JP6249794B2 - Bot determination device, bot determination method, and program - Google Patents

Description

  The present invention relates to a bot determination apparatus, a bot determination method, and a program for determining whether or not a user posting a message is a bot in SNS (Social Networking Service).

  There is a large amount of information (spam) such as advertisements that hinders information retrieval on the Internet, and various studies have been conducted to remove them. On SNS (Social Networking Service), there are a large number of systems called bots that automatically speak based on rules, and the utterances that are automatically output on behalf of users are collected from SNS. It is a great hindrance to doing.

  In the existing mail / URL spam filter, if positive examples and negative examples are given in advance, it is possible to discriminate spam using supervised learning such as Bayesian estimation (for example, see Non-Patent Document 1). . Since these are basically determined by whether or not they are similar to the correct answer example of spam, in order to use on SNS where the word used changes every day, the teacher data that indicates whether spam is correct or not is updated daily. There was a problem that it was necessary to go and a lot of labor was required.

  In order to deal with this problem, a method has been proposed (for example, refer to Non-Patent Document 2) in which characteristics such as meta information such as a text and URL reference destination and structure information such as header / tag / script are relatively small. However, if the above characteristics of spam change, it is necessary to update the learning data again.

  On the other hand, there has been proposed a technique for evaluating spam without positive / negative teacher data by evaluating similarity between documents and determining if there are many similar documents as spam (for example, patents). Reference 1 and non-patent reference 3). Patent Document 1 proposes a method for preparing a plurality of receiving accounts for determining e-mail spam, comparing the results received by two accounts, and determining if there is a similar e-mail as spam. Yes. In Non-Patent Document 3, this technique is further generalized to extract a hash feature quantity from the received mail body document, and to extract a similar mail group by calculating the density of mail existing in the feature space. A technique has been proposed.

JP 2007-86821 A

Paul Graham. A plan for spam, In P.A. Graham, Hackers and Painters. O’Reilly. O'Reilly, 2004. Junpei Kitamura, Masaki Aono “Detection of Web Spam Using Similarity Between Websites”, DBSJ Journal, Vol. 8, no. 1, pp. 143-148, 2009. Density-Based Spam Detector, Kenichi Yoshida, Fujimori Adachi, Takashi Wasio, Hiroshi Motoda, Teraki Hamakata, Akihiro Hashimashi. Inf. & Syst. , Vol. E87-D, no. 12, pp. 2678-2688, December 2004 (Japanese Patent Laid-Open No. 2005-202590)

  By the way, unlike e-mail, a user ID is not disposable in SNS. Therefore, it is correct that a user who has many utterances determined to be spam is a bot, and that the user's utterance is determined to be spam. However, in the methods proposed in Non-Patent Documents 1 to 3 and Patent Document 1, spam is determined for each utterance (document), so even a bot user's utterance is spam. If it is not similar to the speech determined to be, it is not determined to be spam, and there is a problem that spam cannot be efficiently removed.

  Therefore, in view of the above problems, the present invention is based on the SNS characteristics that the user ID is not thrown away and the posting of the utterance in accordance with a pre-programmed rule in the SNS, whether or not the user posting the utterance is a bot. An object of the present invention is to provide a bot determination device, a bot determination method, and a program that efficiently determine without using learning data by using the characteristic of the bot to be performed.

  The present invention proposes the following matters in order to solve the above problems. In addition, in order to make an understanding easy, although the code | symbol corresponding to embodiment of this invention is attached | subjected and demonstrated, it is not limited to this.

(1) The present invention relates to a bot determination device that determines whether or not a user posting a comment is a bot in an SNS (Social Networking Service) that accepts and publishes posted comments. 1 bot determination device 100), and a message storage means (for example, message storage unit 120 in FIG. 1) for storing message information including at least posted user information and posted date / time information of a message posted to the SNS; for determining the target user, the talk on the basis of the storage means speech information stored in the other user and determining regularity index calculation means for calculating a regularity index indicating the regularity of the definitive post timing between the target user ( For example, the regularity index calculation unit 130) of FIG. 1 and the regularity index calculated by the regularity index calculation means are set in advance. And a user determination unit (for example, the user determination unit 110 in FIG. 1) for determining whether or not the determination target user is the bot by comparing with a determined threshold value. ing.

  (2) In the bot determination device according to (1), the regularity index calculation unit includes, in the posting date information of the determination target user, based on the statement information stored in the statement storage unit. The nearest posting date / time list acquisition means (for example, the nearest posting date / time list acquisition unit 141 in FIG. 1) that acquires the nearest posting date / time list that associates the posting date / time information of each of the other users closest to the posting date / time information. And an individual phase that calculates an individual correlation coefficient indicating the relevance of the posting timing between each of the other users and the determination target user based on the nearest posting date list acquired by the nearest posting date list acquisition unit. From the correlation number calculating means (for example, the individual correlation coefficient calculating unit 142 in FIG. 1) and the individual correlation coefficient calculated by the individual correlation coefficient calculating means, the determination target user The individual correlation coefficient of another user who has a tendency to post a statement before is extracted based on the nearest posting date list acquired by the nearest posting date list acquisition means, and the maximum correlation coefficient of the extracted individual correlation coefficient Timing correlation coefficient determination means (for example, timing correlation coefficient determination unit 143 in FIG. 1), and the user determination means is set in advance as the timing correlation coefficient. A bot determination device has been proposed in which it is determined whether or not the determination target user is the bot based on the magnitude relationship with the threshold value.

  (3) In the bot determination device according to (1) or (2), the speech information in the speech storage means includes inter-speech relationship information indicating a relationship with other speeches posted to the SNS. In addition, the regularity index calculating means obtains a response date / time list that is a list of post date / time information of response messages of the determination target user based on the message information stored in the message storage means It is a list of post date information of original utterances of the response utterance of the determination target user based on utterance information stored in the utterance storage means (for example, response date / time list acquisition unit 151 in FIG. 1) and the utterance storage unit. An original posting date / time list acquisition unit (for example, an original posting date / time list acquisition unit 152 in FIG. 1) for acquiring an original posting date / time list, a response date / time list acquired by the response date / time list acquisition unit, Based on the original posting date list acquired by the original posting date list acquisition means, the response time until the determination target user posts the response message is calculated, and the average of the calculated response times is calculated by the determination target user. User response time determination means (for example, user response time determination unit 153 in FIG. 1) that is determined as a user response time, wherein the user determination means is a user response time acquired by the user response time determination means; A bot determination device is proposed that determines whether or not the determination target user is the bot based on a magnitude relationship with the preset threshold value.

  (4) In the bot determination device according to (3), the regularity index calculation unit may be configured to have the determination target user and one other user based on the statement information stored in the statement storage unit. A continuous response section extracting means (for example, the continuous response section extracting unit 161 in FIG. 1) for extracting the utterance information of the continuous response section in which utterances are continuously exchanged at a predetermined interval or less, and the continuous response section Based on the utterance information of the continuous response section extracted by the extracting means, the response interval of each utterance included in the continuous response section is calculated, and the replacement latency characteristic amount is calculated from the difference of the calculated response intervals. A temporal feature amount calculating means (for example, a substitute latency feature amount calculating unit 160 in FIG. 1), and the user determining means calculates the substitute latency feature amount calculated by the substitute latency feature amount calculating means from the substitute latency feature amount Target user It proposes a bot determination apparatus characterized by determining whether a said bot.

  (5) In the bot determination device according to (4), the utterance information in the utterance storage unit further includes content information of each utterance posted to the SNS, and the regularity index calculation unit includes the continuous index calculation unit. Based on the content information included in the utterance information of the continuous response section extracted by the response section extraction means, consent determination means for determining whether or not each utterance included in the continuous response section agrees with the previous utterance (for example, 1, and the alternation latency feature value calculating unit weights the response interval according to the determination result of the consent determination unit of each utterance included in the continuous response section. Has proposed a bot determination device characterized by the following.

  (6) In the bot determination device according to any one of (1) to (5), the regularity index calculation means is based on the comment information stored in the comment storage means, and the posting date and time of the determination target user An analysis posting date / time list acquisition unit (for example, an analysis posting date / time list acquisition unit 171 in FIG. 1) that acquires a list of information, and a list of posting date / time information of the determination target user acquired by the analysis posting date / time list acquisition unit is a spectrum. A spectrum analysis unit (for example, the spectrum analysis unit 172 in FIG. 1) that analyzes and acquires the maximum amplitude value of the amplitude spectrum as a periodicity index, and the user determination unit acquires the period acquired by the spectrum analysis unit It is characterized by determining whether or not the determination target user is the bot based on a magnitude relationship between a sex index and the preset threshold value. It has proposed that bot determination device.

  (7) In the bot determination device according to (6), the spectrum analysis unit uses the discrete Fourier transform method or the piccolo method, and the posting date / time of the determination target user acquired by the analysis posting date / time list acquisition unit It proposes a bot determination device characterized by analyzing a list of information and calculating the periodicity index.

  (8) In the bot determination device according to (6) or (7), the speech storage unit further includes inter-speech relationship information indicating a relationship with other utterances posted to the SNS. In addition, the analysis posting date / time list acquisition unit acquires a list of posting date / time information of statements other than response messages for the determination target user based on the inter-speech relationship information of the speech information stored in the speech storage unit A bot determination device characterized by the above is proposed.

(9) The present invention relates to a bot determination device that determines whether or not the user who posted the utterance is a bot in an SNS (Social Networking Service) that accepts and publishes the posted comments. 2 bot determination device 200), which stores message information storing means for storing message information including at least posted user information and posted date information of messages posted to the SNS, and a determination target user stored in the message storage means. based on the speech information that is, definitive regularity index calculating means for calculating a plurality of types of regularity index indicating the regularity of the post timing (e.g., regularity index of 2 between the other users and the determination target user A calculation unit 131), a user determined to be a bot, and a user determined to be not the bot An integrated model learning means for learning an integrated model obtained by weighting and integrating each of the plurality of types of regularity indices using a predetermined learning algorithm using the plurality of types of regularity indices as teacher data (for example, FIG. 2). Integrated model learning unit 210), the determination target user based on the plurality of types of regularity indexes calculated by the regularity index calculation means for the determination target user and the integrated model learned by the integrated model learning means Has proposed an integrated model determination means (for example, an integrated model determination unit 230 in FIG. 2) that determines whether or not is a bot.

(10) The present invention relates to a bot determination method in a bot determination device that determines whether or not a user who has posted a comment is a bot in an SNS (Social Networking Service) that accepts and discloses posted comments. The bot determination device stores message information including at least posted user information and posted date and time information of a message posted to the SNS, regularity index calculation means, user determination means, A first step (for example, step S1 in FIG. 2) for receiving an input of a determination target user for determining whether or not the user is a bot; and for the target user, based on the speech information stored in said speech storage means, determination target user and other users A second step of calculating the regularity index indicating the regularity of the definitive post timing between the (e.g., from step S2 of FIG. 2 S7) and, said user determination means, rules that are calculated in the second step A third step (for example, steps S8 to S10 in FIG. 2) for determining whether or not the determination target user is the bot by comparing the sex index with a preset threshold; A bot determination method including

(11) The present invention relates to a bot determination method in a bot determination device that determines whether or not a user who has posted a comment is a bot in a social networking service (SNS) that accepts and publishes the posted comment. , A bot determination device that stores utterance information including at least posted user information and posted date information of the utterance posted to the SNS, and regularity index calculation A first step (for example, step S1 in FIG. 2) for receiving an input of a determination target user for determining whether the user is a bot, and the rule. The sex index calculating means stores the speech stored in the speech storage means for the determination target user. Based on the information, a second step of calculating the regularity index indicating the regularity of the definitive post timing between the other users and the determination target user (e.g., S7 from step S2 of FIG. 2), the user determination unit A third step of determining whether or not the determination target user is the bot by comparing the regularity index calculated in the second step with a preset threshold value ( For example, a program for causing a computer to execute steps S8 to S10) of FIG. 2 is proposed.

  According to the present invention, in the SNS, whether or not the user posting the speech is a bot, whether the user ID is not disposable, the SNS characteristics, and the posting of the speech is performed according to pre-programmed rules. Using the characteristics, it is possible to make an efficient determination without requiring learning data. As a result, in the SNS, bot remarks can be removed, and spam can be efficiently removed.

It is a figure which shows the function structure of the bot determination apparatus which concerns on the 1st Embodiment of this invention. It is a figure which shows the bot determination processing flow using the timing correlation coefficient in the bot determination apparatus which concerns on the 1st Embodiment of this invention. It is a figure which shows the bot determination processing flow using the user response time in the bot determination apparatus which concerns on the 1st Embodiment of this invention. It is a figure which shows the bot determination processing flow using the alternative latency feature-value in the bot determination apparatus which concerns on the 1st Embodiment of this invention. It is a figure which shows the bot determination processing flow using the spectrum analysis in the bot determination apparatus which concerns on the 1st Embodiment of this invention. It is a figure which shows the function structure of the bot determination apparatus which concerns on the 2nd Embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. Note that the constituent elements in the present embodiment can be appropriately replaced with existing constituent elements and the like, and various variations including combinations with other existing constituent elements are possible. Therefore, the description of the present embodiment does not limit the contents of the invention described in the claims.

<First Embodiment>
<Functional configuration of bot determination device>
FIG. 1 is a diagram illustrating a functional configuration of a bot determination device 100 according to the first embodiment of the present invention. The bot determination device 100 is a device that determines whether or not a user who has posted a comment is a bot in SNS (Social Networking Service), and includes a user determination unit 110, a message storage unit 120, and a regularity index calculation. Unit 130 and threshold value storage unit 180. Here, the bot in the present invention means an automatic speech system by a machine that speaks information (spam) such as advertisements.

  The user determination unit 110 starts processing in response to input of a user ID in the SNS of a user (hereinafter referred to as a determination target user) that determines whether or not the user is a bot, or speech information of the determination target user. Then, the user ID or the utterance information of the determination target user is transmitted to the regularity index calculation unit 130. When the determination target user's utterance information is input, the user determination unit 110 stores the determination target user's utterance information in the utterance storage unit 120. The comment information will be described in a comment storage unit 120 described later.

  When receiving the regularity index from the regularity index calculation unit 130, the user determination unit 110 receives the regularity index received from the regularity index calculation unit 130 and the threshold value stored in the threshold value storage unit 180. And whether or not the determination target user is a bot. The regularity index is an index representing the regularity of the posting timing of the determination target user, and the user determination unit 110 compares the regularity index with a threshold value, so that the speech timing of the determination target user has regularity. It is determined whether or not.

  Since posting of utterances by bots is performed according to pre-programmed rules, regularity occurs in the posting timing. Therefore, when the user determination unit 110 determines that there is regularity, the determination target user is a bot. On the other hand, if it is determined that there is no regularity, the determination target user is determined as a non-bot non-bot.

  The comment storage unit 120 stores the comment information of the comments posted to the SNS. Here, the utterance information includes the user ID of the user who posted the utterance, post date information of the utterance posted to the SNS, inter-speech relation information indicating the relationship with other utterances posted to the SNS, and the content of the utterance. The content information indicating is included. The message information stored in the message storage unit 120 may be automatically acquired periodically from the SNS server, or may be manually registered by the administrator of the bot determination device 100.

  Here, the relationship between each utterance and other utterances refers to a relationship in which each utterance re-posts (including replies) or quotes (including references) utterances posted by other users. In the present invention, the utterance in the above is called a response utterance. For example, the inter-speech relationship information includes information that identifies whether the response is a response speech and the original speech to be responded to, and uses the posting rules applied within the posting site, followed by the player's post This can be realized by extracting posts including the site user name.

  The regularity index calculation unit 130 includes a timing correlation coefficient calculation unit 140, a user response time calculation unit 150, an alternating latency feature value calculation unit 160, and a periodicity index calculation unit 170. The regularity index calculation unit 130 includes at least one of the timing correlation coefficient calculation unit 140, the user response time calculation unit 150, the alternation latency feature value calculation unit 160, and the periodicity index calculation unit 170. That's fine.

  The timing correlation coefficient calculation unit 140 calculates a timing correlation coefficient that represents the correlation of the posting timing of a statement between the determination target user and another user as a regularity index. By the way, since the posting of the utterance by the bot is performed according to a pre-programmed rule, the posting of the utterance by the bot often has a regularity of following the posting of the utterance by the specific user. In this case, since a correlation appears in the posting timing of the speech between the bot and the specific user, it is possible to determine whether the determination target user is a bot using the timing correlation coefficient as the regularity index. it can.

  As shown in FIG. 1, the timing correlation coefficient calculation unit 140 includes a nearest posting date / time list acquisition unit 141, an individual correlation coefficient calculation unit 142, and a timing correlation coefficient determination unit 143.

  The nearest posting date / time list acquisition unit 141 adds the posting date / time information of each of the other users closest to each posting date / time information to each posting date / time information of the determination target user based on the statement information stored in the statement storage unit 120. Acquires the nearest posting date list associated.

  When acquiring the user ID of the determination target user from the user determination unit 110, the nearest neighbor posting date list acquisition unit 141 acquires the determination target user posting date information from the statement storage unit 120 based on the acquired user ID. The nearest neighbor posting date / time list acquisition unit 141 may acquire all posting date / time information stored in the comment storage unit 120, or specify a period and acquire only posting date / time information within the period. Also good.

  When the utterance information is acquired from the user determination unit 110, the nearest posting date / time list acquisition unit 141 replaces the posting date information of the determination target user acquired from the utterance storage unit 120 with the utterance information from the user determination unit 110. The posted date information included may be used. Further, the nearest posting date / time list acquisition unit 141 acquires the posting date / time information of the determination target user from the statement storage unit 120 based on the user ID of the determination target user included in the remark information acquired from the user determination unit 110. Also good.

  The nearest posting date / time list acquisition unit 141 acquires the user IDs of users other than the determination target user from the statement storage unit 120 based on the user ID of the determination target user. When speech information is acquired from the user determination unit 110, user IDs of users other than the determination target user are acquired from the speech storage unit 120 based on the user ID of the determination target user included in the speech information. At this time, only users other than the determination target users who have a predetermined number of statements may be set as other users.

  The nearest neighbor posting date list acquisition unit 141 acquires posting date information for each other user based on the acquired user ID of the other user. Then, the nearest posting date / time list acquisition unit 141 extracts each posting date / time information closest to each posting date / time information of the determination target user from the posting date / time information of the other users, and extracts each posting date / time information of the determination target user. The association and the same processing are performed for all other users to create the nearest posting date list.

  The individual correlation coefficient calculation unit 142 is a determination target user that indicates the relevance of the posting timing between the determination target user and each of the other users based on the nearest posting date / time list acquired by the nearest posting date / time list acquisition unit 141. The individual correlation coefficient between the posting date / time information and the posting date / time information of other users is calculated. Specifically, by substituting the i-th posting date / time information of the determination target user and the i-th posting date / time information of the other user from the nearest posting date / time list into the following formula (1), the determination target user and the other user The individual correlation coefficient is calculated. The individual correlation coefficient calculation unit 142 calculates a correlation coefficient with the determination target user for all other users.

  The timing correlation coefficient determination unit 143 selects the individual correlation coefficient of another user who has a tendency to post a comment before the determination target user from among the individual correlation coefficients calculated by the individual correlation coefficient calculation unit 142. Extracted based on the nearest posting date / time list acquired by the side posting date / time list acquisition unit 141, and the maximum correlation coefficient of the extracted individual correlation coefficient is determined as a timing correlation coefficient.

  Specifically, the timing correlation coefficient determination unit 143 determines the average value of the posting date information of the determination target user and each of the other users based on the nearest posting date list acquired by the nearest posting date list acquisition unit 141. The average value of the posting date information is calculated. Next, the timing correlation coefficient determination unit 143 specifies another user whose average value of the posting date information is smaller than the average value of the posting date information of the determination target user. It can be determined that another user whose average value of the posting date information is smaller than the average value of the posting date information of the determination target user has a tendency to post before the determination target user. Then, the timing correlation coefficient determination unit 143 determines the maximum value of the individual correlation coefficient between the identified other user and the determination target user as the timing correlation coefficient.

  Since the posting of the utterance by the bot follows the posting of the utterance of the other user, the timing correlation coefficient used to determine whether or not the bot is the maximum individual correlation coefficient with the other user that the target user follows Determine as. If the average value of the posting date information is smaller than the average value of the posting date information of the determination target user and the timing correlation coefficient determination unit 143 cannot identify the other user, the other user that the determination target user follows Therefore, it can be determined that the determination target user is not a bot. In this case, it is desirable to replace the timing correlation coefficient with 0 or a negative value in order to cause the user determination unit 110 to always determine that the determination target user is not a bot.

  The larger the timing correlation coefficient, the more relevant the posting timing between the determination target user and other users. The user determination unit 110 compares the magnitude correlation between the timing correlation coefficient calculated as the regularity index and the threshold value stored in the threshold value storage unit 180, and the user determination unit 110 has a timing correlation coefficient larger than the threshold value. In this case, it is determined that the determination target user is a bot. On the other hand, if the timing correlation coefficient is equal to or less than the threshold value, it is determined that the determination target user is not a bot.

  The user response time calculation unit 150 calculates the response time of the determination target user as a regularity index. By the way, since posting of utterances by bots is performed according to pre-programmed rules, bots often have regularity that responds faster than humans. It can be determined whether or not the user is a bot.

  As illustrated in FIG. 1, the user response time calculation unit 150 includes a response date / time list acquisition unit 151, an original posting date / time list acquisition unit 152, and a user response time determination unit 153.

  When the response date and time list acquisition unit 151 acquires the user ID of the determination target user from the user determination unit 110, the response date and time list acquisition unit 151 is a response message included in the inter-sentence relationship information stored in the message storage unit 120 in association with the acquired user ID. Based on the information on whether or not there is a response date / time list, which is a list of post date / time information of response speech of the determination target user, is acquired from the statement storage unit 120.

  When the response date and time list acquisition unit 151 acquires the utterance information of the determination target user from the user determination unit 110, the response date and time of posting the response utterance of the determination target user based on the inter-utterance relation information included in the acquired utterance information Get a response date list that is a list of information. The response date / time list acquisition unit 151 responds based on the inter-sentence relationship information stored in the speech storage unit 120 in association with the user ID of the determination target user included in the speech information acquired from the user determination unit 110. The date / time list may be acquired from the comment storage unit 120.

  When the original posting date / time list acquisition unit 152 acquires the user ID of the determination target user from the user determination unit 110, the response target included in the inter-sentence relationship information stored in the statement storage unit 120 in association with the acquired user ID. Based on the information specifying the original utterance, the original posting date / time list, which is a list of the posting date / time information of the original utterance of each response utterance of the determination target user, is acquired from the utterance storage unit 120. The i-th posting date / time information in the original posting date / time list is the original message of the response message corresponding to the i-th posting date / time information in the response date / time list.

  When the original posting date / time list acquisition unit 152 acquires the utterance information of the determination target user from the user determination unit 110, the original posting date / time list acquisition unit 152 is based on information that specifies the original utterance of the response target included in the inter-utterance relationship information of the acquired utterance information. Thus, an original posting date list that is a list of original posting date information of each response message of the determination target user is acquired. The original posting date / time list acquisition unit 152 is a response included in the inter-sentence relationship information stored in the speech storage unit 120 in association with the user ID of the determination target user included in the speech information acquired from the user determination unit 110. The response date / time list may be acquired from the message storage unit 120 based on information specifying the target original message.

  The user response time determination unit 153 calculates a response time until the determination target user posts each response message based on the corresponding posting date information in the response date list and the original posting date list, and calculates the calculated response time Is determined as the user response time of the determination target user.

  Since the user whose response time is always fast is likely to be a bot, the user determination unit 110 compares the user response time calculated by the user response time calculation unit 150 with a threshold value and compares the user response time with the threshold value. When the response time is small, it is determined that the determination target user is a bot. On the other hand, when the user response time is equal to or greater than the threshold value, it is determined that the determination target user is a non-bot.

  The alternation latency feature quantity calculation unit 160 calculates the alternation latency feature quantity of the determination target user as a regularity index. By the way, when a person talks between two parties, the alternation latency is synchronized, but since the posting of utterances by the bot is performed according to pre-programmed rules, there is a regularity that the alternation latency is not synchronized Therefore, it is possible to determine whether or not the determination target user is a bot using the alternation latency characteristic amount as the regularity index.

  As shown in FIG. 1, the replacement latency feature value calculation unit 160 includes a continuous response section extraction unit 161, an agreement determination unit 162, and a replacement latency feature value determination unit 163.

  The continuous response section extraction unit 161 continuously repeats posting of comments a predetermined number of times toward each other at near real-time intervals with the determination target user and one other user based on the speech information stored in the speech storage unit 120. The comment information of the response section is extracted from the comment storage unit 120.

  Specifically, the continuous response section extraction unit 161 firstly specifies the user ID of the determination target user and the response target original utterance included in the inter-speech relationship information stored in the utterance storage unit 120 and the posting. Based on the date and time information, a user who has posted a message that is in a response relationship with the determination target user's statement and whose posting interval with the determination target user's message is within a predetermined time is extracted as a continuous response user candidate. When speech information is acquired from the user determination unit 110, first, the user ID of the determination target user included in the speech information acquired from the user determination unit 110 is acquired.

  Next, the continuous response section extraction unit 161 extracts the utterance of one continuous response user candidate extracted based on the information specifying the original utterance to be responded included in the inter-speech relation information stored in the utterance storage unit 120. Then, the utterance information of the utterances having a response relationship is extracted from the utterances of the determination target user, and the utterance time series list arranged in time series is created based on the posting date information included in the extracted utterance information. The continuous response section extraction unit 161 creates a speech time series list for all continuous response user candidates.

  Then, the continuous response section extraction unit 161 identifies, as a continuous response section, a section in which a predetermined number of utterances with a posting interval within a predetermined time are included as a continuous response section from the created all utterance time series list. Extract speech information for the section.

  The consent determination unit 162 determines whether each utterance in the continuous response section agrees or disagrees with the original utterance based on the content information included in the utterance information extracted by the continuous response section extraction unit 161. The determination may be made including neutrality that is neither consent nor disagreement. The determination result is given to the comment information.

  The alternation latency feature quantity calculation unit 160 calculates the response interval of each utterance included in the continuous response section based on the posting date information included in the utterance information of the continuous response section extracted by the continuous response section extraction unit 161. Then, the calculated response interval is substituted into the following equation (2) to calculate the alternation latency characteristic amount.

  When the determination result of the consent determination unit 162 is given to the utterance information in the continuous response section, the alternation latency feature quantity calculation unit 160 responds to each utterance included in the continuous response section based on the determination result. Weight intervals. Then, the weighted response interval is substituted into the above-described equation (2) to calculate the alternate latency feature amount.

  Response time in the case of disagreement with the other party tends to be longer than response time in the case of consent (Non-patent document 4: Takashi Matsuyama, Hiroaki Kawashima, Takaaki Hirayama, “Information processing of sensitivity to time and time sense”) , IEICE Journal 92 (11), 952-954, 2009-11-01). Therefore, it is possible to calculate the alternation latency feature amount with higher accuracy by weighting the response time depending on whether the utterance is agreement or disagreement.

  When humans interact with each other, the alternation latency is synchronized. Therefore, when the alternation latency feature value converges to the threshold value “0”, the user determination unit 110 determines that the determination target user is not On the other hand, if it is determined to be a bot and does not converge to the threshold value “0”, the determination target user is determined to be a bot.

  The periodicity index calculation unit 170 includes an analysis posting date list acquisition unit 171 and a spectrum analysis unit 172.

  When the analysis posting date and time list acquisition unit 171 acquires the user ID of the determination target user from the user determination unit 110, the determination target user is based on the remark information stored in the remark storage unit 120 in association with the acquired user ID. Get a list of post date information as an analysis post date list. The analysis posting date list acquisition unit 171 may acquire all posting date information stored in the comment storage unit 120, or may specify only a period and acquire only posting date information within the period. Good.

  When the utterance information is acquired from the user determination unit 110, the analysis posting date / time list acquisition unit 171 receives the utterance information acquired from the user determination unit 110 instead of the posting date / time information of the determination target user acquired from the utterance storage unit 120. You may acquire the analysis posting date list from. The analysis posting date / time list acquisition unit 171 may acquire the analysis posting date / time list from the message storage unit 120 based on the user ID of the determination target user included in the message information acquired from the user determination unit 110.

  The analysis posting date / time list acquisition unit 171 analyzes and posts a list of posting date / time information of statements other than response messages for the determination target user based on the inter-speech relationship information included in the speech information stored in the speech storage unit 120. You may acquire as a date list. Thereby, the presence or absence of periodicity can be determined with higher accuracy. This is because a utterance of a response to another user is posted without depending on a preset rule, so that it becomes noise to determine the presence or absence of periodicity.

  The spectrum analysis unit 172 performs spectrum analysis on the analysis posting date list acquired by the analysis posting date list acquisition unit 171. Specifically, the spectrum analysis unit 172 regards the posting of the message as an output of a constant output impulse response signal, performs spectrum analysis, and acquires an amplitude spectrum as an analysis result. For the spectrum analysis, for example, Fourier transform or piccolo method is used. Then, the spectrum analysis unit 172 acquires the maximum amplitude value of the acquired amplitude spectrum as a regularity index.

  The user determination unit 110 determines a bot if the maximum amplitude value acquired by the spectrum analysis unit 172 is greater than the threshold value, and determines that it is a non-bot if the maximum amplitude value is equal to or less than the threshold value. Since the maximum amplitude value increases as the utterance is posted in a specific cycle, it can be determined whether or not the determination target user is a bot by comparing the maximum amplitude value with a threshold value.

  The threshold value storage unit 180 stores a preset regularity index threshold value. Specifically, when the inter-user correlation coefficient is used for the bot determination, the inter-user correlation coefficient threshold value is used. When the response time is used for the bot determination, the response time threshold value and the bot determination are used. When the alternate latency feature quantity is used, the threshold value of the alternate latency feature quantity “0” is stored. When the maximum amplitude value is used for bot determination, the threshold value of the maximum amplitude value is stored.

<Bot decision processing flow using timing correlation coefficient>
FIG. 2 is a diagram showing a bot determination processing flow using a timing correlation coefficient in the bot determination apparatus 100 according to the first embodiment of the present invention.

  First, in step S1, the user determination unit 110 determines whether the user ID of the determination target user or the remark information of the determination target user is input from the outside. If it is determined that it has been input (YES), the process proceeds to step S2. On the other hand, if it is determined that it has not been input (NO), the process ends.

  Next, in step S <b> 2, the nearest posting date / time list acquisition unit 141 acquires the posting date / time information of the determination target user. When the user ID of the determination target user is received from the user determination unit 110, the posting date / time information is acquired from the message storage unit 120 based on the received user ID. On the other hand, when the utterance information of the determination target user is acquired from the user determination unit 110, the posting date information may be acquired from the utterance storage unit 120 based on the user ID included in the utterance information. Post date information may be obtained from

  Next, in step S <b> 3, the nearest posting date / time list acquisition unit 141 acquires the user IDs of users other than the determination target user from the statement storage unit 120 based on the user ID of the determination target user.

  Next, in step S4, the nearest posting date / time list acquisition unit 141, based on the posting date / time information of the determination target user acquired in step S2 and the user ID of each other user acquired in step S3, the message storage unit Based on the utterance information stored in 120, the utterance storage unit includes a nearest posting date / time list in which each posting date / time information of the determination target user is associated with each posting date / time information of the other users closest to the posting date / time information. Obtain from 120.

  Next, in step S5, the individual correlation coefficient calculation unit 142 indicates the relevance of the posting timing between the determination target user and other users based on the nearest posting date / time list acquired in step S4. The individual correlation coefficient between the posting date / time information and the posting date / time information of each other user is calculated.

  Next, in step S6, the individual correlation coefficient calculation unit 142 determines whether or not individual correlation coefficients have been calculated for all other users included in the nearest posting date / time list acquired in step S4. If the individual correlation coefficients of all other users have been calculated, the process proceeds to step S7. On the other hand, if the individual correlation coefficients of all other users have not yet been calculated, the process returns to step S5. .

  Next, in step S7, the timing correlation coefficient determination unit 143 determines the individual correlation coefficient of another user who has a tendency to post a comment before the determination target user from the individual correlation coefficients calculated in step S5. Are extracted based on the nearest posting date / time list acquired in step S4, and the maximum correlation coefficient of the extracted individual correlation coefficients is determined as the timing correlation coefficient.

  Next, in step S <b> 8, the user determination unit 110 determines whether or not the timing correlation coefficient calculated in step S <b> 7 is greater than the threshold value stored in the threshold value storage unit 180. If it is determined that the timing correlation coefficient is greater than the threshold value (YES), the process proceeds to step S9. On the other hand, if it is determined that the timing correlation coefficient is equal to or less than the threshold value (NO), step S10 is performed. Proceed with the process.

  Next, in step S9, the user determination unit 110 determines that the determination target user is a bot, and outputs it.

  Next, in step S10, the user determination unit 110 determines that the determination target user is a non-bot and outputs it.

<Bot decision processing flow using user response time>
FIG. 3 is a diagram showing a bot determination processing flow using the user response time in the bot determination apparatus 100 according to the first embodiment of the present invention. Note that the same processes as those in the bot determination process flow using the timing correlation coefficient described with reference to FIG. 2 are denoted by the same reference numerals, and detailed description thereof is omitted.

  Next, in step S11, the response date / time list acquisition unit 151 acquires the response date / time list of the determination target user. When the user ID of the determination target user is received from the user determination unit 110, information indicating whether the response is included in the inter-sentence relation information stored in the statement storage unit 120 in association with the received user ID. Based on this, the response date / time list is acquired from the message storage unit 120. On the other hand, when the utterance information of the determination target user is acquired from the user determination unit 110, the response date and time list may be acquired from the utterance storage unit 120 in the same manner based on the user ID included in the utterance information. You may acquire a response date list from the utterance information acquired from the determination part 110. FIG.

  Next, in step S12, the original posting date / time list acquisition unit 152 acquires the original posting date / time list. When the user ID of the determination target user is received from the user determination unit 110, the response target original statement included in the inter-sentence relationship information stored in the statement storage unit 120 in association with the received user ID is specified. Based on the information, the original posting date list is acquired from the comment storage unit 120. On the other hand, when the utterance information of the determination target user is acquired from the user determination unit 110, the response date and time list may be acquired from the utterance storage unit 120 in the same manner based on the user ID included in the utterance information. The original posting date / time list may be acquired from the statement storage unit 120 based on the information specifying the original message to be responded included in the inter-speech relationship information included in the information.

  Next, in step S13, the user response time determination unit 153 determines whether the determination target user responds based on the corresponding posting date information in the response date list acquired in step S11 and the original posting date list acquired in step S12. The response time until posting a comment is calculated, and the average of the calculated response times is determined as the user response time of the determination target user.

  Next, in step S14, the user determination unit 110 determines whether or not the user response time calculated in step S13 is smaller than the threshold value stored in the threshold value storage unit 180. If it is determined that the user response time is smaller than the threshold value (YES), the process proceeds to step S9. On the other hand, if the user response time is determined to be equal to or greater than the threshold value (NO), the process proceeds to step S10. To proceed.

<Bot decision processing flow using alternate latency feature>
FIG. 4 is a diagram showing a bot determination processing flow using the alternate latency feature quantity in the bot determination apparatus 100 according to the first embodiment of the present invention. Note that the same processes as those in the bot determination process flow using the timing correlation coefficient described with reference to FIG. 2 are denoted by the same reference numerals, and detailed description thereof is omitted.

  Next, in step S <b> 21, the continuous response section extraction unit 161 selects a user who has posted a comment that is in a response relationship with the determination target user's comment and that has a posting interval within a predetermined time with the determination target user's comment. Extract as continuous response user candidates.

  Next, in step S22, the continuous response section extraction unit 161 extracts remark information that is in a response relationship from the remark information of the one continuous response user candidate extracted in step S21 and the remark information of the determination target user, Create a utterance time series list arranged in time series.

  Next, in step S23, the continuous response section extraction unit 161 identifies, as a continuous response section, a section in which a predetermined number of utterances with a posting interval within a predetermined time continue from the utterance time series list created in step S22. Extract utterance information in the response section.

  Next, in step S24, the continuous response section extraction unit 161 determines whether or not the processing in steps S22 and S23 has been performed for all the continuous corresponding user candidate users extracted in step S21. If the process is performed for all the continuous corresponding user candidate users (YES), the process proceeds to step S25. On the other hand, if the individual correlation coefficients of all other users are not yet calculated (NO), The process returns to step S22.

  Next, in step S25, the alternation latency feature quantity calculation unit 160 combines all the individual continuous response sections extracted in step S23 as one continuous response section, and the response interval of each comment included in the continuous response section. And the calculated response interval is substituted into the above-described equation (2) to determine the alternation latency feature amount.

  Next, in step S <b> 26, the user determination unit 110 determines whether or not the alternation latency feature value calculated in step S <b> 25 converges to the threshold value “0” stored in the threshold value storage unit 180. . If it is determined that the alternation latency feature value converges to the threshold value “0” (YES), the process proceeds to step S9. On the other hand, if it is determined not to converge to the threshold value “0” (NO) Advances the process to step S10.

<Bot decision processing flow using spectrum analysis>
FIG. 5 is a diagram showing a bot determination processing flow using spectrum analysis in the bot determination apparatus 100 according to the first embodiment of the present invention. Note that the same processes as those in the bot determination process flow using the timing correlation coefficient described with reference to FIG. 2 are denoted by the same reference numerals, and detailed description thereof is omitted.

  Next, in step S31, the analysis posting date list acquisition unit 171 acquires the posting date information of the determination target user as an analysis posting date list. When the user ID of the determination target user is received from the user determination unit 110, the analysis posting date list is acquired from the statement storage unit 120 based on the received user ID. On the other hand, when the utterance information of the determination target user is acquired from the user determination unit 110, the analysis posting date list may be acquired from the utterance storage unit 120 based on the user ID included in the utterance information, or the acquired utterance An analysis posting date list may be acquired from the information.

  Next, in step S32, the spectrum analysis unit 172 performs spectrum analysis on the analysis posting date list acquired in step S31.

  Next, in step S33, the maximum amplitude value of the amplitude spectrum acquired as the analysis result of step S32 is acquired as the regularity index.

  Next, in step S <b> 34, the user determination unit 110 determines whether or not the maximum amplitude value calculated in step S <b> 33 is greater than the threshold value stored in the threshold value storage unit 180. If it is determined that the maximum amplitude value is greater than the threshold value (YES), the process proceeds to step S9. On the other hand, if it is determined that the user response time is equal to or less than the threshold value (NO), the process proceeds to step S10. To proceed.

  As described above, according to the present embodiment, in the SNS, the user ID is not thrown away, and the bot posts a statement according to the rule, and the correlation and response of the posting timing with other users. Efficiency of whether or not the user who posted the utterance is a bot, without learning data, due to the regularity of posting timing such as time, the alternation latency of the utterance, and the periodicity of the posting interval within the user Can be judged well. As a result, in the SNS, bot remarks can be removed, and spam can be efficiently removed.

<Second Embodiment>
A second embodiment of the present invention will be described with reference to FIG. The bot determination device according to the present embodiment improves the determination accuracy by integrating the correlation coefficient between users, the response time, and the maximum amplitude value described in the first embodiment to perform the bot determination. Device. In addition, about the component which attaches | subjects the same code | symbol as 1st Embodiment, since it has the same function, the detailed description is abbreviate | omitted.

<Functional configuration of bot determination device>
FIG. 6 is a diagram illustrating a functional configuration of the bot determination device 200 according to the second embodiment of the present invention. The bot determination device 200 is a device that determines whether or not a user who has posted a comment is a bot in SNS (Social Networking Service), and includes a regularity index calculation unit 131, a message storage unit 120, and an integrated model. A learning unit 210, an integrated model storage unit 220, and an integrated model determination unit 230 are provided. Here, the bot in the present invention means an automatic speech system by a machine that speaks information (spam) such as advertisements.

  The regularity index calculation unit 131 includes a timing correlation coefficient calculation unit 140, a user response time calculation unit 150, and a periodicity index calculation unit 170.

  The integrated model learning unit 210 uses a plurality of types of regularity indices for each of a user determined to be a bot and a user determined to be a non-bot as teacher data, using a predetermined learning algorithm. Learn an integrated model that weights and integrates each sex index. The plurality of types of regularity indices include at least two of a timing correlation coefficient, a user response time, and a periodicity index.

  The integrated model learning unit 210 uses, for example, AfaBoost (Non-Patent Document 5: Yoav Freund, Robert E. Shapire. an Application to Boost ", 1995).

  The integrated model storage unit 220 stores the integrated model learned by the integrated model learning unit 210. An integrated model may be stored for each combination of a plurality of types of regularity indices.

  The integrated model determination unit 230 determines whether the determination target user is a bot based on a plurality of types of regularity indexes calculated by the regularity index calculation unit 131 for the determination target user and the integrated model stored in the integrated model storage unit 220. It is determined whether or not there is.

  As described above, according to the present embodiment, whether or not the user who is posting a utterance is a bot using an integrated model based on regularity of posting timing learned using teacher data. Can be determined efficiently. As a result, in the SNS, bot remarks can be removed, and spam can be efficiently removed.

  Note that the bot determination device of the present invention can be realized by recording the processing of the bot determination device on a recording medium readable by a computer system, causing the bot determination device recorded on the recording medium to read and execute the processing. . The computer system here includes an OS and hardware such as peripheral devices.

  Further, the “computer system” includes a homepage providing environment (or display environment) if a WWW (World Wide Web) system is used. The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line.

  The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, and what is called a difference file (difference program) may be sufficient.

  The embodiments of the present invention have been described in detail with reference to the drawings. However, the specific configuration is not limited to the embodiments, and includes designs and the like that do not depart from the gist of the present invention.

100, 200 Bot determination device 110 User determination unit 120 Statement storage unit 130, 131 Regularity index calculation unit 140 Timing correlation coefficient calculation unit 141 Nearest neighbor posting date / time list acquisition unit 142 Individual correlation coefficient calculation unit 143 Timing correlation coefficient determination Unit 150 User response time calculation unit 151 Response date and time list acquisition unit 152 Original posting date and time list acquisition unit 153 User response time determination unit 160 Replacement latency characteristic calculation unit 161 Continuous response section extraction unit 162 Consent determination unit 163 Replacement latency characteristic determination Unit 170 periodicity index calculation unit 171 analysis posting date list acquisition unit 172 spectrum analysis unit 180 threshold calculation unit 210 integrated model learning unit 220 integrated model storage unit 230 integrated model determination unit

Claims (11)

In a SNS (Social Networking Service) that accepts and publishes posted comments, it is a bot determination device that determines whether or not a user posting a comment is a bot.
Message storage means for storing message information including at least posted user information and posted date information of the message posted to the SNS;
For determining the target user, based on the speech information stored in said speech storage means, and regularity index calculating means for calculating a regularity index indicating the regularity of the definitive post timing between the other users and the determination target user ,
User determination means for determining whether or not the determination target user is the bot by comparing the regularity index calculated by the regularity index calculation means and a preset threshold;
A bot determination device comprising: The regularity index calculating means is
The nearest posting date / time in which each posting date / time information of the determination target user is associated with the posting date / time information of each of the other users closest to the posting date / time information based on the statement information stored in the statement storage unit. The nearest posting date / time list acquisition means for acquiring a list,
Based on the nearest posting date list acquired by the nearest posting date list acquisition unit, an individual correlation coefficient that calculates an individual correlation coefficient indicating a posting timing relationship between each of the other users and the determination target user A calculation means;
Among the individual correlation coefficients calculated by the individual correlation coefficient calculation means, the individual correlation coefficient of another user having a tendency to post a statement before the determination target user is obtained by the nearest posting date list acquisition means. A timing correlation coefficient determining means for extracting based on the acquired nearest posting date list and determining the maximum correlation coefficient of the extracted individual correlation coefficient as a timing correlation coefficient;
With
The said user determination means determines whether the said determination target user is the said bot from the magnitude relationship between the said timing correlation coefficient and the said preset threshold value. The bot determination device described in 1. The speech information of the speech storage means further includes inter-speech relationship information indicating a relationship with other utterances posted to the SNS,
The regularity index calculating means is
A response date and time list acquisition unit that acquires a response date and time list that is a list of posting date and time information of the response speech of the determination target user based on the message information stored in the message storage unit;
Based on the utterance information stored in the utterance storage means, original posting date / time list acquisition means for acquiring an original posting date / time list that is a list of posting date / time information of the original utterance of the response utterance of the determination target user;
Based on the response date list acquired by the response date list acquisition means and the original posting date list acquired by the original posting date list acquisition means, the response time until the determination target user posts the response message is calculated. And a user response time determining means for determining an average of the calculated response times as a user response time of the determination target user;
With
The user determination means determines whether or not the determination target user is the bot from the magnitude relationship between the user response time acquired by the user response time determination means and the preset threshold value. The bot determination device according to claim 1, wherein the bot determination device according to claim 1. The regularity index calculating means is
Based on the utterance information stored in the utterance storage means, the utterance information of the continuous response section in which utterances are continuously exchanged at a predetermined interval or less between the determination target user and the one other user. Continuous response interval extracting means for extracting;
Based on the utterance information of the continuous response section extracted by the continuous response section extraction means, the response interval of each utterance included in the continuous response section is calculated, and the alternate latency feature quantity is calculated from the difference of the calculated response intervals. Alternate latency feature quantity calculating means for calculating;
With
The said user determination means determines whether the said determination object user is the said bot from the substitution latency feature-value calculated by the said substitution latency feature-value calculation means. Bot judgment device. The message information of the message storage means further includes content information of each message posted to the SNS,
The regularity index calculating means is
Consent determination means for determining whether or not each utterance included in the continuous response section agrees with the previous utterance based on the content information included in the utterance information of the continuous response section extracted by the continuous response section extraction means. With
5. The bot determination according to claim 4, wherein the alternate latency characteristic amount calculation unit weights the response interval according to a determination result of the consent determination unit of each utterance included in the continuous response section. apparatus. The regularity index calculating means is
Analysis posting date / time list acquisition means for acquiring a list of posting date / time information of the determination target user based on the statement information stored in the comment storage means;
A spectrum analysis unit that performs spectrum analysis on a list of post date information of the determination target user acquired by the analysis post date list acquisition unit, and acquires a maximum amplitude value of an amplitude spectrum as a periodicity index;
With
The user determination means determines whether the determination target user is the bot from the magnitude relationship between the periodicity index acquired by the spectrum analysis means and the preset threshold value. The bot determination device according to any one of claims 1 to 5, characterized in that:   The spectrum analyzing unit analyzes a list of post date information of the determination target user acquired by the analysis post date list acquiring unit using a discrete Fourier transform method or a piccolo method, and calculates the periodicity index. The bot determination device according to claim 6, wherein The speech storage means further includes inter-speech relationship information indicating a relationship with other utterances posted to the SNS in the utterance information,
The analysis posting date / time list acquisition unit acquires a list of posting date / time information of a statement other than a response message for the determination target user based on the inter-speech relationship information of the message information stored in the message storage unit. The bot determination device according to claim 6 or 7, wherein the bot determination device according to claim 6 or 7 is characterized. In a SNS (Social Networking Service) that accepts and publishes posted comments, it is a bot determination device that determines whether or not a user posting a comment is a bot.
Message storage means for storing message information including at least posted user information and posted date information of the message posted to the SNS;
For determining the target user, based on the speech information stored in said speech storage means, regularities of calculating a plurality of types of regularity index indicating the regularity of the definitive post timing between between other users and determining target user Index calculation means;
Using each of the plurality of types of regularity indices of the user determined to be a bot and each of the users determined to be not the bot as teacher data, a predetermined learning algorithm is used to calculate each of the plurality of types of regularity indices. An integrated model learning means for learning an integrated model weighted and integrated;
It is determined whether or not the determination target user is a bot based on the plurality of types of regularity indexes calculated by the regularity index calculation means and the integrated model learned by the integrated model learning means for the determination target user. An integrated model determination means to
A bot determination device comprising: In a SNS (Social Networking Service) that accepts and publishes posted comments, a bot determination method in a bot determination device that determines whether or not a user posting a comment is a bot,
The bot determination device includes message storage means for storing message information including at least posting user information and posting date information of messages posted to the SNS, regularity index calculation means, and user determination means,
A first step in which the user determination means receives an input of a determination target user who determines whether or not the bot;
The regularity index calculation means for the determination target user, based on the speech information stored in said speech storage means, regularity index indicating the regularity of the definitive post timing between the determination target user and other users A second step of calculating
The user determination means determines whether or not the determination target user is the bot by comparing the regularity index calculated in the second step with a preset threshold value. 3 steps,
A bot determination method including In order to cause a computer to execute a bot determination method in a bot determination device that determines whether or not a user posting a utterance is a bot in an SNS (Social Networking Service) that accepts and publishes a posted utterance A program,
The bot determination device includes message storage means for storing message information including at least posting user information and posting date information of messages posted to the SNS, regularity index calculation means, and user determination means,
A first step in which the user determination means receives an input of a determination target user who determines whether or not the bot;
The regularity index calculation means for the determination target user, based on the speech information stored in said speech storage means, regularity index indicating the regularity of the definitive post timing between the determination target user and other users A second step of calculating
The user determination means determines whether or not the determination target user is the bot by comparing the regularity index calculated in the second step with a preset threshold value. 3 steps,
A program that causes a computer to execute.

Images