Keyword identification method and device, storage medium and program product

Abstract

The embodiment of the invention provides a keyword recognition method and device, a storage medium and a program product. In the embodiment of the invention, the keyword is identified from the interaction information of the AI intelligent customer service and the user by combining the intention identification model and the keyword identification model with the corresponding cue word. Wherein the intention recognition model can be more accurately guided to understand the real intention of the user by using the first prompt word, so that the real demand of the user is accurately grasped; the keyword recognition model quickly recognizes the candidate keywords and the corresponding second adjustment operation according to the second cue word, so that the search range and recognition time of the keywords are greatly shortened; by identifying the intention of the user, the keyword matched with the intention of the user and the adjustment operation of the keyword are determined in a targeted manner, so that the keyword identification efficiency and accuracy are improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a keyword recognition method, device, storage medium and program product. Background Art

[0002] With the rapid development of Internet technology, online communication between people is becoming more and more frequent, especially information transmission in the form of dialogue and chat. In some service scenarios, merchants send a request to intelligent customer service to modify keywords in information units (such as posts) in the form of dialogue. Intelligent customer service identifies keywords and operations to be performed on keywords from the dialogue content, so that the service backend can perform corresponding operations on the identified keywords in the merchant's post.

[0003] At present, the text processing method based on term frequency (TF) and inverse document frequency (IDF) determines the TF-IDF value of each word according to the term frequency (TF) and inverse document frequency (IDF) of each word in the conversation content. The larger the TF-IDF value, the higher the possibility that the word is a keyword, so the keywords in the conversation content can be identified. However, the accuracy of identifying keywords in this way is low and cannot meet the usage requirements. Summary of the invention

[0004] Multiple aspects of the present application provide a keyword recognition method, device, storage medium and program product to improve the accuracy of keyword recognition.

[0005] The embodiment of the present application provides a keyword recognition method, including: obtaining at least one round of interaction information between a user and an AI intelligent customer service; filling the at least one round of interaction information into a first prompt word template including multiple adjustment operations to generate a first prompt word, where different adjustment operations correspond to different user intentions, and the first prompt word template also includes: first task description information, which is used to describe the task that the intention recognition model needs to perform; inputting the first prompt word into the intention recognition model, and identifying the user intention corresponding to at least one round of interaction information from multiple adjustment operations according to the first task description information, where the user intention includes at least one first adjustment operation; filling the at least one round of interaction information into a second prompt word template to generate a second prompt word, and the second prompt word template includes second task description information, which is used to describe the task that the keyword recognition model needs to perform; inputting the second prompt word into the keyword recognition model, and identifying at least one candidate keyword and its corresponding at least one second adjustment operation from at least one round of interaction information according to the second task description information; if at least one second adjustment operation matches the user intention, the second adjustment operation matching the user intention and its corresponding candidate keyword are respectively used as the target adjustment operation and target keyword identified from the at least one round of interaction information.

[0006] An embodiment of the present application also 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 each step in the keyword recognition method provided in the embodiment of the present application.

[0007] The embodiment of the present application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor implements the steps in the keyword identification method provided in the embodiment of the present application.

[0008] The embodiment of the present application also provides a computer program product, including a computer program / instruction. When the computer program / instruction is executed by a processor, the processor is caused to implement each step in the keyword identification method provided in the embodiment of the present application.

[0009] In the embodiment of the present application, the intention recognition model and the keyword recognition model are combined with the corresponding prompt words to identify keywords from the interactive information between the AI ​​intelligent customer service and the user. Among them, by using the first prompt word, the intention recognition model can be more accurately guided to understand the user's true intention, so as to accurately grasp the user's real needs; the keyword recognition model quickly identifies the candidate keywords and their corresponding second adjustment operations based on the second prompt word, greatly shortening the search range and recognition time of the keywords; by identifying the user's intention, the keywords and their adjustment operations that match the user's intention are specifically determined, thereby improving the efficiency and accuracy of keyword recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] 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 on the present application. In the drawings:

[0011] Figure 1 A flowchart of a keyword identification method provided for an exemplary embodiment of the present application;

[0012] Figure 2 A schematic diagram of a post-processing flow corresponding to a keyword recognition model provided for an exemplary embodiment of the present application;

[0013] Figure 3 A schematic structural diagram of an electronic device provided for an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0015] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0016] The various models involved in this application (including but not limited to language models or large models) are in compliance with relevant laws and standards.

[0017] Currently, manual customer service can modify the keywords of the posts published by merchants according to their needs. Merchants can find potential service demanders through posts with keywords to ensure the normal operation of the merchants. However, manual customer service may not respond in time, which will eventually lead to the loss of high-quality potential service demanders and merchants from the platform and a poor experience. In this context, how to quickly and accurately extract keywords from massive conversation content has become a problem that needs to be solved urgently.

[0018] An implementation method: Currently, most keyword extraction tools are based on traditional text processing methods. For example, they use word frequency (TF) and inverse document frequency (IDF) to mark the parts of speech in the text through grammatical analysis, thereby screening out keywords such as nouns and verbs, and extract keywords by finding related words in the text to form a vocabulary chain.

[0019] Traditional text processing methods (IF-IDF) are not effective for processing unstructured, colloquial text such as conversational chats, and are unable to capture deeper and more subtle features in the data. They are not accurate enough in understanding contextual information and are unable to determine the type of operation to be performed on keywords.

[0020] Another implementation method is to train a classifier or regression model to identify keywords. The model training method relies heavily on manual feature engineering and manually labeled data, resulting in insufficient accuracy in actual applications.

[0021] In order to solve the problem of low accuracy in identifying keywords, in the embodiment of the present application, the intention recognition model and the keyword recognition model are combined with the corresponding prompt words to identify keywords from the interaction information between AI intelligent customer service and users. Among them, by using the first prompt word, the intention recognition model can be more accurately guided to understand the user's true intention, so as to accurately grasp the user's real needs; the keyword recognition model quickly identifies the candidate keywords and their corresponding second adjustment operations based on the second prompt word, greatly shortening the search range and recognition time of keywords; by identifying the user's intention, the keywords and their adjustment operations that match the user's intention are determined in a targeted manner, thereby improving the efficiency and accuracy of keyword recognition.

[0022] Among them, the prompt word template provides the model with rich semantic information and contextual clues, which helps the model better understand the diversity and complexity of tasks and improve the accuracy of model recognition. Furthermore, there is no need to pre-train the model, nor to rely on manual feature engineering and manually labeled data, nor does it require complicated data pre-processing to understand the context.

[0023] A solution provided by an embodiment of the present application is described in detail below in conjunction with the accompanying drawings.

[0024] Figure 1 The flowchart of a keyword identification method provided by the exemplary embodiment of the present application is as follows. Figure 1 As shown, the method includes:

[0025] 101. Obtain at least one round of interaction information between the user and the AI ​​intelligent customer service;

[0026] 102. Fill at least one round of interaction information into a first prompt word template including multiple adjustment operations to generate a first prompt word, where different adjustment operations correspond to different user intentions, and the first prompt word template further includes: first task description information, which is used to describe the task that the intention recognition model needs to perform;

[0027] 103. Input the first prompt word into the intention recognition model, and identify the user intention corresponding to at least one round of interaction information from a plurality of adjustment operations according to the first task description information, where the user intention includes at least one first adjustment operation;

[0028] 104. Fill at least one round of interaction information into a second prompt word template to generate a second prompt word, wherein the second prompt word template includes second task description information, which is used to describe the task that the keyword recognition model needs to perform;

[0029] 105. Input the second prompt word into the keyword recognition model, and identify at least one candidate keyword and its corresponding at least one second adjustment operation from at least one round of interaction information according to the second task description information;

[0030] 106. If at least one second adjustment operation matches the user intention, the second adjustment operation matching the user intention and its corresponding candidate keyword are respectively used as the target adjustment operation and the target keyword identified from at least one round of interaction information.

[0031] In this embodiment, a life service platform is provided, which can be implemented through a mobile phone application (APP), a web page, or a mini program, without limitation. The life service platform is a link between merchants and service demanders. Service demanders can obtain the required services on the platform, and merchants can publish service units on the life service platform. Service units include but are not limited to: posts, personal dynamics, automatic replies, stories, activities, topics, etc. The content involved in the service unit may include but is not limited to: moving, cleaning, confinement nanny, maintenance, renting, used cars and other fields.

[0032] In this embodiment, the execution subject of the keyword recognition method can be a server or a user terminal (such as a merchant terminal). The user can have a dialogue with artificial intelligence (AI) intelligent customer service (referred to as AI intelligent customer service), and AI intelligent customer service provides customer service to the user. For example, if the user is a merchant, AI intelligent customer service can provide the merchant with publishing information units, deleting keywords in information units, adding keywords and modifying keywords, recharging, renewing, answering questions, etc. If the user is a service demander, AI intelligent customer service can provide recommended service units, appointment services, answering questions and ordering services for the service demander.

[0033] Among them, no matter which user is implemented, the user can have a conversation with AI intelligent customer service, and AI intelligent customer service can identify keywords from the conversation content and provide services to users based on the identified keywords. For example, in the keyword adjustment scenario, the keyword adjustment service in the service unit can be provided to the merchant based on the identified keywords. For another example, in the service unit recommendation scenario, the service unit recommendation service can be provided to the service demander based on the identified keywords.

[0034] It should be noted that AI intelligent customer service can be deployed on the user terminal or on the server side, without any limitation.

[0035] In this embodiment, the execution subject of the keyword identification method may be a server or a user terminal. The following description will be made by taking the execution subject being the server as an example.

[0036] Users can communicate with AI intelligent customer service through user terminals to express adjustment operations for keywords. Keywords are important words in information units that can reflect the theme, content or purpose of the post. Keywords are very important for understanding the core information of the post and help the life service platform to recommend content. Keywords vary according to the user's service category. Among them, the user's service category may include but is not limited to: moving, decoration, nanny, cleaning, car rental, maintenance and driving school. Among them, the keywords corresponding to the moving category may include but are not limited to: moving company, long-distance moving, short-distance moving, furniture disassembly and assembly, packaging service, warehousing service, moving expenses and appointment moving. The keywords corresponding to the decoration category may include but are not limited to: decoration company, interior design, decoration style, decoration materials, budget quotation, construction team, supervision service, environmentally friendly decoration and whole house customization. The keywords corresponding to the nanny category may include but are not limited to: live-in nanny, hourly worker, childcare nanny, confinement nanny, elderly care and tutor. The keywords under the cleaning category may include but are not limited to: daily cleaning, deep cleaning, land reclamation cleaning, carpet cleaning, air conditioning cleaning, glass cleaning and disinfection services. Keywords for car rental categories may include, but are not limited to: car rental companies, short-term rentals, long-term rentals, self-driving tour rentals, business rentals, wedding rentals, and airport transfer rentals. Keywords for maintenance categories may include, but are not limited to: home appliance maintenance, mobile phone maintenance, computer maintenance, fault diagnosis, door-to-door service, replacement of parts outside the warranty period, and data recovery. Keywords for driving school categories may include, but are not limited to: driving school registration, driving school fees, coach evaluation, driving license exams for subjects one to four, driving practice site simulation exams, and make-up exam arrangements.

[0037] The adjustment operations corresponding to the keywords may include but are not limited to: modification operation, addition operation, deletion operation, etc. Among them, the addition operation is used to add the identified keyword to the corresponding information unit; the deletion operation is used to delete the keyword from the information unit; the modification operation is used to modify the existing keywords in the information unit, for example, the modification operation includes refinement operation and replacement operation, the refinement operation is used to concretize the generalized keywords, such as refining "cleaning" into "deep cleaning" or "daily cleaning"; the replacement operation is used to replace the existing keywords in the existing information unit with more precise or professional terms.

[0038] Among them, the user terminal can conduct a dialogue between the user and the AI ​​intelligent customer service, generating at least one round of interaction information between the user and the AI ​​intelligent customer service. At least one round of interaction information can revolve around how to perform adjustment operations in the information unit for keywords. For example, at least one round of interaction information may include but is not limited to: Merchant: "Hello, I want to add a word to the post"; AI intelligent customer service: "OK, what do you want to add?"; Merchant: "Please help me add a 'range hood cleaning' to the post, thank you"; AI intelligent customer service: "OK".

[0039] In this embodiment, the user terminal may provide at least one round of interaction information between the user and the AI ​​intelligent customer service to the server. The server receives the at least one round of interaction information, and performs intent recognition on the at least one round of interaction information through the intent recognition model. The intent recognition model is used to perform intent recognition on at least one round of interaction information input by the user, and is mainly used to identify the adjustment operations performed by the user on the keywords. Different adjustment operations correspond to different intentions, that is, to identify which or which adjustment operations correspond to at least one round of interaction information. For the convenience of distinction and description, the adjustment operation included in the user intention recognized by the intent recognition model is referred to as the first adjustment operation. For example, the intent recognition model may be a large language model (LLM). LLM is a type of deep learning model specially designed to process and generate natural language text. These models are usually based on a neural network architecture and capture the complex patterns and structures of the language by training on a large amount of text data. LLM relies on the Transformer architecture. The main components of Transformer include: encoder, decoder, self-attention layer, and feed-forward neural network. The encoder is used to understand the semantics of the input text. The decoder is used to generate the output text. The self-attention layer allows the word at each position to pay attention to the words at other positions in the sentence, thereby better capturing contextual information. The feedforward neural network performs a nonlinear transformation on the representation of each position. When asked to generate text, the LLM can generate the next most likely word from left to right, word by word, based on a given prompt. Different sampling strategies can be applied in this process, such as greedy search, beam search, or temperature-adjusted sampling to control the diversity of the generated text.

[0040] Among them, the intention recognition model corresponds to a first prompt word template. The first prompt word template is used to guide the intention recognition model to generate text content that meets expectations (such as user intention), and the intention recognition model can more accurately identify the user's intention to adjust the operation of keywords. The first prompt word template may include but is not limited to: role, task, background, and requirements. Among them, the role is the profession or identity played by the intention recognition model, what personality characteristics the profession has, or what special skills the identity has, for example, the role played by the intention recognition model is a classifier; the background refers to some contextual information required by the role to complete the task, so that the intention recognition model can better understand the task background and related details. For example, adjustment operations for keywords, posts published by merchants, service categories corresponding to merchants, description information of merchants, etc. Tasks refer to specific tasks or goals completed by the role, for example, identifying the first adjustment operation that the user wants to perform from multiple adjustment operations. In order to facilitate description and distinction, the task information in the first prompt word template is called the first task description information, and the first task description information is used to describe the tasks that the intention recognition model needs to perform. Restrictions refer to specific requirements or restrictions for the role to complete the task, such as content requirements for output data and time limits for output data. Among them, content requirements may include but are not limited to: format requirements, output data type requirements, removal of sensitive information, removal of modal particles, etc.

[0041] In this embodiment, at least one round of interaction information can be filled into the first prompt word template to generate the first prompt word. For example, placeholder information corresponding to at least one round of interaction information can be reserved in advance in the first prompt word template, and the placeholder information can be directly replaced with at least one round of interaction information to generate the first prompt word; for example, the placeholder information includes a placeholder identifier and prompt information, such as $[prompt information], and the prompt text is used to briefly describe the content represented by the placeholder information, and the prompt information can be replaced by at least one round of interaction information to generate the first prompt word. For another example, the first prompt word template includes a blank prompt word, and the blank prompt word is a special placeholder information, which can appear in the form of a blank placeholder, such as "__" "____", etc. At least one round of interaction information can be directly filled into the blank prompt word to generate the first prompt word.

[0042] In this embodiment, the first prompt word is input into the intention recognition model, and in the intention recognition model, the user intention corresponding to the at least one round of interaction information is identified from a plurality of adjustment operations according to the first task description information. The user intention includes at least one first adjustment operation.

[0043] In this embodiment, in addition to identifying the user intent through the intent recognition model, the keyword recognition model can also be used to identify the keywords of the adjustment operation to be performed from at least one round of interaction information. The keyword recognition model is used to identify the keywords of the adjustment operation to be performed from at least one round of interaction information, and further, it can also identify which adjustment operation is performed on the keyword. For the convenience of distinction and description, the adjustment operation identified by the keyword recognition model is referred to as the second adjustment operation. For example, the keyword recognition model can be a large language model (Large Language Model, LLM). Among them, the introduction to LLM can be found in the aforementioned embodiment, which will not be repeated here.

[0044] It should be noted that the keyword recognition model can not only recognize keywords, but also recognize the adjustment operations performed on the keywords. In order to improve the accuracy of identifying the adjustment operations, the second adjustment operation output by the keyword recognition model can be matched with the user intent (including at least one first adjustment operation) recognized by the intent recognition model, and the second adjustment operation of the user intent in the match and the keyword for performing the second adjustment operation can be determined to improve the accuracy of identifying keywords.

[0045] In this embodiment, the keyword recognition model corresponds to a second prompt word template, and the second prompt word template is used to guide the keyword recognition model to generate text content that meets expectations (such as keywords and their corresponding adjustment operations). The second prompt word template may include, but is not limited to: roles, tasks, backgrounds, and requirements. Among them, the role is the profession or identity played by the keyword recognition model, what personality characteristics the profession has, or what special skills the identity has, for example, the role played by the keyword recognition model is a keyword extraction expert; the background refers to some contextual information required by the role to complete the task, so that the keyword recognition model can better understand the task background and related details. For example, the description information of the keywords included in each service category, the posts published by the merchant, the service categories corresponding to the merchant, the description information of the merchant, etc. The task refers to the specific task or goal completed by the role, for example, the adjustment operation and keywords that the user wants to perform are identified from at least one round of interactive information. In order to facilitate description and distinction, the task information in the second prompt word template is called the second task description information, and the second task description information is used to describe the tasks that the keyword recognition model needs to perform. Restrictions refer to the specific requirements or restrictions for the role to complete the task, for example, the content requirements of the output data and the time limit for the output data. Among them, content requirements may include but are not limited to: format requirements, output data type requirements, removal of sensitive information, removal of modal particles, etc.

[0046] In this embodiment, at least one round of interaction information can be filled into the second prompt word template to generate a second prompt word. For example, placeholder information corresponding to at least one round of interaction information can be reserved in advance in the second prompt word template, and the placeholder information can be directly replaced with at least one round of interaction information to generate the second prompt word; for example, the placeholder information includes a placeholder identifier and prompt information, such as $[prompt information], and the prompt text is used to briefly describe the content represented by the placeholder information, and the prompt information can be replaced with at least one round of interaction information to generate the second prompt word. For another example, the second prompt word template includes a blank prompt word, and the blank prompt word is a special placeholder information, which can appear in the form of a blank placeholder, such as "__" "____", etc. At least one round of interaction information can be directly filled into the blank prompt word to generate the second prompt word.

[0047] In this embodiment, the second prompt word is input into the keyword recognition model, and at least one candidate keyword and its corresponding at least one second adjustment operation are identified from at least one round of interaction information according to the second task description information. One candidate keyword corresponds to one second adjustment operation, indicating that the second adjustment operation can be performed on the candidate keyword.

[0048] In this embodiment, in order to identify keywords and their corresponding adjustment operations that are consistent with the actual needs of the user and avoid erroneous adjustments caused by misunderstanding the user's intention, at least one second adjustment operation can be matched with the user's intention. The matching can be for any second adjustment operation, and it is determined whether the any second adjustment operation is the same as at least one first adjustment operation included in the user's intention. If so, it is determined that the any second adjustment operation matches the user's intention; if not, it is determined that the any second adjustment operation does not match the user's intention. If at least one second adjustment operation matches the user's intention, the second adjustment operation that matches the user's intention is used as the target adjustment operation identified from at least one round of interaction information, and the candidate keyword corresponding to the second adjustment operation that matches the user's intention is used as the target keyword identified from at least one round of interaction information, so as to achieve precise matching, which helps to improve the accuracy of keyword recognition.

[0049] In the embodiment of the present application, the intention recognition model and the keyword recognition model are combined with the corresponding prompt words to identify keywords from the interactive information between the AI ​​intelligent customer service and the user. Among them, by using the first prompt word, the intention recognition model can be more accurately guided to understand the user's true intention, so as to accurately grasp the user's real needs; the keyword recognition model quickly identifies the candidate keywords and their corresponding second adjustment operations based on the second prompt word, greatly shortening the search range and recognition time of the keywords; by identifying the user's intention, the keywords and their adjustment operations that match the user's intention are specifically determined, thereby improving the efficiency and accuracy of keyword recognition.

[0050] Furthermore, by identifying keywords in the conversation content through intent recognition models and keyword recognition models, the life service platform can respond to the needs of users (such as merchants) in a timely manner, improve the accuracy of users in acquiring business opportunities (such as potential service demanders), increase user experience, and protect the interests of the platform.

[0051] In an optional embodiment, the first prompt word template further includes: third task description information, which is used to describe the task information executed by the intent recognition model. The implementation method of filling at least one round of interaction information into the first prompt word template including multiple adjustment operations to generate the first prompt word is not limited. An implementation method of filling at least one round of interaction information into the first prompt word template including multiple adjustment operations to generate the first prompt word includes: obtaining multiple information units corresponding to the user; filling at least one round of interaction information and multiple information units into the first prompt word template including multiple adjustment operations to generate the first prompt word; wherein, when the first prompt word is input into the intent recognition model, in addition to identifying the user intent corresponding to at least one round of interaction information from multiple adjustment operations according to the first task description information, at least one information unit corresponding to at least one adjustment operation included in the user intent can also be identified from multiple information units according to the third task description information; and the information unit corresponding to the target adjustment operation is used as the target information unit, and the target adjustment operation for the target keyword is performed in the target information unit.

[0052] The execution entity may maintain in advance the association between the user's identification information and its corresponding information unit. The user's identification information may include, but is not limited to, an identifier (ID), a universally unique identifier (Universally Unique Identifier), version information, validity period, source, label, geographic location, etc. The identification information of the information unit may include, but is not limited to, a name, ID, title, subject, and category, etc. The user's identification information may be combined with the association to obtain multiple information units corresponding to the user.

[0053] It should be noted that the description information and definition information of the information unit may also be added to the background information of the first prompt word template to assist in subsequently identifying the information unit corresponding to the first adjustment operation from multiple information units.

[0054] The implementation of filling the first prompt word template with multiple information units may refer to the aforementioned implementation of filling the first prompt word template with at least one round of interaction information, which will not be described in detail here.

[0055] The implementation method of performing the target adjustment operation for the target keyword in the target information unit is not limited. For example, the target information unit is implemented as a post published by a merchant, the post corresponds to the cleaning category, the target adjustment operation is implemented as an add operation, and the target keyword is "range hood cleaning", then the keyword "range hood cleaning" can be added to the post.

[0056] Among them, through the third task description information, it is possible to more accurately identify which information units are associated with specific adjustment operations, enhance the understanding of user intentions and the pertinence of responses. Furthermore, clearly performing target adjustment operations for target keywords in target information units helps to more directly meet user needs and provide more effective solutions.

[0057] In an optional embodiment, keyword recognition may be performed for all or part of at least one round of interaction information, or keyword recognition may be performed for interaction information round by round. Among them, a round of interaction information between a user and an AI intelligent customer service may include one interaction information of the user and one interaction information of the AI ​​intelligent customer service, and there is an association relationship between the two interaction information, for example, the AI ​​intelligent customer service outputs question information, and the user outputs reply information, and for another example, the user outputs instruction information, and the AI ​​intelligent customer service responds to the instruction information. An exemplary description is given below.

[0058] In one example, at least one round of interaction information is fully filled into a first prompt word template including multiple adjustment operations to generate a first prompt word; the first prompt word is input into an intention recognition model, and the user intention corresponding to at least one round of interaction information is identified from multiple adjustment operations according to the first task description information; at least one round of interaction information is fully filled into a second prompt word template to generate a second prompt word; the second prompt word is input into a keyword recognition model, and at least one candidate keyword and its corresponding at least one second adjustment operation are identified from at least one round of interaction information according to the second task description information; if at least one second adjustment operation matches the user intention, the second adjustment operation matching the user intention and its corresponding candidate keyword are respectively used as the target adjustment operation and target keyword identified from at least one round of interaction information. Among them, by inputting complete interaction information, the intention recognition model and the keyword recognition model can obtain a more comprehensive dialogue background and context, which helps to improve the understanding accuracy of the model, and can significantly improve the accuracy and coherence of the model understanding and response.

[0059] Another example is to perform keyword recognition on the interaction information in rounds. For any round of interaction information, any round of interaction information and its corresponding historical interaction information are filled into a first prompt word template including multiple adjustment operations to generate a first prompt word; wherein the historical interaction information can be the interaction information before any round of interaction information in at least one round of interaction information, or can be the historical interaction information generated between the user and the intelligent customer service at other times, and this is not limited to this; according to the first task description information, the user intent corresponding to any round of interaction information is identified from at least two adjustment operations; any round of interaction information and its corresponding historical interaction information are filled into a second prompt word template to generate a second prompt word; according to the second task description information, at least one candidate keyword and its corresponding at least one second adjustment operation are identified from any round of interaction information; if at least one second adjustment operation does not match the user intent, the historical interaction information is updated according to any round of interaction information, and keyword recognition is continued for the next round of interaction information until keyword recognition is performed on at least one round of interaction information.

[0060] The implementation method of updating the historical interaction information according to any round of interaction information is not limited. For example, the any round of interaction information can be added to the historical interaction information to update the historical interaction information.

[0061] It should be noted that, in the process of performing keyword identification on the interactive information one round at a time, candidate keywords and the second adjustment operation can be identified from one round of interactive information, or candidate keywords and the second adjustment operation are not identified from one round of interactive information, or the second adjustment operation is identified from the previous round of interactive information and the candidate keywords are identified from the next round, or the second adjustment operation is identified from the next round and the candidate keywords are identified from the previous round. There is no limitation on this.

[0062] The interaction information of the current round is considered together with the historical interaction information each time, so that the model can better adapt to the dynamic changes of the conversation and capture the user's latest needs or intention changes in a timely manner. For conversations with long spans or multiple rounds, the round-by-round processing method can effectively manage the amount of information and ensure that each stage receives appropriate attention without causing inefficiency or misunderstanding due to processing too much information at one time. The method of processing interaction information round by round and combining historical interaction information not only improves the accuracy of keyword recognition and user intent understanding, but also enhances the system's dynamic adaptability and personalized service level, thereby providing users with a more intelligent, efficient and friendly interaction environment.

[0063] Optionally, at least one round of interaction information and its corresponding historical interaction information are filled into the second prompt word template to generate the second prompt word. The implementation method is not limited. In this optional embodiment, the server maintains a service knowledge base, which includes service reference information under each service category, wherein the service reference information is used to describe or define the keywords under the service category to assist the keyword recognition model in identifying the keywords under the service category. For example, the keywords and their description information under the nanny category may include but are not limited to: child care: providing daily care for infants to teenagers, including feeding, bathing, accompanying play, etc.; housework management: responsible for tasks such as cleaning, washing, organizing clothes, and buying daily necessities in the home; elderly care: special care for the elderly, such as health monitoring, assistance in moving, medication reminders, etc.; meal preparation: preparing nutritionally balanced meals according to the needs and preferences of family members; special needs care: professional care for family members with special needs (such as disabled or chronic patients). Keywords and descriptions under the moving category include but are not limited to: Furniture disassembly and assembly: provide furniture disassembly and reinstallation services to ensure the safety and integrity of items during transportation; Packing services: use professional materials to pack customers' items to reduce the risk of damage during transportation; Long-distance relocation: suitable for long-distance moving needs across cities or provinces, providing full-process logistics solutions; Warehousing services: provide temporary or long-term cargo storage space to ensure the safe storage of items; International moving: handles the transfer of personal or corporate items across national borders, covering complex procedures such as customs declaration.

[0064] Based on this, an implementation method of filling at least one round of interaction information and its corresponding historical interaction information into a second prompt word template to generate a second prompt word includes: obtaining the target service category information corresponding to the user; matching the service reference information corresponding to the target service category from the service knowledge base according to the target service category information; filling at least one round of interaction information and its corresponding historical interaction information, as well as the service reference information into the second prompt word template to generate a second prompt word. Among them, each user corresponds to target service category information, and the target service category information can be selected by the user in the process of registering a resource account on the life service platform. The number of target service category information corresponding to the user can be one or more.

[0065] Accordingly, an exemplary implementation of identifying at least one candidate keyword and its corresponding at least one second adjustment operation from any round of interaction information according to the second task description information includes: identifying at least one candidate keyword and its corresponding at least one second adjustment operation from any round of interaction information according to the second task description information combined with service reference information. The service reference information provides professional vocabulary and concepts in a specific field, which helps to improve the recognition accuracy and relevance of the keyword recognition model.

[0066] Further optionally, any round of interaction information and its corresponding historical interaction information and its corresponding historical interaction information, as well as service reference information are filled into a second prompt word template to generate a second prompt word, including: parsing any round of interaction information to obtain user unilateral information and intelligent customer service unilateral information; describing the user unilateral information and the intelligent customer service unilateral information respectively according to a preset structured grammatical format to obtain first structured information and second structured information; filling the first structured information, the second structured information, any round of interaction information and its corresponding historical interaction information, and the service reference information into the second prompt word template to generate a second prompt word.

[0067] Among them, user unilateral information may include but is not limited to: user's questions or commands, etc., and intelligent customer service unilateral information may include but is not limited to: replies and responses provided by AI. Among them, the structured grammatical format may include but is not limited to: Domain Specific Language-JavaScriptObject Notation (JSON), Ordered Graph Data Language (OGDL), Data Serialization Language (YAML) and Extensible Markup Language (XML), etc.

[0068] For example, when the user is a merchant, the chat history between the merchant and the AI ​​intelligent customer service is divided into the format of "Merchant:#specific chat content#", "Customer service:#specific chat content#", where "#specific chat content#" can be replaced with the real conversation between the merchant and the AI ​​intelligent customer service. By clearly distinguishing the interaction information between the user and the AI ​​intelligent customer service and performing structured processing, confusion can be reduced, making it easier for the model to understand the intentions of each participant, and helping to identify keywords and adjust operations accordingly.

[0069] Among them, timestamps are used to determine the order in which interactive information occurs, which helps to understand the conversation flow and context.

[0070] In an optional embodiment, the second prompt word includes multiple operation lists corresponding to multiple adjustment operations, and one adjustment operation corresponds to one operation list; for example, the delete operation corresponds to the delete list, the add operation corresponds to the add list, and the modify operation corresponds to the modify list, etc. Based on this, the method provided in the embodiment of the present application also includes: for any second adjustment operation, adding at least one candidate keyword corresponding to it to the operation list corresponding to any second adjustment operation; judging whether there is a non-empty list in the multiple operation lists; if there is a non-empty list, judging whether the user intention includes the second adjustment operation corresponding to any non-empty list for any non-empty list; if it does, determining that the second adjustment operation corresponding to any non-empty list matches the user intention.

[0071] Among them, the above implementation only needs to check the non-empty list instead of traversing all possible operations, which greatly simplifies the processing flow and improves the response speed of the system. Especially when there are a large number of candidate keywords, this method can significantly reduce the amount of calculation. Optionally, if the user's intention does not include the second adjustment operation corresponding to any non-empty list, it is determined that the second adjustment operation corresponding to any non-empty list does not match the user's intention, and a notification message is sent to the manual customer service terminal for the manual customer service terminal to interact with the user terminal, and ask the user for keywords and their corresponding adjustment operations; if there is no non-empty list in the multiple operation lists, it means that the multiple operation lists are all empty lists, then it is determined that the target keyword is not identified, and the keyword identification is continued for the next round of interaction information, and when the set interaction round is reached, a notification message is sent to the manual customer service terminal for the manual customer service terminal to interact with the user terminal. The set interaction rounds can be 5 times, 10 times or 15 times, etc., and there is no limitation on this.

[0072] When automatic processing cannot meet user needs, timely switching to manual customer service can ensure that the service will not be interrupted, maintaining the consistency and satisfaction of the user experience. By setting the interaction rounds, it is possible to avoid wasting computing resources by endlessly trying to parse. When necessary, manual support is requested to achieve the best balance between automatic processing and manual intervention. It not only takes advantage of the speed and efficiency of AI, but also retains the flexibility and depth of manual customer service.

[0073] Optionally, the second prompt word also includes: output format information, the output format information indicates that the recognition result information is output according to a preset grammatical format, the preset grammatical format includes: a start identifier, an end identifier, and multiple operation list identifiers corresponding to multiple adjustment operations; according to the second task description information, identifying at least one candidate keyword and its corresponding at least one second adjustment operation from any round of interaction information, including: according to the second task description information, identifying keywords and their adjustment operations for any round of interaction information to obtain recognition result information; judging whether the recognition result includes a start identifier, an end identifier, and multiple operation list identifiers corresponding to multiple adjustment operations; if so, extracting at least one candidate keyword and its corresponding at least one second adjustment operation from the recognition result information according to the start identifier, the end identifier, and the multiple operation list identifiers.

[0074] For example, the recognition result information is output in the following json format, {"Add keywords": [keywords that the merchant requires to be added], "Delete keywords": [business keywords that the merchant requires to be deleted]}. Among them, "{" and "}" are examples of start identifiers and end identifiers, respectively, and "[]" is an example of an operation list identifier.

[0075] In an optional embodiment, the second task description information in the second prompt word defines the output data format of the keyword recognition model. In order to ensure the accuracy of the output result of the keyword recognition model, the output of the keyword recognition model needs to be post-processed. The following is an exemplary introduction, taking the recognition result information as a string, and the adjustment operation including the addition operation and the deletion operation as an example. Figure 2 shown.

[0076] 1) Use regular expressions to match the recognition result information (e.g., string) output by the keyword recognition model, such as determining whether the recognition result information is a set string (e.g., json format); for example, determining whether the string includes a start identifier, an end identifier, and multiple operation list identifiers corresponding to multiple adjustment operations. If included, the recognition result information is considered to be the set json string. If any identifier is included, the recognition result information is considered not to belong to the set json string, and the manual customer service is notified to extract and modify the keywords.

[0077] 2) If the recognition result information is the set json string, determine whether the list of added keywords and the list of deleted keywords are both empty; if both are empty, notify the AI ​​intelligent customer service to ask the merchant for the keywords that need to be adjusted;

[0078] 3) If the add list is not empty and the delete list is empty, it is necessary to determine whether the user's intention includes the add operation. If not, notify the manual customer service to ask the merchant about the keywords that need to be adjusted. Otherwise, output the add list.

[0079] 4) If the addition list is empty and the deletion list is not empty, it is necessary to determine whether the user's intention includes the deletion operation. If not, notify the manual customer service to ask the merchant for the keywords that need to be adjusted, otherwise output the deletion list.

[0080] 5) If both the add list and the delete list are not empty, directly output the add list and the delete list. It should be noted that the keyword recognition method used in the embodiment of the present application can be implemented on the terminal in addition to the server, and this is not limited. Figure 1 The detailed implementation and beneficial effects of each step in the method have been described in detail in the aforementioned embodiments and will not be elaborated here.

[0081] It should be noted that the execution subject of each step of the method provided in the above embodiment can be the same device, or the method can be executed by different devices. For example, the execution subject of steps 101 to 103 can be device A; for another example, the execution subject of steps 101 and 102 can be device A, and the execution subject of step 103 can be device B; and so on.

[0082] In addition, in some of the processes described in the above embodiments and the accompanying drawings, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel, and the sequence numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the sequence numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0083] Figure 3 A schematic diagram of the structure of an electronic device provided by an exemplary embodiment of the present application is shown in FIG. Figure 3 As shown, the device includes: a memory 34 and a processor 35.

[0084] The memory 34 is used to store computer programs and can be configured to store various other data to support operations on the electronic device. Examples of such data include instructions for any application or method used to operate on the electronic device.

[0085] The processor 35 is coupled to the memory 34 and is used to execute the computer program in the memory 34 to execute the keyword recognition method recorded in the aforementioned embodiments. Please refer to the records of the aforementioned embodiments for details, which will not be described again here.

[0086] Regarding the embodiments of the present application Figure 3 The detailed implementation and beneficial effects of each step in the illustrated device have been described in detail in the aforementioned embodiments and will not be elaborated on here.

[0087] Further, if Figure 3 As shown, the electronic device also includes: a communication component 36, a display 37, a power component 38, an audio component 39 and other components. Figure 3 Only some components are shown schematically, which does not mean that the electronic device only includes Figure 3 In addition, Figure 3 The components in the dashed box are optional components, not mandatory components, and the specific components depend on the product form of the electronic device. The electronic device of this embodiment can be implemented as a terminal device such as a desktop computer, a laptop computer, a smart phone, or an IOT (Internet of Things) device and a smart wearable device (such as a smart watch, a smart bracelet), or a conventional server, a cloud server, or a server array. If the electronic device of this embodiment is implemented as a terminal device such as a desktop computer, a laptop computer, a smart phone, etc., it can include Figure 3 If the electronic device of this embodiment is implemented as a server device such as a conventional server, a cloud server or a server array, it may not include Figure 3 Components within the dashed box.

[0088] Accordingly, the embodiment of the present application further provides a computer-readable storage medium storing a computer program, which can achieve the above-mentioned Figure 1 The steps in the method embodiment shown may be executed by an electronic device.

[0089] Accordingly, the embodiment of the present application also provides a computer program product, including a computer program / instruction, which, when executed by a processor, causes the processor to implement the above Figure 1 The steps in the method embodiment shown may be executed by an electronic device.

[0090] The above-mentioned memory can 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 storage, flash memory, magnetic disk or optical disk.

[0091] The above-mentioned communication component is configured to facilitate wired or wireless communication between the device where the communication component is located and other devices. The device where the communication component is located can 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 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be based on Radio Frequency Identification (RFID) technology, Infrared Data Association (IrDA) technology, Ultra Wide Band (UWB) technology, Bluetooth (BT) technology and other technologies.

[0092] The above-mentioned display includes a screen, and the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from a user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure associated with the touch or slide operation.

[0093] The power supply assembly provides power to various components of the device where the power supply assembly is located. The power supply assembly may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device where the power supply assembly is located.

[0094] The above-mentioned audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (Microphone, MIC), and when the device where the audio component is located is in an operating mode, such as a call mode, a recording mode, and a speech recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in a memory or sent via a communication component. In some embodiments, the audio component also includes a speaker for outputting an audio signal.

[0095] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-readable storage media (including but not limited to disk storage, compact disc read-only memory (Compact Disc Read-Only Memory, CD-ROM), optical storage, etc.) containing computer-usable program code.

[0096] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, 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, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0097] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0098] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0099] In a typical configuration, a computing device includes one or more processors (Central Processing Unit, CPU), input / output interface, network interface and memory.

[0100] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0101] Computer readable media include permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. 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 (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0102] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0103] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A keyword recognition method, characterized in that: include: Obtain at least one round of interaction information between the user and the AI ​​intelligent customer service; Filling the at least one round of interaction information into a first prompt word template including multiple adjustment operations to generate a first prompt word, where different adjustment operations correspond to different user intentions, and the first prompt word template further includes: first task description information, which is used to describe the task that the intention recognition model needs to perform; Inputting the first prompt word into the intention recognition model, and identifying the user intention corresponding to the at least one round of interaction information from the multiple adjustment operations according to the first task description information, wherein the user intention includes at least one first adjustment operation; Filling the at least one round of interaction information into a second prompt word template to generate a second prompt word, wherein the second prompt word template includes second task description information, which is used to describe the task that the keyword recognition model needs to perform; Inputting the second prompt word into the keyword recognition model, and identifying at least one candidate keyword and its corresponding at least one second adjustment operation from the at least one round of interaction information according to the second task description information; If the at least one second adjustment operation matches the user intention, the second adjustment operation matching the user intention and its corresponding candidate keyword are respectively used as the target adjustment operation and target keyword identified from the at least one round of interaction information.

2. The method according to claim 1, characterized in that The first prompt word template further includes: third task description information, filling the at least one round of interaction information into the first prompt word template including multiple adjustment operations to generate the first prompt word, including: Obtain multiple information units corresponding to the user; Filling the at least one round of interaction information and the multiple information units into a first prompt word template including multiple adjustment operations to generate a first prompt word; In the case where the first prompt word is input into the intention recognition model, the method further includes: identifying, according to the third task description information, at least one information unit corresponding to at least one first adjustment operation from the plurality of information units; and The information unit corresponding to the target adjustment operation is used as the target information unit, and the target adjustment operation for the target keyword is performed in the target information unit.

3. The method according to claim 1, characterized in that Filling the at least one round of interaction information into a first prompt word template including a plurality of adjustment operations to generate a first prompt word includes: For any round of interaction information, fill the any round of interaction information and its corresponding historical interaction information into a first prompt word template including multiple adjustment operations to generate a first prompt word; Identifying the user intention corresponding to the at least one round of interaction information from the at least two adjustment operations according to the first task description information includes: identifying a user intention corresponding to any round of interaction information from the at least two adjustment operations according to the first task description information; Filling the at least one round of interaction information into a second prompt word template to generate a second prompt word includes: Filling any round of interaction information and its corresponding historical interaction information into a second prompt word template to generate a second prompt word; According to the second task description information, identifying at least one candidate keyword and at least one second adjustment operation corresponding to the candidate keyword from the at least one round of interaction information includes: identifying at least one candidate keyword and at least one second adjustment operation corresponding to it from any round of interaction information according to the second task description information; If the at least one second adjustment operation does not match the user intention, the historical interaction information is updated according to any round of interaction information, and keyword recognition is continued for the next round of interaction information until keyword recognition is performed on at least one round of interaction information.

4. The method according to claim 3, characterized in that Filling the at least one round of interaction information and its corresponding historical interaction information into a second prompt word template to generate a second prompt word includes: Obtaining target service category information corresponding to the user; According to the target service category information, matching the service reference information corresponding to the target service category from a service knowledge base; Filling the at least one round of interaction information and its corresponding historical interaction information, and the service reference information into a second prompt word template to generate a second prompt word; According to the second task description information, identifying at least one candidate keyword and at least one second adjustment operation corresponding to it from the any round of interaction information includes: According to the second task description information in combination with the service reference information, at least one candidate keyword and its corresponding at least one second adjustment operation are identified from any round of interaction information.

5. The method according to claim 4, characterized in that Filling the any round of interaction information and its corresponding historical interaction information, and the service reference information into a second prompt word template to generate a second prompt word, including: Parse any round of interaction information to obtain user information and intelligent customer service information; Describing the user unilateral information and the intelligent customer service unilateral information respectively according to a preset structured grammatical format to obtain first structured information and second structured information; The first structured information, the second structured information, the any round of interaction information and its corresponding historical interaction information, and the service reference information are filled into a second prompt word template to generate a second prompt word.

6. The method according to claim 3, characterized in that The second prompt word includes a plurality of operation lists corresponding to a plurality of adjustment operations, and one adjustment operation corresponds to one operation list; the method further includes: For any second adjustment operation, adding at least one candidate keyword corresponding to the second adjustment operation to an operation list corresponding to the second adjustment operation; Determine whether there is a non-empty list in the multiple operation lists; If there is a non-empty list, for any non-empty list, determining whether the user intention includes a second adjustment operation corresponding to any non-empty list; If included, it is determined that the second adjustment operation corresponding to any non-empty list matches the user intention.

7. The method according to claim 6, characterized in that The method further comprises: If not, determining that the second adjustment operation corresponding to any non-empty list does not match the user intention, and sending notification information to the manual customer service terminal so that the manual customer service terminal interacts with the user terminal; If there is no non-empty list, it is determined that the target keyword is not recognized, and the keyword recognition continues for the next round of interaction information. When the set interaction round is reached, a notification message is sent to the manual customer service terminal for the manual customer service terminal to interact with the user terminal.

8. The method according to claim 6, characterized in that The second prompt word further includes: output format information, the output format information indicating outputting the recognition result information according to a preset grammatical format, the preset grammatical format including: a start identifier, an end identifier, and multiple operation list identifiers corresponding to the multiple adjustment operations; Identifying at least one candidate keyword and its corresponding at least one second adjustment operation from any round of interaction information according to the second task description information includes: According to the second task description information, identifying keywords and adjustment operations thereof for any round of interaction information to obtain identification result information; Determining whether the recognition result includes the start identifier, the end identifier, and multiple operation list identifiers corresponding to the multiple adjustment operations; If so, at least one candidate keyword and its corresponding at least one second adjustment operation are extracted from the recognition result information according to the start identifier, the end identifier and the plurality of operation list identifiers.

9. 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 8.

10. 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 8.

11. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the processor is caused to implement the steps in any one of the methods of claims 1-8.

Images