Detection device, detection method, and program

Abstract

A detection device according to one aspect of the present disclosure detects that a specific situation has occurred between a first person and a second person, the detection device comprising: a creation unit that, on the basis of text expressing an utterance made by at least one of the first person and the second person, creates a command statement including a command indicating that an assessment is to be made in regard to whether the specific situation has occurred; an assessment unit that, on the basis of the command statement, assesses whether the specific situation has occurred by using a trained machine learning model; and a notification unit that, when it is assessed that the specific situation has occurred, notifies a prescribed notification recipient that the specific situation has occurred.

Description

Detection device, detection method, and program

[0001] The present disclosure relates to a detection device, a detection method, and a program.

[0002] For contact centers (which may also be called "call centers"), there are known technologies for detecting keywords from the results of speech recognition performed during phone calls and technologies for recognizing customer emotions (see, for example, Non-Patent Document 1). These technologies make it possible to detect the occurrence of certain situations (e.g., the occurrence of customer harassment, the occurrence of compliance violations, etc.).

[0003] ForeSight Voice Mining, Internet <URL: https: / / www.ntt-tx.co.jp / products / foresight_vm / >

[0004] However, conventional techniques may not be able to accurately detect the occurrence of a particular situation.

[0005] The present disclosure has been made in consideration of the above points, and aims to accurately detect the occurrence of a specific situation.

[0006] A detection device according to one aspect of the present disclosure is a detection device that detects the occurrence of a specific situation between a first person and a second person, and includes: a creation unit that creates an instruction document including an instruction to determine whether or not the specific situation has occurred based on text representing an utterance of at least one of the first person and the second person; a determination unit that determines whether or not the specific situation has occurred using a trained machine learning model based on the instruction document; and a notification unit that, if it is determined that the specific situation has occurred, notifies a specified notification destination that the specific situation has occurred.

[0007] The occurrence of a particular situation can be detected with high accuracy.

[0008] 1 is a diagram illustrating an example of the overall configuration of a contact center system according to the present embodiment; FIG. 2 is a diagram illustrating an example of the functional configuration of a detection device according to the present embodiment; FIG. 3 is a flowchart illustrating an example of the operation of a detection device according to the present embodiment; FIG. 4 is a diagram illustrating an example (part 1) of a prompt template; FIG. 5 is a diagram illustrating an example (part 2) of a prompt template; FIG. 6 is a diagram illustrating an example (part 1) of output information; and FIG. 7 is a diagram illustrating an example (part 2) of output information.

[0009] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings.

[0010] <Background and Issues with Prior Art> In recent years, inappropriate verbal and physical behavior by customers and operators, such as customer harassment by customers and remarks by operators that violate compliance, has become a problem at contact centers. Taking appropriate follow-up and countermeasures against such inappropriate verbal and physical behavior will lead to maintaining a healthy working environment for contact center employees and improving the quality of customer service.

[0011] For example, by using the technology described in Non-Patent Document 1, etc., it is possible to detect the occurrence of a specific situation (e.g., the occurrence of customer harassment, the occurrence of compliance violations, etc.). For example, by detecting certain keywords (e.g., keywords representing inappropriate remarks that constitute customer harassment) from the results of speech recognition of utterances during a call, the occurrence of customer harassment can be detected. Furthermore, for example, by recognizing certain emotions of customers (e.g., emotions such as anger), the occurrence of customer harassment can be detected.

[0012] However, conventional technologies may not be able to accurately detect the occurrence of certain situations. For example, even if a certain keyword indicating an inappropriate remark is detected, it does not necessarily mean that customer harassment is occurring, depending on the context. For example, even if the keyword "kill" indicating an inappropriate remark is detected, the customer may be talking about "insecticide." Also, for example, even if emotion recognition recognizes that a customer is angry, it does not necessarily mean that customer harassment is occurring.

[0013] Therefore, the following describes a contact center system 1 that can accurately detect the occurrence of customer harassment by assuming the occurrence of customer harassment as a specific situation and using generative AI (artificial intelligence). Generative AI (also sometimes referred to as "generative AI") is a trained machine learning model that receives commands called prompts as input and generates and outputs information according to the prompts. There are various machine learning models that can realize generative AI, but machine learning models called large language models (LLMs) are often used. Note that prompts may also be called "instructions," for example. However, the present invention is not limited to generative AI. Instead of generative AI, a model that adapts a pre-trained model such as BERT (Bidirectional Encoder Representations from Transformers) for detecting customer harassment may also be used.

[0014] However, the occurrence of a specific situation is not limited to the occurrence of customer harassment. For example, the occurrence of a specific situation may be the occurrence of a compliance violation or the occurrence of any other negative situation.

[0015] <Example of Overall Configuration of Contact Center System 1> An example of the overall configuration of the contact center system 1 according to this embodiment will be described with reference to Fig. 1. Fig. 1 is a diagram showing an example of the overall configuration of the contact center system 1 according to this embodiment.

[0016] 1, the contact center system 1 according to this embodiment includes a detection device 10, one or more operator terminals 20, one or more supervisor terminals 30, a PBX (Private Branch eXchange) 40, a connection control device 50, and a customer terminal 60. The detection device 10, each operator terminal 20, each supervisor terminal 30, the PBX 40, and the connection control device 50 are installed in a contact center environment E, which is a system environment of a contact center. Note that the contact center environment E is not limited to a system environment within the same building, and may be, for example, a system environment within multiple geographically separated buildings.

[0017] The detection device 10 is a server or a group of servers that detects the occurrence of customer harassment (an example of a specific situation) when it occurs in a call between an operator and a customer.

[0018] The operator terminals 20 are various terminals such as PCs (personal computers) used by operators whose job it is to answer calls with customers. Each operator terminal 20 is assumed to have installed therein a program or the like that causes the operator terminal 20 to function as an IP (Internet Protocol) telephone.

[0019] The supervisor terminal 30 is a terminal such as a PC used by a supervisor whose job is to monitor, supervise, or support operators. Usually, there is one supervisor for several to a dozen operators. Each supervisor terminal 30 is assumed to have installed therein a program or the like that enables the supervisor terminal 30 to function as an IP telephone.

[0020] The PBX 40 is a telephone exchange (IP-PBX) and is connected to a communication network 70 including a Voice over Internet Protocol (VoIP) network and a Public Switched Telephone Network (PSTN).

[0021] The connection control device 50 transfers a connection request from a customer terminal 60 to one of the one or more operator terminals 20. Furthermore, when the operator terminal 20 responds to the connection request with a connection response, the connection control device 50 establishes a connection between the customer terminal 60 and the operator terminal 20. This enables a call between the customer terminal 60 and the operator terminal 20. The connection control device 50 may have, for example, an interactive voice response (IVR) function.

[0022] When transferring a connection request from a customer terminal 60 to an operator terminal 20, the connection control device 50 determines the operator terminal 20 to which the connection request is to be transferred in accordance with transfer settings in which transfer rules, etc. Examples of transfer rules include randomly determining the operator terminal 20 to be the transfer destination from among operator terminals 20 that are not currently in a call, or determining the operator terminal 20 to be the transfer destination in accordance with the content of an inquiry entered via automated voice response.

[0023] The connection control device 50 also relays packets (e.g., packets containing audio data, video data, etc.) between the operator terminal 20 and the PBX 40 , and also captures and transmits the packets to the detection device 10 .

[0024] The customer terminal 60 is a terminal such as a PC, a smartphone, a mobile phone, or a landline phone used by a customer.

[0025] 1 is an example and is not intended to be limiting. For example, telephones (such as fixed IP telephones or mobile IP telephones) used by operators and supervisors may be present in the contact center environment E. Furthermore, for example, the functions of the PBX 40 may be realized by a cloud service or the like.

[0026] <Example of Functional Configuration of Detection Device 10> An example of the functional configuration of the detection device 10 according to this embodiment will be described with reference to Fig. 2. Fig. 2 is a diagram showing an example of the functional configuration of the detection device 10 according to this embodiment.

[0027] As shown in FIG. 2 , the detection device 10 according to this embodiment includes a speech recognition unit 101, an emotion recognition unit 102, an utterance information creation unit 103, a judgment instruction unit 104, a situation judgment unit 105, a judgment result information creation unit 106, and a notification unit 107. These functional units are implemented, for example, by one or more programs installed in the detection device 10, which are executed by a computing device such as a central processing unit (CPU). The detection device 10 according to this embodiment also includes an utterance information storage unit 108, a template storage unit 109, and a judgment result information storage unit 110. These storage units are implemented, for example, by storage areas of storage devices such as hard disk drives (HDDs), solid state drives (SSDs), flash memory, etc. At least one of these storage units may be implemented by a storage area of ​​a storage device (e.g., a storage device provided in a database server, etc.) communicatively connected to the detection device 10.

[0028] The speech recognition unit 101 performs speech recognition processing on the speech data contained in the packets transmitted from the connection control device 50, and creates time-information-attached speech text for each speaker (i.e., the customer's time-attached speech text and the operator's time-attached speech text). The speech text is a text that represents the speech recognition results for each utterance. For simplicity, the term "time-attached" will be omitted below, and the time-attached speech text will also be simply referred to as "speech text." The speech recognition unit 101 may create the speech text using known speech recognition technology.

[0029] The emotion recognition unit 102 performs emotion recognition processing on the voice data included in the packets transmitted from the connection control device 50, and creates emotion information representing the emotion of the speaker of each utterance. The emotion recognition unit 102 may perform emotion recognition processing using, in addition to the voice data, for example, the speech text created by the voice recognition unit 101. Emotion information may represent various emotions, such as "anger," "sadness," "joy," "fear," and "surprise." However, for simplicity, these emotions are categorized into three categories: "intimidating," "friendly," and "neutral," and the emotion information is assumed to represent "intimidating," "friendly," or "neutral." The emotion recognition unit 102 may create emotion information using known emotion recognition technology.

[0030] The utterance information creation unit 103 creates information including an utterance text and emotional information corresponding to the utterance text (i.e., emotional information that expresses the emotion of the speaker of the utterance expressed by the utterance text) as utterance information, and stores the information in the utterance information storage unit 108. In addition to the utterance text and emotional information, the utterance information also includes, for example, an utterance ID, which is an example of identification information that identifies the utterance information. That is, the utterance information is expressed in a format such as (utterance ID, utterance text, emotional information).

[0031] The judgment instruction unit 104 uses the utterance information stored in the utterance information storage unit 108 and the template stored in the template storage unit 109 to create a prompt to be input to the generation AI for determining whether customer harassment is occurring. Here, the template refers to a prompt template. The prompt is created by specifying or setting the utterance text and emotional information included in the utterance information as input sentences (input information) for the template, which describes a command statement for causing the generation AI to determine whether customer harassment is occurring.

[0032] The situation determination unit 105 determines whether or not customer harassment is occurring in accordance with the prompt created by the determination instruction unit 104. Here, the situation determination unit 105 is realized by a process that a program or module that realizes the generation AI causes a computing device to execute.

[0033] The determination result information creation unit 106 creates information including the determination result by the situation determination unit 105 as determination result information, and stores it in the determination result information storage unit 110. In addition to the determination result, the determination result information includes, for example, a determination result ID, which is an example of identification information for identifying the determination result information, and a set of utterance IDs (hereinafter also referred to as an "utterance ID set") included in the utterance information used in the determination by the situation determination unit 105. That is, the determination result information is expressed in a format such as (determination result ID, determination result, utterance ID set), for example.

[0034] The notification unit 107 determines whether notification is necessary and the notification destination (the operator terminal 20 or the supervisor terminal 30) using the determination result by the situation determination unit 105. For example, when a determination result indicating that customer harassment is occurring in a call being handled by a certain operator is obtained, the notification unit 107 determines that "notification is necessary" and determines the supervisor terminal 30 used by a supervisor who monitors the operator, etc., as the notification destination.

[0035] Furthermore, if it is determined that a notification is necessary, the notification unit 107 transmits a predetermined notification content to the determined notification destination.

[0036] The utterance information storage unit 108 stores the utterance information created by the utterance information creation unit 103. The template storage unit 109 stores templates (prompt templates) created in advance. The judgment result information storage unit 110 stores the judgment result information created by the judgment result information creation unit 106.

[0037] <Example of Operation of the Detection Device 10> An example of operation of the detection device 10 according to this embodiment will be described with reference to FIG. 3. FIG. 3 is a flowchart illustrating an example of operation of the detection device 10 according to this embodiment. It is assumed below that a call is taking place between a certain operator terminal 20 and a certain customer terminal 60. In this case, the following steps S101 to S109 may be repeatedly executed at predetermined time intervals, such as several seconds or several tens of seconds, or each time a certain amount of voice data contained in packets transmitted from the connection control device 50 is accumulated. Alternatively, for example, after a time-information-attached utterance text is created in step S101, steps S102 to S109 may be repeatedly executed for each sentence, utterance (delimiter), conversation (e.g., question, response), or other predetermined time unit or data volume unit in the utterance text.

[0038] The speech recognition unit 101 performs speech recognition processing on the speech data contained in the packet transmitted from the connection control device 50, and creates a speech text with time information for each speaker (step S101).

[0039] The emotion recognition unit 102 performs emotion recognition processing on the voice data contained in the packets transmitted from the connection control device 50, and creates emotion information representing the emotion of the speaker of each utterance (step S102).

[0040] The utterance information creation unit 103 uses the utterance text created in step S101 above and the emotion information created in step S102 above to create utterance information including the utterance text and the emotion information corresponding to the utterance text, and stores the information in the utterance information storage unit 108 (step S103).

[0041] The determination instruction unit 104 creates a prompt using the utterance information stored in the utterance information storage unit 108 and the template stored in the template storage unit 109 (step S104). That is, the determination instruction unit 104 creates a prompt by specifying or setting the utterance text and emotional information included in the utterance information as an input sentence (input information) for the template.

[0042] Example of a prompt template (part 1) An example of a prompt template (part 1) will be described with reference to Fig. 4. Fig. 4 is a diagram showing an example of a prompt template (part 1).

[0043] The template 1000 shown in FIG. 4 includes a command statement 1100, an output format 1200, and an input statement 1300.

[0044] Command statement 1100 is a command statement for causing the generation AI to determine whether customer harassment is occurring. Command statement 1100 describes a command statement to determine whether an input sentence corresponds to customer harassment, and to output the determination result and the items corresponding to customer harassment (i.e., the basis for the determination result) according to a format.

[0045] The output format 1200 is a format for outputting the determination result indicating whether or not something constitutes customer harassment, and the items that constitute customer harassment. The output format 1200 describes that the determination result is output as either "applicable" or "not applicable," and that "inappropriate content," "mental attacks," "intimidating behavior," "discriminatory behavior," and "sexual behavior" are output as "Yes" or "No."

[0046] Information to be determined as to whether it constitutes customer harassment is described in input sentence 1300. For example, utterance text (a customer's utterance text) to be determined as to whether it constitutes customer harassment and emotion information corresponding to the utterance text are set or specified in input sentence 1300.

[0047] A prompt is created by setting or specifying the customer's utterance text and emotion information corresponding to the utterance text for the input sentence 1300 of the template 1000 shown in FIG.

[0048] Example of Prompt Template (Part 2) An example of a prompt template (Part 2) will be described with reference to Fig. 5. Fig. 5 is a diagram showing an example of a prompt template (Part 2).

[0049] 5 includes a command statement 2100, an output format 2200, and an input statement 2300. This command statement causes the generation AI to determine whether customer harassment is occurring. The command statement 2100 determines whether an input statement constitutes customer harassment, and outputs the determination result, items that constitute customer harassment, and a summary of the occurrence of customer harassment (i.e., a summary of the input statement) according to a format.

[0050] Output format 2200 is a format for outputting the determination result indicating whether or not something constitutes customer harassment, the items that constitute customer harassment, and a summary of the occurrence of customer harassment. In addition to outputting the determination result as either "applicable" or "not applicable," output format 2200 describes how to output "yes" or "no" for "inappropriate content," "mental attacks," "intimidating behavior," "discriminatory behavior," and "sexual behavior," and how to output the occurrence of customer harassment.

[0051] Input sentence 2300 describes information to be determined as to whether it constitutes customer harassment. For example, a sequence of spoken text to be determined as to whether it constitutes customer harassment and emotional information corresponding to the spoken text are set or specified in input sentence 2300. The sequence of spoken text and emotional information corresponding to the spoken text may be, for example, a sequence of spoken text and emotional information from the start of the call to the present, or a sequence of spoken text and emotional information for a recent predetermined time period (e.g., a time period from one minute ago to the present). Furthermore, the sequence of spoken text and emotional information may be a sequence of spoken text and emotional information from a customer, or a sequence of spoken text and emotional information from both a customer and an operator.

[0052] A prompt is created by setting or specifying a sequence of spoken text and emotion information corresponding to that spoken text for an input sentence 2300 of the template 2000 shown in FIG.

[0053] Returning to the explanation of FIG. 3 , following step S104, the situation determination unit 105 uses the prompt created in step S104 to determine whether customer harassment is occurring (step S105). That is, the situation determination unit 105 inputs the prompt created in step S104 to the generation AI, causing the generation AI to determine whether customer harassment is occurring. As a result, the generation AI determines whether customer harassment is occurring according to the prompt, and outputs information according to the output format described in the prompt. Hereinafter, the information output according to the output format described in the prompt will also be referred to as "output information."

[0054] Example of Output Information (Part 1) An example of output information when a prompt created from the template 1000 shown in Fig. 4 is input to the generation AI will be described with reference to Fig. 6. Fig. 6 is a diagram showing an example of output information (Part 1).

[0055] 6 is information output from the generation AI in accordance with the output format 1200. The output information 3000 includes a determination result 3100 indicating whether or not customer harassment has occurred, and a value 3200 of an item corresponding to customer harassment.

[0056] Example of Output Information (Part 2) An example of output information when a prompt created from the template 2000 shown in Fig. 5 is input to the generation AI will be described with reference to Fig. 7. Fig. 7 is a diagram showing an example of output information (Part 2).

[0057] 7 is information output from the generation AI in accordance with the output format 2200. The output information 4000 includes a determination result 4100 indicating whether or not customer harassment has occurred, values ​​4200 for each item corresponding to customer harassment, and a summary 4300 indicating the occurrence status of customer harassment. Note that the summary 4300 is, for example, a summary of the series of utterance text set or specified in the input sentence 2300 (or a summary that takes into account the series of emotional information for that summary).

[0058] Returning to the description of Fig. 3 , following step S105, the determination result information creation unit 106 creates determination result information using the determination result in step S105 (or output information including the determination result) and stores the created determination result information in the determination result information storage unit 110 (step S106). Note that the determination result information creation unit 106 may create determination result information in a format such as (determination result ID, determination result, utterance ID set) or the like, or may create determination result information in a format such as (determination result ID, output information, utterance ID set) or the like using the output information.

[0059] The notification unit 107 determines whether or not a notification is necessary and who to notify using the determination result in step S105 (or output information including the determination result) (step S107). For example, if the determination result indicating whether or not customer harassment has occurred is "yes," the notification unit 107 determines whether or not a notification is necessary and who to notify, such as "need for notification" and "notification destination" as follows: "need for notification" = "notification required" and "destination of notification" = "supervisor terminal 30" used by a supervisor who monitors the operator.

[0060] The notification unit 107 determines whether or not it has been determined in step S107 that notification is necessary (step S108).

[0061] If it is determined in step S108 that notification is necessary, the notification unit 107 transmits a predetermined notification content (e.g., notification content including output information) to the notification destination determined in step S107 (step S109). This detects the occurrence of customer harassment. That is, for example, a supervisor monitoring the operator can learn that customer harassment is occurring from the output information included in the notification content. This allows the supervisor to take appropriate action, such as providing appropriate advice to the operator or taking charge of the operator's call. Furthermore, since the output information also includes values ​​for each item corresponding to customer harassment, it is possible to know what type of customer harassment is occurring. Furthermore, if the output information includes a summary describing the occurrence of customer harassment, the occurrence can be easily understood from the content of the summary.

[0062] <Modifications> Modifications of the above embodiment will be described below. Note that the following modifications can be combined with multiple modifications as appropriate, as long as they do not contradict each other.

[0063] In the above embodiment, the occurrence of customer harassment is detected as an example of the occurrence of a specific situation, but the occurrence of some positive situation may also be detected. An example of the occurrence of a positive situation is, for example, a situation in which a contract for a product or service proposed by an operator in a contact center performing outbound operations can be expected (e.g., a customer becoming interested in a product or service).

[0064] In this case, in step S107 of Figure 3, the notification destination may be determined to be the operator terminal 20 used by the operator, or the supervisor terminal 30 used by the supervisor who monitors the operator, or both.

[0065] Modification 2 In the above embodiment, the detection device 10 includes the situation determination unit 105. However, the situation determination unit 105 may be realized by a cloud service or the like.

[0066] Variation 3: For example, image data of an image of a customer, video data of a video of a customer, or sound data representing environmental sounds in the customer's environment (e.g., sounds around the customer) may be set or specified as input sentences (input information) included in the prompt template. This allows the generation AI to determine whether customer harassment is occurring by taking into account the customer's gestures and the amount of change in those gestures (i.e., the intensity of the gestures), for example, when image data or video data is set or specified. Furthermore, for example, when sound data representing environmental sounds in the customer's environment is set or specified, the generation AI can determine whether customer harassment is occurring by taking into account specific sounds, such as the sound of a customer tapping on a desk.

[0067] Variation 4: The output information of the generation AI may include the judgment result, each item corresponding to customer harassment (i.e., the basis for the judgment result), as well as the accuracy of each item and the overall accuracy. Here, the accuracy of each item refers to a probability or score that indicates the likelihood of the value (Yes / No) of that item. The overall accuracy refers to, for example, the average value of the accuracy of each item.

[0068] Variation 5: Because the determination result information and the utterance information are associated with each other via an utterance ID, for determination result information that includes the determination result "corresponds," it is possible to acquire utterance text from the utterance information that corresponds to the determination result information. For this reason, the detection device 10 according to this embodiment may have a functional unit (e.g., a functional unit called an "information providing unit" or the like) that, for example, in response to a request from the supervisor terminal 30 or the like, identifies utterance information that corresponds to the determination result information that includes the determination result "corresponds," acquires utterance text from the identified utterance information, and displays it on the supervisor terminal 30.

[0069] Furthermore, the utterance information creation unit 103 may create and store utterance information including, in addition to the utterance text, audio data, video data, etc. corresponding to the utterance text. This allows the information providing unit to cause the supervisor terminal 30 to play audio data, video data, etc. in addition to the utterance text in response to a request from the supervisor terminal 30, etc.

[0070] Variation 6: The judgment result included in the judgment result information is the result of a judgment made by the generation AI, and therefore may be incorrect. Therefore, the judgment result included in the judgment result information may be presented to a supervisor, operator, etc., to have them judge whether the judgment result is correct. Furthermore, the correctness of the judgment result may be added to the judgment result information that includes the judgment result.

[0071] Variation 7 The generation AI may be retrained or additionally trained using the judgment result information to which the correctness or incorrectness of the judgment result has been added in Variation 6. That is, the correctness or incorrectness of the judgment result may be used as training data, and the generation AI may determine whether a specific situation has occurred based on the input information when the judgment result was obtained, and the generation AI may be retrained or additionally trained based on the judgment result and the training data.

[0072] Modification 8: The detection device 10 according to this embodiment may use the determination result included in the determination result information to change the transfer rules set in the connection control device 50 or change the settings of the automated voice response of the connection control device 50. For example, for a customer whose call has been identified as a result of determination result information containing the determination result "corresponds," the transfer rules may be changed so that the call is connected to a specific operator (e.g., a veteran operator). Furthermore, for a customer whose call has been identified as a result of determination result information containing the determination result "corresponds," the content of the automated voice response may be changed to content that matches the customer.

[0073] Variation 9 In the above embodiment, the generation AI determines whether a specific situation has occurred, but the generation AI may also determine the possibility that a specific situation will occur in the future. In this case, the generation AI may use, as input information, not only the utterance text and emotional information of the current call, but also, for example, the content of past calls with the customer.

[0074] This makes it possible, for example, to notify operators or supervisors if there is a high possibility that a specific situation will occur in the future, making it possible to prevent negative situations (e.g., customer harassment) from occurring in advance.

[0075] Variation 10: In the above embodiment, emotion information is used by the generation AI to determine whether customer harassment is occurring. However, emotion information need not be used. That is, the detection device 10 does not need to have the emotion recognition unit 102. In this case, the determination instruction unit 104 can create a prompt by specifying or setting the utterance text included in the utterance information as an input sentence (input information) for the template.

[0076] <Summary> As described above, the detection device 10 according to the present embodiment uses generative AI to enable detection of the occurrence of specific situations (e.g., the occurrence of customer harassment) with greater accuracy than existing technologies. Therefore, by using the detection device 10 according to the present embodiment, it becomes possible to respond to specific situations quickly and effectively, and to avoid negative situations and utilize positive situations.

[0077] Furthermore, the detection device 10 according to this embodiment does not require detailed configuration to detect specific situations, thereby reducing the costs required for such configuration. For example, when detecting specific situations using a rule base such as keyword search, detailed rule configuration is required to accurately detect specific situations (i.e., to detect specific situations while reducing both false positives and false negatives). For example, if you want to determine that the keyword "kill" corresponds to customer harassment, but not the keyword "insecticide," you must configure detailed keyword matching rules. This increases the costs (e.g., personnel costs, etc.) required for such configuration. In contrast, the detection device 10 according to this embodiment uses generation AI, eliminating the need for detailed configuration to detect specific situations, thereby reducing the costs required for such configuration.

[0078] Although the above embodiment is directed to a contact center, this is merely an example, and the above embodiment can be applied to other places besides a contact center. For example, the above embodiment can be applied to online sales, online seminars, online learning, etc.

[0079] The present invention is not limited to the above-described specifically disclosed embodiments, and various modifications, changes, and combinations with known technologies are possible without departing from the scope of the claims.

[0080] REFERENCE SIGNS LIST 1 Contact center system 10 Detection device 20 Operator terminal 30 Supervisor terminal 40 PBX 50 Connection control device 60 Customer terminal 70 Communication network 101 Speech recognition unit 102 Emotion recognition unit 103 Utterance information creation unit 104 Judgment instruction unit 105 Situation judgment unit 106 Judgment result information creation unit 107 Notification unit 108 Utterance information storage unit 109 Template storage unit 110 Judgment result information storage unit E Contact center environment

Claims

1. A detection device that detects the occurrence of a specific situation between a first person and a second person, comprising: a creation unit that creates an instruction document including an instruction to determine whether or not the specific situation has occurred based on text representing an utterance by at least one of the first person and the second person; a determination unit that determines whether or not the specific situation has occurred using a trained machine learning model based on the instruction document; and a notification unit that, when it is determined that the specific situation has occurred, notifies a predetermined notification destination that the specific situation has occurred.

2. The detection device described in claim 1, wherein the creation unit creates the instruction by setting or specifying the text as input information to the trained machine learning model using a template created in advance.

3. The detection device of claim 1 or 2, wherein the occurrence of the specific situation includes at least one of the occurrence of harassment of the second person by the first person, the occurrence of a compliance violation, and the occurrence of a situation in which a contract for a product or service proposed by the second person to the first person can be expected.

4. A detection method in which a detection device that detects the occurrence of a specific situation between a first person and a second person executes the following steps: a creation step of creating an instruction document including an instruction to determine whether or not the specific situation has occurred based on text representing utterances by at least one of the first person and the second person; a determination step of determining whether or not the specific situation has occurred using a trained machine learning model based on the instruction document; and a notification step of notifying a predetermined notification destination that the specific situation has occurred if it is determined that the specific situation has occurred.

5. A program that causes a detection device that detects the occurrence of a specific situation between a first person and a second person to execute the following steps: a creation procedure for creating an instruction document including an instruction to determine whether or not the specific situation has occurred based on text representing utterances by at least one of the first person and the second person; a determination procedure for determining whether or not the specific situation has occurred using a trained machine learning model based on the instruction document; and a notification procedure for notifying a predetermined notification destination that the specific situation has occurred if it is determined that the specific situation has occurred.

Images