Customer important opinion extraction system and customer important opinion extraction method

Abstract

To extract important customer opinions.SOLUTION: In a customer important opinions extraction system, an aggregation unit extracts a classification category in which increase in the number of customer opinions is large from difference in the number of the customer opinions belonging to the classification category at a first timing corresponding to a start time and at a second timing corresponding to an end time of a predetermined period. A keyword review processing unit reviews for a category keyword about the classification category extracted by the aggregation unit. A similarity discrimination processing unit discriminates similarity about the customer opinions using importance of the customer opinions about the classification category extracted by the aggregation unit. An important inquiry label review processing unit reviews importance about the customer opinions belonging to the classification category extracted by the aggregation unit. An important inquiry evaluation processing unit extracts the customer opinions with high importance.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a system and method for extracting important customer opinions from customer opinions collected through marketing activities or call centers. [Background technology]

[0002] In the business of selling products such as devices, in today's world where science and technology have developed and spread, many companies are able to launch competing products with the same functions and quality into the market, and products are becoming commoditized. Therefore, in order to develop products that are supported by customers, many companies are taking measures to increase the attractiveness of their products by differentiating them from other companies' products through product development that takes the perspective of the customers who purchase and use the products into account. For example, marketing activities are carried out to understand market trends and customer needs or identify target customers by conducting market research to collect customer opinions and analyzing the customer opinions. Companies also use customer opinions about their company's products and services and survey results received at call centers as important information to understand customer needs. Such marketing activities and customer opinions (VOC: Voice of customer) collected at call centers contain a lot of text-format document data in natural language rather than numerical values. For example, the free-response section of a survey questionnaire and the information answered by operators at a call center are often written in text format.

[0003] To grasp customer needs and market trends in a timely manner, it is necessary to analyze this huge amount of text data (text data) as quickly and accurately as possible. To achieve this, a method has been devised that utilizes text mining technology to efficiently analyze and grasp customer needs from the perspective of importance.

[0004] For example, Patent Document 1 discloses a technology in which an information processing device that stores customer opinion information including document data expressing customer opinions in natural language is provided with: a morphological analysis unit that breaks down the document data into words and outputs each word with a part of speech associated therewith; a syntactic analysis unit that analyzes the content of the document using the data output by the morphological analysis unit; a clustering unit that uses the processing results of the syntactic analysis unit to classify and output the customer opinion information according to specified customer needs; an evaluation term definition unit that accepts from users keywords for evaluating customer needs and their evaluation values; and a calculation processing unit that calculates a score indicating the importance of the customer needs using the customer opinion information classified according to customer need, the set keywords and their evaluation values. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2007-226568 A Summary of the Invention [Problem to be solved by the invention]

[0006] The technology described in the above patent documents has the following problems. Specifically, the method and system disclosed in Patent Document 1 predefines evaluation terms and their evaluation values, calculates a score indicating the importance of customer needs using methods such as morphological analysis, syntactic analysis, and clustering, and analyzes needs from data indicating customer opinions (VOC), i.e., VOC data. In defining evaluation terms in the importance evaluation, the user needs to set keywords for evaluating customer needs and their evaluation values ​​as appropriate. However, there is a concern that the settings of the keywords and evaluation values ​​may become outdated over time.

[0007] The present invention has been made in consideration of the above circumstances, and an object of the present invention is to update the settings of keywords and evaluation values ​​to meet the ever-changing needs of customers by analyzing accumulated data. [Means for solving the problem]

[0008] A customer important opinion extraction system for extracting customer important opinions from customer opinions submitted by customers, the system comprising: a calculation unit for executing calculation processing; a storage unit accessible to the calculation unit; and an input / output unit, the storage unit storing the customer opinions, data related to the customer opinions, a plurality of classification categories into which the customer opinions are classified, a plurality of category keywords that characterize each of the plurality of classification categories, and a degree of importance assigned to the customer opinions, the customer important opinion extraction system further comprising: a classification unit for classifying the customer opinions into the classification categories based on the plurality of category keywords; and a calculation unit for detecting when an increase in the customer opinions in a predetermined period of time is large. the calculation unit includes a keyword review processing unit that reviews the category keywords, a similarity determination processing unit that determines similarity of the customer opinions, an important inquiry label review processing unit that reviews the importance, and an important inquiry evaluation processing unit that extracts the customer opinions with high importance, the calculation unit extracts classification categories with a large increase in the number of customer opinions from a difference in the number of customer opinions belonging to the classification category between a first timing corresponding to a start time of the predetermined period and a second timing corresponding to an end time of the predetermined period, and the keyword review processing unit extracts the classification categories with a large increase in the number of customer opinions from the difference in the number of customer opinions belonging to the classification category between a first timing corresponding to a start time of the predetermined period and a second timing corresponding to an end time of the predetermined period, extracting category keywords that frequently appear in customer opinions belonging to the above, changing the category keywords based on the extraction result of the frequently appearing category keywords, and changing the contents of the category keywords based on the frequently appearing category keywords based on the hypothesis verification result input from the input / output unit; The similarity determination processing unit determines similarity of the customer opinions for the classification category extracted by the tabulation unit using the importance of the customer opinions, and the important inquiry label review processing unit determines similarity of the customer opinions belonging to the classification category extracted by the tabulation unit. Extracting frequently occurring similar inquiries, and providing a response to the customer opinions based on the number of the extracted similar inquiries. The importance of change death, Based on the hypothesis verification result input from the input / output unit, The importance of change The important inquiry evaluation processing unit extracts and outputs the customer opinions having a high degree of importance. It is characterized by: Effect of the Invention

[0009] According to the present invention, it is possible to extract important customer opinions. Problems, configurations and effects other than those described above will become apparent from the following description of the preferred embodiment of the present invention. [Brief description of the drawings]

[0010] [Figure 1] 1 is a functional block diagram of a customer important opinion extraction system to which an embodiment of the present invention is applied. [Diagram 2] 1 is a diagram showing a representative example of a user's operation of the customer important opinion extraction system of the present embodiment and a processing flow in the customer important opinion extraction system. FIG. [Diagram 3] FIG. 2 is a diagram showing a schematic flow of a process in the customer important opinion extraction system according to the embodiment of the present invention. [Figure 4] FIG. 11 is a diagram showing a flow of an important inquiry area extraction process by aggregating call center inquiry data in the customer important opinion extraction system according to the embodiment of the present invention. [Diagram 5] FIG. 13 is a diagram showing a flow of an important inquiry label review process in the customer important opinion extraction system according to the embodiment of the present invention. [Figure 6] FIG. 11 is a diagram showing a flow of important inquiry data evaluation and extraction processing in an important inquiry area in the customer important opinion extraction system according to the embodiment of the present invention. [Figure 7] FIG. 11 is a diagram showing a flow of a keyword definition review process in the customer important opinion extraction system according to the embodiment of the present invention. [Figure 8] 1 is a diagram illustrating an example of a data configuration of call center inquiry data used in an embodiment of the present invention. FIG. [Figure 9] FIG. 2 is a diagram illustrating an example of a data configuration of customer attribute data used in the embodiment of the present invention. [Figure 10] FIG. 2 is a diagram illustrating an example of a data configuration of device attribute data used in the embodiment of the present invention. [Figure 11] FIG. 10 is a diagram illustrating an example of trouble hypothesis data in the customer important opinion extraction system according to the embodiment of the present invention. [Figure 12] 11 is a diagram illustrating an example of hypothesis verification result data in the customer important opinion extraction system according to the embodiment of the present invention. FIG. [Figure 13] FIG. 2 is a diagram illustrating an example of keyword definitions of the customer important opinion extraction system according to the embodiment of the present invention. [Figure 14] FIG. 13 is a diagram showing an example of an important inquiry label in the customer important opinion extraction system according to the embodiment of the present invention. [Figure 15] 1 is a diagram illustrating an example of a configuration of a customer important opinion extraction system according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. First, a schematic configuration of an embodiment of the present invention will be described with reference to Figs. 1 and 15. Fig. 1 is a functional block diagram of a customer important opinion extraction system to which an embodiment of the present invention is applied. Fig. 15 is a diagram showing an example of the configuration of a customer important opinion extraction system according to an embodiment of the present invention. In this embodiment, an example will be described in which the present invention is applied to a company that manufactures and sells products such as devices as merchandise, and the company collects and analyzes opinions on existing products from customers for the purpose of developing new products, etc.

[0012] The customer important opinion extraction system of this embodiment is a customer important opinion extraction system that extracts customer important opinions from customer opinions that are opinions submitted by customers. As shown in FIG. 15, the customer important opinion extraction system 1 according to this embodiment has a calculation unit 4 such as a CPU (Central Processing Unit) that executes calculation processing, a memory 6 that is a storage element such as a RAM, and a storage unit 8 that is a non-volatile storage element such as a ROM or a hard disk drive and is accessible to the calculation unit 4. The customer important opinion extraction system 1 further has an input / output unit 10 that has an input device such as a keyboard or a mouse, and an output device such as a display that outputs processing results. The calculation unit 4, the memory 6, the storage unit 8, and the input / output unit 10 are mutually connected by a bus line 12, and the calculation unit 4 reads out a program stored in the storage unit 8 into the memory 6 and executes it, thereby controlling the customer important opinion extraction system itself and performing various judgments, calculation processing, and control processing. Furthermore, the customer important opinion extraction system 1 may have a communication unit such as a network interface that handles communication processing with other devices such as external user terminals, although not shown in the figure. The functions of the customer important opinion extraction system 1 shown in FIG. 1 may be implemented by the calculation unit 4 executing a program.

[0013] As shown in FIG. 1, the customer important opinion extraction system 1 according to the present embodiment includes a customer request integrated data lake unit 30 that stores customer opinions (Voice of customer) written in natural language, attributes of the customers who express the opinions, and attributes of devices owned by the customers, and a data statistical processing unit 40 that executes data statistical processing for analyzing customer needs using text data in the customer request integrated data lake unit 30. The customer important opinion extraction system 1 further includes a machine learning model unit 50 that executes customer important opinion extraction processing by machine learning on the data that has been subjected to statistical processing, an extraction processing database unit 20 that stores keyword definitions 201 for statistical processing and important inquiry labels 202 for machine learning, and an input / output unit 10 through which notifications of extracted customer important opinions and hypothesis verification results are input to the customer request data lake. The customer important opinion extraction system accepts input of data indicating customer opinions (VOC) such as survey results from customers and inquiry information from customers received at a call center via the input / output unit 10. The customer important opinion extraction system analyzes customer needs using the data indicating the received customer opinions, and notifies the user of the customer important opinion extraction system of the extraction results via the input / output unit 10. Note that a user of the customer important opinion extraction system may be simply referred to as a "user."

[0014] Specifically, the data statistical processing unit 40 has a morphological analysis unit 41, a classification unit 42, and a compilation unit 43. The machine learning model unit 50 has a keyword review processing unit 51, a top inquiry label review processing unit 52, a similarity determination processing unit 53, and a top inquiry evaluation processing unit 54.

[0015] The customer request integrated data lake unit 30 stores data indicating customer opinions (VOC), such as customer inquiry data collected at a call center, end user questionnaires, and various reports. Here, a plurality of customer opinion data including information expressing customer opinions in natural language is the above-mentioned customer inquiry data. The customer opinion data includes customer opinions. The inquiry data from customers may be called call center inquiry data 301, or simply inquiry data or inquiries. In the example shown in FIG. 1, the customer request integrated data lake unit 30 stores call center inquiry data 301, which is data received at a call center or the like and converted into a file, as customer opinions. In addition, the customer request integrated data lake unit 30 stores data indicating customer opinions (VOC), that is, customer attribute data 302 as information for identifying the type of customer from whom the VOC data is submitted, and device attribute data 303 as information for identifying the type of device the VOC data is submitted from, for aggregation of inquiry data based on the characteristics of customer opinions. Regarding the information included in the customer request integrated data lake unit 30, for example, in the above description, call center inquiry data 301 is described with the word "data" omitted, as in call center inquiry data 301 in Fig. 1. This also applies to customer attribute data 302, device attribute data 303, hypothesis verification result data 311, and problem hypothesis data 312 in Fig. 1. Also, in the above description, the call center inquiry data 301 is named call center inquiry data for convenience, but data showing customer opinions (VOC) such as end user questionnaires and various reports may be stored.

[0016] An example of the data structure of the call center inquiry data 301 stored in the customer request integrated data lake unit 30 will now be described with reference to Fig. 8. The call center inquiry data 301 includes customer opinions.

[0017] FIG. 8 is a diagram showing an example of a data configuration of call center inquiry data 301 used in an embodiment of the present invention. In FIG. 8, data in which the contents of inquiries from customers who are users of a product are collected is illustrated as the call center inquiry data 301. As shown in the figure, the call center inquiry data 301 is a database in which call center inquiry results for a product are collected and registered. As a specific example, the call center inquiry data 301 has a field 3011 for registering a "reception ID" for identifying the collected customer opinions, and fields 3012 and 3013 for identifying customers. Furthermore, one record is formed by including fields 3014, 3015, 3016, 3017, and 3018 for registering the contents of inquiries from customers, responses to inquiries, results of corresponding actions, and completion dates and times when the actions are completed. In the example shown in FIG. 8, as information for identifying a customer, a "customer code" for uniquely identifying a customer is registered in field 3012, and a "customer name" for the name of the customer represented by the customer code is registered in field 3013. In addition, in field 3014, "date and time of receipt" is registered as information on the date and time of the call center inquiry from the customer, and in field 3015, "contents of receipt" is registered as information on the contents of the call center inquiry. In addition, in field 3016, "answer" is registered as information on the answer to the customer in response to the inquiry, in field 3017, "response action" is registered as information on the action taken as a result of the answer, and in field 3018, "completion date and time" is registered as the date and time when the inquiry response was completed. In addition to these, information on the call center reception may be registered as column 301n, etc. Here, n is a positive integer value assigned to represent the field number. The positive integer value n assigned to represent this field number is the same in the following. Customer name 3013, date and time of receipt 3014, and contents of receipt 3015 correspond to specific customer opinions.

[0018] An example of the data configuration of the customer attribute data 302 stored in the customer request integrated data lake unit 30 will be described with reference to FIG.

[0019] FIG. 9 is a diagram showing an example of the data configuration of the customer attribute data 302 used in the embodiment of the present invention. In FIG. 9, data on the attributes of a customer who has made a call center inquiry is shown as the customer attribute data 302. As shown in FIG. 9, the customer attribute data 302 is a database in which attribute information of a customer who is a user of a product is registered when the inquiry from the customer is about the product. As a specific example, the customer attribute data 302 includes fields 3021 and 3022 for registering a "customer code" and a "customer name" for identifying the customer, respectively. Furthermore, one record is formed by including fields 3023, 3024, 3025, 3026, and 3027 for registering classification information based on the industry and function to which the customer belongs, and facility size, region, and maintenance contract information. In the example shown in FIG. 9, as information for identifying the customer, a "customer code" for uniquely identifying the customer is registered in field 3021, and a "customer name" which is the name of the customer represented by the customer code is registered in field 3022. In addition, in field 3023, "industry classification" is registered as information regarding the industry to which the customer belongs, and in field 3024, "functional classification" is registered as information regarding the function of the facility to which the customer belongs. Furthermore, in field 3025, "size" is registered as information regarding the size of the customer's facility, in field 3026, "area" is registered as information regarding the area to which the customer belongs, and in field 3027, "maintenance contract" is registered as information regarding whether a maintenance contract has been concluded with the customer. Other information regarding customer attributes may be registered as column 302n, etc.

[0020] An example of the data structure of the device attribute data 303 stored in the customer request integrated data lake unit 30 will be described with reference to FIG.

[0021] FIG. 10 is a diagram showing an example of the data configuration of the device attribute data 303 used in the embodiment of the present invention. In FIG. 10, the device attribute data 303 is exemplified by data on the attributes of the device related to the inquiry owned by the customer who made the inquiry in the call center inquiry data 301. As shown in FIG. 10, the device attribute data 303 is a database in which attribute information of the device used by the user who is the customer is registered. As a specific example, the device attribute data 303 includes a field 3031 for registering a "customer code" for identifying the customer. Furthermore, the device attribute data 303 includes fields 3032 and 3033 for registering information on the device model and the serial number of the device for identifying the device used by the customer, and fields 3034, 3035, 3036, 3037, and 3038 for registering information on the purpose of use of the product for each customer and the environment in which the device is installed, the manufacturing date and installation date of the device, and the number of years the device has been in operation, etc., forming one record. In the example shown in FIG. 10, in field 3032, "device type" is registered as information for identifying the type of device owned by the customer, and in field 3033, "serial number" is registered as information for uniquely identifying the device owned by the customer. In addition, in field 3034, "product use purpose" is registered as information regarding the use of the device, and in field 3035, "installation environment" is registered as information regarding the installation environment of the device, for example, use in cold regions, installation outdoors, etc. In addition, in field 3036, "manufacturing date" is registered as information regarding the manufacturing date of the device, and in field 3037, "installation date" is registered as information regarding the date the device was installed in the customer's facility. In addition, in field 3038, "years of operation" is registered as information regarding the period of time the device has been used from the installation date to the present. In addition to these, information regarding device attributes may be registered as column 303n, etc.

[0022] The customer attribute information and device attribute information shown in FIGS. 9 and 10 are merely examples. In this embodiment, the customer attribute information is data that characterizes the customer who made an inquiry to the call center. For example, data called a profile, such as the customer's gender, physical characteristics, hobbies, and occupation, may be used as the customer attribute information. In addition, data other than the profile, such as data indicating the characteristics of the business, such as the type of report content related to the inquiry to the call center and its importance and urgency, may be used as the customer attribute information. These attribute information are input by the analyst, that is, the user, in advance, assuming an analysis, and the respondent or writer is asked to input them, and they may not be added after the customer opinions are collected. However, once the customer opinions are collected, the addition of the information may not be prohibited. Since the items to be collected differ depending on the product to be newly developed based on the customer opinions, these attribute information may not be common to all the call center inquiry data 301 stored in the customer request integrated data lake unit 30. Therefore, it is important to set appropriate items to be collected arbitrarily, taking into account the detailed analysis based on the evaluation results described later. It is desirable to code the attribute information as a description method, taking into account the efficiency of the subsequent analysis. For example, this coding may represent an industry classification name in the customer attribute data 302 as a code such as "110101." The code may be represented in a binary format. By adopting such coding, it is possible to speed up processing.

[0023] Returning to FIG. 1, the description will be continued. The data statistical processing unit 40 performs data analysis processing using the call center inquiry data 301, the customer attribute data 302, and the device attribute data 303 stored in the customer request integrated data lake unit 30, using an existing text mining technique. Specifically, the data statistical processing unit 40 has a morphological analysis unit 41, a classification unit 42, and a counting unit 43. The morphological analysis unit 41 performs word separation processing of text-format document data, that is, word decomposition processing. The classification unit 42 executes counting of the frequency of occurrence of keywords among words appearing in each call center inquiry data according to the category classification registered in the keyword definition 201 in the extraction processing database unit 20 and the keywords that characterize the classification. In other words, the classification unit 42 classifies customer opinions into classification categories based on multiple category keywords. Here, an example of the data configuration of a table of keywords registered in the keyword definition 201 in the extraction processing database unit 20 is shown in FIG. 13. Here, the category into which multiple customer opinions are classified is the above-mentioned category classification. The above-mentioned category classification can be called classification categories for classifying customer opinions. There are multiple classification categories. Keywords that characterize each of the multiple classification categories are keywords registered in the keyword definition 201. These keywords can be called category keywords. There may be multiple category keywords.

[0024] Fig. 13 is a diagram illustrating an example of keyword definition 201 of the customer important opinion extraction system according to the embodiment of the present invention. Fig. 13 illustrates an example of keyword definition 201 as a table, which is used by the customer important opinion extraction system to tally up related keywords in order to extract categories with high customer needs.

[0025] As shown in Fig. 13, the table of keyword definition 201 is a database in which keywords for a given category classification are registered. Specifically, the table of keyword definition 201 includes a category classification 2011 in which a field "category classification" for classifying various call center inquiry data 301 into similar category classifications is registered, and a keyword 2012 in which a "keyword" for registering a keyword associated with each category classification is registered, forming one record. Examples of category classifications and their keywords include the keyword "education" for the category classification "training" and the keyword "method" for the category classification "operation". In keyword definition 201, multiple keywords may be defined for one category.

[0026] Returning to Fig. 1, the explanation will be continued. The aggregation unit 43 aggregates the number of inquiries for each predetermined target customer attribute and device attribute for the results categorized by keyword by the classification unit 42, and extracts categories with a large number of inquiries for each category and a high growth rate of the number of inquiries in a certain period of time, i.e., a high growth rate. For example, the aggregation unit 43 extracts classification categories for which the increase in customer opinions in a certain period of time is greater than a predetermined threshold. The processing of the data statistics processing unit 40 will be explained in detail later.

[0027] The machine learning model unit 50 uses a machine learning model learned using teacher data to determine the similarity of the importance 2027 of the important inquiry label 202 in the extraction processing database unit 20 based on the call center inquiry data 301 collected by the data statistical processing unit 40, extracts important inquiries, and notifies the user. Specifically, the machine learning model unit 50 has a keyword review processing unit 51, an important inquiry label review processing unit 52, a similarity determination processing unit 53, and an important inquiry evaluation processing unit 54. The keyword review processing unit 51 reviews keywords based on categories with high growth rates extracted by the data statistical processing unit 40, and reviews the keyword definition 201. This review processing of the keyword definition 201 is executed by reading the contents of the hypothesis verification result data 311, which is the result of verifying the problem hypothesis by interviewing customers who have made inquiries to the call center, based on the problem hypothesis data 312 registered in the customer request integrated data lake unit 30. The processing of the keyword review processing unit 51 will be described in detail later. FIG. 11 shows an example of the data structure registered in the problem hypothesis data 312 in the customer request integrated data lake unit 30, and FIG. 12 shows an example of the data structure registered in the hypothesis verification result data 311.

[0028] 11 is a diagram illustrating an example of problem hypothesis data 312 in the customer important opinion extraction system according to the embodiment of the present invention. This diagram illustrates, in the form of a table, problem hypothesis data 312, which is the result of formulating a customer problem hypothesis based on important inquiry data output from the customer important opinion extraction system. Here, the customer problems can be said to represent the customer needs.

[0029] As shown in FIG. 11, the table of problem hypothesis data 312 is a database in which the results of proposing hypotheses for customer problems and their causes and solutions are registered. As a specific example, the table of problem hypothesis data 312 includes a field 3121 for registering a "hypothesis ID" for identifying the proposed problem hypothesis for the customer, and a field 3122 for registering a "target customer attribute" for identifying the attribute of the customer to be the subject of the hypothesis. Furthermore, it is a single record including fields 3123, 3124, 3125, 3126, and 3127 for registering a specific problem hypothesis, its category classification, reference data, causes, and solutions. In the example shown in FIG. 11, "problem" is registered in field 3123 as specific information related to the problem hypothesis, and "category classification" is registered in field 3124 as category classification information for the problem hypothesis. Furthermore, in field 3125, a "reference data reception ID" is registered as call center reception information referenced to formulate a problem hypothesis, in field 3126, a "cause" is registered as information on an event that is the cause of the problem hypothesis, and in field 3127, a "solution" is registered as information on a solution to the problem hypothesis. In addition to these, information on hypotheses about customer problems may be registered as column 312n, etc.

[0030] 12 is a diagram illustrating an example of hypothesis verification result data 311 in the customer important opinion extraction system according to the embodiment of the present invention. This diagram illustrates, in the form of a table, hypothesis verification result data 311, which is a result of verifying a problem hypothesis by interviewing customers who have made inquiries to a call center, based on problem hypothesis data 312 in the customer important opinion extraction system.

[0031] As shown in FIG. 12, the table of hypothesis verification result data 311 is a database in which the results of verifying a hypothesis of a customer's problem are registered. As a specific example, the table of hypothesis verification result data 311 includes a field 3111 for registering a "hypothesis ID" for identifying a proposed customer's problem hypothesis, and fields 3112, 3113, 3114, and 3115 for identifying a verification target customer and customer attributes for which the problem hypothesis was verified. Furthermore, a field 3116 for registering whether the problem hypothesis is applicable or not as a verification result, a field 3117 for registering a reason for non-application, and the like are included to form one record. In the example shown in FIG. 12, a "customer code" is registered in the field 3112 as information for uniquely identifying a verification target customer for which the problem hypothesis was verified, and a "customer attribute" is registered in the field 3113 as information for identifying the target customer attribute of the verification target customer. Furthermore, in field 3114, "customer name" which is the name of the customer represented by the customer code is registered, and in field 3115, "verification partner" is registered regarding the department and position of the person who conducted the verification by interview, etc., among the customers who are the verification targets. Other items registered in this "verification partner" may be gender, age, etc. Furthermore, in field 3116, the result of verifying the hypothesis, whether the problem hypothesis is applicable or not is registered, and in field 3117, the reason when the problem hypothesis is not applicable is registered. In addition to these, information regarding the hypothesis verification result may be registered as column 311n, etc.

[0032] Of the data stored in the customer request integrated data lake unit 30, call center inquiry data 301 is an example of customer opinion data that includes customer opinions, and various data other than the call center inquiry data 301, such as customer attribute data 302 and device attribute data 303, are data related to customer opinions.

[0033] Returning to FIG. 1, the description will be continued. The important inquiry label review processing unit 52 reviews the important inquiry label 202 registered in the extraction processing database unit 20 based on inquiries that frequently appear in a category with a high growth rate extracted by the data statistical processing unit 40. The important inquiry label review processing unit 52 also executes the review processing of the important inquiry label 202 by reading the contents of the hypothesis verification result data 311, which is the result of verifying a problem hypothesis by interviewing customers who have made inquiries to the call center. The processing of the important inquiry label review processing unit 52 will be described in detail later. FIG. 14 is a diagram showing an example of the important inquiry label 202 in the customer important opinion extraction system according to the embodiment of the present invention. FIG. 14 shows an example of the data configuration registered in the important inquiry label 202 in the extraction processing database unit 20.

[0034] As shown in FIG. 14, the table of the important inquiry label 202 is a database in which data to which, for example, an expert has assigned importance to some or all of the call center inquiry data 301 is registered. This database includes importance data assigned to customer opinions. Here, an example of information to which importance has been assigned to a plurality of customer opinions is the table of the important inquiry label 202 described above. The table of the important inquiry label 202 may be created in advance when the customer important opinion extraction system is in operation. Here, the expert is a person who has sufficient skills to give an appropriate answer, including technical content, to a call center inquiry from a customer. The important inquiry label 202 is information to which importance has been assigned to a plurality of customer opinions. As a specific example, the table of the important inquiry label 202 includes a field 2021 for registering a "reception ID" for identifying the aggregated customer opinions, and fields 2022 and 2023 for identifying the customer. Furthermore, fields 2024, 2025, and 2026 for registering the date and time of receipt of an inquiry from a customer, the device used by the customer, the content of the inquiry from the customer, and the date and time of completion of the response, and importance 2027 for registering the importance given by an expert to the content of the inquiry from the customer are combined into one record. In the example shown in FIG. 14, as information for identifying a customer, a "customer code" for uniquely identifying a customer is registered in field 2022, and a "customer name" which is the name of the customer represented by the customer code is registered in field 2023. In addition, a "receipt date and time" is registered in field 2024 as information regarding the date and time of the customer's call center inquiry, a "device model" is registered in field 2025 as information regarding the device used by the customer, and a "receipt content" is registered in field 2026 as information regarding the content of the call center inquiry. The importance of inquiry data registered in Importance 2027 is an item that is entered by statistically judging market trends that experts have as tacit knowledge, the impact of past inquiries and their contents on customer operations, the complexity and difficulty of work when local response is required, etc. The greater the impact an inquiry has on customers, the more important the importance may be.Here, the importance level is given a three-level range of high, medium, and low, but there is no restriction on how this value should be taken, and it may be set arbitrarily.

[0035] Returning to FIG. 1, the description will be continued. The similarity discrimination processing unit 53 performs similarity discrimination on customer opinions. That is, the similarity discrimination processing unit 53 performs similarity discrimination on customer opinions using importance data having importance related to customer opinions for the classification categories extracted by the aggregation unit 43. The similarity discrimination processing unit 53 performs similarity discrimination processing of inquiry text data by a machine learning model for inquiries in a category extracted by the data statistical processing unit 40, which has a large number of inquiries and a high growth rate, which is the rate at which the number of inquiries grows in a certain period. During learning of this machine learning model, the important inquiry label 202 reviewed by the important inquiry label review processing unit 52 is used as teacher data. The important inquiry label review processing unit 52 reviews the importance data. As a result of the similarity discrimination processing, each inquiry data is assigned an importance similar to the text data to which the important inquiry label 202 is assigned, and the important inquiry evaluation processing unit 54 extracts inquiry data determined to have a high importance, and notifies the user via the input / output unit 10 as important inquiry data. The important inquiry evaluation processing unit 54 extracts the customer opinions with a high importance. The important inquiry evaluation processing unit 54 reviews importance data regarding customer opinions belonging to the classification categories extracted by the aggregation unit 43. The important inquiry evaluation processing unit 54 extracts and outputs customer opinions with high importance. The processing of the similarity discrimination processing unit 53 and the important inquiry evaluation processing unit 54 will be described in detail later.

[0036] Next, the process executed by the customer important opinion extraction system of this embodiment will be described with reference to FIGS. 2, 3, 4, 5, 6 and 7. FIG.

[0037] FIG. 2 is a diagram showing a representative example of the operation of a user who uses the customer important opinion extraction system of this embodiment and a processing flow in the customer important opinion extraction system. FIG. 3 is a diagram showing a schematic flow of processing in the customer important opinion extraction system of the embodiment of the present invention. FIG. 4 is a diagram showing a flow of important inquiry area extraction processing by call center inquiry data aggregation in the customer important opinion extraction system of the embodiment of the present invention. FIG. 5 is a diagram showing a flow of important inquiry label review processing in the customer important opinion extraction system of the embodiment of the present invention. FIG. 6 is a diagram showing a flow of important inquiry data evaluation / extraction processing in an important inquiry area in the customer important opinion extraction system of the embodiment of the present invention. FIG. 7 is a diagram showing a flow of review processing of keyword definition 201 in the customer important opinion extraction system of the embodiment of the present invention.

[0038] First, a representative example of the operation of a user who uses the customer important opinion extraction system of this embodiment and the processing flow executed by the customer important opinion extraction system will be described with reference to FIG. 2. First, the user performs "keyword initial input" (S1) of the customer important opinion extraction system, and as a result, an initial keyword is defined in the system as "initial keyword registration" (S2). This initial keyword is referred to in the processing of "data statistical processing" (S4). Here, the keyword defined as the initial setting and the analysis data in the customer request integrated data lake unit 30 are read in the processing of "analysis data" (S3), and the customer important opinion extraction system 1 executes "data statistical processing" (S4). Here, the analysis data is the call center inquiry data 301. Similarly, the user also performs "important inquiry label initial input" (S5) which becomes the teacher data for machine learning in "machine learning processing" (S7). In response to this, an initial important inquiry label is defined in the system in "initial important inquiry label registration" (S6).

[0039] After that, the customer important opinion extraction system 1 executes the "machine learning process" (S7) using the defined important inquiry label 202 and the analysis data that has been subjected to the above-mentioned statistical process. Here, the "machine learning process" (S7) learns to output the contents of the important inquiry label 202 to be recommended in the "keyword ranking recommendation" (S8), the keywords to be corrected in the "keyword correction process" (S10), the similar inquiry record number ranking to be recommended in the "similar inquiry record number ranking recommendation" (S11), and the important inquiry label 202 to be corrected in the important inquiry label correction process (S13), using the important inquiry label 202 and the result of the "data statistical process" (S4) as inputs.

[0040] This machine learning process performs automatic correction processing in the "keyword correction processing" (S10) based on frequently occurring keywords, etc., and recommends keyword rankings to the user in the keyword ranking recommendation (S8). The user checks the recommendation result, and in the "keyword correction input" (S9), inputs a keyword to correct the initially defined keyword as necessary. Here, the "keyword" recommended in step S8 is a candidate to be used as a keyword. It should be noted that the candidates to be used as keywords may include words other than those listed in the keyword definition 201 at the time of recommendation. This consideration also applies to step S26 described later. In response to the input of the corrected keyword, in the customer important opinion extraction system 1, in the keyword correction processing (S10), the initially defined keyword is corrected to the keyword input in the "keyword correction input" (S9). Similarly, based on the results of the "machine learning process" (S7), the system also performs a correction process for the important inquiry label 202 in the "important inquiry label correction process" (S13), and a recommendation of a similar inquiry record count ranking is performed to the user in the "similar inquiry record count ranking recommendation" (S11). The user checks the recommendation result, and performs correction input of the defined important inquiry label 202 as necessary ("important inquiry label correction input" (S12)), and performs the "important inquiry label correction process" (S13) defined in the customer important opinion extraction system 1.

[0041] The specific content of the "important inquiry label correction input" (S12) is, for example, a correction of the importance 2027 in the important inquiry label 202 illustrated in FIG. 14. In FIG. 2, two inputs, namely, the "machine learning process" (S7) and the "keyword correction input" (S9), can be input to the keyword correction process (S10). In this case, when an input from the "machine learning process" (S7) is first made and then the "keyword correction input" (S9) is made, the information input from the "keyword correction input" (S9) may be treated as the corrected keyword in the "keyword correction process" (S10). Here, the user may be able to set on the customer important opinion extraction system 1 which information is to be given priority between the information input from the "machine learning process" (S7) and the information input from the "keyword correction input" (S9) in the "keyword correction process" (S10). Even after this, when there is an input from the machine learning process and an input from the user side, the user may be able to set on the customer important opinion extraction system 1 the priority between the two inputs. For example, two inputs, the "machine learning process" (S7) and the "important inquiry label correction input" (S12), may be input to the "important inquiry label correction process" (S13). In this case, when an input is first made from the "machine learning process" (S7) and then the "important inquiry label correction input" (S12) is made, the information input from the "important inquiry label correction input" (S12) may be treated as a corrected keyword in the "important inquiry label correction process" (S13) based on the settings in the customer important opinion extraction system 1.

[0042] The important inquiries extracted by the "important inquiry extraction" (S14) as a result of the "machine learning process" (S7) are notified to the user, and the user checks the extracted data in the procedure of "important inquiry data confirmation" (S15), and based on the data, formulates a hypothesis about the customer's problem in the procedure of "customer problem hypothesis formulation" (S16). The formulated hypothesis is registered in the system by the user in the procedure of "hypothesis registration / reference" (S17). At that time, the user can refer to customer problem hypotheses formulated in the past as necessary. This formulated hypothesis is verified by the user through interviews with the customer in "hypothesis verification" (S18). The result of the hypothesis verification is input to the system by the user in the process of "hypothesis verification result input" (S19). As a result, the customer important opinion extraction system 1 registers the hypothesis verification result on the customer important opinion extraction system 1 side in the process of "hypothesis verification result registration" (S20). In the subsequent extraction of important inquiries, "revised keyword reference" (S21) is performed for the keywords revised in the "keyword revision process" (S10), and "data for analysis" (S22) is used to perform "data statistical processing" (S23). Here, the "revised keyword reference" (S21) process refers to the keywords revised in the "keyword revision process" (S10), and the data for analysis may be the call center inquiry data 301. In addition, the customer important opinion extraction system 1 performs "revised important inquiry label reference" (S24) for the important inquiry label 202 revised in the "important inquiry label revision process" (S13), and performs "machine learning processing" (S25) using the analysis data that has been subjected to statistical processing. The system learns the contents of the "machine learning processing" (S25) from the hypothesis verification results registered in the "registration of hypothesis verification results" (S20), the contents of the revised important inquiry label 202 in the "revised important inquiry label reference" (S24), and the results of the "data statistical processing" (S23) as training data.In addition, the customer important opinion extraction system 1 performs learning to output the contents of each of the keywords to be recommended in the "keyword ranking recommendation" (S26), the keywords to be corrected in the "keyword correction process" (S28), the similar inquiry record number ranking to be recommended in the "similar inquiry record number ranking recommendation" (S29), and the important inquiry label 202 to be corrected in the "important inquiry label correction process" (S31).

[0043] Through these processes, the customer important opinion extraction system 1 performs processes related to the correction of the defined keywords (S26 to S28) and the correction of the important inquiry label (S29 to S31) as necessary, similar to the processes in S8 to S13. In addition, as a result of the "machine learning process" (S25), the important inquiries extracted in the "important inquiry extraction" (S32) are notified to the user, and the user can check the extracted data in the "important inquiry data confirmation" (S33) and, in the same manner as in S16 and after, consider a means to solve the customer's troubles by formulating and verifying a hypothesis based on the extracted customer important opinion. The process flow executed by the customer important opinion extraction system will be described in detail below.

[0044] As shown in FIG. 3, in the customer important opinion extraction system of this embodiment, the customer important opinion extraction system 1 first determines whether the keyword definition 201 and the important inquiry label 202 to be registered in the extraction processing database unit 20 have been initially registered in "Initial registration?" (210). If it is determined that they have been initially registered, the process proceeds to "Important inquiry area extraction by aggregation of call center inquiry data" (401). If the keyword definition 201 and the important inquiry label 202 have not been initially registered, the customer important opinion extraction system 1 executes "Keyword initial setting read process" (211) and "Important inquiry label initial setting read" (221) processes prepared by the user. These read processes are executed via the input / output unit 10. Next, the customer important opinion extraction system reviews the teacher data settings for the importance evaluation of inquiry data to the latest information in the process of "Review of important inquiry label" (521), and then presents the extracted important inquiry data to the analyst, i.e., the user, in the process of "Important inquiry data evaluation and extraction in important inquiry area" (531). Next, the customer important opinion extraction system 1 executes the registration of the customer problem hypothesis input by the user in "Hypothesis registration / reference" (S250). When registering the problem hypothesis, the system can execute a display process of the customer problem hypothesis formulated in the past, as necessary, in response to the user's request. Next, the customer important opinion extraction system 1 reads the revised version of the keywords by the process of "revised keyword reading" (251). Next, the customer important opinion extraction system 1 reads the revised version of the important inquiry label 202 by the process of "revised important inquiry label reading" (261). Next, the customer important opinion extraction system 1 proceeds to the process of "important inquiry area extraction by aggregation of call center inquiry data" (451). This process is executed using the revised version of the keywords. Next, the customer important opinion extraction system 1 executes a review of the teacher data setting for the importance evaluation of the inquiry data by the process of "review of important inquiry label" (551).Next, the customer important opinion extraction system 1 presents the extracted important inquiry data to the analyst, i.e., the user, through the process of "evaluating and extracting important inquiry data in important inquiry domain" (561). The processes of steps 551 and 561 are executed using the revised keyword and the revised important inquiry label 202. It should be noted that, regarding the revised keyword and the revised important inquiry label 202, if no revision is made, the original keyword and important inquiry label 202 without revision are used.

[0045] In the following, the processes of "important inquiry area extraction by call center inquiry data aggregation" (401 and 451), "review of important inquiry labels" (521 and 551), and "evaluation and extraction of important inquiry data in important inquiry areas" (531 and 561) will be described. The process of step 401 is executed in the data statistical process (S4). The process of step 521 is executed in the machine learning process (S7). The process of step 531 is executed in the important inquiry extraction (S14). The process of step 451 is executed in the data statistical process (S23). The process of step 551 is executed in the machine learning process (S25). The process of step 561 is executed in the important inquiry extraction (S32). The data statistical process (S4) and the data statistical process (S23) are both processes executed by the data statistical processing unit 40. The machine learning process (S7), the machine learning process (S25), the important inquiry extraction (S14), and the important inquiry extraction (S32) are processes executed by the machine learning model unit 50.

[0046] FIG. 4 is a specific example of a flowchart of the process of "important inquiry area extraction by call center inquiry data aggregation" (401 and 451) in FIG. 3. As shown in FIG. 4, the customer important opinion extraction system 1 performs the process of "important inquiry area extraction by call center inquiry data aggregation" (401) by reading out the call center inquiry data 301, customer attribute data 302, and device attribute data 303 to be analyzed from the customer request integrated data lake unit 30, and then the morphological analysis unit 41 first performs a morphological analysis process to analyze the document data included in the call center inquiry data 301 in "word extraction by morphological analysis unit for call center inquiry data" (411). The read call center inquiry data 301 and customer attribute data 302 are stored in a buffer memory (not shown) provided in the data statistical processing unit 40. The system performs a word segmentation process, that is, a word decomposition process, on the call center inquiry data 301 in the morphological analysis unit 41. The morphological analysis technique executed by the morphological analysis unit 41 is implemented by existing techniques, provided that appropriate word decomposition is possible, and therefore a description thereof will be omitted here.

[0047] Next, in the process of "extracting inquiries in which keywords appear" (421), the customer important opinion extraction system 1 extracts call center inquiry data 301 in which the keywords appear in the classification unit 42. Furthermore, the customer important opinion extraction system 1 classifies the call center inquiry data into each category in the process of "classifying and sorting by category" (422). Here, the process of step 422 classifies customer opinions into categories, i.e., classification categories, based on keywords, i.e., category keywords. That is, in step 422, the customer important opinion extraction system 1 classifies the classification categories based on the results of performing morphological analysis on the customer opinions. The classification unit 42 classifies the customer opinions into classification categories based on a plurality of category keywords.

[0048] The process of "extraction of inquiries in which keywords appear" (421) is executed using the word list data, which is the morphological analysis result output from morphological analysis unit 41 and is the result of morphological analysis of each call center inquiry data 301. That is, first, keywords registered in keyword definition 201 in extraction process database unit 20 are extracted from this word list. In this way, in step 421, call center inquiry data 301 from which keywords have been extracted is extracted.

[0049] Next, the process of "classification and sorting by category" (422) is executed. That is, the customer important opinion extraction system 1 refers to the correspondence between the category classification 2011 and the keyword 2012 defined in the keyword definition 201, and determines the category classification to which the extracted keyword belongs as the category to which the call center inquiry data 301 belongs. As a result of extracting keywords from the above-mentioned word list, there may be a case where a plurality of keywords exist in one call center inquiry data 301. In this case, the customer important opinion extraction system 1 determines the category to which the largest number of keywords apply when applying the plurality of keywords to the keywords 2012 in the keyword definition 201 as the category to which the call center inquiry data 301 belongs. The method of processing the classification and sorting by category (422) is not limited to the above-mentioned one. When classifying call center inquiry data 301 into categories, rather than simply deciding which category has the most keywords, as described above, it is possible to assign a weight to each keyword, and when the multiple keywords described above are applied to keywords 2012 in keyword definition 201, the sum of the weights of the applied keywords belonging to each category is calculated, and the category with the greatest weight is decided as the category to which call center inquiry data 301 belongs.

[0050] The classification unit 42 also acquires, in association with the word list data, the record of the call center inquiry data 301 from which the word list was extracted, and the customer attribute data 302 and the device attribute data 303 associated with the record. For association between the call center inquiry data 301 and the customer attribute data 302 and the device attribute data 303, the customer important opinion extraction system 1 can use a customer code. Here, in this embodiment, the category classification is set by an analyst, i.e., a user, as a major classification of problems associated with the business of a customer, i.e., a device user, such as device malfunctions and operation methods, but it is desirable to decide on the category classification after sharing knowledge with experts who are familiar with the customer business.

[0051] Then, in "aggregating the number of inquiries by category classification x target customer / device attribute" (431), the aggregation unit 43 multiplies the category classification by the target customer / device attribute for the categorized call center inquiry data 301 output from "classifying and sorting by category" (422), and executes the process of aggregating the number of inquiries in each area. The aggregation unit 43 extracts classification categories in which customer opinions are classified based on the category keywords, and among customer opinions belonging to the classification categories that have more than a predetermined number of customer opinions, classification categories in which the number of customer opinions at the second timing is greater than the number of customer opinions at the first timing. Here, the second timing is the current time, and the first timing may be a past time point going back from the current time. The first timing and the second timing will be described in detail later. Specifically, this process is, for example, as follows. As a result of executing the process of "classifying and sorting by category" (422), a category to which each call center inquiry data belongs is assigned, and a set of call center inquiry data with attribute data is formed in which the call center inquiry data is associated with the customer attribute data 302 and the equipment attribute data 303 corresponding to the call center inquiry data. For each of the attribute data-attached call center inquiry data for each category, the customer important opinion extraction system 1 classifies data belonging to a data item selected from the customer attribute data 302 of the customer corresponding to the call center inquiry data, such as customer name and reception details. Furthermore, for each of the selected data, the customer important opinion extraction system 1 classifies data belonging to a data item selected from the equipment attribute data 303, such as equipment model and installation environment. A collection of data resulting from these two classifications is regarded as one area. This area corresponds to a collection of call center inquiry data for each category for which the process of "classifying and sorting by category" (422) has been executed, the collection having at least one data item selected from the corresponding customer attribute data 302 and at least one data item selected from the corresponding device attribute data 303. By tallying up the number of call center inquiry data for each area created in this way, the number of inquiries for each category can be found.By executing such a process, the contents of the call center inquiry data 301 included in each category can be analyzed.

[0052] As described above, the process of step 431 is to extract and tally the number of pieces of call center inquiry data for each category based on the customer opinions, customer attributes, and device attributes. The specific process of step 431 is not limited to the above example, and the number of pieces of call center inquiry data may be extracted and tallyed by various methods. The process of step 431 may be to extract the number of customer opinions for each classification category based on a combination of customer opinions, data on customer attributes, and data on product attributes.

[0053] Here, in this embodiment, the items of customer attributes and the items of device attributes to be multiplied with the call center inquiry data 301 are not specified. However, it is desirable to specify attributes according to the customer segment defined by the analyst, i.e., the user, such as the industry to which the customer belongs and the size of the customer, in consideration of the targeting of subsequently executing hypothesis formulation and hypothesis verification. Furthermore, the customer important opinion extraction system 1 "extracts the top M categories with the largest number of inquiries" (432) as a result of the category classification and aggregation in the aggregation unit 43. Here, M may be a positive integer and a threshold value previously set by the analyst, i.e., the user. Furthermore, the customer important opinion extraction system 1 executes a process of evaluating the growth rate of each category for the M categories extracted in step 432 in "comparing the number of inquiries with the same category as of N months ago and evaluating the growth rate of each category" (433). In this process, the customer important opinion extraction system 1 calculates, for example, the number of call center inquiry data at an arbitrarily specified time point for each of the M categories to which the call center inquiry data belongs. Furthermore, the customer important opinion extraction system 1 calculates the number of call center inquiry data belonging to the same category for the N months prior to that point in time, using the same procedure as described above. From these two data number calculation results, the customer important opinion extraction system 1 calculates the rate of change in the number of call center inquiry data divided by N, that is, the growth rate. In this way, the customer important opinion extraction system 1 evaluates the growth rate for each of the M categories. Here, N may be a positive real number and may be a period set in advance by the analyst, i.e., the user. The period for evaluating the growth rate can be set arbitrarily, but it can also be set according to legal amendments, such as 12 months.

[0054] Next, the customer important opinion extraction system 1 executes the process of "automatically extracting the category with the highest growth rate" (434). In step 434, the customer important opinion extraction system 1 extracts the category with the highest growth rate from the result of the growth rate of each category calculated in step 433. The process of step 434 is executed by the aggregation unit 43. As for the results aggregated by the aggregation unit 43, not only the category with the highest growth rate but also the growth rate ranking by category, the inquiry number ranking by category, and the call center inquiry data 301 classified into each category may all be passed to the machine learning model unit 50, or some of the data may be passed as necessary. That is, the process in step 434 may be a process of extracting a classification category with a large increase in the number of changes from the change in the number of customer opinions belonging to the classification category in a predetermined period. Also, as described later, the process in step 434 may be a process of extracting a classification category in which the number of customer opinions belonging to the classification category is greater than a predetermined number and the number of customer opinions belonging to the classification category has increased by a predetermined amount compared to the number of customer opinions belonging to the classification category at a predetermined past time point, for classification categories in which customer opinions are classified based on category keywords.

[0055] The above-mentioned process from step 411 to step 434 is an example of extracting a classification category with a large increase from the change in the number of customer opinions belonging to the classification category in a predetermined period. The predetermined period is a first timing and a second timing corresponding to the start and end of the predetermined period, respectively. In other words, for example, the customer important opinion extraction system 1 evaluates the increase from the change in the number of customer opinions belonging to the classification category at both timings, with the current time when the customer important opinion extraction system is used being the second timing, which is the end time of the predetermined period, and a past time from the current time being the first timing. Also, as long as the process is to extract a classification category with a large increase, the process may not be the one described from step 411 to step 434. In step 431, the process may be simplified and the customer important opinion extraction system 1 may calculate the number of call center inquiry data belonging to each category. In step 433, the concept of growth rate is used for explanation, but this is a ratio between the number of call center inquiry data a under a given condition at the time when the customer important opinion extraction system is used and the number of call center inquiry data b under the same given condition at a time several months before that time, for example, "a / b", which is treated as the growth rate. However, this does not necessarily have to use the concept of a ratio, and it can be simply evaluated, for example, by the increase from the number of call center inquiry data a under a given condition at the present time to the number of call center inquiry data b under the same given condition at a certain time in the past. In the above explanation of step 433, the concept of a ratio of growth rate is used, but the word growth rate may be understood to mean an increase. In addition, in step S434, the process of "automatically extracting a category with the highest growth rate" has been explained as described above. However, in this process, instead of "automatically extracting a category with the highest growth rate", it may simply be "extracting a category with a high growth rate". That is, for example, the top K categories with the highest growth rates may be extracted and then processed further.In addition, "automatic extraction" expresses that the process is performed automatically, and may be understood as "extraction." This "automatic" applies to other processes as well.

[0056] So far, the processing of "important inquiry area extraction by aggregating call center inquiry data" (401) has been described. Next, the processing of "important inquiry area extraction by aggregating call center inquiry data" (451) will be described. In the processing of step 451, any items that have been updated or modified from the processing of step 401, such as keyword definition 201, are used. In the processing of step 451, any items that overlap with the processing of step 401 may be omitted.

[0057] FIG. 5 shows a specific example of a flowchart for the process of "reviewing important inquiry labels" (521 and 551). Next, the process of "reviewing important inquiry labels" (521) will be described in detail with reference to FIG. 5. The system first executes the process of "automatically extracting similar inquiries that frequently appear in the category with the highest growth rate" (5211) for the category with the highest growth rate automatically extracted by the aggregation unit 43. Here, a method for automatically extracting similar inquiries that frequently appear will be described. First, the customer important opinion extraction system 1 converts each sentence of the inquiry data into a vector, and quantitatively expresses the semantic distance between sentences by the vector. If sentences can be converted into a fixed-length vector expression, it is possible to extract sentences with a close inter-vector distance by applying the vector expression to a machine learning algorithm such as clustering. Note that the technology for determining similar sentences executed by "automatically extracting similar inquiries that frequently appear in the category with the highest growth rate" (5211) is realized by existing technology under the condition that sentences of any length can be converted into fixed-length vectors and the similarity of the vectors can be calculated, and therefore a detailed description thereof will be omitted here. Thereafter, the customer important opinion extraction system 1 determines whether or not there are K or more similar inquiries based on the extraction result of the similar inquiry records in "Are there K or more similar inquiries?" (5212), and if necessary, counts and ranks the number of extracted similar inquiry records, and notifies, i.e., recommends, the analyst, i.e., the user ("Notify the user of the ranking of the number of frequently occurring similar inquiry records" (5213)). K may be a positive integer and may be a threshold number set in advance by the analyst, i.e., the user. The number of similar inquiry records may be the same as the number of similar call center inquiry data. Here, if the important inquiry label 202 is simply reviewed manually, for example, by the intuition or experience of an expert, it depends on the knowledge that the analyst, i.e., the user or the expert, can know. However, by executing the review process of the important inquiry label 202 as described above, the customer important opinion extraction system 1 can perform recommendations to the user within a range including frequently occurring similar call center inquiry data in the category with the highest growth rate.In other words, by monitoring the time series changes and trend changes in the ranking of frequently asked questions, the system can recommend changes to users and give them notice. As a result, it becomes possible to review the important question labels 202 based on more timely market conditions and trends surrounding users, and extract important opinions that better match the needs of users.

[0058] In the process of step 5211 described above, the customer important opinion extraction system 1 extracts frequently-asked similar inquiries for the customer opinions belonging to the classification category extracted in step 434. In addition, in the process of step 5213, the customer important opinion extraction system 1 outputs, i.e., recommends, the frequently-asked similar inquiries extracted in step 5211 via the input / output unit 10.

[0059] Based on the recommended results, the user judges whether or not to revise the important inquiry label 202 ("Revise important inquiry label based on notification?" (5214)), and if necessary, "accepts revision of important inquiry label" (5215) based on input from the user. That is, the customer important opinion extraction system 1 can revise the important inquiry label 202 based on input from the input / output unit 10. The revision content accepted in step 5214 may be temporarily stored in a buffer memory. Here, the customer important opinion extraction system can make the judgment in step 5214 by judging input from the analyst, i.e., the user, regarding whether or not the revision is necessary.

[0060] In addition, in step 5215, the customer important opinion extraction system 1 can modify the contents of the important inquiry label 202, specifically, the importance level, based on the similar inquiries extracted in step 5211, the inquiries with K or more similar inquiries extracted in step 5212, and the high-ranked inquiries notified to the user in step 5213. Regarding which of the modification contents input by the user in step 5215 and the contents automatically modified by the customer important opinion extraction system 1 is to be adopted with priority, a setting of whether or not to change the important inquiry label 202 is provided, and a process of which is to be prioritized in the setting is set, and the customer important opinion extraction system 1 can execute the process according to the setting. That is, in the process of 5215, the customer important opinion extraction system 1 can change the importance level based on the input from the input / output unit 10 according to the contents of the important inquiry label 202, specifically, the setting of whether or not to change the importance level. Furthermore, the customer important opinion extraction system 1 changes the importance level based on the extraction result of frequently appearing similar customer opinions according to the setting of whether or not to change the importance level. For example, the customer important opinion extraction system 1 changes the importance of the important inquiry label 202 based on frequently similar customer opinions from a "medium" importance level before the change to a "high" importance level.

[0061] Furthermore, the customer important opinion extraction system 1 determines whether or not there is a hypothesis verification result for the problem hypothesis in "Hypothesis verification result available?" (5216), and if there is no hypothesis verification result, executes the process of "Automatic correction of important inquiry label" (5218). In this process, the customer important opinion extraction system 1 corrects the important inquiry label 202 based on the contents of the corrected important inquiry label 202 received in step 5215. The customer important opinion extraction system 1 can determine whether or not there is a hypothesis verification result by whether or not the registration of the hypothesis verification result (S20) in FIG. 2 has been completed. If there is a hypothesis verification result, the user can execute the process of feedback of the hypothesis verification result (5217) to the important inquiry label 202, thereby executing the process of correcting the important inquiry label (5218) that takes into account the latest hypothesis verification result, that is, the change in the customer's business with respect to the expert's past knowledge. Specifically, for example, the system can execute a process of lowering the importance label by one level for the reception ID (3124) of the important inquiry data referenced and associated with the hypothesis ID (3111) whose problem was not applicable in the hypothesis verification result in the hypothesis verification result data 311 shown in FIG. 12. When there is a hypothesis verification result, the position of the call center inquiry data to be verified changes based on the accuracy of the hypothesis as a result of the hypothesis verification, and as a result, the position of the ranking recommended in step 5213 may change. Therefore, when there is a hypothesis verification result, the customer important opinion extraction system 1 can create a proposal to change the contents of the important inquiry label 202 to be corrected in the process of step 5217. For example, when the hypothesis verification result shows that the problem applicability 3116 in the hypothesis verification result data 311 is "applicable", this proposal raises the importance 2027 of the relevant important inquiry label 202 by one level. Similarly, for example, if the problem applicability 3116 is “not applicable”, the importance 2027 of the relevant important inquiry label 202 is lowered by one level. Also, in step 5217, the customer important opinion extraction system 1 outputs the contents of the hypothesis verification result via the input / output unit 10, and the user can confirm the output contents and input necessary corrections for the important inquiry label 202.In the process of step 5218 following step 5217, the contents in the important inquiry label 202 can be automatically amended to, for example, the proposed change created in step 5217. The presence or absence of a hypothesis verification result is determined based on whether or not the registration (S20) of the hypothesis verification result in Fig. 2 has been completed. Regarding whether or not to give priority to the amendment content input by the user in step 5217 or the amendment content automatically amended by the customer important opinion extraction system 1 in step 5218, a setting is provided as to whether or not to make an amendment to the important inquiry label 202, and processing such as which is to be given priority is set in the setting, and the customer important opinion extraction system 1 executes processing in accordance with the setting.

[0062] The processing from steps 5216 to 5218 is processing in which the customer important opinion extraction system 1 changes the content of the importance data in accordance with the setting of whether or not the importance data needs to be changed based on the content of the hypothesis verification result input from the input / output unit 10, and changes the content of the importance data based on the input from the input / output unit 10, based on the setting of whether or not the importance data needs to be changed.

[0063] So far, the process of "reviewing important inquiry labels" (521) has been described. Next, the process of "reviewing important inquiry labels" (551) will be described. In the process of step 551, the keyword definitions 201, important inquiry labels 202, etc., that have been updated or modified in the process of step 521 are used. The process of step 451 can be omitted if it overlaps with the process of step 401.

[0064] Next, the process of "important inquiry data evaluation and extraction in important inquiry area" (531) will be described in detail with reference to FIG. 6. FIG. 6 is a specific example of the process of "important inquiry data evaluation and extraction in important inquiry area" (531 and 561). First, the customer important opinion extraction system 1 executes "similarity discrimination process of inquiry data in extracted priority review category" (5311) for inquiry data in the category with the highest growth rate extracted in the process of step 434 (this is called "priority review category") in the aggregation unit 43 using the important inquiry label 202 updated by the important inquiry label review processing unit 52. The process of step 5311 is a process of extracting customer opinions similar to customer opinions with high importance by machine learning using importance data as training data for customer opinions belonging to the classification category with a large increase extracted in the process of step 434. Here, in the similarity discrimination process, the customer important opinion extraction system 1 executes importance discrimination from the similarity with the target call center inquiry data 301 by machine learning using the important inquiry label 202 as training data. That is, the customer important opinion extraction system 1 extracts inquiry data similar to inquiry data with high importance 2027 raised to the important inquiry label 202 from the inquiry data in the aforementioned priority consideration target category. The inquiry data extracted here is assigned a similarity based on the degree of similarity. In the similarity determination processing unit, the customer important opinion extraction system 1 extracts customer opinions similar to customer opinions with high importance in the importance data by machine learning, assigns importance to the extracted customer opinions with high importance in the importance data according to the level of importance, and after the assignment of the importance, the important inquiry evaluation processing unit 54 outputs the customer opinions with a predetermined importance for the customer opinions. Note that, regarding the similarity determination processing technology executed by the similarity determination processing unit 53, for example, a method of converting text data into a vector representation, i.e., a numerical representation, which is often used in natural language processing, can be cited as a method, but the method is not particularly limited.Furthermore, in order to extract important questions with greater accuracy, parameter tuning according to the machine learning method and algorithm is essential; however, since this can be achieved using existing established methods and technologies, we will not explain this here.

[0065] Next, based on the result of the similarity discrimination process, the customer important opinion extraction system 1 executes the process of "assigning importance to inquiry data" (5411) for each inquiry data. In step 5411, importance is assigned to customer opinions similar to the customer opinions with high importance extracted in step 5311. That is, in step 5411, the customer important opinion extraction system 1 outputs inquiry data similar to the inquiry data with high importance 2027 via the input / output unit 10, and the user understands the output contents and inputs and sets an appropriate importance for the similar inquiry data via the input / output unit 10. Also, in step 5411, the customer important opinion extraction system 1 automatically assigns importance to inquiry data similar to the inquiry data with high importance 2027 extracted in step 5311. For example, the automatically assigned importance is assigned to inquiry data extracted as being similar to a certain inquiry data with high importance 2027 raised to the important inquiry label 202, with the same importance as that certain inquiry data. The importance can also be given based on the degree of similarity between the two data to be compared. For example, if the degree of similarity is high, the degree of similarity to be given is increased by one level, and if the degree of similarity is low, the degree of similarity to be given is decreased by one level. In step 5411, regarding whether to give priority to the input from the user or the automatic input from the customer important opinion extraction system 1 in giving the importance, a setting is provided for, for example, whether or not to change the degree of importance to be given, and processing such as which is given priority in the setting is set, and the customer important opinion extraction system 1 executes the processing according to the setting.

[0066] For the inquiry data to which importance has been assigned in step 5411, in "automatically extract inquiry data (top L items) with high importance" (5412), the customer important opinion extraction system 1 extracts and outputs customer opinions with a predetermined importance from the customer opinions. That is, in step 5412, the customer important opinion extraction system 1 executes a process of sorting and automatically extracting the top L items of inquiry data with high importance from the inquiry data to which importance has been assigned in step 5411, and outputs the inquiry data with high importance via the input / output unit 10. Here, L is a positive integer and is a threshold value arbitrarily set in advance by the analyst, i.e., the user. In addition, the customer important opinion extraction system 1 does not necessarily execute the process of sorting and outputting the top L items here, and can, for example, extract inquiry data with high importance and output the result as it is. Here, the process of outputting the inquiry data with high importance via the input / output is executed in important inquiry extraction S14 in FIG. 2. The process of S15 is executed by referring to the inquiry data with high importance output by this process. The processes in S32 and S33 are similar to those in S14 and S15, respectively.

[0067] So far, the process of "evaluating and extracting important inquiry data in important inquiry regions" (531) has been described. Next, the process of "evaluating and extracting important inquiry data in important inquiry regions" (561) will be described. In the process of step 561, the important inquiry labels 202 and the like that have been updated or modified from the process of step 531 are used. The process of step 561 may be omitted if it overlaps with the process of step 531.

[0068] Next, the flow of the process of reviewing the keyword definition 201 in the keyword review processing unit 51 of the machine learning model unit 50 will be described in detail with reference to FIG. 7. In the keyword review processing unit 51, the customer important opinion extraction system 1 executes the review of the category keywords. First, the customer important opinion extraction system 1 executes the process of "aggregating frequently occurring keywords in the inquiry data in the category with the highest growth rate" (511), which is the process of aggregating frequently occurring keywords in the call center inquiry data 301 in the category with the highest growth rate automatically extracted by the aggregation unit 43. After executing this process, the customer important opinion extraction system 1 executes the process of ranking the frequently occurring keywords and "notifying the analyst, i.e., the user, of the ranking of the frequently occurring keywords" (512). This notification is a recommendation for the user. Note that the ranking of the keywords notified in step 512 here is a ranking of candidates to be used as keywords. It should be noted that the candidates to be used as keywords may include words other than those listed in the keyword definition 201 at the time of recommendation. Here, the process of outputting keyword candidates that are relatively frequently included in customer opinions belonging to the extracted category is the one described above. That is, the process of step 512 is a process of extracting and outputting category keywords that frequently appear in customer opinions belonging to the extracted classification category with a large increase. In this way, instead of simply reviewing the keywords manually, for example, by the intuition or experience of an expert, the system can recommend a change to the keyword definition 201 to the user by watching the time-series changes and trend changes of the frequent keyword rankings. Based on the recommended results, the user judges whether or not to correct the keyword definition 201 as shown in "Modify keyword definition based on notification?" (513), and when the user executes the correction input, the system accepts the correction of the keyword definition 201 in "Accept correction of keyword definition" (514). Here, the judgment in step 513 is a judgment of input of the presence or absence of the correction from the analyst, i.e., the user, as the customer important opinion extraction system. The process described here is a process of reviewing the keyword according to the keyword correction input.Also, here, instead of the above-mentioned keyword correction input, the customer important opinion extraction system 1 may review the keywords for keyword candidates that are included relatively frequently. This process is automatically performed by the customer important opinion extraction system 1 in response to a process performed by a user. Also, regarding the keyword correction input and this automatic review, a process of prioritizing one of them based on, for example, a setting of whether or not a change is required for the category keyword can be executed. The process of step 514 is a process in which the customer important opinion extraction system 1 changes the category keyword based on the extraction result of the frequently occurring category keyword based on the setting of whether or not a change is required for the category keyword. Also, the process of step 514 is a process in which the customer important opinion extraction system 1 changes the category keyword based on the input from the input / output unit 10 based on the setting of whether or not a change is required for the category keyword.

[0069] In addition, in "Hypothesis verification result available?" (515), the customer important opinion extraction system 1 determines whether or not there is a hypothesis verification result for the problem hypothesis, and executes a process of correcting the keyword definition 201 in step 517. If there is no hypothesis verification result, the customer important opinion extraction system 1 executes a process of "automatic correction of keyword definition" (517). In the process of step 517, the customer important opinion extraction system 1 directly corrects the keyword in the keyword definition 201 to the keyword accepted in the "acceptance of keyword definition modification" (514). If there is a hypothesis verification result, the customer important opinion extraction system 1 can execute a keyword correction process in step 517 that takes into account the latest hypothesis verification result and changes in customer business by performing a process of "feedback of hypothesis verification result to keyword definition" (516). In this case, as a result of executing the hypothesis verification, the positioning of the call center inquiry data 301 changes depending on whether the hypothesis is correct or not, and as a result, the positioning of the keyword recommended in step 512 may change. Therefore, if there are hypothesis verification results, a proposal for changing the keyword to be corrected can be created in the process of step 516. A proposal for changing the keyword may be, for example, adding the most frequently appearing keyword in that category. In the process of step 517 following step 516, the keyword in keyword definition 201 is corrected to the keyword of this proposed change. The presence or absence of hypothesis verification results may be determined based on whether or not the registration of hypothesis verification results in FIG. 2 (S20) has been completed.

[0070] As for the above-mentioned correction of the keyword definition 201 accepted in step 514 and the keyword correction process in step 517, the customer important opinion extraction system 1 may be configured to give priority to the correction in step 517 as described above. Alternatively, before executing the automatic correction in step 517, a process of inquiring the user as to whether or not to execute the automatic correction may be provided, and if the automatic correction is not to be executed, the customer important opinion extraction system 1 may not execute step 517 and may utilize the correction in step 514 as necessary. Also, if the process proceeds to NO in step 513 and the correction in step 514 is not executed, the customer important opinion extraction system 1 may also provide a process of inquiring the user as to whether or not to execute the automatic correction in step 517 before executing the automatic correction in step 517, and if the automatic correction is not to be executed, may not execute step 517 and may perform a different process.

[0071] As described above, the processing from steps 515 to 517 is processing in which the customer important opinion extraction system 1 changes the contents of the category keywords in accordance with frequently occurring category keywords based on the setting of whether or not the category keywords need to be changed, in accordance with the contents of the hypothesis verification results input from the input / output unit 10, and changes the contents of the category keywords based on the input from the input / output unit 10, in accordance with the setting of whether or not the category keywords need to be changed.

[0072] As described above, according to this embodiment, the keyword correction, which was conventionally performed based on the past knowledge of an expert, can be performed by "automatic correction of keyword definition" (517) that takes into account changes in customer business without involving an expert. Specifically, for example, for a problem hypothesis in which the cause or solution of a problem that was a hypothesis in the hypothesis verification result is matched in the problem suitability test, a process is performed to add a newly appearing word to the keyword definition 201 in the category classification to which the call center inquiry data 301 in which the problem occurred belongs.

[0073] In this way, according to this embodiment, important opinions can be evaluated and extracted by an analyst, i.e., a user, from among a large amount of accumulated customer opinions written in natural language, without an expert having to make a judgment one by one. In other words, according to this embodiment, it is possible to evaluate and extract important opinions that follow changes in customer needs without relying on the analyst's intuition or experience. As a result, it is possible to reflect many customer opinions that match the historical background and customer needs in products and services, and it is possible to quickly introduce products and services to the market and increase their appeal. In addition, it is possible to develop products that accurately reflect customer opinions in the early stages of the product life cycle, where product planning and development are carried out, and to develop new solutions and services that match customer needs and lead to timely market introduction.

[0074] The present invention is not limited to the embodiment described above, and various modifications are possible within the scope of the gist of the present invention. For example, data of web inquiries and the like can be stored in the customer request integrated data lake unit 30 of this embodiment, and important opinions of customers can be extracted using the data. Alternatively, IoT data and the like that can be acquired from devices used by customers can be stored in the customer request integrated data lake unit 30, and analysis can be performed in combination with the call center inquiry data 301. In this way, it is possible to prevent important opinions related to the devices from being overlooked.

[0075] In addition, communication when a customer uses a mobile terminal to submit their opinion via a call center inquiry, and communication between a user of this system and the server on which this system runs, can be performed using a general public line network, whether wired or wireless. For example, the fifth generation mobile communication system, so-called 5G (5th Generation), which enables "multiple simultaneous connections" and "ultra-low latency", can be used for such communication. Furthermore, by utilizing the characteristics of new mobile phone systems beyond 5G, the effects of the present invention can be expected to be improved.

[0076] In the present embodiment, the extraction processing database unit 20 and the customer request integrated data lake unit 30 are provided with functions for storing data separately, but this is merely an example. For example, the extraction processing database unit 20 and the customer request integrated data lake unit 30 may be integrated.

[0077] As described above, according to the embodiment of the present invention, in the process of extracting important customer opinions, it has been necessary to rely on experts with business knowledge to set keywords for personal extraction and extract important opinions, but it is possible to automatically extract important opinions that take these into consideration. Furthermore, it is possible to capture changes in keywords and important opinions that characterize customer needs that change from moment to moment.

[0078] The present invention is not limited to the above-described embodiment, and includes various modified forms and equivalent configurations within the spirit of the appended claims. For example, the above-described embodiment has been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to all of the configurations described. In addition, a part of the configuration of a certain embodiment and a modified form may be replaced with a configuration of another embodiment and a modified form. In addition, the configuration of another embodiment and a modified form may be added to the configuration of a certain embodiment and a modified form. In addition, a part of the configuration of each embodiment and a modified form may be added, deleted, or replaced with another configuration.

[0079] In addition, each of the aforementioned configurations, functions, processing units, processing means, etc. may be realized in hardware, for example by designing some or all of them as an integrated circuit, or may be realized in software by a processor interpreting and executing a program that realizes each function.

[0080] Information such as programs, tables, and files that realize each function can be stored in a storage device such as a memory, a hard disk, or an SSD (Solid State Drive), or in a recording medium such as an IC card, an SD card, or a DVD.

[0081] In addition, the control lines and information lines shown are those considered necessary for the explanation, and do not necessarily show all the control lines and information lines necessary for implementation. In reality, it can be considered that almost all components are connected to each other. [Explanation of symbols]

[0082] 1...customer important opinion extraction system, 4...calculation unit, 8...storage unit, 10...input / output unit, 20...extraction processing database unit, 30...customer request integrated data lake unit, 40...data statistics processing unit, 41...morphological analysis unit, 42...classification unit, 43...aggregation unit, 50...machine learning model unit, 51...keyword review processing unit, 52...important inquiry label review processing unit, 53...similarity discrimination processing unit, 54...important inquiry evaluation processing unit

Claims

1. A customer important opinion extraction system that extracts customer important opinions from customer opinions received from customers, A computing unit that executes computational processing, a storage unit accessible by the computing unit, and an input / output unit, the storage unit stores the customer opinions, data related to the customer opinions, a plurality of classification categories into which the customer opinions are classified, a plurality of category keywords that characterize each of the plurality of classification categories, and a degree of importance assigned to the customer opinions; The customer important opinion extraction system includes: a classification unit configured to classify the customer opinions into classification categories based on the plurality of category keywords; The calculation unit includes a tabulation unit that extracts the classification category having a large increase in the customer opinions over a predetermined period of time; a keyword review processing unit for reviewing the category keywords; the computing unit has a similarity determination processing unit that determines similarity of the customer opinions; the calculation unit includes a label review processing unit for reviewing the importance of an important inquiry; An important inquiry evaluation processing unit that extracts the customer opinions having a high degree of importance, the counting unit extracts classification categories with a large increase in the number of customer opinions from a difference in the number of customer opinions belonging to the classification categories between a first timing corresponding to a start time of the specified period and a second timing corresponding to an end time of the specified period; The keyword review processing unit: extracting category keywords that frequently appear in the customer opinions that belong to the classification category extracted by the tabulation unit; changing the category keywords based on the extraction result of the frequently occurring category keywords; based on the hypothesis verification result input from the input / output unit, the content of the category keyword is changed based on the frequently occurring category keyword; the similarity determination processing unit determines similarity between the customer opinions for the classification categories extracted by the counting unit, using the importance of the customer opinions; The important inquiry label review processing unit: extracting similar inquiries that frequently appear in the customer opinions belonging to the classification category extracted by the tabulation unit; Changing the importance of the customer opinion based on the number of similar inquiries extracted; changing the importance of the customer opinion based on the hypothesis verification result input from the input / output unit; The important inquiry evaluation processing unit extracts and outputs the customer opinions having high importance. A customer important opinion extraction system.

2. 2. The customer important opinion extraction system according to claim 1, The data related to the customer opinion includes customer attributes and product attributes corresponding to the customer opinion; the classification unit extracts the number of the customer opinions for each classification category based on the customer opinions, attributes of the customers, and attributes of the products; A customer important opinion extraction system.

3. 2. The customer important opinion extraction system according to claim 1, the tabulation unit extracts, from among customer opinions belonging to classification categories in which the number of classified customer opinions is greater than a predetermined number, the classification category in which the number of customer opinions at the second timing has increased by more than a predetermined number compared to the number of customer opinions at the first timing; A customer important opinion extraction system.

4. 2. The customer important opinion extraction system according to claim 1, the similarity determination processing unit extracts customer opinions similar to the customer opinions with high importance using a machine learning model that has been trained using customer opinions belonging to a classification category with a large increase in the number of customer opinions and their importance as training data; assigning the importance of the customer opinion having a high importance to the extracted customer opinion; the important inquiry evaluation processing unit outputs customer opinions having a predetermined level of importance with respect to the customer opinions; A customer important opinion extraction system.

5. A customer important opinion extraction system as described in claim 1, The important inquiry evaluation processing unit includes: Extracting frequent and similar customer opinions that belong to the classification category extracted as having a large increase; modifying the importance based on the frequent and similar customer opinions; a customer important opinion extraction system that changes the importance level based on the content of a hypothesis verification result input from the input / output unit;

6. The customer important opinion extraction system according to claim 1, The classification unit classifies the classification categories based on a result of performing a morphological analysis on the customer opinions. A customer important opinion extraction system.

7. A customer important opinion extraction method executed by a customer important opinion extraction system that extracts customer important opinions from customer opinions submitted by customers, comprising: The customer important opinion extraction system includes a calculation unit that executes calculation processing, a storage unit that is accessible by the calculation unit, and an input / output unit; the storage unit stores the customer opinions, data related to the customer opinions, a plurality of classification categories into which the customer opinions are classified, a plurality of category keywords that characterize each of the plurality of classification categories, and a degree of importance assigned to the customer opinions; The customer important opinion extraction method includes: a classification step in which the calculation unit classifies the customer opinions into classification categories based on the plurality of category keywords; a counting step in which the calculation unit extracts classification categories with a large increase in the number of customer opinions from a difference in the number of customer opinions belonging to the classification categories between a first timing corresponding to a start time of a predetermined period and a second timing corresponding to an end time of the predetermined period; a keyword review step in which the calculation unit extracts category keywords that appear frequently in the customer opinions belonging to the classification category extracted in the compilation step, changes the category keywords based on the extraction result of the frequently appearing category keywords, and changes the contents of the category keywords based on the frequently appearing category keywords based on a hypothesis verification result input from the input / output unit; a similarity determination step in which the calculation unit determines similarity of the customer opinions for the classification categories extracted in the compilation step by using importance of the customer opinions; an important inquiry label review step in which the calculation unit extracts similar inquiries that frequently appear in the customer opinions belonging to the classification category extracted in the compilation step, changes the importance of the customer opinions based on the number of the extracted similar inquiries, and changes the importance of the customer opinions based on a hypothesis verification result input from the input / output unit; and an important inquiry evaluation step in which the calculation unit extracts and outputs the customer opinions having a high degree of importance.

Images