage unit 17 stores information (hereinafter referred to as "candidate post information") including "post content", "user name", "profile text", "post date and time", "reply", "number of replies", "number of reposts", "number of blocks", "account creation date and time", "number of 'Likes' given by the posting user" and "cumulative number of posts" of post related information that meets the conditions of steps S103 to S105 in Fig. 5 among the post related information stored in the post related information storage unit 16.

[0039] [Other SNS information acquisition section 13] The other SNS information acquisition unit 13 acquires the user name of each account registered in the other SNS from the other SNS server 30 at regular time intervals (every 12 hours), for example, and stores the acquired user name in the other SNS information storage unit 18. Note that, if the acquired user name has already been stored in the other SNS information storage unit 18, the other SNS information acquisition unit 13 does not need to store the user name in the other SNS information storage unit 18.

[0040] Fig. 7 is a diagram showing an example of the configuration of the other SNS information storage unit 18. As shown in Fig. 7, the other SNS information storage unit 18 stores information including the "user name" of each account registered in the other SNS server 30 (hereinafter referred to as "other SNS information").

[0041] [Experience Estimation Part 14] The experience value estimation unit 14 receives the candidate posted information stored in the flame candidate storage unit 17 and the other SNS information stored in the other SNS information storage unit 18 as input, and calculates the experience value of the user related to the candidate posted information. If the experience value is equal to or less than the threshold value γ (i.e., if the user is highly likely to be inexperienced), the experience value estimation unit 14 outputs the posted content and reply of the candidate posted information. The experience value calculated by the experience value estimation unit 14 is an index indicating the level of experience in using the Internet, including not only the target SNS but also other SNSs. The threshold value γ is a value that is set in advance based on the experience value calculated by the experience value estimation unit 14. As an example, the average is calculated from the values ​​(d) of each experience value of the experience value estimation unit 14, and the average is set as the threshold value.

[0042] FIG. 8 is a flowchart for explaining an example of a processing procedure executed by the experience value estimation unit 14. The processing procedure of FIG. 8 may be executed following the processing by the flame candidate determination unit 12, or may be executed at regular intervals (for example, every 12 hours). In the former case, the candidate posted information to be processed in FIG. 8 is the candidate posted information last output by the flame candidate determination unit 12 to the flame candidate storage unit 17. In the latter case, the candidate posted information to be processed in FIG. 8 is the candidate posted information stored in the flame candidate storage unit 17 in the most recent regular time. When there are multiple pieces of applicable candidate posted information, the processing of FIG. 8 may be executed for each of the multiple pieces of candidate posted information. In the following, one "candidate posted information" to be processed is referred to as "target candidate posted information".

[0043] In step S201, the experience value estimation unit 14 calculates an experience value a for a user related to the target candidate posted information (hereinafter referred to as the "target user") based on the "posted content," "cumulative number of posts," "number of reposts," "number of likes given by the posting user," and "number of blocks" of the target candidate posted information.

[0044] The experience value a is calculated using the following formula (1) when the “post content” of the target candidate posting information does not include a hashtag, and is calculated using the following formula (2) when the “post content” of the target candidate posting information includes a hashtag. Experience value a = (T + I + RT + 10B) / average cumulative number of posts ··· (1) Experience value a = (2T + I + RT + 10B) / average cumulative number of posts (2) The meaning of each variable is as follows: T: "Cumulative number of posts" RT: "Number of reposts" I: "The number of likes given by the posting user" B: "Number of blocks" In other words, if the "post content" contains a hashtag, the weight (contribution) of the "cumulative number of posts" to the experience value a is doubled. The average cumulative number of posts is the average of the cumulative number of posts for each user. The experience value a is normalized by dividing it by the average cumulative number of posts.

[0045] Next, the experience value estimation unit 14 calculates the experience value b based on the number of days elapsed from the “account creation date and time” of the target candidate posted information to the “posting date and time” of the target candidate posted information (S202). The experience value b is calculated based on, for example, the following formula (3). B=P / 365 Here, P is the number of days that have passed since the "account creation date and time" to the "post date and time." The denominator, 365, is the number of days in a year. Dividing by 365 normalizes the experience value b.

[0046] Next, the experience value estimation unit 14 calculates the experience value c based on the similarity between the “user name” of the target candidate posting information (hereinafter referred to as the “target user name”) and each “user name” of the other SNS information stored in the other SNS information storage unit 18 (Figure 7) (hereinafter referred to as the “other user name”) (S203).

[0047] That is, even if the target user is determined to have little experience with the target SNS, if the target user uses other SNSs, the target user's Internet experience value is considered to be high. Therefore, in step S203, the accuracy of the experience value is improved by checking whether the target user uses other SNSs. In this embodiment, in order to check whether the target user uses other SNSs, if there is an account in the other SNS with a user name similar to the target user name, the account is determined to be the target user's account. That is, in this case, the target user is determined to use other SNSs.

[0048] Specifically, the experience value estimation unit 14 calculates the similarity between each other user name and the target user name. For example, the Levenshtein distance may be used as the similarity. The Levenshtein distance D(C1, C2) between the character string C1 and the character string C2 is evaluated based on the minimum number of steps required to transform the character string C1 into the character string C1 by inserting, deleting, or replacing characters. When calculating the Levenshtein distance between the other user name and the target user name, if the other user name or the target user name contains alphabets, the experience value estimation unit 14 standardizes each alphabet to lower case. The experience value estimation unit 14 also deletes characters used as delimiters, such as "_" and ".", from the other user name and the target user name. The experience value estimation unit 14 then calculates the Levenshtein distance between the other user name and the target user name, and normalizes the Levenshtein distance by dividing the Levenshtein distance by the number of characters in the other user name or the target user name that has the larger number of characters. Therefore, the normalization of the Levenshtein distance D(C1, C2) is expressed by the following formula: D normalize (C1,C2)=d(C1,C2) / max(length(C1),length(C2)) The normalized Levenshtein distance (similarity) is calculated for each of the other user names. The experience value estimation unit 14 sets the maximum value of the normalized Levenshtein distances calculated for each of the other user names as the experience value c.

[0049] Next, the experience value estimation unit 14 calculates a weighted sum of the experience value a, the experience value b, and the experience value c as the experience value d (S204). In the weighted sum, the weights for the experience value b and the experience value c may be calculated by applying an increasing function to the age of the target user. This is because it is generally said that the older a person is, the more likely they are to use the Internet appropriately. In this case, the post related information collection unit 11 may also collect the age of the target user from the target SNS and include it in the post related information.

[0050] The empirical value d may be calculated based on any two of the empirical values ​​a, b, and c, or may be calculated based on any one of them. For example, any one of the empirical values ​​a, b, and c may be directly set as the empirical value d.

[0051] Next, the experience value estimation unit 14 judges whether the experience value d is less than the threshold value γ (S205). If the experience value d is less than the threshold value γ (Yes in S205), the experience value estimation unit 14 outputs the "post content" and the "reply" of the target candidate posted information to the inflaming potential determination unit 15 (S206). On the other hand, if the experience value d is equal to or greater than the threshold value γ (No in S205), step S206 is not executed. Therefore, in this case, the process by the inflaming potential determination unit 15 is not executed.

[0052] [Flame-warning Judgment Section 15] When the experience value estimation unit 14 outputs a post content and a reply, the flame war potential determination unit 15 inputs the post content (hereinafter referred to as the "target post content") and the reply (hereinafter referred to as the "target reply"), and when it determines that the post content has a high flame war potential, outputs the post content to the corpus memory unit 19.

[0053] Specifically, the flame war potential determination unit 15 performs morphological analysis on the target post content and the reply, and determines whether or not a word having a negative meaning is included in the group of morphemes obtained as a result of the morphological analysis. The determination of whether or not a word has a negative meaning depends on the Japanese polarity dictionary used. For example, when a Japanese polarity dictionary in which a flag of "positive" or "negative" is simply assigned to a word is used, a word to which a flag of "negative" is assigned may be determined as a word having a negative meaning. On the other hand, when a Japanese polarity dictionary in which a degree of negativity (score) is assigned is used, a threshold value for the score may be set, and a word having a negative meaning may be determined based on the threshold value. For example, when the range of the score is -1 (negative) to 1 (positive), a word having a score less than 0 may be determined as a word having a negative meaning.

[0054] If the morpheme group (i.e., at least one of the target post content and the reply) contains a word with a negative meaning, the flame potential determination unit 15 determines that the target post content is highly flame potential, and adds the target post content to a corpus (a collection of flame potential posts) stored in the corpus memory unit 19 as flame potential post content.

[0055] Fig. 9 is a diagram showing a configuration example of the corpus storage unit 19. As shown in Fig. 9, the corpus storage unit 19 stores flaming post contents that the flaming potential determination unit 15 has determined to be highly flaming likely. The example in Fig. 9 shows an example in which the following post contents are stored as flaming post contents. "I don't think anyone would be affected if an earthquake were to occur in ___. I think it's better to stop trying to raise money through earthquakes." "I'm not a mascot character who's rude to people for taking photos." By training a model such as a neural network using machine learning to learn the contents of each flaming post included in such a corpus and using the trained model, it is expected that flaming caused by careless postings can be prevented. For example, at the time a post is made, the trained model can estimate the possibility of the post causing a flaming, and if there is a high possibility of the post causing a flaming, the poster can be notified of the possibility of the post causing a flaming before the post is made public. In this case, the poster can review the post in response to such a notification.

[0056] As described above, according to the present embodiment, it is possible to determine the possibility that each posted content will cause a flaming incident due to carelessness, and therefore it is possible to collect posted content that is likely to cause a flaming incident due to carelessness.

[0057] The target SNS is an example of a first SNS. The other SNS is an example of a second SNS. The flame candidate determination unit 12 is an example of a first determination unit. The threshold value γ is an example of a first threshold. The experience value estimation unit 14 is an example of an estimation unit. The flame likelihood determination unit 15 is an example of a second determination unit. The threshold value α or threshold value β is an example of a second threshold.

[0058] Although the embodiment of the present invention has been described in detail above, the present invention is not limited to such specific embodiment, and various modifications and variations are possible within the scope of the gist of the present invention described in the claims. [Explanation of symbols]

[0059] 10 Flaming Judgment Device 11 Post-related Information Collection Department 12 Flaming candidate determination section 13 Other SNS information acquisition department 14 Experience Estimation Unit 15 Flaming Judgment Department 16 Post related information storage section 17 Burning Candidate Memory Unit 18 Other SNS information storage unit 19 Corpus Storage 20 Social Networking Server 30 Other SNS servers 100 Drive device 101 Recording media 102 Auxiliary storage 103 Memory device 104 processors 105 Interface device B Bus

Claims

1. a first determination unit that determines whether or not a posting frequency to a first SNS by a first user who has posted a certain message to the first SNS has a predetermined deviation from an average posting frequency of all users who use the first SNS; an estimation unit that estimates an experience value indicating a level of experience regarding the use of the Internet for the first user based on a possibility of the first user using a second SNS different from the first SNS when the predetermined difference exists; a second determination unit that, when the experience value is less than a first threshold, determines whether or not a negative expression is included in the content of the certain post, and, if the negative expression is included, adds the content of the certain post to a set of post contents that may cause a flame war; A flaming potential determination device comprising:

2. The first determination unit further determines whether or not a number of reactions from other users to the certain post is equal to or greater than a second threshold value; The estimation unit estimates the empirical value when the predetermined deviation exists and the number of the responses is equal to or greater than the second threshold value.

2. The flaming potential determination device according to claim 1.

3. The estimation unit estimates the experience value based on a similarity between a username of the first user in the first SNS and a username in the second SNS.

3. The flaming potential determination device according to claim 1 or 2.

4. The estimation unit further estimates the experience value based on at least one of a number of times the content of the certain post has been reposted, a period of time the first user has used the first SNS, and a number of accounts blocked by the first user in the first SNS.

4. The flaming potential determination device according to claim 1, wherein the flaming potential determination device is a device for determining whether or not a flaming potential is present.

5. the second determination unit determines whether or not a negative word is included in a content of the certain post or a post in response to the certain post, and adds the content of the certain post to the set if the negative word is included.

5. The flaming potential determination device according to claim 1, wherein the flaming potential determination device is a device for determining whether or not a flaming potential is present.

6. a first determination step of determining whether or not a posting frequency to a first SNS by a first user who has posted a certain message to the first SNS has a predetermined deviation from an average posting frequency of all users who use the first SNS; an estimation step of estimating an experience value indicating a level of experience regarding the use of the Internet for the first user based on a possibility of the first user using a second SNS different from the first SNS when the predetermined difference exists; a second determination step of determining whether or not a negative expression is included in the content of the certain post when the experience value is less than a first threshold, and adding the content of the certain post to a set of post contents that may cause a flame war if the negative expression is included; A method for determining whether a content is likely to explode, the method being characterized by being executed by a computer.

7. A program characterized by causing a computer to function as a flaming tendency determination device described in any one of claims 1 to 5.

Images