KR20230116144A - Consultation text summary system - Google Patents

Abstract

A consultation text summary system according to the present invention includes an input interface; and a processor that executes the counseling application program and controls the counseling screen to be displayed on the display. The processor calculates the importance weight for each keyword derived by morphologically analyzing the entire counseling sentence of the first call, calculates the weight for each sentence by substituting the calculated importance weight for each sentence, and calculates the weight for each sentence. The summary sentences may be configured in a first method of arranging sentences in order and extracting summary sentences from the sentences according to a selected summary rate, and in a second method in order of utterance time.

Description

Consultation text summary system}

The present invention relates to a counseling system for providing a counseling service by a counselor. More specifically, the invention relates to a method for summarizing a consultation text and a system for summarizing a consultation text. The present invention is a research conducted with the support of the Information and Communication Planning and Evaluation Institute with financial resources from the government (Ministry of Science and ICT) in 2021 (No.2021-0-00052, development and commercialization of an intelligent platform for contact center counseling assistants)

Recently, services based on advanced technologies such as artificial intelligence and robots are emerging like a flood due to the 4th industrial revolution. In fact, many people talk about what artificial intelligence can do, and even replacing the work of call center counselors is being discussed.

In addition, as various counseling channels have emerged in addition to telephone counseling, the term contact center has become more natural than the term call center.

In this regard, it is required to evaluate counseling on the content and counseling attitude of the counselor. On the other hand, even when summarizing and writing the contents of consultation after the end of consultation, in the case of a long call, many contents may be omitted if reliance on memory. In order to solve this problem, in the present invention, only important sentences can be extracted and provided to counselors for reminder purposes. Alternatively, even when the consultation manager needs to check the consultation contents, only important sentences are summarized so that the consultation contents search time can be reduced.

Regarding the existing counseling evaluation method, when the counseling evaluation performed the previous day is presented the next day after evaluating the counseling content recorded the previous day, there is a problem in that it is difficult to correct the wrong counseling content or correct the counseling attitude during counseling.

Therefore, in the present invention, the real-time counseling evaluation result is presented immediately after the call is terminated. In addition, in the present invention, real-time counseling evaluation results are presented immediately after the end of the corresponding unit in units of sentences, paragraphs, or consultation contents.

Accordingly, an object of the present invention is to provide a method for summarizing a consultation text and a system for summarizing a consultation text.

In addition, an object of the present invention is to present real-time counseling evaluation results immediately after a call is terminated.

In addition, an object of the present invention is to present real-time counseling evaluation results immediately after the end of a corresponding unit in units of sentences, paragraphs, or counseling contents.

A consultation text summary system according to the present invention includes an input interface configured to acquire real-time conversation through a microphone and a processor operatively coupled to the input interface and controlling a consultation screen to be displayed on a display by executing a consultation application program. . The processor calculates the importance weight for each keyword derived by morphologically analyzing the entire counseling sentence of the first call, calculates the weight for each sentence by substituting the calculated importance weight for each sentence, and calculates the weight for each sentence. The summary sentences may be configured in a first method of arranging sentences in order and extracting summary sentences from the sentences according to a selected summary rate, and in a second method in order of utterance time.

According to an embodiment, upon receiving a call termination signal from the STT engine, the processor performs preprocessing and parsing on the entire counseling sentence based on the closing sentence of each sentence through recognizing the silent section, recognizing the ending sentence, and removing the corresponding word. Sentence unit separation and sentence reconstruction are performed for each speaker, and natural language processing may be performed on the reconstructed sentence through morpheme analysis, noun unit extraction, and stopword filtering.

According to an embodiment, the processor classifies the reconstructed sentence into noun units according to the natural language processing, assigns a weight to a spoken sentence associated with a customer question, and based on term frequency-inverse document frequency (TF-IDF) Thus, it is possible to calculate the weight of a specific word in the noun unit within the reconstructed sentence.

According to an embodiment, the processor generates a sentence-term matrix associated with the weight of the specific word for the reconstructed sentence, and constructs a correlation matrix-based graph according to the generated sentence-word matrix. can create

According to an embodiment, the processor lists the sentences in the order of the highest weight for each sentence according to a text rank algorithm, and extracts summarized sentences from the sentences according to a selected summary rate. A first method and utterance Summary sentences of the document may be constructed in a second manner in chronological order.

According to an embodiment, the processor returns the final evaluation result of the call unit for all counseling sentences performed according to the summary sentences through the API of the consultation application program, and displays the final evaluation result of the call unit on the display. And the counseling application program may be controlled to display the summary sentences.

According to an embodiment, the processor may configure the final evaluation result of the call unit and the summary sentences by performing morpheme analysis, keyword analysis, natural language processing, and attribute classification using a keyword pattern-based analysis module.

According to an embodiment, the processor performs a summary of the original consultation text based on the information output from the keyword pattern-based analysis module, and when performing the summary of the original consultation text, the processor does not interwork with the AI-based analysis module and the matching rate analysis module. , The deep learning module can be interlocked only when the sentences are arranged in the order of the highest weight for each sentence according to the Text Rank algorithm.

According to an embodiment of the present invention, a method for summarizing a consultation text and a system for summarizing a consultation text may be provided based on artificial intelligence (AI).

In addition, according to an embodiment of the present invention, real-time counseling evaluation results may be presented in a manner in which the sentences are listed in the order of highest weight and the corresponding sentences are extracted according to a defined summary rate.

1 is a block diagram of a consultation text summary system according to an embodiment of the present invention.
2 is a detailed configuration diagram of a consultation text summary system according to an embodiment of the present invention.
3 is a flowchart of a method for summarizing a consultation text performed in the consultation text summary system according to the present invention.
4 shows a conceptual diagram of technical differences of the counseling text summary system according to the present invention and some configurations for implementing the same.
5 shows a detailed model and module configuration for summarizing a counseling text in the counseling text summary system according to the present invention.
6a to 6c show configurations related to differentiated technical features of the consultation text summary system according to the present invention.

Specific structural or functional descriptions of the embodiments according to the concept of the present invention disclosed in this specification are only illustrated for the purpose of explaining the embodiments according to the concept of the present invention, and the embodiments according to the concept of the present invention It can be embodied in various forms and is not limited to the embodiments described herein.

Embodiments according to the concept of the present invention can apply various changes and have various forms, so the embodiments are illustrated in the drawings and described in detail in this specification. However, this is not intended to limit the embodiments according to the concept of the present invention to specific disclosure forms, and includes all changes, equivalents, or substitutes included in the spirit and technical scope of the present invention.

Terms such as first or second may be used to describe various components, but the components should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another, e.g. without departing from the scope of rights according to the concept of the present invention, a first component may be termed a second component and similarly a second component may be termed a second component. A component may also be referred to as a first component.

It is understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, but other elements may exist in the middle. It should be. On the other hand, when a component is referred to as “directly connected” or “directly connected” to another component, it should be understood that no other component exists in the middle. Other expressions describing the relationship between components, such as "between" and "directly between" or "adjacent to" and "directly adjacent to", etc., should be interpreted similarly.

Terms used in this specification are used only to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, terms such as "comprise" or "having" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in this specification, but one or more other features It should be understood that it does not preclude the possibility of the presence or addition of numbers, steps, operations, components, parts, or combinations thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and unless explicitly defined in this specification, it should not be interpreted in an ideal or excessively formal meaning. don't

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the scope of the patent application is not limited or limited by these examples. Like reference numerals in each figure indicate like elements.

1 is a block diagram of a consultation text summary system according to an embodiment of the present invention. As shown in FIG. 1 , the counseling text summary system according to an embodiment of the present invention is a system composed of a customer terminal 100 , a server 200 , a counselor computer 300 and a communication network 400 . The communication network 400 may be a wired and/or wireless communication network that communicatively connects the customer terminal 100 and the counselor computer 300 . In addition, the communication network 400 may be a wired and/or wireless communication network that communicatively connects the customer terminal 100 and the server 200 or the server 200 and the counselor computer 300 .

Meanwhile, when the processor 330 receives a call termination signal from the STT engine 540, the counselor computer 300 implemented with the counseling text summary system according to the present invention generates a summary text for the corresponding call text. To this end, the processor 330 interworks with the STT engine 540 to receive the original counseling text when the call is terminated, generate a document summary through sentence-by-sentence separation and sentence reconstruction, and perform final evaluation on a call-by-call basis. In addition, the counseling application program 600 may receive a final evaluation for each call in association with the processor 330 . This series of procedures will be described in detail in FIG. 3 .

Meanwhile, FIG. 2 is a detailed configuration diagram of a consultation text summary system according to an embodiment of the present invention.

1 and 2, the counselor computer 300 may be configured to include a communication unit 310, an input interface 320, a processor 330, a display 340 and a memory 350. The server 200 is operably coupled with a plurality of customer terminals 100 and counselor computers 300 and may be composed of a plurality of servers.

The communication unit 310 may be configured to be communicatively coupled with a plurality of customer terminals 100 and the server 200, and may be implemented as a wired and/or wireless communication network. The input interface 320 may be configured to acquire real-time conversations of the counselor and/or the customer through the microphone 321 . The input interface 320 is not limited thereto and may further include other input means capable of controlling the counselor computer 300, such as a keyboard and a mouse.

Processor 330 may be operably coupled with input interface 320 . The processor 330 may control a counseling application program to be executed so that the counseling screen is displayed on the display 340 . The memory 350 may be configured to store counseling-related information and to access the information when a counseling application program is executed.

The counseling text summary system according to the present invention can evaluate the counseling content related to the spoken sentence during the counseling process in conjunction with the KMS (Knowledge Management Service) system. In this regard, in the present invention, only important conversation contents may be provided in order to reduce the time required to check the entire contents of consultation after the end of consultation.

On the other hand, even when the counselor summarizes and writes the counseling contents after the end of the consultation, in the case of a long call, many contents may be omitted if the counselor relies on memory. To solve this problem, only important sentences may be extracted and provided to the counselor for reminder purposes.

In addition, even when the consultation manager needs to check the consultation contents, only the important sentences are summarized so that the consultation contents search time can be reduced.

Implementation of the consultation text summary system according to the present invention can be implemented as follows. It is possible to analyze the morpheme of the entire counseling sentence of one call and calculate the importance weight for each keyword derived through this. After the keyword weight is calculated, the weight for each sentence can be calculated by substituting it into the sentence again. A method in which sentences are listed in the order of high weight for each sentence, and the sentences are extracted according to the defined summary rate (e.g., if the original text is 100 lines and the summary rate is 50%, the order of the most important sentences (primary sort), Only 50 dialogue sentences remain in chronological order of speech (secondary sorting).

Specifically, in the present invention, when an entire counseling sentence for one call is input, keywords and sentence weights are calculated, and a final summary sentence is provided according to the user's selection. In this regard, as an example, the user-selected summary rate may be 50%, 70%, or 5-line summary.

In relation to the concept and implementation method of the present invention described above, the counseling text summary system of the present invention may be configured as shown in FIG. 3 . In this regard, FIG. 3 shows a flow chart of a method for summarizing a consultation text performed in the system for summarizing a consultation text according to the present invention.

Referring to FIGS. 2 and 3 , the processor 330 interworking with the STT engine 540 may run the counseling application program 600 to perform NLP Pre-Processing and TextRank analysis combination. Specifically, the processor 330 may perform a batch request of the original call counseling text, pre-processing and parsing of sentence ending sentences, and separation/reconstruction of sentence units for each speaker.

Thereafter, the processor 330 may perform preprocessing associated with natural language processing (NLP), classification/customer question utterance sentence weight calculation, term frequency-inverse document frequency (TF-IDF) keyword weight calculation, and sentence-word matrix generation. . TF-IDF is a weight used in information retrieval and text mining, and represents a statistical value that indicates how important a word is in a specific document when there is a document group consisting of several documents.

Thereafter, the processor 330 may transfer the creation of a graph based on a correlation matrix, the application of the TextRank algorithm, the generation of a document summary, and the return of the final evaluation result in each call to the counseling application program 600 through an API. Meanwhile, the processor 330 performing the aforementioned operations may be driven by the counseling application program 600 or may be a separate software module that performs a summary of the counseling text, for example, the keyword pattern-based analysis module 630 of FIG. 5 . .

Referring to the technical features described above, the counseling text summary system according to the present invention will be described. The text summary system may include an input interface 320 configured to acquire real-time conversation via a microphone. The consultation text summary system may include a processor 330 operatively coupled to an interface 320 . The processor 330 may control the counseling application program 600 to be executed so that the counseling screen is displayed on the display 340 .

The processor 330 may calculate an importance weight for each keyword derived by morphologically analyzing the entire counseling sentence of the first call. The processor 330 may calculate a weight for each sentence by substituting the calculated importance weight for each sentence. The processor 330 may configure the summary sentences in a first method of arranging the sentences in the order of the sentence-specific weight and extracting summary sentences from the sentences according to the selected summary rate and in a second method in order of utterance time. there is.

When the processor 330 receives the call termination signal from the STT engine 540, the processor 330 performs preprocessing and parsing on the entire counseling sentence based on the closing sentence of each sentence through recognizing the silent section, recognizing the final sentence, and removing the corresponding word. can The processor 330 may perform sentence unit separation and sentence reconstruction for each speaker, and perform natural language processing on the reconstructed sentence through morpheme analysis, noun unit extraction, and stopword filtering.

The processor 330 may classify the reconstructed sentences into noun units according to natural language processing and assign weights to spoken sentences associated with customer questions. The processor 330 may calculate a weight of a specific word in the noun unit within the reconstructed sentence based on term frequency-inverse document frequency (TF-IDF).

The processor 330 may generate a sentence-term matrix associated with the weight of the specific word for the reconstructed sentence. The processor 330 may generate a correlation matrix-based graph according to the generated sentence-word matrix.

The processor 330 may list the sentences in the order of the highest weight for each sentence according to a text rank algorithm. The processor 330 may compose summary sentences of the document in a first method of extracting summary sentences from the sentences according to the selected summary rate and in a second method in order of utterance times. The selected summary rate may be a predefined value in the application or configured to be input by the user.

The processor 330 may return the final evaluation result of each call for all counseling sentences performed according to the summary sentences through the API of the counseling application program 600 . The processor 330 may control the counseling application program 600 to display the final evaluation result of the call unit and the summary sentences on the display 340 .

The processor 330 may configure the final evaluation result of the call unit and the summary sentences by performing morpheme analysis, keyword analysis, natural language processing, and attribute classification using a keyword pattern-based analysis module.

The method of summarizing the consultation text performed by the system for summarizing the consultation text according to the present invention may aim at a certain level of satisfaction or a certain level of response performance. In this regard, FIG. 4 shows a conceptual diagram of technical differences of the counseling text summary system according to the present invention and some configurations for implementing the same. Referring to FIG. 4, the satisfaction of the method for summarizing the original text according to the present invention can be verified through qualitative evaluation, and the response performance can be aimed at within 1 second, but is not limited thereto and can be changed according to the application. .

Meanwhile, the method for summarizing the original text according to the present invention is not simply limited to the TestRank sentence weight-based reduction. Therefore, the counseling text summary method according to the present invention can be performed through TextRank combination after preprocessing such as filtering ruleset and segmented sentence combination.

To this end, in the present invention, natural language pre-processing (NLP) (S310) can be performed in a state in which sentence unit separation and sentence reconstruction have been performed for each speaker. Subsequently, classification/customer question utterance sentence weight calculation (S320), TF-IDF (Term Frequency-Inverse Document Frequency) keyword weight calculation (S330), and sentence-word matrix generation (S340) may be performed. TF-IDF is a weight used in information retrieval and text mining, and represents a statistical value that indicates how important a word is in a specific document when there is a document group consisting of several documents.

Thereafter, generation of a graph based on a correlation matrix (S350), application of a TextRank algorithm (S360), generation of a document summary (S370), and return of final evaluation results in units of calls (S380) may be performed. Through these series of processes, in the present invention, the counseling text summary method can be performed through the counseling text summary system.

Referring to the technical features described above, the counseling text summary system according to the present invention will be described. The text summary system may include an input interface 320 configured to acquire real-time conversation via a microphone. The consultation text summary system may include a processor 330 operatively coupled to an interface 320 . The processor 330 may control the counseling application program 600 to be executed so that the counseling screen is displayed on the display 340 .

Meanwhile, various modules including the STT engine 540 and the counseling application program 600 according to the present invention may be implemented for summarizing the counseling text. In this regard, FIG. 5 shows a detailed model and module configuration for summarizing a counseling text in the counseling text summary system according to the present invention.

2 to 5 , when the counseling application program 600 is driven by the processor 330, the counseling application program 600 may be layered as follows. As another example, the counseling application program 600 may be implemented in the server 200 .

The counseling application program 600 is configured to include an input interface 610, a learning/evaluation model 620, a keyword pattern-based analysis module 630, an AI-based analysis module 640, and a real-time processing/interworking module 650. It can be. The keyword pattern-based analysis module 630 may include the KMP matching rate analysis module 631 or the matching rate analysis module 631 may be configured separately from the keyword pattern-based analysis module 630 . The AI-based analysis module 640 may include the deep learning module 641 or the deep learning module 641 may be configured separately from the AI-based analysis module 640 .

The input interface 610 may be configured to obtain consultation text, application metadata, KMS-related data, and administrator-defined data. The learning/evaluation model 620 may be referred to as a KMS learning model. The learning/evaluation model 620 may include a knowledge data learning model 621 that learns knowledge data. The learning/evaluation model 620 may be configured to include a classification data learning model 622 that classifies and learns original counseling texts and a counseling evaluation model 623 including a script and an evaluation criterion model.

The processor 330 may perform keyword pattern-based analysis for performing morphological analysis, keyword analysis, natural language processing, and attribute classification of spoken sentences using the knowledge data learning model 621 . The processor 330 may control similar knowledge contents to be recommended within a predetermined range through word embedding of spoken sentences, similarity analysis, and detailed tuning using the knowledge data learning model 621 . To this end, the processor 330 may interwork with the AI-based analysis module 640.

In this regard, a knowledge management platform (KMP) associated with the consultation text summary system may include a knowledge data learning model 621 for learning knowledge data. The KMP may be configured to include a classification data learning model 622 that classifies and learns counseling texts and a counseling evaluation model 623 including a script and an evaluation criterion model.

The processor 330 may perform morpheme analysis, keyword analysis, natural language processing, and attribute classification using the keyword pattern-based analysis module 630 to construct the final evaluation result of the call unit and the summary sentences.

The processor 330 may perform a summary of the consultation text based on the information output from the keyword pattern-based analysis module 630 . When performing the summary of the consultation text, it may be implemented so as not to interwork with the AI-based analysis module 640 and the matching rate analysis module 631. Accordingly, while performing a summary of the consultation text through the keyword pattern-based analysis module 630, at the same time through the AI-based analysis module 640 and the matching rate analysis module 631, other services, such as knowledge content recommendation, real-time consultation evaluation or consultation type can be classified. The deep learning module 641 may interwork with the processor 330 only when the sentences are arranged in the order of the sentence-specific weight according to a text rank algorithm.

Accordingly, the method of summarizing the consultation text by the system for summarizing the consultation text according to the present invention may be performed by a combination of 1) + 5) + 8), but is not limited thereto.

On the other hand, the processor 330 may ensure the accuracy of the consultation text summary through the application of the AI analysis engine's differentiation technology.

6A shows a configuration applying a BERT + GPT mixed model. 6B shows a configuration for combining byte pair encoding and morpheme analysis during conversion and tokenization of written language data. 6C shows a configuration to which a string matching improvement algorithm considering free speech characteristics is applied. To this end, a sentence search screen area within the counseling text summary algorithm is defined, and accordingly, a string matching improvement algorithm considering free speech characteristics may be implemented.

Referring to FIGS. 2 to 6A , the processor 330 may analyze a matching rate according to a keyword pattern associated with essential counseling items defined in the KMP using a matching rate analysis model associated with the KMP. The processor 330 may apply the weight obtained through the Bidirectional Encoder Representations from Transformers (BERT) method to the Generative Pre-trained Transformer (GPT) method. The processor 330 may perform decoding on each sentence associated with the counseling contents, morpheme analysis, keyword analysis, natural language processing, and attribute classification. Meanwhile, when the BERT + GPT mixed model is applied, the original counseling text or the feature keyword of the original counseling text may be the original text collected by the STT engine 540 and uttered by the counselor. Therefore, in principle, the counseling text summary method according to the present invention can be performed through matching with a dictionary model pattern string without AI processing by an AI engine or the like.

Through the configuration of applying the BERT + GPT mixed model, it is possible to improve the accuracy and speed of the original text summary during morpheme analysis, keyword analysis, natural language processing, and attribute classification by performing decoding on each sentence related to the counseling content.

Referring to FIGS. 3 to 5 and 6B , the AI model learning engine 510 may data-encode written data of the counseling text into colloquial data. In addition, the AI model learning engine 510 may perform term expansion learning using a language model that has already been trained on the encoded counseling text data. In this regard, when performing the method of summarizing the counseling text according to the present invention, colloquial data may be the original text collected by the counselor through the STT engine 540. Therefore, in principle, the counseling text summary method according to the present invention can be performed through matching with a dictionary model pattern string without AI processing by an AI engine or the like.

However, the AI engine may be used to learn terms and extract feature keywords from the colloquial speech collected from the STT engine 540 with the original consultation text.

Meanwhile, the process of performing AI learning by extracting feature keywords from the original text data is performed by the AI model learning engine 510 and the classification/analysis model 530, or the ruleset model 520 and the classification/analysis model 530 ) can be performed by Keyword extraction-based AI learning performed by the ruleset model 520 and the classification/analysis model 530 will be described as follows. In this regard, feature keywords may be extracted from original counseling text data through the ruleset model 520 . A secondary search of a morpheme dictionary for token attribute extraction may be performed. AI learning may be performed on the feature keywords extracted through the ruleset model 520 based on the classification/analysis model 530. In this regard, feature keywords may be separately stored in an AI model and a dictionary model.

Referring to FIG. 6C , in the method of summarizing the original counseling text according to the present invention, the original text string of the counseling text and the pattern string stored in the dictionary model can be compared with different matching intervals for each stage. In this regard, the original counseling text string is the original text uttered by the counselor collected through the STT engine 540. Therefore, the counseling text summary method according to the present invention can be performed through matching with a dictionary model pattern string without AI processing by an AI engine or the like. The comparison of pattern strings with different matching intervals for each stage is applied with a string matching improvement algorithm considering free speech characteristics.

In consideration of free speech characteristics, the string columns of the original counseling text strings (S1 to S3) in the first step (S100) to the third step (S300) may be configured to be shifted by a predetermined interval. In other words, the first consultation text string (S1) of the first step (S100) to the third consultation text string (S3) of the third step (S300) are transitions of the same counseling sentence at a predetermined time interval. For example, the second counseling original text string S2 of the second step S200 may be obtained by shifting the first counseling original text string S1 of the first step S100 by a specific character interval, for example, a two-character interval. . Meanwhile, the third consultation original text string S3 of the third step S300 may be in the form of shifting the second consultation original text string S2 of the second step S200 within one character interval, for example, by a 0.5 character interval. .

Sentence search screen areas within the counseling text summary algorithm are defined differently by differentiating the first matching section (M1) to the third matching section (M3) of the first counseling text string (S1) to the third counseling text string (S3) and can be implemented. For example, the first matching section M1 of the first counseling text string S1 and the second matching section M2 of the second counseling text string S2 are configured not to overlap each other, so that the main sentence of the counseling text is It is possible to search for a location where the ignition is fired. The second matching section (M2) of the second counseling text string (S2) and the second matching section (M3) of the third counseling text string (S3) are configured to partially overlap, so that the main sentence of the counseling text is uttered. and check pattern matching in detail.

In the above, the counseling text summary method and the counseling text summary system according to the present invention have been reviewed. The technical effects of the consultation text summary method and the consultation text summary system according to the present invention are as follows.

According to an embodiment of the present invention, a method for summarizing a consultation text and a system for summarizing a consultation text may be provided based on artificial intelligence (AI).

In addition, according to an embodiment of the present invention, it is possible to morpheme-analyze the entire counseling sentence of a call and calculate an importance weight for each keyword derived through the morpheme analysis.

The features and effects of the present invention described above will become more apparent through the following detailed description in conjunction with the accompanying drawings, and accordingly, those skilled in the art to which the present invention belongs can easily practice the technical idea of the present invention. You will be able to.

Since the present invention can make various changes and have various embodiments, specific embodiments will be illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the present invention to specific embodiments, and should be understood to include all modifications, equivalents, or substitutes included in the spirit and technical scope of the present invention.

According to the software implementation, not only the procedures and functions described in this specification, but also the design and parameter optimization of each component may be implemented as a separate software module. The software code may be implemented as a software application written in any suitable programming language. The software code may be stored in a memory and executed by a controller or processor.

100: customer terminal
200: server
300: counselor computer
310: communication department
320: input interface
330: processor
340: display
350: memory
400: communication network
540: STT engine
600: counseling application program
610: input interface
620: learning/evaluation model
621: knowledge data learning model
622: classification data learning model
623: consultation evaluation model
630: keyword pattern based analysis module
640: AI-based analysis module
650: real-time processing / interlocking module

Claims (8)

In the full text summary system,
an input interface configured to acquire a real-time conversation through a microphone; and
a processor operably coupled to the input interface and controlling a counseling application program to be executed and displayed on a display;
the processor,
Calculate the importance weight for each keyword derived by morphologically analyzing the entire counseling sentence of the first call,
Calculate the weight for each sentence by substituting the calculated importance weight for each sentence;
The counseling text summary system configures the summary sentences in a first method of arranging the sentences in the order of the highest weight for each sentence and extracting summary sentences from the sentences according to the selected summary rate and in a second method in order of utterance time. According to claim 1,
the processor,
When a call termination signal is received from the STT engine, pre-parsing is performed on the entire counseling sentence based on the closing sentence of each sentence through recognizing the silent section, recognizing the final ending, and removing the corresponding word,
Sentence unit separation and sentence reconstruction are performed for each speaker,
The counseling text summary system for performing natural language processing on the reconstructed sentence through morpheme analysis, noun unit extraction, and stopword filtering. According to claim 2,
the processor,
Classifying the reconstructed sentences into noun units according to the natural language processing, assigning weights to uttered sentences related to customer questions,
Based on TF-IDF (Term Frequency-Inverse Document Frequency), the counseling text summary system calculates the weight of the specific word of the noun unit within the reconstructed sentence. According to claim 3,
the processor,
For the reconstructed sentence, a sentence-term matrix associated with the weight of the specific word is generated;
A counseling text summary system for generating a correlation matrix-based graph according to the generated sentence-word matrix. According to claim 4,
the processor,
Sentences are arranged in the order of highest weight for each sentence according to a text rank algorithm;
A system for compiling summary sentences of a document in a first method of extracting summary sentences from the sentences according to a selected summary rate and in a second method in order of utterance time. According to claim 5,
the processor,
Returning the final evaluation result of each call for all counseling sentences performed according to the summary sentences through the API of the counseling application program,
and controlling the counseling application program to display the final evaluation result of the call unit and the summary sentences on the display. According to claim 6,
the processor,
A counseling text summary system that configures the final evaluation result of the call unit and the summary sentences by performing morpheme analysis, keyword analysis, natural language processing, and attribute classification using a keyword pattern-based analysis module. According to claim 7,
the processor,
Performing a summary of the consultation text based on the information output from the keyword pattern-based analysis module;
When performing the summary of the counseling text, the AI-based analysis module and the matching rate analysis module are not interlocked, and the deep learning module interlocks only when the sentences are listed in the order of high weight for each sentence according to the Text Rank algorithm Characterized in that, the consultation text summary system.

Images