Conversation assistance processing method and device

Abstract

The present disclosure provides a conversation assistance processing method, including: obtaining audio stream data of a real-time conversation with a user; converting the audio stream data into a conversation text in real time; detecting the conversation text based on a sensitive word detection algorithm to identify sensitive risk words appearing in the conversation text; determining a rescue speech set for the sensitive risk word in a preset rescue speech library when the sensitive risk word meets a preset warning standard; sending a conversation intervention request and the rescue speech set to an auxiliary conversation terminal to request the auxiliary conversation terminal to access the real-time conversation, and performing conversation assistance processing based on the rescue speech set. The present disclosure also provides a conversation assistance processing device, an electronic device, and a computer storage medium.

Description

Technical Field

[0001] The present disclosure relates to the field of big data technology, and in particular to a conversation assistance processing method and device. Background Art

[0002] With the rapid development of the financial industry and other industries, the forms of user services have diversified. As a common form of user service, outbound call conversations can provide efficient consulting services to a large number of users, as well as convenient outbound product promotion to a large number of users.

[0003] In the process of implementing the technical solution disclosed in the present invention, the inventors found that due to the high mobility of outbound call agents, during the real-time communication between outbound call agents and users, problems such as unsatisfactory conversation results and high user complaint rates often occur due to the outbound call agents' lack of professional knowledge or insufficient processing experience. Summary of the invention

[0004] One aspect of the present disclosure provides a conversation assistance processing method, including: obtaining audio stream data of a real-time conversation with a user; converting the audio stream data into a conversation text in real time; detecting the conversation text based on a sensitive word detection algorithm to identify sensitive risk words appearing in the conversation text; when the sensitive risk words meet preset warning standards, determining a rescue speech set for the sensitive risk words in a preset rescue speech library; sending a conversation intervention request and the rescue speech set to an auxiliary conversation end to request the auxiliary conversation end to access the real-time conversation, and performing conversation assistance processing based on the rescue speech set.

[0005] Optionally, the detection of the conversation text based on the sensitive word detection algorithm to identify sensitive risk words appearing in the conversation text includes: preprocessing the conversation text to obtain a set of keywords in the conversation text, wherein the preprocessing includes at least word segmentation, stop word removal and punctuation removal; based on a preset sensitive feature word library, using a similarity matching algorithm to determine keywords whose similarity to the sensitive feature words is higher than a preset threshold as the sensitive risk words.

[0006] Optionally, the method for constructing the sensitive feature word library includes: obtaining audio stream data of the complained conversation; converting the audio stream data of the complained conversation into the complained conversation text; extracting keywords whose frequency characteristics meet preset conditions from the complained conversation text to form the sensitive feature word library.

[0007] Optionally, the method for determining whether the sensitive risk word meets the preset warning standard includes: calculating the risk score of the conversation text according to the sensitive feature words associated with each of the sensitive risk words, and according to the preset weights associated with each of the sensitive feature words; when the risk score reaches a preset threshold, determining that the sensitive risk word meets the warning standard.

[0008] Optionally, the method for constructing the rescue speech library includes: obtaining audio stream data of the rescue conversation; converting the audio stream data of the rescue conversation into a rescue conversation text; extracting at least one rescue speech having an occurrence frequency higher than a preset threshold from the rescue conversation text; determining a category label associated with each of the rescue speeches according to a preset classification standard; and summarizing the at least one rescue speech based on the category label to obtain the rescue speech library.

[0009] Optionally, determining a rescue speech set for the sensitive risk words in a preset rescue speech library includes: determining a target category label that matches the risk-sensitive word; in the rescue speech set associated with the target category label, determining a preset number of rescue speech with the highest recommendation weights to form a rescue speech set for the sensitive risk words, wherein the recommendation weight associated with any of the rescue speech is determined by the frequency of occurrence of the rescue speech in the rescue conversation text.

[0010] Optionally, the method also includes: sending the latest conversation sub-text of a preset round in the conversation text to the auxiliary conversation end; the sending of the rescue speech set to the auxiliary conversation end includes: sending the rescue speech set sorted by recommendation weight to the auxiliary conversation end, so that the auxiliary conversation end performs conversation assistance processing based on the latest conversation sub-text of the preset round and the rescue speech set sorted by recommendation weight.

[0011] Another aspect of the present disclosure provides a conversation assistance processing device, including: an acquisition module, used to acquire audio stream data of a real-time conversation with a user; a first processing module, used to convert the audio stream data into a conversation text in real time; a second processing module, used to detect the conversation text based on a sensitive word detection algorithm to identify sensitive risk words appearing in the conversation text; a third processing module, used to determine a rescue speech set for the sensitive risk word in a preset rescue speech library when the sensitive risk word meets a preset warning standard; and a fourth processing module, used to send a conversation intervention request and the rescue speech set to an auxiliary conversation terminal to request the auxiliary conversation terminal to access the real-time conversation, and perform conversation assistance processing based on the rescue speech set.

[0012] Optionally, the second processing module includes: a first processing submodule, used to preprocess the conversation text to obtain a set of keywords in the conversation text, wherein the preprocessing includes at least word segmentation, stop word removal and punctuation removal; a second processing submodule, used to determine, based on a preset sensitive feature word library, using a similarity matching algorithm to determine keywords whose similarity with the sensitive feature words is higher than a preset threshold as the sensitive risk words.

[0013] Optionally, the method for constructing the sensitive feature word library includes: obtaining audio stream data of the complained conversation; converting the audio stream data of the complained conversation into the complained conversation text; extracting keywords whose frequency characteristics meet preset conditions from the complained conversation text to form the sensitive feature word library.

[0014] Optionally, the third processing module is also used to: calculate the risk score of the conversation text based on the sensitive feature words associated with each of the sensitive risk words, and based on the preset weights associated with each of the sensitive feature words; and when the risk score reaches a preset threshold, determine that the sensitive risk word meets the warning criteria.

[0015] Optionally, the method for constructing the rescue speech library includes: obtaining audio stream data of the rescue conversation; converting the audio stream data of the rescue conversation into a rescue conversation text; extracting at least one rescue speech having an occurrence frequency higher than a preset threshold from the rescue conversation text; determining a category label associated with each of the rescue speeches according to a preset classification standard; and summarizing the at least one rescue speech based on the category label to obtain the rescue speech library.

[0016] Optionally, the third processing module includes: a third processing sub-module, used to determine a target category label that matches the risk-sensitive word; a fourth processing sub-module, used to determine a preset number of rescue speeches with the highest recommendation weights in the rescue speech set associated with the target category label, to form a rescue speech set for the sensitive risk word, wherein the recommendation weight associated with any of the rescue speeches is determined by the frequency of occurrence of the rescue speech in the rescue conversation text.

[0017] Optionally, the fourth processing module is also used to: send the latest conversation sub-text of a preset round in the conversation text to the auxiliary conversation end; send a set of rescue words sorted by recommendation weight to the auxiliary conversation end through the fourth processing module, so that the auxiliary conversation end performs conversation assistance processing based on the latest conversation sub-text of the preset round and the set of rescue words sorted by recommendation weight.

[0018] Another aspect of the present disclosure provides an electronic device, comprising one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method of an embodiment of the present disclosure.

[0019] Another aspect of the present disclosure provides a computer-readable storage medium storing computer-executable instructions, which are used to implement the method of the embodiment of the present disclosure when executed.

[0020] Another aspect of the present disclosure provides a computer program product, including computer-readable instructions, wherein the computer-readable instructions are used to perform the conversation assistance processing method of an embodiment of the present disclosure when executed. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] For a more complete understanding of the present disclosure and its advantages, reference will now be made to the following description taken in conjunction with the accompanying drawings, in which:

[0022] Figure 1 The system architecture of the conversation assistance processing method and device according to the embodiment of the present disclosure is schematically shown;

[0023] Figure 2 A flowchart of a conversation assistance processing method according to an embodiment of the present disclosure is schematically shown;

[0024] Figure 3 A flowchart of another conversation assistance processing method according to an embodiment of the present disclosure is schematically shown;

[0025] Figure 4 A schematic diagram schematically shows a display interface of an auxiliary conversation terminal according to an embodiment of the present disclosure;

[0026] Figure 5 A block diagram of a conversation assistance processing device according to an embodiment of the present disclosure is schematically shown;

[0027] Figure 6 A block diagram of an electronic device according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0028] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0029] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "include", "comprising", etc. used herein indicate the existence of features, operations, operations and / or components, but do not exclude the existence or addition of one or more other features, operations, operations or components.

[0030] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0031] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0032] Some block diagrams and / or flow charts are shown in the accompanying drawings. It should be understood that some of the blocks in the block diagrams and / or flow charts or combinations thereof can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable conversational assistance processing device, so that these instructions, when executed by the processor, can create a device for implementing the functions / operations described in these block diagrams and / or flow charts. The technology of the present disclosure can be implemented in the form of hardware and / or software (including firmware, microcode, etc.). In addition, the technology of the present disclosure can take the form of a computer program product on a computer-readable storage medium storing instructions, which can be used by an instruction execution system or in combination with an instruction execution system.

[0033] The embodiment of the present disclosure provides a conversation assistance processing method and a processing device capable of applying the method. The method may specifically include the following operations: first, obtain audio stream data of a real-time conversation with a user, convert the audio stream data into a conversation text in real time, then detect the conversation text based on a sensitive word detection algorithm to identify sensitive risk words appearing in the conversation text, then, if the sensitive risk words meet the preset warning criteria, determine a rescue speech set for the sensitive risk words in a preset rescue speech library, and finally, send a conversation intervention request and a rescue speech set to the auxiliary conversation end to request the auxiliary conversation end to access the real-time conversation, and perform conversation assistance processing based on the rescue speech set.

[0034] Figure 1 The system architecture of the conversation assistance processing method and device according to the embodiment of the present disclosure is schematically shown. It should be noted that: Figure 1 What is shown is merely an example of a system architecture to which the embodiments of the present disclosure can be applied, in order to help those skilled in the art understand the technical content of the present disclosure, but it does not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios.

[0035] like Figure 1 As shown, the system architecture 100 includes a real-time conversation terminal 102, a processor 103 and an auxiliary conversation terminal 104. The real-time conversation terminal 102 is used to conduct a real-time conversation with a user (such as user 101 in the figure). The processor 103 is used to obtain audio stream data of the real-time conversation between the real-time conversation terminal 102 and the user, and to determine whether the real-time conversation reaches a preset warning standard based on the audio stream data. If so, the auxiliary conversation terminal 104 is requested to access the real-time conversation and perform conversation assistance processing.

[0036] Specifically, the processor 103 obtains the audio stream data of the real-time conversation between the real-time conversation terminal 102 and the user, and converts the audio stream data into conversation text in real time. Then, the processor 103 detects the conversation text based on the sensitive word detection algorithm to identify the sensitive risk words appearing in the conversation text, and when the sensitive risk words meet the preset warning standards, the processor 103 determines the rescue speech set for the sensitive risk words in the preset rescue speech library. Finally, the processor 103 sends a conversation intervention request and a rescue speech set to the auxiliary conversation terminal 104 to request the auxiliary conversation terminal 104 to access the real-time conversation, and perform conversation assistance processing based on the rescue speech set.

[0037] It should be noted that the conversation assistance processing method and device of the embodiment of the present disclosure can be used in the financial field, and can also be used in any field other than the financial field. The present disclosure will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0038] Figure 2 A flowchart of a conversation assistance processing method according to an embodiment of the present disclosure is schematically shown. Figure 2 As shown, method 200 may include operations S210 to S250.

[0039] In operation S210, audio stream data of a real-time conversation with a user is acquired.

[0040] In this embodiment, specifically, the audio stream data of the real-time conversation with the user is obtained. The real-time conversation can specifically be the real-time call data in the customer service reception process. The customer service reception process can include, for example, the business consultation reply process, the after-sales service processing process, the opinion complaint reception process, the product outbound call promotion process, etc. The real-time call data between the real-time conversation end and the user is obtained, and the real-time call data is detected for sensitive risk words. The sensitive risk words indicate that the real-time call quality inspection is unqualified and there is a possibility of being complained by the user.

[0041] When it is determined that a real-time conversation reaches the preset warning standard based on sensitive risk words, a request is sent to the auxiliary conversation end to access the real-time conversation for in-process control, so that sensitive complaint events can be resolved in a timely manner, thereby effectively reducing the post-event customer complaint rate and effectively improving the user's customer service reception experience. In practical applications, the real-time conversation end is generally the conversation end of the front desk operator. The front desk operators have a relatively high turnover rate and may lack professional skills or emergency handling experience. The auxiliary conversation end can be the conversation end of the back desk operator with high professional skills and rich emergency handling experience, or it can also be the conversation end of high-title personnel, such as the conversation end of the account manager.

[0042] Next, in operation S220, the audio stream data is converted into conversation text in real time.

[0043] In this embodiment, specifically, the obtained audio stream data is processed in real time to convert it into text by using Automatic Speech Recognition (ASR) technology, and a conversation list between the front desk operator and the user is obtained. Further, the operator's statements and the user's statements in the conversation list are separated to obtain the operator's conversation text and the user's conversation text.

[0044] Next, in operation S230, the conversation text is detected based on a sensitive word detection algorithm to identify sensitive risk words appearing in the conversation text.

[0045] In this embodiment, specifically, the conversation text is preprocessed to obtain a keyword set in the conversation text. The preprocessing may include, for example, word segmentation, stop word removal, punctuation removal, etc. Stop words are words automatically filtered out during the process of processing natural language data to save storage space and improve retrieval efficiency. Stop words may include, for example, modal particles, adverbs, prepositions, conjunctions, etc. Exemplarily, stop words such as "de" and "zai" appearing in the conversation text are removed.

[0046] Based on a preset sensitive feature word library, a similarity matching algorithm is used to determine keywords with a similarity higher than a preset threshold to the sensitive feature words as sensitive risk words. Sensitive risk words indicate the possibility of a real-time conversation being complained about afterwards, and it is necessary to intervene and assist in processing the real-time conversation to achieve reducing the post-event user complaint rate through in-process control.

[0047] A method for constructing a sensitive feature word library includes: obtaining the audio stream data of the complained conversation; converting the audio stream data of the complained conversation into the complained conversation text; extracting keywords in the complained conversation text whose word frequency features meet preset conditions to form a sensitive feature word library.

[0048] Specifically, the complaint records of past users against outbound call services are collected, and the audio stream data of the complained conversation is obtained based on the complaint number, customer information, agent number, etc. recorded in the complaint record. The audio stream data is converted into text to obtain the complained conversation text, and the complained conversation text is preprocessed such as word segmentation, stop word removal, and punctuation removal to obtain a set of keywords in the complained conversation text.

[0049] Next, from the keyword set in the complained conversation text, keywords whose frequency features meet the preset conditions are extracted to form a sensitive feature word library. The frequency feature indicates the importance and distinguishing ability of the keyword in the complained conversation text. For example, the TF-IDF value is used to characterize the frequency features of each keyword. Specifically, TF-IDF= TF×IDF, TF=t / T, IDF= N / (n+1), where TF represents the word frequency, IDF represents the inverse document frequency, t represents the frequency of occurrence of a keyword in a complained conversation text, T represents the total number of words in the complained conversation text, N represents the total number of complained conversation texts obtained, and n represents the total number of complained conversation texts containing the keyword. The larger the TF-IDF value, the higher the importance of the keyword, the stronger the distinguishing ability, and the more representative it is of the complained conversation text. Keywords with TF-IDF values ​​greater than the preset threshold in the complained conversation text are extracted to form a sensitive feature word library.

[0050] Next, in operation S240, when the sensitive risk word meets the preset warning standard, a rescue speech set for the sensitive risk word is determined in the preset rescue speech library.

[0051] In this embodiment, specifically, since different sensitive feature words have different risk levels, different weight values ​​can be preset for different sensitive feature words. The risk score of the conversation text is calculated based on the sensitive feature words associated with each sensitive risk word and the preset weight associated with each sensitive feature word. When the risk score reaches a preset threshold, it is determined that the sensitive risk word meets the preset warning standard.

[0052] When the sensitive risk words meet the preset warning criteria, a set of rescue words for the sensitive risk words is determined in the preset rescue words library. Specifically, rescue words matching the sensitive risk words are determined in the rescue words library, in order to reduce the probability of user complaints afterwards by delivering the rescue words to the users.

[0053] Among them, the method for constructing a rescue speech library includes: obtaining audio stream data of the rescue conversation; converting the audio stream data of the rescue conversation into a rescue conversation text; extracting at least one rescue speech with an occurrence frequency higher than a preset threshold from the rescue conversation text; determining the category label associated with each rescue speech according to a preset classification standard; and summarizing at least one rescue speech based on the category label to obtain a rescue speech library.

[0054] Specifically, based on past user complaint records regarding outbound call services, rescue conversation records with good subsequent resolution effects are collected. Based on the follow-up number, customer information, agent number, and other contents recorded in the rescue conversation record, the audio stream data of the rescue conversation is obtained. The audio stream data of the rescue conversation is converted into text to obtain the rescue conversation text. Since there may be multiple audio stream data of the rescue conversation obtained, a rescue conversation text set containing multiple rescue conversation texts may be obtained after the text conversion process. In the rescue conversation text set, at least one rescue speech with an occurrence frequency higher than a preset threshold is determined to form a rescue speech library. The higher the occurrence frequency of the rescue speech, the more significant the rescue effect of the speech, the higher the importance, and the more worthy of recommendation for such rescue speech.

[0055] The source scenarios of sensitive feature words can generally be divided into business consultation scenarios, business processing scenarios, outbound call promotion scenarios and other types. Sensitive feature words in business consultation scenarios may include words such as bad attitude, slow response, untimely, verbose, complaint, unfamiliar, and waste of time. Sensitive feature words in business processing scenarios may include words such as privacy leakage, violation, lag, overdue, error, verification failure, and repeated submission. Sensitive feature words in outbound call promotion scenarios may include words such as disturbing, unnecessary, disgusting, affecting, and repeating. The rescue words adapted to the sensitive feature words conform to the source scenarios of the sensitive feature words. Therefore, the applicable source scenarios can be used as preset classification standards, and the category labels associated with each rescue word can be determined based on the preset classification standards. Based on the aforementioned category labels, at least one rescue word is summarized to obtain a rescue word library.

[0056] Next, in operation S250, a session intervention request and a rescue speech set are sent to the auxiliary session end to request the auxiliary session end to access the real-time session and perform session assistance processing based on the rescue speech set.

[0057] In this embodiment, specifically, a session intervention request is sent to the auxiliary session terminal to request the auxiliary session terminal to access the real-time session. At the same time, a rescue speech set for sensitive risk words is sent to the auxiliary session terminal, so that the auxiliary session terminal performs session assistance processing based on the rescue speech set. By delivering rescue speech to users, sessions that may evolve into complaint events are processed in the process, which can effectively reduce the incidence of actual complaints in outbound call services. Optionally, the latest conversation subtext of a preset round in the conversation text can also be sent to the auxiliary session terminal, so that the auxiliary session terminal can grasp the user's interaction intention based on the latest conversation subtext of the preset round, and select or provide applicable rescue speech on its own.

[0058] Through the disclosed embodiment, the audio stream data of the real-time conversation with the user is obtained; the audio stream data is converted into a conversation text in real time; the conversation text is detected based on the sensitive word detection algorithm to identify the sensitive risk words appearing in the conversation text; when the sensitive risk words meet the preset warning standard, a rescue speech set for the sensitive risk words is determined in the preset rescue speech library; a conversation intervention request and a rescue speech set are sent to the auxiliary conversation end to request the auxiliary conversation end to access the real-time conversation, and the conversation auxiliary processing is performed based on the rescue speech set. By identifying the sensitive risk words appearing in the conversation text, and when the sensitive risk words meet the preset warning standard, a rescue speech set for the sensitive risk words is determined in the preset rescue speech library, and the auxiliary conversation end is requested to access the real-time conversation, and the conversation auxiliary processing is performed based on the rescue speech set. By introducing natural language processing technology and big data analysis technology, the real-time conversation between the outbound call seat and the user is effectively monitored, and the real-time conversation that may evolve into a complaint event is processed in the process, which can effectively reduce the incidence of actual complaints in the outbound call service, which is conducive to improving the service quality of the financial outbound call service and improving the communication efficiency between the outbound call seat and the user. In addition, it is also helpful to control the training costs for outbound call agents, shorten the training time, and can well adapt to the rapid development needs of financial business.

[0059] Optionally, after the sensitive feature word library is constructed, the audio stream data of the complained session that has evolved into a complaint event can be fully utilized, and the sensitive feature word library can be updated by performing fixed-point analysis on the audio stream data of the complained session, specifically including updating the sensitive feature words in the sensitive feature word library and updating the preset weights associated with the sensitive feature words. Similarly, after the rescue speech library is constructed, the audio stream data of the rescue session with a good solution effect can be fully utilized, and the rescue speech library can be updated by performing fixed-point analysis on the audio stream data of the rescue session.

[0060] Figure 3 The flowchart of another conversation assistance processing method according to an embodiment of the present disclosure is schematically shown. Figure 3As shown, method 300 may include operations S210 - S220 , S310 - S330 , and S250 .

[0061] In operation S210, audio stream data of a real-time conversation with a user is acquired.

[0062] Next, in operation S220, the audio stream data is converted into a conversation text in real time.

[0063] Next, in operation S310, the conversation text is preprocessed to obtain a set of keywords in the conversation text, and based on a preset sensitive feature word library, a similarity matching algorithm is used to determine keywords whose similarity with sensitive feature words is higher than a preset threshold as sensitive risk words.

[0064] In this embodiment, specifically, after the conversation text is preprocessed, a set of keywords in the conversation text is obtained. Using a similarity matching algorithm, keywords with a similarity to sensitive feature words higher than a preset threshold are determined based on a preset sensitive feature word library as sensitive risk words in the conversation text. Alternatively, a search is performed in the sensitive feature word library to determine whether there are sensitive feature words matching each keyword, and if so, the corresponding keyword is used as a sensitive risk word in the conversation text. Optionally, each keyword in the conversation text can be converted into a first feature vector, and each sensitive feature word in the sensitive feature word library can be converted into a second feature vector by word embedding. According to the cosine similarity algorithm, each first feature vector is matched with each second feature vector for similarity. When the similarity of two feature vectors is higher than a preset threshold, it is determined that the keywords and sensitive feature words corresponding to the two feature vectors are semantically similar, and the keyword is used as a sensitive risk word in the conversation text.

[0065] Next, in operation S320, the risk score of the conversation text is calculated based on the sensitive feature words associated with each sensitive risk word and the preset weight associated with each sensitive feature word. When the risk score reaches a preset threshold, it is determined that the sensitive risk word meets the preset warning standard.

[0066] In this embodiment, specifically, different sensitive feature words have different risk levels, therefore, the weight values ​​pre-assigned to each sensitive feature word may be different, and the weight values ​​pre-assigned to different sensitive feature words can be reasonably set according to the complaint history records in the outbound call business field. Exemplarily, the risk level of the sensitive feature word "violation" is higher than that of the sensitive feature word "disturbance", therefore, a higher weight value can be preset for the sensitive feature word "violation". The sensitive feature word library summarizes different sensitive feature words, and also summarizes the preset weights associated with each sensitive feature word. By setting up a sensitive feature word library, it is convenient to identify sensitive risk words appearing in conversation texts, and it is convenient to calculate the risk score of conversation texts based on each sensitive risk word.

[0067] Next, in operation S330, when the sensitive risk word meets the preset warning standard, the target category label that matches the risk sensitive word is determined, and in the rescue speech set associated with the target category label, a preset number of rescue speech with the highest recommendation weight is determined to form a rescue speech set for the sensitive risk word.

[0068] In this embodiment, specifically, sensitive feature words generally have associated source scenario features, and the source scenarios include, for example, business consultation scenarios, business processing scenarios, outbound call recommendation scenarios, and other types. Therefore, the target category label that matches the sensitive risk word can be determined based on the source scenario features of the sensitive risk word. A rescue speech set associated with the target category label is determined in the rescue speech library, and a preset number of rescue speeches with the highest recommendation weights are determined in the rescue speech set to form a rescue speech set for the sensitive risk word.

[0069] For example, for the sensitive risk words "verification failure" and "duplicate submission" in business processing scenarios, the target category label that matches the sensitive risk words is determined to be "comforting + solution words for business processing scenarios", and the rescue words that match the sensitive risk words include "Don't worry, we will help you verify the situation right away. Can you provide the order number, please?"

[0070] The recommendation weight associated with any rescue words is determined by the frequency of occurrence of the rescue words in the rescue conversation text. The higher the frequency of occurrence, the more times the rescue words are used to effectively resolve user complaints and the more positive the effect of the words. Therefore, the recommendation weight for the rescue words is higher.

[0071] Next, in operation S250, a session intervention request and a rescue speech set are sent to the auxiliary session end to request the auxiliary session end to access the real-time session and perform session assistance processing based on the rescue speech set.

[0072] In this embodiment, specifically, a session intervention request is sent to the auxiliary session end to request the auxiliary session end to access the real-time session for session assistance processing. At the same time, a rescue speech set sorted by recommendation weight is sent to the auxiliary session end, so that the auxiliary session end uses the rescue speech set as a reference and combines the latest session subtext of the preset round in the session text to perform real-time session assistance processing. Through the session assistance method, the real-time session that may evolve into a complaint incident can be resolved in the process, which can effectively reduce the incidence of actual complaints in outbound call services, which is conducive to improving the communication efficiency between outbound call seats and users, and at the same time, it is conducive to improving the user's interactive experience in financial outbound call services.

[0073] Figure 4A schematic diagram of an auxiliary conversation terminal display interface according to an embodiment of the present disclosure is schematically shown, wherein the display interface displays a conversation intervention request, the latest conversation sub-text of a preset round in the conversation text, business keywords and risk-sensitive words in the conversation text, and a rescue speech set sorted by recommendation weight, so that the auxiliary conversation terminal can access the real-time conversation between the outbound call agent and the user and perform conversation assistance processing based on the latest conversation sub-text of the preset round and the rescue speech set.

[0074] Figure 5 A block diagram of a conversation assistance processing device according to an embodiment of the present disclosure is schematically shown.

[0075] like Figure 5 As shown, the apparatus 500 includes an acquisition module 501 , a first processing module 502 , a second processing module 503 , a third processing module 504 and a fourth processing module 505 .

[0076] Specifically, the acquisition module 501 is used to acquire the audio stream data of the real-time conversation with the user; the first processing module 502 is used to convert the audio stream data into conversation text in real time; the second processing module 503 is used to detect the conversation text based on the sensitive word detection algorithm to identify the sensitive risk words appearing in the conversation text; the third processing module 504 is used to determine the rescue speech set for the sensitive risk words in the preset rescue speech library when the sensitive risk words meet the preset warning criteria; the fourth processing module 505 is used to send a conversation intervention request and a rescue speech set to the auxiliary conversation end to request the auxiliary conversation end to access the real-time conversation, and perform conversation assistance processing based on the rescue speech set.

[0077] Through the disclosed embodiment, the audio stream data of the real-time conversation with the user is obtained; the audio stream data is converted into a conversation text in real time; the conversation text is detected based on the sensitive word detection algorithm to identify the sensitive risk words appearing in the conversation text; when the sensitive risk words meet the preset warning standard, a rescue speech set for the sensitive risk words is determined in the preset rescue speech library; a conversation intervention request and a rescue speech set are sent to the auxiliary conversation end to request the auxiliary conversation end to access the real-time conversation, and the conversation auxiliary processing is performed based on the rescue speech set. By identifying the sensitive risk words appearing in the conversation text, and when the sensitive risk words meet the preset warning standard, a rescue speech set for the sensitive risk words is determined in the preset rescue speech library, and the auxiliary conversation end is requested to access the real-time conversation, and the conversation auxiliary processing is performed based on the rescue speech set. By introducing natural language processing technology and big data analysis technology, the real-time conversation between the outbound call seat and the user is effectively monitored, and the real-time conversation that may evolve into a complaint event is processed in the process, which can effectively reduce the incidence of actual complaints in the outbound call service, which is conducive to improving the service quality of the financial outbound call service and improving the communication efficiency between the outbound call seat and the user. In addition, it is also helpful to control the training costs for outbound call agents, shorten the training time, and can well adapt to the rapid development needs of financial business.

[0078] As a feasible embodiment, the second processing module includes: a first processing sub-module, used to pre-process the conversation text to obtain a set of keywords in the conversation text, wherein the pre-processing includes at least word segmentation, stop word removal and punctuation removal; a second processing sub-module, used to determine, based on a preset sensitive feature word library, keywords whose similarity with the sensitive feature words is higher than a preset threshold by using a similarity matching algorithm, as sensitive risk words.

[0079] As a feasible embodiment, a method for constructing a sensitive feature word library includes: obtaining audio stream data of the complained conversation; converting the audio stream data of the complained conversation into the complained conversation text; extracting keywords whose word frequency characteristics meet preset conditions in the complained conversation text to form a sensitive feature word library.

[0080] As a feasible embodiment, the third processing module is also used to: calculate the risk score of the conversation text based on the sensitive feature words associated with each sensitive risk word and the preset weight associated with each sensitive feature word; when the risk score reaches a preset threshold, determine that the sensitive risk word meets the warning standard.

[0081] As a feasible embodiment, a method for constructing a rescue speech library includes: obtaining audio stream data of a rescue conversation; converting the audio stream data of the rescue conversation into a rescue conversation text; extracting at least one rescue speech whose occurrence frequency is higher than a preset threshold from the rescue conversation text; determining a category label associated with each rescue speech according to a preset classification standard; and summarizing at least one rescue speech based on the category label to obtain a rescue speech library.

[0082] As a feasible embodiment, the third processing module includes: a third processing sub-module, used to determine a target category label that matches the risk-sensitive word; a fourth processing sub-module, used to determine a preset number of rescue words with the highest recommendation weights in the rescue word set associated with the target category label, to form a rescue word set for sensitive risk words, wherein the recommendation weight associated with any rescue word is determined by the frequency of occurrence of the rescue word in the rescue conversation text.

[0083] As a feasible embodiment, the fourth processing module is also used to: send the latest conversation sub-text of a preset round in the conversation text to the auxiliary conversation end; send a set of rescue words sorted by recommendation weight to the auxiliary conversation end through the fourth processing module, so that the auxiliary conversation end performs conversation assistance processing based on the latest conversation sub-text of the preset round and the set of rescue words sorted by recommendation weight.

[0084] It should be noted that in the embodiments of the present disclosure, the implementation of the device part is the same as or similar to the implementation of the method part, and will not be repeated here.

[0085] According to the embodiments of the present invention, any one or more of the modules, or at least part of the functions of any one of them, can be implemented in one module. According to the embodiments of the present invention, any one or more of the modules can be split into multiple modules for implementation. According to the embodiments of the present invention, any one or more of the modules can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented by hardware or firmware in any other reasonable way of integrating or packaging the circuit, or can be implemented in any one of the three implementation methods of software, hardware and firmware, or in any appropriate combination of any of them. Or according to the embodiments of the present invention, one or more of the modules can be at least partially implemented as a computer program module, and when the computer program module is run, the corresponding function can be executed.

[0086] For example, any multiple of the acquisition module 501, the first processing module 502, the second processing module 503, the third processing module 504, and the fourth processing module 505 can be combined in one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the acquisition module 501, the first processing module 502, the second processing module 503, the third processing module 504, and the fourth processing module 505 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware, and firmware, or in any appropriate combination of any of them. At least one of the acquisition module 501 , the first processing module 502 , the second processing module 503 , the third processing module 504 and the fourth processing module 505 may be at least partially implemented as a computer program module, and when the computer program module is executed, a corresponding function may be executed.

[0087] Figure 6 A block diagram of an electronic device according to an embodiment of the present disclosure is schematically shown. Figure 6 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0088] like Figure 6 As shown, the electronic device 600 includes a processor 610 and a computer-readable storage medium 620. The electronic device 600 can execute the method according to the embodiment of the present disclosure.

[0089] Specifically, the processor 610 may include, for example, a general-purpose microprocessor, an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 610 may also include an onboard memory for cache purposes. The processor 610 may be a single processing module or multiple processing modules for executing different actions of the method flow according to an embodiment of the present disclosure.

[0090] The computer-readable storage medium 620 may be, for example, a non-volatile computer-readable storage medium, and specific examples include but are not limited to: a magnetic storage device, such as a magnetic tape or a hard disk (HDD); an optical storage device, such as a compact disk (CD-ROM); a memory, such as a random access memory (RAM) or a flash memory; and the like.

[0091] The computer-readable storage medium 620 may include a computer program 621 , which may include code / computer-executable instructions that, when executed by the processor 610 , cause the processor 610 to perform a method according to an embodiment of the present disclosure or any variation thereof.

[0092] The computer program 621 may be configured to have, for example, a computer program code including a computer program module. For example, in an example embodiment, the code in the computer program 621 may include one or more program modules, for example, including module 621A, module 621B, ... It should be noted that the division method and number of modules are not fixed, and those skilled in the art may use appropriate program modules or program module combinations according to actual conditions, and when these program module combinations are executed by the processor 610, the processor 610 may execute the method according to the embodiment of the present disclosure or any variation thereof.

[0093] According to an embodiment of the present disclosure, at least one of the acquisition module 501, the first processing module 502, the second processing module 503, the third processing module 504, and the fourth processing module 505 can be implemented as a reference. Figure 6 The computer program modules described herein, when executed by the processor 610, can implement the corresponding operations described above.

[0094] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0095] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0096] It will be appreciated by those skilled in the art that, although the present disclosure has been shown and described with reference to specific exemplary embodiments of the present disclosure, it will be appreciated by those skilled in the art that, without departing from the spirit and scope of the present disclosure as defined by the appended claims and their equivalents, various changes in form and detail may be made to the present disclosure. Therefore, the scope of the present disclosure should not be limited to the above-described embodiments, but should be determined not only by the appended claims, but also by the equivalents of the appended claims.

Claims

1. A conversation assistance processing method, comprising: Get the audio stream data of the real-time conversation with the user; Converting the audio stream data into conversation text in real time; Detecting the conversation text based on a sensitive word detection algorithm to identify sensitive risk words appearing in the conversation text; When the sensitive risk word meets the preset warning standard, a rescue speech set for the sensitive risk word is determined in a preset rescue speech library; Sending a session intervention request and the rescue speech set to the auxiliary session end to request the auxiliary session end to access the real-time session and perform session assistance processing based on the rescue speech set; Sending the latest conversation subtext of a preset round in the conversation text to the auxiliary conversation end; The sending the rescue speech set to the auxiliary conversation terminal includes: Sending the rescue speech set sorted by recommendation weight to the auxiliary conversation terminal, so that the auxiliary conversation terminal performs conversation assistance processing based on the latest conversation subtext of the preset round and the rescue speech set sorted by recommendation weight; The method for constructing the rescue speech library includes: Acquire audio stream data of a rescue conversation; convert the audio stream data of the rescue conversation into a rescue conversation text; extract at least one rescue speech whose occurrence frequency is higher than a preset threshold from the rescue conversation text; determine a category label associated with each rescue speech according to a preset classification standard; and summarize the at least one rescue speech based on the category label to obtain the rescue speech library.

2. The method according to claim 1, wherein: The detecting the conversation text based on the sensitive word detection algorithm to identify the sensitive risk words appearing in the conversation text includes: Preprocessing the conversation text to obtain a keyword set in the conversation text, wherein the preprocessing at least includes word segmentation, stop word removal, and punctuation mark removal; Based on a preset sensitive feature word library, a similarity matching algorithm is used to determine keywords whose similarity to the sensitive feature words is higher than a preset threshold value to serve as the sensitive risk words.

3. The method according to claim 2, wherein: The method for constructing the sensitive feature word library includes: Get the audio stream data of the complained session; Converting the audio stream data of the complained conversation into the complained conversation text; Keywords whose frequency characteristics meet preset conditions are extracted from the complained conversation text to form the sensitive feature word library.

4. The method according to claim 2, wherein: The method for determining whether the sensitive risk word meets the preset warning standard includes: Calculating a risk score of the conversation text according to the sensitive feature words associated with each of the sensitive risk words and according to a preset weight associated with each of the sensitive feature words; When the risk score reaches a preset threshold, it is determined that the sensitive risk word meets the warning standard.

5. The method according to claim 1, wherein: The step of determining a rescue speech set for the sensitive risk word in a preset rescue speech library includes: Determine a target category label that matches the risk-sensitive word; In the rescue speech set associated with the target category label, a preset number of rescue speech with the highest recommendation weight is determined to form a rescue speech set for the sensitive risk word, The recommendation weight associated with any of the rescue words is determined by the frequency of occurrence of the rescue words in the rescue conversation text.

6. A conversation assistance processing device, comprising: An acquisition module is used to acquire audio stream data of a real-time conversation with a user; A first processing module, used for converting the audio stream data into conversation text in real time; A second processing module is used to detect the conversation text based on a sensitive word detection algorithm to identify sensitive risk words appearing in the conversation text; A third processing module is used to determine a rescue speech set for the sensitive risk word in a preset rescue speech library when the sensitive risk word meets the preset warning standard; The method for constructing the rescue speech library comprises: obtaining audio stream data of the rescue conversation; converting the audio stream data of the rescue conversation into a rescue conversation text; extracting at least one rescue speech with an appearance frequency higher than a preset threshold from the rescue conversation text; determining a category label associated with each rescue speech according to a preset classification standard; and summarizing the at least one rescue speech based on the category label to obtain the rescue speech library; A fourth processing module, configured to send a session intervention request and the rescue speech set to the auxiliary session end, so as to request the auxiliary session end to access the real-time session, and perform session assistance processing based on the rescue speech set; The fourth processing module is also used to: send the latest conversation sub-text of a preset round in the conversation text to the auxiliary conversation end; send a set of rescue words sorted by recommendation weight to the auxiliary conversation end through the fourth processing module, so that the auxiliary conversation end performs conversation assistance processing based on the latest conversation sub-text of the preset round and the set of rescue words sorted by recommendation weight.

7. An electronic device comprising: one or more processors; as well as a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enables the processor to implement the method according to any one of claims 1 to 5.

Images