Customer service method and device, storage medium and program product

Abstract

The embodiment of the invention provides a customer service method and device, a storage medium and a program product. In the embodiment of the invention, in the process of identifying the target intention information matched with the current round of question information of the user, the intention identification model based on artificial intelligence is guided through the model cue word, and the intention is subjected to layered identification aiming at the current round of question information, namely, the target intention type matched with the current round of question information is identified firstly; and then identifying a target intention matched with the question information of the round under the target intention type. Furthermore, the model cue word can guide the intention recognition model to carry out layered intention recognition on the keyword of the question information of the round. Therefore, the intention understanding ability of the intelligent customer service staff can be improved, the service quality of the intelligent customer service staff is improved, the problem of manual customer service staff is reduced, the pressure of the manual customer service staff is relieved, and the labor cost is saved.

Description

Technical Field

[0001] This application relates to the field of Internet technologies, and in particular, to a customer service scheduling method, device, storage medium, and program product. Background Art

[0002] With the rise of the Internet, the demand for intelligent customer service has also been increasing day by day. Intelligent customer service can share the workload of human customer service, reduce the work pressure of human customer service, and save labor costs. Currently, in the process of communication between users and customer service, first, the intelligent customer service provides services to users. When the services of the intelligent customer service cannot meet the needs of users, users can summon human customer service to provide services for them. However, existing intelligent customer service has the problem of inaccurate understanding of user intentions, resulting in more problems still being transferred to human customer service, and the advantages of intelligent customer service have not been fully utilized. Summary of the Invention

[0003] Multiple aspects of this application provide a customer service method, device, storage medium, and program product to improve the intention understanding ability of intelligent customer service, improve the service quality of intelligent customer service, reduce the problem of transferring to human customer service, relieve the pressure on human customer service, and save labor costs.

[0004] An embodiment of this application provides a customer service method, including: receiving the current round of question information input by a user on a customer service session interface; inputting the current round of question information and a first model prompt word into an intention recognition model based on artificial intelligence, where the first model prompt word is used to prompt the intention recognition model to provide at least intention hierarchical recognition services, and the intention hierarchical recognition services include intention classification operations and intention recognition operations; under the prompt of the first model prompt word, performing the intention classification operation according to the first keyword in the current round of question information to determine a target intention type adapted to the current round of question information from multiple known intention types; and performing the intention recognition operation under the target intention type to identify target intention information adapted to the current round of question information under the target intention type; sending the target intention information to an intelligent customer service for the intelligent customer service to output a reply message corresponding to the question information according to the target intention information.

[0005] An embodiment of this application further provides an electronic device, including: a memory and a processor; the memory is used to store a computer program; the processor is coupled to the memory and is used to execute the computer program to implement the steps in the above method.

[0006] An embodiment of this application further provides a computer-readable storage medium storing a computer program, which causes a processor to implement the steps in the above method when the computer program is executed by the processor.

[0007] The embodiments of the present application also provide a computer program product, which includes computer programs / instructions. When the computer programs / instructions are executed by a processor, the processor is enabled to implement the steps in the above method.

[0008] In the embodiments of the present application, in the process of identifying the target intent information adapted to the user's current question information, through the model prompt words, the intent recognition model based on artificial intelligence is guided to perform hierarchical intent recognition on the current question information, that is, first identify the target intent type adapted to the current question information, and then identify the target intent adapted to the current question information under the target intent type. Further, the model prompt words can guide the intent recognition model to perform hierarchical intent recognition on the keywords of the current question information. Thereby, the intent understanding ability of the intelligent customer service can be improved, the service quality of the intelligent customer service can be improved, the problems directed to the artificial customer service can be reduced, the pressure on the artificial customer service can be alleviated, and the labor cost can be saved. Description of the Drawings

[0009] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation to the present application. In the drawings:

[0010] Figure 1 It is a schematic flowchart of the customer service method provided by an exemplary embodiment of the present application;

[0011] Figure 2 It is a schematic flowchart of the hierarchical intent recognition process provided by an exemplary embodiment of the present application;

[0012] Figure 3 It is a schematic structural diagram of an electronic device provided by an exemplary embodiment of the present application. Detailed Embodiments

[0013] To make the objectives, technical solutions, and advantages of the present application clearer, the technical solutions of the present application will be clearly and completely described below in conjunction with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0014] It should be noted that the user information involved in this application (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) are all information and data that have been authorized by the user or fully authorized by all parties. And the collection, use, and processing of relevant data need to comply with relevant laws, regulations, and standards of relevant countries and regions, and corresponding operation entrances are provided for users to choose to authorize or reject.

[0015] In customer service, during the process of communication between the user and the customer service, usually, the intelligent customer service first provides services to the user. When the services of the intelligent customer service cannot meet the user's needs, the user can summon a human customer service to provide services for him / her.

[0016] Generally, the methods for the intelligent customer service to recognize intentions mainly include rule-based methods and deep learning-based methods. The rule-based method mainly detects whether some keywords that can express the user's intention are included in the user's question. Taking the user's question "Help me buy a movie ticket" as an example, if the keywords "buy" and "ticket" are detected at the same time, it can be judged that the user wants to buy a movie ticket. The deep learning-based method mainly uses the BERT model for multi-classification, takes the user's question as the input of the model, and the model labels the user's intention.

[0017] However, the rule-based method has the problem of poor scalability. The user may have multiple different expressions for a question, and it is unrealistic to list all combinations of keywords. Although the classification accuracy of the BERT model is higher than that of the rule-based method, it needs a large number of training samples for training to achieve better results, and it has a limit on the length of the input data, thus affecting the input of training samples and having low training efficiency, which is not very suitable for the rapid iteration of current intelligent customer service.

[0018] Therefore, the above-mentioned solutions all have the problem of insufficient understanding of the user's intention, resulting in more problems still going to the human customer service, and the advantages of the intelligent customer service have not been fully utilized.

[0019] To solve the above technical problems, in the embodiments of this application, during the process of identifying the target intention information adapted to the user's current round of question information, through model prompt words, an intention recognition model based on artificial intelligence is guided to perform hierarchical intention recognition on the current round of question information, that is, first identify the target intention type adapted to the current round of question information, and then identify the target intention adapted to the current round of question information under the target intention type. Further, the model prompt words can guide the intention recognition model to perform hierarchical intention recognition on the keywords of the current round of question information. Thus, the intention understanding ability of the intelligent customer service can be improved, the service quality of the intelligent customer service can be improved, the problem of going to the human customer service can be reduced, the pressure on the human customer service can be alleviated, and the labor cost can be saved.

[0020] The following will combine with the accompanying drawings to provide a detailed description of a solution provided by an embodiment of the present application.

[0021] Figure 1 It is a schematic flowchart of a customer service method provided for an exemplary embodiment of the present application. As Figure 1 shown, the method includes:

[0022] 101. Receive the current round of question information input by the user on the customer service session interface;

[0023] 102. Input the current round of question information and the first model prompt word into an intention recognition model based on artificial intelligence. The first model prompt word is used to prompt the intention recognition model to provide at least an intention hierarchical recognition service, and the intention hierarchical recognition service includes an intention classification operation and an intention recognition operation;

[0024] 103. Under the prompt of the first model prompt word, perform an intention classification operation according to the first keyword in the current round of question information to determine the target intention type adapted to the current round of question information from multiple known intention types; and perform an intention recognition operation under the target intention type to recognize the target intention information adapted to the current round of question information under the target intention type;

[0025] 104. Send the target intention information to the intelligent customer service for the intelligent customer service to output a reply message corresponding to the question information according to the target intention information.

[0026] Generally, some applications provide customer service, and customer service is the service (question and answer service) that these applications provide for users to have a conversation with the customer service. Among them, the user refers to a person who seeks the services or products he needs from the target application. Or, the user can also be a merchant who cooperates with the application program and promotes the services or goods it can provide in the target application, such as posting a post in the target application.

[0027] For merchants, they can purchase memberships of the target application. After becoming a member of the target application, they can enjoy membership privileges. Membership privileges can be, for example, the assured investment privilege, the daily booth privilege, the precise service privilege, etc. Among them, the assured investment privilege is a promotion service mainly used to help merchants increase the exposure rate and traffic of posts. It can help merchants quickly obtain more high-quality traffic through functions such as intelligent product selection and intelligent keyword selection. The daily booth privilege is a post promotion product purchased by the day, which can meet the needs of merchants for 24-hour online promotion of posts. The precise privilege is a digital marketing solution usually used to help merchants achieve precise reach to target customers. It can provide more efficient promotion effects for merchants through big data analysis and intelligent algorithms. For users who are looking for services or products they need from the target application, they can purchase a membership of the platform to enjoy discount privileges, or become a regular user of the target application only through registration to enjoy the privileges of regular users, such as the privilege of chatting with the customer service. The above is only an exemplary illustration, but not limited thereto.

[0028] In the embodiments of the present application, the type of the target application is not limited. The target application can be any APP running on a terminal device. For example, a housekeeping APP, a house viewing APP, a transportation APP, a live broadcast APP, and so on. Or, the target application can also be a web page or a small program running on the basis of an APP. In addition, the type of the terminal device is not limited in this embodiment. The terminal device can be a smart handheld device, such as a smart phone, a tablet computer, and so on. Or, the terminal device can also be a desktop device, such as a laptop computer, a desktop computer, and so on; or, the terminal device can also be a smart wearable device, such as a smart watch, a smart bracelet, and so on; or, the terminal device can also be various smart home appliances with a display screen, such as a smart TV, a smart large screen, or a smart robot, and so on. The target application can provide a customer service conversation interface to facilitate users to have a conversation with the intelligent customer service based on the customer service conversation interface.

[0029] In the embodiments of the present application, users can input each round of question information through the customer service conversation interface, and each round of question information can be displayed on the customer service conversation interface. The implementation form of the question information is not limited in this embodiment. The question information can be, for example, text, picture, or video. The specific content of the question information is also not limited. For example, the question information can be question information for which the user seeks an answer from the intelligent customer service, such as "How much time is left for my membership privileges?" Another example is that the question information can also be request information for which the user requests the intelligent customer service to provide some services, such as "Based on the following detailed information of the commodity or service given, can you help me generate a post and publish this post?"

[0030] In the embodiments of the present application, in order to improve the ability to understand intentions and the accuracy of intention recognition, improve the service quality of intelligent customer service, and reduce the problem of transferring to human customer service, an intention recognition model based on artificial intelligence is introduced. The intention recognition model based on artificial intelligence is used to identify the target intention information corresponding to the question information of each round of the user, and then the target intention information is sent to the intelligent customer service for the intelligent customer service to output the response information corresponding to the question information of each round according to the target intention information. Correspondingly, an AI-based customer service scheduler can also be introduced in this embodiment. The AI customer service scheduler is mainly used to call the intention recognition model to perform intention recognition, and then provide the recognized target intention information to the intelligent customer service.

[0031] Among them, the intention recognition model based on artificial intelligence can be obtained by training an initial intention recognition model based on a large number of samples. Each sample includes question information and intention information adapted to the question information. In this embodiment, multiple intention information during intention recognition can be pre-configured, that is, multiple intention information is known. The pre-configured multiple intention information can be filtered from the intention information involved in historical data. Moreover, when filtering multiple intention information from the intention information involved in historical data, an intention information de-duplication operation can be performed to avoid duplicate intention information in multiple intention information.

[0032] In addition, intention information has an intention complexity. Different intention information has different intention complexities. The intention complexity represents the difficulty level of solving the question information related to the intention information. What the intelligent customer service can solve are question information whose complexity is adapted to its response ability. For some question information with a relatively high difficulty level, the intelligent customer service may not have the response ability and can only be solved by human customer service. Based on this, when sending the target intention information to the intelligent customer service, it can first be determined whether the complexity of the target intention information exceeds the response ability of the intelligent customer service; if the complexity of the target intention information does not exceed the response ability of the intelligent customer service, the target intention information is sent to the intelligent customer service; if the complexity of the target intention information exceeds the response ability of the intelligent customer service, the user is prompted to select a human customer service for conversation. For the specific implementation method of determining whether the complexity of the target intention information exceeds the response ability of the intelligent customer service when sending the target intention information to the intelligent customer service, reference can be made to the relevant descriptions in the following embodiments, which will not be elaborated here.

[0033] In the embodiments of the present application, since the intent recognition model may have multiple functions, such as identifying keywords in the question information, semantic analysis, identifying the intent type corresponding to the question information, identifying the intent information corresponding to the question information, etc. Based on this, in order to improve the recognition efficiency of the intent recognition model, the model prompt words can be used to guide the intent recognition model to perform the operations required for each task. The model prompt words at least include: the first model prompt word, and the first prompt word can be used to prompt the intent recognition model to at least provide the intent recognition service.

[0034] As Figure 2 shown, in order to further improve the intent recognition ability of the intent recognition model, hierarchical intent recognition can be performed on each round of question information. First, identify the target intent type adapted to each round of question information, and then identify the target intent information adapted to each round of question information under the target intent type. Then the first prompt word can also be used to prompt the intent recognition model to at least provide the hierarchical intent recognition service, and the hierarchical intent recognition service includes intent classification operations and intent recognition operations. Based on this, after receiving the current round of question information input by the user on the customer service session interface, the current round of question information and the first model prompt word are input into the intent recognition model based on artificial intelligence. Under the prompt of the first model prompt word, according to the first keyword of the current round of question information, perform the intent classification operation to determine the target intent type adapted to the current round of question information from multiple known intent types. In this embodiment, the multiple intent types during intent recognition can be pre-configured, that is, the multiple intent types are known, and the pre-configured multiple intent types can be filtered from the intent types involved in the historical data. When filtering multiple intent types from the intent types involved in the historical data, an intent type deduplication operation can be performed to avoid duplicate intent types in the multiple intent types. The intent type can be, for example: promotion questions, phone questions, cost questions, etc.

[0035] It should be noted that the target intent type may include multiple intent types, and there may be duplicate intent types in the multiple intent types. Then, after obtaining the target intent type and when the target intent type includes multiple intent types, a deduplication operation can be performed on the multiple intent types. If there are the same or similar intent types, select one intent type as the target intent type.

[0036] For the specific implementation of performing the intent classification operation according to the first keyword of the current round of question information under the prompt of the first model prompt word to determine the target intent type adapted to the current round of question information from multiple known intent types, reference can be made to the relevant descriptions of the following embodiments, which will not be elaborated here.

[0037] In practical applications, the first model prompt can be, for example, "Please act as an expert in intent recognition, identify the target intent information corresponding to each question information, and when identifying the target intent information corresponding to each question information, first identify the target intent type corresponding to each question information from multiple pre-set intent types. After identifying the target intent type, identify the target intent information from multiple intent information under the pre-set target intent type".

[0038] In the embodiments of the present application, after obtaining the target intent type adapted to the current round of question information, under the prompt of the first prompt, the intent recognition model can also perform an intent recognition operation under the target intent type to identify the target intent information adapted to the current round of question information under the target intent type. The target intent information can include multiple intent information, and there may be duplicate intent information among the multiple intent information. Then, after obtaining the target intent information and when the target intent information includes multiple intent information, a deduplication operation can be performed on the multiple intent information. If there are the same or similar intent information, one of the intent information is selected as the target intent information. For example, the multiple intent information included in the promotion question can be: poor promotion effect, promotion operation problem, promotion cost problem, coupon usage problem, etc. The multiple intent information included in the phone question can be: few calls problem, no call problem, phone signal problem, etc. The multiple intent information included in the cost question can be: promotion cost problem, phone cost problem, etc. For the specific implementation manner of performing an intent recognition operation under the target intent type to identify the target intent information adapted to the current round of question information under the target intent type, reference can be made to the relevant descriptions in the following embodiments, and details are not described here for the time being.

[0039] In the embodiments of the present application, after obtaining the target intent information corresponding to the current round of question information, the target intent information is sent to the intelligent customer service for the intelligent customer service to output the reply information corresponding to the question information according to the target intent information.

[0040] Optionally, the intelligent customer service corresponds to an AI Q&A service model based on artificial intelligence. When the intelligent customer service outputs the reply information corresponding to the question information according to the target intent information, it can be that the AI Q&A service model outputs the reply information corresponding to the question information according to the target intent information and then sends it to the target application in the role of the intelligent customer service and displays it to the user through the customer service conversation interface. For a more specific implementation manner of sending the target intent information to the intelligent customer service for the intelligent customer service to output the reply information corresponding to the question information according to the target intent information, reference can be made to the relevant descriptions in the following embodiments, and details are not described here for the time being.

[0041] In the technical solutions provided by the above embodiments of the present application, during the process of identifying the target intent information adapted to the user's current round of question information, through the model prompt words, the intent recognition model based on artificial intelligence is guided to perform hierarchical intent recognition on the current round of question information, that is, first identify the target intent type adapted to the current round of question information, and then identify the target intent under the target intent type that is adapted to the current round of question information. Further, the model prompt words can guide the intent recognition model to perform hierarchical intent recognition on the keywords of the current round of question information. Thereby, the intent understanding ability of the intelligent customer service can be improved, the service quality of the intelligent customer service can be improved, the problems of transferring to the human customer service can be reduced, the pressure on the human customer service can be alleviated, and the labor cost can be saved.

[0042] In practical applications, after receiving the user's current round of question information, the previous round of conversation may have ended or not ended yet. To avoid missing the user's question information and improve the conversation experience between the user and the intelligent customer service, after receiving the current round of question information, it is possible to identify whether the previous round of conversation has ended and perform corresponding processing operations according to the recognition result. Based on this, sending the target intent information to the intelligent customer service for the intelligent customer service to output the response information corresponding to the current round of question information according to the target intent information, including: inputting the second model prompt words into the intent recognition model, and under the guidance of the second model prompt words, determining whether the previous round of conversation information has ended. The second model prompt words are used to prompt the intent recognition model to provide the recognition service of whether the previous round of question information has ended; in the case where it is determined that the previous round of conversation information has not ended, outputting the target intent information corresponding to the previous round of question information and the target intent information corresponding to the current round of question information; sending the target intent information corresponding to the current round of question information and the target intent information corresponding to the previous round of question information to the intelligent customer service for the intelligent customer service to output the corresponding response information according to the target intent information corresponding to the current round of question information and the target intent corresponding to the previous round of question information respectively.

[0043] It should be noted that in the case where it is determined that the previous round of conversation information has not ended, the present embodiment does not limit the output order of the response information corresponding to the current round of question information and the response information corresponding to the previous round of question information by the intelligent customer service. For example, the intelligent customer service can first output the response information corresponding to the current round of question information and then output the response information corresponding to the previous round of question information. Another example is that the intelligent customer service can first output the response information corresponding to the previous round of question information and then output the response information corresponding to the current round of question information. The intelligent customer service first outputs the response information corresponding to the previous round of question information and then outputs the response information corresponding to the current round of question information can be used as a preferred implementation manner, which can avoid the situation of jumping responses to the user's question information and improve the conversation experience between the user and the intelligent customer service.

[0044] Further optionally, the information of the previous conversation between the user and the intelligent customer service includes the user's previous question information and the intelligent customer service's previous reply information. The previous reply information may contain an indication of whether the previous conversation has ended, or the combined information of the previous question information and the previous reply information may contain an indication of whether the previous conversation has ended. Thus, it is possible to identify whether the previous conversation has ended based on the previous reply information or the combined information of the previous question information and the previous reply information. Based on this, input the second model prompt into the intent recognition model to determine whether the previous conversation has ended, including: obtaining the information of the previous conversation between the user and the intelligent customer service; inputting the previous conversation information and the second model prompt into the intent recognition model, and under the prompt of the second model prompt, identify whether there is an indication of whether the previous conversation has been completed from the previous reply information or from the previous question information and the previous reply information; if it exists, determine that the previous conversation has ended; if it does not exist, determine that the previous conversation has not ended.

[0045] Optionally, obtaining the information of the previous conversation between the user and the intelligent customer service includes: automatically triggering the obtaining of the information of the previous conversation between the user and the intelligent customer service when receiving the current question information; or periodically obtaining the conversation information between the user and the intelligent customer service to obtain the information of each conversation between the user and the intelligent customer service and the current question information. The period for obtaining the conversation information can be, for example, 30 seconds, 1 minute, N minutes, etc.

[0046] In an alternative embodiment, under the prompt of the second model prompt, identifying whether there is an indication of whether the previous conversation has been completed from the previous reply information or from the previous question information and the previous reply information includes: identifying the previous reply information from the previous conversation information, performing word segmentation on the previous reply information to obtain a plurality of first word segments, and determining a second keyword from the first word segments; identifying the semantic features of the second keyword, and determining whether the keyword has an indication of having been completed according to the semantic features of the second keyword; if the second keyword has an indication of having been completed, determine that there is an indication of whether the previous conversation has been completed in the previous reply information; if the second keyword does not have an indication of having been completed, determine that there is no indication of whether the previous conversation has been completed in the previous reply information. In this embodiment, the second model prompt can be, for example, "Please act as a dialogue recognition expert and, based on the semantic features of the keywords included in the previous reply information of the intelligent customer service, determine whether the keyword has an indication of having completed the previous conversation. If it contains such an indication, it means the previous conversation has been completed; otherwise, it means the previous conversation has not been completed."

[0047] Optionally, perform word segmentation on the previous round of response information to obtain multiple first word segments, and determine the second keywords from the first word segments, including: determining the keywords of the previous round of response information according to TF-IDF (Term Frequency-Inverse Document Frequency); or, determining the keywords of the previous round of response information based on a rule-based method; or, determining the keywords of the previous round of response information based on a machine learning method; or, determining the keywords of the previous round of response information based on a deep learning method. Among them, TF-IDF is a method for measuring the importance of vocabulary, which combines term frequency (TF) and inverse document frequency (IDF). TF (term frequency) refers to the frequency of a word appearing in historical Q&A information, reflecting the importance of the word in historical Q&A information. IDF (inverse document frequency): measures the general importance of a word and is achieved by calculating the rarity of the word in the entire historical Q&A information. The TF-IDF formula is: TF-IDF(t, d) = TF(t, d) × IDF(t), where t is the word and d is the document. The rule-based methods include but are not limited to: stop word filtering method, part-of-speech filtering method, pattern matching method, etc.

[0048] Among them, determining the keywords of the previous round of response information according to TF-IDF includes: for the previous round of response information, perform word segmentation operations to obtain multiple word segments included in each question information; calculate the TF value of each word segment and calculate the IDF value of each word segment; based on the TF value and IDF value of each word segment, calculate the TF-IDF value of each word segment, and select the word with the highest TF-IDF value as the keyword.

[0049] Among them, when determining the keywords of the previous round of response information based on a rule-based method, the stop word filtering method is to remove common and meaningless words in the text (such as "of", "is", "and", etc.) and only retain words that may be meaningful. Specifically, it can be implemented as: maintaining a stop word list and filtering out these words after word segmentation; the part-of-speech filtering method is to extract keywords according to parts of speech (such as nouns, verbs, adjectives), and these parts of speech usually can better express the core content of the text. Specifically, it can be implemented as: using a part-of-speech tagging tool (such as Stanford NLP, HanLP, etc.) to extract words of specific parts of speech; the pattern matching method is to extract keywords according to predefined patterns (such as "completed word + noun"). Specifically, it can be implemented as: defining rules, such as "already completed + noun", "verb + noun", etc., such as "already replied" and "already done according to your requirements".

[0050] Among them, when determining the keywords of the previous round of response information based on machine learning methods, there are supervised learning methods and unsupervised learning methods. Among them, for supervised learning methods, a model can be trained using labeled keyword data, and the model learns how to extract keywords from text. Specifically, it can be implemented as follows: prepare labeled data (text and its keywords); extract text features (such as TF-IDF vectors); train a model (such as SVM, Naive Bayes, deep learning model); use the model to predict keywords. For unsupervised learning methods, keywords can be automatically extracted through the structure and semantic information of the text without labeled data. Specifically, it can be implemented as follows: use a graph-based ranking (TextRank) algorithm, that is, similar to PageRank, construct a graph through the co-occurrence relationship between words, and calculate the importance of words; or use the topic LDA (Latent Dirichlet Allocation) algorithm, that is, extract topic words as keywords through the three-layer structure of document-topic-word.

[0051] Among them, when determining the keywords of the previous round of response information based on deep learning methods, an architecture of WordEmbedding + neural network and a Transformer architecture (such as BERT) can be used. Among them, using the architecture of Word Embedding + neural network and the Transformer architecture converts words into semantic features (vectors) through pre-trained word embedding algorithms (such as Word2Vec, GloVe, BERT), and then extracts keywords through neural network algorithms (such as CNN, RNN, Transformer). Specifically, it can be implemented as follows: use a pre-trained model (such as BERT) to obtain the embedding vector of the word; input it into a neural network model (such as Bi-LSTM, Transformer); use the attention mechanism (Attention) to highlight important words; output keywords. Using the Transformer architecture (such as BERT) utilizes the BERT model to generate dynamic embeddings of words through context information, which can better capture the semantic information of words. Specifically, it can be implemented as: use BERT to encode the text; extract the output of the hidden layer, calculate the importance of each word; select keywords according to the importance score. It should be noted that when using an algorithm model to determine the keywords of the previous round of response information, the algorithm model belongs to the network layer included in the intent recognition model.

[0052] Optionally, identifying the semantic features of the second keyword includes: encoding the second keyword based on a self-attention learning mechanism to obtain the semantic features of the second keyword.

[0053] For example. The previous round of question information is, for example, "How much time is left for my membership benefits?" The previous round of reply information can be, for example, "There are N months left for your membership benefits. Your current question has been answered. Do you have any other needs?" Among them, "There are N months left" and "has been answered" can both be used as keywords indicating that the conversation has been completed.

[0054] It should be noted that the above TF-IDF, rule-based method, machine learning-based method, and deep learning-based method can be used alone or in combination to determine the keywords of the previous round of reply information. It should also be noted that the specific implementation manner of determining the second keyword and the specific implementation manner of semantic feature extraction in this optional embodiment are also applicable to the determination of keywords and the semantic feature extraction process in the above or following embodiments, and will not be elaborated in the following embodiments.

[0055] In another optional embodiment, the previous round of reply information and the previous round of response information are identified from the previous round of conversation information, the previous round of question information and the previous round of response information are tokenized to obtain a plurality of second tokens, and the third keyword is determined from the second tokens; the semantic features of the third keyword are identified, and whether the keyword has a meaning indication of completion is determined according to the semantic features of the third keyword; if the third keyword has a meaning indication of completion, it is determined that there is a meaning indication of whether the previous round of conversation has been completed in the previous round of question information and the previous round of response information; if the third keyword does not have a meaning indication of completion, it is determined that there is no meaning indication of whether the previous round of conversation has been completed in the previous round of question information and the previous round of response information. In this embodiment, the second model prompt can be, for example, "Please act as a dialogue recognition expert. Based on the semantic features of the keywords included in the combined information of the previous round of question information and the previous round of response information of the intelligent customer service, determine whether the keyword has a meaning indication of completing the previous round of conversation. If it contains, it means that the previous round of conversation has been completed; otherwise, it means that the previous round of conversation has not been completed." In addition, for this embodiment, the specific implementation manner of tokenizing the previous round of question information and the previous round of response information to obtain a plurality of second tokens, determining the third keyword from the second tokens, identifying the semantic features of the third keyword, and determining whether the keyword has a meaning indication of completion according to the semantic features of the third keyword can refer to the relevant description of the previous optional embodiment.

[0056] For example. Suppose the user's previous question was, "Based on the following details of the goods or services provided, could you help me post a message, please?" And the intelligent customer service's previous reply was, "I have already posted it for you. Please confirm whether you can view the posted message. If you can view it, it means the post was successfully published." Further, suppose the user's previous reply was, "I can view the posted message on my end!" This example determines whether there is an indication of whether the previous conversation has been completed by combining each question and reply message contained in the previous conversation between the user and the intelligent customer service. Among them, the combination of the keywords "help post a message", "whether can view", and "can view" contained in the previous conversation information can determine that there is an indication in the previous conversation that the previous conversation has been completed.

[0057] Further optionally, identify whether there is an indication of whether the previous conversation has been completed from the previous reply message or from the previous question message and the previous reply message. Among the two identification methods, one of the methods can be selected first to identify whether there is an indication of whether the previous conversation has been completed. If the identification result cannot be obtained by this identification method, then the other identification method can be continued to be selected. The selection order of the two identification methods can be not limited. Preferably, priorities can be set for the two methods. Identifying whether there is an indication of whether the previous conversation has been completed from the previous reply message is set as the first priority, and identifying whether there is an indication of whether the previous conversation has been completed from the previous question message and the previous reply message is set as the second priority. In practical applications, the identification method corresponding to the first priority is preferably selected first. If the identification result cannot be obtained by the identification method corresponding to the first priority, then the identification method corresponding to the second priority is continued to be selected for identification. By preferentially selecting the identification method corresponding to the first priority, only the previous reply message needs to be identified, which can improve the data processing efficiency and the reply efficiency of the intelligent customer service.

[0058] Continuing with the above embodiment of determining the target intent type adapted to the user's current question message from multiple known intent types, under the prompt of the first model prompt word, perform an intent classification operation according to the first keyword in the current question message to determine the target intent type adapted to the user's current question message from multiple known intent types, including: performing word segmentation processing on the current question message to obtain multiple third word segments, and respectively extracting features from the multiple third word segments and each known intent type to obtain the semantic features of the multiple third word segments and the semantic features of each known intent type; calculating the first similarity between the semantic feature of each third word segment and the semantic features of each target intent type. If there is a first target similarity greater than the first similarity threshold, use at least one intent type corresponding to the first target similarity as the target intent type.

[0059] In an optional embodiment, under the target intent type, an intent recognition operation is performed to identify target intent information adapted to the current round of question information under the target intent type, including: extracting features of each known intent information under the target intent type to obtain semantic features of each intent information under each intent type; calculating a second similarity between the semantic feature of each third sub-word and the semantic features of each intent information under each intent type, and if there is a second target similarity greater than the second similarity threshold, using at least one intent information corresponding to the second target similarity as the target intent information.

[0060] Further optionally, the target intent information may include multiple intent information. To avoid duplicate intent information in the target intent information and improve the working efficiency of the AI Q&A service model, when the target intent information is multiple, the semantic features of the multiple intent information included in the target intent information can be extracted, the similarity of the semantic features of the multiple intent information can be calculated, and the intent information with a semantic similarity difference less than the preset similarity threshold is used as the same or similar intent information. Further, one intent information can be selected from at least two same or similar intent information as one intent information in the target intent information, and other intent information with the same or similar semantic features as this intent information can be discarded. Or, when the target intent information is multiple, the semantic features of the multiple intent information included in the target intent information can be extracted, the similarity of the semantic features of the multiple intent information can be calculated, and the intent information with a semantic similarity difference less than the preset similarity threshold is used as the same or similar intent information. Further, the semantic feature similarity between each intent information in at least two same or similar intent information and the first keyword is calculated, and the one with the largest similarity is used as one intent information in the target intent information, and other intent information with the same or similar semantic features as this intent information can be discarded.

[0061] Further optionally, the known multiple intent information is pre-configured with an intent complexity, and different intent information has different intent complexities. The intent complexity represents the difficulty level of solving the question information related to this intent information. The AI Q&A service model can solve question information with a complexity adapted to its capabilities. For some question information with a higher difficulty level, the AI Q&A service model may not be able to solve it and can only be solved by a human customer service. Based on this, sending the target intent information to the intelligent customer service includes: identifying the complexity of the target intent information corresponding to the current round of question information based on the pre-configured intent complexity level; if the intelligent customer service is adapted to the complexity of the target intent information corresponding to the current round of question information, sending the target intent information corresponding to the current round of question information to the intelligent customer service.

[0062] Further optionally, if the intelligent customer service is adapted to the complexity of the target intent information corresponding to the current round of question information, the target intent information corresponding to the current round of question information is sent to the intelligent customer service, including: inputting the target intent information corresponding to the current round of question information and the third model prompt word into the AI Q&A service model based on artificial intelligence corresponding to the intelligent customer service, where the third model prompt word is used to prompt the AI service model to provide Q&A services for users; under the prompt of the third model prompt word, generating target service information according to the target intent information corresponding to the current round of question information, and after modifying the target service information, outputting it to the customer service conversation interface as the reply information for the current round of question information. Among them, when modifying the target service information, it can be modified in terms of the style, tone, intonation, etc. of the target service information to make the target service information more anthropomorphic and improve the dialogue experience between the user and the intelligent customer service.

[0063] In practical applications, for example, the third model prompt word can be "Please act as a dialogue expert, generate service information according to the target intent information corresponding to the current round of question information, and output the modified information after modifying the service information".

[0064] It should be noted that while inputting the target intent information corresponding to the current round of question information and the third model prompt word into the AI Q&A service model based on artificial intelligence corresponding to the intelligent customer service, the current round of question information can also be input into the AI Q&A service model, so that under the guidance of the third model prompt word, the AI Q&A service model can generate target service information by combining the current round of question information and the target intent information corresponding to the current round of question information.

[0065] It should also be noted that the generation method of the previous round of reply information can refer to the generation method of the reply information corresponding to the current round of question information, which will not be elaborated here.

[0066] Further optionally, in the case where it is determined that the intelligent customer service is not adapted to the complexity of the target intent information, at least the current round of question information is sent to the human customer service for the human customer service to reply to the question information.

[0067] Further optionally, in the case where it is determined that the intelligent customer service is not adapted to the complexity of the target intent information, the target intent information of the current round of question information can also be sent to the human customer service for the human customer service to reply to the question information with reference to the target intent, thereby improving the reply efficiency of the human customer service.

[0068] Further optionally, based on the historical information related to customer service in advance, multiple intent types and multiple intent information under each intent type are constructed; a deduplication and priority sorting process is performed on the multiple intent types and the multiple intent information under each intent type to obtain the processed multiple intent types and the multiple intent information under each intent type.

[0069] Figure 3 The structural schematic diagram of the electronic device provided by the exemplary embodiment of the present application. As Figure 3 shown, it includes: a memory 30a and a processor 30b; the memory 30a is used for storing a computer program; the processor 30b is coupled with the memory 30a and is used for executing the computer program to implement the following steps:

[0070] Receiving the current round of question information input by the user on the customer service session interface; inputting the current round of question information and the first model prompt word into the intention recognition model based on artificial intelligence, the first model prompt word is used to prompt the intention recognition model to provide at least an intention hierarchical recognition service, and the intention hierarchical recognition service includes an intention classification operation and an intention recognition operation; under the prompt of the first model prompt word, performing an intention classification operation according to the first keyword in the current round of question information to determine the target intention type adapted to the current round of question information from multiple known intention types; and performing an intention recognition operation under the target intention type to recognize the target intention information adapted to the current round of question information under the target intention type; sending the target intention information to the intelligent customer service for the intelligent customer service to output the reply information corresponding to the question information according to the target intention information.

[0071] In an optional embodiment, when the processor sends the target intention information to the intelligent customer service for the intelligent customer service to output the reply information corresponding to the current round of question information according to the target intention information, it is specifically used for: inputting the second model prompt word into the intention recognition model to determine whether the previous round of conversation has ended; the second model prompt word is used to prompt the intention recognition model to provide the recognition service of whether the previous round of question information has ended; in the case of determining that the previous round of conversation has not ended, outputting the target intention information corresponding to the previous round of question information and the target intention information corresponding to the current round of question information; sending the target intention information corresponding to the current round of question information and the target intention information corresponding to the previous round of question information to the intelligent customer service for the intelligent customer service to output the corresponding reply information according to the target intention information corresponding to the current round of question information and the target intention corresponding to the previous round of question information respectively.

[0072] Further optionally, when the processor inputs the second model prompt word into the intention recognition model to determine whether the previous round of conversation has ended, it is specifically used for: obtaining the previous round of conversation information between the user and the intelligent customer service, and the previous round of conversation information includes the previous round of question information of the user and the previous round of reply information of the intelligent customer service; inputting the previous round of conversation and the second model prompt word into the intention recognition model, and under the prompt of the second model prompt word, identifying whether there is a meaning representation indicating whether the previous round of conversation has been completed from the previous round of reply information or from the previous round of question information and the previous round of reply information; if it exists, determining that the previous round of conversation has ended; if it does not exist, determining that the previous round of conversation has not ended.

[0073] In an alternative embodiment, when the processor, under the prompt of the second model prompt word, identifies whether there is an indication of whether the previous conversation has been completed from the previous response information or from the previous question information and the previous response information, it is specifically used for: identifying the previous response information from the previous conversation, performing word segmentation processing on the previous response information to obtain a plurality of first word segments, and determining a second keyword from the first word segments; identifying the semantic features of the second keyword, and determining whether the keyword has an indication of having been completed according to the semantic features of the second keyword; if the second keyword has an indication of having been completed, determining that there is an indication of whether the previous conversation has been completed in the previous response information; if the second keyword does not have an indication of having been completed, determining that there is no indication of whether the previous conversation has been completed in the previous response information.

[0074] In another alternative embodiment, when the processor, under the prompt of the second model prompt word, identifies whether there is an indication of whether the previous conversation has been completed from the previous response information or from the previous question information and the previous response information, it is specifically used for: identifying the previous reply information and the previous response information from the previous conversation, performing word segmentation processing on the previous question information and the previous response information to obtain a plurality of second word segments, and determining a third keyword from the second word segments; identifying the semantic features of the third keyword, and determining whether the keyword has an indication of having been completed according to the semantic features of the third keyword; if the third keyword has an indication of having been completed, determining that there is an indication of whether the previous conversation has been completed in the previous question information and the previous response information; if the third keyword does not have an indication of having been completed, determining that there is no indication of whether the previous conversation has been completed in the previous question information and the previous response information.

[0075] In an alternative embodiment, when the processor, under the prompt of the first model prompt word, performs an intent classification operation according to the first keyword in the question information to determine the target intent type adapted to the user's current question information from multiple known intent types, it is specifically used for: performing word segmentation processing on the current question information to obtain a plurality of third word segments, and respectively performing feature extraction on the plurality of third word segments and each known intent type to obtain the semantic features of the plurality of third word segments and the semantic features of each known intent type; calculating the first similarity between the semantic features of each third word segment and the semantic features of each target intent type, and if there is a first target similarity greater than the first similarity threshold, taking at least one intent type corresponding to the first target similarity as the target intent type.

[0076] In an optional embodiment, when the processor performs an intent recognition operation under a target intent type to identify target intent information adapted to the current round of question information under the target intent type, it is specifically configured to: extract features from each known intent information under the target intent type to obtain semantic features of each intent information under each intent type; calculate a second similarity between the semantic feature of each third sub-word and the semantic features of each intent information under each intent type, and if there is a second target similarity greater than the second similarity threshold, use at least one intent information corresponding to the second target similarity as the target intent information.

[0077] In an optional embodiment, when the processor sends the target intent information to the intelligent customer service, it is specifically configured to: identify the complexity of the target intent information corresponding to the current round of question information based on a pre-configured intent complexity level; if the intelligent customer service is adapted to the complexity of the target intent information corresponding to the current round of question information, send the target intent information corresponding to the current round of question information to the intelligent customer service.

[0078] Further optionally, if the intelligent customer service is adapted to the complexity of the target intent information corresponding to the current round of question information, when the processor sends the target intent information corresponding to the current round of question information to the intelligent customer service, it is specifically configured to: input the target intent information corresponding to the current round of question information and a third model prompt word into an AI question-answering service model based on artificial intelligence corresponding to the intelligent customer service, where the third model prompt word is used to prompt the AI service model to provide a question-answering service for the user; under the prompt of the third model prompt word, generate target service information according to the target intent information corresponding to the current round of question information, and after modifying the target service information, output it to the customer service conversation interface as a reply message to the current round of question information.

[0079] Further optionally, the processor is further configured to: in the case of determining that the intelligent customer service is not adapted to the complexity of the target intent information, send at least the current round of question information to the human customer service for the human customer service to reply to the question information.

[0080] Further optionally, the processor is further configured to: pre-construct multiple intent types and multiple intent information under each intent type according to historical information related to customer service; perform a deduplication and priority sorting process on the multiple intent types and multiple intent information under each intent type to obtain the processed multiple intent types and multiple intent information under each intent type.

[0081] Further, as Figure 3 shown, the electronic device further includes: other components such as a communication component 30c, a display 30d, a power supply component 30e, and an audio component 30f. Figure 3 Only some components are schematically shown, and it does not mean that the electronic device only includes Figure 3 the components shown.

[0082] The detailed implementation manners and beneficial effects provided in the embodiments of the present application have been described in detail in the foregoing embodiments, and will not be elaborated herein.

[0083] An exemplary embodiment of the present application further provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to implement the steps in the foregoing method embodiments.

[0084] An exemplary embodiment of the present application further provides a computer program product, which includes computer programs / instructions that, when executed by a processor, cause the processor to be able to implement the steps in the foregoing method embodiments.

[0085] The foregoing memory may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic memory, flash memory, a magnetic disk, or an optical disc.

[0086] The foregoing communication component is configured to facilitate communication between the device where the communication component is located and other devices in a wired or wireless manner. The device where the communication component is located may access a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G / LTE, 5G and other mobile communication networks, or a combination thereof. In an exemplary embodiment, the communication component receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component further includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID) technology, Infrared Data Association (IrDA) technology, Ultra Wide Band (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0087] The above-mentioned display includes a screen, which may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors can sense not only the boundaries of touch or swipe actions but also detect the duration and pressure associated with the touch or swipe operation.

[0088] The above-mentioned power supply component provides power for various components of the device where the power supply component is located. The power supply component may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the device where the power supply component is located.

[0089] The above-mentioned audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC), which is configured to receive external audio signals when the device where the audio component is located is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signals can be further stored in the memory or sent via the communication component. In some embodiments, the audio component further includes a speaker for outputting audio signals.

[0090] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk memories, Compact Disc Read-Only Memories (CD-ROMs), optical memories, etc.) that contain computer-usable program code.

[0091] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate for implementing in the process Figure 1 one process or multiple processes and / or blocksFigure 1 means for the functions specified in one or more blocks.

[0092] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction means that implements the functions specified in one Figure 1 one or more processes and / or blocks Figure 1 means for the functions specified in one or more blocks.

[0093] These computer program instructions may also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are performed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one Figure 1 one or more processes and / or blocks Figure 1 means for the functions specified in one or more blocks.

[0094] In a typical configuration, a computing device includes one or more processors (Central Processing Unit, CPU), an input / output interface, a network interface, and memory.

[0095] The memory may include non-permanent memory in the computer-readable medium, in the form of random access memory (Random Access Memory, RAM) and / or non-volatile memory such as read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0096] Computer-readable media include permanent and non-permanent, removable and non-removable media that can store information by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change random access memory (Phase-change Random Access Memory, PRAM), static random access memory (SRAM), dynamic random access memory (Dynamic Random Access Memory, DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory

[0097] (EEPROM), flash memory, or other memory technologies, CD-ROM (Compact Disc Read-Only Memory), DVD (Digital Video Disc), or other optical storage, magnetic cassette tapes, magnetic disk storage, or other magnetic storage devices, or any other non-transitory medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer-readable medium does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0098] It should also be noted that the term "comprising", "including", or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but also other elements not expressly listed, or elements that are inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the statement "comprising one..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0099] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. A customer service method, characterized in that: include: Receive the current round of question information entered by the user on the customer service session interface; Inputting the question information of this round and the first model prompt word into the intention recognition model based on artificial intelligence, wherein the first model prompt word is used to prompt the intention recognition model to at least provide an intention hierarchical recognition service, wherein the intention hierarchical recognition service includes an intention classification operation and an intention recognition operation; Under the prompt of the first model prompt word, the intent classification operation is performed according to the first keyword in the current round of question information to determine a target intent type adapted to the current round of question information from a plurality of known intent types; and performing the intention recognition operation under the target intention type to recognize the target intention information adapted to the current round of question information under the target intention type; The target intention information is sent to the intelligent customer service, so that the intelligent customer service can output reply information corresponding to the question information according to the target intention information.

2. The method according to claim 1, characterized in that The target intention information is sent to the intelligent customer service, so that the intelligent customer service can output the reply information corresponding to the current round of question information according to the target intention information, including: Inputting the second model prompt word into the intention recognition model to determine whether the last round of dialogue has ended; the second model prompt word is used to prompt the intention recognition model to provide a recognition service of whether the last round of question information has ended; When it is determined that the previous round of dialogue has not ended, outputting the target intention information corresponding to the previous round of question information and the target intention information corresponding to the current round of question information; The target intention information corresponding to the current round of question information and the target intention information corresponding to the previous round of question information are sent to the intelligent customer service, so that the intelligent customer service can output corresponding reply information according to the target intention information corresponding to the current round of question information and the target intention corresponding to the previous round of question information.

3. The method according to claim 2, characterized in that Inputting the second model prompt word into the intention recognition model to determine whether the previous round of dialogue has ended, including: Acquire the last round of dialogue information between the user and the intelligent customer service, wherein the last round of dialogue information includes the last round of question information of the user and the last round of reply information of the intelligent customer service; Inputting the previous round of dialogue information and the second model prompt word into the intention recognition model, and under the prompt of the second model prompt word, identifying whether there is an intention indication indicating whether the previous round of dialogue has been completed from the previous round of reply information or from the previous round of question information and the previous round of reply information; If it exists, it is determined that the previous round of dialogue has ended; if it does not exist, it is determined that the previous round of dialogue has not ended.

4. The method according to claim 3, characterized in that Under the prompt of the second model prompt word, identifying whether there is an indication indicating whether the previous round of dialogue has been completed from the previous round of reply information or from the previous round of question information and the previous round of reply information, including: Identify the previous round of reply information from the previous round of dialogue information, perform word segmentation processing on the previous round of reply information to obtain multiple first word segmentations, and determine a second keyword from the first word segmentations; identify the semantic feature of the second keyword, and determine whether the keyword has a meaning of completion according to the semantic feature of the second keyword; if the second keyword has a meaning of completion, determine whether the previous round of reply information contains a meaning indicating whether the previous round of dialogue has been completed; if the second keyword does not have a meaning of completion, determine whether the previous round of reply information does not contain a meaning indicating whether the previous round of dialogue has been completed; or, The previous round of reply information and the previous round of reply information are identified from the previous round of dialogue information, the previous round of question information and the previous round of reply information are segmented to obtain multiple second segmented words, and a third keyword is determined from the second segmented words; the semantic feature of the third keyword is identified, and whether the keyword has a completed meaning expression according to the semantic feature of the third keyword; if the third keyword has a completed meaning expression, it is determined that the previous round of question information and the previous round of reply information contain a meaning expression indicating whether the previous round of dialogue has been completed; if the third keyword does not have a completed meaning expression, it is determined that the previous round of question information and the previous round of reply information do not contain a meaning expression indicating whether the previous round of dialogue has been completed.

5. The method according to claim 1, characterized in that Under the prompt of the first model prompt word, the intent classification operation is performed according to the first keyword in the question information to determine the target intent type adapted by the user's question information in this round from multiple known intent types, including: Performing word segmentation processing on the question information of the current round of questioning to obtain a plurality of third word segmentations, and performing feature extraction on the plurality of third word segmentations and each known intent type respectively to obtain semantic features of the plurality of third word segmentations and semantic features of each known intent type; A first similarity between the semantic features of each third word segment and the semantic features of each target intent type is calculated. If there is a first target similarity greater than a first similarity threshold, at least one intent type corresponding to the first target similarity is used as the target intent type.

6. The method according to claim 1, characterized in that Under the target intent type, performing the intent recognition operation to identify the target intent information adapted to the current round of question information under the target intent type includes: Extracting features of each known intention information under the target intention type to obtain semantic features of each intention information under each intention type; Calculate the second similarity between each third word segmentation semantic feature and the semantic features of each intent information under each intent type. If there is a second target similarity greater than the second similarity threshold, take at least one intent information corresponding to the second target similarity as the target intent information.

7. The method according to any one of claims 1 to 6, characterized in that: Sending the target intention information to the intelligent customer service includes: Based on a pre-configured intent complexity level, identifying the complexity of the target intent information corresponding to the current round of question information; If the intelligent customer service is adapted to the complexity of the target intention information corresponding to the current round of question information, the target intention information corresponding to the current round of question information is sent to the intelligent customer service.

8. The method according to claim 7, characterized in that If the complexity of the intelligent customer service is adapted to the target intention information corresponding to the current round of question information, sending the target intention information corresponding to the current round of question information to the intelligent customer service includes: Input the target intention information and the third model prompt word corresponding to the question information of this round into the AI ​​question-answering service model based on artificial intelligence corresponding to the intelligent customer service, wherein the third model prompt word is used to prompt the AI ​​service model to provide question-answering service for the user; Under the prompt of the third model prompt word, target service information is generated according to the target intention information corresponding to the current round of question information, and the target service information is modified and output to the customer service conversation interface as reply information for the current round of question information.

9. The method according to claim 7, characterized in that: Also includes: When it is determined that the intelligent customer service is not compatible with the complexity of the target intent information, at least the current round of question information is sent to the manual customer service so that the manual customer service can reply to the question information.

10. The method according to any one of claims 1 to 6, characterized in that: Also includes: Construct multiple intent types and multiple intent information under each intent type in advance based on historical information related to customer service; A deduplication and priority sorting process is performed on multiple intent types and multiple intent information under each intent type to obtain multiple processed intent types and multiple intent information under each intent type.

11. An electronic device, characterized in that: include: Memory and processor; The memory is used to store a computer program; the processor is coupled to the memory and is used to execute the computer program to implement the steps in the method according to any one of claims 1 to 10.

12. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the processor is caused to implement the steps in the method according to any one of claims 1 to 10.

13. A computer program product, characterized in that The computer program product comprises a computer program / instruction, which, when executed by a processor, enables the processor to implement the steps of any one of the methods of claims 1-10.

Images