Intelligent auxiliary method and system suitable for call center manual service agent

Abstract

The invention relates to the technical field of natural language processing, in particular to an intelligent assistance method and system suitable for a call center manual service agent, and the method comprises the steps: obtaining a target dialogue text; recording the segmented words in the target dialogue text after the stop words are removed as target segmented words; obtaining all co-occurrence words of each target segmented word according to stop word features and emotion intensity degrees of statements where each target segmented word is located; according to the co-occurrence frequency of each target segmented word and each co-occurrence word and the occurrence frequency of each co-occurrence word, obtaining an associated word of each target segmented word; according to the associated word similarity degree and the pinyin similarity degree between the target segmented words, dividing all the target segmented words into a plurality of entity sets, then obtaining entity vocabularies of each entity set, and then replacing all the target segmented words with the entity vocabularies corresponding to the entity sets where the target segmented words are located. According to the method, the co-occurrence word of each target segmented word is accurately obtained, so that the entity vocabulary corresponding to each target segmented word is accurately obtained, and the error correction accuracy of the dialogue text is improved.

Description

Technical Field

[0001] This application relates to the field of natural language processing technology, specifically to intelligent assistance methods and systems applicable to human agents in call centers. Background Technology

[0002] In multi-turn dialogue scenarios, the call center intelligent assistance system automatically performs speech recognition, generates dialogue transcripts, avoids the problem of human agents forgetting information in multi-turn dialogues, and quickly extracts key information from customer inquiries, providing human agents with accurate knowledge base recommendations.

[0003] The accuracy of speech recognition in call center intelligent assistance systems is affected by factors such as the speaker and the environment. Furthermore, the error rate for proper nouns, homophones, and uncommon words is high during speech-to-text conversion, resulting in erroneous information in the dialogue transcripts obtained by the system and thus lower accuracy in providing knowledge assistance to human agents. Traditional language models, through data-driven methods, can automatically correct words in dialogue transcripts. However, call center dialogue transcripts are often highly colloquial, and nested entities are easily truncated by stop words or emotional words. Existing language models lack long-distance dependencies and struggle to capture features between nested entity words across long lexical intervals, resulting in low accuracy during error correction. Summary of the Invention

[0004] To address the aforementioned technical problems, the purpose of this application is to provide an intelligent assistance method and system suitable for call center human agents, and the specific technical solution adopted is as follows: In a first aspect, embodiments of this application provide an intelligent assistance method suitable for call center human agents, the method comprising the following steps: Obtain the target dialogue text during the conversation between the human agent and the customer; Based on the number of stop words in each sentence of the target dialogue text and the positional differences of adjacent stop words, the redundancy feature value of each sentence is obtained. Combined with the intensity of emotion in each sentence, the spoken evaluation value of each sentence is obtained, and then the co-occurrence interval length of each sentence is obtained. The segmented words in the target dialogue text after removing stop words are recorded as target segments. All co-occurring words of each target segment are obtained based on the co-occurrence interval length of the sentences in which each target segment is located. Based on the co-occurrence frequency of each target word and its co-occurring words, and the occurrence frequency of each co-occurring word in the target dialogue text, the related word filtering value between each target word and its co-occurring words is obtained, thereby obtaining the related words of each target word; based on the number of identical related words in the related words of any two target words, and the similarity of their pinyin, all target words in the target dialogue text are divided into multiple entity sets; based on the average level of the spoken evaluation value of all sentences containing each target word in each entity set, the entity vocabulary of each entity set is obtained, and then all target words in the target dialogue text are replaced with the entity vocabulary corresponding to their respective entity sets.

[0005] Preferably, the formula for calculating the redundancy feature value of each statement is: In the formula, The redundant feature value of the j-th statement. Let j be the number of stop words in the j-th statement. Let be the total number of all words in the j-th statement. and These represent the position values ​​of the i-th and (i-1)-th stop words in the j-th statement, respectively.

[0006] Preferably, the formula for calculating the spoken assessment value of each statement is as follows: In the formula, Let j be the spoken evaluation value of the j-th statement. The redundant feature value of the j-th statement. Let be the sentiment intensity value of the j-th statement, and Norm() be the normalization function. The process of obtaining the sentiment intensity value of each statement is as follows: obtain the sentiment category and category intensity of all words in the target dialogue text; take the maximum value of the maximum category intensity of all sentiment categories corresponding to all words of each statement as the sentiment intensity value of each statement.

[0007] Preferably, the formula for calculating the co-occurrence interval length of each statement is: In the formula, Let j be the co-occurrence interval length of the j-th statement. It is the preset initial interval length. It is the floor function. Let j be the spoken evaluation value of the j-th statement. These are preset correction parameters.

[0008] Preferably, all co-occurring words of each target word refer to all target words in the sentence in which each target word is located, whose distance from the target word is no greater than the co-occurrence interval length of the sentence in which it is located.

[0009] Preferably, the formula for calculating the correlation word filtering value between each target word segment and its co-occurring words is as follows: In the formula, Filter the associated words between the target word segment x and its co-occurring word y. The co-occurrence frequency of the target word segment x and its co-occurring word y. It is the frequency of co-occurrence of word y in the target dialogue text; where the co-occurrence frequency of target word x and its co-occurrence word y is the ratio of the number of times target word x and its co-occurrence word y co-occur in the target dialogue text to the total number of times target word x appears.

[0010] Preferably, the associated words of each target word refer to the co-occurring words among all co-occurring words of each target word whose associated word screening value is greater than a preset segmentation threshold.

[0011] Preferably, the specific process of dividing all target words in the target dialogue text into multiple entity sets is as follows: The set of all related words for each target word is denoted as the related word set for each target word; the pronunciation string of each target word is obtained. If the number of related words in the intersection of the related word sets of any two target word segments is greater than a preset first threshold, then the two target word segments satisfy the first judgment condition; if the edit distance between the pronunciation strings of any two target word segments is less than a preset second threshold, then the two target word segments satisfy the second judgment condition. If any two target word segments simultaneously satisfy the first and second judgment conditions, then the two target word segments are determined to be related to the same entity, and these two target word segments are assigned to the same entity set; all target word segments in the target dialogue text that are related to the same entity are recorded as an entity set, thus obtaining the entity set of all targets in the target dialogue text.

[0012] Preferably, the process of obtaining the entity vocabulary of each entity set is as follows: the average of the spoken evaluation values ​​of all sentences containing each target word is used as the recognition disturbance value of each target word; the target word with the smallest recognition disturbance value in each entity set is used as the entity vocabulary of each entity set.

[0013] Secondly, embodiments of this application also provide an intelligent assistance system suitable for call center human agents, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the intelligent assistance method suitable for call center human agents described above.

[0014] This application has at least the following beneficial effects: This application constructs a spoken language evaluation value by analyzing the distribution features of stop words and the emotional level features in each sentence, and then adaptively adjusts the co-occurrence interval length of each sentence. It gives a larger co-occurrence interval length to sentences with strong emotions and significant stop word redundancy, effectively avoiding obvious spoken language representation features in the dialogue process and improving the accuracy of co-occurrence word acquisition. For sentences with insignificant spoken language features, the co-occurrence interval length is reduced to avoid unnecessary computational complexity due to excessively large co-occurrence interval length. This solves the problem that traditional language models are insufficient in long-distance dependencies and have difficulty capturing the lexical features of nested entities in sentences across long-distance lexical intervals. This application determines whether two target word segments belong to the same entity set based on the similarity of related word sets and the edit distance of pronunciation, and then obtains the entity vocabulary of each entity set. This allows for error correction of highly colloquial recorded text, effectively solving the problem of speech transcription errors caused by significant colloquial features in dialogue recorded text, improving the accuracy of dialogue recorded text, avoiding errors in human agents' retrieval due to recognition errors, and providing more accurate dialogue information for human agents. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating the steps of an intelligent assistance method for call center human agents, provided as an embodiment of this application; Figure 2 This is a flowchart illustrating the process of obtaining entity vocabulary for each entity set provided in one embodiment of this application. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by this application to achieve the intended purpose of the invention, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the intelligent assistance method and system for call center human agents proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0019] The following description, in conjunction with the accompanying drawings, details the specific solutions provided in this application for the intelligent assistance method and system applicable to human agents in call centers.

[0020] Please see Figure 1 The diagram illustrates a flowchart of an intelligent assistance method for call center human agents according to an embodiment of this application. The method includes the following steps: Step 1: Obtain the target dialogue text during the conversation between the human agent and the customer.

[0021] The call center microphone simultaneously captures the voices of customers and human agents. The call center intelligent auxiliary system first filters out background noise through noise reduction processing, and then uses the voice recognition function to obtain the dialogue record text in real time through voice recognition between customers and human agents.

[0022] During the dialogue between the customer and the live agent, a preset time interval is designated as an error correction moment. All dialogue records from the start of the dialogue to the current error correction moment are recorded as the target dialogue text for that moment. Error correction is performed on the corresponding target dialogue text at each correction moment. The preset time interval ranges from 10s to 60s; in this embodiment, 20s is used. The target dialogue text is segmented using the jieba Chinese word segmentation library. Specifically, after importing the jieba Chinese word segmentation library, the jieba.cut function is used for word segmentation. The HMM (Hidden Markov Model) parameter of the jieba.cut function is set to the default value True, and the cut_all parameter is set to False to determine whether to use the full mode (the cut_all parameter defaults to False; cut_all=False indicates the use of jieba's precise segmentation mode). After jieba segmentation, all the segmented words in the target dialogue text are obtained.

[0023] Step 2: Based on the number of stop words in each sentence of the target dialogue text and the positional differences of adjacent stop words, obtain the redundancy feature value of each sentence, and combine it with the emotional intensity of each sentence to obtain the spoken evaluation value of each sentence, and then obtain the co-occurrence interval length of each sentence; record the segmented words in the target dialogue text after removing stop words as target segmented words; obtain all co-occurring words of each target segmented word based on the co-occurrence interval length of the sentence in which each target segmented word is located.

[0024] The accuracy of speech recognition in call center intelligent assistance systems is affected by factors such as the speaker and the environment. When call center agents provide services, customers often use highly colloquial language, and factors such as their emotions, accents, elision, and unclear pronunciation can all lead to transcription errors. Furthermore, call center applications are often in industries such as product after-sales service, insurance consultation, and e-commerce, which have domain-specific vocabulary. Some of these proprietary terms are nested entities with long word combinations, making it difficult for customers to express themselves clearly and completely. This results in fragmented speech, severely interfering with the speech recognition performance of the call center intelligent assistance system.

[0025] The redundancy of each sentence can be determined by using stop words in the sentences. Therefore, the Harbin Institute of Technology stop word list is used to mark whether all word segments in the target dialogue text are stop words.

[0026] In a preferred embodiment, the redundancy feature value of each statement is obtained based on the number of stop words in each statement of the target dialogue text and the difference in the order of adjacent stop words, which is used to characterize the redundancy of each statement.

[0027] In this embodiment, the redundant feature value of the j-th statement is denoted as Its specific expression is: In the formula, The redundant feature value of the j-th statement. Let j be the number of stop words in the j-th statement. Let be the total number of all words in the j-th statement. and Let represent the positional values ​​of the ith and (i-1)th stop words in the j-th statement, respectively. It should be noted that when the j-th statement contains only one stop word, let the redundancy feature value of the j-th statement be . .

[0028] in, This value reflects the proportion of stop words in the j-th statement. The larger the value, the more stop words there are in the statement, and the more redundant the statement is. This value reflects the distribution characteristics of stop words in the j-th statement. The larger the value, the more concentrated the stop words are in the whole statement. The more likely the redundant stop words appearing in a concentrated manner will cover the nested entities corresponding to the proper nouns, causing the nested entities to become fragmented. In this case, the redundancy of the statement is greater.

[0029] Further, during the artificial seat service process, the customer conversation may contain more emotional words, which often intersperse between sentence entities. The stronger the emotion, the more likely it is to have a stronger impact on speech recognition, but emotional words are often not in the stop list. In this application, all word segments of the target dialogue text are used as the input of the sentence matching algorithm based on the Chinese Sentiment Lexical Ontology. Among them, the data label categories in the Chinese Sentiment Lexical Ontology are respectively joy, goodness, anger, sorrow, fear, disgust, and surprise. The category intensity under each emotion category can be specifically divided into five levels: 1, 3, 5, 7, and 9, where 9 represents the maximum intensity and 1 represents the minimum intensity. The emotion category and category intensity of all word segments in the target dialogue text are obtained. The maximum value among the maximum values of the category intensities under all emotion categories corresponding to all word segments of each sentence is used as the emotion intensity value of each sentence.

[0030] As a preferred implementation manner, according to the redundancy feature value of each sentence and the degree of emotion intensity in each sentence, the oral evaluation value of each sentence is obtained to represent the degree of oralization of each sentence.

[0031] In this embodiment, the oral evaluation value of the j-th sentence is denoted as , and its specific expression is: ; In the formula, is the oral evaluation value of the j-th sentence, is the redundancy feature value of the j-th sentence, is the emotion intensity value of the j-th sentence, and Norm( ) is a normalization function.

[0032] The oral evaluation value reflects the degree of oralization of the j-th sentence. The larger the value, the higher the frequency of occurrence of stop words such as modal particles and pause words in the sentence, and the stronger the emotion. In the scenario where the sentence is truncated by emotion modifiers and stop words, it is necessary to try to expand the neighborhood length when searching for co-occurring words later to avoid the problem that there are no co-occurring words in a smaller search range.

[0033] Further, stop word removal processing is performed on the target dialogue text, and all word segments after removing stop words are denoted as target word segments. During the process of converting speech to text, due to the strong oralization of speech, some target word segments in the converted text may be incorrect. For example, a certain target word segment in the target dialogue text is "share", but in fact, the correct word in the speech is "binning", so the entity word of "share" is "binning". In order to correct the incorrect target word segments in the target dialogue text into correct words, the entity words of each target word segment can be obtained according to the similarity of co-occurring words and the similarity of pronunciation of each target word segment.

[0034] First, in order to accurately obtain the co-occurring words of all target words in the target dialogue text, as a preferred implementation, the co-occurrence interval length of each sentence is obtained based on the spoken evaluation value of each sentence, which is used to characterize the farthest interval length between each target word and its co-occurring words.

[0035] In this embodiment, the co-occurrence interval length of the j-th statement is denoted as... Its expression is: In the formula, Let j be the co-occurrence interval length of the j-th statement. This is a preset initial interval length; in this embodiment, it is set to 3. It is the floor function. Let j be the spoken evaluation value of the j-th statement. This is a preset correction parameter, which is set to 0.5 in this embodiment.

[0036] like When the value is greater than 1, a longer co-occurrence interval is given to dialogue sentences with strong emotions and significant stop word redundancy. This effectively avoids the impact of obvious spoken language representation features on speech recognition accuracy during dialogue, expands the search length of co-occurrence words, and is beneficial for mining the sentence features of word segmentation and accurately obtaining the co-occurrence words of each target word. If When the value is less than 1, the computational complexity is not increased due to excessively large search lengths.

[0037] Furthermore, taking the target segment x in the target dialogue text as an example, all target segment words in the sentence containing the target segment x whose distance from the target segment word is no greater than the co-occurrence interval length of the sentence containing the target segment word are considered as co-occurring words of the target segment word x.

[0038] Similarly, obtain the co-occurring words of all target segmented words in the target dialogue text.

[0039] Step 3: Based on the co-occurrence frequency of each target word and its co-occurring words, and the occurrence frequency of each co-occurring word in the target dialogue text, obtain the related word filtering value between each target word and its co-occurring words, thereby obtaining the related words of each target word; based on the number of identical related words in the related words of any two target words, and the similarity of the pinyin, divide all target words in the target dialogue text into multiple entity sets; based on the average level of the spoken evaluation value of all sentences containing each target word in each entity set, obtain the entity vocabulary of each entity set, and then replace all target words in the target dialogue text with the entity vocabulary corresponding to their respective entity sets.

[0040] Furthermore, based on the co-occurrence frequency of each target word and its co-occurring words, and the occurrence frequency of each co-occurring word in the target dialogue text, the associated word filtering value of each target word and its co-occurring words is obtained, which is used to characterize the degree of association between each target word and its co-occurring words. The ratio of the number of times target word x co-occurs with its co-occurring word y in the target dialogue text to the total number of times target word x appears is taken as the co-occurrence frequency of target word x and its co-occurring word y, denoted as . .

[0041] In this embodiment, the correlation word filtering value between the target word segment x and its co-occurring word y is denoted as Its specific expression is: In the formula, Filter the associated words between the target word segment x and its co-occurring word y. The co-occurrence frequency of the target word segment x and its co-occurring word y. It represents the frequency of co-occurrence of word y in the target dialogue text.

[0042] calculate The purpose is to prevent the co-occurring word y from being easily misjudged as a related word of the target segmentation word x when it is a common word. The larger the value, the more significant the co-occurrence characteristic between the target word segment x and the co-occurring word y, and the lower the probability that the co-occurring word is a common word. Therefore, the target word segment x and the co-occurring word y are more likely to belong to nested entities in a call center scenario. The co-occurring word y plays an important role in text correction for the target word segment x, and should therefore be used as a word segmentation related word for the target word segment x. Related word selection value The larger.

[0043] Furthermore, the correlation word filtering value between each target word and all its co-occurring words is used as the input of the Otsu's maximum inter-class variance algorithm to obtain the segmentation threshold, which is denoted as the preset segmentation threshold. Co-occurring words with correlation word filtering values ​​greater than the preset segmentation threshold are used as the correlation words of each target word. The set of all correlation words of each target word is denoted as the correlation word set of each target word.

[0044] During the speech-to-text process in call centers, the conversation transcripts are often influenced by the customer's emotions, accent, elision, and unclear pronunciation. This results in the transcripts having strong emotional content, redundant stop words, and a pronounced colloquial style, which can easily interfere with the speech recognition process. Consequently, some speech may be erroneously transcribed, and the same nested entity may be incorrectly identified as multiple word combinations with different meanings and similar pronunciations.

[0045] Chinese Pinyin consists of initials, finals, and tones. In call center customer service scenarios, customers' speech tends to be more conversational and they are not sensitive to pitch variations. Therefore, this application takes each target word segment as input and uses the pypinyin library in the Python programming language, with the Pinyin style set to NORMAL (normal style, consisting of initials and finals, without tones), to output the pronunciation strings of each target word segment.

[0046] Furthermore, if the number of related words in the intersection of the related word sets of any two target word segments is greater than a preset first threshold, then the two target word segments are more likely to be words in the same nested entity, and the two target word segments satisfy the first judgment condition; wherein, the preset first threshold ranges from [2,4], and is set to 2 in this embodiment. If the edit distance between the pronunciation strings of any two target word segments is less than a preset second threshold, then the two target word segments are more likely to be homophones, and the two target word segments satisfy the second judgment condition; wherein, the preset second threshold ranges from [2,5], and is set to 3 in this embodiment.

[0047] If any two target word segments simultaneously satisfy both the first and second judgment conditions, then the two target word segments are determined to have a same entity relationship, and thus belong to the same entity set. It should be noted that: if any two target word segments a and b have a same entity relationship, and target word x also has a same entity relationship with target word a or target word b, then target word x has a same entity relationship with both target word a and target word b, and target word a, b, and x all belong to the same entity set.

[0048] Calculate whether any two target words in the target dialogue text conform to the same entity relationship. Record all target words in the target dialogue text that conform to the same entity relationship as an entity set, and obtain all entity sets in the target dialogue text.

[0049] For example, in the after-sales field of call center products, the customer's actual dialogue is "the fuse in the box". During the speech recognition transcription process, due to the heavy colloquialism, "box" is converted into "box", "fly to", "share" etc. in the dialogue record text. Then, the words "box", "fly to", "share" etc. form an entity set used for text correction, and the corresponding entity word in this entity set is "box".

[0050] The average spoken evaluation value of all sentences containing the target word x is used as the recognition perturbation value of the target word x, reflecting the degree of interference to the target word x during speech recognition transcription. The smaller the recognition perturbation value of the target word, the weaker the interference from factors such as emotion, accent, elision, and unclear pronunciation during speech recognition transcription, the higher the accuracy of speech recognition transcription, and the higher the confidence of the transcribed target word, enabling error correction in call center scenarios.

[0051] The target word with the smallest perturbation value in each entity set is taken as the entity vocabulary of each entity set, and all target words in the target dialogue text are replaced with the entity vocabulary corresponding to their respective entity set. The flowchart for obtaining the entity vocabulary of each entity set is as follows. Figure 2 As shown, this allows for text correction of some incorrectly identified entity words in the target dialogue text during multi-turn conversations in call centers, improving the accuracy of the target dialogue text and avoiding errors in retrieval by human agents due to identification errors, thus providing accurate and intelligent assistance to human agents.

[0052] Based on the same inventive concept as the above method, this application also provides an intelligent assistance system suitable for call center human agents, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the above-described intelligent assistance method suitable for call center human agents.

[0053] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments of this specification have been described above. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0054] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0055] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. An intelligent assistance method suitable for human agents in call centers, characterized in that, The method includes the following steps: Obtain the target dialogue text during the conversation between the human agent and the customer; Based on the number of stop words in each sentence of the target dialogue text and the positional differences of adjacent stop words, the redundancy feature value of each sentence is obtained. Combined with the intensity of emotion in each sentence, the spoken evaluation value of each sentence is obtained, and then the co-occurrence interval length of each sentence is obtained. The segmented words in the target dialogue text after removing stop words are recorded as target segments. All co-occurring words of each target segment are obtained based on the co-occurrence interval length of the sentences in which each target segment is located. Based on the co-occurrence frequency of each target word and its co-occurring words, and the occurrence frequency of each co-occurring word in the target dialogue text, the related word filtering value between each target word and its co-occurring words is obtained, thereby obtaining the related words of each target word; based on the number of identical related words in the related words of any two target words, and the similarity of their pinyin, all target words in the target dialogue text are divided into multiple entity sets; based on the average level of the spoken evaluation value of all sentences containing each target word in each entity set, the entity vocabulary of each entity set is obtained, and then all target words in the target dialogue text are replaced with the entity vocabulary corresponding to their respective entity sets.

2. The intelligent assistance method for call center human agents as described in claim 1, characterized in that, The formula for calculating the redundancy feature value of each statement is as follows: In the formula, The redundant feature value of the j-th statement. Let j be the number of stop words in the j-th statement. Let be the total number of all words in the j-th statement. and These represent the position values ​​of the i-th and (i-1)-th stop words in the j-th statement, respectively.

3. The intelligent assistance method for call center human agents as described in claim 1, characterized in that, The formula for calculating the oral assessment score of each statement is as follows: In the formula, Let j be the spoken evaluation value of the j-th statement. The redundant feature value of the j-th statement. Let be the sentiment intensity value of the j-th statement, and Norm() be the normalization function. The process of obtaining the sentiment intensity value of each statement is as follows: obtain the sentiment category and category intensity of all words in the target dialogue text; take the maximum value of the maximum category intensity of all sentiment categories corresponding to all words of each statement as the sentiment intensity value of each statement.

4. The intelligent assistance method for call center human agents as described in claim 1, characterized in that, The formula for calculating the co-occurrence interval length of each statement is as follows: In the formula, Let j be the co-occurrence interval length of the j-th statement. It is the preset initial interval length. It is the floor function. Let j be the spoken evaluation value of the j-th statement. These are preset correction parameters.

5. The intelligent assistance method for call center human agents as described in claim 1, characterized in that, All co-occurring words of each target word refer to all target words within the sentence in which each target word is located, whose distance from the target word is no greater than the co-occurrence interval length of the sentence in which it is located.

6. The intelligent assistance method for call center human agents as described in claim 1, characterized in that, The formula for calculating the correlation word filtering value between each target word segment and its co-occurring words is as follows: In the formula, Filter the associated words between the target word segment x and its co-occurring word y. The co-occurrence frequency of the target word segment x and its co-occurring word y. It is the frequency of co-occurrence of word y in the target dialogue text; where the co-occurrence frequency of target word x and its co-occurrence word y is the ratio of the number of times target word x and its co-occurrence word y co-occur in the target dialogue text to the total number of times target word x appears.

7. The intelligent assistance method for call center human agents as described in claim 1, characterized in that, The associated words of each target word segment refer to the co-occurring words among all co-occurring words of each target word segment whose associated word filtering value is greater than the preset segmentation threshold.

8. The intelligent assistance method for call center human agents as described in claim 1, characterized in that, The specific process of dividing all target words in the target dialogue text into multiple entity sets is as follows: The set of all related words for each target word is denoted as the related word set for each target word; the pronunciation string of each target word is obtained. If the number of related words in the intersection of the related word sets of any two target word segments is greater than a preset first threshold, then the two target word segments satisfy the first judgment condition; if the edit distance between the pronunciation strings of any two target word segments is less than a preset second threshold, then the two target word segments satisfy the second judgment condition. If any two target word segments simultaneously satisfy the first and second judgment conditions, then the two target word segments are determined to be related to the same entity, and these two target word segments are assigned to the same entity set; all target word segments in the target dialogue text that are related to the same entity are recorded as an entity set, thus obtaining the entity set of all targets in the target dialogue text.

9. The intelligent assistance method for call center human agents as described in claim 1, characterized in that, The process of obtaining the entity vocabulary of each entity set is as follows: the average of the spoken evaluation values ​​of all sentences containing each target word is used as the recognition disturbance value of each target word; the target word with the smallest recognition disturbance value in each entity set is used as the entity vocabulary of each entity set.

10. An intelligent auxiliary system for call center human agents, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent assistance method for call center human agents as described in any one of claims 1-9.

Images