Information processing method, information processing device, information processing program

Abstract

In an information processing method, language information and non-language information of a plurality of users constituting a group are collected, a power level indicating a magnitude of power of each of the plurality of users within the group is evaluated based on at least one of the language information and the non-language information, relationship information indicating a relationship between the plurality of users is acquired based on the power level of each of the plurality of users, a detection sensitivity for detecting inappropriate behavior from communication between the plurality of users is set based on the relationship information, it is determined whether inappropriate behavior is included in the communication based on the detection sensitivity, and information indicating that inappropriate behavior is detected is outputted when inappropriate behavior is included in the communication.

Description

Technical Field

[0001] This disclosure relates to technology for detecting inappropriate speech or behavior. Background Technology

[0002] Patent Document 1 discloses a management system for inappropriate submissions as follows: In order to exclude inappropriate submissions on the Internet, when a request for investigation is received from a document distribution system regarding a notification that a submission is inappropriate, an investigation is conducted using a prohibited information statistics database to determine the reliability of the notification, and based on the investigation, a notification of rejection, permission, or reservation of the notification is sent to the document distribution system.

[0003] However, the technology described in Patent Document 1 was developed with the aim of excluding inappropriate submissions on the Internet and did not involve any research on how to properly detect inappropriate speech or behavior within a group.

[0004] Prior art literature

[0005] Patent documents

[0006] Patent Document 1: Japanese Patent Application Publication No. 2013-178706 Summary of the Invention

[0007] This disclosure was made to address such a problem, with the aim of providing a technique that can properly detect inappropriate speech and behavior within a group.

[0008] One aspect of this disclosure is an information processing method in a computer. In this method, verbal and nonverbal information of multiple users constituting a group is collected. Based on at least one of the verbal and nonverbal information, a power level representing the relative strength of each user within the group is evaluated. Based on the power level of each user, relational information representing the relationships between users is obtained. Based on the relational information, a detection sensitivity is set for detecting inappropriate speech or behavior in interactions between users. The detection sensitivity is used as a benchmark to determine whether the inappropriate speech or behavior is present in the interaction. If the inappropriate speech or behavior is present in the interaction, information indicating that the inappropriate speech or behavior has been detected is output.

[0009] According to this disclosure, it is possible to properly detect inappropriate words and actions. Attached Figure Description

[0010] Figure 1 This is a block diagram illustrating an example of the structure of the information processing system according to Embodiment 1.

[0011] Figure 2 This is a diagram illustrating an example of level correspondence information.

[0012] Figure 3 This is a diagram illustrating an example of a data structure that shows information about inappropriate speech or behavior.

[0013] Figure 4 This is a flowchart illustrating the overall processing of the information processing system according to Embodiment 1.

[0014] Figure 5 This is a block diagram illustrating an example of the structure of the information processing system according to Embodiment 2.

[0015] Figure 6 This is a flowchart illustrating the overall processing of the information processing system according to Embodiment 2.

[0016] Figure 7 This is a block diagram illustrating an example of the structure of the information processing system according to Embodiment 3.

[0017] Figure 8 This is a flowchart illustrating the overall processing of the information processing system according to Embodiment 3.

[0018] Figure 9 This is a block diagram illustrating an example of the structure of the information processing system according to Embodiment 4.

[0019] Figure 10 This is a flowchart illustrating the overall processing of the information processing system according to Embodiment 4. Detailed Implementation

[0020] (The process of realizing this disclosure)

[0021] When one member of a group makes inappropriate remarks or actions towards another member, it is possible for relationships among the members to deteriorate, and the productivity of group activities to decrease. To prevent this, it is advisable to monitor communication within the group and detect inappropriate remarks or actions.

[0022] However, whether a particular word or action is considered inappropriate is influenced by the relationships among the members of the group. For example, whether a word or action is perceived as humorous or inappropriate can vary depending on the relationships among the members. Indiscriminately testing all words or actions that raise concerns about inappropriateness without considering this factor could hinder free communication within the group, resulting in decreased productivity of group activities.

[0023] The present invention was made in view of the above-mentioned problems, and its object is to provide a technique capable of properly detecting inappropriate speech and behavior in a group.

[0024] To address the above issues, the following technology is disclosed.

[0025] (1) An information processing method in one aspect of the present disclosure is an information processing method in a computer. In the information processing method, language information and non-language information of multiple users constituting a group are collected. Based on at least one of the language information and the non-language information, a power level representing the strength of each of the multiple users within the group is evaluated. Based on the power level of each of the multiple users, relational information representing the relationship between users among the multiple users is obtained. Based on the relational information, a detection sensitivity for detecting inappropriate speech or behavior in communication between users among the multiple users is set. The detection sensitivity is used as a benchmark to determine whether the inappropriate speech or behavior is contained in the communication. If the inappropriate speech or behavior is contained in the communication, information for informing that the inappropriate speech or behavior has been detected is output.

[0026] According to this information processing method, the detection sensitivity for inappropriate speech and behavior is set based on relational information. That is, the detection sensitivity is adjusted according to the relationships between users. Furthermore, this detection sensitivity is used as a benchmark to determine whether inappropriate speech and behavior is included in the communication. In this way, inappropriate speech and behavior is detected based on the relationships between users. Therefore, it is possible to appropriately detect inappropriate speech and behavior within a group.

[0027] (2) In the information processing method described in (1) above, it is also possible to collect attribute information representing the attributes of the multiple users in the collection of non-verbal information, obtain level correspondence information representing the correspondence between the attribute and the level of the power in the evaluation of the power level, and evaluate the level of the power of the multiple users based on the attribute information and the level correspondence information.

[0028] According to this structure, attribute information representing user attributes is used to evaluate user power levels, thus enabling more accurate evaluation of power levels.

[0029] (3) In the information processing method described in (1) or (2) above, it is also possible to collect the speaking volume of each of the multiple users in the collection of language information, and to evaluate the power level of each of the multiple users based on the speaking volume in the evaluation of power level.

[0030] According to this structure, the amount of a user's speech is used to evaluate the user's power level, thus enabling a more accurate evaluation of the power level.

[0031] (4) In any of the information processing methods described in (1) to (3) above, the detection sensitivity may be set as follows: when the first user communicates with the second user whose power level is lower than that of the first user, a level difference representing the difference in power levels between the first user and the second user is calculated, and the detection sensitivity when the first user communicates with the second user is set high based on the magnitude of the level difference.

[0032] According to this structure, the detection sensitivity can be set to high when a user with a lower power level communicates with a user with a higher power level. That is, when a user with a higher power level communicates with a user with a lower power level, the greater the difference in their power levels, the higher the detection sensitivity. Thus, the behavior of the user with the higher power level is more likely to be detected as inappropriate, thereby prompting the user with the higher power level to pay attention to their own behavior when communicating with a user with a lower power level.

[0033] (5) In the information processing method described in (4) above, the detection sensitivity may be set low when the second user communicates with the first user based on the size of the grade difference.

[0034] According to this structure, the detection sensitivity when a user with lower power level communicates with a user with higher power level can be set to a lower value based on the size of the power difference. That is, when a user with lower power level communicates with a user with higher power level, the greater the difference in their power levels, the lower the detection sensitivity. In this way, the behavior of the user with lower power level is less likely to be detected as inappropriate, thus promoting positive communication between users with lower power level and users with higher power level.

[0035] (6) In any of the information processing methods described in (1) to (5) above, it may also be possible to further calculate the amount of communication between users among the plurality of users based on at least one of the language information and the non-language information, and adjust the detection sensitivity according to the amount of communication in the setting of the detection sensitivity.

[0036] According to this structure, since the detection sensitivity is adjusted based on the amount of communication, it is possible to set the detection sensitivity to more accurately reflect interpersonal relationships within the group. As a result, inappropriate words and actions within the group can be detected more appropriately.

[0037] (7) The information processing apparatus according to another aspect of this disclosure includes: a collection unit for collecting language information and non-language information of multiple users constituting a group; a ranking evaluation unit for evaluating a power ranking that represents the strength of each of the multiple users within the group based on at least one of the language information and the non-language information; an acquisition unit for acquiring relational information representing the relationships between the multiple users based on the power ranking of each of the multiple users; a sensitivity setting unit for setting a detection sensitivity for detecting inappropriate speech or behavior in communication between the users based on the relational information; a determination unit for determining whether the inappropriate speech or behavior is included in the communication based on the detection sensitivity; and an output unit for outputting information indicating that the inappropriate speech or behavior has been detected if the communication contains the inappropriate speech or behavior.

[0038] Based on this structure, an information processing apparatus can be provided that can achieve the same effect as the information processing method described above.

[0039] (8) In another aspect of this disclosure, the information processing program enables a computer to function as an information processing device, the information processing program causing the computer to perform the following processing: collecting verbal and nonverbal information of multiple users constituting a group; evaluating a power level representing the strength of each of the multiple users within the group based on at least one of the verbal and nonverbal information; acquiring relational information representing the relationships between the users based on the power levels of each of the multiple users; setting a detection sensitivity for detecting inappropriate speech or behavior in communication between the users based on the relational information; determining whether the inappropriate speech or behavior is included in the communication based on the detection sensitivity; and outputting information indicating that the inappropriate speech or behavior has been detected if the inappropriate speech or behavior is included in the communication.

[0040] Based on this structure, an information processing program can be provided that achieves the same effect as the information processing method described above.

[0041] This disclosure can also be implemented as an information processing system that operates through such an information processing program. Furthermore, it goes without saying that such a computer program can be distributed via computer-readable non-transitory recording media such as CD-ROMs or communication networks such as the Internet.

[0042] Furthermore, the embodiments described below are all specific examples of this disclosure. The numerical values, shapes, constituent elements, steps, and order of steps shown in the following embodiments are examples and are not intended to limit this disclosure. In addition, any constituent elements in the following embodiments that are not described in the independent claims representing the highest-level concept are described as arbitrary constituent elements. Moreover, the contents of each embodiment can be combined in all embodiments.

[0043] (Implementation Method 1)

[0044] Figure 1 This is a diagram illustrating an example of an information processing system 100 according to Embodiment 1. The information processing system 100 is a system for appropriately detecting inappropriate speech or behavior within a group. Furthermore, a group refers to a community composed of multiple members, such as a workplace, family, local community, sports club, etc. In addition, a community composed of service providers and service recipients, such as a hospital or large supermarket, is also considered a group. Furthermore, inappropriate speech or behavior refers to statements or actions that cause distress to at least one of the members' mental or physical well-being.

[0045] The following describes an example of the application of the information processing system 100 to company 1 (an example of a group). In the following example, the personnel constituting company 1 will be referred to as users.

[0046] like Figure 1 As shown, the information processing system 100 according to this embodiment includes multiple user terminals 2, multiple sensors 3, and a server 10 (an example of an information processing device). The multiple user terminals 2, multiple sensors 3, and server 10 are interconnected and can communicate with each other via a network NW. The network NW is, for example, a public communication line such as the Internet. However, the network NW can also be a local area network (LAN).

[0047] Multiple user terminals 2 are multiple information processing devices operated by multiple users, such as multiple desktop personal computers, laptop personal computers, or mobile terminals such as smartphones and tablets. Each user can send and receive messages using online chat tools such as email by operating user terminals 2. Furthermore, each user can participate in online meetings by operating user terminals 2.

[0048] Multiple sensors 3 each have one or more microphone devices and one or more camera devices. The microphone devices record sounds generated within company 1 as audio data. For example, the microphone devices record users' voices as audio data. In addition, the microphone devices capture various sounds such as lip-smacking sounds and table-tapping sounds. The camera devices record users' actions within company 1 in the form of image data or animation data. This data is transmitted to server 10 via network NW.

[0049] Multiple sensors 3 can be installed, for example, on the walls or ceiling of company 1. Alternatively, multiple sensors 3 can also be mounted on the aforementioned multiple user terminals 2. That is, if the user terminals 2 are equipped with microphones and cameras, they can also be used as multiple sensors 3.

[0050] Server 10, for example, is a cloud server consisting of one or more computers. Figure 1 As shown, server 10 has communication line 20, memory 30 and processor 40.

[0051] Communication line 20 is a communication interface circuit corresponding to communication methods using networks such as Ethernet (registered trademark), connecting server 10 to network NW.

[0052] The memory 30 is composed of non-volatile, rewritable storage devices such as solid-state drives and hard disk drives. Figure 1 As shown, the memory 30 in this embodiment has an attribute information storage unit 31, a level correspondence information storage unit 32, and an inappropriate speech and behavior information storage unit 33.

[0053] The attribute information storage unit 31 stores attribute information representing the attributes of multiple users. Attribute information includes, for example, a user's age, job title, duration of a user's affiliation with a group, and the blood relationship between a user and other users.

[0054] The level correspondence information storage unit 32 stores level correspondence information D1, which represents the correspondence between a user's attributes and their power level. Power levels will be discussed later. Figure 2 This is a diagram illustrating an example of level-corresponding information D1. For example... Figure 2As shown, in the level correspondence information D1 of this embodiment, the power level value assigned to a position such as a trainee is 2, the power level value assigned to a position such as a general manager is 6, and the power level value assigned to a position such as a director is 8. Furthermore, in level correspondence information D1, the power level value assigned to a position such as an assistant manager is 4, the power level value assigned to a position such as a manager is 5, the power level value assigned to a position such as an executive director is 9, and the power level value assigned to a position such as a chairman is 10. In addition, level correspondence information D1 may also store various attribute-power level correspondences.

[0055] The inappropriate speech and behavior information storage unit 33 stores inappropriate speech and behavior information D2, which represents examples of speech and behavior that can be equivalent to inappropriate speech and behavior, and the correspondence between the inappropriate speech and behavior level and the detection sensitivity level. Figure 3 This is a diagram illustrating an example of the data structure for inappropriate speech and behavior information D2. (See diagram for example.) Figure 3 As shown, the inappropriate speech and behavior information D2 has: a first column CL1 that registers examples of speech and behavior that can be equivalent to inappropriate speech and behavior, a second column CL2 that registers the level of inappropriate speech and behavior, and a third column CL3 that registers the level of detection sensitivity.

[0056] In column 1 (CL1), various statements and actions are registered as examples of speech or behavior that can be equated with inappropriate speech or behavior. Specifically, such as... Figure 3 As shown, examples of speech or behavior that can be equivalent to inappropriate speech or behavior include statements such as "jerk," violent behavior, statements such as "go back," and smacking lips.

[0057] Column 2, CL2, lists the inappropriate speech / behavior levels corresponding to examples of speech / behavior that could be considered equivalent to inappropriate speech / behavior. The inappropriate speech / behavior level is an indicator of the likelihood that a particular speech / behavior is equivalent to another inappropriate speech / behavior. In this example, the inappropriate speech / behavior levels are divided into 10 stages, from level 1 to level 10. The numerical value of the inappropriate speech / behavior level corresponds to the likelihood of it being equivalent to inappropriate speech / behavior. For example, as... Figure 3 As shown, the inappropriateness level of speech and behavior, such as "Please cheer up," which is less likely to be considered inappropriate, is 1. On the other hand, the inappropriateness level of speech and behavior, such as "You jerk," which is more likely to be considered inappropriate, is 10. Additionally, Figure 3The structures of the first column CL1 and the second column CL2 shown are merely examples, and the degree of the inappropriate behavior level set for which behavior can be changed as appropriate. Furthermore, the inappropriate behavior level may be determined based on the harmfulness score output by the AI model disclosed in Non-Patent Document 1 "Kobayashi Koga, 7 others, [online], 'Proposal and Evaluation of Japanese Harmful Expression Schema', The Association for Natural Language Processing 2023, [retrieved on March 18, Reiwa 6], Internet <URL: https: / / www.anlp.jp / proceedings / annual_meeting / 2023 / pdf_dir / D4-1.pdf>".

[0058] In the third column CL3, detection sensitivity levels corresponding to the inappropriate behavior levels are registered. The detection sensitivity level is an index indicating the level of detection sensitivity for detecting inappropriate behavior. In this example, the detection sensitivity level is divided into 10 stages from level 1 to level 10. The sensitivity setting unit 44 described later sets the value of the detection sensitivity level within the range of 1 to 10. The value set by the sensitivity setting unit 44 is used as a threshold for the detection unit 45 described later to detect inappropriate behavior. For example, the sensitivity setting unit 44 sets the value of the detection sensitivity level to 3. In this case, the detection unit 45 described later detects behaviors with an inappropriate behavior level value of 3 or higher as inappropriate behaviors. On the other hand, the detection unit 45 does not detect behaviors with an inappropriate behavior level value less than 3 as inappropriate behaviors. With reference to Figure 3 for description, when the value of the detection sensitivity level is set to 3, the detection unit 45 detects behaviors with an inappropriate behavior level value of 3 or higher, such as the utterance "bastard", the utterance "freeloader", and the act of smacking one's lips, as inappropriate behaviors. On the other hand, the detection unit 45 does not detect behaviors with an inappropriate behavior level value set to 1 or 2, such as "Is this okay?" and "Pull yourself together", as inappropriate behaviors. In this way, the range of behaviors detected as inappropriate can be changed by adjusting the value of the detection sensitivity level.

[0059] Overall, the detection sensitivity can be increased by setting the value of the detection sensitivity level to a smaller value. On the other hand, the detection sensitivity can be decreased by setting the value of the detection sensitivity level to a larger value.

[0060] The processor 40 is, for example, a CPU. The processor 40 has an acquisition unit 41, a grade evaluation unit 42, a relational information acquisition unit 43 (an example of an acquisition unit), a sensitivity setting unit 44, a detection unit 45 (an example of a determination unit), and an output unit 46. The acquisition unit 41 to the output unit 46 can be implemented by the processor 40 executing a given program stored in the memory 30, or they can be configured using dedicated hardware circuitry.

[0061] The acquisition unit 41 acquires verbal and non-verbal information. Verbal information refers to information used in communication between users. Examples of verbal information include spoken words and text contained in messages sent and received by users. Non-verbal information refers to information related to communication between users that is not verbal. Examples of non-verbal information include facial expressions, eye contact, posture, gestures, body language, and attribute information. In addition, tone of voice and volume can also be examples of non-verbal information. The acquisition unit 41 acquires verbal and non-verbal information from the aforementioned multiple sensors 3.

[0062] The rating department 42 evaluates the power levels of multiple users within their respective groups based on at least one of linguistic and non-linguistic information. Furthermore, power within a group can refer to, for example, authority, influence, voice, and political power.

[0063] In this embodiment, the rating evaluation unit 42 obtains rating correspondence information D1, which represents the correspondence between a user's attributes and the level of their power, from the rating correspondence information storage unit 32, and evaluates the level of power of multiple users based on the attribute information and the rating correspondence information D1.

[0064] The relational information acquisition unit 43 acquires power balance information (an example of relational information) representing the power balance among multiple users (an example of relational information) based on the respective power levels of each user. Power balance refers to the distribution of power within a group. Power balance becomes unbalanced when a particular individual or organization has greater power than other members. If this imbalance persists, it may lead to problems such as power mismatch, feelings of unfairness, and biased decision-making.

[0065] The sensitivity setting unit 44 sets the detection sensitivity based on power balance information, which is used to detect inappropriate speech or behavior in interactions between multiple users. Specifically, the sensitivity setting unit 44 sets the detection sensitivity by setting a value for the detection sensitivity level. Details will be described later.

[0066] Furthermore, when the first user communicates with a second user whose power level is lower than the first user, the sensitivity setting unit 44 calculates the power level difference between the first user and the second user. Moreover, the sensitivity setting unit 44 sets the detection sensitivity high when the first user communicates with the second user, based on the magnitude of the power level difference.

[0067] Furthermore, setting the detection sensitivity high means setting the detection sensitivity level value to a value lower than the standard value for the detection sensitivity level. In this embodiment, the standard value for the detection sensitivity level is set to 5. Therefore, setting the detection sensitivity high means setting the detection sensitivity level value to a value less than 5. However, the standard value for the detection sensitivity level is not limited to 5 and can be appropriately changed.

[0068] Furthermore, the sensitivity setting unit 44 sets the detection sensitivity to low when the second user communicates with the first user, based on the magnitude of the level difference.

[0069] In addition, setting the detection sensitivity to low means setting the detection sensitivity level value to a value greater than the standard value for the detection sensitivity level.

[0070] The detection unit 45 uses the detection sensitivity set by the sensitivity setting unit 44 as a benchmark to determine whether there are inappropriate words or actions in the communication between users.

[0071] When inappropriate speech or behavior is detected in the communication, the output unit 46 outputs an alarm message. The alarm message is used to inform the user that inappropriate speech or behavior has been detected. For example, the output unit 46 outputs a message such as "Inappropriate speech or behavior" as an alarm message to the user terminal 2. Alternatively, the output unit 46 may output an alarm message by causing a wearable device worn by the user (e.g., a smartwatch) to vibrate. In addition, the output unit 46 may output an image of an indicator simulating a pointer and numerical scale as an alarm message to the user terminal 2. For example, if the inappropriate speech or behavior level is set to 5 for the detected inappropriate speech or behavior, the output unit 46 may output an image of an indicator showing the pointer pointing to 5 as an alarm message to the user terminal 2.

[0072] In this embodiment, the output unit 46 outputs the aforementioned alarm information only to the user terminal 2 of the user who made the inappropriate remarks. However, the output unit 46 may also output the alarm information to all user terminals 2 of multiple users. Alternatively, if there is a manager who is equivalent to the user who made the inappropriate remarks, the output unit 46 may output the alarm information only to the manager's user terminal 2.

[0073] Regarding the processing of the information processing system 100 configured as described above, refer to... Figure 4 Please provide an explanation. Figure 4 This is a flowchart illustrating the overall image of the processing of the information processing system 100 according to Embodiment 1.

[0074] In step S1, the acquisition unit 41 acquires the language and non-language information of each user. That is, the acquisition unit 41 acquires the speech and actions of each user. For example, the acquisition unit 41 acquires audio data containing the speech "Good morning" and text messages containing "Thank you for your hard work" as language information. Furthermore, for example, the acquisition unit 41 acquires the gaze and facial orientation of each user as non-language information. The gaze and facial orientation of each user are information represented by a 3D vector. Furthermore, techniques for detecting each user's gaze and facial orientation are disclosed, for example, in Non-Patent Document 2, "Matsumoto Yoshio, et al., 'Development of a Real-Time Face and Gaze Measurement System and Its Application in Smart Interfaces', Journal of the Society for Information Processing, Computer Vision & Image Media, October 2006, vol.47, no. SIG15 (CVIM16), pp10-21." Additionally, for example, the acquisition unit 41 acquires and records image data or animation data showing behaviors such as "looking down" as non-verbal information.

[0075] Furthermore, in step S1, the acquisition unit 41 acquires the attribute information of each user as non-verbal information. For example, the acquisition unit 41 acquires attribute information indicating that user A's job title is trainee, user B and user C's job titles are general manager, and user D's job title is director. As described above, this attribute information is stored in the attribute information storage unit 31 of the memory 30.

[0076] In step S2, the acquisition unit 41 determines the user. That is, the acquisition unit 41 determines the subject and object of each speech or action acquired in step S1. In other words, the acquisition unit 41 determines from which user each speech or action acquired in step S1 was directed at which user. The acquisition unit 41 performs the above determination process, for example, using non-verbal information such as the user's gaze and facial orientation, as well as linguistic information such as the content of each user's speech, acquired in step S1. Furthermore, if text is acquired as linguistic information in step S1 and the text contains reference information, the acquisition unit 41 can also determine the user by referring to the reference information. Reference information refers to information containing an identifier for identifying the user, such as a string consisting of "@ + username".

[0077] Hereinafter, the user who is the subject of a given speech or action will be called the subject user, and the user who is the object of a given speech or action will be called the object user.

[0078] In step S3, the detection unit 45 detects behaviors that could be considered inappropriate from the behaviors of each user collected in step S1. If the detection unit 45 detects behaviors that could be considered inappropriate (Yes in step S3), the process proceeds to step S4. If the detection unit 45 does not detect behaviors that could be considered inappropriate (No in step S3), the process returns to step S1.

[0079] The processing in step S3 will be explained in detail. The detection unit 45 in this embodiment compares the statements and actions of each user collected in step S1 with... Figure 3 The first column CL1 shown detects speech or behavior that could be considered inappropriate. For example, in step S1, consider a case where sound data containing the phrase "enough is enough" is collected as language information. In this case, firstly, the detection unit 45 uses existing speech recognition technology to textualize the phrase "enough is enough." Furthermore, if the detection unit 45 registers a word in the first column CL1 that exceeds the baseline similarity to the textualized speech, it determines this as "yes" in step S3. Here, as... Figure 3 As shown, the phrase "enough is enough" is registered in column 1, CL1. Therefore, in this example, the detection unit 45 determines that inappropriate speech or behavior has been detected (Yes in step S3).

[0080] Back Figure 4 In step S4, the detection unit 45 stores the value of the inappropriate speech level set for the speech detected in step S3 as the value of the comparison object level in a given storage area of ​​the memory 30. For example, imagine a case where speech such as "That's enough" is detected in step S3. Here, as Figure 3 As shown in columns 1 (CL1) and 2 (CL2), the behavior of saying "enough is enough" is considered inappropriate and its level is set to 4. Therefore, in this case, the detection unit 45 stores the comparison object level as 4.

[0081] In step S5, the rating evaluation unit 42 evaluates the power level of the user who issued the inappropriate statement or behavior. That is, the rating evaluation unit 42 evaluates the power level of both the user who issued the statement or behavior that is equivalent to the inappropriate statement or behavior and the user who was the object of the statement or behavior.

[0082] The following example illustrates the processing in step S5 by evaluating the power levels of the subject and object users of the phrase "Let's stop here." In step S2, it is assumed that user A is the subject user of the phrase "Let's stop here" and user B is the object user. In this case, the power level evaluation unit 42 evaluates the power levels of user A and user B according to the following process: First, the power level evaluation unit 42 refers to the attribute information collected by the acquisition unit 41 in step S1 and determines that user A's position is a trainee and user B's position is a general manager. Next, the power level evaluation unit 42 retrieves the power level information from the power level correspondence storage unit 32 in the memory 30. Figure 2 The indicated level corresponds to information D1. Furthermore, referring to information D1, the Level Evaluation Department 42 determined that the power level assigned to the trainee was "2" and the power level assigned to the general manager was "6". Based on this information, the Level Evaluation Department 42 rated user A (trainee)'s power level as 2 and user B (general manager)'s power level as 6.

[0083] Back Figure 4 In step S6, the relational information acquisition unit 43 acquires power balance information. Power balance information refers to a dataset of the power level values ​​of the subject user and the object user. For example, when the subject user is user A and the object user is user B, the power balance information is represented as "2:6". Furthermore, when the subject user is B and the object user is C, the power balance information is represented as "6:6". Additionally, when the subject user is A and the object user is D, the power balance information is represented as "2:8".

[0084] In step S7, the sensitivity setting unit 44 determines whether the power balance is balanced. That is, it determines whether the power level values ​​of the subject user and the object user are the same. If the power balance between the subject user and the object user is balanced (yes in step S7), the process proceeds to step S11. If the power balance between the subject user and the object user is unbalanced (no in step S7), the process proceeds to step S8.

[0085] In step S8, the sensitivity setting unit 44 determines whether the power level of the subject user is higher than that of the object user. If the power level of the subject user is higher than that of the object user (yes in step S8), the process proceeds to step S9. If the power level of the subject user is lower than that of the object user (no in step S8), the process proceeds to step S10.

[0086] In step S9, the sensitivity setting unit 44 calculates the following formula 1 to set the value of the detection sensitivity level.

[0087] L1 = sp·(u1-u2)……(Equation 1)

[0088] L1 represents the detection sensitivity level value. That is, L1 is the threshold used to detect inappropriate speech or behavior. s represents the standard value of the detection sensitivity level. As described above, the standard value involved in this embodiment is set to 5. p represents the first correction coefficient used to adjust the detection sensitivity level value. The first correction coefficient is, for example, 0.5. In addition, the first correction coefficient can be changed within the range of 0 < p < 1. u1 represents the power level value of the user with the higher power level in the power balance information obtained in step S6. u2 represents the power level value of the user with the lower power level in the power balance information obtained in step S6.

[0089] Furthermore, the detection sensitivity levels involved in this embodiment are divided with a lower limit of 1 and an upper limit of 10. Therefore, when the value of L1 is less than 1, the sensitivity setting unit 44 obtains "L1 = 1" as the solution to Equation 1. Similarly, when the value of L1 is greater than 1, the sensitivity setting unit 44 obtains "L1 = 10".

[0090] In step S10, the sensitivity setting unit 44 calculates the following formula 2 to set the value of the detection sensitivity level.

[0091] L1 = s + p · (u1 - u2) …… (Equation 2)

[0092] Regarding L1, s, p, u1, u2, they are the same as those explained in Equation 1, so the explanation is omitted.

[0093] In step S11, the sensitivity setting unit 44 sets the standard value of the detection sensitivity level as the value of the detection sensitivity level.

[0094] In step S12, the detection unit 45 determines whether the value of the comparison object level stored in step S4 is greater than or equal to the detection sensitivity level value set in step S9, S10, or S11. If the value of the comparison object level is greater than or equal to the detection sensitivity level value (yes in step S12), the process proceeds to step S13. If the value of the comparison object level is less than the detection sensitivity level value (no in step S12), the process returns to step S1.

[0095] In step S13, the output unit 46 outputs alarm information.

[0096] The following examples illustrate the processing of steps S7 to S13.

[0097] (Example 1) In Example 1, imagine the following situation: through the processing from step S1 to step S6, the speech "Enough is enough" is detected as speech that can be equivalent to inappropriate speech, and it is determined that the speech was sent from user A to user B.

[0098] In Example 1, user A's power level (power level 2) and user B's power level (power level 6) are not the same value, so the process in step S7 is determined to be negative, and the process proceeds to step S8. Moreover, since user A's power level is lower than user B's power level, the process in step S8 is determined to be negative, and the process proceeds to step S10.

[0099] In step S10, Equation 2 is calculated. As described above, in this embodiment, the standard value of the detection sensitivity level is set to 5, and the first correction coefficient p is set to 0.5. Therefore, 5 is substituted into s in Equation 2, and 0.5 is substituted into p. Furthermore, user B (power level 6) is equivalent to a user with a higher power level, and user A (power level 2) is equivalent to a user with a lower power level, so 6 is substituted into u1, and 2 is substituted into u2. Therefore, the sensitivity setting unit 44 obtains the solution "L1 = 7" by calculating Equation 2. Based on this solution, the sensitivity setting unit 44 sets the detection sensitivity level for communication between user A and user B to 7. That is, the sensitivity setting unit 44 sets the value of the detection sensitivity level so that when user A communicates with user B, only inappropriate speech or behavior with a value of 7 or higher is detected as inappropriate speech or behavior.

[0100] Furthermore, in step S12, the detection unit 45 determines the relationship between the value of the inappropriate speech level (the value of the comparison object level) set for the speech "Let's stop this" and the value of the detection sensitivity level set in step S10. As described above, the value of the inappropriate speech level set for the speech "Let's stop this" is 4, and the value of the detection sensitivity level set in step S10 is 7. Therefore, in this example, the determination in step S12 is negative, and the process returns to step S1. That is, in Example 1, the speech "Let's stop this" from user A to user B is not detected as inappropriate speech.

[0101] (Example 2) In Example 2, imagine the following situation: Through the processing from step S1 to step S6, the speech "Enough is enough" is detected as speech that can be equivalent to inappropriate speech, and it is determined that the speech was sent from user B to user A.

[0102] In Example 2, the power level of user A (power level 2) and the power level of user B (power level 6) are not the same. Therefore, the determination in step S7 is negative, and the process proceeds to step S8. Moreover, since the power level of user B is higher than that of user A, the determination in step S8 is positive, and the process proceeds to step S9.

[0103] In step S9, Equation 1 is calculated. Similar to Example 1, 5 is substituted into s, 0.5 into p, 6 into u1, and 2 into u2 in Equation 1. Therefore, the sensitivity setting unit 44 obtains the solution "L1 = 3" by calculating Equation 1. Based on this solution, the sensitivity setting unit 44 sets the detection sensitivity level for communication between user B and user A to 3. That is, the sensitivity setting unit 44 sets the detection sensitivity level such that when user B communicates with user A, only inappropriate speech or behavior with a level of 3 or higher is detected as inappropriate.

[0104] Furthermore, in step S12, the detection unit 45 determines the relationship between the inappropriate behavior level set for the phrase "enough is enough" and the detection sensitivity level set in step S9. The inappropriate behavior level set for "enough is enough" is 4, and the detection sensitivity level set in step S9 is 3; therefore, in this example, it is determined to be inappropriate in step S12. That is, in Example 2, the phrase "enough is enough" sent from user B to user A is detected as inappropriate behavior. Moreover, in step S13, a message such as "this is inappropriate behavior" is output as an alarm message to user B's user terminal 2.

[0105] (Example 3) In Example 3, imagine the following situation: Through the processing of steps S1 to S6, the speech "Enough is enough" is detected as speech that can be equivalent to inappropriate speech, and it is determined that the speech was sent from user B to user C.

[0106] In Example 3, User B (Power Level 6) and User C (Power Level 6) have the same power level value, so in step S7, it is determined that this is true, and the process proceeds to step S11. Furthermore, in step S11, the sensitivity setting unit 44 sets 5, which is the standard value for the detection sensitivity level, as the detection sensitivity level value. That is, the sensitivity setting unit 44 sets the detection sensitivity level value so that when User B communicates with User C, only inappropriate speech or behavior with a value of 5 or higher is detected as inappropriate speech or behavior.

[0107] Furthermore, in step S12, the detection unit 45 determines the relationship between the inappropriate speech level set for the speech "Let's stop this" and the detection sensitivity level set in step S11. The inappropriate speech level set for "Let's stop this" is 4, and the detection sensitivity level set in step S11 is 5. Therefore, in this example, the determination in step S12 is negative, and the process returns to step S1. That is, in Example 3, the speech "Let's stop this" from user B to user C is not detected as inappropriate speech.

[0108] As described above, according to the information processing system 100 of Embodiment 1, a detection sensitivity level for detecting inappropriate speech or behavior is set based on power balance information. Specifically, in Equations 1 and 2 above, the values ​​of u1 and u2 are determined based on power balance information. Furthermore, in the information processing system 100, inappropriate speech or behavior is detected based on the detection sensitivity level determined based on power balance information. Thus, by setting the detection sensitivity based on power balance information, the detection sensitivity is set to reflect the interpersonal relationships within the group. As a result, inappropriate speech or behavior can be detected appropriately.

[0109] Furthermore, in the information processing system 100 according to Embodiment 1, when a first user communicates with a second user whose power level is lower than the first user, the above-mentioned Equation 1 is calculated. Here, "u1-u2" is calculated in Equation 1 to determine the power level difference between the first user and the second user. Moreover, in Equation 1, the larger the value of the power level difference calculated by "u1-u2", the smaller the detection sensitivity level value. That is, the detection sensitivity is set high. Thus, the greater the power level difference between the first user (i.e., the user with higher power level) and the second user (i.e., the user with lower power level), the easier it is for the behavior of the user with higher power level to be detected as inappropriate. As a result, when a user with higher power level communicates with a user with lower power level, the user with higher power level can be prompted to pay attention to their own behavior.

[0110] Furthermore, in the information processing system 100 according to Embodiment 1, when a second user communicates with a first user, the above-mentioned Equation 2 is calculated. Here, "u1-u2" is calculated in Equation 2 to determine the level difference between the first user and the second user. Moreover, in Equation 2, the larger the value of the level difference calculated by "u1-u2", the larger the value of the inappropriate speech level set as the detection sensitivity. That is, the detection sensitivity is set low. Thus, the larger the level difference, the more difficult it is to detect the speech of the second user, i.e., the user with the lower power level, as inappropriate speech. As a result, it is possible to promote positive communication between the user with the lower power level and the user with the higher power level.

[0111] (Implementation Method 2)

[0112] The information processing system 100S according to Embodiment 2 will be described. The information processing system 100S adjusts the value of the detection sensitivity level based on the amount of communication between users within a certain period. Furthermore, in Embodiment 2, the same symbols are used to mark the same components as in Embodiment 1, and descriptions are omitted.

[0113] Figure 5 This is a block diagram illustrating an example of the structure of an information processing system 100S. For example... Figure 5 As shown, the server 10S of the information processing system 100S has a processor 40S. The processor 40S replaces the acquisition unit 41 described above and has an acquisition unit 41S, and replaces the sensitivity setting unit 44 described above and has a sensitivity setting unit 44S.

[0114] The acquisition unit 41S in Embodiment 2, like the acquisition unit 41 described above, acquires both verbal and non-verbal information. Furthermore, the acquisition unit 41S acquires first information based on at least one of the verbal and non-verbal information. The first information is used to calculate the amount of communication between users within a certain period, and includes, for example, the time each user spends attending online meetings, the time each user spends communicating using online chat tools such as email, the time each user spends attending real-world meetings, and the time each user spends having conversations in the real world. A real-world meeting refers to a meeting held in the real world by at least three users. A real-world conversation refers to a one-on-one conversation between users in the real world. When acquiring the first information, the acquisition unit 41S may also refer to a predetermined schedule that records the activities of each user. This schedule information is stored, for example, in the user terminal 2.

[0115] The sensitivity setting unit 44S in Embodiment 2 is similar to the sensitivity setting unit 44 in Embodiment 1 in that it sets the value of the detection sensitivity level based on the power balance information.

[0116] Furthermore, the sensitivity setting unit 44S according to Embodiment 2 calculates the amount of communication between users based on the first information obtained based on at least one of verbal information and non-verbal information. Specifically, the sensitivity setting unit 44S calculates the amount of communication between users among multiple users based on at least one of the following: the time each user spends participating in online meetings, the time each user spends communicating using online chat tools, the time each user spends participating in real-world meetings, and the time each user spends having conversations in the real world.

[0117] Furthermore, the sensitivity setting unit 44S in Embodiment 2 adjusts the value of the detection sensitivity level based on the aforementioned AC quantity.

[0118] Regarding the processing of the information processing system 100S configured as described above, refer to... Figure 6 Please provide an explanation. Figure 6 This is a flowchart illustrating the overall processing of the information processing system 100S according to Embodiment 2. Furthermore, the following describes the process performed once per day. Figure 6 The example shown illustrates the processing. However, the process... Figure 6 The timing of the processing shown can be changed appropriately.

[0119] In step S101, the acquisition unit 41S acquires both verbal and non-verbal information. Furthermore, the acquisition unit 41S obtains the first information based on at least one of the verbal and non-verbal information. Additionally, while acquiring the first information, the acquisition unit 41S can also acquire the number of characters in the text sent by each user during their online chat communication, replacing the time required for communication using the online chat tool.

[0120] In step S102, the sensitivity setting unit 44S calculates the AC quantity c(t) over a certain period using the following formula 3. The certain period is, for example, one day.

[0121] c(t) = c1·k1 + c2·k2 + c3·k3 + c4·k4 ... (Equation 3)

[0122] In Equation 3 above, c1 represents the total participation time of a given user in an online meeting. k1 represents the first weighting coefficient used to correct c1. c2 represents the total participation time of a given user in communication using an online chat tool. k2 represents the second weighting coefficient used to correct c2. c3 represents the total participation time of a given user in a real-world meeting. k3 represents the third weighting coefficient used to correct c3. c4 represents the total conversation time of a given user in the real world. k4 represents the fourth weighting coefficient used to correct c4. Furthermore, in this embodiment, the values ​​of k1 to k4 are set such that k1 < k2 < k3 < k4. However, the relative sizes of k1 to k4 can be appropriately changed.

[0123] As an example, consider calculating the amount of communication c(t) between user A and user B. In this case, substituting c1 represents the total time user A spends participating in the online meeting with user B. Substituting c2 represents the total time required for user A and user B to communicate using online chat. Substituting c3 represents the total time user A spends participating in the real-world meeting with user B. Substituting c4 represents the total real-world conversation time between user A and user B.

[0124] Furthermore, in step S101, after the acquisition unit 41S obtains the number of characters in the text sent by each user in their online chat communication, the sensitivity setting unit 44S converts the number of text characters into the time required for communication using the online chat tool. For example, based on the language information acquired in step S101, the acquisition unit 41S calculates the average number of characters spoken by a user within one minute. Moreover, the acquisition unit 41S performs a conversion process based on this average number of spoken characters and the number of text characters. Specifically, if user A's average number of spoken characters within one minute is 100, and user A sends 300 characters of text in the online chat tool, the acquisition unit 41S calculates that the time required for user A to communicate in the online chat tool is 3 minutes.

[0125] In step S103, the sensitivity setting unit 44S adjusts the value of the detection sensitivity level based on the AC quantity c(t) over a certain period. Specifically, the sensitivity setting unit 44S calculates the following equation 4.

[0126] L3 = L1 + l·c(t) …… (Equation 4)

[0127] L3 represents the detection sensitivity level value adjusted based on the AC quantity over a certain period. L1 represents the detection sensitivity level value set based on power balance information. Examples of setting the detection sensitivity level value based on power balance information are explained in steps S1 to S11 of Embodiment 1, and therefore are omitted from the explanation. l represents the second correction coefficient used to correct the value of L3. As described above, c(t) represents the AC quantity over a certain period.

[0128] Furthermore, in step S103, the sensitivity setting unit 44S sets L3, calculated by formula 4, as the value of the detection sensitivity level.

[0129] As explained above, the information processing system 100S according to Embodiment 2 can dynamically adjust the value of the detection sensitivity level, which is statically set based on power balance information, based on the communication volume c(t) within a certain period. That is, the value of the detection sensitivity level set by the processing through steps S9, S10, or S11 according to Embodiment 1 can be adjusted based on the communication volume c(t) within a certain period. Specifically, in Equation 4 above, the larger the value of the communication volume c(t) within a certain period, the larger the value of L3. In other words, the detection sensitivity decreases. Thus, the detection sensitivity can be reduced between users who are considered to have a close relationship due to a large value of communication volume c(t) within a certain period. Therefore, the detection sensitivity can be set to more accurately reflect the interpersonal relationships within a group. As a result, inappropriate words and actions can be detected more appropriately.

[0130] (Implementation Method 3)

[0131] The information processing system 100T according to Embodiment 3 will be described. The information processing system 100T analyzes users' emotions based on their facial expressions during communication and adjusts the detection sensitivity based on the analysis results. Furthermore, in Embodiment 3, the same symbols are used to mark the same components as in Embodiment 2, and descriptions are omitted.

[0132] Figure 7 This is a block diagram illustrating the structure of the information processing system 100T. (For example...) Figure 7 As shown, the server 10T of the information processing system 100T includes a memory 30T and a processor 40T. The memory 30T includes an expression information storage unit 34. The processor 40T replaces the acquisition unit 41 of Embodiment 1 with an acquisition unit 41T, and replaces the sensitivity setting unit 44 of Embodiment 1 with a sensitivity setting unit 44T. In addition, the processor 40T includes an emotion analysis unit 47T.

[0133] The facial expression information storage unit 34 stores various template images representing human facial expressions. Specifically, the facial expression information storage unit 34 stores template images representing expressions such as anger, crying, laughing, and happiness. In addition, the facial expression information storage unit 34 may also store template images representing expressions of surprise, expressions of drowsiness, etc.

[0134] The acquisition unit 41T acquires user facial expression information, representing the facial expressions of users during communication, as non-verbal information. This user facial expression information can be constructed, for example, by writing image data or animation data of the user's facial expressions.

[0135] The Emotion Analysis Unit 47T analyzes users' emotions during communication based on their facial expressions. Furthermore, based on the analysis results, it evaluates whether the communication conducted within a certain period was positive or negative.

[0136] The sensitivity setting unit 44T in Embodiment 3, like the sensitivity setting unit 44T in Embodiment 1, sets the detection sensitivity level value based on power balance information. Furthermore, the sensitivity setting unit 44T adjusts the detection sensitivity level value based on the evaluation results of the emotion analysis unit 47T.

[0137] Regarding the processing of the information processing system 100T as described above, please refer to... Figure 8 Please provide an explanation. Figure 8 This is a flowchart illustrating the overall processing of the information processing system 100T according to Embodiment 3.

[0138] In step S201, the acquisition unit 41T acquires language information and non-language information. In particular, the acquisition unit 41T acquires user facial expression information as non-language information.

[0139] In step S202, the emotion analysis unit 47T analyzes the user's emotions during communication based on the user's facial expression information. An example of the processing performed by the emotion analysis unit 47T will be described below.

[0140] For example, imagine a 3-minute conversation between user A and user B, which is collected by the acquisition unit 41T as verbal information. Furthermore, imagine a scenario where facial expressions representing the two users during the conversation are collected by the acquisition unit 41T as nonverbal information. In this case, the emotion analysis unit 47T segments the conversation every minute and categorizes it into conversation interval A, conversation interval B, and conversation interval C.

[0141] Next, the emotion analysis unit 47T extracts the feature values ​​of the template images stored in the expression information storage unit 34 and the feature values ​​of user A's expression information in conversation interval A, and calculates the similarity between the two based on these feature values. Furthermore, the emotion analysis unit 47T analyzes the user's emotion based on the template image most similar to the user's expression information. For example, if the template image representing anger has the highest similarity to user A's expression information, the emotion analysis unit 47T analyzes that user A is angry. Or, if the template image representing a smile has the highest similarity to user A's expression information, the emotion analysis unit 47T analyzes that user A is smiling. The emotion analysis unit 47T uses the same method to analyze user B's emotion. Moreover, if emotions such as anger or crying are most frequently observed in conversation interval A, the emotion analysis unit 47T evaluates it as a negative exchange in conversation interval A. Conversely, if emotions such as laughter or happiness are most frequently observed in conversation interval A, the emotion analysis unit 47T evaluates it as a positive exchange in conversation interval A. The same processing is performed for conversation intervals B and C.

[0142] The Emotion Analysis Department 47T performs the above processing on all communications between User A and User B within a certain period (e.g., 1 day).

[0143] In step S203, the sensitivity setting unit 44T calculates the following formula 5.

[0144] L4 = L1 + c1(t)·m1 - c2(t)·m2……(Equation 5)

[0145] L4 represents the detection sensitivity level adjusted based on the user's emotion. L1 represents the detection sensitivity level set based on power balance information. c1(t) represents the total time of positive communication within a certain period. m1 represents the third correction coefficient used to correct c1(t). c2(t) represents the total time of negative communication within a certain period. m2 represents the fourth correction coefficient used to correct c2(t).

[0146] Furthermore, in step S203, the sensitivity setting unit 44T sets L4, calculated by Equation 5 above, as the value of the detection sensitivity level.

[0147] As explained above, the information processing system 100T according to Embodiment 3 can dynamically adjust the value of the detection sensitivity level, which is statically set based on power balance information, based on the emotions expressed by the user's facial expressions. That is, the value of the detection sensitivity level set through steps S9, S10, or S11 according to Embodiment 1 can be adjusted based on the user's emotions. Specifically, in Equation 5 above, the value of L4 increases between users who have engaged in positive communication and are considered to have positive emotions towards each other. That is, the detection sensitivity decreases. On the other hand, the value of L4 decreases between users who have engaged in negative communication and are considered to have negative emotions towards each other. That is, the detection sensitivity increases. According to this structure, the detection sensitivity can be set to more accurately reflect the interpersonal relationships within a group. As a result, inappropriate words and actions can be detected more appropriately.

[0148] (Implementation Method 4)

[0149] The information processing system 100U according to Embodiment 4 will be described. The information processing system 100U analyzes users' emotions based on messages sent and received between users, and adjusts the detection sensitivity based on these emotions. Furthermore, in Embodiment 4, the same symbols are used to mark the same components as in Embodiment 1, and descriptions are omitted.

[0150] Figure 9 This is a block diagram illustrating an example of the structure of an information processing system 100U. (See diagram for example.) Figure 9 As shown, the server 10U of the information processing system 100U includes a memory 30U and a processor 40U. The memory 30U includes an emotion dictionary storage unit 35. Furthermore, the processor 40U replaces the acquisition unit 41 of Embodiment 1 with an acquisition unit 41U, and replaces the sensitivity setting unit 44 of Embodiment 1 with a sensitivity setting unit 44U. In addition, the processor 40U includes an emotion analysis unit 47U.

[0151] The emotion dictionary storage unit 35 stores an emotion dictionary containing words that represent human emotions. Emotional information is associated with each word in the emotion dictionary. Emotional information refers to whether a word is positive or negative, represented in this example by -1, 0, or +1. -1 indicates a negative word, +1 indicates a positive word, and 0 indicates neutral.

[0152] The acquisition unit 41U collects the text contained in messages sent and received between users as language information.

[0153] The Sentiment Analysis Unit 47U analyzes users' emotions based on the text contained in messages sent and received between users.

[0154] Similar to the sensitivity setting unit 44 in Embodiment 1, the sensitivity setting unit 44U sets the detection sensitivity level value based on power balance information. Furthermore, the sensitivity setting unit 44U adjusts the detection sensitivity level value based on the analysis results of the emotion analysis unit 47U.

[0155] Regarding the processing of the information processing system 100U as described above, please refer to... Figure 10 Please provide an explanation. Figure 10 This is a flowchart illustrating the overall image of the processing of the information processing system 100U according to Embodiment 4.

[0156] In step S301, the acquisition unit 41U acquires both verbal and nonverbal information. At this time, the acquisition unit 41U acquires the text contained in messages sent and received between users via online chat tools over a certain period (e.g., one day). For example, the acquisition unit 41U acquires messages exchanged between user A and user B.

[0157] In step S302, the emotion analysis unit 47U analyzes the user's emotions during communication based on the aforementioned text. An example of the processing performed by the emotion analysis unit 47U will be described below.

[0158] The sentiment analysis unit 47U first performs morpheme analysis on the aforementioned text, using vector transformation techniques such as Word2Vec to convert the obtained words into numerical vectors. Next, the words registered in the sentiment dictionary stored in the sentiment dictionary storage unit 35 are vectorized using the same method. Then, the similarity between the words extracted from the text and the words registered in the sentiment dictionary is calculated. Cosine distance, Euclidean distance, and other distances can be used to calculate the similarity. Furthermore, based on the aforementioned similarity, the word most similar to the word extracted from the text is retrieved from the words registered in the sentiment dictionary. Moreover, based on the sentiment information associated with this most similar word, the extracted word is evaluated as either a positive or negative word.

[0159] For example, imagine a text sent from user A to user B containing the word "good morning," and the word "hello" is registered in the sentiment dictionary. Here, we assume that the word with the highest similarity to "good morning" (closest, such as cosine distance) is "hello." Furthermore, we assume that in the sentiment dictionary, the sentiment information "+1" is associated with words like "hello." That is, we assume that words like "hello" registered in the sentiment dictionary represent positive emotions. In this case, the sentiment analysis unit 47U evaluates "good morning" as a positive word. The sentiment analysis unit 47U performs the above processing on all words contained in all messages exchanged between user A and user B within a certain period. Moreover, the sentiment analysis unit 47U calculates Equation 6 below.

[0160] L5 = L1 + d1(t)·n1 - d2(t)·n2……(Equation 6)

[0161] L5 represents the detection sensitivity level adjusted for user sentiment. L1 represents the detection sensitivity level set based on power balance information. d1(t) represents the total number of positive words detected from the text. n1 represents the fifth correction coefficient used to correct d1(t). d2(t) represents the total number of negative words detected from the text. n2 represents the sixth correction coefficient used to correct d2(t). According to Equation 6 above, the more positive words are contained in the text sent and received between users within a certain period, the larger the value of L5 becomes. That is, the detection sensitivity decreases. On the other hand, the more negative words are contained in the text sent and received between users within a certain period, the smaller the value of L5 becomes. That is, the detection sensitivity increases.

[0162] Furthermore, in step S303, the sensitivity setting unit 44U sets L5, calculated by Equation 6 above, as the value of the detection sensitivity level.

[0163] Based on the above structure, among users whose text messages contain more positive words and who are considered to have positive emotions towards each other, the detection sensitivity can be adjusted to be lower. Conversely, among users whose text messages contain more negative words and who are considered to have negative emotions towards each other, the detection sensitivity can be adjusted to be higher. Therefore, the detection sensitivity can be set to more accurately reflect interpersonal relationships within a group. As a result, inappropriate words and actions can be detected more appropriately.

[0164] The present disclosure can be adapted to the following variations.

[0165] (Modified example)

[0166] (1) The information processing system 100 can also be used to determine whether a user's actions constitute inappropriate speech or behavior. For example, the user's actions can be recorded in the form of animation data using one or more of the aforementioned camera devices, and the detection unit 45 analyzes the animation data using existing image recognition technology, which includes... Figure 3 In cases where actions registered in column 1, CL1 (violent behavior, lip-smacking behavior, table-banging behavior, etc.) are detected by the detection department 45... Figure 4 In step S3, it is determined to be yes.

[0167] In addition, the information processing system 100 can also be used to determine whether words contained in text sent and received in online chat tools or the like are equivalent to inappropriate speech or behavior.

[0168] (2) In Implementation 1, an example of calculating a user's power level based on the user's job title was described, but a user's power level can also be calculated based on the amount of time a user speaks. For example, the total speaking time of each user within a certain period can be calculated based on the amount of time a user speaks, and the power level can be evaluated from high to low in order of the total speaking time from long to short.

[0169] In addition, users' power levels can also be evaluated based on the volume of their voices. For example, the average volume of each user's voice can be calculated, and the power level can be evaluated from high to low according to the average volume.

[0170] In addition, a user's power level can also be evaluated based on the content of their posts.

[0171] (3) As an example of the relationship between users, the balance of power is listed, but the relationship is not limited to this. For example, the relationship can also be a superior-subordinate relationship, etc.

[0172] (4) The methods used by the sentiment analysis department to analyze users' emotions are not limited to the examples mentioned above. For example, the sentiment analysis department may also analyze users' emotions based on the content of their statements.

[0173] Industrial availability

[0174] This disclosure is useful in the technical field of detecting inappropriate speech or behavior.

Claims

1. An information processing method, specifically an information processing method in a computer. In the information processing method, Collect verbal and nonverbal information from multiple users who constitute a group. Based on at least one of the linguistic information and the non-linguistic information, an evaluation is made of the power hierarchy representing the relative strength of each of the multiple users within the group. Based on the respective power levels of the multiple users, relational information representing the relationships between the users is obtained. The detection sensitivity is set based on the aforementioned relational information to detect inappropriate speech or behavior in interactions between users among the multiple users. The inappropriate words or actions in the communication are determined based on the aforementioned detection sensitivity. If the communication includes the inappropriate words or actions, information is output to indicate that the inappropriate words or actions have been detected.

2. The information processing method according to claim 1, wherein, In the collection of non-verbal information, attribute information representing the attributes of each of the multiple users is collected. In the evaluation of the power level, level correspondence information representing the correspondence between the attribute and the power level is obtained. Based on the attribute information and the level correspondence information, the power level of each of the multiple users is evaluated.

3. The information processing method according to claim 1 or 2, wherein, In the process of collecting the language information, the amount of speech given by each of the multiple users is collected. In the evaluation of the power level, the power level of each of the multiple users is evaluated based on the amount of speech.

4. The information processing method according to claim 1 or 2, wherein, In setting the detection sensitivity, when communication occurs between a first user and a second user whose power level is lower than that of the first user, a level difference representing the difference in power levels between the first user and the second user is calculated. The detection sensitivity will be set high when the first user communicates with the second user, based on the magnitude of the grade difference.

5. The information processing method according to claim 4, wherein, In setting the detection sensitivity, the detection sensitivity is set low when the second user communicates with the first user, based on the magnitude of the grade difference.

6. The information processing method according to claim 1 or 2, wherein, Further, based on at least one of the linguistic information and the non-linguistic information, the amount of communication between users among the plurality of users is calculated. In setting the detection sensitivity, the detection sensitivity is adjusted according to the AC quantity.

7. An information processing device, comprising: The data collection department collects verbal and nonverbal information from multiple users who constitute the group. The rating department evaluates the power levels representing the strength of each of the plurality of users within the group, based on at least one of the linguistic information and the non-linguistic information. The acquisition unit acquires relational information representing the relationships between users among the multiple users, based on the respective power levels of the multiple users. The sensitivity setting unit sets the detection sensitivity for detecting inappropriate speech or behavior from interactions between users among the plurality of users based on the relational information. The determination unit uses the detection sensitivity as a benchmark to determine whether the inappropriate words or actions are included in the communication. and The output unit outputs information indicating that the inappropriate speech or behavior has been detected, in the case that the communication includes such inappropriate speech or behavior.

8. An information processing program that enables a computer to function as an information processing device, said information processing program causing the computer to perform the following processes: Collect verbal and nonverbal information from multiple users who constitute a group. Based on at least one of the linguistic information and the non-linguistic information, an evaluation is made of the power hierarchy representing the relative strength of each of the multiple users within the group. Based on the respective power levels of the multiple users, relational information representing the relationships between the users is obtained. The detection sensitivity is set based on the aforementioned relational information to detect inappropriate speech or behavior in interactions between users among the multiple users. The inappropriate words or actions in the communication are determined based on the aforementioned detection sensitivity. If the communication includes the inappropriate words or actions, information is output to indicate that the inappropriate words or actions have been detected.

Images