KR101080840B1 - System for evaluating communication quality and Its using method - Google Patents

Abstract

The present invention relates to a dialogue quality measurement system and a dialogue quality measurement method, and by evaluating quantitatively the similarity of dialogue contents between dialogue persons in various industrial sites, to evaluate dialogue fidelity and understanding among dialogue persons, The present invention relates to a system capable of measuring the performance of a job and a method of measuring the same.

More specifically, the present invention provides a conversation quality measurement system including a plurality of conversation terminal and a conversation quality measurement module, wherein the conversation quality measurement module is connected to the conversation terminal by wire or wirelessly and connects the plurality of conversation terminals. Voice receiving unit for receiving a conversation message through; A chat message storage unit for storing a chat message received in the voice receiver as a voice file; A voice file conversion unit for converting and displaying the voice file stored in the conversation message storage unit into a text file, and storing the converted text file; A conversation quality evaluation unit for creating and storing a recording using a text file stored in the voice file converting unit and measuring conversation quality using the recording and sentence similarity calculation logic; And a display unit connected to the voice file converting unit and the conversation quality evaluating unit to display the text file, the recording, the similarity evaluation process, and the similarity evaluation result on a terminal screen. The conversation quality evaluation unit comprises: a recording storage unit for creating a recording using a text file stored in the voice file converting unit, and classifying the recording according to a conversation time, a talker, a message type, or a message content; A morpheme evaluation unit for dividing the conversation message of the transcript into morpheme units, merging for each talker, and performing sentence similarity calculation using similarity calculation logic on the morpheme sets merged for each talker; And performing semantic expansion on the conversation message of the transcript, separating the semantic extended conversation message into morpheme units, merging by each talker, and calculating similarity with respect to the merged morpheme set by the talker of the semantic extended chat message. It provides a dialogue quality measurement system comprising a; semantic evaluation unit for performing sentence similarity calculation using logic.

Sentence similarity, dialogue quality, dialogue quality measurement method, cosine similarity, power plant

Description

System for evaluating communication quality and Its using method

The present invention relates to a dialogue quality measurement system and a dialogue quality measurement method, and by quantitatively evaluating the similarity of the dialogue contents between the dialogues in various industrial sites such as factories, plants, and power plants, the dialogue fidelity and understanding of the dialogues. The present invention relates to a system and a method of measuring the performance of a job based on the evaluation.

Recently, due to the rapid development of various energy utilization technologies and electronic technologies and the advancement of building technology, various industrial facilities are being enlarged, and large industrial facilities are being created considering the efficiency and surplus creation, but the stability and smoothness The process should be guaranteed

In particular, as seen in the Chernobyl nuclear power plant accident in Russia, large industrial facilities always present potential high risks, so reducing the risk as much as possible and ensuring maximum stability are top priority.

In various plants, such as power plants and factories handling dangerous goods, safety devices are provided using various electronic equipment. However, the control of the safety devices is ultimately determined by humans, so the precise control and operation of the safety devices are performed. This is important.

In particular, the communication quality (Communication Quality) of the operator who controls the safety device in the emergency situation, such as emergency, can be the main factor to determine the situation response performance, prevent or prevent the occurrence of accidents.

If there is a way to quantitatively evaluate the quality of communication, it can be used to establish improved communication between operators and to reduce human error due to inadequate communication. When quality communication is defined as the right understanding and understanding of each other's conversations, it can be judged based on the fidelity and similarity of conversation messages.

Also, by measuring the communication quality continuously, it is possible to improve the job skills of operators in high-risk facilities and to help prevent accidents.

Therefore, i) a system and method for measuring conversational quality between conversational persons at various industrial facilities, and ii) a system and method for measuring conversational quality suitable for various industrial sites. There is a need for systems and methods that assist in prevention.

The present invention quantitatively evaluates the similarity of the conversation contents related to the job between the dialogers in various industrial sites, thereby evaluating the fidelity and understanding of the dialogue between the dialogers and measuring the performance of the job based on this. And to provide a measuring method for that purpose.

In addition, the present invention provides a system having a similarity analysis by morpheme classification, similarity analysis by expansion of meanings, or a similarity analysis method that combines them with respect to the conversation messages between the dialogue persons, so that the customized job performances suitable for various industrial sites can be evaluated. Another aim is to provide a method of measuring the system.

In addition, the present invention quantitatively evaluates the similarity of contents of conversations generated by the operators of various high-risk facilities such as nuclear power plants to perform corresponding driving tasks in an emergency situation, and repeats the process of measuring the performance of the high-risk facilities. Another aim is to provide a system that helps prevent accidents.

The technical objects to be achieved by the present invention are not limited to the above-mentioned technical problems, and other technical subjects which are not mentioned can be clearly understood by those skilled in the art from the description of the present invention .

In order to solve the above-mentioned problems of the prior art, the present invention, a conversation quality measuring system having a plurality of conversation terminals and conversation quality measurement module, the conversation quality measurement module, the conversation terminal and the wired or wireless connection A voice receiver configured to receive a chat message through the plurality of talker terminals; A chat message storage unit for storing a chat message received in the voice receiver as a voice file; A voice file conversion unit for converting and displaying the voice file stored in the conversation message storage unit into a text file, and storing the converted text file; A conversation quality evaluation unit for creating and storing a recording using a text file stored in the voice file converting unit and measuring conversation quality using the recording and sentence similarity calculation logic; And a display unit connected to the voice file converting unit and the conversation quality evaluating unit to display the text file, the recording, the similarity evaluation process, and the similarity evaluation result on a terminal screen. The conversation quality evaluation unit comprises: a recording storage unit for creating a recording using a text file stored in the voice file converting unit, and classifying the recording according to a conversation time, a talker, a message type, or a message content; A morpheme evaluation unit for dividing the conversation message of the transcript into morpheme units, merging for each talker, and performing sentence similarity calculation using similarity calculation logic on the morpheme sets merged for each talker; And performing semantic expansion on the conversation message of the transcript, separating the semantic extended conversation message into morpheme units, merging by each talker, and calculating similarity with respect to the merged morpheme set by the talker of the semantic extended chat message. It provides a dialogue quality measurement system comprising a; semantic evaluation unit for performing sentence similarity calculation using logic.

In the present invention, the dialogue quality evaluation unit calculates a similarity evaluation result by applying a predetermined correction value to the similarity evaluation results of the morpheme evaluation unit and the semantic evaluation unit, and synthesizes the unit job similarity evaluation results according to the classification of conversation contents. A final similarity evaluation unit for calculating job similarity evaluation results; Conversation quality measurement system further comprises a.

In the present invention, the talker terminal includes a talk quality measurement system, characterized in that any one selected from a mobile phone, a computer, a notebook, an electronic notebook, a personal digital assistant (PDA) or a radio.

In the present invention, the dialogue message includes a dialogue quality measurement system, characterized in that the dialogue message transmitted or received between the dialogue using a plurality of terminals in a factory, power plant, fire department, police station, hospital, government office or construction site.

The present invention relates to a method for measuring conversation quality using a conversation quality measurement system, comprising: receiving a conversation message transmitted or received between a plurality of conversation terminal terminals, creating a recording using the conversation message, a conversation time, a conversation person and A first step of classifying a conversation message type according to the message content and subdividing and storing the conversation message into unit jobs; A second step of dividing the classified conversation message into morpheme units to perform sentence similarity calculation; Performing semantic expansion on the classified conversation message, and performing sentence similarity calculation by separating the semantic extended conversation message into morpheme units; And a fourth step of combining the similarity evaluation result by the morpheme classification and the similarity evaluation result by the semantic expansion, and calculating a final job similarity by applying a predetermined correction value. It provides a conversation quality measurement method comprising a.

In the present invention, the second step may include: dividing the recording into unit conversations of a dialogue person, separating dialogue messages of each dialogue unit into morpheme units, merging the separated morphemes into each dialogue unit, and each dialogue unit. Conversation quality measurement method, characterized in that the step of calculating the unit job similarity evaluation results by performing a sentence similarity calculation using similarity calculation logic for the merged morpheme set for each.

In the present invention, the third step may include: dividing the recording into unit conversations of a talker, performing semantic expansion on the conversation message of each talker, and separating the conversation message of each talker's conversation into morpheme units. Merging the separated morphemes for each talker and calculating sentence similarity calculations using semantic expansion using sentence similarity calculation logic for the set of morphemes merged for each talker. Conversation quality measurement method characterized in that the.

The present invention includes a method of measuring a conversation quality, further comprising the step of repeatedly performing the above conversation quality measuring method and calculating a plurality of unit job similarity evaluation results.

In the present invention, the similarity calculation logic extracts a feature vector from a voice or a voice-converted character of each conversation, and calculates similarity between sentences using a cosine similarity calculation algorithm. It includes.

In the present invention, the fourth step is to summarize the unit job similarity calculated through the sentence similarity calculation of the second step, and to obtain the total number of words of the morpheme words performed in the analysis, each unit job similarity total Dividing by the number of words, calculating proportional unit duty similarity for the whole job, summing the values of the proportional unit duty similarity calculated from each unit duty similarity, and determining job similarity by morphological analysis; Compiling the unit job similarity calculated through the sentence similarity degree calculation of the third step, and calculating the total number of words of the morpheme words performed in the semantic expansion analysis, each unit job similarity by the semantic expansion to the semantic expansion analysis Calculating proportional unit job similarity by dividing each word by the total number of morpheme words, Summing the values of proportional unit job similarities by each semantic expansion calculated from the unit job similarity by each semantic expansion to determine job similarity by semantic expansion, determining a predetermined correction value and the correction value The conversation quality measurement method comprising the step of calculating the final job similarity by applying a.

The final job similarity in the present invention is the following formula,

And generates a result, and f is a correction value, and includes a conversation quality measuring method, which satisfies a condition of 0 <f <1 .

According to the present invention, by quantitatively evaluating the similarity of the conversation contents related to the job between the dialogers in various industrial sites, it is possible to evaluate the fidelity and understanding of the dialogue between the dialogers, and measure the performance of the job based on this. It is effective to provide a system and a method of measuring the same.

In addition, according to the present invention, a system having a similarity analysis method by i) similarity analysis by morpheme classification, ii) similarity analysis by semantic expansion, or iii) combining the conversation messages between the dialogue persons is suitable for various industrial sites. It is effective to provide a system that can measure customized job performance.

In addition, the present invention quantitatively evaluates the similarity of contents of conversations generated by operators of various high-risk facilities such as nuclear power plants, factories, and plants to perform corresponding driving tasks in an emergency situation, and measures the performance of tasks. Repetition has the effect of providing a system that helps prevent accidents in high-risk facilities.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. Prior to this, terms or words used in the specification and claims should not be construed as having a conventional or dictionary meaning, and the inventors should properly explain the concept of terms in order to best explain their own invention. Based on the principle that can be defined, it should be interpreted as meaning and concept corresponding to the technical idea of the present invention. Therefore, the embodiments described in the specification and the drawings shown in the drawings are only one of the most preferred embodiments of the present invention and do not represent all of the technical idea of the present invention, various modifications that can be replaced at the time of the present application It should be understood that there may be equivalents and variations.

1 is a block diagram of a dialogue quality measurement system according to an embodiment of the present invention.

The present invention provides a dialogue quality measurement system including a dialogue quality measurement module 102 including a dialogue quality evaluation unit and a plurality of dialogue terminal terminals 101.

The conversation quality measurement module 102, i) voice connection unit 103 is connected to the talker terminal 101 by wire or wireless, and receives a chat message through a plurality of talker terminal 101 ii) the voice receiver ( Dialogue message storage unit 104 for storing the conversation message received in 103) as a voice file iii) It is possible to convert the voice file stored in the conversation message storage unit 104 to a text file and display the converted text, Voice file conversion unit 105 for storing a file iv) Dialog quality for creating and recording a recording using a text file stored in the voice file conversion unit 105 and measuring conversation quality using the recording and the similarity calculation logic. Evaluator 106 and v) is connected to the voice file conversion unit 105 and the conversation quality evaluation unit 106 to display the text file, recording, similarity evaluation process and similarity evaluation results on the terminal screen And a vector unit (107).

In addition, the dialogue quality measurement module 102 may store an algorithm related to the similarity calculation logic and use it for sentence similarity calculation.

The conversation quality evaluation unit 106 included in the conversation quality measurement module 102 includes: i) creating a recording using a text file stored in the voice file converting unit, and recording the recording in conversation time, conversation person, message type, or the like. Recording storage unit 108 to classify and store according to the message content ii) The conversation message of the recording is divided into morpheme units, merged by each conversation, and the sentence using the similarity calculation logic for the morpheme set merged for each conversation Morpheme evaluation unit 109 for performing similarity calculation iii) performing semantic expansion on the conversation message of the transcript, separating the semantic extended conversation message into morpheme units, merging each conversational person, and the semantic extended conversation A semantic evaluation unit that performs sentence similarity calculation using similarity calculation logic on the merged morpheme sets for each talker of a message (1 10) vi) The similarity evaluation result of the morphological evaluation unit and the semantic evaluation unit is applied to calculate a similarity evaluation result by applying a predetermined correction value, and the final job similarity evaluation result is synthesized by synthesizing the unit job similarity evaluation results according to the classification of conversation contents. The final similarity evaluator 111 may be calculated.

In addition, the talker terminal 101 may be any one selected from a mobile phone, a computer, a notebook, an electronic notebook, a personal digital assistant (PDA) or a radio. However, the present invention is not limited thereto, and any means capable of transmitting or receiving a conversation between the talkers at a separated distance through a wired or wireless communication means may be used in the present invention.

The voice receiver 103 may adopt any means that can receive the contents transmitted and received between the talker terminal 101.

The conversation message may be a conversation message transmitted or received between the talkers using a plurality of terminals in a factory, a power plant, a fire station, a police station, a hospital, a public office, or a construction site. However, the present invention is not limited thereto, and if the site is associated with high-risk facilities such as various industrial facilities, the conversation quality measuring system of the present invention may be used.

2 is an exemplary view showing a process of calculating sentence similarity using a cosine similarity calculation method according to an embodiment of the present invention.

The present invention quantitatively evaluates the similarity of the conversation messages related to the job among the dialogers in various industrial sites, thereby evaluating the fidelity and understanding of the dialogue between the dialogers and measuring the performance of the job based thereon. And a measuring method thereof.

In the present invention, in order to quantitatively evaluate the similarity of the conversation message, a similarity calculation logic may be used. In particular, a cosine similarity calculation method is performed by extracting a feature vector from a voice or a voice-converted character of each conversation. It is possible to calculate the similarity between sentences using.

In the present invention, the similarity of the sentence by the similarity evaluation logic may be calculated using the similarity evaluation logic using a vector widely used in mathematics and computer engineering. Similarity evaluation logic is a technique of expressing dialogue sentences between operators as a mathematical vector, calculating the distance between each vector, and expressing the results as similarity between sentences. It is mainly used in fields such as internet search in engineering field.

For example, when a web document is to be searched on the Internet, when a user enters a specific keyword into a web search engine, the keyword and the search engine calculate an associative similarity with the indexed results of various documents in the web space. Based on this, the documents most relevant to the search term are selected and output as the results of the search engine. In this case, the documents with high similarity to the search word are placed first in the arrangement of the search results because the similarity evaluation results using the similarity calculation technique are used.

Although there are many similarity calculation techniques that can be employed in the present invention, in particular, a cosine similarity calculation method can be used, and the cosine similarity calculation method can be calculated by Equation 1 below.

At this time, the d _ik is a sentence related to the question of the speaker 1, q _ik is a sentence related to the response of the speaker 2. Therefore, the cosine similarity is calculated by extracting the feature vector from the sentence related to the question and the sentence related to the response, respectively, and extracting Vector Text 1 and Vector Text 2.

1 is an example of evaluating dialogue similarity calculated using a similarity evaluation technique using the cosine similarity calculation method described above with a dialogue message between two operators of a power plant.

The conversation between the two operators is the response of the plant manager in response to the orders issued by the plant manager to the plant manager. In order to evaluate the similarity of the conversation message, when the morpheme (index) is separated for each sentence, it is extracted like sentence 1 and sentence 2. For each extracted morphological index, expressing it as a vector space and obtaining a vector of each morpheme, Vector Text 1 and Vector Text 2 are extracted, and the distance between these vectors is represented by a cosine similarity equation. When calculated using, it can be seen that the calculation result shows 0.6708204, which is 67.1% conversation sentence similarity.

The cosine similarity calculation algorithm may be used to calculate sentence similarity by sentence analysis and unit job similarity by semantic expansion analysis, which will be described later.

3 is a flow chart of a conversation quality measuring method according to an embodiment of the present invention.

The present invention relates to a method for measuring conversation quality using a conversation quality measurement system, comprising: receiving a conversation message transmitted or received between a plurality of conversation terminal terminals, creating a recording using the conversation message, and classifying a conversation message type. A first step (s310) of subdividing and storing the classified conversation messages into morpheme units to perform sentence similarity calculation (s320), and to extend the meaning of the classified conversation messages. After performing the above, the third step of performing sentence similarity calculation by dividing the semantic extended dialogue message into morpheme units (S330) and the similarity evaluation result by the morpheme classification and the similarity evaluation result by the semantic expansion are predetermined correction values. Calculate the similarity evaluation result by applying the result, and the final job similarity by combining the unit job similarity evaluation results It provides a dialogue quality measuring method including the step 4 (s340) whether to perform.

That is, the first step (s310) is a preparation step for acquiring the conversational data between the conversations and performing the conversation quality evaluation, the second step is a sentence similarity evaluation step by the sentence morphological distinction between the conversations, the third step It is a step of evaluating sentence similarity by sentence semantic expansion, and the fourth step may be referred to as a step for obtaining final job similarity by statistically processing the results of dialogue similarity evaluation between the dialogue speakers obtained in the second and third stages, respectively.

The first step (s310) is a preparation step for acquiring conversation data and evaluating the conversation quality. I) First, a conversation message is obtained, which is a conversation message between the conversation receivers, and then the conversation message. The voice file is stored in the storage unit, the voice file is converted into a text file, and then stored in the voice file conversion unit.

ii) A transcript is then prepared, which can include the speaker, the start and end times, and the conversation. Recordings can be created using a text file stored in the voice file converter.

iii) After creating the transcript, the message types are classified according to the contents of the conversation message, and the contents are divided into unit tasks according to the contents of the entire task and the conversation and stored in the transcript storage unit.

In the present invention, subdividing and storing the unit job can be stored as a unit job for one call between conversations. In addition, if necessary, a plurality of currencies may be treated as a unit job and stored as a currency related to a series of processes.

When the detailed steps of the first step are completed, the second to fourth steps are started, which will be described later with reference to FIGS. 5 to 7.

4 is a flow chart of a conversational quality measurement method for obtaining a final job similarity according to an embodiment of the present invention.

First, after receiving and storing the conversation message between the dialoguers, the transcript is recorded according to the unit job such as the dialogue person, the contents of the conversation, and the duration of the conversation.

By using the recordings classified according to the unit job, the conversation messages of each talker are separated into morphological units and merged by the talker (s402), by morphological classification by similarity calculation logic such as a cosine similarity calculation algorithm. In this case, the unit job similarity calculation process may be performed repeatedly, instead of once, to calculate a plurality of unit job similarity diagrams.

Thereafter, a plurality of proportional unit job similarities are obtained by dividing by the total number of morpheme words performed for each unit job similarity (s404), and the job similarity is obtained by summing all the plurality of proportional unit job similarities. )

Meanwhile, by using the recordings classified according to the unit duties, the conversation messages of the respective conversations are meaningfully expanded, and the semantic extended conversation messages are separated into morphological units (m406) and merged for each conversation (s406). The unit job similarity is calculated by means of semantic expansion (s407). At this time, the unit job similarity is calculated not only once but also by the semantic expansion. Can be.

Subsequently, for each unit job similarity due to each semantic expansion, the proportional unit job similarity due to a plurality of semantic expansions is obtained by dividing by the total number of words of the morpheme words performed in the semantic expansion analysis (s408), and the proportional unit by the plurality of semantic expansions. The final value of job similarity by semantic expansion is obtained by adding up all job similarities. (S409)

As described above, when job similarity by morpheme classification and job similarity by semantic expansion are obtained, the correction value is determined by the person in charge of the power plant or the manager of the cab (s410). Can be given as an approximation to 1, and can be given as an approximation to 0 in order to give more weight to the results of the semantic analysis.

When the correction value is determined as described above (S410), the final job similarity is calculated. (S411) The final job similarity can be obtained by Equation 2 below.

In the above formula, f is a correction value, and may be regarded as a value satisfying the condition of 0 <f <1.

5 is an exemplary view showing a sentence similarity evaluation process by morpheme classification according to an embodiment of the present invention.

The conversation quality measuring method of the present invention includes a second step of performing sentence similarity calculation by dividing the classified conversation message into morpheme units.

The second step may include: dividing the recordings into unit conversations of a talker, separating the conversation messages of each talker into morpheme units, merging the separated morphemes into each talker, and merging by each talker. It may be a step of calculating unit duty similarity by performing sentence similarity calculation using similarity calculation logic on the morpheme set.

In more detail, the second step is divided into steps as follows.

i) Prepare the transcripts prepared in the first step in the original form with the operator (conversator), and divide them by unit conversation of the conversation (see 510), and ii) morph the conversation messages of each operator from the transcripts divided by unit conversation. In this case, morphological separation refers to the smallest unit of linguistic morphological level that imparts a function of meaning in constructing the contents of a sentence. iii) merge the separated morphemes for each talker. In other words, the dialogue contents for each dialogue unit are merged in the dialogue contents indicating unit tasks (see 530). Iv) The sentence similarity calculation is performed using the similarity calculation logic on the morpheme set merged for each dialogue group. In the present invention, the similarity calculation may be performed using a cosine similarity calculation algorithm by extracting a feature vector from the set of separated morphemes.

When the above calculation process is performed, unit job similarity is calculated by morphological classification, which is not limited to one-time operation, and it is possible to calculate a plurality of unit job similarities repeatedly without limitation, and these unit job similarities are summed. It can be calculated as proportional unit job similarity, and finally used to generate final job similarity.

6 is an exemplary view showing a sentence similarity evaluation process by the semantic expansion analysis according to an embodiment of the present invention.

The conversation quality measuring method of the present invention includes a third step of performing semantic expansion on a classified conversation message and then calculating sentence similarity by separating the semantic extended conversation message into morpheme units.

The third step may include: dividing the recording into unit conversations of a talker, performing semantic expansion on the conversation message of each talker, separating the conversation messages of each talker-expanded talker into morpheme units, Merging the separated morphemes for each talker and calculating sentence similarity using sentence similarity calculation logic for the set of morphemes merged for each talker to calculate unit job similarity due to semester expansion. Conversation quality measurement method.

In the present invention, the semantic expansion of the sentence means a method of evaluating the similarity of the sentences by expanding the morphemes according to the actual conversation contents.

Semantic expansion refers to a method of estimating that meaning has been delivered by extending the meaning of the original conversation, and includes synonym expansion and sentence receiving slogan expansion.

For example, (i) when dialog 1 instructs' lock valve 30 times', and dialog 2 responds with 'yes, ok', the similarity evaluation step by the morphological analysis of the second step stems from 'valve 30, Morphological separation, like 'lock' and 'yes'.

(ii) On the other hand, in the similarity evaluation step by means of semantic expansion, the answer of dialog 2 is 'Yes, I know' to the instruction of 'lock valve 30 times' assuming that it understands and accepts the instructions of dialog 1. Means that the valve will be locked 30 times, and its meaning is extensible, and it is an example of meaning expansion and morphological analysis by dividing it into 'valve 30, lock' and 'valve 30, lock'. have.

In more detail, the third step may be classified as follows.

i) Prepare the transcripts prepared in the first step in the text form with the operator (conversator), divide them by operator conversation (see 610), and ii) Extend the meaning of the conversation messages of each operator of the recording divided by unit conversation. (See 620) and iii) separate the conversation message of each driver whose meaning has been extended (see 620). Iv) Through the above process, the separated morphemes are merged for each talker. In other words, the conversation contents indicating unit tasks are merged for each conversation user (see 630). V) The sentence similarity calculation is performed using the similarity calculation logic on the semantic set merged for each conversation person. In the present invention, the similarity calculation may be performed using a cosine similarity calculation algorithm by extracting a feature vector from the set of separated morphemes.

When the above calculation process is performed, the unit job similarity due to the semantic expansion is calculated, which is not limited to one-time, and it is possible to repeatedly calculate the unit job similarity due to the plurality of semantic expansions. Unit job responsiveness by can be used to calculate the final job responsiveness later.

7 is an exemplary view showing an evaluation process for calculating the final job similarity according to an embodiment of the present invention.

The conversation quality measuring method according to the present invention calculates the similarity evaluation result by applying a predetermined correction value to the similarity evaluation result by morpheme classification and the similarity evaluation result by semantic expansion, and the unit job similarity evaluation result is synthesized and the final job A fourth step of performing similarity assessment.

In more detail, the fourth step includes: i) arranging the unit job similarities calculated through the sentence similarity calculation of the second step, and calculating the total number of words of the morpheme words performed in the analysis; Calculating the proportional unit duty similarity for the whole job by dividing the unit duty similarity by the total number of words, iii) summing the values of the proportional unit duty similarity calculated from each unit duty similarity, Determining job similarity, iv) arranging the unit job similarities calculated through the sentence similarity calculation of the third step, and obtaining the total number of words of the morpheme words performed in the semantical expansion analysis, and v) in the semantic expansion. Each unit job similarity by dividing by the total number of words of the morpheme words performed in the semantic expansion analysis, the proportional unit duty by each semantic expansion Calculating a degree, vi) deciding a job likelihood by means of semantical expansion by summing values of proportional unit duty similarities by means of each semantic expansion calculated from the unit duty similarity by means of each semantic expansion, vii) Determining a predetermined correction value and viii) calculating a final job similarity by applying the correction value.

The correction value is a value between 0 and 1, which can be determined by the main administrator of the conversational quality measurement system according to the necessity of the invention. If you want to give more weight to the results of the analysis, you can give a value close to zero.

In addition, the final job similarity may be calculated by Equation 3 below.

In the above formula, f is a correction value, and may be regarded as a value satisfying the condition of 0 <f <1.

Referring to FIG. 7, the conversation messages are classified and sorted according to the unit job (see 710) (see 720), and the conversation messages according to the classified unit job are calculated for sentence similarity by morphological classification (see 730). ), Each unit job similarity is obtained from the dialogue message according to each unit job, and each proportional unit job similarity is obtained by dividing this by the total number of morpheme words used in the analysis.

The plurality of proportional unit job likelihoods are added together to determine a job likelihood value based on morphological classification.

On the other hand, the sentence similarity is also calculated by means of the expansion of the semantics (see 740), and the semantic expansion is performed on the dialogue messages according to the unit jobs, and the unit job similarity by the semantic expansion is obtained by classifying the morphemes. You lose. By dividing this by the total number of words of morpheme words used in the expansion of semantics, the proportional unit job similarity by each semantic expansion is obtained.

The proportional unit job similarity due to the plurality of semantic expansions is added together to determine the job similarity value due to the semantic expansion.

After that, the principal manager determines a correction value suitable for an industrial site, and uses the correction value and Equation 2 to obtain a final job similarity.

Referring to FIG. 5 and FIG. 6, as shown in the sentence similarity calculation results for the unit tasks of the second and third stages, the dialogue similarity evaluation result by the morphological analysis of the second stage for the same conversation content is 49.07. %, And the dialogue similarity evaluation result by the semantic expansion of the third stage is 78.26%, which is different from each other.

Therefore, in summary, the final job similarity in the fourth step is i) correcting and statistically processing two unit job similarity evaluation results at a predetermined ratio in consideration of correction values, and ii) distributing the evaluation results of these unit jobs to all jobs. This may be referred to as a step of performing statistical processing.

The present invention has been described above in connection with specific embodiments of the present invention, but this is only an example and the present invention is not limited thereto. Those skilled in the art can change or modify the described embodiments without departing from the scope of the present invention, and within the equivalent scope of the technical spirit of the present invention and the claims to be described below. Various modifications and variations are possible.

1 is a block diagram of a conversational quality measurement system according to an embodiment of the present invention.

2 is an exemplary view illustrating a process of calculating sentence similarity using a cosine similarity calculation method according to an embodiment of the present invention.

3 is a flow chart of a conversational quality measuring method according to an embodiment of the present invention.

4 is a flow chart of a conversational quality measurement method for obtaining a final job similarity diagram according to an embodiment of the present invention.

5 is an exemplary view showing a sentence similarity evaluation process by morpheme classification according to an embodiment of the present invention.

6 is an exemplary view showing a sentence similarity evaluation process by the semantic expansion analysis according to an embodiment of the present invention.

Figure 7 is an exemplary view showing an evaluation process for calculating the final job similarity according to an embodiment of the present invention.

{Description of major symbols in the drawing}

101: chat terminal

102: dialogue quality measurement module

103: voice receiver

104: conversation message storage

105: voice file conversion unit

106: Conversation Quality Evaluation Department

107: display unit

108: recording storage unit

109: morphological evaluation

110: Semantic Evaluation Department

111: final similarity evaluation unit

Claims (11)

In the dialogue quality measurement system having a plurality of dialogue terminals and dialogue quality measuring module, The dialogue quality measurement module, A voice receiver connected to the talker terminal by wire or wirelessly and receiving a chat message through a plurality of talker terminals; A chat message storage unit for storing a chat message received in the voice receiver as a voice file; A voice file conversion unit for converting and displaying the voice file stored in the conversation message storage unit into a text file, and storing the converted text file; A conversation quality evaluation unit for creating and storing a recording using a text file stored in the voice file converting unit and measuring conversation quality using the recording and sentence similarity calculation logic; And A display unit connected to the voice file converting unit and the conversation quality evaluating unit to display the text file, the recording, the similarity evaluation process, and the similarity evaluation result on a terminal screen; Including, The conversation quality evaluation unit, A recording storage unit for creating a recording using a text file stored in the voice file converting unit, and classifying and storing the recording according to conversation time, dialog, message type, or message content; A morpheme evaluation unit for dividing the conversation message of the transcript into morpheme units, merging for each talker, and performing sentence similarity calculation using similarity calculation logic on the morpheme sets merged for each talker; And The semantic expansion is performed on the conversation message of the transcript, the semantic extended conversation message is divided into morpheme units, merged by each talker, and the similarity calculation logic for the merged morpheme set by the talker of the extended chat message. A semantic evaluation unit to perform sentence similarity calculation using a; Conversation quality measurement system comprising a. According to claim 1, The conversation quality evaluation unit, The similarity evaluation result of the similarity evaluation result of the morpheme evaluation unit and the semantic evaluation unit is calculated by applying a predetermined correction value, and the final similarity degree of the final job similarity evaluation result is calculated by combining the unit job similarity evaluation results according to the dialogue content classification. Evaluation unit; Conversation quality measurement system, characterized in that it further comprises. The method of claim 1, The talker terminal is any one selected from a mobile phone, a computer, a notebook, an electronic notebook, a PDA (Personal Digital Assistance) or a radio. The system of claim 1, wherein the conversation message is a conversation message transmitted or received between conversations using a plurality of terminals at a factory, a power plant, a fire station, a police station, a hospital, a government office, or a construction site. In the method of measuring the chat quality using the chat quality measuring system, Receiving a conversation message transmitted or received between a plurality of conversation terminal terminals to create a recording using the conversation message, classify the conversation message type according to the conversation time, the conversation and the contents of the message, subdivided into unit duties and stored; Stage 1; A second step of dividing the classified conversation message into morpheme units to perform sentence similarity calculation; Performing semantic expansion on the classified conversation message, and performing sentence similarity calculation by separating the semantic extended conversation message into morpheme units; And A fourth step of combining the similarity evaluation result by the morpheme classification and the similarity evaluation result by the semantic expansion, and calculating a final job similarity by applying a predetermined correction value; Conversation quality measurement method comprising a. The method of claim 5, wherein the second step, Dividing the transcripts into unit conversations of a dialog; Dividing the conversation message of each talker into morpheme units; Merging the separated morphemes for each talker; And Calculating a unit job similarity evaluation result by performing sentence similarity calculation using similarity calculation logic on the morpheme sets merged for each conversation person; Conversation quality measurement method, characterized in that. The method of claim 5, wherein the third step, Dividing the transcripts into unit conversations of a talker; Performing semantic expansion on the conversation message of each talker; Dividing the conversation message of each of the semantic extensions into morpheme units; Merging the separated morphemes for each talker; And Calculating a unit job similarity evaluation result by means of semantic expansion by performing sentence similarity calculation using a similarity calculation logic on the set of morphemes merged for each talker; Conversation quality measurement method, characterized in that. The method according to claim 6 or 7, And repeatedly performing the dialogue quality measuring method to calculate a plurality of unit job similarity evaluation results. The method according to claim 6 or 7, The similarity calculation logic is a conversation quality measurement method, characterized in that for calculating the similarity between sentences using a cosine similarity (Cosine Similarity) calculation algorithm by extracting a feature vector from the voice of each conversation or the voice-converted characters. The method of claim 5, wherein the fourth step, Arranging the unit job similarities calculated through the sentence similarity degree calculation of the second step, and obtaining the total number of words of the morpheme words performed in the analysis; Calculating a proportional unit job likelihood ratio for all jobs by dividing each unit job likelihood by a total number of words; Determining a job similarity by morphological analysis by summing values of respective proportional unit job similarities calculated from each unit job similarity diagram; Arranging the unit job similarities calculated through the sentence similarity degree calculation in the third step, and obtaining the total number of words of the morpheme words performed in the semantic expansion analysis; Calculating proportional unit job similarity by each semantic expansion by dividing each unit job similarity by the semantic expansion by the total number of words of the morpheme words performed in the semantic expansion analysis; Deciding a job similarity by means of semantically expanding by summing values of proportional unit duty similarities by means of each semantic expansion calculated from the unit duty similarity by means of each semantic expansion; Determining a predetermined correction value; And Calculating a final job similarity by applying the correction value; Conversation quality measurement method comprising a. The method of claim 10, The final job similarity is expressed by the following equation,

And generating a result, and f is a correction value, and satisfies a condition of 0 <f <1 .

Images