JP2023009563A - Harassment prevention system and harassment prevention method - Google Patents

Abstract

To provide a technique for effectively preventing harassment behavior occurring under a remote work environment.SOLUTION: A server 10 constituting a harassment prevention system 1 comprises: a storage device 101 for holding evaluation information which determines an influence degree to harassment on each event which accompanies conversation and a calculation formula of a harassment level with an influence degree of each event as a variable; and an arithmetic unit 104 which acquires voice data of utterance, extracts prescribe vocabulary indicating a conversation participant, specifies the number of conversation participants and an utterance object, extracts data corresponding to each event from the voice data, compares the data with evaluation information, determines the influence degree on each event and calculates the harassment level by applying the influence degree to the calculation formula.SELECTED DRAWING: Figure 2

Description

The present invention relates to a harassment prevention system and a harassment prevention method.

With the spread of remote work in recent years, the number of opportunities to hold online meetings is increasing rapidly. Participants in such online meetings often behave differently from traditional face-to-face meetings.
Therefore, it is difficult for each participant to immediately recognize how his or her remarks are received by other participants, based on their facial expressions, reactions, and atmosphere.

This situation is particularly conspicuous in the case of a conference in which communication is performed only by voice, not in a conference where video can be confirmed in real time. Naturally, the problem that it is difficult to grasp the other party's reaction also leads to the possibility that one's own remarks, etc., unintentionally or intentionally become harassment acts toward other participants.

In the case of online meetings in a remote work environment, there is a possibility that colleagues and superiors in the organization will not even notice the existence of the problem, let alone point out and give guidance to the above harassment behavior.

Conventional technologies related to the avoidance of such harassment acts include a system that provides technology that can detect harmful acts such as power harassment, sexual harassment, and bullying in the workplace environment and support countermeasures (see Patent Document 1). is proposed.

The system is a harmful behavior detection system comprising a computer for observing and detecting harmful behaviors including power harassment, sexual harassment, and bullying among people in a work environment, the computer monitoring the behavior of people around a subject. voice data obtained by inputting voice, voice information including words representing conversation content extracted based on voice recognition from the voice data, emotional information representing emotions classified based on emotion recognition from the voice data, and Acquiring data including at least one of vital data of a subject, date and time information, and location information of the subject, and analyzing words and emotions of others around the subject based on the acquired data and at least one of the five elements of the subject's words, emotions, and vital data to calculate an index value related to the harmful behavior, and based on the index value, the evaluation of the harmful behavior Estimating the state of the harmful act including at least one of the victim and the perpetrator and the presence or absence or degree of the harmful act, and generating coping data for coping with the harmful act according to the estimated state This is the output.

In addition, a harassment detection program that appropriately detects the occurrence of harassment, a program that provides a harassment detection system and a harassment detection method (see Patent Document 2), and the like have been proposed.

This program acquires voice data related to a first person and matching information related to the first person, which is voice data related to the first person and matching information for matching the positions of the first person and the second person, Acquiring the physical information data and the verification information regarding the second person, detecting a first suspicion of harassment of the first person based on the voice data obtained regarding the first person, A second suspicion of harassment of said second person is detected based on the physical information data obtained with respect to the two persons, and the result of matching the time of detection of said first suspicion with the time of detection of said second suspicion. , and the occurrence of harassment between the first person and the second person based on the matching result of the matching information obtained about the first person and the matching information obtained about the second person It detects.

JP 2020-123204 A Japanese Patent Application Laid-Open No. 2020-009238

However, even if the conventional technology is adopted, it is not possible to observe the positions of the participants and the distance to each other when it is assumed that the participants participate in the conference at remote locations. effect cannot be expected. These points are the same even when adopting the technology that uses the voices around the participants as input.
After all, no technology is provided to effectively deal with harassment that occurs in today's remote work environment.

SUMMARY OF THE INVENTION Accordingly, an object of the present invention is to provide a technique for effectively preventing harassment that occurs in a remote work environment.

A harassment prevention system of the present invention that solves the above problems holds evaluation information that defines the degree of influence of various events accompanying conversation on harassment, and a calculation formula for a harassment level that uses the degree of influence of the various events as variables. Acquisition of voice data of utterances by conversation participants from a storage device and a predetermined device, extracting predetermined vocabulary indicating conversation participants from the voice data, and specifying the number of participants in the conversation and the utterance target based on the vocabulary , a process of extracting data corresponding to the various events from the voice data, comparing the data with the evaluation information, and determining the degree of influence of the various events; and an arithmetic device that executes processing for calculating the level.
Further, in the harassment prevention method of the present invention, the information processing device stores evaluation information that defines the degree of influence of various events accompanying conversation on harassment, and a formula for calculating the harassment level using the degree of influence of the various events as a variable. Stored in a storage device, voice data of conversation participants' utterances is acquired from a predetermined device, predetermined vocabulary indicating conversation participants is extracted from the voice data, and the number of conversation participants and the utterance target are specified based on the vocabulary a process of extracting data corresponding to the various events from the audio data and comparing the data with the evaluation information to determine the degree of impact related to the various events; and applying the determined degree of impact to the calculation formula. and executing a process of calculating the harassment level.

According to the present invention, it is possible to effectively prevent harassment that occurs in a remote work environment.

It is a figure which shows the network configuration example of the harassment prevention system of this embodiment. It is a figure which shows the hardware structural example of the server in this embodiment. It is a figure which shows the hardware structural example of the terminal in this embodiment. It is a figure which shows the structural example of the word evaluation (NG word) table of this embodiment. It is a figure which shows the structural example of the word evaluation (following word) table of this embodiment. It is a figure which shows the structural example of the emotion evaluation table of this embodiment. It is a figure which shows the structural example of the interception value table of this embodiment. It is a figure which shows the structural example of the silence length reference table of this embodiment. It is a figure which shows the structural example of the conversation object evaluation table of this embodiment. It is a figure which shows the structural example of the harassment determination table of this embodiment. It is a figure which shows the structural example of the input information table of this embodiment. It is a figure which shows the flow example of the harassment prevention method in this embodiment. It is a figure which shows the example of harassment degree calculation in this embodiment. It is a figure which shows the flow example of conversation directionality determination in this embodiment. It is a figure which shows the hardware structural example of the server in this embodiment. It is a figure which shows the example of the participant attribute information in this embodiment. It is a figure which shows the structural example of the table for definition update of the individual harassment model of this embodiment. It is a figure which shows the flow example of the harassment prevention method in this embodiment. It is a figure which shows the hardware structural example of the server in this embodiment. It is a figure which shows the flow example of the harassment prevention method in this embodiment. It is a figure which shows the example of a screen in this embodiment.

<Network configuration>
Embodiments of the present invention will be described in detail below with reference to the drawings. FIG. 1 is a network configuration diagram including a harassment prevention system 1 of this embodiment. A harassment prevention system 1 shown in FIG. 1 is a computer system that can effectively prevent harassment behavior occurring in a remote work environment.

In the harassment prevention system 1 of the present embodiment, as shown in FIG. 1, a server 10 serving as a harassment prevention device and terminals 20 operated by participants in an online conference are communicably connected via a network 5. It is configured. In addition, the harassment prevention system 1 of this embodiment shall include the server 10 as the minimum structure.

The server 10 acquires at least audio data, more preferably video data, of the content spoken through the terminal 20 during the online conference, and based on these data, performs a series of processing such as determination of the harassment level and output of the result. Run.

On the other hand, the terminal 20 distributes the above-described voice data and the like during the online conference to the server 10 via the network 5 in real time or at fixed time intervals. In addition, the server 10 receives a notification of the harassment level determination result, etc., outputs this on a suitable output device such as a display or a speaker, and presents it to the conference participants and the like.
<Hardware Configuration: Server>
Also, the hardware configuration of each device constituting the harassment prevention system 1 of this embodiment will be described. FIG. 2 is a diagram showing a hardware configuration example of the server 10 in this embodiment.

As illustrated in FIG. 2 , the server 10 includes a storage device 101 , a memory 103 , a CPU (Central Processing Unit) 104 as an arithmetic device, an input/output I/F 105 and a network I/F 106 .

Among them, the storage device 101 is composed of an appropriate non-volatile storage element such as an SSD (Solid State Drive) or a hard disk drive.

Also, the memory 103 is composed of a volatile memory element such as a RAM (Random Access Memory).

The CPU 104 is an arithmetic unit that reads programs 110 to 114 stored in the storage device 101 into the memory 103 and executes them, performs overall control of the device itself, and performs various determinations, calculations, and control processes.

Also, the input/output I/F 105 is an interface for exchanging data with an input device that receives key input and voice input from the user as needed, and an output device such as a display that displays processed data.

Also, the network I/F 106 is a communication device that connects to the network 5 and performs communication processing with the terminal 20 .

In the storage device 101, in addition to programs 110 to 114 for implementing the functions necessary for the server 10 of this embodiment, various data obtained from the terminal 20, tables used for determination, etc. are stored as information. It is stored in storage area 115 .

The data held in this information storage area 115 include word evaluation information (NG word) 120, word evaluation information (following word) 121, emotion evaluation information 122, interruption evaluation information 123, silence length evaluation information 124, conversation object Evaluation 125, harassment judgment criteria 126, table 127 for updating definition of individual harassment model, input information table 128, and participant attribute information 129 are assumed. However, the details of these data, tables, etc. will be described later.

Among the programs held in the storage device 101, the input processing program 110 implements a process of acquiring and shaping voice data from the terminal 20 and a function of acquiring data relating to the participants of the online conference.

Also, the conversation direction determination program 111 implements a function of determining the destination of speech in an online conference.

Also, the harassment judgment program 112 implements a function of judging the degree of harassment regarding the conversation content of the speaker during the online conference based on the features of each event obtained from the voice data.

In addition, the feedback generation program 113 has a function of generating a message to be displayed to the speaker during the online conference based on the determination result of the harassment determination program 112, and identifies the participant who made a statement determined to be harassment. implement the functionality to

The output processing program 114 also implements a function of distributing the message generated by the feedback generation program 113 to the terminal 20 of the speaker, which is an appropriate display destination, and displaying the message.
<Hardware Configuration: Terminal>
Also, the hardware configuration of the terminal 20 that constitutes the harassment prevention system 1 of this embodiment will be described. FIG. 3 is a diagram showing a hardware configuration example of the terminal 20 in this embodiment.

As illustrated in FIG. 3, the terminal 20 includes a storage device 201, a memory 203, a CPU (Central Processing Unit) 204 serving as an arithmetic unit, an input I/F 205,
It has an output I/F 206 and a network I/F 207 .

Among these, the storage device 201 is composed of an appropriate non-volatile storage element such as an SSD (Solid State Drive) or a hard disk drive.

Also, the memory 203 is composed of a volatile memory element such as a RAM (Random Access Memory).

The CPU 204 is an arithmetic unit that reads programs 210 to 214 stored in the storage device 201 into the memory 203 and executes them, performs overall control of the device itself, and performs various determinations, calculations, and control processes.

In addition, the input I/F 205 is an input device itself (eg, microphone, camera, keyboard, mouse, touch panel, etc.) that receives key input and voice input from the user as necessary, or executes data exchange with such an input device. It is an interface to

Also, the output I/F 206 is an output device (eg, a display, a speaker, etc.) itself for displaying processed data, or an interface for exchanging data with such an output device.

A network I/F 207 is a communication device that connects to the network 5 and performs communication processing with the server 10 .

In addition to the programs 210 to 214 for implementing the functions necessary for the terminal 20 of this embodiment, the storage device 201 also stores audio data and video images of users who operate the terminal 20 to participate in the online conference. Data and the like are stored in the information storage area 215 .

Of the programs held in the storage device 201, a voice data acquisition program 210 implements a function of acquiring voice data, which is speech content, from a microphone, which is the input I/F 205. FIG.

Also, the image data acquisition program 211 implements a function of acquiring image data of conference participants, who are users, from the camera, which is the input I/F 205 . However, this image data acquisition program 211 is not essential.

Also, the input data acquisition program 212 implements a function of acquiring the user's key input as text data from the keyboard, which is the input I/F 205 . However, this input data acquisition program 212 is not essential.

Also, the harassment button input detection program 213 implements a function of detecting that the harassment notification button has been pressed on the input I/F 205 . The harassment notification button is an interface that is pressed when a user who is a conference participant recognizes that he or she has been harassed.

Also, the data transmission program 214 implements a function of transmitting the data acquired by the above-described programs and the identification information of the terminal 20 (which may also include the concept of identification information of participants) to the server 10 .
<Data structure example>
Next, various data held in the storage device 101 by the server 10 of this embodiment will be described with reference to FIGS. 4A to 4G and 5. FIG.

FIG. 4A shows a data configuration example of the word evaluation information (NG word) 120 in this embodiment. The word evaluation information (NG words) 120 of the present embodiment is a table that defines the possibility of leading to harassment behavior, that is, the degree of impact, regarding words that are assumed to be included in utterances in an online conference.

The degree of influence defined by the word evaluation information (NG word) 120 is an evaluation value of each word defined by a knowledgeable person or obtained by statistically processing past case data. As a property of the evaluation value, the higher the possibility of being linked to harassment, the larger the absolute value.

FIG. 4B shows a data configuration example of the word evaluation information (subsequent word) 121 in this embodiment. The word evaluation information (subsequent word) 121 of the present embodiment is a word assumed to be included in the utterance of the online conference, and the person who received the NG word after the utterance of the NG word (that is, the victim of harassment) This is a table in which the degree of influence is defined as the degree of likelihood of words that are likely to be uttered by a person who is likely to be a person to respond.

The degree of influence defined by the word evaluation information (subsequent word) 121 is an evaluation value of each word defined by a knowledgeable person or obtained by statistically processing past case data. As a property of the evaluation value, the higher the possibility of appearing as a reaction to the NG word, the larger the absolute value.

FIG. 4C shows a data configuration example of the emotion evaluation information 122 in this embodiment. The emotion evaluation information 122 of the present embodiment is obtained by inputting the above-described voice data into a predetermined emotion analysis program, and evaluates the intentions of the above-mentioned NG words and subsequent words for each type of emotion indicated by the voice data. It is a table defined in the form of values. In other words, even if the same word is used, the degree of association with harassment behavior differs depending on the emotional intention of the voice at that time.

FIG. 4D shows a data configuration example of the obstruction evaluation information 123 in this embodiment. The interruption evaluation information 123 of the present embodiment defines the possibility that an event in which the other party interrupts a conversation during an online conference leads to harassment.

FIG. 4E shows a data configuration example of the silence length evaluation information 124 in this embodiment. The silence length evaluation information 124 of the present embodiment indicates that after the other party (the person who uttered the NG word) uttered during the conversation in the online conference, the person who received this (probably was shocked by the NG word and hurt) remained silent. It defines the possibility that the event in the state will lead to harassment.

FIG. 4F shows a data configuration example of the conversation target evaluation information 125 in this embodiment. Conversation target evaluation information 125 of the present embodiment identifies NG words and speaks them depending on whether the conversation between conference participants in the online conference was one-on-one or with a plurality of people. Depending on whether it was addressed to the victim or addressed to the victim without being specifically addressed to him or her, the mental shock received will differ, and this will define the possibility of leading to harassment behavior.

It should be noted that the content of each evaluation information shown in FIGS. 4A to 4F is, for example, a standard harassment model that can be generally applied to conference participants of an online conference. On the other hand, an individual harassment model is obtained by tuning the contents of the evaluation information for each conference participant or each member's organization. In this individual harassment model, the numerical values stipulated in each of the above evaluation information are adjusted according to, for example, regional characteristics such as the birthplace and location of the meeting participants, and social positions such as positions within the organization ( details later).

FIG. 4G shows a data configuration example of the harassment judgment criteria 126 in this embodiment. The harassment determination criteria 126 of this embodiment are criteria for determining whether or not the degree of harassment calculated by the server 10 based on the evaluation values for each event shown in FIGS. 4A to 4F is harassment.

FIG. 5 shows a data configuration example of the input information table 128 in this embodiment. The input information table 128 in this embodiment is a table in which voice data and the like obtained by the server 10 from the terminal 20 are stored. Note that this voice data is data obtained by recording the speech voices of the conference participants via a microphone whose input I/F 205 is the terminal 20 .

Each record in the input information table 128 has values such as the speech start time, the speech end time, voice data, the interruption flag, and the speaking partner, using the participant ID that uniquely identifies the conference participant as a key. are associated.

It should be noted that such a record is voice data of one section from the timing when the voice of a conference participant appears until the voice is interrupted.

Also, the interruption flag indicates that an event has occurred in which the speech is interrupted by a conference participant other than the conference participant. If another speaker starts to speak before the voice interruption is detected, it is determined as an interruption, and the "interruption flag" of the record of the other speaker becomes "1".

Also, the utterance partner is a value indicating whether the other conference participants who are conversing with the conference participant leading the utterance in the record are an individual or a plurality of persons.
<Flow example: Main flow>
The actual procedure of the harassment prevention method according to this embodiment will be described below with reference to the drawings. Various operations corresponding to the harassment prevention method described below are realized by a program that is read into a memory or the like and executed by the server 10 that mainly constitutes the harassment prevention system 1 . This program is composed of codes for performing various operations described below.

FIG. 6 is a diagram showing a flow example of a harassment prevention method according to this embodiment. In this case, the server 10 determines who the speaking partner is (that is, the direction of the conversation) from the voice features and voice data of the participant attribute information 129 (described in detail in FIG. 10) (s1). The detailed logic of this determination will be described later with reference to FIG.

As a result of the above determination, if the utterance partner is neither an individual nor a plurality of persons (s2: No), the server 10 concludes the processing as a soliloquy of the conference participant.

On the other hand, as a result of the above determination, if the utterance partner is either an individual or a plurality of people (s2: Yes), the server 10 determines from the participant attribute information 129 and the voice data that each conference participant appearing in the conversation is Determine who it is (s3).

Subsequently, the server 10 extracts event data (element data) used for calculating the degree of harassment from the above-described voice data (s4), and converts various element data into the evaluation information already described in FIGS. 4A to 4F. The evaluation values are determined by application, and the degree of harassment is calculated by applying these to the formula for calculating the degree of harassment (s5).

In the specific example shown in this embodiment, an NG word included in the utterance content indicated by the audio data, a subsequent word uttered corresponding to the NG word, emotion determined from the atmosphere of the voice, presence or absence of interruption in conversation, length of silence Since evaluation values for events such as conversation target (for individuals or for multiple people) can be obtained, these are substituted into the formula.

In that case, for example, a mathematical expression such as degree of harassment=((NG word+presence/absence of interruption+individual or plural)×emotion)+((following word+length of silence)×following emotion) is assumed. FIG. 7 shows examples of specific calculations for each of the following patterns: "NG words exist", "No NG words but people around them feel fear", and "Not harassment". For example, in the pattern of "with NG words", based on the values of each event of NG word, presence or absence of interruption, individual or plural, emotion, subsequent word, silence length, and subsequent emotion, in "Example 1", (( -0.7 - 0.1 - 0.1) x 1.5) + ((-0.5 - 0.1) x 1.5) = -2.25, etc. ing. Similarly, in the pattern of "not an NG word, but people around me feel fear", based on the values of each event: NG word, presence/absence of interruption, single or multiple, emotion, subsequent word, silence length, and subsequent emotion , in "Example 1", ((-0.1-0-0) x 1.5) + ((-0.5-0) x 1.5) = -0.9, etc. is calculated. Similarly, in the "not harassment" pattern, based on the values of each event of NG word, interruption presence, individual or multiple, emotion, subsequent word, silence length, and subsequent emotion, in "example 1", (( -0.1 - 0 - 0) x 0.1) + ((-0.1 - 0) x 0.1) = -0.02 and so on to calculate the degree of harassment.

The harassment degree calculation method described here is an example, and is not limited to this. For example, a technique based on machine learning can also be adopted. In that case, using a supporting vector machine (SVM) or neural network (NN), etc., these (evaluation value of each event) and the result (degree of harassment) are judged based on the model trained by the SVM or NN in advance. can also

Of course, unlike evaluation information that is generated and utilized across conference participants and their affiliated organizations (Figures 4A to 4F: standard model), it is generated and utilized individually for each conference participant and affiliated organization. It is also possible to assume a configuration in which evaluation information is assumed and a judgment model is learned based on this. Configuration and processing based on such an individual model will be described later. In such individual models, weights (evaluation values for each event) and thresholds for harassment judgments can be customized individually (individual user groups).

Subsequently, the server 10 determines whether the degree of harassment calculated in s5 is greater than the harassment determination criteria 126 (FIG. 4G) (s6).

As a result of the above determination, if the value of the degree of harassment is not equal to or greater than the value of the harassment criterion (s6: No), the server 10 terminates the process.

On the other hand, if the value of the degree of harassment is equal to or greater than the value of the harassment criterion (s6: Yes), feedback corresponding to the degree of harassment is generated (s7). In this feedback, in addition to a message indicating that harassment is taking place, a message warning about the degree of harassment and its danger is predetermined according to the degree of harassment.

Subsequently, the server 10 transmits the feedback value generated in s7 to the terminal 20 of the speaker (s8), and ends the process.
<Flow example: Conversation direction determination>
Details of the directionality determination (s1) in the flow of FIG. 6 will now be described with reference to FIG. Here, when determining whether the voice data includes a personal pronoun or a personal name, the server 10 extracts words included in the voice data using an existing voice recognition algorithm, and determines whether the word corresponds to a personal pronoun or a personal name. It is assumed that the identification is based on a dictionary (stored by the server 10 in the storage device 101) that defines personal pronouns and personal names.

The server 10 extracts one unprocessed record (for example, the n-th record in the input information table 128) from the input information table 128 (s20).

In addition, the server 10, for example, refers to the participant IDs included in each record of the input information table 128 (for each online conference), and counts the types thereof to determine the number of participants in this online conference. It is determined whether or not there are two or more participants in the conference (s21).

As a result of this determination, if the number of participants is not two or more (s21: No), the server 10 determines that the utterance indicated by the record n is a soliloquy (s22), and shifts the process to s32.

On the other hand, as a result of the above determination, if there are two or more participants (s21: Yes), the server 10 checks whether there is an unprocessed record n+1 registered at the next time in the input information table 128. is searched and determined (s23).

As a result of the above determination, if there is no unprocessed record n+1 (s23: No), the server 10 waits, for example, 10 seconds (s24) and returns the process to s23.

On the other hand, as a result of the above determination, if there is an unprocessed record (s23: Yes), the server 10 determines whether the message content of the nth record includes a personal name or a first person pronoun (s25).

As a result of the above determination, if the message indicated by the n-th record includes a personal name or a first-person pronoun (s25: Yes), the server 10 determines that the message is directed to an individual ( s26), and the process transitions to s32.

On the other hand, as a result of the above determination, if the message indicated by the n-th record does not include a personal name or a first person pronoun (s25: No), the server 10 determines that the message indicated by the n-th record does not include a second person pronoun. It is determined whether it is included (s27).

As a result of the above determination, if the utterance content indicated by the n-th record includes a second-person pronoun (s27: Yes), the server 10 identifies the utterance as directed to a plurality of people (s28), and proceeds to s32. transition.

On the other hand, as a result of the above determination, if the statement content indicated by the n-th record does not include a second-person pronoun (s27: No), the server 10 determines that a person other than the speaker of the n-th record will It is determined whether the speaker is speaking (s29).

As a result of this determination, if a person other than the speaker of the n-th record has spoken within a certain period of time (s29: Yes), the server 10 identifies the speech as directed to an individual or a plurality of people (s30), The processing is shifted to s32.

On the other hand, as a result of the above determination, if no person other than the speaker of the n-th record has spoken within a certain period of time (s29: No), the server 10 specifies that the direction of the conversation is unknown (s31). , the process goes to s32.

Finally, the server 10 sets a value in the utterance partner field of the record based on the utterance direction, that is, the utterance partner specified in the above-described processing (s32), and ends the processing.
<Updating the individual harassment model>
Here, a configuration in which the server 10 uses an individual harassment model will be described with reference to FIGS. 9 to 12. FIG. In this case, the server 10 has an individual harassment model update program 116 in addition to the configuration shown in FIG. This individual harassment model update program 116 implements a function of updating the evaluation information, that is, the individual harassment model, according to the determination result of the degree of harassment.

The server 10 also holds, in the information storage area 115, participant attribute information 129 shown in FIG. 10 and an individual harassment model definition update table 127 shown in FIG. Processing using these will be described later based on the flow of FIG.

The participant attribute information 129 is a table storing information about each participant of the online conference. Each record uses an ID such as an employee number that uniquely identifies a conference participant as a key, and includes the name, e-mail address, date of birth, hometown or current location, department to which the conference participant belongs, position, voice characteristics, are associated with values such as Such information can be assumed to be set in the server 10 by the organization to which the conference participants belong, for example.

Further, the table 127 for updating the definition of the individual harassment model defines contents for increasing or decreasing the evaluation value in each evaluation information of, for example, NG words and subsequent words, according to the attributes of the conference participants themselves or their relationships with their conversation partners. It will be More specifically, as shown in FIG. 11, for each target attribute and target word, a table defining the rate at which the evaluation value for the word is increased or decreased from the value in the general model shown in FIGS. 4A to 4F. be.

FIG. 12 shows an example flow of a harassment prevention method based on this premise. Here, only differences from the flow already shown based on FIG. 6 will be explained.

In this case, through steps s40 and s41, the server 10 determines each conference participant appearing in the conversation, that is, the speaker and the target person to whom the speech is directed, from the participant attribute information 129 and the voice data (s42). ). For example, in the input information table 128, among the records that are chronologically continuous and have different participant IDs, the target person is determined from the record regarding the person who spoke in response to one of the utterances. can be identified. Of course, in such a method, it is also conceivable that the conference participants are determined by voice recognition from the voice data, and the speaker and the target person who is the receiver are specified based on the chronological order of the conversation between them.

In addition, the server 10 uses the attribute information of the speaker and the subject to whom the utterance is directed from the participant attribute information 129 through the above-described step s42 and the subsequent s43 as element data, that is, as a variable in the calculation formula of the degree of harassment. (s44).

Subsequently, the server 10 applies the event data (element data) obtained in step s44 to the evaluation information already described in FIGS. to calculate the degree of harassment (s45).

If the value of the degree of harassment calculated here is equal to or greater than the value of the harassment judgment criteria 126 (s46: Yes), the individual harassment model, that is, various evaluation information for the conference participants (word evaluation information (NG word) 120 to conversation object The evaluation value in the evaluation information 125) is updated by increasing or decreasing based on the definition update table 127 of the individual harassment model. Of course, as the value of the degree of harassment increases, the rate of change defined in the definition update table 127 of the individual harassment model may be further increased before updating the evaluation value.

By operating in this way, unlike evaluation information generated and utilized across conference participants and their affiliated organizations (Figures 4A to 4F: standard model), individual evaluation information for each conference participant and affiliated organization It is also possible to envisage a configuration in which evaluation information to be generated and utilized is assumed, and a judgment model is learned based on this. In other words, in such an individual model, the weight (evaluation value of each event) and the threshold for harassment determination can be customized individually (individual user groups).
<Regarding sensitivity to harassment>
Here, when a conference participant in an online conference recognizes a harassing act in response to another's remarks, an operation will be described in which he/she can declare that fact on the screen of the online conference tool. This reporting is carried out by pressing the harassment notification button 1504 on the terminal 20 operated by the reporting person, for example, on the screen 1500 of the web meeting tool as shown in FIG.

The server 10 receives the press data of the harassment notification button 1504 on the terminal 20 through the network I/F 106 and detects it by the function of the input processing program 110 . In addition, the server 10 uses the function of the individual harassment susceptibility learning program 117, based on the timing of pressing the harassment notification button 1504, and based on the most recently obtained voice data (from the online conference), NG words that can be estimated to be harassment targets. is specified, and based on this, each evaluation information, which is an individual harassment model, is updated. Therefore, the hardware configuration example of the server 10 in this case further includes an individual harassment tendency learning program 117 as a program held in the storage device 101, as shown in FIG. Other configurations are the same as those of the server 10 that has already been described.

A flow example in this case is shown in FIG. The server 10 determines who the speaking partner is (that is, the direction of the conversation) from the voice features and voice data of the participant attribute information 129 (described in detail in FIG. 10) (s50).

In addition, the server 10 extracts data (element data) of each event used for calculating the degree of harassment from the time when the harassment notification button 1504 was pressed or the voice data (record of the input information table 128) immediately before that (s51). ).

Subsequently, the server 10 acquires the attribute information of the speaker and the subject to whom the speech is directed from the participant attribute information 129 as element data (s52).

In addition, the server 10 advances learning of a learning model in a predetermined machine learning engine (those held or available by the server 10) using the element data obtained up to s52 as learning data (s53).

This learning model is a model in which the evaluation values in various evaluation information for the conference participants are increased or decreased and updated to suitable values. More specifically, for example, when the harassment notification button 1504 is pressed more than a certain frequency for a certain NG word within a certain period of time, the NG word in the word evaluation information (NG word) 129 until then is updated by increasing the evaluation value by a predetermined percentage. Alternatively, if conference participants with a common attribute press the harassment notification button 1504 with respect to the same NG word more than a certain frequency, word evaluation information (NG Word) 120 also increases the evaluation value of the NG word.

In addition, the server 10 applies the element data obtained in s52 to the corresponding evaluation information (updated in s53) to obtain an evaluation value, and substitutes it into the calculation formula to calculate the degree of harassment. do. The server 10 generates feedback according to the degree of harassment obtained here (s54). In this feedback, in addition to a message indicating that harassment is taking place, a message warning about the degree of harassment and its danger is predetermined according to the degree of harassment.

Subsequently, the server 10 transmits the feedback (screen 1500 in FIG. 15) generated in s54 to the speaker's terminal 20 (s55), and ends the process.

A screen 1500 shown in FIG. 15 includes an interface for displaying a list of conference participants' face images 1501, own face images 1502, and feedback 1503 described above, in addition to the harassment notification button 1504 already described. When a conference participant feels a mental burden or the like due to the remarks of others, the conference participant presses the harassment notification button 1504 and reports to the server 10 . In response to this, the server 10 identifies the utterance of the NG word immediately before the declaration, and executes the flow described above.

According to this, when a conversation participant individually feels harassment, a notification is received directly, and based on the utterance that is the target act and its content, etc., it is applied to the above-mentioned conversation participant and affiliated organization Evaluation information can be updated as individual items. As a result, it becomes possible to more effectively prevent harassment that occurs in a remote work environment.

Although the best mode for carrying out the present invention has been specifically described above, the present invention is not limited to this, and can be variously modified without departing from the scope of the invention.

According to this embodiment, for example, the perpetrator of a harassment act can be notified in real time of a sign of harassment in a conversation, and can obtain an opportunity to restrain himself before a problem occurs. Of course, if you have unintentionally made other people feel uncomfortable, you can quickly notice that fact.

On the other hand, from the harassment victim's point of view, the system warns the perpetrator of the harassment act on behalf of the person concerned. can solve the situation. Therefore, it leads to lowering the mental threshold for reporting actions related to harassment behavior.

In addition, for an organization that intends to prevent harassment behavior, this system creates an environment in which warnings are automatically issued. Furthermore, it is easy to detect the existence of harassment behavior without waiting for a report by the person concerned, and to be able to deal with the problem before it becomes serious.

In other words, it becomes possible to effectively prevent harassment that occurs in a remote work environment.

At least the following will be clarified by the description of this specification. That is, in the harassment prevention system of this embodiment, the computing device may distribute the calculation result of the harassment level to the terminal of the conversation participant or the organization to which the conversation participant belongs.

According to this, the existence of harassment that occurs in an online meeting in a remote work environment can be clearly recognized among the meeting participants, and can trigger reflection and countermeasures. As a result, it becomes possible to more effectively prevent harassment that occurs in a remote work environment.

Further, in the harassment prevention system of the present embodiment, the storage device stores, as the evaluation information, words included in a conversation, emotions indicated by voices of the conversation, interruptions in the conversation, length of silence, the number of utterance targets, and the calculating device extracts data corresponding to each of the plurality of events from the voice data when determining the degree of impact. and collate it with the evaluation information to determine the degree of impact for each of the plurality of events.

According to this, important elements of conversations in an online meeting environment can be grasped from multiple angles and accurately, and reflected in the harassment judgment logic, so that the existence of harassment can be supplemented with good accuracy. As a result, it becomes possible to more effectively prevent harassment that occurs in a remote work environment.

Further, in the harassment prevention system of the present embodiment, when specifying the number of participants in the conversation and the utterance target, the computing device extracts a personal name or personal pronoun, which is a vocabulary indicating the conversation participant in the voice data, The number of participants in the conversation and the utterance target may be specified based on the presence or absence and number of the personal name or personal pronoun.

According to this, it is possible to efficiently and accurately identify conversation participants and utterance targets, and the efficiency and accuracy of harassment determination can be easily improved. As a result, it becomes possible to more effectively prevent harassment that occurs in a remote work environment.

Further, in the harassment prevention system of the present embodiment, the storage device further holds attribute information of the conversation participant and adjustment information of the degree of influence based on the attribute of the conversation participant, and the arithmetic device is, when determining the degree of influence, if the number of participants in the conversation is two or more, the attribute information of each of the speaker identified as the utterance target and the receiver of the utterance is extracted from the storage device, and the attribute information may increase or decrease the degree of impact on the predetermined event based on the adjustment information, based on at least one of the social standpoints and regional characteristics of the speaker and the receiver indicated by .

According to this, even if the same word is used, it can be expected to be effective in improving the accuracy of harassment determination by appropriately coping with the situation where the way of receiving is greatly different depending on the region of origin or social position. As a result, it becomes possible to more effectively prevent harassment that occurs in a remote work environment.

Further, in the harassment prevention system of the present embodiment, as a result of calculating the harassment level based on the degree of influence that has been increased or decreased, if the calculated harassment level is equal to or higher than a predetermined standard, With respect to a predetermined event involved in the calculation of the harassment level, the evaluation information is updated by increasing or decreasing the degree of influence according to a predetermined rule so as to be a value unique to each of the conversation participants or to the organization to which the conversation participant belongs. It may be said that it is a thing.

According to this, in response to the fact that the degree of impact of events may change and the results of harassment judgments may vary depending on the attributes of conversation participants and individual circumstances, evaluation information to be applied to the conversation participants and the organization to which they belong can be made individually. can be updated as As a result, it becomes possible to more effectively prevent harassment that occurs in a remote work environment.

Further, in the harassment prevention system of the present embodiment, when a harassment report is received from a conversation participant's terminal during a conversation, the computing device extracts the vocabulary from voice data at or immediately before the harassment report is received. to identify the number of participants in the conversation and the target of utterance based on the vocabulary, extract data corresponding to the various events from the voice data, and extract the data corresponding to the various events including the number of participants in the conversation and the target of utterance. By executing regression analysis or machine learning using the corresponding data as an explanatory variable and the content of the harassment report as an objective variable, the degree of influence of the various events in the evaluation information is evaluated by the conversation participant who made the harassment report. It may be assumed that it is to be updated for individuals.

According to this, when a conversation participant individually feels harassment, a notification is received directly, and based on the utterance that is the target act and its content, etc., it is applied to the above-mentioned conversation participant and affiliated organization Evaluation information can be updated as individual items. As a result, it becomes possible to more effectively prevent harassment that occurs in a remote work environment.

Further, in the harassment prevention system of the present embodiment, when receiving the harassment report, the computing device generates feedback content according to the content of the harassment report, and converts the feedback content into a conversation with the executor of the harassment. It may be assumed that the information is notified to the terminal of the participant.

According to this, by immediately giving a warning to the person involved in the harassment act, it can be expected that the person's behavior will be changed immediately and in the future. As a result, it becomes possible to more effectively prevent harassment that occurs in a remote work environment.

Further, in the harassment prevention method of the present embodiment, the information processing device may distribute the calculation result of the harassment level to the terminal of the conversation participant or the organization to which the conversation participant belongs.

Further, in the harassment prevention method of the present embodiment, the information processing device stores, in the storage device, as the evaluation information, words included in a conversation, emotions indicated by voices in the conversation, interruptions in the conversation, length of silence, and the number of utterance target persons, and holds data that defines the degree of impact on harassment for each of a plurality of events, and data corresponding to each of the plurality of events from the voice data when determining the degree of impact may be extracted and compared with the evaluation information to determine the degree of impact for each of the plurality of events.

Further, in the harassment prevention method of the present embodiment, when the information processing device specifies the number of participants in the conversation and the utterance target, it extracts personal names or personal pronouns, which are vocabulary indicating conversation participants in the voice data. , the number of participants in the conversation and the utterance target may be specified based on the presence or absence and number of the personal name or personal pronoun.

Further, in the harassment prevention method of the present embodiment, the information processing device further retains attribute information of the conversation participant and adjustment information of the degree of influence based on the attribute of the conversation participant in the storage device, When determining the degree of influence, if the number of participants in the conversation is two or more, the attribute information of each of the speaker identified as the utterance target and the receiver of the utterance is extracted from the storage device and indicated by the attribute information. , based on at least one of social standpoints and regional characteristics of the speaker and the receiver, the degree of influence of the predetermined event may be increased or decreased based on the adjustment information.

Further, in the harassment prevention method of the present embodiment, when the information processing device calculates the harassment level based on the degree of influence that has been increased or decreased, and the calculated harassment level is equal to or higher than a predetermined standard, With respect to a predetermined event involved in the calculation of the harassment level, updating the evaluation information by increasing or decreasing the degree of influence according to a predetermined rule so that the value is unique to each of the conversation participants or to the organization to which the conversation participant belongs. You can do it.

Further, in the harassment prevention method of the present embodiment, when the information processing device receives a harassment report from a conversation participant's terminal during a conversation, the vocabulary is extracted from the voice data at or immediately before receiving the harassment report. Extracting, specifying the number of participants in the conversation and the target of utterance based on the vocabulary, extracting data corresponding to the various events from the voice data, and extracting the various events including the number of participants in the conversation and the target of utterance By performing regression analysis or machine learning with the data corresponding to the harassment report as an explanatory variable and the content of the harassment report as an objective variable, the degree of influence of the various events in the evaluation information is determined by the conversation participation that made the harassment report. It may be updated for individuals.

Further, in the harassment prevention method of the present embodiment, when the information processing device receives the harassment report, it generates feedback content according to the content of the harassment report, and sends the feedback content to the harassment executor. It may be notified to the terminals of the conversation participants.

1 harassment prevention system 5 network 10 server (harassment prevention device)
101 storage device 103 memory 104 arithmetic device 105 input/output I/F
106 Network I/F
110 Input processing program 111 Conversation direction determination program 112 Harassment determination program 113 Feedback generation program 114 Output processing program 115 Information storage area 116 Individual harassment model update program 117 Individual harassment sensitivity learning program 120 Word evaluation information (NG word)
121 word evaluation information (subsequent word)
122 Emotion evaluation information 123 Interruption evaluation information 124 Silence length evaluation information 125 Conversation object evaluation information 126 Harassment criteria 127 Individual harassment model definition update table 128 Input information table 129 Participant attribute information 20 Terminal 201 Storage device 203 Memory 204 Calculation Device 205 Input I/F
206 Output I/F
207 Network I/F
210 voice data acquisition program 211 image data acquisition program 212 input data acquisition program 213 harassment button input detection program 214 data transmission program 215 information storage area

Claims (10)

a storage device that holds evaluation information that defines the degree of influence of various events accompanying conversation on harassment, and a formula for calculating a harassment level using the degree of influence of the various events as a variable;
A process of acquiring voice data of utterances by conversation participants from a predetermined device, extracting predetermined vocabulary indicating conversation participants from the voice data, and specifying the number of participants in the conversation and the utterance target based on the vocabulary, and the voice data Extracting data corresponding to the various events from the data, comparing it with the evaluation information, determining the degree of impact related to the various events, and applying the determined degree of impact to the calculation formula to calculate the harassment level. a computing device that performs processing;
A harassment prevention system comprising: The computing device is
Distributing the calculation result of the harassment level to the terminal of the conversation participant or the organization to which the conversation participant belongs,
The harassment prevention system according to claim 1, characterized by: The storage device
As the evaluation information, the degree of impact on harassment for each of at least one or more of the words contained in the conversation, the emotion indicated by the voice of the conversation, the interruption of the conversation, the length of silence, and the number of utterance targets. keep what you stipulate,
The computing device is
When determining the degree of impact, data corresponding to each of the plurality of events is extracted from the audio data and compared with the evaluation information to determine the degree of impact for each of the plurality of events.
The harassment prevention system according to claim 1, characterized by: The computing device is
When specifying the number of participants in the conversation and the utterance target, personal names or personal pronouns, which are vocabulary indicating the participants in the conversation, are extracted from the audio data, and the number of participants in the conversation is based on the presence or absence and number of the personal names or personal pronouns. and to specify the utterance target,
The harassment prevention system according to claim 1, characterized by: The storage device
Further holding attribute information of the conversation participant and adjustment information of the degree of influence based on the attribute of the conversation participant,
The computing device is
When determining the degree of influence, if the number of participants in the conversation is two or more, the attribute information of each of the speaker identified as the utterance target and the receiver of the utterance is extracted from the storage device and indicated by the attribute information. , based on at least one of the social standpoints and regional characteristics of the speaker and the receiver, and increasing or decreasing the degree of impact on the predetermined event based on the adjustment information;
The harassment prevention system according to claim 1, characterized by: The computing device is
As a result of calculating the harassment level based on the increased or decreased degree of influence, if the calculated harassment level is equal to or higher than a predetermined standard, the conversation participant with respect to a predetermined event involved in the calculation of the harassment level or to update the evaluation information by increasing or decreasing the degree of influence according to a predetermined rule so that it becomes a value specific to the organization to which the conversation participant belongs,
The harassment prevention system according to claim 5, characterized by: The computing device is
When a harassment report is received from a conversation participant's terminal during a conversation, the vocabulary is extracted from the voice data at or immediately before the harassment report is received, and the number of conversation participants and the utterance target are calculated based on the vocabulary. data corresponding to the various events are extracted from the voice data, the data corresponding to the various events including the number of participants in the conversation and the utterance target are used as explanatory variables, and the content of the harassment report is used as the objective variable By performing regression analysis or machine learning as described above, the degree of impact on the various events in the evaluation information is updated for the individual conversation participant who made the harassment report.
The harassment prevention system according to claim 1, characterized by: The computing device is
When the harassment report is received, feedback content is generated according to the content of the harassment report, and the feedback content is notified to the terminal of the conversation participant who is the executor of the harassment.
The harassment prevention system according to claim 7, characterized by: The information processing device
holding in a storage device evaluation information that defines the degree of influence on harassment of various events that accompany the conversation, and a formula for calculating the harassment level using the degree of influence of the various events as a variable;
A process of acquiring voice data of utterances by conversation participants from a predetermined device, extracting predetermined vocabulary indicating conversation participants from the voice data, and specifying the number of participants in the conversation and the utterance target based on the vocabulary, and the voice data Extracting data corresponding to the various events from the data, comparing it with the evaluation information, determining the degree of impact related to the various events, and applying the determined degree of impact to the calculation formula to calculate the harassment level. perform processing,
A harassment prevention method characterized by: The information processing device
Distributing the calculation result of the harassment level to the terminal of the conversation participant or the organization to which the conversation participant belongs;
The harassment prevention method according to claim 9, characterized by:

Images