Dialogue analysis device, dialogue analysis program, and dialogue analysis method

Abstract

For a conversation between a contact center operator and a customer, the number of occurrences of a combination of an operator response group into which an operator utterance is classified and a customer response group into which a customer utterance is classified is tallied. [Solution] The dialogue analysis device disclosed herein comprises a dialogue division unit that divides dialogue data recording dialogue between an operator and a customer into multiple dialogue parts including operator utterances and customer utterances, and a counting unit that counts the number of occurrences of combinations of operator response groups and customer response groups by classifying operator utterances included in the dialogue parts into the same operator response group for each similar operator response, and by classifying customer utterances included in the dialogue parts into the same customer response group for each similar customer response.

Description

[Technical Field]

[0001] The present disclosure relates to a dialogue analysis device that analyzes an operator's response by evaluating the customer's reaction in a dialogue between the operator and a customer at a contact center, call center, or the like. [Background technology]

[0002] In contact center operations, it is desirable to improve not only customer satisfaction but also the skills of operators. Conventionally, systems that evaluate conversations between contact center operators and customers have been known. For example, the automatic scoring device for operator-customer conversations in prior art document 1 analyzes the audio of conversations between contact center operators and customers, scores the conversations according to the characteristics of the contact center, and extracts audio files in order of ranking according to the evaluation. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2014-123813 Summary of the Invention [Problem to be solved by the invention]

[0004] Typically, contact center operations require operators to appropriately change intonation, words, and other aspects of their speech depending on the content of the conversation and the customer. This requires operators to understand the content and context of the conversation, and then respond appropriately based on the customer's reaction. However, conventional technology analyzes one-on-one conversations between a call center operator and a customer to evaluate the conversation between the operator and the customer, and the same evaluation may not be obtained when the same operator serves another customer.

[0005] The present disclosure has been made in consideration of the above-mentioned problems, and aims to identify combinations of operator response groups into which the operator's utterances included in the dialogue portion of a dialogue between a contact center operator and a customer are classified, and customer response groups into which the customer's utterances are classified, and to tally the number of occurrences of each combination. [Means for solving the problem]

[0006] The dialogue analysis device according to the present disclosure includes a dialogue division unit that divides dialogue data recording a dialogue between an operator and a customer into a plurality of dialogue portions each including an operator utterance and a customer utterance; a dialogue analysis unit that classifies the operator utterances included in the plurality of dialogue portions into the same operator response group for each similar operator response, and classifies the customer utterances included in the plurality of dialogue portions into the same customer response group for each similar customer response, thereby identifying, for each dialogue portion, a combination of the operator response group into which the operator utterances included in the dialogue portion have been classified and the customer response group into which the customer utterances have been classified; and a counting unit that counts the number of occurrences of combinations of operator response groups and customer response groups for each combination in the dialogue data. [Effects of the Invention]

[0007] According to the dialogue analysis device of the present disclosure, for a dialogue between a contact center operator and a customer, it is possible to identify combinations of operator response groups into which the operator's utterances included in the dialogue portion are classified and customer response groups into which the customer's utterances are classified, and to count the number of occurrences of each combination. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a configuration diagram of a dialogue analysis system according to a first embodiment. [Figure 2] FIG. 1 is a functional configuration diagram of a dialogue analysis device according to a first embodiment. [Figure 3] FIG. 1 is a hardware configuration diagram of a dialogue analysis device according to a first embodiment. [Figure 4] 3 shows an example of dialogue data of the dialogue analysis device according to the first embodiment. [Figure 5] 3 shows an example of dialogue partial data of the dialogue analysis device according to the first embodiment. [Figure 6] 3 shows an example of dialogue portion analysis data of the dialogue analysis device according to the first embodiment. [Figure 7] 3 shows an example of dialogue portion customer evaluation index data of the dialogue analysis device according to the first embodiment. [Figure 8] 3 shows an example of table generation information of the dialogue analysis device according to the first embodiment. [Figure 9] 3 is a dialogue analysis processing flow diagram of the dialogue analysis device according to the first embodiment. [Figure 10] FIG. 3 is a flowchart showing a customer evaluation process performed by the dialogue analysis device according to the first embodiment. [Figure 11] 10 shows an example of a dialogue analysis result (combination summary table) screen display of the dialogue analysis device according to the first embodiment. [Figure 12] 3 is a flowchart showing the dialogue partial classification process of the dialogue reaction analysis device according to the first embodiment. [Figure 13] 4 is a flowchart showing the process of extracting a representative utterance from an operator-handling group performed by the dialogue analysis device according to the first embodiment. [Figure 14] 2 shows an example (part 2) of dialogue portion analysis data of the dialogue analysis device according to the first embodiment. [Figure 15] 10 shows an example of a dialogue analysis result (dialogue list) screen display of the dialogue analysis device according to the first embodiment. [Figure 16] FIG. 10 is a dialogue analysis processing flow diagram of the dialogue analysis device according to the second embodiment. [Figure 17] 10 shows an example of speech part analysis data of the dialogue analysis device according to the second embodiment. [Figure 18] 10 shows an example of table generation information of the dialogue analysis device according to the second embodiment. [Figure 19] 10 shows an example of a dialogue analysis result (combination summary table) screen display of the dialogue analysis device according to the second embodiment. [Figure 20] FIG. 11 is a dialogue analysis processing flow diagram of the dialogue analysis device according to the third embodiment. [Figure 21] 11 shows an example of speech part analysis data of the dialogue analysis device according to the third embodiment. [Figure 22] 13 shows an example of table generation information of the dialogue analysis device according to the third embodiment. [Figure 23] 11 shows an example of a dialogue analysis result (combination summary table) screen display of the dialogue analysis device according to the third embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] [Embodiment 1] A first embodiment will be described below. In the first embodiment, a dialogue analysis system, which is a system configuration of a dialogue analysis device and a display terminal, classifies operator utterances included in dialogue portions in a dialogue between an operator and a customer into operator response groups based on similar operator responses, and classifies customer utterances included in dialogue portions into customer response groups based on similar customer responses, thereby identifying combinations of operator response groups and customer response groups in each dialogue portion and tallying the number of occurrences of each combination. Furthermore, a process for evaluating customer response groups based on evaluation information in which customers evaluate the operator responses will be described.

[0010] ***Configuration of the First Embodiment*** 1 is a diagram showing a schematic configuration of a dialogue analysis system 1 according to Embodiment 1. The dialogue analysis system 1 includes a dialogue analysis device 10 and a display terminal 20. The dialogue analysis device 10 and the display terminal 20 are connected via a network 2. Communication is carried out between the dialogue analysis device 10 and the display terminal 20 via the network 2. Specific examples of the network 2 include a LAN or the Internet. LAN is an abbreviation for Local Area Network. The dialogue analysis device 10 is a device that specifies and executes instructed processing. A specific example of the dialogue analysis device 10 is a computer such as a server.

[0011] 2 is a block diagram showing an example configuration of a dialogue analysis device 10. The dialogue analysis device 10 includes a memory unit 11, a communication unit 12, a dialogue segmentation unit 13, a dialogue analysis unit 14, a counting unit 15, a customer evaluation unit 16, and a table generation unit 17. The memory unit 11 stores dialogue data 111, dialogue portion data 112, dialogue portion analysis data 113, customer evaluation index data 114, dialogue portion customer evaluation index data 115, group combination data 116, and customer response group evaluation data 117. With this configuration, the dialogue analysis device 10 provides a dialogue analysis service that identifies combinations of operator response groups and customer response groups in a dialogue and counts the number of occurrences of each combination.

[0012] The components of the dialogue analysis device 10 will be described below with reference to Fig. 2. The communication unit 12 connects to the network 2 and performs various communications required for dialogue analysis. The communication unit 12 transmits information on the results of analysis by the dialogue analysis device 10 to the display terminal 20.

[0013] The dialogue division unit 13 divides the dialogue data 111 into dialogue portions including at least one utterance by the operator (operator utterance) and at least one utterance by the customer (customer utterance). For example, the dialogue division unit 13 divides the dialogue data 111 into a plurality of dialogue portions by performing semantic analysis or context analysis on the dialogue data 111 to divide the dialogue into dialogue portions based on the content of the dialogue, or by setting a predetermined time for division and dividing the dialogue at each predetermined time. Then, the dialogue division unit 13 generates dialogue portion data 112, which is information on the plurality of dialogue portions into which the dialogue data 111 is divided, and stores the data in the storage unit 11.

[0014] Here, when the dialogue division unit 13 acquires the emotional value corresponding to each utterance, it may divide the dialogue data 111 into a plurality of dialogue parts based on the change in the emotional values ​​of the operator and the customer. The emotional value indicates an emotional state, specifically, an emotional state such as joy, sadness, expectation, anger, fear, disgust, trust, surprise, etc., expressed as a numerical value.

[0015] The dialogue analysis unit 14 classifies the operator utterances included in each dialogue segment into one of a plurality of operator response groups for each similar operator response, and classifies the customer utterances included in each dialogue segment into one of a plurality of customer response groups for each similar customer response, based on the dialogue segment data 112. After classification, the unit 14 identifies the combination of the operator response group and the customer response group for each dialogue segment, and generates dialogue segment analysis data 113 including information on the combinations. Then, the dialogue analysis unit 14 generates dialogue portion analysis data 113 including information on operator response groups that classify operator utterances included in multiple dialogue portions and customer response groups that classify customer utterances included in multiple dialogue portions, and stores the data in the memory unit 11. Here, the dialogue analysis unit 14 also has a function of identifying a representative utterance that is representative of each of the multiple agent response groups. When the representative utterance is identified, the dialogue analysis unit 14 may generate dialogue portion analysis data 113 that includes representative utterance indicators that indicate indicators of representative utterances in each agent response group.

[0016] When the dialogue analysis unit 14 classifies utterances into operator response groups and customer response groups, it analyzes based on the dialogue portion data 112, acquires features obtained from the utterances included in the dialogue portions, and classifies each utterance into a group. For example, the features can be acquired by using TF-IDF (Term Frequency - Inverse Document Frequency) to acquire features according to the appearance frequency and word importance of words included in the utterances, or by using Word2vec to acquire features according to the similarity between utterances, and then quantifying the features obtained from the utterances.

[0017] The dialogue analysis unit 14 can also analyze each utterance based on the dialogue data 111 or the dialogue portion data 112, and obtain an emotional value corresponding to each utterance to use as a feature. For example, if the dialogue data 111 is text data, a language model such as BERT (Bidirectional Encoder Representations from Transformers) is used to quantify the emotional state of the speaker from the content of their utterance, and obtain the emotional value. When calculating the emotional value from the audio file itself that records the call, the speaker's emotional value may be obtained by analyzing the speaker's voice using voice recognition.

[0018] The counting unit 15 calculates the number of occurrences of combinations of operator response groups and customer response groups in the dialogue portions based on the dialogue portion analysis data 113. Group combination data 116 including the obtained number of occurrences of each combination of operator response groups and customer response groups is generated and stored in the storage unit 11.

[0019] The customer evaluation unit 16 generates dialogue portion customer evaluation index data 115 by calculating a customer evaluation index for each dialogue portion based on customer evaluation index data 114 in which the customer evaluated the operator's response, and stores the data in the memory unit 11. Furthermore, based on the dialogue portion analysis data 113 and the dialogue portion customer evaluation index data 115, a customer evaluation index is calculated for each customer reaction group to generate customer reaction group evaluation data 117, which is stored in the storage unit 11.

[0020] The table generation unit 17 generates combination summary table information based on the group combination data 116 and the customer response group evaluation data 117, with the number of occurrences of combinations of operator response groups and customer response groups as table data, and displays the combination summary table information on the display terminal 20.

[0021] Here, the display terminal 20 is an information processing device used by a user who performs dialogue analysis. The display terminal 20 displays the results of dialogue analysis performed by the dialogue analysis device 10 in accordance with operations by the user. Although not shown, the display terminal 20 includes a storage unit, a communication unit, a display unit, and an operation reception unit. Note that the display terminal 20 may only have the functions of a communication unit that receives the results of the dialogue analysis from the dialogue analysis device 10 and a display unit that displays the results.

[0022] In this embodiment, a conversation between an operator and a customer at a contact center is described as an example, but the present invention can also be applied to a conversation between multiple people as long as the speaker is identified.

[0023] Each component of the dialogue analysis device 10 configured as described above is configured by a computer including a processor 51, a memory 52, and a signal input / output unit 53, as shown in FIG. 3. The functions of the storage unit 11, the communication unit 12, the dialogue segmentation unit 13, the dialogue analysis unit 14, the aggregation unit 15, the customer evaluation unit 16, and the table generation unit 17 are realized by this computer. That is, the computer's memory 52 stores programs for realizing the functions of the communication unit 12, the dialogue segmentation unit 13, the dialogue analysis unit 14, the aggregation unit 15, the customer evaluation unit 16, and the table generation unit 17. In addition, various information stored in the storage unit 11 is stored in the memory 52. ​​The processor 51 executes arithmetic processing related to the functions of the dialogue analysis device 10 based on the programs stored in the memory 52.

[0024] 4 is an example of a data configuration diagram of the dialogue data 111. The dialogue data 111 is data in which a call between an operator and a customer is recorded from start to finish, and the speaker is identified by speaker recognition. Specifically, the dialogue data 111 is text data, and includes a call number that identifies the call between the operator and the customer from start to finish, the date and time of the dialogue, the start and end times of the utterances, the speaker, and the content of the utterance. The dialogue data 111 may be an audio file recording the conversation. Alternatively, an emotion value may be acquired in advance from a call file using existing technology, and the dialogue data 111 may further include an emotion value corresponding to the utterance and be stored in the storage unit 11. Alternatively, the dialogue analysis unit 14 may analyze each utterance based on the dialogue data 111, calculate an emotional value corresponding to each utterance, and generate dialogue data 111 including the emotional value, which may then be stored in the memory unit 11.

[0025] 5 is an example of a data configuration diagram of the dialogue portion data 112. The dialogue portion data 112 is text data obtained by dividing a dialogue between an operator and a customer from start to finish based on the dialogue data 111 into dialogue portions each including at least one successive utterance by the operator and at least one successive utterance by the customer. Specifically, the dialogue portion data 112 includes a call number that identifies the call, a dialogue number that identifies the dialogue portion, the speaker, the start time and end time of the utterance, the content of the utterance, and an emotional value that indicates the emotional state obtained from the utterance.

[0026] Here, the emotional values ​​(A), (B), and (C) shown in the example of Fig. 5 are numerical values ​​that represent each of the emotions set in advance. For example, the emotional value (A) represents anger, the emotional value (B) represents hope, and the emotional value (C) represents joy, and so on. These represent emotional values ​​corresponding to a plurality of emotions obtained from an utterance using known techniques. When the dialogue analysis unit 14 analyzes a dialogue without using an emotional value, the dialogue portion data 112 may be data that does not include an emotional value.

[0027] 6 is an example of a data configuration diagram of the dialogue portion analysis data 113. The dialogue portion analysis data 113 is data in which each utterance in each dialogue portion obtained by the dialogue analysis unit 14 is classified into an operator response group and a customer reaction group. Specifically, it includes the call number of the call, the dialogue number of the divided dialogue portion, and information on the operator response group and the customer reaction group.

[0028] 7 is an example of a data configuration diagram of the dialogue portion customer evaluation index data 115. The dialogue portion customer evaluation index data 115 is data of customer evaluation indexes for each dialogue portion calculated by the customer evaluation unit 16 based on the customer evaluation index data 114 in which the customer evaluates the operator's response. Specifically, the dialogue portion customer evaluation index data 115 includes information on the call number, the dialogue number, and the customer evaluation index for each dialogue portion. The customer evaluation index data 114 is, for example, customer satisfaction, and is data that can be acquired by a customer evaluating the operator's response in a questionnaire format after the customer has spoken to an operator.

[0029] FIG. 8 is an example of a data configuration diagram of group combination data 116 and customer response group evaluation data 117. Group combination data 116 is information on the number of occurrences and customer response evaluations obtained by calculating the number of occurrences of each combination of agent response group and customer response group. Specifically, group combination data 116 includes operator response groups, customer response groups, and the number of occurrences. Customer response group evaluation data 117 is information on the evaluation of customer response groups obtained by customer evaluation unit 16. Specifically, customer response group evaluation data 117 includes information on customer response groups, correlation values ​​of customer response groups, and customer response evaluations. The customer response evaluation is an evaluation label that indicates a high evaluation, such as "good," a low evaluation, or a "neutral" evaluation that is between good and bad.

[0030] ***Operation of the First Embodiment*** 9 is a processing flow diagram of the dialogue analysis device 10 that analyzes dialogue between agents and customers based on the dialogue data 111 and customer evaluation index data 114, classifies each agent's utterance into operator response groups based on similar operator responses, and classifies each customer's utterance into customer response groups based on similar customer responses, thereby identifying combinations of operator response groups and customer response groups and counting the number of occurrences of each combination. The operation of the dialogue analysis device 10 will be described using this flow diagram.

[0031] The operation procedure of the dialogue analysis device 10 according to the embodiment 1 corresponds to the dialogue analysis method according to the embodiment 1. Furthermore, the program that realizes the operation of the dialogue analysis device 10 according to the embodiment 1 corresponds to the dialogue analysis program according to the embodiment 1.

[0032] First, the dialogue division unit 13 acquires, based on the dialogue data 111, emotion values ​​corresponding to each utterance of the operator and the customer in the dialogue data 111. The dialogue data 111 is divided into dialogue segments, which are dialogues that include at least one successive operator utterance and one successive customer utterance, based on the content of the utterances, changes in emotional values, etc. A dialogue number that uniquely identifies the dialogue segment is assigned to each dialogue segment, and dialogue segment data 112 is generated and stored in the storage unit 11 (step S101).

[0033] For example, a case where the dialogue data 111 is divided into dialogue portions based on the content of the utterances will be described. For the dialogue data 111 including the speakers and the contents of the dialogue such as "Customer, the letter I received is too detailed and difficult to understand," "Operator, I'm sorry," "Operator, I understand how you feel," and "Customer, what should I check?", the dialogue division unit 13 acquires the emotional value of each utterance, and determines that the emotional value corresponding to "Customer, the letter I received is too detailed and difficult to understand" is "emotional value (A) 9, emotional value (B) 3, emotional value (C) 1," the emotional value corresponding to "Operator, I'm sorry" is "emotional value (A) 2, emotional value (B) 6, emotional value (C) 2," the emotional value corresponding to "Operator, I understand how you feel" is "emotional value (A) 1, emotional value (B) 2, emotional value (C) 8," and the emotional value corresponding to "Customer, what should I check?" is "emotional value (A) 2, emotional value (B) 2, emotional value (C) 7." The utterance is analyzed semantically and contextually to determine that it is a series of conversations and divided into conversation parts. At this time, the conversation part is assigned the conversation number "BN005" of the call number "TN001."

[0034] Although the customer's speech in this dialogue portion is not continuous, in the classification process in the next step S103, the speech portions "The letter I received is too small to understand" and "Where should I check" are treated as customer speech portions in the dialogue portion corresponding to dialogue number "BN005" of call number "TN001". Similarly, for operator speech, speeches from two or more operators included in a dialogue portion are treated as a single speech portion and classification process is performed.

[0035] The dialogue may be divided when a change in the speaker's emotional value occurs, or when the emotional value changes from the point where it stops changing. For example, the dialogue may be divided when the customer's emotional value (A) changes from 9 to 2 and the operator's emotional value (C) changes from 2 to 8.

[0036] Next, the dialogue analysis unit 14 generates operator utterance set information that collects operator utterances contained in multiple dialogue parts based on the dialogue part data 112, and classifies each of the operator utterances belonging to the operator utterance set information into one of multiple operator response groups for each similar operator response (step S102). Each of the multiple agent response groups contains a collection of similar agent utterances. Therefore, an utterance of a given agent is classified into an agent response group containing utterances of similar agents.

[0037] Next, the dialogue analysis unit 14 generates customer utterance set information that collects customer utterances contained in multiple dialogue parts based on the dialogue part data 112, and classifies each customer utterance belonging to the customer utterance set information into one of multiple customer reaction groups according to similar customer reactions (step S103). Each of the multiple customer response groups contains a collection of similar customer utterances. Therefore, the utterance of a customer is classified into a customer response group containing utterances of customers similar to the utterance of the customer.

[0038] Next, for each dialogue portion, the dialogue analysis unit 14 identifies a combination of an operator response group of the operator's utterance for each dialogue portion obtained in step S102 and a customer response group of the customer's utterance obtained in step S103. The dialogue analysis unit 14 generates dialogue portion analysis data 113 including information on the combination of the operator response group and the customer reaction group for each dialogue portion, and stores the data in the storage unit 11 (step S104). Specifically, as shown in Figure 6, the dialogue portion analysis data 113 includes information on the combination of an operator response group and a customer response group for each dialogue number that identifies dialogue portions such as "BN001," "BN002," and so on, where there are multiple dialogue numbers for the call number "TN001." For example, the call number "TN001" and dialogue number "BN005" are stored as the operator response group [3] and customer response group B corresponding to the dialogue portion.

[0039] Next, the aggregation unit 15 calculates the number of occurrences of combinations of operator response groups and customer response groups based on the dialogue portion analysis data 113. This is calculated for each combination of operator response groups and customer response groups, and the number of occurrences of each combination is aggregated. The aggregation unit 15 generates group combination data 116 including the aggregated operator response groups and customer response groups and their numbers of occurrences, and stores this in the storage unit 11 (step S105).

[0040] 8, the number of occurrences of customer response group B for operator response group [1] in group combination data 116 is 50. The respective combinations of customer response groups for operator response group [1] are "[1] and A," "[1] and B," "[1] and C," "[1] and D," and "[1] and E." The aggregation unit 15 has aggregated the results as follows: the combination of "[1] and A" appeared 5 times, the combination of "[1] and B" appeared 50 times, the combination of "[3] and A" appeared 100 times, the combination of "[3] and B" appeared 8 times, and the combination of "[3] and C" appeared 9 times. That is, the group combination data 116 stores the results of the counting unit 15 calculating the number of occurrences for each combination of the agent response groups [1] to [5] and the customer response groups A to E.

[0041] The processing flow of the customer evaluation unit 16 and table generation unit 17 of the dialogue analysis device 10, which evaluates the customer's reaction in a dialogue between an operator and the customer and displays the dialogue analysis results on the screen, will be described with reference to the processing flow diagram shown in FIG.

[0042] First, the customer evaluation unit 16 calculates a customer evaluation index for each interaction portion from the customer evaluation index data 114 in which the customer evaluates the call with the operator. The customer evaluation unit 16 generates interaction portion customer evaluation index data 115 including the call number, the interaction number, and the customer evaluation index of the interaction portion, and stores it in the storage unit 11 (step S201).

[0043] Specifically, when the customer evaluation index data 114 is data for one call, where one customer evaluation index is configured from the start to the end of the call, the customer evaluation indexes for multiple conversation segments for one call will all be the same. Therefore, the customer evaluation index for each conversation segment is calculated using the customer evaluation index for the call number in the customer evaluation index data 114.

[0044] Furthermore, if the customer evaluation index data 114 is obtained at predetermined time intervals, it is composed of multiple customer evaluation indexes for one call. In this case, the customer evaluation index for each dialogue portion is calculated using the customer evaluation index in the customer evaluation index data 114 corresponding to the time from the start to the end of the dialogue portion. For example, as shown in Figure 7, the call number "TN001" has a customer evaluation index of "3" for one call. In this case, the customer evaluation index of "3" is calculated for all conversation numbers of the call number "TN001." On the other hand, for call number "TN002," customer evaluation indices are obtained at predetermined time intervals, and the calculated customer evaluation indices for conversation number "BN001" and "BN002" are "5" and "2," respectively. Note that when customer evaluation indices are obtained at predetermined time intervals, if the predetermined time is not the same as the start time and end time of a conversation segment, the customer evaluation indices for each conversation segment may be calculated by, for example, calculating the average value of the customer evaluation indices for the conversation segment.

[0045] Next, the customer evaluation unit 16 evaluates the customer's reaction by calculating a correlation value, which indicates the strength of the relationship between each customer reaction group and the dialogue portion customer evaluation index, based on the dialogue portion customer evaluation index data 115. The evaluation results for each customer reaction group are generated as customer reaction group evaluation data 117 and stored in the storage unit 11 (step S202).

[0046] Specifically, the customer response groups are evaluated such that the group with the highest correlation value is the group with the highest customer evaluation and the group with the lowest correlation value is the group with the lowest customer evaluation. For example, the customer evaluation unit 16 calculates the correlation value between the customer response group and the customer evaluation index as the average value of the customer evaluation indexes of the dialogue portions belonging to the customer response group. As shown in the customer response group evaluation data 117 of Figure 8, if the correlation values ​​of each customer response group are "4.9" for Group A, "1.2" for Group B, "3.0" for Group C, "3.1" for Group D, and "2.9" for Group E, it is determined that "4.9" for Group A has the highest customer response and "1.2" for Group B has the lowest customer response. The number of groups with high correlation values ​​and the number of groups with low correlation values ​​may be two or more groups in descending order of evaluation.

[0047] Next, the table generating unit 17 generates a combination summary table based on the group combination data 116 and the customer reaction group evaluation data 117, and transmits it to the display terminal 20 via the communication unit 12 (step S203). Specifically, the combination summary table is a summary table with each operator response group on the vertical axis and each customer response group on the horizontal axis, and the number of occurrences of the customer response group corresponding to each operator response group is displayed in the corresponding section.

[0048] An example of a combination summary table is shown in Figure 11. In the example screen, customer response group A had the highest rating, so it is displayed as "Rating: Good." In addition, the number of times that group A appeared in combination with operator response group [3] was 100, so it is displayed as "100." This means that customer response group A had the best customer response rating, and among those, operator response group [3] appeared the most, so it can be evaluated as having the tendency for the operator's response to be the best. On the other hand, the number of times that group B appeared in combination with operator response group [1] was 50, so customer response group B had the worst customer response rating, and among those, operator response group [1] appeared the least, so it can be evaluated as having the tendency for the operator's response to be the worst.

[0049] The combination summary table may be displayed in a way that is easy for the user to understand by changing the display mode such as the text color or background color to indicate whether the customer reaction group is good or bad. In this way, it is possible to visualize in an easy-to-understand manner which customer reaction group has received a high or low evaluation. The combination summary table may also be a summary table with each customer response group on the vertical axis and each operator response group on the horizontal axis. In step S203, the table generating unit 17 may transmit information on the group combination data 116 and the customer reaction group evaluation data 117 to the display terminal 20, and operate the display terminal 20 to configure information for displaying the combination summary table.

[0050] 12 is a process flow diagram for classifying utterances in a dialogue portion into a plurality of customer interaction groups along with a plurality of operator interaction groups in the dialogue analysis unit 14 of the dialogue analysis device 10. The operation will be explained using this flow diagram.

[0051] First, the dialogue analysis unit 14 generates operator utterance collection information that collects utterances of the operator (operator utterances) included in a plurality of dialogue portions based on the dialogue portion data 112 (step S301).

[0052] Next, the dialogue analysis unit 14 calculates the similarity between each of the operator utterances based on the feature amounts obtained from the operator utterances (step S302). Specifically, the feature amount of the operator's utterance is an emotional value obtained by analyzing the operator's utterance.

[0053] Next, the dialogue analysis unit 14 classifies similar agent utterances into a plurality of agent response groups based on the similarity between the agent utterances (step S303).

[0054] The dialogue analysis unit 14 generates customer utterance collection information that collects customer utterances (customer utterances) included in a plurality of dialogue portions based on the dialogue portion data 112 (step S304). Next, the dialogue analysis unit 14 calculates the similarity between each customer utterance based on the feature amount of the utterance of each customer (step S305). Specifically, the feature amount of the customer's speech is an emotional value obtained by analyzing the customer's speech.

[0055] Next, the dialogue analysis unit 14 classifies similar customer utterances into a plurality of customer response groups based on the similarity between the customer utterances (step S306).

[0056] Next, the dialogue analysis unit 14 identifies the combination of the operator response group into which the operator utterances in each dialogue part are classified and the customer response group into which the customer utterances are classified, generates dialogue part analysis data 113, and stores it in the memory unit (step S307).

[0057] The classification method described above uses clustering analysis, which classifies data into multiple groups based on specific rules. A typical clustering analysis method is the k-means method. The values ​​of k, which is the number of agent response groups to be classified, and k', which is the number of customer response groups, are integers, set in advance as classification parameters, and stored in the memory unit 11 of the dialogue analysis device 10. Alternatively, the dialogue analysis unit 14 may acquire analysis parameters including k and k'.

[0058] Here, the above-described classification method will be specifically described. In the following explanation, an example will be described in which feature quantities other than emotional values ​​are defined as word importance corresponding to the word frequency and importance in the utterance, and the emotional values ​​and word importance corresponding to the operator utterances and customer utterances in the dialogue portion are used to analyze the dialogue between the operator and the customer. In addition, analysis using emotional values ​​is important in improving customer satisfaction and the quality of operator responses, so emotional values ​​are used in this example. The word importance is a feature obtained by the dialogue analysis unit 14.

[0059] For example, if the operator response groups are set to be classified into five groups (k=5), they will be classified into five operator response groups [1], [2], [3], [4], and [5] based on the similarity between each operator utterance calculated from the word importance of each operator utterance and the emotional value corresponding to each operator utterance. Similarly, if the customer reaction groups are set to be classified into five groups (k'=5), the customer reaction groups will be classified into customer reaction groups A, B, C, D, and E.

[0060] In this embodiment, a method using the non-hierarchical clustering k-means method is described, but classification into operator response groups and customer response groups may also be performed using a hierarchical clustering method such as the Ward method.

[0061] 13 is a process flow diagram for extracting representative utterances from each agent interaction group in the dialogue analysis unit 14. The operation will be explained using this flow diagram. This process is obtained from the results of the classification process in step S103 or steps 301 to S307, and will be explained as the process following step S103 or steps 301 to S307.

[0062] First, as a process following step S303, the dialogue analysis unit 14 expresses the utterances of the operators belonging to each agent handling group as points in a one or more dimensional plane or space, and calculates the point that is the center of gravity (average position) of the agent handling group (step S401). Note that the point that is the center of gravity indicates the average position of the multiple points that belong to the agent handling group.

[0063] For example, when calculating the center of gravity using the K-means method, the dialogue analysis unit 14 first randomly sets representative points for k agent response groups and calculates the distance between each point and the representative point of each set agent response group. The point belonging to the group that is closest in distance to the representative point of the initially set group is set as a temporary representative point. This process is repeated, and the point where the representative point is fixed is set as the center of gravity. Therefore, the center of gravity indicates the average position of all points of the agent response groups.

[0064] Next, the dialogue analysis unit 14 defines the average position calculated in step S401 as the representative point of the group, calculates the distance between each point of the group handled by the agents, and generates dialogue portion analysis data 113A including the representative utterance index of each dialogue portion (step S402). The representative utterance index is calculated for each dialogue portion.

[0065] 14 shows an example of the data structure of the dialogue portion analysis data 113A including the representative utterance index. When the representative utterance index is calculated in step S402, the data includes the call number, dialogue number, operator response group, customer response group, and representative utterance index.

[0066] As mentioned above, the representative point is the point that is the average position of the group of operators who are served by that agent, so the utterance closest to the representative point is defined as the representative utterance of the group of operators who are served by that agent. Also, the closer a point is to the representative point, the more similar the utterance is to the representative utterance.

[0067] Note that, in the same way as the representative point of the operator response group is identified, a representative utterance index, which is the distance between each point of the customer response group and the representative point of the customer response group, may be calculated using each point belonging to the customer response group, and the customer response group representative utterance may be defined. Alternatively, the representative utterance index may be calculated by summing or averaging the distance between each point belonging to each operator response group and the representative point and the distance between each point belonging to the customer response group and the representative point.

[0068] Next, the dialogue analysis unit 14 extracts multiple operator utterances that are close to the representative point of the operator handling group based on the dialogue portion analysis data 113A. A representative dialogue list is generated by listing dialogue portions including operator utterances that are representative utterances of the operator handling group in order of proximity to the representative utterance (representative order), and the representative dialogue list is displayed on the display terminal 20 in representative order (step S403).

[0069] Here, Figure 15 is an example of a screen display of a representative dialogue list for the combination group of customer response group A with the operator response group [3] that received a high evaluation of customer response. The dialogue list screen in Figure 15 shows that representative order 1 is the dialogue that represents the combination group of operator response group [1] and customer response group A. The closer the representative order number is to 1, the more representative the dialogue becomes. In other words, the dialogue portion with representative order 1 is the dialogue handled by the operator that had the best customer response, and this corresponds to representative order "1", call number "TN001", and dialogue number "BN005" in Figure 15. This screen may also be displayed by selecting the number of combinations of operator response groups and customer reaction groups displayed in the combination summary table on the dialogue analysis results (combination summary table) screen shown in FIG.

[0070] 15, the representative dialogue list may display the emotional state corresponding to the emotional value as "anger," "fear," etc., based on the emotional value index for the utterance. For example, if the emotional value of the utterance indicates a high anger value, the emotional state is displayed in the list as "Anger: High."

[0071] ***Effects of the First Embodiment*** According to the present disclosure, for a dialogue between a contact center operator and a customer, it is possible to identify combinations of operator response groups into which the operator's utterances included in the dialogue portion are classified and customer response groups into which the customer's utterances are classified, and to count the number of occurrences of each combination. Furthermore, it is possible to identify the operator response group for each customer response group according to the customer's evaluation, which makes it easier to grasp the trend of customer responses and the tendency of the operators to respond according to the customer responses of each group.

[0072] Furthermore, it is possible to identify the conversations between customers and representative operators of the customer response group and the operator response group. This makes it possible to identify the conversations between operators when the customer response is good, and on the other hand, to identify the conversations between operators when the customer response is bad, and to analyze the good and bad conversations in comparison. It is possible to analyze the conversations with operators that bring customer satisfaction, and to accumulate the responses of that group as know-how.

[0073] [Embodiment 2] The following describes embodiment 2. In embodiment 2, utterances by an operator included in a plurality of dialogue portions are classified into one of a plurality of first agent handling groups based on a first feature amount obtained from the utterances of the operator, and utterances by an operator included in a plurality of dialogue portions are classified into one of a plurality of second agent handling groups based on a second feature amount (different from the first feature amount) for the utterances of the operator. The present embodiment differs from the first embodiment in that it classifies customer utterances included in multiple dialogue segments into one of multiple first customer response groups based on a first feature obtained from the customer utterances, and classifies customer utterances included in multiple dialogue segments into one of multiple second customer response groups based on a second feature (different from the first feature) in response to the customer utterances, thereby identifying combinations of the first operator response group and the second operator response group and the first customer response group and the second customer response group for each dialogue segment, and counts the number of times that these combinations appear in the dialogue data. Note that the following description will only focus on these differences, and will omit a description of the same configuration as the first embodiment.

[0074] Hereinafter, as an example, a process will be described in which a first feature is defined as the inter-utterance similarity obtained from the utterances, a second feature different from the first feature is defined as the emotional value obtained from the utterances, and the dialogue portion is analyzed using the inter-utterance similarity obtained from the operator's utterance and the customer's utterance, and also using the emotional value obtained from the operator's utterance and the customer's utterance. The inter-utterance similarity is a feature acquired by the dialogue analysis unit 14. In addition, classifying each utterance using the emotional values ​​corresponding to the operator's utterance and the customer's utterance is important in improving customer satisfaction, response quality, etc.

[0075] Referring to the processing flow diagram shown in Figure 16, for each dialogue part in embodiment 2, the processing flow will be described, in which the operator's utterances are classified into a first operator response group based on the inter-utterance similarity between each utterance and into a second operator response group based on the emotional value, and the customer's utterances are classified into a first customer response group based on the inter-utterance similarity between each utterance and into a second customer response group based on the emotional value, thereby tallying the number of occurrences of combinations of the first operator response group and the second operator response group and the first customer response group and the second customer response group. The process up to generating the dialogue portion data 112 is the same as step S101 in the first embodiment, and will therefore be described as the process following step S101.

[0076] First, the dialogue analysis unit 14 generates operator utterance set information that collects utterances of operators included in multiple dialogue portions based on the dialogue portion data 112. The dialogue analysis unit 14 classifies the utterances of the operators for each dialogue portion into first operator handling groups based on the inter-utterance similarity obtained between each utterance of the operators (step S501).

[0077] Next, the dialogue analysis unit 14 classifies the utterances of the agents for each dialogue part into second agent handling groups based on the emotion values ​​for the utterances of the agents from the agent utterance collection information (step S502).

[0078] Next, the dialogue analysis unit 14 extracts customer utterances included in the multiple dialogue portions from the dialogue portion data 112 based on the dialogue portion data 112, and generates customer utterance set information. The dialogue analysis unit 14 classifies the customer utterances included in the multiple dialogue portions into first customer reaction groups based on the inter-utterance similarity of the customer utterances (step S503).

[0079] Next, the dialogue analysis unit 14 calculates the similarity between each customer utterance based on the emotional value for each customer utterance from the customer utterance collection information, and classifies the customer utterances included in multiple dialogue parts into multiple second customer reaction groups (step S504).

[0080] Next, the dialogue analysis unit 14 identifies the combination of the first operator response group, the second operator response group, the customer response group, and the second customer response group for each dialogue part, generates dialogue part analysis data 113B including information on the combination, and stores it in the memory unit 11 (step S505).

[0081] Here, the dialogue analysis unit 14 may identify a representative point of the second agent handling group and calculate a representative utterance index corresponding to each dialogue portion, similar to steps S401 and S402. The representative utterance index may be a representative utterance index corresponding to the first agent handling group. Alternatively, the representative utterance index may be a representative utterance index corresponding to the first customer response group, or a representative utterance index corresponding to the second customer response group. When the representative utterance index is calculated, in addition to information on the combination of the first operator response group, the second operator response group, the customer response group, and the second customer response group for each dialogue part, information including the representative utterance index is generated as dialogue part analysis data 113B and stored in the memory unit 11.

[0082] The aggregation unit 15 calculates the number of occurrences of combinations of the first operator response group, the second operator response group, the first customer response group, and the second customer response group based on the dialogue portion analysis data 113B. Group combination data 116B including the number of occurrences of combinations of the first operator response group, the second operator response group, the first customer response group, and the second customer response group is generated and stored in the storage unit 11 (step S506).

[0083] An example of the data configuration of group combination data 116B in embodiment 2 is shown in Fig. 18. The operator response group item of group combination data 116B contains information combining first operator response groups classified into five groups ([1] to [5]) with second operator response groups classified into three groups ([1]' to [3]'), and the customer response group item contains information combining first customer response groups classified into five groups (A to E) with second customer response groups classified into three groups (A' to C'), as well as the number of times the operator response groups and customer response groups appeared.

[0084] Here, the customer evaluation unit 16 may determine the customer response evaluations of the first customer response group and the second customer response group, similar to the processing of steps S201 to S202 in embodiment 1. In this case, first customer response group evaluation data 117B1 and second customer response group evaluation data 117B2 are generated and stored in the storage unit 11.

[0085] 18 is an example of the data configuration of first customer response group evaluation data 117B1 in embodiment 2. It is information including correlation values ​​and customer response evaluations of first customer response groups A to E. Second customer response group evaluation data 117B2 is an example of the data configuration of second customer response group evaluation data 117B2 in embodiment 2. It is information including correlation values ​​and customer response evaluations of second customer response groups A' to C'.

[0086] In the second embodiment, the table generating unit 17 generates a combination summary table based on the group combination data 116B, the first customer reaction group evaluation data 117B1, and the second customer reaction group evaluation data 117B2. An example of displaying a combination summary table as a dialogue analysis result in the second embodiment will be described with reference to FIG. The operator response groups are composed of 15 groups, which are combinations of the five first operator response groups [1] to [5] and the three second operator response groups [1]' to [3]'. Similarly, the customer response groups are composed of 15 groups, which are combinations of the five first customer response groups A to E and the three second customer response groups A' to C'. These 15 operator response groups and 15 customer response groups are tabulated and displayed on the display terminal 20 as a combination summary table with the operator response groups on the horizontal axis and the customer response groups on the vertical axis. The customer response groups are labeled with a first customer response group evaluation label and a second customer response group evaluation label. In the example of Figure 9, the combination of "operator response group [3]" and "operator response groups [1]' to [3]'", which is a combination of "customer response group A (evaluation: good)" and "group A' (good)", is analyzed to be a group with good customer response.

[0087] In the second embodiment, the utterances of each operator are classified into a first operator response group and a second operator response group, and the utterances of each customer are classified into a first customer response group and a second customer response group, and the combination of each group is identified. However, it is also possible to classify the utterances of the operator and the customer three or more times based on features different from the features used previously, and identify the combination of the operator response group and the customer response group that is the combination of each group that has been classified three or more times.

[0088] ***Effects of the Second Embodiment*** For a conversation between a contact center operator and a customer, the system can identify combinations of a first operator response group classified by a first feature of the operator's utterance included in the conversation portion and a second operator response group classified by the feature of the second feature, and a first customer response group classified by the first feature of the customer's utterance included in the conversation portion and a second customer response group classified by the second feature, and can tabulate the number of occurrences of each combination.

[0089] [Embodiment 3] The third embodiment will be described below. In the third embodiment, utterances of operators included in multiple dialogue segments are classified into one of multiple first operator response groups based on a first feature obtained from the utterances, and then utterances of operators belonging to the first operator response group are further classified into one of multiple second operator response groups based on a second feature obtained from the utterances (different from the first feature). Then, utterances of customers included in multiple dialogue segments are classified into one of multiple first customer response groups based on the first feature obtained from the utterances, and then utterances of customers belonging to each first customer response group are further classified into one of multiple second customer response groups based on the second feature obtained from the utterances (different from the first feature). This differs from the first embodiment in that combinations of second operator response groups and second customer response groups in each dialogue segment are identified and the number of occurrences of each combination is tallied. The following description will only focus on these differences, and a description of the same configuration as the first embodiment will be omitted.

[0090] Hereinafter, as an example, a process will be described in which a first feature is defined as the inter-utterance similarity obtained from the utterances, a second feature different from the first feature is defined as the emotional value obtained from the utterances, and the dialogue portions are analyzed and classified using the inter-utterance similarity obtained from the operator's utterance and the customer's utterance, and the dialogue portions are analyzed using the emotional values ​​of the operator's utterances and the customer's utterances belonging to each classified group. Classifying each utterance using the emotional values ​​corresponding to the operator's utterance and the customer's utterance is important in improving customer satisfaction and response quality.

[0091] Referring to the processing flow diagram shown in Figure 20, we will explain the processing flow in embodiment 3, which classifies each dialogue portion into a first operator response group, a second operator response group, a first customer response group, and a second customer response group, identifies combinations of the second operator response group and the second customer response group, and tallies the number of occurrences of each combination. The process up to generating the dialogue portion data 112 is the same as step S101 in the first embodiment.

[0092] First, the dialogue analysis unit 14 generates operator utterance set information that collects utterances of operators included in multiple dialogue portions based on the dialogue portion data 112. The dialogue analysis unit 14 classifies the utterances of the operators in the operator utterance set information into similar first agent response groups based on the inter-utterance similarity obtained between the utterances of the operators (step S601).

[0093] Next, the dialogue analysis unit 14 generates first agent handling group utterance set information for each first agent handling group, which is a collection of a plurality of utterances belonging to the first agent handling group (step S602).

[0094] Next, the utterances of the agents in the first agent handling group utterance collection information are classified into similar second agent handling groups based on the emotion value obtained from the utterances of each agent (step S603).

[0095] Next, the dialogue analysis unit 14 generates customer utterance set information that collects customer utterances included in multiple dialogue portions based on the dialogue portion data 112. The dialogue analysis unit 14 classifies the customer utterances in the customer utterance set information into similar first customer reaction groups based on the inter-utterance similarity obtained between the utterances of each customer (step S604).

[0096] Next, the dialogue analysis unit 14 generates first customer response group utterance set information for each first customer response group, which is a collection of multiple utterances belonging to the first customer response group (step S605).

[0097] Next, the first customer response group utterance set information is classified into a second customer response group based on the emotion value obtained from the utterance of each customer (step S606).

[0098] Next, the dialogue analysis unit 14 identifies the combination of the first operator response group, the second operator response group, the customer response group, and the second customer response group for each dialogue part, generates dialogue part analysis data 113C including information on the combination, and stores it in the memory unit 11 (step S607).

[0099] Here, the dialogue analysis unit 14 may calculate a representative point for the second group of operators, and calculate a representative utterance index corresponding to each dialogue portion, as in steps S401 and S402 of the first embodiment. The representative utterance index may be a representative utterance index corresponding to the first group of operators. Alternatively, the representative utterance index may be a representative utterance index corresponding to the first customer response group, or a representative utterance index corresponding to the second customer response group. When the representative utterance index is calculated, in addition to information on the combination of the first operator response group, the second operator response group, the customer response group, and the second customer response group for each dialogue part, information including the representative utterance index is generated as dialogue part analysis data 113C and stored in the memory unit 11.

[0100] Specifically, the operator's utterances "I'm sorry" and "I understand your feelings" for call number "TN001" and conversation number "BN001" are classified into the first operator response group "[3]" in step S601, and are then classified into the first operator response group "[9]" based on the similarity between multiple utterances belonging to the first operator response group "[3]". The customer utterances for call number "TN001" and conversation number "BN001," "The letter I received is too detailed and difficult to understand," and "Where should I check?" are classified into the first customer response group "A" in step S605, and are classified into the second customer response group "b" in step S606 based on the similarity between multiple utterances belonging to the first customer response group "A." That is, if the first operator response groups are classified into five, they are classified into five groups, first operator response groups [1] to [5], and if the utterances belonging to the first operator response groups are further classified into three groups for each first operator response group, they are classified into 15 groups, second operator response groups 1 to 15. Also, if the first customer response groups are classified into five, they are classified into five groups, first customer response groups [1] to [5], and if the utterances belonging to the first customer response groups are further classified into three groups for each first customer response group, the second customer response groups are classified into 15 groups, second operator response groups a to o.

[0101] The aggregation unit 15 calculates the number of occurrences of the combination of the second agent handling group and the second customer response group based on the dialogue portion analysis data 113C, generates group combination data 116C including information on the number of occurrences of each agent handling group and customer response group, and stores the data in the storage unit 11 (step S608).

[0102] Group combination data 116C in Fig. 22 is an example of the data configuration of group combination data in Embodiment 3. The operator response group item of group combination data 116C contains information on second operator response groups (1 to 15) classified into 15 groups corresponding to the first operator response groups ([1] to [5]), and the customer response group item contains information on second customer response groups (a to o) classified into 15 groups corresponding to the first customer response groups (A to E), as well as the number of times the operator response group and customer response group appeared.

[0103] Here, the customer evaluation unit 16 may determine the customer reaction evaluations of the first customer reaction group and the second customer reaction group, similar to the processing of steps S201 to S202 in embodiment 1. In this case, first customer reaction group evaluation data 117C1 and second customer reaction group evaluation data 117C2 are generated and stored in the storage unit 11.

[0104] In the third embodiment, the table generating unit 17 generates a combination summary table based on the group combination data 116C, the first customer reaction group evaluation data 117C1, and the second customer reaction group evaluation data 117C2. An example of displaying a combination summary table as a dialogue analysis result in the third embodiment will be described with reference to FIG. The operator response groups are composed of the following combinations: second operator response groups 1-3 for the first operator response group [1], second operator response groups 4-6 for the first operator response group [2], second operator response groups 7-9 for the first operator response group [3], second operator response groups 10-12 for the first operator response group [4], and second operator response groups 13-15 for the first operator response group [5].

[0105] Similarly, the customer response groups are composed of combinations of groups such as second customer response groups a-c for the first customer response group A, second customer response groups d-f for the first customer response group B, second customer response groups g-i for the first customer response group C, second customer response groups j-l for the first customer response group D, and second customer response groups m-o for the first customer response group E.

[0106] The combinations of these 15 operator response groups and 15 customer response groups are tallied and displayed as a combination summary table on the display terminal 20, with the horizontal axis representing the operator response group and the vertical axis representing the customer response group. The customer response groups are labeled with a first customer response group evaluation label and a second customer response group evaluation label, and in the example of Figure 23, the combination of "Customer Response Group A (Evaluation: Good)" and "Group a (Good)", that is, "Operator Response Group [3]" and "Operator Response Groups 7-9", is a group that has been analyzed to have a good customer response.

[0107] In the third embodiment, each agent utterance is classified into a set of utterances belonging to the first agent response group based on the first feature, and then further classified into a second agent response group based on the second feature. Also, each customer utterance is classified into a set of utterances belonging to the first customer response group based on the first feature, and then further classified into a second customer response group based on the second feature, and combinations of the second agent response group and the second customer response group are identified. However, the second agent response group may be further classified into a third or more agent response groups based on a feature different from the first feature and the second feature, and similarly, the second customer response group may be classified into a third or more customer response groups, and the final combination of the agent response group and the customer response group classified may be identified.

[0108] ***Effects of the Third Embodiment*** For a conversation between a contact center operator and a customer, the system can identify combinations of each utterance of an operator belonging to a first operator response group classified by a first feature of the operator's utterance included in the conversation portion with a second operator response group classified by a second feature different from the first feature of the operator's utterance included in the conversation portion, and each utterance of a customer belonging to a first customer response group classified by a first feature of the customer's utterance included in the conversation portion with a second customer response group classified by a second feature different from the first feature of the customer's utterance included in the conversation portion, and count the number of occurrences of each combination.

[0109] ***Other Configurations*** <Variation 1> As a first modification, the dialogue analysis device 10 and the display terminal 20 described in each embodiment may be integrated into a single device.

[0110] <Variation 2> As a second variant example, the memory unit 11 containing some or all of the data and information of the dialogue data 111, dialogue portion data 112, dialogue portion analysis data 113, dialogue portion customer evaluation index data 115, group combination data 116, and customer response group evaluation data 117 described in each embodiment may be configured to be stored in an external memory unit.

[0111] <Variation 3> As a third modification, some or all of the functions of the dialogue analysis device 10 described in each embodiment may be implemented by being built on the cloud.

[0112] In addition, in each embodiment, a conversation between an operator and a customer at a contact center is given as an example, and the process of analyzing the conversation is explained, but this is not limited to contact centers, and the conversation can be configured to be analyzed as long as it is a conversation between a customer and an operator or other person who handles customer inquiries.

[0113] In addition, the term "unit" in the above description may be read as a "circuit," "process (step)," "procedure," "process," or "processing circuit."

[0114] The embodiments of the present disclosure have been described above. Some of these embodiments may be combined and implemented. Also, one or some of them may be partially implemented. Note that the present disclosure is not limited to the above embodiments, and various modifications are possible as needed. [Explanation of symbols]

[0115] 1 Dialogue analysis system, 2 network, 10 dialogue analysis device, 11 memory unit, 12 communication unit, 13 dialogue segmentation unit, 14 dialogue analysis unit, 15 aggregation unit, 16 customer evaluation unit, 17 table generation unit, 20 display terminal, 51 processor, 52 memory, 53 signal input / output unit, 111 dialogue data, 112 dialogue part data, 113 dialogue part analysis data, 114 customer evaluation index data, 115 dialogue part customer evaluation index data, 116 group combination data, 117 customer response group evaluation data.

Claims

1. a dialogue dividing unit that divides dialogue data recorded from a dialogue between an operator and a customer into a plurality of dialogue parts each including an operator utterance and a customer utterance; The operator utterances included in the plurality of dialogue portions are classified into the same operator response group according to similar responses by the operator, and The customer utterances included in the plurality of dialogue portions are classified into the same customer reaction group according to similar reactions of the customer, a dialogue analysis unit that identifies, for each of the plurality of dialogue portions, a combination of the operator response group into which the operator utterances included in the dialogue portion are classified and the customer response group into which the customer utterances are classified; a counting unit that counts the number of occurrences of a combination of the agent response group and the customer response group in the dialogue data; A dialogue analysis device comprising:

2. a table generating unit that generates a combination summary table in which the number of occurrences calculated by the calculation unit is used as table data; The conversation analysis device according to claim 1 .

3. and a customer evaluation unit that evaluates the customer's evaluation of the customer reaction group by extracting a correlation value indicating the strength of a relationship between each customer reaction group and the customer evaluation index data based on the customer reaction group into which the customer utterances included in the dialogue portion are classified and the customer evaluation index data indicating the degree of satisfaction the customer has achieved with the dialogue portion. The conversation analysis device according to claim 1 .

4. The dialogue analysis unit classifies the operator utterances included in the plurality of dialogue portions into the same operator response group based on the similarity between the operator utterances. The conversation analysis device according to claim 1 .

5. The dialogue analysis unit classifies the customer utterances included in the plurality of dialogue portions into the same customer reaction group based on the similarity between the customer utterances. The conversation analysis device according to claim 1 .

6. the dialogue analysis unit analyzes each of the operator utterances included in the plurality of dialogue portions, and acquires an emotion value representing an emotional state of the operator in each of the operator utterances; The method is characterized in that a similarity between the operator utterances is calculated based on the emotion value acquired for the operator utterances, and a plurality of the operator utterances are classified into the same operator response group. The conversation analysis device according to claim 4 .

7. the dialogue analysis unit analyzes each of the customer utterances included in the plurality of dialogue portions, and acquires an emotion value representing an emotional state of the customer in each customer utterance; a similarity between the customer utterances is calculated based on the emotion value acquired for the customer utterances, and the plurality of customer utterances are classified into the same customer reaction group; The conversation analysis device according to claim 5 .

8. The dialogue portion includes two or more speech portions of the customer, The dialogue analysis unit analyzes the change in the emotional value in each of the two or more utterance portions to evaluate the customer's reaction to the dialogue portion, and classifies the plurality of customer utterances into the same customer reaction group. The conversation analysis device according to claim 7.

9. When each of the plurality of operator utterances classified into one of the operator response groups is expressed as a point in at least one or more dimensional space based on the feature amount obtained from the utterance of the operator, the dialogue analysis unit: Among the plurality of points belonging to the operator response group, an utterance corresponding to a point closest to an average position of the plurality of points is determined as a representative utterance of the operator response group. The conversation analysis device according to claim 6.

10. When each of the plurality of customer utterances classified into one of the customer response groups is expressed as a point in at least one or more dimensional space based on the feature amount obtained from the customer utterance, the dialogue analysis unit: Among the plurality of points belonging to the customer reaction group, an utterance corresponding to a point closest to an average position of the plurality of points is determined as a representative utterance of the customer reaction group. The conversation analysis device according to claim 7.

11. the dialogue analysis unit analyzes each of the operator utterances and the customer utterances included in the plurality of dialogue portions, and acquires first feature amounts and second feature amounts corresponding to the operator utterances and the customer utterances included in the dialogue portions, respectively; calculating a similarity between the agent utterances based on the first feature amount acquired about the agent utterances, and classifying the agent utterances into the same first agent handling group; calculating a similarity between the operator utterances included in the plurality of dialogue portions based on second feature amounts acquired about the operator utterances, and classifying the operator utterances into the same second operator response group; and calculating a similarity between the customer utterances included in the plurality of dialogue portions based on the first feature amount acquired for the customer utterances, and classifying the customer utterances into the same first customer reaction group; calculating a similarity between the customer utterances included in the plurality of dialogue portions based on the second feature amount acquired for the customer utterances, and classifying the customer utterances into the same second customer reaction group; Identifying a combination of the first operator response group, the second operator response group, the first customer response group, and the second customer response group for each of the dialogue portions; The counting unit counts the number of occurrences of combinations of the first operator response group, the second operator response group, the first customer response group, and the second customer response group in the dialogue data. The conversation analysis device according to claim 1 .

12. the dialogue analysis unit analyzes each of the operator utterances and the customer utterances included in the plurality of dialogue portions, and acquires first feature amounts and second feature amounts corresponding to the operator utterances and the customer utterances included in the dialogue portions, respectively; calculating a similarity between the operator utterances included in the plurality of dialogue portions based on the first feature amount obtained from the operator utterances, and classifying the operator utterances into the same first operator response group; Further, a similarity between the operator utterances belonging to the first operator handling group is calculated based on the second feature amount acquired about the operator utterances, and the operator utterances are classified into the same second operator handling group; calculating a degree of similarity between the customer utterances included in the plurality of dialogue portions based on the first feature amount acquired for the customer utterances, and classifying the customer utterances into the same first customer reaction group; Further, the customer utterances belonging to the first customer response group are classified into the same second customer response group by calculating a similarity between the customer utterances based on the second feature amount acquired for the customer utterances, For each of the dialogue portions, a combination of the second agent response group and the second customer response group included in the dialogue portion is identified; The counting unit counts the number of occurrences of a combination of the second agent response group and the second customer response group in the dialogue data for each combination. The conversation analysis device according to claim 1 .

13. a dialogue division process for dividing dialogue data recorded in a computer into a plurality of dialogue parts including operator utterances and customer utterances; The operator utterances included in the plurality of dialogue portions are classified into the same operator response group according to similar responses by the operator, and The customer utterances included in the plurality of dialogue portions are classified into the same customer reaction group according to similar reactions of the customer, a dialogue analysis process for identifying, for each of the plurality of dialogue portions, a combination of the operator response group into which the operator utterances included in the dialogue portion are classified and the customer response group into which the customer utterances are classified; a counting process for counting the number of occurrences of a combination of the agent response group and the customer response group in the dialogue data; A dialogue analysis program that executes

14. A dialogue analysis method executed by a computer of a dialogue analysis device, a dialogue division step of dividing dialogue data recorded from a dialogue between an operator and a customer into a plurality of dialogue parts including operator utterances and customer utterances; The operator utterances included in the plurality of dialogue portions are classified into the same operator response group according to similar responses by the operator, and The customer utterances included in the plurality of dialogue portions are classified into the same customer reaction group according to similar reactions of the customer, a dialogue analysis step of identifying, for each of the plurality of dialogue portions, a combination of the operator response group into which the operator utterances included in the dialogue portion are classified and the customer response group into which the customer utterances included in the dialogue portion are classified; a counting step of counting the number of occurrences of a combination of the agent response group and the customer response group in the dialogue data for each combination; A dialogue analysis method with

Images