User interaction method and system, vehicle, equipment and storage medium
Sorting and recommending stores through large language models solves the problem that the existing dialogue system cannot accurately understand user needs, and achieves more efficient information push and user interaction experience.
Patent Information
- Application Number
- CN202411024043.3
- 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
The existing dialogue system is not very intelligent, it is difficult to handle complex multi-round dialogue content, and it is impossible to accurately understand the real needs of users, resulting in low accuracy in information push and poor user interaction experience.
A large language model is used to sort stores based on user intentions, and through multiple data recall and fusion sorting, target recommended store data is generated and interact with users to improve the accuracy and user experience of information push.
By accurately understanding the real needs of users, we provide target recommendation store data that is more in line with user intentions, improve the accuracy of information push and user interaction experience, and enhance the diversity and accuracy of personalized recommendations.
Smart Images

Figure CN118964732A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a user interaction method, system, vehicle, device and storage medium. Background Art
[0002] With the rapid development of artificial intelligence technology, artificial intelligence-based dialogue systems have emerged. When users purchase products and receive services, most users often find it difficult to describe their specific needs. Relevant personnel usually need to have multiple rounds of dialogue with users to guide users to clarify their real needs; however, in the application process of existing dialogue systems, due to the low level of intelligence of the dialogue system, it is difficult to handle complex multi-round dialogue content and cannot accurately understand the real needs of users, resulting in the inability of pushed information to meet user needs, low accuracy of information push, and poor user interaction experience. Summary of the invention
[0003] The present invention provides a user interaction method, system, vehicle, device and storage medium to solve the problem that the existing conversation system cannot accurately understand the real needs of users, resulting in low accuracy of information push and poor user interaction experience.
[0004] In a first aspect, an embodiment of the present application provides a user interaction method, characterized by comprising:
[0005] Perform multi-channel data recall based on the user's store query information to obtain multi-channel store recall data;
[0006] Based on the store query information and the store information of each recalled store in the multi-channel store recall data, a large language model is used to sort the stores based on user intent to obtain the target recommended store data;
[0007] Interact with users by recommending store data based on goals.
[0008] Optionally, based on the store query information and the store information of each recalled store in the multi-channel store recall data, a large language model is used to perform store sorting based on user intent to obtain target recommended store data, including:
[0009] Fusion and sorting of multiple store recall data to obtain target recall store data;
[0010] Based on the store query information and the store information of each recalled store in the target recalled store data, a large language model is used to sort the stores based on user intent to obtain the target recommended store data.
[0011] Optionally, based on the store query information and the store information of each recalled store in the target recalled store data, a large language model is used to perform store sorting based on user intent to obtain target recommended store data, including:
[0012] Generate model prompt information of the large language model based on the user's portrait information, store query information, and store information of each recalled store in the target recalled store data. The model prompt information includes store information of each recalled store, store query information, number of filtered stores, and instructions for generating recommendation reasons.
[0013] The model prompt information is input into the large language model to perform store sorting and recommendation reason generation based on user intention, and obtain the target recommended store data that meets the number of screened stores. The target recommended store data includes the recommendation information, sorting number and store information of each recommended store.
[0014] Optionally, multiple store recall data are merged and sorted to obtain target recall store data, including:
[0015] Based on the user's portrait information, the multi-channel store recall data is optimized and fused and sorted to obtain the optimized store recall data;
[0016] The inverse sorting fusion method is used to fuse and sort the multi-channel store recall data and the optimized store recall data to obtain the target recall store data.
[0017] Optionally, the multi-channel store recall data is optimized, fused and sorted based on the user's portrait information to obtain optimized store recall data, including:
[0018] Determine store characteristic information of each recalled store according to store information of each recalled store in the multi-channel store recall data;
[0019] Based on the user's portrait information, a large language model is used to filter user interest information to obtain portrait feature information;
[0020] Based on the portrait feature information and store feature information, the multi-channel store recall data is optimized, fused and sorted to obtain the optimized store recall data.
[0021] Optionally, the multi-channel store recall data includes first store recall data for structured query statement recall based on store query information, and second store recall data for semantic vector recall based on store query information. Based on the portrait feature information and the store feature information, the multi-channel store recall data is optimized and fused and sorted to obtain optimized store recall data, including:
[0022] Determine the spatial distance information between each recalled store and the user based on the store information of each recalled store and the vehicle driving trajectory data;
[0023] According to the spatial distance information, portrait feature information and store feature information, a pre-trained re-ranking model is used to fuse and rank the first store recall data and the second store recall data to obtain the optimized store recall data.
[0024] Optionally, the multi-channel store recall data includes first store recall data and second store recall data, and the multi-channel data recall is performed based on the user's store query information to obtain the multi-channel store recall data, including:
[0025] The store query information is converted into a structured query statement, and the structured store information is recalled based on the structured query statement to obtain first store recall data;
[0026] When the first store recall data does not meet the quantity input requirement of the large language model, the store query information is converted into semantic vector information, and the vectorized store information is recalled based on the semantic vector information to obtain the second store recall data.
[0027] Optionally, after obtaining the first store recall data, the method further includes:
[0028] When the first store recall data meets the quantity input requirement of the large language model, the large language model is used to sort the first store recall data based on user intent, obtain target recommended store data, and interact with the user based on the target recommended store data.
[0029] Optionally, interact with the user based on the target recommended store data, including:
[0030] Feedback the target recommended store data to the user, and after receiving the user's input information again, use the large language model to perform context-based intent recognition and input expansion on the user's input information to obtain store inquiry information and the inquiry target store;
[0031] Based on the store inquiry information and the store information of the target store, a large language model is used to generate reply answers, and the reply information of the current input information is obtained and pushed to the user.
[0032] Optionally, the target recommended store data includes the ranking number, store information and recommendation information of each recommended store, and the target recommended store data is fed back to the user, including:
[0033] Based on the store information and recommendation information of each recommended store, a store picture and text display page is generated according to the ranking number of each recommended store;
[0034] The store picture and text display page is used as a response to the user's input and displayed to the user through the interactive interface of the terminal device.
[0035] Optionally, the target recommended store data includes the ranking number of the recommended store, store information and recommendation information, and the recommendation information includes the personalized recommendation reason of the recommended store and the comprehensive recommendation information that integrates the characteristics of the first m recommended stores. Feedback of the target recommended store data to the user includes:
[0036] The personalized recommendation reasons of each recommended store are inserted into the store information of each recommended store in the form of a prominent label, and a store graphic display page is generated according to the ranking number of each recommended store;
[0037] The store's graphic and text display page is used as a user's input response and displayed to the user through the interactive interface of the terminal device, and the comprehensive recommendation information is voice broadcasted through the virtual character in the interactive interface.
[0038] Optionally, before performing multi-channel data recall based on the user's store query information and obtaining the multi-channel store recall data, the method further includes:
[0039] Use the large language model to identify the question type of the user's current input information and obtain the question type of the input information, which includes store query type and store introduction type;
[0040] When the question type of the input information is a store query type, a large language model is used to perform context-based expansion and rewriting of the input information to obtain the user's store query information, and multiple data are recalled based on the user's store query information to obtain multiple store recall data.
[0041] In a second aspect, an embodiment of the present application provides a user interaction system, including a vehicle-mounted terminal and a server, wherein the server is used to:
[0042] Perform multi-channel data recall based on the user's store query information to obtain multi-channel store recall data;
[0043] Based on the store query information and the store information of each recalled store in the multi-channel store recall data, a large language model is used to sort the stores based on user intent to obtain the target recommended store data;
[0044] Recommend store data based on goals and interact with users through terminal devices.
[0045] In a third aspect, an embodiment of the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned user interaction method when executing the computer program.
[0046] In a fourth aspect, an embodiment of the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned parameter aggregation method based on federated learning when executing the computer program.
[0047] In a fifth aspect, an embodiment of the present application provides a readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the above-mentioned user interaction method are implemented.
[0048] In one solution provided by the above-mentioned user interaction method, system, vehicle, device and storage medium, multi-channel data recall is performed based on the user's store query information to obtain multi-channel store recall data, and then based on the store query information and the store information of each recalled store in the multi-channel store recall data, a large language model is used to sort the stores based on the user's intention, and the target recommended store data is obtained, and finally the user is interacted with based on the target recommended store data. By performing multi-channel data recall to enhance the retrieval of user input information, and using a large language model to sort the recalled stores based on the user's intention, the user's real needs can be accurately understood, and the target recommended store data that is more in line with the user's intention can be obtained, thereby improving the accuracy of information push and thus improving the user interaction experience. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.
[0050] Figure 1 is a schematic diagram of an application environment of a user interaction method in one embodiment of the present invention;
[0051] Figure 2 is another application environment schematic diagram of the user interaction method in one embodiment of the present invention;
[0052] Figure 3 is a flow chart of a user interaction method in one embodiment of the present invention;
[0053] Figure 4 yes Figure 3 A schematic diagram of an implementation process of step S10;
[0054] Figure 5 yes Figure 3 A schematic diagram of an implementation process of step S20;
[0055] Figure 6 yes Figure 3 A schematic diagram of an implementation process of step S30;
[0056] Figure 7 yes Figure 1 or Figure 2 A structural diagram of a server in FIG.
[0057] Figure 8 It is a structural schematic diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION
[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0059] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or their collections. It should also be understood that the term "and / or" used in the present specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.
[0060] In addition, in the description of the present 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.
[0061] References to "one embodiment" or "some embodiments" etc. described in the present specification mean that one or more embodiments of the present invention 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.
[0062] It should be understood that the order of execution of the steps in the following embodiments does not imply a precedence 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 embodiments of the present invention.
[0063] In order to illustrate the technical solution of the present invention, specific embodiments are provided below for illustration.
[0064] It is important to understand that with the improvement of people's living standards, vehicle users have higher and higher requirements for vehicle intelligence. For this reason, some car companies have developed artificial intelligence-based dialogue systems for vehicles to meet users' different car needs, such as smart navigation, smart store recommendations, smart music recommendations, etc. However, due to technical limitations, the existing dialogue systems lack natural and smooth dialogue capabilities and can only answer some preset template questions. It is difficult to handle complex multi-round dialogue content, so it is impossible to accurately understand the real needs of users, and the pushed information cannot meet user needs, resulting in low accuracy of information push and poor user interaction experience.
[0065] The user interaction method provided by the embodiment of the present invention can be applied in Figure 1 In the user interaction system shown, the user interaction system includes a terminal device and a server, wherein the terminal device communicates with the server through a network.
[0066] Among them, in the actual application process, the user can use the terminal device to interact with the application server to find various products, services or information required by the user. That is, the terminal device is used to interact with the user and obtain the user's input information. The server is used to: recall multiple data based on the user's store query information to obtain multiple store recall data; based on the store query information and the store information of each recalled store in the multiple store recall data, use a large language model to sort the stores based on the user's intention, and obtain the target recommended store data; based on the target recommended store data, control the terminal device to interact with the user.
[0067] In this embodiment, after obtaining the user's input information sent by the terminal device, the server performs multi-channel data recall based on the user's store query information to obtain multi-channel store recall data, and then uses a large language model to perform store sorting based on user intent based on the store query information and the store information of each recalled store in the multi-channel store recall data to obtain target recommended store data, and finally interacts with the user through the terminal device based on the target recommended store data. By performing multi-channel data recall to enhance the retrieval of user input information, and using a large language model to sort the recalled stores based on user intent, the real needs of users can be accurately understood, and target recommended store data that better meets user intent can be obtained, thereby improving the accuracy of information push and thus improving the user interaction experience.
[0068] The terminal devices include but are not limited to various personal computers, laptops, smart phones, tablet computers, vehicle-mounted terminals and portable wearable devices. The server can be implemented as an independent server or a server cluster consisting of multiple servers.
[0069] In one embodiment, the terminal device of the user interaction system may be a vehicle-mounted terminal on a vehicle. Figure 2 As shown, a vehicle is provided, which includes an on-board terminal, such as a vehicle computer. The on-board terminal communicates with an external server through a network to implement the user interaction method in this embodiment. The user can log in to the vehicle application through the on-board terminal, and then perform user interaction through the on-board terminal. The on-board terminal needs to obtain the input information of the user in the on-board terminal and send it to the server. The server obtains store query information based on the user's input information recognition, and then performs multi-channel data recall based on the user's store query information to obtain multi-channel store recall data, and then uses a large language model to perform store sorting based on user intention based on the store query information and the store information of each recalled store in the multi-channel store recall data, obtains target recommended store data and sends it to the on-board terminal. The on-board terminal obtains the target recommended store data fed back by the server based on the input information, and interacts with the user based on the target recommended store data. By performing multi-channel data recall to realize retrieval enhancement of user input information, and using a large language model to sort the recalled stores based on user intention, the real needs of the user can be accurately understood, and the target recommended store data that is more in line with the user's intention can be obtained, thereby improving the accuracy of information push, and thus improving the vehicle user interaction experience.
[0070] In one embodiment, if Figure 3 As shown, a user interaction method is provided, which is applied in Figure 1 The server in the example is used as an example to illustrate the process, including the following steps:
[0071] S10: Perform multi-channel data recall based on the user's store query information to obtain multi-channel store recall data.
[0072] When a user uses a user interaction system (including a conversation system and a question-and-answer system) to query information, it is necessary to obtain the input information entered by the user through a terminal device. After the server obtains the input information sent by the terminal device (such as a car computer), it will perform intent recognition on the user's input information to obtain the user's store query information. The store query information can be the store query information extracted by the server from the user's current input information, or it can be the store query information obtained by the server through joint intent recognition based on the current input information and its previous input information.
[0073] Then, the server performs multi-channel data recall based on the user's store query information to obtain multi-channel store recall data. Specifically, it includes: converting the store query information into a structured query statement, and recalling the structured store information based on the structured query statement to obtain first store recall data; converting the store query information into semantic vector information, and recalling the vectorized store information based on the semantic vector information to obtain second store recall data.
[0074] The server converts the store query information into a structured query statement, namely a SQL (Structured Query Language) statement, and recalls the structured store information based on the structured query statement to obtain the first store recall data. Store recall based on SQL statements can perform store statistics calculation, sorting, filtering and other functions, thereby obtaining the first store recall data after the stores are sorted. SQL statements are suitable for information query by field and have high accuracy.
[0075] After obtaining the store query information, the large language model can be used to divide the store query information into fields, and then SQL statements can be generated based on the fields to perform accurate store recall based on the fields on the structured store information in the database to obtain the first store recall data. For example, if the store query information is: Find a welfare store suitable for parents and children in Shanghai, the converted SQL statement is: SELECT * FROM P_DATABASE WHERE city = 'Shanghai' and 'tag' LIKE '% Parent-child %'. Based on this SQL statement, store recall data that meets all fields can be quickly found with high accuracy.
[0076] Among them, in order to ensure that accurate store recall can be performed based on structured query statements in the future, it is necessary to perform structured information processing on the store information of all stores in advance to obtain structured store information, and then store it in the database for direct use in the future. That is, a large language model can be used to perform text semantic classification processing on the store information of all stores to obtain multiple semantic labels of different dimensions; then the semantic labels are used as key field data columns, and all stores are classified based on multiple semantic labels to obtain structured store information. In this structured store information, multiple stores are included under the key field data column, and the same store can be under different key field data columns.
[0077] For example, multiple semantic tags of different dimensions include first-class tags, second-class tags, third-class tags, and n-class tags, which are successively refined. Among them, first-class tags include store type tags such as food, training, home, and health. Taking food stores as an example, the food tag can also include second-class tags such as store name, shelf time, location, per capita price, and store rating. It can also include third-class tags or fourth-class tags such as Tibetan style, yak beef, matsutake, Medog stone pot, plateau food, fresh fish, Hubei cuisine, fried and steamed, high quality, and special dishes.
[0078] Among them, it is necessary to use a large language model to extract store information of all stores, obtain various types of text information of each store, and then input various types of text information of the store into the semantic coding model for semantic coding to obtain the semantic vector of the store, and store the semantic vectors of all stores as vectorized store information in the database for subsequent direct use.
[0079] The server can convert the store query information into semantic vector information through the semantic coding model, and perform vector similarity-based store recall on the vectorized store information based on the semantic vector information to obtain the second store recall data. That is, the semantic vector information of the store query information is calculated, and the similarity is calculated with the semantic vectors of each store in the database, and the stores with similarity greater than a preset value are determined, and the stores are sorted in ascending order according to the similarity to obtain the second store recall data.
[0080] In this embodiment, the store query information is converted into a structured query statement, and the structured store information is recalled based on the structured query statement to obtain the first store recall data, which can improve the accuracy of the store recall structure, and the store query information is converted into semantic vector information, and the vectorized store information is recalled based on the semantic vector information to obtain the second store recall data. That is, the multi-channel store recall data includes the first store recall data based on the store query information by performing a structured query statement recall, and the second store recall data based on the store query information by performing a semantic vector recall, which can increase the accuracy and diversity of the store recall.
[0081] In other embodiments, the multi-channel store recall data also includes third store recall data based on the store query information for keyword recall. For example, the store query information may be subjected to keyword extraction, and then keyword matching may be performed based on at least two extracted keywords and the store information of each store in the database, and each recalled store may be sorted according to the number of matched keywords to obtain the third store recall data. This may further increase the diversity of store recalls.
[0082] S20: Based on the store query information and the store information of each recalled store in the multi-channel store recall data, a large language model is used to perform store sorting based on user intention to obtain target recommended store data.
[0083] After obtaining the multi-channel store recall data, the server will use a large language model to sort the stores based on user intent based on the store query information and the store information of each recalled store in the multi-channel store recall data to obtain the target recommended store data.
[0084] Among them, the prompt word generation of the large language model can be performed based on the store query information and the store information of each recalled store in the multi-channel store recall data to obtain the model prompt information, and the model prompt information at least includes the store query information and the store information of each recalled store; in other embodiments, the model prompt information package may include output data requirements, such as compliance with user intent, output store quantity, and output data format (such as JSON). Then, the model prompt information is input into the large language model for user intent recognition, and the recalled stores are sorted based on user intent to obtain the target recommended store data. The target recommended store data includes multiple recommended stores, the sorting number of each recommended store and the store information. Among them, the store information includes information such as store name, address, per capita price (per capita consumption), store type, store story title and store story content.
[0085] S30: Interact with the user based on the target recommended store data.
[0086] After obtaining the target recommended store data, the server can directly send the target recommended store data to the terminal device, display the target recommended store data to the user through the terminal device, and realize the interaction with the user. In other embodiments, the server can also process the target recommended store data, such as image and text processing, voice processing, etc., and send the processed target recommended store data to the terminal device, display the processed target recommended store data to the user through the terminal device, and realize the interaction with the user.
[0087] In this embodiment, multi-channel data recall is performed based on the user's store query information to obtain multi-channel store recall data, and then based on the store query information and the store information of each recalled store in the multi-channel store recall data, a large language model is used to sort the stores based on the user's intention to obtain target recommended store data, and finally based on the target recommended store data, interaction with the user is performed through the terminal device. By performing multi-channel data recall, the retrieval enhancement of the user input information is achieved, and the recalled stores are sorted based on the user's intention. The large language model can better understand the user's complex queries and diverse intentions, accurately understand the user's real needs, and obtain target recommended store data that is more in line with the user's intention, thereby improving the accuracy of personalized information push, and then improving the user interaction experience. In addition, the target recommended store data can also achieve a good balance in diversity, accuracy and user satisfaction to meet the personalized needs of users.
[0088] In one embodiment, the multiple store recall data includes first store recall data and second store recall data. Figure 4 As shown, in step S10, multiple data are recalled based on the user's store query information to obtain multiple store recall data, including the following steps:
[0089] S11: converting the store query information into a structured query statement, and performing data recall on the structured store information based on the structured query statement to obtain first store recall data.
[0090] S12: When the first store recall data does not meet the quantity input requirement of the large language model, the store query information is converted into semantic vector information, and the vectorized store information is recalled based on the semantic vector information to obtain the second store recall data.
[0091] Among them, the acquisition process of the first store recall data and the second store recall data is as described above and will not be repeated here. In this embodiment, after performing a store recall based on an SQL statement to obtain the first store recall data, the store query information is converted into semantic vector information, and before the vectorized store information is recalled based on the semantic vector information, it is necessary to determine whether the number of recalled stores in the first store recall data does not meet the number input requirement of the large language model.
[0092] For example, if the quantity input requirement of the large language model is 10 stores, if the number of recalled stores in the first store recall data is less than 10, the quantity input requirement of the large language model is not met, and multi-channel data recall is required, that is, the store query information needs to be converted into semantic vector information, and the vectorized store information needs to be recalled based on the semantic vector information to obtain the second store recall data to execute steps S20 to S30.
[0093] If the number of recalled stores in the first store recall data is greater than or equal to 10, the quantity input requirement of the large language model is met. That is, when the first store recall data meets the quantity input requirement of the large language model, there is no need to perform multi-channel data recall, that is, there is no need to perform data recall on the vectorized store information based on the semantic vector information. Only the large language model is used to sort the first store recall data based on user intention, and select a preset number of stores (such as the first L) with the highest ranking to obtain target recommended store data, and then interact with the user based on the target recommended store data.
[0094] In this embodiment, the store query information is converted into a structured query statement, and the structured store information is recalled based on the structured query statement to obtain accurate first store recall data. When the first store recall data does not meet the quantity input requirements of the large language model, the store query information is converted into semantic vector information, and the vectorized store information is recalled based on the semantic vector information to obtain the second store recall data, which can improve the diversity of store recall data and ensure that more accurate and diverse target recommended store data can be output later. When the first store recall data meets the quantity input requirements, there is no need to perform multi-channel recall. On the basis of ensuring the accuracy of the recall result, the recall calculation amount is reduced, thereby improving the speed of subsequent information push and reducing the system load.
[0095] In one embodiment, if Figure 5 As shown, in step S20, based on the store query information and the store information of each recalled store in the multi-channel store recall data, a large language model is used to perform store sorting based on user intention to obtain target recommended store data, including the following steps:
[0096] S21: Fusion and sorting of multiple store recall data to obtain target recall store data.
[0097] For example, the multi-channel store recall data includes first store recall data and second store recall data. The first store recall data is obtained by converting the store query information into a structured query statement and recalling the structured store information based on the structured query statement. The second store recall data is obtained by converting the store query information into semantic vector information and recalling the vectorized store information based on the semantic vector information.
[0098] The reciprocal rank fusion algorithm (RRF) is used to fuse and sort the recall data of the first store and the recall data of the second store to obtain the target recall store data. Based on the reciprocal rank fusion algorithm, the first reciprocal ranking score of each recalled store in the recall data of the first store is determined, and the second reciprocal ranking score of each recalled store in the recall data of the second store is determined. The first reciprocal ranking score and the second reciprocal ranking score of the same recalled store are combined to obtain the comprehensive reciprocal ranking score of each recalled store. All recalled stores are re-sorted (in descending order) based on the comprehensive reciprocal ranking score of each recalled store to obtain the initial sorted store data, and then according to the number of input requirements of the large language model, a certain number of recalled stores with high rankings are selected from the initial sorted store data to obtain the target recall store data.
[0099] S22: Based on the store query information and the store information of each recalled store in the target recalled store data, a large language model is used to perform store sorting based on user intent to obtain target recommended store data.
[0100] After obtaining the target recalled store data, based on the store query information and the store information of each recalled store in the target recalled store data, a large language model is used to sort the stores based on user intent to obtain the target recommended store data.
[0101] That is, based on the store query information and the store information of each recalled store in the target recalled store data, the prompt word of the large language model can be generated to obtain the model prompt information, and then the model prompt information is input into the large language model for user intent recognition, and the recalled stores are sorted based on user intent to obtain the target recommended store data. The target recommended store data includes multiple recommended stores, the sorting number of each recommended store and the store information. Among them, the store information includes information such as store name, address, per capita price (per capita consumption), store type, store story title and store story content.
[0102] In this embodiment, by fusing and sorting the multi-channel store recall data, the target recall store data is obtained, and then based on the store query information and the store information of each recalled store in the target recall store data, the large language model is used to sort the stores based on user intent to obtain the target recommended store data. First, the multi-channel store recall data is fused and sorted, and then the large language model is used to sort the stores based on user intent, which can reduce the data processing volume of the large language model input, save computing resources and time, increase the speed of subsequent information push, and reduce the system load.
[0103] In addition, directly inputting multi-channel store recall data into the large language model may cause the language model to rely too much on the specific features or preferences of a certain channel of store recall data, increasing the risk of overfitting. Using the fused target recall store data as the input of the large language model can balance the contribution of each channel of store recall data at a higher level and reduce the possibility of overfitting.
[0104] In one embodiment, step S22, i.e., based on the store query information and the store information of each recalled store in the target recalled store data, a large language model is used to perform store sorting based on user intent to obtain target recommended store data, including the following steps:
[0105] S221: Generate model prompt information of the large language model based on the user's portrait information, store query information, and store information of each recalled store in the target recalled store data.
[0106] The server needs to generate model prompt information of the large language model based on the user's portrait information, store query information and the store information of each recalled store in the target recalled store data. The model prompt information includes the store information of each recalled store, the store query information and the generation target requirements, and the generation target requirements include the number of filtered stores (filtering out a preset number of stores), the recommendation reason generation instructions and the output format and other requirements.
[0107] S222: Input the model prompt information into the large language model, perform store sorting and generate recommendation reasons based on user intention, and obtain target recommended store data that meets the number of screened stores.
[0108] Then, the model prompt information is input into the large language model to perform store ranking and recommendation reason generation based on user intention, and obtain the target recommended store data that meets the number of screened stores, where the target recommended store data includes the recommendation information, ranking number and store information of each recommended store. Among them, the recommendation information includes the personalized recommendation reason of the recommended store and the comprehensive recommendation information that integrates the characteristics of all recommended stores.
[0109] Specifically, the model prompt information is input into the large language model so that the large language model can achieve the following: sort the recalled stores based on the user's intention based on the store query information, and filter the sorted recalled stores based on the number of filtered stores to obtain initial recommended store data; respond to the recommendation reason generation instruction, generate the recommendation reason for the recommended stores in the initial recommended store data based on the user's portrait information, obtain the recommendation information of each recommended store, and summarize and output it in the output format to obtain the model output data. Then, the server parses the model output data of the large language model according to the preset format to obtain the target recommended store data.
[0110] In other embodiments, the model prompt information includes the store information of each recall store, store query information, vehicle-related information, user profile information, and generation target requirements, which include requirements such as the number of screened stores, instructions for generating recommendation reasons, and output formats. By increasing the diversity of the input information of the large language model, the large language model can more accurately identify the user's potential intentions when performing store sorting and recommendation reason generation based on user intentions, thereby obtaining target recommended store data that better meets user needs.
[0111] Among them, when a large language model (LLM) is used to sort stores based on user intent and obtain target recommended store data, the model prompt information prompt of the large language model and the model output data output of the large language model can be shown in Table 1:
[0112] Table 1
[0113]
[0114]
[0115] The model prompt information and model output data in Table 1 are only exemplary. In other embodiments, the model prompt information and model output data may be other, which will not be described in detail here.
[0116] In this embodiment, based on the user's portrait information, store query information and the store information of each recalled store in the target recalled store data, the model prompt information of the large language model is generated, the model prompt information is input into the large language model, the store ranking and recommendation reason generation based on the user's intention are performed, and the target recommended store data that meets the number of screened stores is obtained. The target recommended store data includes the recommendation information, ranking number and store information of each recommended store. Among them, combining the user's portrait information and the store information in the target recalled store data, the user's preferences and needs can be understood more accurately. The large language model can take into account the user's personalized characteristics in the recommendation process, thereby providing a recommendation result that is more in line with the user's interests and intentions. The model prompt information includes the number of screened stores and the indication of the generation of the recommendation reason, which enables the large language model to generate detailed recommendation reasons, which not only increases the interpretability of the recommendation results, but also helps users understand why these stores are recommended, and provides relevant information to support user decision-making. The target recommended store data finally generated includes recommendation information, ranking number and store information, which enables users to intuitively understand the reasons and results of the recommendation, enhances the user's sense of trust, and improves the overall user interaction experience.
[0117] In one embodiment, in step S21, multiple store recall data are merged and sorted to obtain target recall store data, which specifically includes the following steps:
[0118] S211: Optimize, merge and sort the multiple store recall data based on the user's portrait information to obtain optimized store recall data.
[0119] After obtaining the multi-channel store recall data, the server needs to optimize and merge the multi-channel store recall data based on the user's portrait information to obtain the optimized store recall data. The user's portrait information includes the user's basic information (such as age, gender, occupation, family members and their member information, etc.), the user's store preference type, such as food, entertainment, and the user's store consumption habits.
[0120] For example, the multi-channel store recall data includes the first store recall data and the second store recall data. The server will use a large language model to filter the user's portrait information based on user interest information to obtain portrait feature information. Based on the portrait feature information, the server will then perform interest-based store matching and optimized sorting on the first store recall data and the second store recall data to obtain optimized store recall data that sorts the recalled stores according to the degree of matching of user interest preferences.
[0121] S212: Adopting the inverse sorting fusion method, the multi-channel store recall data and the optimized store recall data are fused and sorted to obtain the target recall store data.
[0122] After obtaining the optimized store recall data, the server uses a reverse sorting fusion method to fuse and sort the multi-channel store recall data and the optimized store recall data to obtain the target recall store data.
[0123] For example, multi-channel store recall data includes first store recall data and second store recall data. After obtaining the optimized store recall data, the server needs to first determine the reciprocal ranking score of each recalled store in the first store recall data in the first store recall data, the reciprocal ranking score of each recalled store in the second store recall data in the second store recall data, and the reciprocal ranking score of each recalled store in the optimized store recall data in the optimized store recall data based on a reciprocal ranking algorithm. For each recalled store, all reciprocal ranking scores of the recalled store are summed (or weighted summed) to obtain a comprehensive reciprocal ranking score. All recalled stores are reordered based on the comprehensive reciprocal ranking score to obtain target recall store data.
[0124] The calculation formula of the reciprocal ranking score is: score = 1 / (rank + k). Score represents the reciprocal ranking score of a recalled store in a certain data path; rank represents the ranking position of the recalled store in a certain data path; k is a pre-calibrated constant less than 60.
[0125] In this embodiment, the multi-channel store recall data is optimized, fused and sorted based on the user's portrait information to obtain the optimized store recall data. The multi-channel store recall data and the optimized store recall data are fused and sorted using the inverse sorting fusion method to obtain the target recall store data. The steps of fusing and sorting the multi-channel store recall data to obtain the target recall store data are clarified. The portrait information can accurately reflect the user's interests and preferences. The optimized store recall data is obtained by optimizing and sorting based on the portrait information, so that the optimized store recall data better matches the user's personalized needs and improves the recall effect of the subsequent target recall store data. The use of the inverse sorting fusion method for multi-dimensional recall data fusion can weigh the importance of data from different sources and ensure that the output target recall store data has good performance in multiple indicators (such as recall rate, accuracy, etc.).
[0126] In one embodiment, step S211, i.e., optimizing and merging multiple store recall data based on user portrait information to obtain optimized store recall data, specifically includes the following steps:
[0127] S2111: Determine store characteristic information of each recalled store based on the store information of each recalled store in the multi-channel store recall data.
[0128] The server can obtain the store information of each recalled store in the multi-channel store recall data, and then use the large language model to capture the key information of each recalled store in the multi-channel store recall data to determine the store feature information of each recalled store. The store feature information is information that can affect the user's store selection.
[0129] For example, the store characteristic information of each recalled store includes the store rating, the store average price, the number of various store reviews (such as good reviews, bad reviews, and medium reviews), whether the store is currently open, whether the store has discounts, and whether the store provides exclusive benefits for special users (car owners). In this embodiment, the store characteristic information is only an example. In other embodiments, the store characteristic information may also include information that can affect the user's store selection, which will not be repeated here.
[0130] S2112: Based on the user's portrait information, a large language model is used to filter the user's interest information to obtain portrait feature information.
[0131] At the same time, the server will use the large language model to filter the user interest information of each recalled store based on the user's portrait information and the store information of each recalled store in the multi-channel store recall data, and obtain the portrait feature information corresponding to each recalled store. The portrait feature information represents the interest level of the user (car owner) or family members in each store.
[0132] For example, portrait feature information may include the number of times the car owner has visited the store or used coupons; whether the type of store is the type preferred by the car owner; the degree of match between the store and the car owner's age group; whether the store's main content matches the car owner's interests; whether the store's type is the type preferred by the car owner's family members, such as the user's input information is: recommend a food store that Doudou may like, where Doudou is the car owner's daughter; the degree of match between the store and the car owner's family members' age group; whether the store's main content matches the interests of the car owner's family members; whether the store's main content matches the interests of the people currently in the car (the car owner or family members or the car owner's friends).
[0133] In this embodiment, the above-mentioned portrait feature information is only for illustrative purposes. In other embodiments, the portrait feature information may also include other information for the user (car owner) or family members' interest in each store, which will not be repeated here.
[0134] S2113: Determine the spatial distance information between each recalled store and the user based on the store information of each recalled store and the vehicle driving trajectory data.
[0135] At the same time, the server will also determine the spatial distance information between each recall store and the user based on the store information of each recall store in the multi-channel store recall data, as well as the vehicle-related data (including vehicle driving trajectory data, vehicle occupants, vehicle status, etc.). Among them, the vehicle driving trajectory data includes the vehicle's current location information, the vehicle's historical trajectory information, and the vehicle's future predicted trajectory information (such as navigation information). The spatial distance information between each recall store and the user is the distance information that represents the location space or time space between the store and the user (car owner).
[0136] For example, spatial distance information includes the distance between the vehicle's current location and the store, the distance between the vehicle's current destination and the store, whether the store is along the vehicle's current navigation route, the number of times the owner has navigated to the store, etc. In this embodiment, the spatial distance information is only an example. In other embodiments, the spatial distance information may also include other information, which will not be described in detail here.
[0137] S2114: Based on the portrait feature information, store feature information and / or spatial distance information, the multi-channel store recall data are optimized, fused and sorted to obtain optimized store recall data.
[0138] Finally, based on the portrait feature information, store feature information and / or spatial distance information, the multi-channel store recall data are optimized, fused and sorted to obtain the optimized store recall data.
[0139] Among them, based on the portrait feature information and store feature information corresponding to each recalled store, the multi-channel store recall data can be optimized, fused and sorted to obtain the optimized store recall data. Alternatively, based on the portrait feature information and spatial distance information, the multi-channel store recall data can be optimized, fused and sorted to obtain the optimized store recall data. In other embodiments, the multi-channel store recall data can also be optimized, fused and sorted based on the portrait feature information, store feature information and spatial distance information to obtain the optimized store recall data, increase the diversity of information, and comprehensively consider the influence of user interest preferences, vehicle trajectory and the quality of the store itself on the recall effect, so as to improve the accuracy of the optimized store recall data.
[0140] In this embodiment, store characteristic information of each recalled store is determined based on the store information of each recalled store in the multi-channel store recall data, and based on the user's portrait information, a large language model is used to screen user interest information to obtain portrait characteristic information, and according to the store information of each recalled store and the vehicle driving trajectory data, the spatial distance information between each recalled store and the user is determined, and based on the portrait characteristic information, as well as the store characteristic information and / or spatial distance information, the multi-channel store recall data is optimized, fused and sorted to obtain optimized store recall data, which can comprehensively consider the influence of user interest preferences, vehicle trajectory and / or the quality of the store itself on the recall effect, thereby improving the accuracy of the optimized store recall data, and further improving the accuracy of subsequent information push.
[0141] In one embodiment, the multi-channel store recall data includes first store recall data based on structured query statement recall of store query information, and second store recall data based on semantic vector recall of store query information. A pre-trained re-ranking model may be obtained first, wherein the re-ranking model may be a neural network model trained in advance based on multiple portrait feature information and multiple store feature information using an algorithm such as the Extreme Gradient Boosting (XGBoost) algorithm; then all recalled stores, the portrait feature information and the store feature information corresponding to each recalled store are input into the re-ranking model to re-rank the stores, and the recalled store ranking information output by the re-ranking model is obtained, and the optimized ranking number of each recalled store is obtained. Then, each recalled store is re-ranked according to the optimized ranking number of each recalled store, and optimized store recall data including the store information of each recalled store is obtained.
[0142] In one embodiment, the multi-channel store recall data includes first store recall data for structured query statement recall based on store query information, and second store recall data for semantic vector recall based on store query information. A pre-trained reordering model can be obtained first, wherein the reordering model can be pre-trained based on multiple spatial distance information, multiple portrait feature information, and multiple store feature information using algorithms such as the XGBoost algorithm to obtain a neural network model; then all recalled stores, the spatial distance information, portrait feature information, and store feature information corresponding to each recalled store are input into the reordering model to reorder the stores, and the recalled store ranking information output by the reordering model is obtained, and the ranking number of each recalled store after optimization is obtained. Then, each recalled store is reordered according to the ranking number of each recalled store after optimization, and optimized store recall data including the store information of each recalled store is obtained. The influence of user interest preferences and the quality of the vehicle trajectory store itself on the recall effect can be comprehensively considered to improve the accuracy of the optimized store recall data.
[0143] In one embodiment, if Figure 6 As shown, in step S30, that is, interacting with the user based on the target recommended store data, specifically includes the following steps:
[0144] S31: Feedback the target recommended store data to the user, and after receiving the user's input information again, use the large language model to perform context-based intent recognition and input expansion on the user's input information to obtain store inquiry information and the inquiry target store.
[0145] After obtaining the target recommended store data, the server feeds back the target recommended store data to the terminal device (vehicle computer), and displays the target recommended store data to the user through the terminal device so that the user can view and conduct further interactive dialogue based on the target recommended store data.
[0146] After feeding back the target recommended store data to the user, it is necessary to determine whether the user's input information is received again from the terminal device; after receiving the user's input information again, the user's input information is used to identify the intent and expand the input in combination with the context using the large language model to obtain the store inquiry information and the target store to be inquired. That is, for example, the large language model performs intent identification on the current input information and context information (including the target recommended store data), and determines the store that the user wants to query (i.e. the target store to be inquired) and the specific information about the store that the user wants to query; then, based on the intent identification result, the current input information is semantically expanded to obtain the store inquiry information.
[0147] For example, the current input information is: "What is the second store like?" According to the context, the second store is the second recommended store B in the target recommended store data. It is fully determined that the store the user intends to inquire about is store B, and the information to be queried is the general introduction information of store B. Then, the current input information "What is the second store like?" can be rewritten as "Introduction to store B."
[0148] S32: Based on the store inquiry information and the store information of the target store, a reply answer is generated using a large language model to obtain reply information for the current input information and push it to the user.
[0149] After determining the store inquiry information and the target store to be inquired, the large language model is used to generate a reply answer based on the store inquiry information and the store information of the target store to be inquired, and the reply information of the current input information is obtained and pushed to the user. That is, the model prompt information of the large language model is generated by the store inquiry information and the store information of the target store to be inquired, and the model prompt information is input into the large language model to generate a reply answer, and the output data of the large language model is obtained. The server parses the output data to obtain the reply information of the current input information, and feeds back the reply information of the current input information to the terminal device, and pushes the reply information of the current input information to the user through the terminal device.
[0150] For example, the user's current input information is: What time does the first store open? The first store (the query target store) is: Xijiao Santai Teppanyaki, and the store query information is: What time does Xijiao Santai Teppanyaki open? The store information includes the store name, sort number (index), address, distance, per capita consumption (per capita price), store type, discount information, store details (including business hours), etc. The model prompt information includes store query information, store information of the query target store, and answer generation requirements, which include output format and output field requirements.
[0151] Then, the model prompt information prompt of the large language model LLM and the output data out of the large language model LLM can be shown in Table 2:
[0152] Table 2
[0153]
[0154]
[0155] As mentioned above, the answer part is the specific reply text to the user's question (such as store inquiry information). referred_shop_index is used to highlight the store referred to by the current reply when it is displayed on the front end. Tags are used to classify user questions, highlight specific tags, and prompt users.
[0156] In this embodiment, the model prompt information and output data are only for illustrative purposes. In other embodiments, the model prompt information and output data may also be abstract and will not be described in detail here.
[0157] In this embodiment, the target recommended store data is fed back to the user, and after receiving the user's input information again, the user's input information is used to identify the intent and expand the input in context using the large language model to obtain the store inquiry information and the target store for inquiry, and then based on the store inquiry information and the store information of the target store for inquiry, the large language model is used to generate a reply answer, and the reply information of the current input information is obtained and pushed to the user. For the user's specific inquiry about the store, the use of the large language model for reply generation can provide more accurate and detailed answers, improve the accuracy of the reply information, the completeness of the information, the timeliness of the system response, and the user satisfaction.
[0158] In one embodiment, the target recommended store data includes the ranking number, store information and recommendation information of each recommended store. In step S31, the target recommended store data is fed back to the user, which specifically includes the following steps:
[0159] S311: Based on the store information and recommendation information of each recommended store, generate a store picture and text display page according to the ranking number of each recommended store;
[0160] S312: The store picture and text display page is presented to the user via an interactive interface of a terminal device as a response to the user's input.
[0161] In this embodiment, the target recommended store data includes the ranking number, store information and recommendation information of each recommended store. After obtaining the target recommended store data, the server will generate a store graphic display page according to the ranking number of each recommended store based on the store information and recommendation information of each recommended store, and then send the store graphic display page to the terminal device as a user input response, and display it to the user through the interactive interface of the terminal device. Among them, in the store graphic display page, the recommendation information of each recommended store can be inserted into the display area of each recommended store in the form of a prominent label, so that the user can quickly notice the reason for the recommendation of the recommended store.
[0162] In this embodiment, based on the store information and recommendation information of each recommended store, a store picture and text display page is generated according to the ranking number of each recommended store, and the store picture and text display page is used as a user input response and displayed to the user through the interactive interface of the terminal device. The store information and recommendation information of each recommended store are displayed to the user through the store picture and text display page, which increases the diversity and accuracy of the pushed information, allows users to know the recommendation reasons of the recommended store, and further improves the user interaction experience.
[0163] In one embodiment, the target recommended store data includes not only the ranking number, store information and personalized recommendation reasons (i.e., recommendation information) of each recommended store, but also a comprehensive recommendation reason (i.e., comprehensive recommendation reason) that integrates the characteristics of the first m (e.g., first 3) recommended stores. In step S31, the target recommended store data is fed back to the user, which specifically includes the following steps:
[0164] S301: inserting the personalized recommendation reason of each recommended store into the store information of each recommended store in the form of a prominent label, and generating a store picture and text display page according to the ranking number of each recommended store;
[0165] S302: The store text and picture display page is presented to the user as a user input response through an interactive interface of a terminal device, and the comprehensive recommendation information is voice broadcasted through a virtual character in the interactive interface.
[0166] The user interaction system in this embodiment has a virtual character, which can be a digital person. The virtual character is displayed through the interactive interface of the terminal device (such as the vehicle computer) and interacts with the user by voice, which can enhance the user interaction experience.
[0167] After obtaining the target recommended store data, the server will insert the personalized recommendation reasons of each recommended store into the store information of each recommended store in the form of a highlight label (such as yellow, red, etc., a highlight image label that is different from the page text color), generate a store graphic display page according to the sorting number of each recommended store, and then send the store graphic display page to the terminal device as a user's input response, and display it to the user through the interactive interface of the terminal device; at the same time, the comprehensive recommendation information that integrates the characteristics of the first m recommended stores is sent to the terminal device, and the comprehensive recommendation information is voice broadcast through the virtual character in the interactive interface.
[0168] In this embodiment, the personalized recommendation reason for each recommended store is inserted into the store information of each recommended store in the form of a prominent label, and a store picture and text display page is generated according to the ranking number of each recommended store. The store picture and text display page is used as a user input response and displayed to the user through the interactive interface of the terminal device. The comprehensive recommendation information is voice broadcast through the virtual character in the interactive interface, thereby further increasing the diversity of push information and the information display effect, and improving the user interaction experience.
[0169] In one embodiment, before step S10, that is, before performing multi-channel data recall based on the user's store query information to obtain multi-channel store recall data, the method specifically includes the following steps:
[0170] S01: Using a large language model to identify the question type of the user's current input information, and obtain the question type of the input information.
[0171] The user interaction system or server includes two functions (tools): store query function and store introduction function. These two functions can be functions implemented in multiple programming languages or encapsulated functions in the form of APIs. Among them, the store query function is used to query (search) stores based on user input information, and the store introduction function is used to accurately inquire about the displayed stores after displaying stores or displaying store lists. The two functions can realize different granularity information acquisition needs for users and meet the user interaction experience.
[0172] When the user inputs relevant information through the terminal device, the terminal device will send the user's current input information to the server. After obtaining the user's current input information, the server needs to identify the question type of the user's current input information based on the large language model to obtain the question type of the input information, and then determine to call different functions to implement different information search logic according to the question type of the current input information, that is, determine different server data processing logic according to the question type of the current input information. Among them, the question types include store query questions and store introduction questions.
[0173] Specifically, the server can combine the functional descriptions of the two functions and the user's current input information into a set of strings as model prompt information, and input the model prompt information into the large language model so that the large language model can identify the function to be called, obtain the output data of the large language model, parse the output data to determine the function to be called, and determine the question type of the input information corresponding to the function to be called. For example, the question type corresponding to the store query function is the store query type, and the question type corresponding to the store introduction function is the store introduction type.
[0174] S02: When the question type of the input information is a store query type, a large language model is used to perform context-based expansion and rewriting of the input information to obtain the user's store query information, and multiple data are recalled based on the user's store query information to obtain multiple store recall data.
[0175] If the question type of the user's current input information is a store query type, it is necessary to call the store query function to perform store query retrieval, that is, the server needs to use a large language model to expand and rewrite the input information based on the context, obtain the user's store query information, and perform multi-channel data recall based on the user's store query information to obtain multi-channel store recall data... until the target store recommendation data is obtained. That is, after obtaining the user's store query information, the server will execute steps S10-S30 to complete the store query.
[0176] If the question type of the user's current input information is a store introduction type, it is necessary to call the store introduction function to further introduce the information of the stores that have appeared in the interactive interface. That is, execute steps S31 to S32, use the large language model to perform context-based intent recognition and input expansion on the user's input information, obtain store inquiry information and the target store for inquiry, and then use the large language model to generate a reply answer based on the store inquiry information and the store information of the target store for inquiry, obtain the reply information of the current input information and push it to the user.
[0177] In this embodiment, the large language model is used to identify the question type of the user's current input information to obtain the question type of the input information. When the question type of the input information is a store query type, the large language model is used to expand and rewrite the input information based on the context to obtain the user's store query information, and multiple data are recalled based on the user's store query information to obtain multiple store recall data. The large language model is used to identify and rewrite the question type to ensure accurate understanding and feedback of the user's questions, thereby improving the accuracy of the system's response.
[0178] It should be understood that the order of execution of the steps in the above embodiment does not necessarily 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 invention.
[0179] In one embodiment, a server is provided, which corresponds one-to-one with the above-mentioned user interaction method. Figure 7 As shown, the server includes a recall module 701, a sorting module 702 and an interaction module 703. The detailed description of each functional module is as follows:
[0180] The recall module 701 is used to recall multiple data based on the user's store query information to obtain multiple store recall data;
[0181] A sorting module 702 is used to sort the stores based on the user's intention using a large language model based on the store query information and the store information of each recalled store in the multi-channel store recall data to obtain target recommended store data;
[0182] The interaction module 703 is used to interact with the user based on the target recommended store data.
[0183] Optionally, the sorting module 702 is specifically configured to:
[0184] Fusion and sorting of multiple store recall data to obtain target recall store data;
[0185] Based on the store query information and the store information of each recalled store in the target recalled store data, a large language model is used to sort the stores based on user intent to obtain the target recommended store data.
[0186] Optionally, the base sorting module 702 is further configured to:
[0187] Generate model prompt information of the large language model based on the user's portrait information, store query information, and store information of each recalled store in the target recalled store data. The model prompt information includes store information of each recalled store, store query information, number of filtered stores, and instructions for generating recommendation reasons.
[0188] The model prompt information is input into the large language model to perform store sorting and recommendation reason generation based on user intention, and obtain the target recommended store data that meets the number of screened stores. The target recommended store data includes the recommendation information, sorting number and store information of each recommended store.
[0189] Optionally, the interaction module 703 is specifically used for:
[0190] Feedback the target recommended store data to the user, and after receiving the user's input information again, use the large language model to perform context-based intent recognition and input expansion on the user's input information to obtain store inquiry information and the inquiry target store;
[0191] Based on the store inquiry information and the store information of the target store, a large language model is used to generate reply answers, and the reply information of the current input information is obtained and pushed to the user.
[0192] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.
[0193] 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.
[0194] In one embodiment, a computer device is provided, which may be a server or a terminal device, and includes a processor, a memory, a network interface, and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data used and generated by the above-mentioned user interaction method. The network interface of the computer device is used to communicate with an external device via a network connection. When the computer program is executed by the processor, a user interaction method is implemented.
[0195] The present application also provides a computer device, such as Figure 8 As shown, the computer device includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor implements the steps in any of the above-mentioned method embodiments when executing the computer program, or implements the functions of the modules / units in the above-mentioned device embodiments when executing the computer program.
[0196] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present application. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program in the computer device.
[0197] Those skilled in the art will understand that Figure 8These are merely examples of the computer device and do not constitute a limitation of the computer device. The computer device may include more or fewer components than those shown in the figure, or a combination of certain components, or different components. For example, the computer device may also include input and output devices, network access devices, buses, etc.
[0198] The processor 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.
[0199] The memory may be an internal storage unit of the computer device, such as a hard disk or memory of the computer device. The memory may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device. Furthermore, the memory may also include both an internal storage unit of the computer device and an external storage device.
[0200] An embodiment of the present application further provides a readable storage medium, wherein the readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.
[0201] 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.
[0202] If the integrated 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, which can 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 can at least include: any entity or device that can carry the computer program code to the camera / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, RandomAccess Memory), electric carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.
[0203] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.
[0204] 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.
[0205] In the embodiments provided in the present application, it should be understood that the disclosed devices / equipment and methods can be implemented in other ways. For example, the device / equipment embodiments described above are merely schematic, for example, the division of the modules or units is only a logical function division, and 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.
[0206] 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.
[0207] 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.
[0208] 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.
[0209] 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.
[0210] 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.
Claims
1. A user interaction method, characterized in that: include: Perform multi-channel data recall based on the user's store query information to obtain multi-channel store recall data; Based on the store query information and the store information of each recalled store in the multiple store recall data, a large language model is used to perform store sorting based on user intention to obtain target recommended store data; Interact with the user based on the target recommended store data.
2. The user interaction method according to claim 1, characterized in that: The method of using a large language model to sort stores based on user intent based on the store query information and the store information of each recalled store in the multiple store recall data to obtain target recommended store data includes: Fusion and sorting of multiple channels of store recall data to obtain target recall store data; Based on the store query information and the store information of each recalled store in the target recalled store data, the large language model is used to perform store sorting based on user intention to obtain the target recommended store data.
3. The user interaction method according to claim 2, characterized in that: The method of using the large language model to perform store sorting based on user intent based on the store query information and the store information of each recalled store in the target recalled store data to obtain the target recommended store data includes: Generate model prompt information of the large language model based on the user's portrait information, the store query information, and the store information of each recalled store in the target recalled store data, wherein the model prompt information includes the store information of each recalled store, the store query information, and a recommendation reason generation indication; The model prompt information is input into the large language model to perform store ranking and recommendation reason generation based on user intention to obtain the target recommended store data, which includes the recommendation information, ranking number and store information of each recommended store.
4. The user interaction method according to claim 2, characterized in that: The step of fusing and sorting the multiple channels of store recall data to obtain target recall store data includes: Based on the user's portrait information, multiple channels of store recall data are optimized, merged and sorted to obtain optimized store recall data; The inverse sorting fusion method is adopted to fuse and sort the multiple store recall data and the optimized store recall data to obtain the target recall store data.
5. The user interaction method according to claim 4, characterized in that: The optimizing, fusing and sorting the multiple store recall data based on the user's portrait information to obtain optimized store recall data includes: determining store characteristic information of each of the recalled stores according to the store information of each of the recalled stores in the multiple channels of store recall data; Based on the user's portrait information, the large language model is used to filter the user's interest information to obtain portrait feature information; Based on the portrait feature information and the store feature information, multiple channels of store recall data are optimized, fused and sorted to obtain the optimized store recall data.
6. The user interaction method according to claim 5, characterized in that: The multiple store recall data include first store recall data based on the store query information for structured query statement recall, and second store recall data based on the store query information for semantic vector recall. The multiple store recall data are optimized, fused and sorted based on the portrait feature information and the store feature information to obtain the optimized store recall data, including: Determine spatial distance information between each of the recalled stores and the user according to the store information of each of the recalled stores and the vehicle driving trajectory data; According to the spatial distance information, the portrait feature information and the store feature information, a pre-trained re-ranking model is used to fuse and rank the first store recall data and the second store recall data to obtain the optimized store recall data.
7. The user interaction method according to any one of claims 1 to 6, characterized in that: The multiple store recall data includes first store recall data and second store recall data. The multiple store recall data is obtained by performing multiple store recall based on the user's store query information, including: Converting the store query information into a structured query statement, and performing data recall on the structured store information based on the structured query statement to obtain the first store recall data; When the first store recall data does not meet the quantity input requirement of the large language model, the store query information is converted into semantic vector information, and the vectorized store information is recalled based on the semantic vector information to obtain the second store recall data.
8. The user interaction method according to any one of claims 1 to 6, characterized in that: The interacting with the user based on the target recommended store data includes: Feeding back the target recommended store data to the user, and after receiving the user's input information again, using the large language model to perform context-based intent recognition and input expansion on the user's input information to obtain store inquiry information and the inquiry target store; Based on the store inquiry information and the store information of the inquiry target store, the large language model is used to generate a reply answer, and the reply information of the current input information is obtained and pushed to the user.
9. The user interaction method according to claim 8, characterized in that: The target recommended store data includes the ranking number, store information and recommendation information of each recommended store, and the feeding back of the target recommended store data to the user includes: Based on the store information and the recommendation information of each of the recommended stores, generating a store picture and text display page according to the ranking number of each of the recommended stores; The store picture and text display page is used as a response to the user's input and displayed to the user through an interactive interface of a terminal device.
10. A user interaction system, characterized in that: It includes a terminal device and a server, wherein the server is used for: Perform multi-channel data recall based on the user's store query information to obtain multi-channel store recall data; Based on the store query information and the store information of each recalled store in the multiple store recall data, a large language model is used to perform store sorting based on user intention to obtain target recommended store data; Based on the target, store data is recommended and interaction is performed with the user through the terminal device.
11. A vehicle, characterized in that: Including vehicle-mounted terminal; The vehicle-mounted terminal is used to obtain input information of the user in the vehicle-mounted terminal and send it to the server; Acquire target recommended store data fed back by the server based on the input information, and interact with the user based on the target recommended store data; The target recommended store data is obtained in the following way: Identifying and obtaining store query information based on the user's input information, and performing multi-channel data recall based on the store query information to obtain multi-channel store recall data; Based on the store query information and the store information of each recalled store in the multiple store recall data, a large language model is used to sort the stores based on user intention to obtain target recommended store data.
12. A computer 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 steps of the user interaction method according to any one of claims 1 to 9 are implemented.
13. A readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the user interaction method according to any one of claims 1 to 9 are implemented.
Citation Information
Patent Citations
Shop recommendation method and device, electronic equipment and storage medium
CN117273868A
Method and device for recommending articles based on large language model and search engine
CN117391824A
Method and system for realizing e-commerce commodity information interaction by using large language model
CN117592489A
Object list recommendation method and device, electronic equipment and readable storage medium
CN118210975A
Cited By
Recommendation method and device of undertaking unit and server
CN119202396A
Shop retrieval method and device based on large model technology, equipment and medium
CN119540922A
Searching method and system for vehicle-mounted system and storage medium
CN120196802A
Search method, system and storage medium for vehicle-mounted system
CN120196802B
