JP6957933B2 - Information processing equipment, information processing methods and information processing programs - Google Patents

Description

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

In the above technical field, Patent Document 1 compares customer information, customer behavior history information, and a customer extraction rule regarded as a customer with a high churn risk, and a customer who satisfies the customer extraction rule has a high churn risk. The technology to extract as a customer is disclosed. Further, Non-Patent Document 1 discloses a method for detecting an utterance section. Non-Patent Document 2 discloses a method of emotion recognition. Non-Patent Document 3 discloses a machine learning method. Non-Patent Document 4 discloses a method for detecting an utterance section.

JP-A-2002-334200

Yusuke Kida and Tatsuya Kawahara, "Voice Activity Detection based on Optimally Weighted Combination of Multiple Features," Proc. INTERSPEECH 2005, pp.2621-2624, 2005. Florian Eyben, Martin Wollmer, and Bjorn Schuller, "openEAR --Introducing the Munich Open-Source Emotion and Affect Recognition Toolkit," 2009 3rd International Conference on Affective Computing and Intelligent Interaction and Workshops Pedregosa et al., "Scikit-learn: Machine Learning in Python," JMLR 12, pp. 2825-2830, 2011. JPYamron, I.Carp, L.Gillick, S.Lowe, and P.van Mulbregt, “A HIDDEN MARKOV MODEL APPROACH TO TEXT SEGMENTATION AND EVENT TRACKING,” IEEE International Conference on Acoustics, Speech and Signal Processing, pp.333- 336, 1998.

However, since the technique described in the above document uses the behavior history information of the customer, it is not possible to predict the churn risk even for the customer who contacts for the first time without the behavior history information.

An object of the present invention is to provide a technique for solving the above-mentioned problems.

In order to achieve the above object, the information processing device according to the present invention is
It acquires the audio data during a call between the customer and the operator, the call storage means for storing,
Based on the voice data, the utterance amount feature amount calculating means for calculating the utterance amount feature amount regarding the utterance amount of the customer and the operator during the call, and the utterance amount feature amount calculation means.
An emotion value feature amount calculating means for calculating an emotion value feature amount relating to an emotional value of the customer and the operator during the call based on the voice data, and an emotion value feature amount calculating means.
A customer extraction model storage means for storing a customer extraction model that predicts the churn risk of the customer based on the utterance amount feature amount and the emotion value feature amount.
Using the customer extraction model, and churn predicting means for predicting a churn risk of the customer from said speech content features and the prior SL emotion value feature quantity,
Equipped with.

In order to achieve the above object, the information processing method according to the present invention
It acquires the audio data during a call between the customer and the operator, the call storage step of storing,
Based on the voice data, the utterance amount feature amount calculation step for calculating the utterance amount feature amount regarding the utterance amount of the customer and the operator during the call, and the utterance amount feature amount calculation step.
An emotion value feature amount calculation step for calculating an emotion value feature amount relating to an emotional value of the customer and the operator during the call based on the voice data, and a step of calculating the emotion value feature amount.
With customer extraction model for predicting the lapse risk of the customer based on said emotion value feature quantity and said utterances amount characteristic amount, a lapse risk of the customer from said speech content features and the prior SL emotion value feature quantity Predicting churn prediction steps and
including.

In order to achieve the above object, the information processing program according to the present invention
It acquires the audio data during a call between the customer and the operator, the call storage step of storing,
Based on the voice data, the utterance amount feature amount calculation step for calculating the utterance amount feature amount regarding the utterance amount of the customer and the operator during the call, and the utterance amount feature amount calculation step.
An emotion value feature amount calculation step for calculating an emotion value feature amount relating to an emotional value of the customer and the operator during the call based on the voice data, and a step of calculating the emotion value feature amount.
With customer extraction model for predicting the lapse risk of the customer based on said emotion value feature quantity and said utterances amount characteristic amount, a lapse risk of the customer from said speech content features and the prior SL emotion value feature quantity Predicting churn prediction steps and
Let the computer run.

According to the present invention, it is possible to predict the churn risk even for a customer who has contacted for the first time without behavior history information.

It is a block diagram which shows the structure of the information processing apparatus which concerns on 1st Embodiment of this invention. It is a block diagram which shows the structure of the information processing apparatus which concerns on 2nd Embodiment of this invention. It is a block diagram which shows the structure of the emotion characteristic calculation part of the information processing apparatus which concerns on 2nd Embodiment of this invention. It is a figure which shows the hardware configuration of the information processing apparatus which concerns on 2nd Embodiment of this invention. It is a flowchart explaining the processing procedure of the information processing apparatus which concerns on 2nd Embodiment of this invention. It is a block diagram which shows the structure of the information processing apparatus which concerns on 3rd Embodiment of this invention. It is a figure which shows the hardware configuration of the information processing apparatus which concerns on 3rd Embodiment of this invention. It is a flowchart explaining the processing procedure of the information processing apparatus which concerns on 3rd Embodiment of this invention.

Hereinafter, embodiments for carrying out the present invention will be described in detail exemplarily with reference to the drawings. However, the configuration, numerical values, processing flow, functional elements, etc. described in the following embodiments are merely examples, and modifications and changes thereof are free, and the technical scope of the present invention is described below. It is not intended to be limited.

[First Embodiment]
The information processing device 100 as the first embodiment of the present invention will be described with reference to FIG. The information processing device 100 is a device that predicts the churn risk in a call based on the utterance volume characteristics and emotions in the call.

As shown in FIG. 1, the information processing device 100 includes a call storage unit 101, an utterance amount feature calculation unit 102, an emotion recognition unit 103, and a churn prediction unit 104. The call storage unit 101 acquires and stores a call between the customer and the operator. The utterance amount feature calculation unit 102 calculates the utterance amount feature amount of the stored call. The emotion recognition unit 103 recognizes the emotions in the utterance section. The churn prediction unit 104 predicts the churn risk in a call based on the utterance volume feature amount and the recognized emotion.

According to this embodiment, it is possible to predict the churn risk even for a customer who has contacted for the first time without behavior history information.

[Second Embodiment]
Next, the information processing apparatus according to the second embodiment of the present invention will be described with reference to FIGS. 2 to 5. FIG. 2 is a block diagram showing a configuration of an information processing device according to the present embodiment.

The information processing device according to the present embodiment can extract customers with a high risk of cancellation, customers who are expected to continue, etc. from calls at a call center or the like, and selectively and appropriately follow up with each customer. It is a device that can be implemented.

The information processing device 200 includes a call storage unit 201, an utterance amount feature calculation unit 202, an emotion recognition unit 203, an emotion feature calculation unit 204, a customer extraction model storage unit 205, a cancellation prediction unit 206, and an extraction customer output unit 207.

The call storage unit 201 acquires a call between the customer and the operator as call voice data. Then, the call storage unit 201 stores the acquired call voice data as a file or stream data.

When storing the call voice data, the call storage unit 201 assigns a call ID (Identifier) (call ID) for specifying the call (call) to the stored call voice data and stores the call voice data. Further, the call storage unit 201 detects the utterance section of the call voice data with respect to the call voice data to be stored, and adds speaker information, that is, operator information or customer information to each of the detected utterance sections. And remember. Furthermore, the call storage unit 201 adds utterance time information, that is, the utterance start time and the utterance end time to each of the detected utterance sections, and stores the utterance voice data to be stored. Therefore, when the call storage unit 201 stores the call voice data, the call ID, the operator information, the customer information, and the utterance time information are added to the stored call voice data and stored.

Various methods are disclosed as methods for detecting the utterance section, and for example, the method described in Non-Patent Document 1 may be used.

Regarding the call ID, for example, in a recent call center, the call ID can be acquired by a CTI (Computer Telephony Integration) system. Further, it is possible to record data as different channels for the signal on the customer side and the signal on the operator side. Therefore, the call ID and the speaker information can be given by using the call ID and the channel information acquired by the CTI system.

Further, the utterance time information can be given by acquiring the absolute time of the detected utterance section or the time information specified by the system for the start and end.

The utterance amount characteristic calculation unit 202 calculates the characteristics of the utterance amount between the operator and the customer from the time information of each utterance section stored in the call storage unit 201. That is, the total time of each utterance section of the operator and the customer in each call, the ratio of the total time of the utterance section between the operator and the customer in each call, the average value of the utterance section lengths of the operator and the customer in each call, and the variance. Use at least one of values, median, mode, or any other value calculated by statistical processing.

The emotion recognition unit 203 has valence (emotion value indicating the quality of emotion, degree of positive and negative emotion), arousal (awakening degree indicating arousal or calmness), for each speech section stored in the call storage unit 201. Calculate at least one emotional value of emotional excitement), conviction, and expectation. Various methods are disclosed as the emotion recognition method, and for example, the method described in Non-Patent Document 2 can be used.

Each emotion is calculated using a numerical value in a predetermined range or a finite number of categories as an emotion value. That is, the emotion recognition unit 203 calculates, for example, a value in the range of -3 to 3 or a category such as {-3, -2, -1, 0, 1, 2, 3} for valence. The same applies to arousal, conviction, and expectation, but the range and category specified for each emotion do not have to be the same.

The emotional feature calculation unit 204 calculates the emotional feature amount of the operator and the customer from the emotional value of each utterance section calculated by the emotional recognition unit 203. The emotion feature is calculated by the sum of the emotion values of the operator and the customer in each call, the emotion value ratio, the average value of the emotion values, the variance value, the median value, the mode value, or other statistical processing. At least one of the values is included.

FIG. 3 is a block diagram showing a configuration of an emotional feature calculation unit of the information processing device according to the present embodiment. The emotional feature calculation unit 204 may calculate the emotional feature amount of the operator and the customer by using only the emotional value of each utterance section included in the specific section in the call voice data according to the configuration shown in FIG. The emotional feature calculation unit 204 shown in FIG. 3 includes a topic section detection unit 241, a narrowing-down unit 242, and a feature calculation unit 243.

The topic section detection unit 241 detects a topic section included in the call voice data stored in the call storage unit 201, and assigns a topic start time and a topic end time to each of the detected topic sections. Topics include, for example, opening, closing, product / service explanation, fee explanation, solicitation, hearing, personal information confirmation, paperwork, and the like.

Various methods are disclosed as methods for detecting the topic section. For example, voice recognition and the methods described in Non-Patent Document 4 can be used in combination. Specifically, a word string obtained by voice recognition can be divided into a plurality of topic sections by using a topic model in which the frequency of appearance of words for various topics is learned in advance.

Further, the topic section detection unit 241 may detect the topic section based on the time information and the number of utterances. For example, the first 30 seconds section and the first 5 utterance sections of a call can be detected as the opening topic section. In addition, the last 30 seconds section and the last 5 utterance sections of the call can be detected as the closing topic section. Further, the latter half 1/3 section of the call (for example, a section of 6 to 9 minutes in the case of a 9-minute call) may be detected as one topic section.

The narrowing-down unit 242 outputs only the utterance section included in the specific topic detected by the topic section detection unit 208 to the feature calculation unit 243 among the utterance sections to which the emotion recognition unit 203 has given the emotion value. For example, only the utterance section included in the closing topic section and their emotional values are output to the feature calculation unit 243.

The feature calculation unit 243 calculates the emotional feature amount of the operator and the customer from the emotional value of each utterance section output from the narrowing-down unit 242. The emotional feature is calculated by the sum of the emotional values of the operator and the customer in a specific topic section, the emotional value ratio, the average emotional value, the variance value, the median value, the mode value, or other statistical processing. At least one of the values to be created is included. For example, the sum or average value of the valence of the customer's utterance and the emotional value of the expectation included in the topic section of the opening or closing is calculated as the emotional feature amount.

The customer extraction model storage unit 205 cancels the contract by using the emotional feature amount of the operator and the customer calculated by the emotional feature calculation unit 204 and the speech amount of the operator and the customer calculated by the speech amount characteristic calculation unit 202 as the feature amount. Store the model predicted by high-risk customers as a customer extraction model.

In the customer extraction model, the correct label information of whether or not the contract is canceled corresponding to each call data is given in advance, and the model for predicting the cancellation risk is learned by the machine learning method. Various methods of machine learning methods are disclosed. For example, Non-Patent Document 3 discloses the details thereof and an executable program. For the churn prediction, a prediction method corresponding to the learning method of the model can be used.

The churn prediction unit 206 predicts (calculates) the churn risk for each call using the customer extraction model.

The extracted customer output unit 207 compares the predicted churn risk of each call with the churn risk threshold specified in advance and the ratio of high churn risk persons in the number of target calls, and the customer with a high churn risk, that is, the churn risk. Extract customers with high calls.

In the description of the present embodiment, an example of predicting churn has been described, but the present embodiment can be similarly applied to continuous prediction.

FIG. 4 is a block diagram illustrating a hardware configuration of the information processing apparatus 200 according to the present embodiment. The CPU (Central Processing Unit) 410 is a processor for arithmetic control, and realizes the functional component of the information processing device 200 of FIG. 2 by executing a program. The CPU 410 has a plurality of processors, and may execute different programs, modules, tasks, threads, and the like in parallel. The ROM (Read Only Memory) 420 stores fixed data such as initial data and programs, and other programs. Further, the network interface 430 communicates with other devices and the like via the network. The CPU 410 is not limited to one, and may be a plurality of CPUs, or may include a GPU (Graphics Processing Unit) for image processing. Further, it is desirable that the network interface 430 has a CPU independent of the CPU 410 and writes or reads transmission / reception data in the area of the RAM (Random Access Memory) 440. Further, it is desirable to provide a DMAC (Direct Memory Access Controller) for transferring data between the RAM 440 and the storage 450 (not shown). Further, the CPU 410 recognizes that the data has been received or transferred to the RAM 440 and processes the data. Further, the CPU 410 prepares the processing result in the RAM 440, and the subsequent transmission or transfer is left to the network interface 430 or the DMAC.

The RAM 440 is a random access memory used by the CPU 410 as a temporary storage work area. The RAM 440 has an area for storing data necessary for realizing the present embodiment. The call ID 441 is data that identifies a call (call) between the operator and the customer, and is assigned to each call. The utterance volume feature 442 is a feature of the utterance volume calculated in a call between the operator and the customer. The emotion value 443 is a calculated emotion value of each utterance section in each call. The emotional feature 444 is an emotional feature calculated from the emotional value of each utterance section, and the customer extraction model 445 is a model for predicting a customer with a high churn risk. The churn prediction 446 is a prediction of the churn risk predicted for each call using the customer extraction model.

The input / output data 447 is data input / output via the input / output interface 460. The transmission / reception data 448 is data transmitted / received via the network interface 430. Further, the RAM 440 has an application execution area 449 for executing various application modules.

The storage 450 stores a database, various parameters, and the following data or programs necessary for realizing the present embodiment. The storage 450 stores the utterance amount feature calculation module 453, the emotion recognition module 454, the emotion feature calculation module 455, and the churn prediction module 457.

The utterance amount characteristic calculation module 453 is a module for calculating the utterance amount characteristic of the call between the operator and the customer. The emotion recognition module 454 is a module for calculating an emotion value in the utterance section of a call between the operator and the customer. The emotional feature calculation module 455 is a module for calculating emotional features from the calculated emotional values of the utterance section. The churn prediction module 457 is a module that predicts the churn risk in a call based on the utterance volume feature amount and the recognized emotion. These modules 453 to 457 are read by the CPU 410 into the application execution area 449 of the RAM 440 and executed. The control program 456 is a program for controlling the entire information processing apparatus 200.

The input / output interface 460 interfaces input / output data with an input / output device. A display unit 461 and an operation unit 462 are connected to the input / output interface 460. Further, a storage medium 464 may be further connected to the input / output interface 460. Further, a speaker 463 which is a voice output unit, a microphone (not shown) which is a voice input unit, or a GPS position determination unit may be connected. The RAM 440 and the storage 450 shown in FIG. 4 do not show programs or data related to general-purpose functions and other feasible functions of the information processing device 200.

FIG. 5 is a flowchart illustrating a processing procedure of the information processing apparatus 200 according to the present embodiment. This flowchart is executed by the CPU 410 of FIG. 4 using the RAM 440, and realizes the functional component of the information processing apparatus 200 of FIG.

In step S501, the information processing apparatus 200 acquires a call between the operator and the customer. In step S503, the information processing device 200 detects the utterance section of the acquired call, and adds speaker information and utterance time information to each of the detected utterance sections. In step S505, the information processing device 200 calculates the characteristics of the amount of speech between the operator and the customer in the call. In step S507, the information processing device 200 calculates the emotion value of the utterance section. In step S509, the information processing device 200 calculates an emotional feature from the calculated emotional value of the utterance section. In step S513, the information processing apparatus 200 predicts the churn risk for each call using the customer extraction model. In step S515, the information processing apparatus 200 extracts the customers of the call having a high churn risk based on the predicted churn risk.

According to this embodiment, by using the customer's inquiries about products and services and the call data at the time of application, it is possible to predict the cancellation even for the customer who cannot acquire the behavior history such as the first contact. ..

Further, according to the present embodiment, information such as which of the operator and the customer is proactively having a conversation as a utterance feature is calculated by using statistical processing including information that cannot be manually ruled. Similarly, a model that estimates churn risk with high accuracy by calculating using statistical processing including information that cannot be manually ruled from emotional values, rather than emotional values that can be interpreted by humans, such as satisfaction and dissatisfaction as emotional characteristics. Can be learned and predicted.

Cancellation or non-cancellation is not determined by inquiries about customers' products and services or exchanges at the time of application. Some customers tend to cancel when they purchase an unintended product through upsell or cross-sell. In such a case, the present inventor is effective in inquiring about the service and conversation information of the exchange at the time of application in predicting the cancellation or continuation, and in that case, the utterance feature and the emotional feature are effective. I found that.

Therefore, in the present embodiment, the predicted (calculated) risk is compared with the threshold value of the churn risk specified in advance and the ratio of high churn risk persons in the number of target calls, and the customers judged to have a high churn risk are extracted. .. As a result, in the present embodiment, it is possible to extract customers who have a high risk of churn, mainly due to telephone correspondence.

The present inventor has also found that whether or not a customer cancels in the future depends on the emotions of the operator and the customer in a specific topic section in the call. For example, we found that the risk of churn is low when the customer's positive emotions and expectations are high in the topic section of opening and closing.

Therefore, in the present embodiment, the churn risk is estimated based on the emotional characteristics focusing on the emotions of the operator and the customer in a specific topic section in the call. As a result, in the present embodiment, customers with a high risk of churn can be extracted with high accuracy.

[Third Embodiment]
Next, the information processing apparatus according to the third embodiment of the present invention will be described with reference to FIGS. 6 to 9. FIG. 6 is a block diagram showing a configuration of an information processing device according to the present embodiment. The information processing apparatus according to the present embodiment is different from the second embodiment in that it has a phrase storage unit, a phrase recognition unit, and a phrase feature calculation unit. Since other configurations and operations are the same as those in the second embodiment, the same configurations and operations are designated by the same reference numerals and detailed description thereof will be omitted.

The information processing device 600 has a phrase storage unit 601, a phrase recognition unit 602, and a phrase feature calculation unit.

The phrase storage unit 601 stores a phrase designated in advance. The phrase storage unit 601 stores phrases that can be regarded as situations for at least one of convincing, non-convincing, brute force, and commitment to fees. For example, if you are satisfied, "OK", "Yes", and so on. In the case of non-convincing, "for the time being", "for the time being", "for the time being", etc. In the case of forcibleness, "let's decide", "this is fine" and so on. In the case of sticking to the price, there are "price", "charge", "free shipping", "consumption tax", "mottainai" and so on.

The phrase recognition unit 602 detects and recognizes the appearance of the phrase stored in the phrase storage unit 601. The phrase recognition unit 602 performs voice recognition and word spotting, detects the appearance of a phrase stored in the phrase storage unit 601 and recognizes it.

The phrase feature calculation unit 603 calculates the phrase features of the operator and the customer. The phrase feature calculation unit 603 uses the recognition result (detection result) by the phrase recognition unit 602 to perform the following calculations for at least one of convincing, non-convincing, forcible, and sticking to the fee. That is, the phrase feature calculation unit 603 calculates the number of appearances of the phrase, which is the number of appearances of the phrase stored in the phrase storage unit 601. Further, the phrase feature calculation unit 603 calculates the number of appearances of the normalized phrase obtained by normalizing the number of appearances of the phrase by the utterance time, and the number of appearances of the phrase for each time interval specified in advance. The phrase feature calculation unit 603 calculates at least one of the number of phrase appearances, the number of normalized phrases appearing, and the number of phrase appearances for each predetermined time interval.

Similar to the second embodiment, the customer extraction model storage unit 205 has the emotional feature amount of the operator and the customer calculated by the emotional feature calculation unit 204 and the utterance of the operator and the customer calculated by the utterance amount feature calculation unit 202. A model is generated by adding the following feature quantities to the quantity features. That is, the customer extraction model storage unit 205 adds the phrase features of the operator and the customer calculated by the phrase feature calculation unit 603 to the emotional features and the utterance features, and predicts the customer with a high churn risk. Generate and store the extraction model. The learning (generation) of the customer extraction model and the prediction using the learned model are the same as those in the second embodiment except that the above-mentioned feature amount (phrase feature amount) is used.

FIG. 7 is a block diagram illustrating a hardware configuration of the information processing apparatus 600 according to the present embodiment. The CPU (Central Processing Unit) 410 is a processor for arithmetic control, and realizes the functional component of the information processing device 600 of FIG. 2 by executing a program.

The RAM 740 is a random access memory used by the CPU 410 as a temporary storage work area. The RAM 740 has an area for storing data necessary for realizing the present embodiment. The storage phrase 741 is a phrase designated and stored in advance.

The storage 750 stores a database, various parameters, and the following data or programs necessary for realizing the present embodiment. The storage 850 stores the phrase 751, the phrase recognition module 752, and the phrase feature calculation module 753. Phrase 751 is a pre-designated phrase. The phrase recognition module 752 is a module that detects and recognizes the appearance of a memorized phrase by voice recognition or word spotting. The phrase feature calculation module 753 uses the phrase recognition result (detection result) to normalize the number of appearances of the phrase and the number of appearances of the phrase by the utterance time, and at least one of the number of appearances of the phrase at a predetermined time interval. It is a module to calculate. These modules 725 to 753 are read by the CPU 410 into the application execution area 449 of the RAM 840 and executed.

FIG. 8 is a flowchart illustrating a processing procedure of the information processing apparatus 600 according to the present embodiment. This flowchart is executed by the CPU 410 of FIG. 7 using the RAM 840, and realizes the functional component of the information processing apparatus 600 of FIG.

In step 801 the information processing apparatus 600 detects and recognizes the appearance of a phrase designated in advance. In step S803, the information processing device 600 calculates the characteristics of the recognized phrase. The characteristics of the phrase are, for example, the number of appearances of the phrase, the number of appearances of the phrase normalized by the utterance time, the number of appearances of the phrase at a predetermined time interval, and the like.

According to the present embodiment, regarding convincing, non-convincing, forcibleness, and commitment to charges, related phrases are detected, phrase features are calculated, and a model is generated by adding these phrase features, so that cancellation or continuation is performed. Can be predicted with high accuracy by taking into account the content of the statement. [Other Embodiments]
Although the invention of the present application has been described above with reference to the embodiment, the invention of the present application is not limited to the above embodiment. Various changes that can be understood by those skilled in the art can be made within the scope of the present invention in terms of the structure and details of the present invention. Also included in the scope of the present invention are systems or devices in which the different features contained in each embodiment are combined in any way.

Further, the present invention may be applied to a system composed of a plurality of devices, or may be applied to a single device. Furthermore, the present invention is also applicable when the information processing program that realizes the functions of the embodiment is supplied directly or remotely to the system or device. Therefore, in order to realize the functions of the present invention on a computer, a program installed on the computer, a medium containing the program, and a WWW (World Wide Web) server for downloading the program are also included in the scope of the present invention. .. In particular, at least a non-transitory computer readable medium containing a program that causes a computer to execute the processing steps included in the above-described embodiment is included in the scope of the present invention.

[Other expressions of the embodiment]
Some or all of the above embodiments may also be described, but not limited to:
(Appendix 1)
A call storage means that acquires and stores calls between the customer and the operator,
The utterance amount feature calculation means for calculating the utterance amount feature amount of the memorized call, and the utterance amount feature calculation means.
An emotion recognition means for recognizing emotions in the utterance section and
A churn prediction means that predicts the churn risk in the call based on the utterance amount feature amount and the recognized emotion.
Information processing device equipped with.
(Appendix 2)
The emotion recognition means further calculates the recognized emotion value of the emotion,
An emotional feature calculation means for calculating an emotional feature, which is an emotional feature, from the emotional value,
A customer extraction model storage means for storing a pre-learned customer extraction model for predicting and extracting a customer with a high churn risk based on the utterance amount feature amount and the emotional feature.
With more
The information processing device according to Appendix 1, wherein the churn prediction means predicts the churn risk based on the customer extraction model.
(Appendix 3)
The emotional feature calculation means
A topic section detection means that detects a specific topic section included in a call,
A narrowing-down means for outputting the emotional value of the utterance section included in the specific topic section, and
A feature calculation means for calculating an emotional feature, which is an emotional feature, from an emotional value output by the narrowing-down means, and a feature calculation means.
The information processing apparatus according to Appendix 2, further comprising.
(Appendix 4)
Phrase storage means to memorize pre-specified phrases,
A phrase recognition means that detects and recognizes the appearance of a phrase stored in the phrase storage means in the call.
Phrase feature calculation means for calculating phrase features that are features of appearing phrases, and
With more
The information processing device according to any one of Supplementary note 1 to 3, wherein the churn prediction means further predicts the churn risk based on the phrase feature.
(Appendix 5)
The call storage means detects the utterance section of the call, adds speaker information and utterance time information to the detected utterance section, and stores the utterance. The information processing apparatus according to item 1.
(Appendix 6)
The information processing apparatus according to any one of Supplementary note 1 to 5, wherein the emotion includes at least one of valence, atomic, convincing and anticipatory.
(Appendix 7)
The topic section includes at least one of the opening, closing, product / service explanation, charge explanation, solicitation, hearing, personal information confirmation and paperwork in the call.
The information processing device according to any one of Supplementary note 3 to 5, wherein the emotion includes at least one of valence and anticipation.
(Appendix 8)
The information processing apparatus according to any one of Supplementary note 2 to 7, wherein the emotional feature includes at least one of an emotional value ratio, an average value, a variance value, a median value, and a mode value of the emotional value.
(Appendix 9)
The phrase feature includes at least one of the number of appearances of the phrase, which is the number of appearances of the phrase, the number of appearances of the normalized phrase obtained by normalizing the number of appearances of the phrase by the speech time, and the number of appearances of the phrase at each predetermined time interval. The information processing apparatus according to any one of items 7 to 7.
(Appendix 10)
A call memory step that captures and stores calls between the customer and the operator,
The utterance amount feature calculation step for calculating the utterance amount feature amount of the memorized call, and
The emotion recognition step for recognizing the emotions in the utterance section,
A churn prediction step that predicts the churn risk in the call based on the utterance amount feature amount and the recognized emotion.
Information processing methods including.
(Appendix 11)
A call memory step that captures and stores calls between the customer and the operator,
The utterance amount feature calculation step for calculating the utterance amount feature amount of the memorized call, and
The emotion recognition step for recognizing the emotions in the utterance section,
A churn prediction step that predicts the churn risk in the call based on the utterance amount feature amount and the recognized emotion.
An information processing program that causes a computer to execute.

Claims (10)

It acquires the audio data during a call between the customer and the operator, the call storage means for storing,
Based on the voice data, the utterance amount feature amount calculating means for calculating the utterance amount feature amount regarding the utterance amount of the customer and the operator during the call, and the utterance amount feature amount calculation means.
An emotion value feature amount calculating means for calculating an emotion value feature amount relating to an emotional value of the customer and the operator during the call based on the voice data, and an emotion value feature amount calculating means.
A customer extraction model storage means for storing a customer extraction model that predicts the churn risk of the customer based on the utterance amount feature amount and the emotion value feature amount.
Using the customer extraction model, and churn predicting means for predicting a churn risk of the customer from said speech content features and the prior SL emotion value feature quantity,
Information processing device equipped with. The call storage means has speaker information indicating whether the speaker is the customer or the operator with respect to the voice data of the utterance section detected based on the voice data, the utterance start time and the utterance end. and speech time information including the time of, granted to the remembers,
The utterance amount feature amount calculating means calculates the utterance amount feature amount by using the speaker information and the utterance time information.
The information processing device according to claim 1, wherein the emotion value feature amount calculating means includes an emotion value calculating means for calculating an emotion value in the utterance section, and calculates the emotion value feature amount using the emotion value. The utterance amount feature amount is
The total time of speech of the customer and the operator,
The ratio of the total time of the customer's utterance to the total time of the operator's utterance,
The average value of the length of the utterance section of the customer and the average value of the length of the utterance section of the operator.
The dispersion value of the length of the utterance section of the customer and the dispersion value of the length of the utterance section of the operator.
The median length of the utterance section of the customer and the median length of the utterance section of the operator.
The mode of the length of the utterance section of the customer and the mode of the length of the utterance section of the operator.
The information processing apparatus according to claim 2, further comprising at least one of the above. The emotional value is according to any one of claims 1 to 3 , which includes at least one of valence (emotion value indicating the quality of emotion) , arousal (alertness indicating arousal or calmness), and convincing and expectation. The information processing device described. The information processing apparatus according to any one of claims 1 to 4 , wherein the emotion value feature amount includes at least one of the ratio, average value, dispersion value, median value, and mode value of the emotion values. The emotion value feature amount calculation means is
A topic section detection means that detects a specific topic section included in a call,
During the speech data of the speech period, a narrowing means for outputting the feeling information value refine the speech data of the speech interval included in the specific topic section,
A feature quantity calculating means for calculating the emotional value feature quantity from the emotion value said narrowing means has outputted,
The information processing apparatus according to claim 2 or 3. The specific topic section, the opening of the topic section that is included in the call, the closing of the topic sections, products and services described topic sections of, topic section of the fee description, solicitation topic section of, hearing of the topic section, the personal information confirmation of the information processing apparatus according to at least one topic section and topic section paperwork to including claim 6. Phrase storage means to memorize pre-specified phrases,
On the basis of the audio data, the phrase storage unit of the stored phrases, recognizing the occurrence during the call, and phrase feature quantity calculating means for calculating the phrase characteristics of the customer and the operator in the call ,
With more
The customer extraction model storage means stores a customer extraction model that predicts the churn risk of the customer based on the utterance amount feature amount, the emotion value feature amount, and the phrase feature amount.
The churn prediction means is any one of claims 1 to 7 , which predicts the churn risk of the customer from the utterance amount feature amount, the emotion value feature amount, and the phrase feature amount using the customer extraction model. The information processing device described in. It acquires the audio data during a call between the customer and the operator, the call storage step of storing,
Based on the voice data, the utterance amount feature amount calculation step for calculating the utterance amount feature amount regarding the utterance amount of the customer and the operator during the call, and the utterance amount feature amount calculation step.
An emotion value feature amount calculation step for calculating an emotion value feature amount relating to an emotional value of the customer and the operator during the call based on the voice data, and a step of calculating the emotion value feature amount.
With customer extraction model for predicting the lapse risk of the customer based on said emotion value feature quantity and said utterances amount characteristic amount, a lapse risk of the customer from said speech content features and the prior SL emotion value feature quantity Predicting churn prediction steps and
Information processing methods including. It acquires the audio data during a call between the customer and the operator, the call storage step of storing,
Based on the voice data, the utterance amount feature amount calculation step for calculating the utterance amount feature amount regarding the utterance amount of the customer and the operator during the call, and the utterance amount feature amount calculation step.
An emotion value feature amount calculation step for calculating an emotion value feature amount relating to an emotional value of the customer and the operator during the call based on the voice data, and a step of calculating the emotion value feature amount.
With customer extraction model for predicting the lapse risk of the customer based on said emotion value feature quantity and said utterances amount characteristic amount, a lapse risk of the customer from said speech content features and the prior SL emotion value feature quantity Predicting churn prediction steps and
An information processing program that causes a computer to execute.

Images