JP7404713B2 - Correspondence presentation device and correspondence presentation method - Google Patents

Description

The present invention relates to a correspondence presentation device and a correspondence presentation method.

Conventionally, support systems for call center operators are known (for example, see Patent Documents 1 and 2).
Patent Document 1: Japanese Patent Application Publication No. 2015-233230 Patent Document 2: Japanese Patent Application Publication No. 2010-224727

However, with conventional support systems, it is necessary to search for and carefully read the contents of countermeasure examples, making it difficult to quickly access information.

In the first aspect of the present invention, there is provided a storage unit that stores a master DB of past question and answer records, a text acquisition unit that acquires text data obtained by voice recognition of the current question and answer record, and a text acquisition unit that acquires the main points of the text data in advance. A hierarchical structure data generation unit that classifies each predetermined category, aggregates the main points, and generates hierarchical structure data for the master DB, and categories into which the hierarchical structure data generation unit classifies the hierarchical structure data corresponding to text data. The display section provides a correspondence presentation device that highlights hierarchical structure data having a high degree of similarity to text data.

The display unit displays the contents of the first category in order of similarity to the text data, and displays the contents of the second category, which is a lower hierarchy than the first category, based on the number of items stored in the master DB. They may be displayed in order of increasing number.

The correspondence presentation device may further include a hierarchy determination unit that determines the number of hierarchies of the hierarchical structure data. The hierarchical structure data generation section may generate hierarchical structure data according to the number of hierarchies determined by the hierarchy determination section.

The hierarchical structure data generation unit may extract the main points of the text data for each category of situation, cause, and countermeasure from the text data of the conversation between the customer and the operator.

The text acquisition unit may modify the text data. The hierarchical structure data generation unit may generate hierarchical structure data based on the corrected text data. The display unit may display the hierarchical structure data corresponding to the corrected text data for each category classified by the hierarchical structure data generation unit.

In the second aspect of the present invention, the steps include: storing a master DB of past question and answer records; obtaining text data obtained by voice recognition of the current question and answer record; A step of classifying each category, aggregating the main points, and generating hierarchical structure data of the master DB, a step of displaying the hierarchical structure data corresponding to the text data for each classified category, and a step of displaying the degree of similarity to the text data. Highlighting hierarchically structured data with a high level of hierarchy.

Note that the above summary of the invention does not list all the features of the invention. Furthermore, subcombinations of these features may also constitute inventions.

An overview of the configuration of the compatible system 200 is shown. An overview of the configuration of the correspondence presentation device 100 is shown. An outline of a flowchart of operations performed by the correspondence presentation device 100 is shown. An outline of a flowchart of customer response using the response presentation device 100 is shown. An example of text data TD voice-recognized by the text acquisition unit 10 is shown. An example of main points classified into categories by the hierarchical structure data generation unit 30 is shown. An example of a summary sentence generated by the hierarchical structure data generation unit 30 is shown. An example of an aggregate sentence generated by the hierarchical structure data generation unit 30 is shown. An example of aggregation of information within a category by the hierarchical structure data generation unit 30 is shown. An example of a master DB stored in the storage unit 20 is shown. An example of text data TD voice-recognized by the text acquisition unit 10 is shown. A display example of aggregated sentences extracted for each category is shown. An example of a flowchart of automatic status search procedure is shown. An example of displaying hierarchical structure data HD on the display unit 40 is shown. An example of a flowchart of automatic cause search procedure is shown. An example of displaying hierarchical structure data HD on the display unit 40 is shown. Another example of a flowchart of a customer response using the response presentation device 100 is shown. An example of text data TD voice-recognized by the text acquisition unit 10 is shown. A display example of aggregated sentences extracted for each category is shown. An example of displaying hierarchical structure data HD on the display unit 40 is shown. An example of a method for modifying text data TD will be shown. An example of displaying hierarchical structure data HD on the display unit 40 is shown.

Hereinafter, the present invention will be described through embodiments of the invention, but the following embodiments do not limit the invention according to the claims. Furthermore, not all combinations of features described in the embodiments are essential to the solution of the invention.

FIG. 1 shows an overview of the configuration of the corresponding system 200. The response system 200 is a system for assisting an operator in responding when a customer 210 makes an inquiry to the call center. In this example, a case will be described in which the support system 200 has a primary reception 230 and a secondary reception 240, but it may have a tertiary reception or higher reception.

A customer 210 contacts an operator at a primary reception desk 230 via a public switched telephone network 220 such as a PSTN (Public Switched Telephone Network) and a private branch exchange 225 such as a PBX (Private Branch eXchange).

The primary reception 230 has a telephone 232 and a reception terminal 234 used by an operator. The secondary reception 240 has a telephone 242 used by an operator and a reception terminal 244. The reception terminal 234 and the reception terminal 244 are connected to a server machine 260 via a LAN (Local Area Network) 250.

The response presentation device 100, which will be described later, assists the primary receptionist 230 and the secondary receptionist 240 in responding to the customer 210. The correspondence presentation device 100 may be implemented in the server machine 260. Server machine 260 is, for example, a personal computer. The results executed by the server machine 260 are displayed on the reception terminal 234 or the reception terminal 244 via a network such as the LAN 250.

FIG. 2A shows an overview of the configuration of the correspondence presentation device 100. The correspondence presentation device 100 includes a text acquisition section 10, a storage section 20, a hierarchical structure data generation section 30, a display section 40, and a hierarchy determination section 50.

The text acquisition unit 10 acquires text data TD obtained by voice recognition of the current question and answer record. The text acquisition unit 10 may convert the input voice into text data TD and acquire it, or may acquire the input text data TD. The text acquisition unit 10 transmits the acquired text data TD of the current question and answer record to the hierarchical structure data generation unit 30. The question and answer record is a record of conversations such as questions and answers between the customer 210 and the operator.

The current question-and-answer record refers to a real-time question-and-answer record in which an operator is interacting with a customer. The current question and answer record may be a question and answer record of the primary receptionist 230 or a question and answer record of the secondary receptionist 240.

The storage unit 20 stores a master DB of past question and answer records. The past question-and-answer record is a question-and-answer record when an operator previously dealt with a customer, which is stored as hierarchical structure data HD, which will be described later. The past question and answer record may be a question and answer record of the primary receptionist 230 or a question and answer record of the secondary receptionist 240. The storage unit 20 transmits the stored master DB to the hierarchical structure data generation unit 30.

The hierarchical structure data generation unit 30 generates hierarchical structure data HD of the master DB. The hierarchical structure data generation section 30 transmits the generated hierarchical structure data HD to the display section 40. Furthermore, the hierarchical structure data generation section 30 may transmit the text data TD from the text acquisition section 10 as is to the display section 40 in real time. Note that the hierarchical structure data generation unit 30 classifies the main points of the text data into predetermined categories, generates a summary sentence for each classified category, and generates an aggregate sentence that aggregates the summary sentences. Details will be described later.

The hierarchical structure data HD is a hierarchical display of the contents of the master DB for each category. The hierarchical structure data HD includes an aggregate sentence generated by the hierarchical structure data generation unit 30. The hierarchical structure data generation unit 30 classifies the main points into each predetermined category and generates the hierarchical structure data HD. The hierarchical structure data HD of this example has three categories. For example, the categories include "situation", "cause", and "remedy" of the case handled by the operator. Note that the category classification in this example is just an example, and the number and contents of categories are not limited to this. “Situation” may be a “question” category, and “handling” may be a “answer” category.

The display unit 40 displays hierarchical structure data HD corresponding to the text data TD. The display section 40 displays the hierarchical structure data HD for each category classified by the hierarchical structure data generation section 30. For example, the display unit 40 displays the hierarchical structure data HD in a tree shape in order to easily understand it. The display unit 40 may highlight the hierarchical structure data HD that is similar to the text data TD. For example, the display unit 40 displays the information in a highlighted manner by adding color or animation such as blinking.

Further, the display unit 40 may display the text data TD acquired by the text acquisition unit 10 as is. That is, the display unit 40 may display the current question and answer record between the customer 210 and the operator in real time. The display unit 40 may display both the text data TD and the hierarchical structure data HD of the current question and answer record. Note that the display unit 40 may have a display for displaying information, or may display information on an external terminal such as the reception terminal 234 and the reception terminal 244.

The hierarchy determination unit 50 determines the number of hierarchies in the hierarchical structure data HD. The hierarchical structure data generation section 30 generates the hierarchical structure data HD according to the number of layers determined by the hierarchy determination section 50. Note that the hierarchy determining unit 50 may determine the number of hierarchies of the hierarchical structure data HD in accordance with input from the outside. For example, the hierarchy determination unit 50 determines the number of hierarchies in the hierarchical structure data HD according to an operator's instruction.

The correspondence presentation device 100 of this example displays hierarchical structure data HD corresponding to the text data TD of the current question and answer record. Therefore, the operator can quickly understand the cause of the response and how to deal with it from the hierarchical structure data HD. Thereby, the correspondence presentation device 100 can assist the operator in responding. The operator can easily find the information (for example, what to do) that he or she needs. Further, the correspondence presentation device 100 can cause the primary receptionist 230 to correspond to an advanced case that the secondary receptionist 240 has responded to. Therefore, even an inexperienced primary receptionist 230 operator can provide answers equivalent to those of an expert or veteran operator to some extent. Therefore, the quality and speed of reception at the primary reception 230 are improved.

Note that the correspondence presentation device 100 may display the modified hierarchical structure data HD by modifying the text data TD of the current question and answer record. In one example, the text acquisition unit 10 modifies the text data TD in response to an instruction from an operator. The hierarchical structure data generation unit 30 generates hierarchical structure data HD based on the modified text data TD. The display unit 40 displays the hierarchical structure data HD corresponding to the corrected text data TD for each category classified by the hierarchical structure data generation unit 30. Thereby, the correspondence presentation device 100 can further improve the quality of the operator's answers.

FIG. 2B shows an outline of a flowchart of operations performed by the correspondence presentation device 100. The correspondence presentation device 100 assists the operator in responding by executing steps S10 to S16.

In step S10, the storage unit 20 stores a master DB of past question and answer records. In step S12, the text acquisition unit 10 acquires text data TD obtained by voice recognition of the current question and answer record. In step S14, the hierarchical structure data generation unit 30 generates the hierarchical structure data HD of the master DB. In step S16, the display unit 40 displays the hierarchical structure data HD corresponding to the text data TD. Thereby, the response presentation device 100 can present the optimal response method to the operator as the hierarchical structure data HD.

FIG. 3 shows an outline of a flowchart of customer response using the response presentation device 100. By executing steps S100 to S118, it is possible to respond to the customer's 210 request. Note that the flowchart of this example is an example of a flowchart of customer response, and the method of customer response is not limited to this example.

In step S100, the text acquisition unit 10 performs voice recognition on records of cases in which the operator was able to respond correctly, and creates text data TD. The text data TD acquired by the text acquisition unit 10 may be the correspondence of the operator of the primary reception 230 or the correspondence of the operator of the secondary reception 240. The text data TD acquired by the text acquisition unit 10 may be the response of operators of both the primary reception 230 and the secondary reception 240.

In step S102, the hierarchical structure data generation unit 30 extracts key points for each category from the text data TD created by the text acquisition unit 10. The hierarchical structure data generation unit 30 can generate hierarchical structure data HD having a hierarchy for each category by extracting key points for each category. For example, the hierarchical structure data generation unit 30 extracts key points for each category of situation, cause, and countermeasure. As a result, parts corresponding to the situation, cause, and countermeasure are extracted from the conversation in the question and answer record.

In step S104, the hierarchical structure data generation unit 30 summarizes and aggregates similar content within the category from the extracted information. For example, the hierarchical structure data generation unit 30 generates a summary sentence that summarizes similar content within a category, and generates an aggregate sentence that aggregates the summary sentences. In this case, the hierarchical structure data generation unit 30 may generate the number of items for each content as the aggregation degree. In step S106, the storage unit 20 registers the hierarchical structure data HD as a master DB in the order of situation, cause, and countermeasure.

In step S108, the text acquisition unit 10 performs voice recognition on the record of the interaction between the customer 210 and the operator in real time to create text data TD. For example, the text acquisition unit 10 creates text data TD of the reception record of the primary receptionist 230.

In step S110, the hierarchical structure data generation unit 30 extracts key points for each category of situation and cause from the created text data TD. In step S112, the hierarchical structure data generation unit 30 automatically searches for content similar to the extracted key points from the situation categories registered in the master DB in step S106. The display unit 40 presents similar content to the operator terminal of the primary receptionist 230. In this specification, similar content is not limited to the most similar content, but may refer to a plurality of content in descending order of similarity. Next, if there is category content of "cause" among the extracted key points, the process advances to step S114, and if there is no category content of "cause" among the extracted key points, the process advances to step S116.

In step S114, if there is category content of "cause" among the extracted key points, the hierarchical structure data generation unit 30 automatically searches for similar content from the cause categories registered in the master DB in step S106. The display unit 40 displays similar content on the terminal of the operator of the primary reception 230.

In step S116, if there is no "cause" category content among the extracted key points, the operator of the primary reception 230 determines the cause by talking with the customer 210 while referring to the content of the multiple causes presented on the terminal. Identify.

In step S118, the operator of the primary reception desk 230 checks the contents of the countermeasures corresponding to the situation and cause presented on the terminal, and replies to the customer with the optimal countermeasure. In this way, the response presentation device 100 can assist the operator in responding by displaying the hierarchical structure data HD corresponding to the text data TD of the current question and answer record.

FIG. 4 shows an example of text data TD voice-recognized by the text acquisition unit 10. The text data TD in this example is an example of audio that is converted into text data by the text acquisition unit 10 in step S100 of FIG.

The text acquisition unit 10 converts the current question and answer record into text data in chronological order. The text acquisition unit 10 distinguishes between operators and customers 210 and converts them into text data. The text acquisition unit 10 may perform voice recognition on records of cases in which the operator was able to respond correctly. The text acquisition unit 10 of this example converts the contents of the conversation between the customer 210 and the operator at the primary reception desk 230 into text data in chronological order. However, the object and method of converting the current question and answer record into text data is not limited to this example.

FIG. 5A shows an example of main points classified into categories by the hierarchical structure data generation unit 30. The gist extraction example of this example is an example of the processing of the hierarchical structure data generation unit 30 in step S102 of FIG.

The hierarchical structure data generation unit 30 extracts key points for each category from the text data TD acquired by the text acquisition unit 10. For example, there are three categories: situation, cause, and countermeasure. The hierarchical structure data generation unit 30 in this example extracts the text "The inverter has stopped" as the "situation", the text "I touched the wiring" as the "cause", and the text "I touched the wiring" as the "measure". Extract the text "Please arrange the wiring and replace it."

Further, the hierarchical structure data generation unit 30 may extract the main points and classify the categories using any method. For example, the hierarchical structure data generation unit 30 extracts key points and classifies categories using a rule-based method that extracts categories when registered keywords appear. Furthermore, the hierarchical structure data generation unit 30 may extract categories using a model trained on classified correct answer examples.

FIG. 5B shows an example of a summary sentence generated by the hierarchical structure data generation unit 30. The gist extraction example of this example is an example of the processing of the hierarchical structure data generation unit 30 in step S104 of FIG.

The hierarchical structure data generation unit 30 generates a summary sentence summarizing the main points extracted for each category. For example, regarding "Status", the content "The inverter has stopped" is summarized as "The inverter has stopped". Regarding the ``cause'', the contents of ``I touched the wiring'' can be summarized as ``I touched the wiring.'' Regarding "Corrective Action", the contents of "Please arrange the wiring and replace it" can be summarized as "Arrange the wiring and replace it".

FIG. 5C shows an example of an aggregate sentence generated by the hierarchical structure data generation unit 30. The gist extraction example of this example is an example of the processing of the hierarchical structure data generation unit 30 in step S104 of FIG.

The hierarchical structure data generation unit 30 generates a summary sentence summarizing the main points extracted for each category. For example, regarding the "situation", the contents of "the inverter has stopped" are aggregated into "the inverter has stopped". Regarding "Cause", if the summary sentence "Touched the wiring" is used as a summary sentence, there is no need to change it. Regarding "Countermeasures," if the summary sentence "arrange and replace wiring" is used as a summary sentence, there is no need to change it.

FIG. 6 shows an example of aggregation of information within a category by the hierarchical structure data generation unit 30. The aggregation example of this example is an example of the processing of the hierarchical structure data generation unit 30 in step S104 of FIG. In this example, a case will be described in which six aggregate sentences, Aggregation Examples 1 to 6, are generated from six summary sentences, Aggregation Examples 1 to 6.

The hierarchical structure data generation unit 30 aggregates similar contents within a category from a summary sentence summarizing the extracted information. For example, the hierarchical structure data generation unit 30 aggregates sentences using a rule-based method that aggregates sentences when registered keywords appear. Furthermore, the hierarchical structure data generation unit 30 may aggregate the sentences by a method of aggregating the aggregated correct answer examples using a trained model.

For example, if the "situation" includes "inverter stopped" and "inverter stopped" as similar contents, these are aggregated into "inverter stopped". Further, in the case where "situation" includes "wires are burnt out" and "cords are burnt out" as similar content, these are aggregated into "wires are burnt out". This allows the main points of similar "situations" to be summarized into a common summary sentence.

Similarly, when "cause" includes "touched wiring" and "touched wiring" as contents to be aggregated, these are aggregated into "touched wiring". In addition, if "cause" includes "long-term use" and "long-term use" as similar content, these are aggregated into "long-term use". Thereby, the main points of similar "causes" can be summarized into a common summary sentence.

Similarly, if "replace the inverter" is included as a similar content in "Countermeasures", it will be summarized as "replace the inverter". In this way, the aggregate sentences to be aggregated may be prepared in advance. By aggregating content, similar content can be treated as the same content. Note that although in this example, the summary sentence and the aggregate sentence are underlined to make the difference clear, they do not need to be underlined.

FIG. 7 shows an example of a master DB stored in the storage unit 20. The master DB in this example is an example of a database stored in the storage unit 20 in step S106 of FIG.

The storage unit 20 registers the situation, cause, and countermeasure in order as a master DB in a hierarchical structure. No. 1~No. 5 correspond to the hierarchical structure data HD shown in FIG. The storage unit 20 may store the items in descending order of number. The storage unit 20 may also store the number of cases for each category and the number of cases for each case. The master DB stores the aggregated sentences aggregated in step S104 of FIG.

FIG. 8 shows an example of text data TD voice-recognized by the text acquisition unit 10. The text data TD of this example is an example of the text that the text acquisition unit 10 acquires in step S108 of FIG.

The text acquisition unit 10 acquires text data TD obtained by voice recognition of the reception records of the customer 210 and the primary receptionist 230 in real time. The text acquisition unit 10 may acquire the text data TD of the question and answer record of the secondary receptionist 240 in real time.

The display section 40 displays a display area 42. Although the display unit 40 of this example displays the display area 42 on the left half of the display screen, the present invention is not limited to this. The display unit 40 displays a display area 42 to the operator of the primary reception desk 230.

The display area 42 displays the conversation between the customer 210 and the primary receptionist 230 in a balloon format. However, the display format of the conversation is not limited to this example. Since the responses of the customer 210 and the primary receptionist 230 are voice recognized and converted into text data in chronological order, the text may be saved as data or may be displayed on a screen. That is, the conversation between the customer 210 and the primary receptionist 230 does not need to be displayed on the screen. In this example, text data TD corresponding to the situation and cause has been acquired, but text data TD corresponding to the countermeasure has not been acquired.

FIG. 9 shows a display example of aggregated sentences extracted for each category. The gist extraction example of this example is an example of a summary sentence displayed by the display unit 40 in step S110 of FIG. 3. The display unit 40 displays information in a display area 44 in addition to the display area 42 .

The display area 44 displays a summary of the conversation between the customer 210 and the primary receptionist 230. Display area 44 is displayed next to display area 42 . In this example, text data TD corresponding to the situation and cause has been acquired, but text data TD corresponding to the countermeasure has not been acquired. Therefore, the countermeasure column for the aggregate sentence is blank.

The aggregate sentence of the conversation between the customer 210 and the primary receptionist 230 may be created by the hierarchical structure data generation unit 30. For example, the hierarchical structure data generation unit 30 creates an aggregate sentence using a rule-based method that extracts registered keywords when they appear.

FIG. 10 shows an example of a flowchart of the automatic situation search procedure. The flowchart of this example shows an example of a process in which the hierarchical structure data generation unit 30 automatically searches for similar content in step S112 of FIG. The correspondence presentation device 100 of this example automatically searches for content similar to the text data TD of “situation” from the category of situations registered in the master DB. Note that the method of automatically searching by calculating the similarity of sentences is not limited to this example. Further, this flowchart may be continued until the conversation ends.

In step S200, the hierarchical structure data generation unit 30 morphologically analyzes the text registered in the master DB. The hierarchical structure data generation unit 30 may morphologically analyze all sentences registered in the master DB. In step S202, the hierarchical structure data generation unit 30 converts the morphologically analyzed word into vector data.

In step S204, the hierarchical structure data generation unit 30 calculates vector data of the text in the master DB. The hierarchical structure data generation unit 30 calculates vector data for each sentence in the master DB. For example, the hierarchical structure data generation unit 30 calculates N pieces of vector data when N pieces of text exist in the same category. The hierarchical structure data generation unit 30 in this example converts all N sentences into vector data, but may convert some sentences into vector data. In steps S200 to S204, vector data is calculated for past question and answer records.

In step S206, the hierarchical structure data generation unit 30 categorizes the text data TD that has been voice-recognized in real time into situations, causes, and countermeasures. If there is text data TD corresponding to the "situation", the hierarchical structure data generation unit 30 proceeds to step S208. In step S208, the hierarchical structure data generation unit 30 generates a summary sentence of the text data TD that has been voice-recognized in real time. For example, the hierarchical structure data generating unit 30 generates a summary sentence of the text data TD corresponding to "situation". In step S210, an aggregated sentence is generated by aggregating the summary sentences. In step S212, the hierarchical structure data generation unit 30 converts the words of the aggregate sentence into vector data.

In step S214, the hierarchical structure data generation unit 30 calculates vector data of sentences categorized into situations. In step S216, the hierarchical structure data generation unit 30 calculates the degree of similarity between the vector data for each sentence calculated in step S204 and the vector data calculated in step S214. In other words, the hierarchical structure data generation unit 30 calculates the degree of similarity between the vector data of the master DB corresponding to the past question and answer record and the vector data of the text data TD corresponding to the current question and answer record. In step S218, sentences of situations registered in the master DB with a high degree of similarity are presented to the operator of the primary reception 230. Not only sentences with a high degree of similarity, but also a plurality of sentences registered in the master DB may be presented to the operator of the primary reception 230 in order of degree of similarity. For example, the hierarchical structure data generation unit 30 uses Doc2Vec to calculate the degree of similarity.

FIG. 11 shows an example of displaying the hierarchical structure data HD on the display unit 40. As shown in FIG. The figure is an example of a screen displayed by the display unit 40 in step S112 of FIG. The display unit 40 in this example presents the search results to the operator of the primary reception 230.

The display unit 40 displays the contents of the hierarchical structure data HD with a high degree of similarity calculated by the hierarchical structure data generation unit 30. The display unit 40 of this example displays contents similar to the extracted key points (for example, the category contents of "situation") in a display area 44 in an emphasized manner. The hierarchical structure data generation unit 30 has acquired "the inverter has stopped" as the aggregate sentence of the status text data TD. Therefore, the display unit 40 highlights "inverter stopped" in the "situation" category of the master DB. Thereby, the operator of the primary reception desk 230 can immediately know the cause and countermeasure related to "inverter stopped" from the hierarchical structure data HD displayed in the display area 44.

Here, the display unit 40 may change the method for determining the priority of the contents displayed in the display area 44 for each category. In one example, the display unit 40 displays the contents of the first category in descending order of similarity. Then, the display unit 40 may display the contents of the second category, which is a lower level hierarchy than the first category, in order of the number of items. For example, if the first category is "situation" and the second category is "cause," the display unit 40 displays the content of the "cause" category, which is a lower hierarchy than "situation." Display them in order. That is, if the number of "touched wiring" registered in the master DB is greater than "used for a long time", "touched wiring" is displayed with priority over "used for a long time". Displaying with priority means displaying in a position in the display area 44 that is easier for the operator to recognize.

Here, the operation of the display section 40 will be explained along the flowchart of the automatic situation search procedure in FIG. Speech recognition is started after vector data is stored for past question and answer records in steps S200 to S204. The text data TD voice-recognized in real time is displayed in the display area 42, and the text data TD is classified into categories (step S206). Then, in the display area 44, an aggregate sentence is displayed if there is text data TD corresponding to the "situation" (step S208). Further, in the display area 44, the degree of similarity of the text data TD recognized in real time is calculated, and aggregated sentences of the sentences of the situations registered in the master DB are displayed in order of the degree of similarity (step S210 ~Step S218).

FIG. 12 shows an example of a flowchart of the automatic cause search procedure. The flowchart of this example shows an example of a process in which the hierarchical structure data generation unit 30 automatically searches for similar content in step S114 or step S116 of FIG. The correspondence presentation device 100 of this example automatically searches for content similar to the text data TD of “cause” from the categories of causes registered in the master DB. This example differs from the case of FIG. 10 in that vector data of sentences classified into the "cause" category is calculated. Note that the method of automatically searching by calculating the similarity of sentences is not limited to this example. Further, this flowchart may be continued until the conversation ends.

In step S300, the hierarchical structure data generation unit 30 morphologically analyzes the text registered in the master DB. The hierarchical structure data generation unit 30 may morphologically analyze all sentences registered in the master DB. In step S302, the hierarchical structure data generation unit 30 converts the morphologically analyzed word into vector data.

In step S304, the hierarchical structure data generation unit 30 calculates vector data of the text in the master DB. The hierarchical structure data generation unit 30 calculates vector data for each sentence in the master DB. For example, the hierarchical structure data generation unit 30 calculates N pieces of vector data when N pieces of text exist in the same category. The hierarchical structure data generation unit 30 in this example converts all N sentences into vector data, but may convert some sentences into vector data. In steps S300 to S304, vector data is calculated for past question and answer records.

In step S306, the hierarchical structure data generation unit 30 categorizes the text data TD that has been voice-recognized in real time into situations, causes, and countermeasures. If there is text data TD corresponding to the "cause", the hierarchical structure data generation unit 30 proceeds to step S308. In step S308, the hierarchical structure data generation unit 30 generates a summary sentence of the text data TD that has been voice-recognized in real time. For example, the hierarchical structure data generation unit 30 generates a summary sentence of the text data TD corresponding to "cause". In step S310, an aggregated sentence is generated by aggregating the summary sentences. In step S312, the hierarchical structure data generation unit 30 converts the words of the aggregate sentence into vector data.

In step S314, the hierarchical structure data generation unit 30 calculates vector data of sentences categorized into causes. In step S316, the hierarchical structure data generation unit 30 calculates the degree of similarity between the vector data for each sentence calculated in step S304 and the vector data calculated in step S314. In other words, the hierarchical structure data generation unit 30 calculates the degree of similarity between the vector data of the master DB corresponding to the past question and answer record and the vector data of the text data TD corresponding to the current question and answer record. In step S318, the cause text registered in the master DB with a high degree of similarity is presented to the operator of the primary reception 230. Not only sentences with a high degree of similarity, but also a plurality of aggregated sentences registered in the master DB may be presented to the operator of the primary reception 230 in order of degree of similarity. For example, the hierarchical structure data generation unit 30 uses Doc2Vec to calculate the degree of similarity.

FIG. 13 shows an example of displaying the hierarchical structure data HD on the display unit 40. As shown in FIG. The figure is an example of a screen displayed by the display unit 40 in step S114 of FIG. 3. The display unit 40 in this example presents the search results to the operator of the primary reception 230.

The display unit 40 highlights and displays in the display area 44 contents registered in the master DB that are similar to the extracted key points (for example, the category contents of "Cause"). The text acquisition unit 10 acquires "I touched the wiring" as the text data TD of the cause. Therefore, the display unit 40 highlights "Touched wiring" which is similar to "Touched wiring" from the "Cause" category of the master DB. As a result, the operator of the primary reception desk 230 can immediately know from the hierarchical structure data HD displayed in the display area 44 what to do in relation to "touching a wire."

Here, the display unit 40 may change the method for determining the priority of the contents displayed in the display area 44 for each category. In one example, the display unit 40 displays the contents of the first category in descending order of similarity. Then, the display unit 40 may display the contents of the second category, which is a lower level hierarchy than the first category, in order of the number of items. For example, if the first category is "Cause" and the second category is "Countermeasure," the display unit 40 displays the content of the "Countermeasure" category, which is a lower hierarchy than "Cause". Display them in order. In other words, if the number of "Check if wires are missing" registered in the master DB is greater than "Arrange wiring and replace", "Check if wires are missing" is changed to "Arrange wires and replace". Display with priority over "Exchange".

Here, the operation of the display unit 40 will be explained along the flowchart of the automatic cause search procedure shown in FIG. Speech recognition is started after vector data is stored for past question and answer records in steps S300 to S304. The text data TD that has been voice-recognized in real time is displayed in the display area 42, and the text data TD is classified into categories (step S306). Then, in the display area 44, if there is text data TD corresponding to the "cause", an aggregate sentence is displayed (step S308). Further, in the display area 44, the degree of similarity of the text data TD recognized in real time is calculated, and aggregated sentences of the sentences of the situations registered in the master DB are displayed in order of the degree of similarity (step S310 ~Step S318).

Note that when the hierarchy determination unit 50 determines that the resolution of the hierarchy is coarse, it may increase the number of categories to make the resolution of the hierarchy finer. A case where the resolution of the hierarchy is coarse is a case where the number of categories is small and the number of hierarchies is small. By fine-tuning the resolution of the hierarchy, more detailed handling can be achieved.

On the other hand, if the hierarchy determination unit 50 determines that the resolution of the hierarchy is fine, it may increase the number of categories to make the resolution of the hierarchy coarser. A case where the resolution of the hierarchy is fine is a case where there are many categories and many hierarchies. By coarsening the resolution of the hierarchy, it becomes easier to classify categories. By appropriately setting the resolution of the hierarchy, the correspondence presentation device 100 can display the hierarchical structure data HD that is easier for the operator to handle on the display unit 40.

Furthermore, the hierarchy determination unit 50 may change the resolution of the hierarchy for each category content. In one example, the hierarchy determining unit 50 increases the resolution of classifications corresponding to categories with a large number of cases, and lowers the resolution of classifications corresponding to categories with a small number of cases. For example, if the situation is one that the company can handle, the resolution of the hierarchy will be increased and the operator will examine the ``cause'' and ``remedy'' in detail. On the other hand, in the case of a situation that cannot be handled by the company, there is no need to understand the "cause" and "remedy" in detail, so the resolution can be lowered. The resolution of the hierarchy may be selected by the operator. However, the hierarchical structure data generation unit 30 generates the hierarchical structure data HD in accordance with the resolution of the hierarchy in advance.

The response presentation device 100 allows the operator of the primary reception 230 to check the contents of the response corresponding to the situation and cause presented on the reception terminal, and allows the customer to respond with the optimal response method. If the cause has not been extracted, the correspondence presentation device 100 can confirm the content of the cause presented on the terminal and proceed with the conversation with the customer 210 while making an analogy.

Further, the correspondence presentation device 100 can provide answers equivalent to those of an expert or veteran operator to some extent even if the operator at the primary reception desk 230 is inexperienced. Further, the correspondence presentation device 100 can display information that is easy for the operator to respond to immediately, compared to a case where only information with a one-to-one correspondence is displayed, such as an FAQ. Therefore, the quality and speed of reception at the primary reception 230 are improved.

FIG. 14 shows another example of a flowchart of customer response using the response presentation device 100. By executing steps S400 to S414, it is possible to appropriately respond to the customer's 210 request even if the voice is misrecognized or misconverted. Note that the flowchart of this example is an example of a flowchart of customer response, and the method of customer response is not limited to this example.

Steps S400 to S408 correspond to steps S106 to S114 in FIG. 3. In this example, points that are different from the operation flowchart of FIG. 3 will be particularly explained.

In step S410, it is determined whether there is any misrecognition or misconversion in the text data TD. In this example, the operator determines whether there is erroneous recognition or erroneous conversion. Text data TD that is likely to be misrecognized or misconverted may be highlighted to notify the operator. If there is misrecognition or misconversion in the text data TD, the process advances to step S412.

In step S412, the erroneous recognition or erroneous conversion of the text data TD is corrected at the terminal of the person in charge of the primary reception 230. For example, erroneously converted homophones are converted into correct wording. The text data TD may be modified all at once, or individually and immediately. Then, the process returns to step S404, and key points are extracted for each category based on the corrected text data TD.

On the other hand, if there is no misrecognition or misconversion in the text data TD, the process advances to step S414, and the operator at the primary reception desk 230 checks the situation presented on the terminal and the contents of the action corresponding to the cause, and takes the optimal action. How to answer customers. In this manner, the correspondence presentation device 100 of this example can correct misrecognition or misconversion of the text data TD, and therefore can provide a more optimal solution to the customer.

FIG. 15A shows an example of text data TD voice-recognized by the text acquisition unit 10. The text data TD of this example is an example of the text that the text acquisition unit 10 acquires in step S402 of FIG. 14.

The text acquisition unit 10 acquires text data TD obtained by voice recognition of the reception records of the customer 210 and the primary receptionist 230 in real time. However, the text acquisition unit 10 erroneously converts what should have been translated into "I touched the wiring" into "I touched the losing battle."

The display area 42 displays the conversation between the customer 210 and the primary receptionist 230 in a balloon format. Furthermore, the display area 42 has an input field for correcting misrecognition or misconversion. In the display area 42 of this example, a correction button 46 and a batch correction button 48 are provided. The correction button 46 and batch correction button 48 will be described later.

FIG. 15B shows a display example of aggregated sentences extracted for each category. The gist extraction example of this example is an example of an aggregated sentence displayed by the display unit 40 in step S404 of FIG. 14. The display unit 40 displays information in a display area 44 in addition to the display area 42 .

The display area 44 displays a summary of the conversation between the customer 210 and the primary receptionist 230. Display area 44 is displayed next to display area 42 . In this example, text data TD corresponding to the situation is acquired. If there was no misrecognition, as shown in FIG. 9, "touched wiring" would have been input as the text data TD corresponding to the cause. However, since "wiring" was converted to "defeat", text data TD corresponding to the cause has not been obtained. Therefore, the column for the cause of the aggregate sentence is blank.

FIG. 15C shows an example of displaying the hierarchical structure data HD on the display unit 40. The display unit 40 in this example presents the search results to the operator of the primary reception 230.

The display unit 40 highlights and displays in the display area 44 contents registered in the master DB that are similar to the extracted key points (for example, the category contents of "situation"). Then, if there is no erroneous conversion of the text data TD, the display unit 40 highlights and displays contents registered in the master DB that are similar to the category contents of "Cause" in the display area 44, as shown in FIG. It was supposed to. However, the text acquisition unit 10 incorrectly converted the text "I touched the wiring" into "I touched the losing battle," so the text is registered in the master DB similar to the category content of "Cause." It is not possible to highlight and display the displayed content.

FIG. 15D shows an example of a method for modifying text data TD. The figure is an example of a screen displayed by the display unit 40 in step S412 of FIG. 14. The operator of the primary reception desk 230 in this example corrects the erroneous conversion in the display area 42.

The modification button 46 is used to individually modify the text data TD. A correction button 46 is provided corresponding to each balloon in the display area 42. In the case of individual correction, the wording in the speech bubble to be corrected is corrected and the corresponding wording is corrected by pressing the correction button 46 placed next to the speech balloon. For example, by correcting "I touched the defeat" to "I touched the wiring" and pressing the correction button 46, the wording is corrected.

The batch correction button 48 is used to collectively correct the text data TD. One batch correction button 48 is provided in the display area 42. In the case of batch correction, input the wording before conversion and the wording after conversion, and press the batch correction button 48. For example, by inputting "defeat" as the wording before conversion, inputting "wiring" as the wording after conversion, and pressing the batch correction button 48, the text data TD displayed in the display area 42 is corrected. .

The operator may decide whether to select individual correction or batch correction. The operator may use both individual correction and batch correction depending on the situation. If the cause of the aggregate sentence has not been extracted, the operator may consider the possibility of misrecognition or misconversion while responding to the customer, and may correct the text data TD in real time.

FIG. 15E shows an example of display of the hierarchical structure data HD on the display unit 40. The figure is an example of a screen displayed by the display unit 40 in step S408 of FIG. 14 after modification. The display unit 40 in this example presents the search results to the operator of the primary reception 230.

The display unit 40 highlights and displays in the display area 44 contents registered in the master DB that are similar to the extracted key points (for example, the category contents of "Cause"). The text acquisition unit 10 acquires the corrected wording "I touched the wiring" as the text data TD of the cause. Therefore, the display unit 40 highlights "Touched wiring" which is similar to "Touched wiring" from the "Cause" category of the master DB. As a result, the operator of the primary reception desk 230 can immediately know from the hierarchical structure data HD displayed in the display area 44 what to do in relation to "touching a wire."

In this way, even if the text data TD is misrecognized or misconverted, the correspondence presentation device 100 can make corrections and present a more optimal countermeasure to the customer. The modification history of the text data TD may be stored in the master DB of the storage unit 20 as information related to past question and answer records. Thereby, the text acquisition unit 10 can reduce the possibility of misrecognition or misconversion based on the modification history of the text data TD.

Although the present invention has been described above using the embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It will be apparent to those skilled in the art that various changes or improvements can be made to the embodiments described above. It is clear from the claims that such modifications or improvements may be included within the technical scope of the present invention.

The order of execution of each process, such as the operation, procedure, step, and stage in the apparatus, system, program, and method shown in the claims, specification, and drawings, is specifically defined as "before" or "before". It should be noted that they can be implemented in any order unless the output of the previous process is used in the subsequent process. Even if the claims, specifications, and operational flows in the drawings are explained using "first," "next," etc. for convenience, this does not mean that it is essential to carry out the operations in this order. It's not a thing.

DESCRIPTION OF SYMBOLS 10... Text acquisition part, 20... Storage part, 30... Hierarchical structure data generation part, 40... Display part, 42... Display area, 44... Display area, 46... Correction button, 48... Batch correction button, 50... Hierarchy determining section, 100... Compatible presentation device, 200... Compatible system, 210... Customer, 220... Public switched telephone network, 225 ... Private branch exchange, 230 ... Primary reception, 232 ... Telephone, 234 ... Reception terminal, 240 ... Secondary reception, 242 ... Telephone, 244 ... Reception terminal, 250 ...LAN, 260...server machine

Claims (5)

a text acquisition unit that acquires text data obtained by voice recognition of the question and answer record;
a hierarchical structure data generation unit that classifies the main points of the text data of past question and answer records into predetermined categories, aggregates the main points, and generates hierarchical structure data of a master DB of past question and answer records ; ,
a storage unit that stores the master DB;
a display unit that displays the hierarchical structure data corresponding to the text data of the current question and answer record for each category classified by the hierarchical structure data generation unit;
The hierarchical structure data generation unit extracts the gist of the text data for each category from the text data of the conversation between the customer and the operator, and automatically generates content similar to the gist of the extracted text data from the corresponding category. search,
The display unit is a correspondence presentation device that highlights the hierarchical structure data having a high degree of similarity to the text data. The display unit displays the contents of the first category in order of similarity to the text data, and displays the contents of the second category, which is a lower hierarchy than the first category, in the master DB. The correspondence presentation device according to claim 1, wherein the correspondence presentation device displays the correspondence in order of the number of stored cases. further comprising a hierarchy determination unit that determines the number of hierarchies of the hierarchical structure data,
The correspondence presentation device according to claim 1 or 2, wherein the hierarchical structure data generation unit generates the hierarchical structure data according to the number of layers determined by the hierarchy determination unit. The text acquisition unit modifies the text data,
The hierarchical structure data generation unit generates the hierarchical structure data based on the corrected text data,
The correspondence presentation device according to any one of claims 1 to 3 , wherein the display unit displays the hierarchical structure data corresponding to the corrected text data for each category classified by the hierarchical structure data generation unit. . a step in which the computer obtains text data obtained by voice recognition of the question and answer record;
a step in which the computer classifies the main points of the text data of past question and answer records into predetermined categories, aggregates the main points, and generates hierarchical structure data of a master DB of past question and answer records ;
a step of a computer storing the master DB;
a step in which the computer extracts the gist of the text data for each category from the text data of the conversation between the customer and the operator;
a step in which the computer automatically searches for content similar to the gist of the extracted text data from the corresponding category;
a step in which the computer displays the hierarchical structure data corresponding to the text data of the current question and answer record for each of the classified categories;
a step of the computer highlighting the hierarchical structure data having a high degree of similarity to the text data;
A correspondence presentation method comprising:

Images