Service entry recommendation method and device of intelligent equipment and terminal equipment

Through the recommended method of service entrance of smart devices, using a large model server to handle user problems and generate search prompt content, select service entrances, solve the problem that users find it difficult to remember multiple service entrances, and improve the efficiency and user experience of smart device services .

CN118964731APending Publication Date: 2024-11-15NIO TECH ANHUI CO LTD
View PDF 8 Cites 0 Cited by

Patent Information

Application Number
CN202411023612.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-26
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

It is difficult for users to remember the various service portals of smart devices, resulting in reduced convenience in obtaining services and affecting user experience.

Method used

Receive and process user problems through the first large model server, generate response results, and generate search prompts based on the introduction information of multiple service portals, input them to the second large model server, and select the service portal that best meets user needs.

Benefits of technology

It improves the efficiency and user satisfaction of smart device services, reduces the time and energy cost of users to find and filter service entrances by themselves, and ensures that the recommended service entrances are highly matched with user needs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118964731A_ABST
    Figure CN118964731A_ABST
Patent Text Reader

Abstract

The invention is suitable for the technical field of artificial intelligence, and provides a service entry recommendation method and device for intelligent equipment and terminal equipment, and the service entry recommendation method comprises the steps: inputting a currently received user question into a first large model server, and obtaining a response result; according to the response result and the user question, generating a retrieval prompt content in combination with introduction information corresponding to a plurality of service entrances; and taking the retrieval prompt content as input of a second large model server, selecting a corresponding service entry from a plurality of service entries by the second large model based on the retrieval prompt content, and outputting corresponding service entry recommendation information. According to the scheme, the time and energy cost of searching and screening the service entrances by the user is reduced, convenience is provided for the user to a great extent, and the use experience of the user on the intelligent equipment is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application belongs to the field of artificial intelligence technology, and in particular, relates to a service entry recommendation method, device and terminal device for smart devices. Background Art

[0002] With the development of artificial intelligence technology, many smart devices, such as smart vehicles and smart phones, can provide users with more and more services. However, as the speed of service updates accelerates, it is difficult for users to clearly remember the location or logo of each service entrance, resulting in reduced convenience for users to obtain services and seriously affecting the user experience. Summary of the invention

[0003] The purpose of this application is to provide a service entry recommendation method, device and terminal device for a smart device, which can provide users with corresponding service entry recommendation information based on user questions, provide convenience to users to a great extent, and enhance users' experience of using smart devices.

[0004] A first aspect of an embodiment of the present application provides a service entry recommendation method for a smart device, the service entry recommendation method comprising:

[0005] Input the currently received user question into the first large model server to obtain the answer result;

[0006] Generate a search prompt content according to the answer result and the user question and in combination with the introduction information corresponding to the multiple service entrances;

[0007] The search prompt content is used as an input of the second largest model server. The second largest model selects a corresponding service entry from multiple service entries based on the search prompt content and outputs corresponding service entry recommendation information.

[0008] In an optional embodiment, before inputting the currently received user question to the first large model server to obtain a response result, the service entry recommendation method includes:

[0009] Determine the corresponding user intent type based on the currently received user questions and the classification model;

[0010] Determining a corresponding question-answering system according to the user intention type;

[0011] Based on the user question and in combination with the question-answering system, determining whether there is corresponding matching answer data in the question-answering system;

[0012] If not, the user question is input into the first large model server.

[0013] In an optional embodiment, the user intent type includes service-related intent, device-related intent, and mixed intent. If the user intent type is service-related intent, the corresponding first large model server includes a first retrieval architecture and a first generation architecture. The currently received user question is input into the first large model server to obtain a response result, which includes:

[0014] Inputting the currently received user question into the first search framework, combining the preset first knowledge resource database and the first question-answer resource database, and obtaining corresponding search results;

[0015] According to the search result, in combination with the first generation architecture, a corresponding response result is obtained.

[0016] In an optional embodiment, if the user intent type is a mixed intent, the corresponding first large model server includes a second retrieval architecture, a second generation architecture, and a large model architecture, and the currently received user question is input into the first large model server to obtain a response result including:

[0017] Inputting the currently received user question into the second search framework, combining with the preset second knowledge resource database and the second question-answer resource database, obtaining corresponding search results;

[0018] Determine whether there is a corresponding response result based on the search result and in combination with the second generation architecture;

[0019] If it does not exist, a search prompt content is generated according to the user question and input into the large model framework, and the large model framework outputs the corresponding answer result.

[0020] In an optional embodiment, if the user intent type is a device-related intent, the corresponding first large model server includes a third retrieval architecture, a third generation architecture, and a large model architecture, and the currently received user question is input into the first large model server to obtain a response result including:

[0021] Inputting the currently received user question into the third search framework, combining with the preset third knowledge resource database and the third question-answer resource database, to obtain corresponding search results;

[0022] Determine whether there is a corresponding answer result according to the search result and in combination with the third generation architecture;

[0023] If it does not exist, a search prompt content is generated according to the user question and input into the large model framework, and the large model framework outputs the corresponding answer result.

[0024] In an optional embodiment, after inputting the currently received user question to the first large model server and obtaining a response result, the service entry recommendation method further includes:

[0025] Inputting the response result into a regular matching module;

[0026] The regular matching module determines the corresponding matching service entry based on the response result, and outputs the corresponding service entry recommendation information.

[0027] In an optional embodiment, after generating a search prompt content according to the response result and the user question in combination with introduction information corresponding to a plurality of service entrances, the service entrance recommendation method includes:

[0028] The search prompt content is used as an input of a second large model server, and the second large model determines whether there is a corresponding service entry in a plurality of service entries based on the search prompt content;

[0029] If not, the service entry recommendation information that matches the user intent is determined based on the user intent corresponding to the user question and in combination with a preset matching list.

[0030] A second aspect of an embodiment of the present application provides a service entry recommendation device for a smart device, the service entry recommendation device comprising:

[0031] A response module, used for inputting the currently received user question into the first large model server to obtain a response result;

[0032] A generating module, configured to generate a search prompt content according to the answer result and the user question and in combination with the introduction information corresponding to the plurality of service entrances;

[0033] The recommendation module is used to use the search prompt content as input to the second largest model server. The second largest model selects a corresponding service entry from multiple service entries based on the search prompt content and outputs the corresponding service entry recommendation information.

[0034] A third aspect of an embodiment of the present application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect when executing the computer program.

[0035] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0036] A fifth aspect of the embodiments of the present application provides a computer program product. When the computer program product is run on a terminal device, the terminal device executes the method described in the first aspect.

[0037] Beneficial effects of this application

[0038] By adopting the service entrance recommendation method, device, terminal device and storage medium for smart devices provided in the present application, user questions are received and processed through the first large model server to ensure the accuracy and depth of question understanding, and to quickly generate answer results. The answer results and user questions are then used to generate search prompt content in combination with the introduction information corresponding to multiple service entrances, and the content is used as the input of the second large model server. The second large model server selects the corresponding service entrance from multiple service entrances based on further parsing the answer results and user questions, thereby ensuring that the recommended service entrance is highly matched with the actual needs of the user, reducing the time and energy cost of users to find and screen service entrances by themselves, and greatly improving the utilization efficiency of smart device services and user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 One of the flow charts of a service entry recommendation method for a smart device provided in an embodiment of the present application;

[0041] Figure 2 A second flowchart of a method for recommending a service entry for a smart device provided in an embodiment of the present application;

[0042] Figure 3 A schematic diagram of the service entry recommendation process provided in an embodiment of the present application;

[0043] Figure 4 A third flowchart of a method for recommending a service entry for a smart device provided in an embodiment of the present application;

[0044] Figure 5 A schematic diagram of the search process provided in the embodiment of the present application;

[0045] Figure 6 A fourth flowchart of a method for recommending a service entry for a smart device provided in an embodiment of the present application;

[0046] Figure 7Flow chart 5 of a method for recommending a service entry for a smart device provided in an embodiment of the present application;

[0047] Figure 8 Flow chart six of a service entry recommendation method for a smart device provided in an embodiment of the present application;

[0048] Fig. 9 Flow chart seven of a method for recommending a service entry for a smart device provided in an embodiment of the present application;

[0049] Fig.10 A schematic diagram of the structure of a service entry recommendation device for a smart device provided in an embodiment of the present application;

[0050] Fig.11 A user intent classification flow chart provided for an embodiment of the present application;

[0051] Fig.12 A schematic diagram of the structure of a terminal device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0053] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0054] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0055] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0056] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0057] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0058] It should be understood that the size of the serial numbers of the steps in this embodiment does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0059] With the development of artificial intelligence technology, many smart devices can provide users with more and more services. Taking smart vehicles as an example, the corresponding series of services for smart vehicles may include repairs, maintenance, car washing, vehicle insurance, chauffeur service, parking, etc. Due to the large number of service types, it is difficult for users to clearly remember all services and corresponding detailed terms. In addition, as the services are constantly updated and upgraded, the icons of each service entrance and their corresponding positions on the vehicle may also change, resulting in reduced convenience for users to call vehicle services, which greatly affects the user experience.

[0060] In this regard, the present application provides a method for recommending service entrances for smart devices, which receives and processes user questions through a first large model server to ensure the accuracy and depth of question understanding, and quickly generates answer results. The answer results and user questions are then used to generate search prompt content in combination with introduction information corresponding to multiple service entrances, and used as input to a second large model server. The second large model server selects a corresponding service entrance from multiple service entrances based on further analysis of the answer results and user questions, thereby ensuring that the recommended service entrance is highly matched with the user's actual needs, reducing the time and energy cost of users searching and screening service entrances by themselves, and greatly improving the utilization efficiency of smart device services and user satisfaction.

[0061] It should be noted that the large model mentioned above refers to a machine learning model with a large number of parameters and complex structure, which can process massive data and complete various complex tasks, such as natural language processing, computer vision, speech recognition, etc. The large model in this application can specifically adopt LLM (Large Language Model) and RAG (Retrieval Augmented Generation Model), etc. Through the use of large models, it is possible to achieve a fast and accurate response to user questions.

[0062] It should also be noted that the smart device in this application refers to a device that integrates computing processing capabilities and network communication technologies, and can provide smart services to users. Specific smart devices may include: smart vehicles, smart phones, smart home devices, and smart wearable devices.

[0063] In order to illustrate the technical solution of the present application, specific embodiments are provided below.

[0064] Reference Figure 1 , shows a flow chart of a service entry recommendation method for a smart device provided in an embodiment of the present application, which specifically includes the following steps:

[0065] Step S101: input the currently received user question into the first large model server to obtain a response result;

[0066] Step S102: generating a search prompt content according to the response result and the user question and combining the introduction information corresponding to the multiple service entrances;

[0067] Step S103: using the search prompt content as input to the second largest model server, the second largest model selects a corresponding service entry from a plurality of service entries based on the search prompt content, and outputs corresponding service entry recommendation information.

[0068] It should be noted that the user questions in step S101 can specifically include opening the designated driver page, closing the designated driver page, placing a designated driver order, and asking designated driver questions, etc., and can be asked by the user to the smart device through voice input or text input. The first large model server used in step S101 is a server with high natural language processing capabilities, which can quickly respond to user questions and generate corresponding answer results to provide basic data support for subsequent steps.

[0069] Since the user's question is expressed broadly, the meaning of the generated answer result content is vague, which makes it impossible to accurately give the service portal recommendation information based on the answer result. Therefore, it is necessary to further combine the large model for processing. Correspondingly, step S102 analyzes the answer result and the content of the user's question through the system, identifies the key information therein, and combines the introduction information of multiple service portals to generate a search prompt content, provide more targeted input data for subsequent steps, and guide the second largest model server to make more accurate service portal selection, wherein the introduction information of the service portal may specifically include the functions, characteristics and usage conditions of each service; step S103 inputs the search prompt content to the second largest model server with a high degree of natural language processing capability, and the second largest model server parses the search prompt content, so as to select the service portal that best meets the user's needs from multiple service portals, and return the corresponding service portal recommendation information to the user, thereby improving the accuracy and pertinence of the service recommendation, wherein the second largest model server can adopt the LLM model, and the service portal recommendation information may specifically include the service portal's logo, link and brief description.

[0070] As an example, in the vehicle voice dialogue system, the first largest model server built-in is the RAG model server, and the second largest model server built-in is the LLM model server. The user question currently received is set as "vehicle engine failure", and the user question is input into the RAG model server for processing. The RAG model server outputs a response result including the cause of the failure and handling suggestions. The specific content of the response result includes "you can seek help from the company's maintenance personnel". Combined with the response result, user questions and introduction information of multiple service portals, retrieval prompt content is generated, and the retrieval prompt content is used as input to the LLM model server. After identification and processing by the LLM model server, the corresponding company maintenance service portal recommendation information is output. The user can directly click on the link in the service portal recommendation information to enter the maintenance personnel communication interface and communicate with the maintenance personnel.

[0071] The above embodiment, through the application of the first large model server and the second large model server, selects the service entrance corresponding to the user question based on a full analysis of the response results and the user question, ensures that the recommended service entrance can highly match the user's actual needs, and greatly improves the use efficiency of smart device services and user experience.

[0072] It should be noted that this application involves the recommendation of service entrances. Different user questions correspond to different user intentions. Therefore, a classification model can be used to classify user intentions related to services, so that corresponding service entrance recommendations can be made for each type of user intention in a targeted manner to improve processing efficiency. Figure 2As an example, before step S101, the service entrance recommendation method provided in the embodiment of the present application includes the following steps:

[0073] Step S201: Determine the corresponding user intention type based on the currently received user question and the classification model;

[0074] Step S202: determining a corresponding question-answering system according to the user intention type;

[0075] Step S203: Based on the user question and in combination with the question-answering system, determining whether there is corresponding matching answer data in the question-answering system;

[0076] Step S204: If it does not exist, the user question is input into the first large model server.

[0077] Specifically, the classification model in step S201 can specifically adopt the BERT (Bidirectional Encoder Representations from Transformers) model. The BERT model can accurately understand user questions and determine their intent types through its powerful language representation ability. Through step S201, the user's real needs can be quickly and accurately identified, providing a basis for subsequent steps; different user intent types correspond to different question-and-answer systems, and each question-and-answer system contains a knowledge base, algorithms and rules in a specific field so that user questions can be answered more accurately. Step S202 calls the corresponding question-and-answer system to process user questions according to the determined user intent type, which can improve the accuracy and processing efficiency of question answering; step S203 searches the question-and-answer system to see if there is answer data matching the user question. If not, execute step S204 to input the user question into the first large model server for further processing, which provides a solution for some user questions that are beyond the capabilities of the current question-and-answer system, and at the same time improves the response capability of smart devices to unknown or complex user questions.

[0078] It should be noted that user intent types may specifically include service-related intents, device-related intents, and mixed intents. Service-related intents refer to user intents that are simply related to the service field, such as chauffeur service, parking service, tire change, maintenance, etc. Mixed intents refer to user intents that overlap service-related intents with other intents, such as vehicle breakdowns, emergencies, etc. Device-related intents refer to user intents that are simply related to the general knowledge field of equipment, such as engine location, vehicle performance, etc. The question-and-answer system may specifically adopt a FAQ (frequently-asked questions) question-and-answer system. Different user intent types each correspond to a FAQ question-and-answer system. Different FAQ question-and-answer systems include different question-and-answer resources. Specifically, the question-and-answer resources within the FAQ question-and-answer system corresponding to service-related intents belong to the service field, the question-and-answer resources within the FAQ question-and-answer system corresponding to device-related intents belong to the general knowledge field of equipment, and the question-and-answer resources within the FAQ question-and-answer system corresponding to mixed intents include the service field and other device-related fields.

[0079] For example, refer to Figure 3 , set the user question as "vehicle engine failure", and use the BERT model to identify that the user intent corresponding to the user question includes vehicle maintenance service consultation and engine usage knowledge, which is a mixed intent. Therefore, based on the user question and the FAQ question-and-answer system corresponding to the mixed intent, determine whether there is corresponding matching answer data in the question-and-answer system. If not, the user question is input into the first large model server for further processing.

[0080] The above steps combine the advantages of classification models and question-answering systems to accurately identify user intentions, quickly retrieve response data, and call on higher-level processing resources to deal with complex problems when necessary, thereby meeting user needs to a great extent.

[0081] In another embodiment that can be implemented in the present application, service entry recommendation information corresponding to the answer data can be set in advance in the FAQ question and answer system, and step S203 is executed to search whether there is answer data matching the user's question in the question and answer system. If so, the service entry recommendation information corresponding to the answer data can be directly fed back to the user. The user can obtain the service entry link or logo through the service entry recommendation information, so that he can smoothly enter the corresponding service interface.

[0082] The following is a detailed description of step S101.

[0083] In the embodiments that can be implemented in the present application, user intent types include service-related intent, device-related intent and mixed intent. The architecture of the first large model server corresponding to different user intent types is different, which are explained one by one below.

[0084] Reference Figure 3 , setting the user intent type corresponding to the current user question as service-related intent, and its corresponding first large model server includes a first retrieval architecture and a first generation architecture, referring to Figure 4 , step S101 may specifically include the following steps:

[0085] Step S301: input the currently received user question into the first search framework, and obtain corresponding search results by combining the preset first knowledge resource database and the first question-and-answer resource database;

[0086] Step S302: Obtain corresponding response results based on the search results and the first generation architecture.

[0087] It should be noted that the first retrieval architecture in the above steps is the part of the first large model server responsible for retrieving relevant information. It has a retrieval function and can search for the most relevant content or answers to the questions in the preset first knowledge resource database and the first question and answer resource database according to the user's questions. The first generation architecture is the part of the first large model server responsible for generating response results. It has a high degree of natural language processing capabilities and can receive the relevant information found by the retrieval architecture, and combine the context of the user's questions to generate a coherent, accurate and user-friendly answer. The specific first retrieval architecture and the first generation architecture can correspond to the RAG model. In addition, in order to improve the RAG model's ability to understand, reason and respond to user questions and various resource data, data in the corresponding fields can be constructed, and the RAG model can be trained by SFT (Supervised Fine-Tuning), thereby improving the model's specialized capabilities.

[0088] It should also be noted that the first knowledge resource database preset in the first search architecture stores a large amount of service-related knowledge data, and the first question-answer resource database stores service-related historical question-answer records and preset question-answer pairs for use by the first search architecture when processing user questions; Figure 5 When establishing the first knowledge resource database and the first question and answer resource database, since the data resources obtained by all parties are relatively divergent and the formats and contents are not uniform, it is necessary to pre-process the data resources, such as dividing and cleaning the service knowledge resources to form high-quality service knowledge resources and importing them into the first knowledge resource database, appropriately reducing the more lengthy question and answer resources in the service question and answer resources, or appropriately supplementing the short question and answer resources, and generalizing them, providing different expression methods to make subsequent retrieval more accurate, and importing the processed high-quality service question and answer resources into the first question and answer resource database, thereby obtaining the first knowledge resource database and the first question and answer resource database.

[0089] For specific search result determination process, please refer to Figure 5 Step S301: After inputting the currently received user question into the first search framework, the first knowledge resource database and the first question-answer resource database can be searched for preliminary search content by using search methods such as vector search and literal search. After deduplication and filtering operations, the preliminary search content is scored and sorted by using the re-rank model. The scored and sorted results are filtered by using a threshold method, and the top K (top K) results are taken. k) results are used as search results. The above operation can ensure that the search results are fast and accurate, and improve the efficiency and accuracy of subsequent generation of answer results. Step S302 assembles the search results into prompt words and inputs them into the first generation architecture, so as to quickly obtain the corresponding answer results. Specifically, when assembling the search content corresponding to the first knowledge resource database, the first generation architecture can be informed that the following paragraph content may contain information and content that can answer the user's question, and you can try to extract the answer from it. When assembling the search content corresponding to the first question and answer resource database, the first generation architecture can be informed that the above search content is the most similar question and answer pair retrieved, which may have a direct or indirect effect on answering the user's question. In addition, the first generation architecture needs to be informed that when the user's question cannot be answered based on the search content, a bottom-up reply needs to be given, rather than answering randomly based on illusions. In addition, it is also necessary to limit the reply of the first generation architecture to be direct and concise, and no irrelevant or non-important information is output.

[0090] It should be noted that vector retrieval can specifically use BGE embedding (general vector model) and the like, and literal retrieval can specifically use Elastic Search (search engine) and the like; the above threshold needs to be selected using a validation set (evaluation dataset). The score of each search content is output in the validation set with positive and negative examples, and a reasonable threshold is selected so that most of the positive examples are above the threshold and most of the negative examples are below the threshold. If it is found difficult to select such a threshold through the validation set, then some positive and negative example data can be constructed, and a re-rank model can be trained using a loss function such as Noise Contrastive Estimation (NCE) loss to increase the score of positive and negative examples, so that the threshold can be selected more reasonably. It should be noted that the thresholds are selected separately for the preliminary search content retrieved from the first knowledge resource database and the first question and answer resource database; in addition, the selection of the k value of top k can be determined according to business needs and the experimental results of the validation set. According to the existing validation results, it can be concluded that k = 1 or 2 can meet user needs.

[0091] Reference Figure 3, set the user intent type corresponding to the current user question to be a mixed intent, and the corresponding first large model server includes a second retrieval architecture, a second generation architecture, and a large model architecture, referring to Figure 6 , step S101 may specifically include the following steps:

[0092] Step S401: input the currently received user question into the second search framework, and obtain corresponding search results by combining the preset second knowledge resource database and the second question-answer resource database;

[0093] Step S402: Determine whether there is a corresponding response result based on the search result and in combination with the second generation architecture;

[0094] Step S403: If it does not exist, a search prompt content is generated according to the user question and input into the large model framework, and the large model framework outputs the corresponding answer result.

[0095] Specifically, the second retrieval architecture in the above steps is the part of the first large model server responsible for retrieving relevant information. It has a retrieval function and can search for the most relevant content or answers to the questions in the preset second knowledge resource database and the second question and answer resource database according to the user's questions. The second generation architecture and the large model architecture are the parts of the first large model server responsible for generating response results and have a high degree of natural language processing capabilities. The second generation architecture can receive the relevant information found by the retrieval architecture, and combine the context of the user's question to generate a coherent, accurate and user-friendly response. The large model architecture can further make a more accurate response based on the response of the second generation architecture. Specifically, the second retrieval architecture and the second generation architecture can correspond to the RAG model, and the large model architecture can correspond to the LLM model. In order to improve the understanding, reasoning and response capabilities of each model to user questions and various resource data, data in the corresponding fields can be constructed, and SFT (Supervised Fine-Tuning) training can be performed on each model to enhance the model's specialized capabilities.

[0096] It should be noted that the establishment process of the second knowledge resource database and the second question and answer resource database is similar to the establishment process of the aforementioned first knowledge resource database and the first question and answer resource database. The difference is that the second knowledge resource database not only stores knowledge data related to the service, but also includes knowledge data in other fields, such as car use knowledge, etc. The second question and answer resource database stores historical question and answer records related to services and other fields, as well as preset question and answer pairs.

[0097] For specific implementation process, please refer to Figure 5In step S401, after the currently received user question is input into the second search framework, the first search content can be retrieved in the second knowledge resource database and the second question-answer resource database respectively by using search methods such as vector search and literal search. After deduplication and filtering operations, the first search content is scored and sorted by using the re-rank model. The scored and sorted results are filtered by using a threshold method, and the top K (top K) are taken. k) results are used as the search results. The above operation can ensure that the search results are fast and accurate, and improve the efficiency and accuracy of the subsequent generation of answer results. Step S402 assembles the search results into prompt words, inputs them into the second generation architecture, and determines whether there is a corresponding answer result. Specifically, when assembling the search content corresponding to the second knowledge resource database, the second generation architecture can be informed that the following paragraph content may contain information and content that can answer the user's question, and the answer can be extracted from it. When assembling the search content corresponding to the first question and answer resource database, the second generation architecture can be informed that the above search content is the most similar question and answer pair retrieved, which may have a direct or indirect effect on answering the user's question. In addition, the second generation architecture needs to be informed that when the user's question cannot be answered based on the search content, a bottom-up reply needs to be given, rather than answering randomly based on illusions. In addition, it is also necessary to limit the reply of the second generation architecture to be direct and concise, and no irrelevant or non-important information is output. If step S402 does not give the corresponding matching answer data, step S403 is executed to assemble a search prompt content using the user's question as the input of the large model architecture, and the large model architecture outputs the corresponding answer result.

[0098] The above steps quickly and accurately obtain relevant information in various databases through refined retrieval, and assemble this information into prompt words and input them into the generated architecture in an attempt to obtain direct and concise responses. If the generated architecture cannot match the response, the large model architecture is used as a supplement to output comprehensive answers based on user questions, thereby ensuring the comprehensiveness and accuracy of data processing.

[0099] Reference Figure 3 , setting the user intent type corresponding to the current user question to device-related intent, and the corresponding first large model server includes a third retrieval architecture, a third generation architecture, and a large model architecture, referring to Figure 7 , step S101 may specifically include the following steps:

[0100] Step S501: input the currently received user question into the third search framework, and obtain corresponding search results by combining the preset third knowledge resource database and the third question-answer resource database;

[0101] Step S502: Determine whether there is a corresponding response result based on the search result and in combination with the third generation architecture;

[0102] Step S503: If it does not exist, a search prompt content is generated according to the user question and input into the large model framework, and the large model framework outputs the corresponding answer result.

[0103] Specifically, the third retrieval architecture in the above steps is the part of the first large model server responsible for retrieving relevant information. It has a retrieval function and can search for the most relevant content or answers to the questions in the preset third knowledge resource database and the third question and answer resource database according to the user's questions. The third generation architecture and the large model architecture are the parts of the first large model server responsible for generating response results. They have a high degree of natural language processing capabilities. The third generation architecture can receive the relevant information found by the retrieval architecture, and combine the context of the user's question to generate a coherent, accurate and user-friendly response. The large model architecture can further make a more accurate response based on the response of the third generation architecture. Specifically, the third retrieval architecture and the third generation architecture can correspond to the RAG model, and the large model architecture can correspond to the LLM model. In order to improve the understanding, reasoning and response capabilities of each model to user questions and various resource data, data in the corresponding fields can be constructed, and SFT (Supervised Fine-Tuning) training can be performed on each model to enhance the model's specialized capabilities.

[0104] It should be noted that the establishment process of the third knowledge resource database and the third question and answer resource database is similar to the establishment process of the aforementioned first knowledge resource database and the first question and answer resource database. The difference is that the third knowledge resource database stores knowledge data related to general knowledge of the equipment, such as car usage knowledge, etc., and the third question and answer resource database stores historical question and answer records related to general knowledge of the equipment and preset question and answer pairs.

[0105] For specific implementation process, please refer to Figure 5In step S501, after the currently received user question is input into the third search framework, the first search content can be retrieved in the third knowledge resource database and the third question-answer resource database respectively by using search methods such as vector search and literal search. After deduplication and filtering operations, the first search content is scored and sorted by using the re-rank model. The scored and sorted results are filtered by using a threshold method, and the top K (top K) are taken. k) results are used as the search results. The above operation can ensure that the search results are fast and accurate, and improve the efficiency and accuracy of the subsequent generation of answer results. Step S502 assembles the search results into prompt words, inputs them into the third generation architecture, and determines whether there is a corresponding answer result. Specifically, when assembling the search content corresponding to the third knowledge resource database, the third generation architecture can be informed that the following paragraph content may contain information and content that can answer the user's question, and the answer can be extracted from it. When assembling the search content corresponding to the third question and answer resource database, the third generation architecture can be informed that the above search content is the most similar question and answer pair retrieved, which may have a direct or indirect effect on answering the user's question. In addition, the third generation architecture needs to be informed that when the user's question cannot be answered based on the search content, a bottom-up reply needs to be given, rather than answering randomly based on illusions. In addition, it is also necessary to limit the reply of the third generation architecture to be direct and concise, and no irrelevant or non-important information is output. If step S502 does not give the corresponding matching answer data, step S503 is executed to assemble a search prompt content using the user's question as the input of the large model architecture, and the large model architecture outputs the corresponding answer result.

[0106] The above steps quickly and accurately obtain relevant information in various databases through refined retrieval, and assemble this information into prompt words and input them into the generated architecture in an attempt to obtain direct and concise responses. If the generated architecture cannot match the response, the large model architecture is used as a supplement to output comprehensive answers based on user questions, thereby ensuring the comprehensiveness and accuracy of data processing.

[0107] In the embodiments that can be implemented in this application, in order to improve the convenience of users to obtain corresponding services, after executing step 101 and obtaining the response result, the response result can be directly input into a regular matching module to determine whether there is a corresponding matching service entry, so that a quick response can be made. Figure 8 The service entry recommendation method provided in the embodiment of the present application also includes the following steps:

[0108] Step S601: input the response result into a regular matching module;

[0109] Step S602: The regular matching module determines the corresponding matching service entry based on the response result, and outputs the corresponding service entry recommendation information.

[0110] It should be noted that the regular matching module is a module that uses regular expressions for pattern matching. Specifically, the regex regular module can be used to identify high-frequency content with a high frequency of occurrence in the response result through the regular matching module, and recommend the corresponding service entry based on the high-frequency content. Exemplarily, if the response result contains content such as "It is recommended to call the official hotline" or "Vehicle insurance services can meet your needs", the corresponding service entry can be matched through the regex regular module, and the corresponding service entry recommendation information can be returned to the user for selection. Since the effective scope of the above matching method is relatively limited, if the corresponding service entry is not matched, it is necessary to execute steps S102 and S103 to further use the large model to make accurate service entry recommendations.

[0111] Reference Figure 3 In another embodiment that can be implemented in the present application, if step S103 is executed, the search prompt content is used as input to the second largest model server, but the second largest model server does not provide a corresponding service entrance, or the user is not satisfied with the recommended service entrance, the problem can be solved by using the preset matching list. For details, refer to Fig. 9 The service entry recommendation method provided in the embodiment of the present application specifically includes the following steps:

[0112] Step S701: taking the search prompt content as input to the second largest model server, and the second largest model determines whether there is a corresponding service entry in multiple service entries based on the search prompt content;

[0113] Step S702: If not, determine the service entry recommendation information that matches the user intent based on the user intent corresponding to the user question and in combination with a preset matching list.

[0114] Specifically, the preset matching list contains user intentions related to multiple service areas, referred to as service intentions, and service entry recommendation information corresponding to the service intentions. The user intention corresponding to the user question is analyzed, and then a search is performed in the preset matching list to determine whether the same service intention exists. If so, the service entry recommendation information corresponding to the service intention is returned to the user. If not, information that the service entry does not exist is returned.

[0115] Through the above embodiments, a solution is provided for the problem that the second largest model server does not provide a corresponding service entrance, or the user is not satisfied with the recommended service entrance, further ensuring that the user can have a higher usage experience.

[0116] Fig.10This is a structural diagram of a service portal recommendation device for a smart device provided in an embodiment of the present application. For the sake of convenience of explanation, only the parts related to the embodiment of the present application are shown.

[0117] The service entry recommendation device may specifically include the following modules:

[0118] The answering module M1 is used to input the currently received user question into the first large model server to obtain the answer result;

[0119] A generating module M2, for generating a search prompt content according to the answer result and the user question, combined with the introduction information corresponding to the plurality of service entrances;

[0120] The recommendation module M3 is used to use the search prompt content as the input of the second largest model server. The second largest model selects a corresponding service entry from multiple service entries based on the search prompt content and outputs the corresponding service entry recommendation information.

[0121] Specifically, the answer module M1 inputs the currently received user question into the first large model server, which quickly responds to the user question and generates a corresponding answer result, providing basic data support for subsequent steps; the generation module M2 systematically analyzes the answer result and the content of the user question, identifies the key information therein, and combines the introduction information of multiple service portals to generate a search prompt content, providing more targeted input data for subsequent steps, and guiding the second large model server to make more accurate service portal selection; the recommendation module M3 inputs the search prompt content into the second large model server with a high degree of natural language processing capabilities, and the second large model service parses the search prompt content, so as to select the service portal that best meets user needs from multiple service portals, and return the corresponding service portal recommendation information to the user, thereby improving the accuracy and pertinence of service recommendations.

[0122] The above-mentioned service portal recommendation device receives and processes user questions through the first large model server to ensure the accuracy and depth of understanding of the questions, and quickly generates answer results. Then, the answer results and user questions are used to combine the introduction information corresponding to multiple service portals to generate search prompt content, and use it as the input of the second large model server. The second large model server selects the corresponding service portal from multiple service portals based on further analysis of the answer results and user questions, thereby ensuring that the recommended service portal is highly matched with the actual needs of the user, reducing the time and energy cost of users to find and screen service portals by themselves, and greatly improving the utilization efficiency of smart device services and user satisfaction.

[0123] It should be noted that if the above-mentioned service entry recommendation device is set in the on-board voice dialogue system of the smart electric vehicle, it has the ability to answer user questions related to the service, and other devices in the on-board voice dialogue system also have the ability to answer some user questions. Therefore, in order to avoid conflicts, the BERT model can be used to distribute service-related user questions to the service entry recommendation device for processing.

[0124] For example, refer to Fig.11 , the user intentions corresponding to user questions can be split into finer granularity, and the service-related user intentions can be distributed to the service entry recommendation device for processing. Taking the designated driver service as an example, the corresponding intentions can be: opening the designated driver page, closing the designated driver page, placing a designated driver order, asking designated driver questions, etc. The service-related intentions are classified and distributed to the corresponding logic for processing.

[0125] Fig.12 A schematic diagram of the structure of a terminal device provided in an embodiment of the present application. The terminal device E1 includes: at least one processor E2 ( Fig.12 Only one is shown in the figure) a processor, a memory E3, and a computer program E4 stored in the memory E3 and executable on the at least one processor E2, and when the processor E2 executes the computer program E4, the steps in the above method embodiment are implemented.

[0126] The terminal device E1 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The terminal device may include, but is not limited to, a processor E2 and a memory E3. Those skilled in the art will understand that Fig.12 This is only an example of the terminal device E1 and does not constitute a limitation on the terminal device E1. The terminal device E1 may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, it may also include input and output devices, network access devices, etc.

[0127] The processor E2 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0128] In some embodiments, the memory E3 may be an internal storage unit of the terminal device E1, such as a hard disk or memory of the terminal device E1. In other embodiments, the memory E3 may also be an external storage device of the terminal device E1, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal device E1. Further, the memory E3 may also include both an internal storage unit and an external storage device of the terminal device E1. The memory E3 is used to store an operating system, an application program, a boot loader, data, and other programs, such as the program code of the computer program. The memory E3 may also be used to temporarily store data that has been output or is to be output.

[0129] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0130] In the above embodiments, the description of each embodiment has its own emphasis. If a rated part is not described or recorded in detail in a certain embodiment, reference may be made to the relevant description of other embodiments.

[0131] An embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.

[0132] An embodiment of the present application provides a computer program product. When the computer program product runs on a mobile terminal, the mobile terminal can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0133] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0134] In the embodiments provided in the present application, it should be understood that the disclosed devices / network equipment and methods can be implemented in other ways. For example, the device / network equipment embodiments described above are merely schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0135] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0136] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above units may be implemented in the form of hardware or software.

[0137] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0138] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

[0139] The relevant user personal information that may be involved in the various embodiments of this application is strictly in accordance with the requirements of laws and regulations, following the principles of legality, legitimacy and necessity, based on the reasonable purposes of business scenarios, to process the personal information that users actively provide during the use of products / services or generated due to the use of products / services, as well as the personal information obtained with the user's authorization.

[0140] The user personal information processed by the applicant will vary depending on the specific product / service scenario, and shall be based on the specific scenario in which the user uses the product / service, which may involve the user's account information, device information, driving information, vehicle information or other related information. The applicant will treat the user's personal information and its processing with a high degree of diligence.

[0141] The Applicant attaches great importance to the security of user personal information and has adopted reasonable and feasible security protection measures that comply with industry standards to protect user information and prevent personal information from being accessed, disclosed, used, modified, damaged or lost without authorization.

Claims

1. A service entry recommendation method for a smart device, characterized in that: The service entry recommendation method includes: Input the currently received user question into the first large model server to obtain the answer result; Generate a search prompt content according to the answer result and the user question and in combination with the introduction information corresponding to the multiple service entrances; The search prompt content is used as an input of the second largest model server. The second largest model selects a corresponding service entry from multiple service entries based on the search prompt content and outputs corresponding service entry recommendation information.

2. The service entry recommendation method according to claim 1, characterized in that: Before inputting the currently received user question into the first large model server to obtain a response result, the service entry recommendation method includes: Determine the corresponding user intent type based on the currently received user questions and the classification model; Determining a corresponding question-answering system according to the user intention type; Based on the user question and in combination with the question-answering system, determining whether there is corresponding matching answer data in the question-answering system; If not, the user question is input into the first large model server.

3. The service entry recommendation method according to claim 2, characterized in that: The user intent type includes service-related intent, device-related intent and mixed intent. If the user intent type is service-related intent, the corresponding first large model server includes a first retrieval architecture and a first generation architecture. The currently received user question is input into the first large model server to obtain a response result including: Inputting the currently received user question into the first search framework, combining the preset first knowledge resource database and the first question-answer resource database, and obtaining corresponding search results; According to the search result, in combination with the first generation architecture, a corresponding response result is obtained.

4. The service entry recommendation method according to claim 3, characterized in that: If the user intent type is a mixed intent, the corresponding first large model server includes a second search architecture, a second generation architecture, and a large model architecture, and the currently received user question is input into the first large model server to obtain a response result including: Inputting the currently received user question into the second search framework, combining with the preset second knowledge resource database and the second question-answer resource database, obtaining corresponding search results; Determine whether there is a corresponding response result based on the search result and in combination with the second generation architecture; If it does not exist, a search prompt content is generated according to the user question and input into the large model framework, and the large model framework outputs the corresponding answer result.

5. The service entry recommendation method according to claim 3, characterized in that: If the user intent type is a device-related intent, the corresponding first large model server includes a third retrieval architecture, a third generation architecture, and a large model architecture, and the currently received user question is input into the first large model server to obtain a response result including: Inputting the currently received user question into the third search framework, combining with the preset third knowledge resource database and the third question-answer resource database, to obtain corresponding search results; Determine whether there is a corresponding answer result according to the search result and in combination with the third generation architecture; If it does not exist, a search prompt content is generated according to the user question and input into the large model framework, and the large model framework outputs the corresponding answer result.

6. The service entry recommendation method according to claim 1, characterized in that: After inputting the currently received user question into the first large model server and obtaining a response result, the service entry recommendation method further includes: Inputting the response result into a regular matching module; The regular matching module determines the corresponding matching service entry based on the response result, and outputs the corresponding service entry recommendation information.

7. The service entry recommendation method according to claim 1, characterized in that: After generating a search prompt content according to the response result and the user question in combination with the introduction information corresponding to the plurality of service entrances, the service entrance recommendation method includes: The search prompt content is used as an input of a second large model server, and the second large model determines whether there is a corresponding service entry in a plurality of service entries based on the search prompt content; If not, the service entry recommendation information that matches the user intent is determined based on the user intent corresponding to the user question and in combination with a preset matching list.

8. A service entry recommendation device for a smart device, characterized in that: The service entrance recommendation device comprises: A response module, used for inputting the currently received user question into the first large model server to obtain a response result; A generating module, configured to generate a search prompt content according to the answer result and the user question and in combination with the introduction information corresponding to the plurality of service entrances; The recommendation module is used to use the search prompt content as input to the second largest model server. The second largest model selects a corresponding service entry from multiple service entries based on the search prompt content and outputs the corresponding service entry recommendation information.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Citation Information

Patent Citations

  • Automobile question-answer interaction system and interaction method based on artificial intelligence

    CN110727776A

  • BERT-based intelligent question and answer implementation method and system

    CN111538824A

  • Voice control method, intelligent cabin, server, vehicle and medium

    CN112242141A

  • Question answering method, question answering system, kitchen appliance and computer readable storage medium

    CN114741493A

  • Intelligent vehicle control system

    CN114940131A