JP6966357B2 - Information processing system, information processing device, information processing method and program - Google Patents

Description

The present invention relates to an information processing system, an information processing device, an information processing method and a program.

Techniques for preventing fraud using telephones (for example, transfer fraud) have been conventionally known. For example, an incoming call from a number other than the number registered in the phone book is notified by a lamp, the caller is notified that a call is being recorded, or a predetermined button is pressed to notify a pre-registered number. Phones having such functions (these functions are also referred to as "fraud prevention functions") are known. In addition to this, an adapter device that provides a fraud prevention function by being attached to a telephone is also known.

Further, there is also known a technique of referring to a database (so-called blacklist) in which telephone numbers used for fraud using telephones are stored and preventing incoming calls from the telephone numbers stored in this database.

Further, when a pre-registered keyword is included in the call content, a technique is known to prevent a transfer fraud using a telephone by sending a caution message to a pre-registered notification destination (for example). , Patent Document 1).

JP-A-2015-220594

However, for example, when using a telephone having a fraud prevention function or an adapter device providing a fraud prevention function, it is necessary for the user to purchase such a telephone or an adapter device and make various settings and installations, which is a burden on the user. It was. Further, for example, when a technique for preventing an incoming call from a telephone number stored in a database is used, fraud using a telephone number not stored in this database cannot be prevented.

On the other hand, for example, when the technology disclosed in Patent Document 1 is used, the user needs to register the keyword in advance, which is a burden on the user, and the pre-registered keyword is included in the call content. It was not possible to prevent new scams that were not included.

As described above, in the conventional technology, some work load (for example, setting work and installation work of the telephone or adapter device, keyword registration work, etc.) is incurred on the user side, and fraud of a new technique is effectively performed. I couldn't prevent it.

The embodiment of the present invention has been made in view of the above points, and an object thereof is to support the prevention of fraud using a telephone.

In order to achieve the above object, an embodiment of the present invention includes a voice recognition means for creating text data by a voice recognition method from voice data indicating a call content in a voice call, and text data created by the voice recognition means. The voice call corresponds to a specific call by using at least one information of the location of the incoming terminal of the voice call and the time zone in which the voice call is taking, and a learned model learned in advance. The accuracy calculation means for calculating the accuracy, the determination means for determining whether or not the accuracy calculated by the accuracy calculation means is equal to or higher than a predetermined threshold, and the determination means determine that the accuracy is equal to or higher than the threshold value. In that case, the trained model is additionally learned by using the notification means for notifying the predetermined terminal that the voice call is a specific call and the text data when the notification is performed by the notification means. Te, possess an updating means for updating the parameters of the learned model and the update unit performs the update in the background of the voice communication, characterized in that.

According to the embodiment of the present invention, it is possible to support the prevention of fraud using a telephone.

It is a figure which shows an example of the whole structure of the specific call notification system which concerns on this embodiment. It is a figure which shows an example of the hardware configuration of a computer. It is a figure which shows an example of the functional structure of the specific call notification device, the voice recognition device, the sentiment analysis device, and the call content analysis device which concerns on this embodiment. It is a sequence diagram which shows an example of the notification processing of the specific call which concerns on this embodiment. It is a sequence diagram which shows an example of the update process of the call content analysis model which concerns on this embodiment. It is a sequence diagram which shows the other example of the notification processing of the specific call which concerns on this embodiment.

Hereinafter, embodiments of the present invention (hereinafter referred to as “the present embodiment”) will be described. In the present embodiment, when a call related to fraud using a telephone (hereinafter, a call related to fraud using a telephone is referred to as a "specific call") is made, a specific call occurs at a notification destination registered in advance. The specific call notification system 1 for notifying this will be described.

Phone fraud is fraud aimed at illegally receiving money or other payments using the phone, such as transfer fraud, oleore fraud, loan deposit fraud, refund fraud, and fictitious billing fraud. , Financial product transaction nominal fraud, etc. The specific call may include a call (so-called nuisance call, mischievous call, etc.) that does not correspond to fraud but causes some discomfort to the called party.

<Overall configuration of specific call notification system 1>
First, the overall configuration of the specific call notification system 1 according to the present embodiment will be described with reference to FIG. FIG. 1 is a diagram showing an example of the overall configuration of the specific call notification system 1 according to the present embodiment.

As shown in FIG. 1, the specific call notification system 1 according to the present embodiment includes a specific call notification device 10, a relay device 20, a terminal 30, a voice recognition device 40, a sentiment analysis device 50, and a call content. The analyzer 60 is included. The specific call notification device 10 and the relay device 20 are included in the system environment E constructed in the telephone station, the communication station, or the like of the telecommunications carrier. Further, the specific call notification device 10, the voice recognition device 40, the sentiment analysis device 50, and the call content analysis device 60 are communicably connected via a network such as the Internet.

The terminal 30 is a telephone of a user who uses a voice call service provided by a telecommunications carrier (this voice call service is also referred to as a "telephone service", a "telephone service", or simply a "telephone"). Mobile phones, smartphones, etc. The terminal 30 may be, for example, a tablet terminal, a PC (personal computer), a wearable device, a game device, or the like, as long as it has a voice call function.

In the present embodiment, the terminal 30 on the calling side of the telephone is referred to as "calling terminal 30a", and the terminal 30 on the receiving side of the telephone is referred to as "incoming terminal 30b". Further, when the call between the outgoing terminal 30a and the incoming terminal 30b is a specific call, the terminal 30 registered in advance other than the incoming terminal 30b as the notification destination that the specific call has occurred is referred to as the "notification destination terminal 30c". Is expressed.

The specific call notification device 10 determines whether or not a specific call has occurred between the outgoing terminal 30a and the incoming terminal 30b, and gives a predetermined notification to the incoming terminal 30b and the notification destination terminal 30c according to the determination result. A computer or computer system. In the following, the service that gives such a notification will also be referred to as a "specific call notification service".

The specific call notification device 10 converts the contents of a call between the calling terminal 30a and the incoming call terminal 30b into text data by the voice recognition device 40. At the same time, the specific call notification device 10 emotionally analyzes the content of the call between the outgoing terminal 30a and the incoming terminal 30b by the emotion analysis device 50. Then, the specific call notification device 10 determines whether or not the call is a specific call from the analysis result by the call content analysis device 60 using these text data and the sentiment analysis result, and notifies according to the determination result. A predetermined notification is given to the destination terminal 30c.

The relay device 20 is a device that relays voice indicating the content of a call between terminals 30, and is, for example, an in-station telephone exchange, an accommodation router, or the like. When a call is started between the calling terminal 30a and the incoming terminal 30b, the relay device 20 mutually relays a voice indicating the content of the call between the outgoing terminal 30a and the incoming terminal 30b, and transmits the voice. Transfer to the specific call notification device 10. However, the relay device 20 may transfer the voice to the specific call notification device 10 only when at least one user of the outgoing terminal 30a and the incoming terminal 30b subscribes to the specific call notification service. ..

The voice recognition device 40 is a computer or a computer system that creates text data by converting voice data representing the contents of a call between the transmitting terminal 30a and the incoming terminal 30b into text.

The sentiment analysis device 50 is a computer or a computer system that analyzes the contents of a call between the transmitting terminal 30a and the incoming terminal 30b and creates an emotion analysis result. The sentiment analysis is, for example, analyzing whether the emotions of the caller or the callee are positive or negative.

The call content analysis device 60 analyzes the call content between the calling terminal 30a and the incoming terminal 30b using the text data created by the voice recognition device 40 and the sentiment analysis result analyzed by the sentiment analysis device 50, and the call is made. Analyze the possibility that is a specific call. Such analysis is performed by classifying whether or not the call is a specific call using a trained model pre-learned by a machine learning method.

As will be described later, the call content analysis device 60 analyzes the possibility of a specific call by using only the text data created by the voice recognition device 40 without using the sentiment analysis result analyzed by the sentiment analysis device 50. You may.

The configuration of the specific call notification system 1 shown in FIG. 1 is an example, and may be another configuration. For example, at least one of the voice recognition device 40, the emotion analysis device 50, and the call content analysis device 60 may be included in the system environment E. Further, the voice recognition device 40, the sentiment analysis device 50, and the call content analysis device 60 do not have to be different devices, and for example, two or more of these devices are configured by one device. It may be configured (or configured as one system).

<Hardware configuration of various devices>
Next, the hardware configuration of the specific call notification device 10, the voice recognition device 40, the sentiment analysis device 50, and the call content analysis device 60 included in the specific call notification system 1 according to the present embodiment will be described. The specific call notification device 10, the voice recognition device 40, the sentiment analysis device 50, and the call content analysis device 60 are realized by using, for example, one or more computers 100 shown in FIG. FIG. 2 is a diagram showing an example of the hardware configuration of the computer 100.

The computer 100 shown in FIG. 2 includes an input device 101, a display device 102, an external I / F 103, a RAM (Random Access Memory) 104, a ROM (Read Only Memory) 105, and a CPU (Central Processing Unit) 106. , Communication I / F 107 and auxiliary storage device 108. Each of these hardware is connected so as to be communicable via the bus B.

The input device 101 is, for example, a keyboard, a mouse, a touch panel, or the like, and is used by a user of the computer 100 to input various operations. The display device 102 is, for example, a display or the like, and displays various processing results. The specific call notification device 10, the voice recognition device 40, the sentiment analysis device 50, and the call content analysis device 60 do not have to have at least one of the input device 101 and the display device 102.

The external I / F 103 is an interface with an external device. The external device includes a recording medium 103a and the like. The computer 100 can read or write the recording medium 103a or the like via the external I / F 103.

The recording medium 103a includes, for example, a flexible disk, a CD (Compact Disc), a DVD (Digital Versatile Disk), an SD memory card (Secure Digital memory card), a USB (Universal Serial Bus) memory card, and the like.

The RAM 104 is a volatile semiconductor memory that temporarily holds programs and data. The ROM 105 is a non-volatile semiconductor memory that can hold programs and data even when the power is turned off. The ROM 105 stores, for example, OS (Operating System) settings, network settings, and the like.

The CPU 106 is an arithmetic unit that reads programs and data from the ROM 105, the auxiliary storage device 108, and the like onto the RAM 104 and executes processing.

The communication I / F 107 is an interface for connecting the computer 100 to the network. The auxiliary storage device 108 is, for example, an HDD (Hard Disk Drive), an SSD (Solid State Drive), or the like, and is a non-volatile storage device that stores programs and data.

The programs and data stored in the auxiliary storage device 108 include, for example, an OS, an application program that realizes various functions on the OS, and the like.

The specific call notification device 10, the voice recognition device 40, the sentiment analysis device 50, and the call content analysis device 60 according to the present embodiment execute various processes described later by using one or more computers 100 shown in FIG. Can be done.

<Functional configuration of various devices>
Next, the functional configurations of the specific call notification device 10, the voice recognition device 40, the sentiment analysis device 50, and the call content analysis device 60 included in the specific call notification system 1 according to the present embodiment will be described with reference to FIG. .. FIG. 3 is a diagram showing an example of the functional configuration of the specific call notification device 10, the voice recognition device 40, the sentiment analysis device 50, and the call content analysis device 60 according to the present embodiment.

<< Specific call notification device 10 >>
As shown in FIG. 3, the specific call notification device 10 according to the present embodiment includes a voice data creation unit 11, a voice recognition request unit 12, a sentiment analysis request unit 13, a call content analysis request unit 14, and a specific call. It has a determination unit 15, a notification unit 16, an update data creation unit 17, and an update request unit 18. Each of these functional units is realized, for example, by a process of causing a CPU or the like to execute one or more programs installed in the specific call notification device 10.

The voice data creation unit 11 creates voice data from the voice transferred from the relay device 20. The voice data creation unit 11 creates voice data in an arbitrary data format that can be processed by the voice recognition device 40 or the sentiment analysis device 50, for example.

The voice recognition request unit 12 transmits a voice recognition request including the voice data created by the voice data creation unit 11 to the voice recognition device 40. The voice recognition request unit 12 transmits a voice recognition request to the voice recognition device 40 by, for example, a streaming method. By using the streaming method, the voice between the transmitting terminal 30a and the incoming terminal 30b can be converted into text in real time.

However, the present invention is not limited to this, and the voice recognition requesting unit 12 divides a call between the calling terminal 30a and the incoming terminal 30b at predetermined time intervals (for example, several seconds), and the voice recognition requesting unit 12 divides the call between the divided sections. A voice recognition request including voice data indicating the voice of the above may be transmitted to the voice recognition device 40. Alternatively, for example, after identifying a voice break during a call (for example, a place where silence occurs for several msec), the call is divided for each specified voice break, and the section between the divided sections is divided. A voice recognition request including voice data indicating voice may be transmitted to the voice recognition device 40. In addition to these, for example, a voice recognition request including voice data indicating voice from the start of a call to the end of a call between the calling terminal 30a and the receiving terminal 30b may be transmitted to the voice recognition device 40.

The sentiment analysis request unit 13 transmits an emotion analysis request including the voice data created by the voice data creation unit 11 to the sentiment analysis device 50. The sentiment analysis request unit 13 transmits the sentiment analysis request to the sentiment analyzer 50 by, for example, a streaming method. By using the streaming method, it is possible to perform voice emotion analysis between the transmitting terminal 30a and the receiving terminal 30b in real time.

However, the present invention is not limited to this, and the sentiment analysis request unit 13 makes an emotion analysis request including voice data indicating voice from the start of the call to the end of the call between the outgoing terminal 30a and the incoming terminal 30b, for example. You may send it to. Further, when the sentiment analysis device 50 performs the sentiment analysis from the text data, the sentiment analysis requesting unit 13 includes the sentiment analysis including the text data converted into text by the voice recognition device 40 instead of the voice data. The request may be sent to the sentiment analyzer 50. Further, when the sentiment analysis device 50 performs the sentiment analysis from both the voice data and the text data, the sentiment analysis request unit 13 makes a sentiment analysis request including both the voice data and the text data. It may be sent to 50.

The call content analysis request unit 14 transmits a call content analysis request including text data converted into text by the voice recognition device 40 and sentiment analysis result data indicating the analysis result by the sentiment analysis device 50 to the call content analysis device 60. do. The call content analysis request unit 14 uses, for example, text data indicating a certain range of text (for example, one word, several words, one phrase, one sentence, etc.) as one unit, and uses this text data as a unit. A call content analysis request including the sentiment analysis result data corresponding to the text data is transmitted to the call content analysis device 60. However, the present invention is not limited to this, and the call content analysis request unit 14 may, for example, transmit a call content analysis request including text data indicating one character to the call content analysis device 60, or a word having a certain meaning. A call content analysis request including text data indicating a collection of the above may be transmitted to the call content analysis device 60. The sentiment analysis result data corresponding to the text data is the sentiment analysis result data obtained from the same voice as the voice used to obtain the text data.

Here, for example, when the call content analysis device 60 analyzes the call content without using the emotion analysis result data, the call content analysis request unit 14 makes a call content analysis request containing only text data. It is transmitted to the analyzer 60.

The specific call determination unit 15 determines whether or not the call between the outgoing terminal 30a and the incoming terminal 30b corresponds to the specific call by using the call content analysis result data showing the analysis result by the call content analysis device 60. ..

When the specific call determination unit 15 determines that the call between the outgoing terminal 30a and the incoming terminal 30b corresponds to the specific call, the notification unit 16 notifies the incoming terminal 30b and the notification destination terminal 30c registered in advance with a predetermined notification. To send. Information that identifies the notification destination terminal 30c (for example, a telephone number, an e-mail address, etc.) is registered in advance by the user of the outgoing terminal 30a or the incoming terminal 30b, for example, when starting to use the specific call notification service. ..

Here, the predetermined notification may be, for example, a notification by e-mail indicating that a call related to fraud has occurred, or an automatic announcement indicating that the call has occurred may be notified by telephone. Also, both of these notifications may be given. Further, the information for the notification destination terminal 30c to make an interrupted call (telephone number for making an interrupted call, etc.) may be notified to the call between the outgoing terminal 30a and the incoming terminal 30b.

The update data creation unit 17 creates update data for updating the call content analysis model 63, which will be described later, every time a call is made between the outgoing terminal 30a and the incoming terminal 30b, for example.

When the update data is updated by the update data creation unit 17, the update request unit 18 transmits an update request including the update data to the call content analysis device 60.

Note that FIG. 3 shows, as an example, a case where one device has each functional unit of the specific call notification device 10, but the present invention is not limited to this. For example, when the specific call notification device 10 is composed of a plurality of devices, each of these functional units includes one of the plurality of devices constituting the specific call notification device 10. You just have to have it.

≪Voice recognition device 40≫
As shown in FIG. 3, the voice recognition device 40 according to the present embodiment has a voice recognition unit 41. The voice recognition unit 41 is realized, for example, by a process of causing a CPU or the like to execute one or more programs installed in the voice recognition device 40.

When the voice recognition unit 41 receives the voice recognition request from the specific call notification device 10, the voice recognition unit 41 converts the voice data included in the voice recognition request into text by a known voice recognition technique, and creates text data. Then, the voice recognition unit 41 transmits the voice recognition result including the created text data to the specific call notification device 10.

≪Sentiment analyzer 50≫
As shown in FIG. 3, the sentiment analysis device 50 according to the present embodiment has an emotion analysis unit 51. The emotion analysis unit 51 is realized, for example, by a process of causing a CPU or the like to execute one or more programs installed in the emotion analysis device 50.

Upon receiving the sentiment analysis request from the specific call notification device 10, the sentiment analysis unit 51 performs sentiment analysis on the voice data (or text data) included in the sentiment analysis request by a known sentiment analysis technique. Create sentiment analysis result data. Then, the sentiment analysis unit 51 transmits the sentiment analysis result including the created sentiment analysis result data to the specific call notification device 10.

The sentiment analysis result data is, for example, data showing a positive ratio or a negative ratio of the emotions of the caller or the called party in the call indicated by the voice data (or text data). However, the present invention is not limited to this, and the sentiment analysis data may be, for example, data showing the ratio of various emotions such as anger, anxiety, joy, and sadness of the sender and the called.

<< Call content analysis device 60 >>
As shown in FIG. 3, the call content analysis device 60 according to the present embodiment has a call content analysis unit 61, a model update unit 62, and a call content analysis model 63. The call content analysis unit 61 and the model update unit 62 are realized, for example, by a process in which one or more programs installed in the call content analysis device 60 are executed by a CPU or the like. Further, the call content analysis model 63 is stored in, for example, an auxiliary storage device of the call content analysis device 60.

When the call content analysis unit 61 receives the call content analysis request from the specific call notification device 10, the call content analysis model 63 uses the call content analysis model 63 to convert the text data included in the call content analysis request into a "specific call" or a "specific call". Classify into one of "not" and create call content analysis result data including the classification result and its accuracy. At this time, when the sentiment analysis result data is included in the call content analysis request, the call content analysis unit 61 also uses the sentiment analysis result data to use the text data as "specific call" or "not a specific call". Classify into one of.

The call content analysis model 63 is a pre-learned model using a known machine learning method. As the call content analysis model 63, for example, a model such as a neural network (particularly, a deep neural network) can be used. However, the call content analysis model 63 is not limited to this, and various classification models used for machine learning such as SVM (Support Vector Machine) can be used. As the call content analysis model 63, a classification model created by a rule-based method may be used.

Here, the call content analysis model 63 is, for example, text data indicating words, phrases, sentences, etc. used in a specific call, and a result in which the text data should be classified (“specific call” or “not a specific call”). It is trained in advance using a large number of training data including and.

When the model update unit 62 receives an update request from the specific call notification device 10, for example, the model update unit 62 updates the call content analysis model 63 using the update data included in the update request. That is, the model update unit 62 updates the parameters of the call content analysis model 63 using the update data. In this way, every time a call is made between the outgoing terminal 30a and the incoming terminal 30b, the parameters of the call content analysis model 63 are updated, so that, for example, a fraud with a new technique is also classified as a specific call. You will be able to.

Note that FIG. 3 shows, as an example, a case where one device has each functional unit of the call content analysis device 60, but the present invention is not limited to this. For example, when the call content analysis device 60 is composed of a plurality of devices, each of these functional units includes one of the plurality of devices constituting the call content analysis device 60. All you have to do is have <Details of processing>
Hereinafter, the details of the processing of the specific call notification system 1 according to the present embodiment will be described.

≪Notification processing of specific call≫
Hereinafter, as an example, it is determined whether or not the call between the outgoing terminal 30a and the incoming terminal 30b is a specific call without using sentiment analysis, and the incoming terminal 30b or the notification is determined according to the determination result. The process of giving a predetermined notification to the destination terminal 30c will be described with reference to FIG. FIG. 4 is a sequence diagram showing an example of notification processing of a specific call according to the present embodiment. In FIG. 4, it is assumed that a call between the calling terminal 30a and the incoming call terminal 30b is started, and the voice of this call is transferred from the relay device 20 to the specific call notification device 10.

The voice data creation unit 11 of the specific call notification device 10 creates voice data from the voice transferred from the relay device 20 (step S101).

The specific call notification device 10 determines, for example, whether or not the user of the incoming terminal 30b is subscribed to the specific call notification service, and only when it is determined that the user is subscribed to the specific call notification service is voice. You may try to create the data. Further, this determination may be performed by another device different from the specific call notification device 10 (for example, a relay device 20, or a device installed between the relay device 20 and the specific call notification device 10).

Next, the voice recognition request unit 12 of the specific call notification device 10 transmits a voice recognition request including the voice data created by the voice data creation unit 11 to the voice recognition device 40 (step S102). At this time, the voice recognition request unit 12 transmits a voice recognition request including voice data to the voice recognition device 40 by using, for example, a streaming method.

Upon receiving the voice recognition request, the voice recognition unit 41 of the voice recognition device 40 converts the voice data included in the voice recognition request into text by a known voice recognition technique, and creates text data (step S103).

Next, the voice recognition unit 41 of the voice recognition device 40 transmits the voice recognition result including the text data created in step S103 to the specific call notification device 10 (step S104).

The call content analysis request unit 14 of the specific call notification device 10 transmits a call content analysis request including text data converted into text by the voice recognition device 40 to the call content analysis device 60 (step S105).

Upon receiving the call content analysis request, the call content analysis unit 61 of the call content analysis device 60 uses the call content analysis model 63 to convert the text data included in the call content analysis request into a "specific call" or "not a specific call". ”, And create call content analysis result data including the classification result and its accuracy (step S106). The accuracy is a value (for example, a percentage) indicating the possibility of being classified into the corresponding classification result. Therefore, the call content analysis result data is represented, for example, as a classification result "specific call", an accuracy of "65%", or the like.

Next, the call content analysis unit 61 of the call content analysis device 60 transmits the call content analysis result including the call content analysis result data created in step S106 to the specific call notification device 10 (step S107).

When the specific call determination unit 15 of the specific call notification device 10 receives the call content analysis result, the specific call content analysis result data included in the call content analysis result is used to make a call between the calling terminal 30a and the incoming call terminal 30b. Determines whether or not corresponds to a specific call (step S108). The specific call determination unit 15 determines, for example, whether or not the accuracy of the classification result "specific call" included in the call content analysis result data is equal to or higher than a preset predetermined threshold, and the accuracy is equal to or higher than the threshold. If, it is determined that the call corresponds to a specific call. On the other hand, the specific call determination unit 15 determines that the call does not correspond to the specific call, for example, when the accuracy is less than the threshold value.

For example, it is assumed that the call content analysis result data obtained in step S107 is the classification result "specific call", the accuracy is "65%", and the preset threshold value is "60". In this case, since the accuracy "65%" is equal to or higher than the threshold value "60", the specific call determination unit 15 determines that the call corresponds to the specific call. On the other hand, for example, it is assumed that the call content analysis result data obtained in step S107 is the classification result "specific call", the accuracy is "55%", and the preset threshold value is "60". In this case, since the accuracy "55%" is less than the threshold value "60", the specific call determination unit 15 determines that the call does not correspond to the specific call.

If it is determined in step S108 above that the call corresponds to a specific call, the notification unit 16 of the specific call notification device 10 transmits a predetermined notification to the notification destination terminal 30c registered in advance (step S109).

As described above, the notification may be, for example, an e-mail indicating that a call related to fraud has occurred, or an automatic announcement indicating that the call has occurred. Also, both of these notifications may be given. At this time, the e-mail may include, for example, the text represented by the text data classified as the specific call in the call content analysis in step S106 above, the accuracy of the classification result, and the like.

Further, the notification destination terminal 30c may notify the call between the calling terminal 30a and the incoming terminal 30b of a telephone number or the like for making an interrupted call. In addition to this, for example, when the angle of the classification result "specific call" is very high in step S108 above, the outgoing terminal 30a and the incoming terminal are premised on the prior consent of the user of the specific call notification service. The call to and from 30b may be forcibly disconnected. Further, such a disconnection of the call may be performed after the above notification is given, or may be performed instead of the above notification.

As described above, in the specific call notification system 1 according to the present embodiment, the call between the outgoing terminal 30a and the incoming terminal 30b is a specific call using the call content analysis model 63 pre-learned by the machine learning method. It is possible to determine whether or not it is. Then, in the specific call notification system 1 according to the present embodiment, when it is determined that the call between the outgoing terminal 30a and the incoming terminal 30b is a specific call, the notification destination terminal 30c registered in advance (for example, the incoming terminal) A predetermined notification is given to the terminal 30) used by the family of 30b or the like.

As a result, in the specific call notification system 1 according to the present embodiment, some work load (for example, setting work and installation work of the telephone or adapter device, keyword registration work, etc.) is performed on the user of the specific call notification service. It becomes possible to prevent a call (specific call) related to a transfer fraud or the like without causing a transfer fraud.

Further, in the specific call notification system 1 according to the present embodiment, by using the call content analysis model 63 learned in advance by the machine learning method, unknown words, phrases, etc. are compared with, for example, a rule-based method. Even if is used, there is a high possibility that it can be classified as a specific call, and transfer fraud and the like can be prevented more effectively. Similarly, by using the call content analysis model 63, it is possible to classify not only words and phrases but also the contexts before and after these words and phrases, for example, in comparison with a rule-based method. Therefore, it becomes possible to classify specific calls with higher accuracy.

<< Update processing of call content analysis model 63 >>
Hereinafter, the process of updating the call content analysis model 63 each time a call is made between the outgoing terminal 30a and the incoming terminal 30b will be described with reference to FIG. FIG. 5 is a sequence diagram showing an example of the update process of the call content analysis model 63 according to the present embodiment. The process of FIG. 5 may be executed, for example, after the end of the call between the calling terminal 30a and the incoming call terminal 30b, or may be executed in the background during the call between the calling terminal 30a and the incoming call terminal 30b. May be done. Alternatively, it may be executed every predetermined period (for example, one week or one month).

First, the update data creation unit 17 of the specific call notification device 10 creates update data for updating the call content analysis model 63 (step S201).

Here, for example, when the call content analysis model 63 is updated by a supervised method, the updated data is classified into, for example, the text data obtained in step S104 of FIG. 4. It is data to which a correct answer label (“specific call” or “not specific call”) indicating a desired result is attached. At this time, as the correct answer label, for example, the information fed back from the notification destination terminal 30c notified in step S109 of FIG. 4 can be used. For example, when feedback indicating "not a specific call" is received from the notification destination terminal 30c notified in step S109 of FIG. 4, a correct answer label indicating "not a specific call" is given to the text data. It's fine. On the other hand, when feedback to the effect that "it is a specific call" is received from the notification destination terminal 30c notified in step S109 of FIG. 4, a correct answer label indicating "specific call" is given to the text data. Just do it. The correct answer label is not limited to the feedback from the notification destination terminal 30c, and may be determined by various methods. For example, whether or not to use the large amount of text data obtained in step S104 of FIG. 4 as update data by an operator or the like using a database or the like listed from a predetermined viewpoint by a known data analysis method or the like, update data. It may be determined whether the correct answer label is "specific call" or "not specific call". From the viewpoint of listing, for example, text data is listed by the source telephone number of a call made to a plurality of called source telephone numbers of a predetermined number or more, and text data having the same or similar call contents is listed. It is possible to do something like that.

On the other hand, for example, when the call content analysis model 63 is updated by the method of unsupervised learning, the update data creation unit 17 may use the text data obtained in step S104 of FIG. 4 as the update data as it is. ..

Next, the update request unit 18 of the specific call notification device 10 transmits an update request including the update data created in step S201 to the call content analysis device 60 (step S202).

Upon receiving the update request, the model update unit 62 of the call content analysis device 60 updates the call content analysis model 63 using the update data included in the update request (step S203). That is, the model update unit 62 performs additional learning of the call content analysis model 63 by a known machine learning method, and updates the parameters of the call content analysis model 63.

Then, the model update unit 62 of the call content analysis device 60 transmits the update result to the specific call notification device 10 (step S204).

As described above, in the specific call notification system 1 according to the present embodiment, the call content analysis model 63 is updated from the call content each time a call is made between the outgoing terminal 30a and the incoming terminal 30b. This makes it possible, for example, to classify a new type of fraud as a specific call.

In particular, in the specific call notification system 1 according to the present embodiment, the specific call notification device 10 included in the system environment E constructed in the telephone office or communication station of the telecommunications carrier creates update data and analyzes the contents of the call. An update request is made to the device 60. Therefore, the call content analysis model 63 can be updated using a huge number of update data created from a huge number of calls, and the call content analysis model 63 can classify specific calls with high accuracy. You can expect to get it.

In addition, in FIG. 5, the case where the call content analysis model 63 is updated every time a call is made between the calling terminal 30a and the incoming call terminal 30b has been described, but for example, it is necessary to update the call content analysis model 63. It may be executed at any time. For example, when it is predicted that a certain kind of fraud will occur in view of social conditions, etc., update data representing a method peculiar to this fraud is created in step S201 above, and the call content analysis model is based on the update data. 63 may be updated.

<< Other examples of notification processing for specific calls >>
Hereinafter, as an example, emotion analysis is also used to determine whether or not the call between the outgoing terminal 30a and the incoming terminal 30b is a specific call, and the notification destination terminal 30c is determined according to the determination result. The process of notifying the above will be described with reference to FIG. FIG. 6 is a sequence diagram showing another example of the notification processing of the specific call according to the present embodiment. In FIG. 6, it is assumed that a call between the calling terminal 30a and the incoming call terminal 30b is started and the voice of this call is transferred from the relay device 20 to the specific call notification device 10. Since steps S101 to S104 and steps S108 to S109 of FIG. 6 are the same as those of FIG. 4, the description thereof will be omitted.

The sentiment analysis request unit 13 of the specific call notification device 10 transmits an emotion analysis request including the voice data created by the voice data creation unit 11 to the sentiment analysis device 50 (step S301). At this time, the sentiment analysis request unit 13 transmits, for example, an emotion analysis request including voice data to the sentiment analysis device 50 by using a streaming method.

When the sentiment analysis device 50 performs sentiment analysis on the text data, the sentiment analysis request unit 13 transmits a sentiment analysis request including the text data obtained from the voice recognition device 40 to the sentiment analysis device 50. ..

When the sentiment analysis device 50 performs sentiment analysis on the text data, for example, the voice recognition device 40 may send the sentiment analysis request including the text data to the sentiment analyzer 50. In this case, the voice recognition device 40 may transmit the sentiment analysis result data obtained from the sentiment analysis device 50 and the text data to the specific call notification device 10.

Upon receiving the sentiment analysis request, the sentiment analysis unit 51 of the sentiment analyzer 50 performs sentiment analysis on the voice data (or text data) included in the sentiment analysis request by a known sentiment analysis technique, and performs sentiment analysis. Create result data (step S302). Here, the sentiment analysis result data includes, for example, a classification of whether the emotion is positive or negative and its accuracy for each speaker (sender or receiver) and phrase. The sentiment analysis result data is not limited to this, and may be data including, for example, classification of various emotions such as anger, anxiety, joy, and sadness of the speaker and their accuracy, or the speaker is lying. The data may include the classification of whether or not the person is speaking and its accuracy.

Next, the sentiment analysis unit 51 of the sentiment analysis device 50 transmits the sentiment analysis result including the sentiment analysis result data created in step S302 to the specific call notification device 10 (step S303).

The call content analysis request unit 14 of the specific call notification device 10 makes a call for a call content analysis request including text data converted into text by the voice recognition device 40 and sentiment analysis result data indicating the analysis result by the sentiment analysis device 50. It is transmitted to the content analyzer 60 (step S304).

When the call content analysis unit 61 of the call content analysis device 60 receives the call content analysis request, the text data is "specific call" or "not a specific call" using the call content analysis model 63 and the sentiment analysis result data. Classify into any of the above, and create call content analysis result data including the classification result and its accuracy (step S305).

At this time, the call content analysis model 63 is a model that classifies the text data into either "specific call" or "not a specific call" by using the sentiment analysis result data, and obtains the classification result and its accuracy. Suppose there is. Therefore, in this case, the call content analysis model 63 is pre-learned using the learning data including the emotion analysis result.

Next, the call content analysis unit 61 of the call content analysis device 60 transmits the call content analysis result including the call content analysis result data created in step S305 to the specific call notification device 10 (step S306).

As described above, in the specific call notification system 1 according to the present embodiment, it is possible to determine whether or not the call between the outgoing terminal 30a and the incoming terminal 30b is a specific call by using the result of the sentiment analysis. can. Therefore, it can be expected that specific calls can be classified with higher accuracy than when the results of sentiment analysis are not used. In particular, for example, in a specific word, phrase, context, etc., it may be difficult to classify a specific call only by text, so by using the result of sentiment analysis, classification with higher accuracy can be expected.

<Modification examples, etc.>
Although the present embodiment has been described above, the following modifications or applications are also possible.

(1) In the call content analysis by the call content analysis device 60, in addition to the text data (and sentiment analysis result data), the telephone number of the calling terminal 30a, the location of the incoming terminal 30b, the call time zone, and the call time At least one of the following information may be further used. As a result, it can be expected that specific calls can be classified with higher accuracy. This is because, for example, it is possible to classify a telephone number used for a transfer fraud or the like, a time zone in which a call such as a transfer fraud is likely to occur, a talk time, or the like. Further, the location of the incoming terminal 30b is because, for example, a call near the ATM of a bank may be a transfer fraud or the like. When performing a call content analysis using at least one of these pieces of information, the call content analysis model 63 needs to be trained in advance using the relevant information as well.

(2) At least one of voice recognition by the voice recognition device 40, emotion analysis by the sentiment analysis device 50, and call content analysis by the call content analysis device 60 is a service provider that provides a call content notification service. It may be realized by an external service (for example, a cloud-type service) provided by a third party different from the above.

(3) The incoming terminal 30b may have some of the functions of the specific call notification device 10, the voice recognition device 40, and the emotion analysis device 50. For example, the incoming terminal 30b may perform voice recognition by the voice recognition device 40, emotion analysis by the emotion analysis device 50, and the like. In this case, the specific call notification device 10 may receive text data, sentiment analysis result data, and the like from the incoming terminal 30b. Alternatively, for example, the incoming terminal 30b may transmit a voice recognition request, a sentiment analysis request, or the like to the voice recognition device 40 or the sentiment analysis device 50.

The present invention is not limited to the above-described embodiment disclosed specifically, and various modifications and modifications can be made without departing from the scope of claims.

1 Specific call notification system 10 Specific call notification device 11 Voice data creation unit 12 Voice recognition request unit 13 Sentiment analysis request unit 14 Call content analysis request unit 15 Specific call judgment unit 16 Notification unit 17 Update data creation unit 18 Update request unit 20 Relay Device 30 Terminal 30a Outgoing terminal 30b Incoming terminal 30c Notification destination terminal 40 Voice recognition device 41 Voice recognition unit 42 Voice recognition model 50 Sentiment analysis device 51 Sentiment analysis unit 52 Sentiment analysis model 60 Call content analysis device 61 Call content analysis unit 62 Model update Part 63 Call content analysis model

Claims (8)

A voice recognition means that creates text data using a voice recognition method from voice data that indicates the content of a voice call.
Using the text data created by the voice recognition means , at least one piece of information about the location of the incoming terminal of the voice call and the time zone in which the voice call is taking, and a pre-learned trained model, the text data is used. An accuracy calculation means for calculating the probability that the voice call corresponds to a specific call, and
A determination means for determining whether or not the accuracy calculated by the accuracy calculation means is equal to or higher than a predetermined threshold value.
When the determination means determines that the accuracy is equal to or higher than the threshold value, the notification means for notifying a predetermined terminal that the voice call is a specific call, and
When the notification by the notification means is given, the update means for additionally learning the trained model using the text data and updating the parameters of the trained model, and
Have a,
The update means is an information processing system characterized in that the update is performed in the background of the voice call. It has an emotion analysis means for analyzing a speaker's emotion in the voice call by an emotion analysis method using at least one of the voice data and the text data.
The accuracy calculation means is
The information processing system according to claim 1, further comprising calculating the probability that the voice call corresponds to a specific call by using the analysis result by the sentiment analysis means. The notification means is
The invention according to claim 1 or 2 , wherein when the determination means determines that the accuracy is equal to or higher than the threshold value, the terminal is notified of the telephone number for interrupting the voice call. Information processing system. The information processing system according to any one of claims 1 to 3 , further comprising a disconnection means for disconnecting the voice call when the determination means determines that the accuracy is equal to or higher than the threshold value. .. The specific call is a call related to fraud using a telephone.
The trained model is a neural network model trained using training data including at least one of words, phrases, and sentences used in a call related to a fraud using the telephone. The information processing system according to any one of claims 1 to 4. A voice recognition means that creates text data using a voice recognition method from voice data that indicates the content of a voice call.
Using the text data created by the voice recognition means , at least one piece of information about the location of the incoming terminal of the voice call and the time zone in which the voice call is taking, and a pre-learned trained model, the text data is used. An accuracy calculation means for calculating the probability that the voice call corresponds to a specific call, and
A determination means for determining whether or not the accuracy calculated by the accuracy calculation means is equal to or higher than a predetermined threshold value.
When the determination means determines that the accuracy is equal to or higher than the threshold value, the notification means for notifying a predetermined terminal that the voice call is a specific call, and
When the notification by the notification means is given, the update means for additionally learning the trained model using the text data and updating the parameters of the trained model, and
Have a,
The update means is an information processing device characterized in that the update is performed in the background of the voice call. A voice recognition procedure that creates text data using a voice recognition method from voice data that indicates the content of a voice call.
Using the text data created by the voice recognition procedure , at least one piece of information about the location of the incoming terminal of the voice call and the time zone during which the voice call is taking, and a pre-learned trained model, the text data is used. The probability calculation procedure for calculating the probability that the voice call corresponds to a specific call, and
A determination procedure for determining whether or not the accuracy calculated by the accuracy calculation procedure is equal to or higher than a predetermined threshold value, and a determination procedure for determining whether or not the accuracy is equal to or higher than a predetermined threshold value.
When it is determined by the determination procedure that the accuracy is equal to or higher than the threshold value, a notification procedure for notifying a predetermined terminal that the voice call is a specific call and a notification procedure.
When the notification is performed by the notification procedure, the update procedure for additionally learning the trained model using the text data and updating the parameters of the trained model, and the update procedure.
The computer runs ,
The update procedure is an information processing method characterized in that the update is performed in the background of the voice call. A program for making a computer function as each means in the information processing system according to any one of claims 1 to 5.

Images