Service providing method for vehicle, service providing device for vehicle, computer program product, and vehicle
By analyzing the intent of passenger requests in the vehicle and performing data domain fusion processing, and utilizing a large language model to understand and respond to user requests, the problem of data silos in in-vehicle service systems is solved, thereby improving the intelligence of in-vehicle services and the user experience.
Patent Information
- Application Number
- CN202511478724.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-16
- Publication Date
- 2026-01-16
AI Technical Summary
Existing in-vehicle service systems cannot effectively combine multi-source heterogeneous data in vehicles with user commands, resulting in low accuracy and personalization of responses.
By acquiring passenger service requests, analyzing the request intent, determining the relevant data domains, performing data fusion processing to generate a fused data table, and executing corresponding in-vehicle application operations based on this data table, the system utilizes a large language model to understand and respond to complex user requests.
It enables intelligent responses to complex query requests from passengers, improves the intelligence level of in-vehicle services and user experience, simplifies human-computer interaction, and enhances the accuracy of data correlation analysis and response.
Smart Images

Figure CN121350203A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of vehicles, and more specifically to a service provision method for vehicles, a service provision device for vehicles, a computer program product, and a corresponding vehicle. Background Technology
[0002] With the rapid development of society and the economy, automobiles have become an important means of transportation for people's daily travel. In recent years, various applications providing services in automobiles have received widespread attention. For example, traditional in-vehicle voice assistant systems can respond to user voice commands using keyword matching, but they can usually only execute simple commands, such as playing music or navigating to a destination, which cannot meet the growing personalized needs of vehicle users. In addition, modern vehicles are usually equipped with a large number of monitoring devices that can collect massive amounts of data, such as trip data, entertainment application data, vehicle status data, and environmental data. However, existing in-vehicle service delivery systems often cannot combine this data with user commands, resulting in low accuracy and personalization of responses.
[0003] Therefore, there is still a real need for continued improvement in the provision of in-vehicle services. Summary of the Invention
[0004] In view of this, the object of the present invention is to provide an improved method for providing services for a vehicle, an improved apparatus for providing services for a vehicle, an improved computer program product, and a corresponding vehicle, so as to at least solve some of the problems in the prior art and / or overcome other possible disadvantages not mentioned herein.
[0005] According to a first aspect of the present invention, a service provision method for a vehicle is provided, wherein the service provision method includes the following steps: S100: obtaining a service request from a vehicle occupant; S200: analyzing the service request to determine the occupant's request intent; S300: determining at least one data domain involved based on the request intent; S400: extracting relevant data from the corresponding data domain of the vehicle based on the at least one data domain and performing fusion processing to generate a fused data table; S500: performing corresponding in-vehicle application operations based on the fused data table and according to the request intent to respond to the service request.
[0006] According to an optional embodiment of the present invention, the service provision method includes the following steps: S220: using a language model to perform natural language understanding on the service request to obtain a query statement containing extraction conditions as the request intent.
[0007] According to an optional embodiment of the present invention, the service provision method includes the following steps: S310: when the extraction conditions involve multiple data domains, determine N data domains corresponding to the vehicle as target data domains, where N is an integer greater than 1; S410: extract relevant data that meets the extraction conditions from one of the N data domains as a base record; S420: extract relevant data that meets the extraction conditions from the remaining data domains of the N data domains, and merge it with the base record according to timestamps or other associated attributes to form the fused data table, wherein the extraction conditions preferably include time information and / or event information identified from the service request as conditions to limit the extraction range of the relevant data.
[0008] According to an optional embodiment of the present invention, in step S420, data in the same data domain is vertically expanded in the fused data table to generate fused data columns that allow the query statement to locate, and data in different data domains is horizontally expanded in the fused data table to generate fused data rows that allow the query statement to locate, wherein each row in the fused data table contains cross-domain data that occurs in the same time period or related events; in step S500, based on the fused data rows and fused data columns located in the fused data table by the query statement, an operation instruction is generated to execute the corresponding vehicle application operation.
[0009] According to an optional embodiment of the present invention, the service provision method includes the following steps: S510: constructing prompt words containing contextual information for the fused data table based on the query statement; S520: formatting the prompt words into a structure suitable for processing by the language model; S530: sending the structured prompt words to the language model for processing to obtain processing results; S540: generating operation instructions based on the processing results to execute corresponding in-vehicle application operations.
[0010] According to an optional embodiment of the present invention, the service provision method includes the following steps: S10: when the processing result of the language model contains multiple possible options and / or uncertain information, a dialogue with the occupant is initiated to obtain the occupant's confirmation; S20: based on the occupant's feedback in the dialogue, the data of the vehicle is updated, the update preferably includes adding tags for specific locations, correcting data associations, and / or supplementing missing information.
[0011] According to an optional embodiment of the present invention, in step S10, the dialogue strategy of the language model and / or the execution mode of the processing result are dynamically adjusted based on the historical interaction behavior of the occupant.
[0012] According to an optional embodiment of the present invention, in step S20, when unmarked locations and / or unmarked events are mentioned in the dialogue, type labels are added to the unmarked locations and / or unmarked events based on the feedback from the occupant.
[0013] According to a second aspect of the present invention, a service provisioning apparatus for a vehicle is provided, wherein the service provisioning apparatus comprises: an acquisition unit configured to acquire a service request from a occupant of the vehicle; an analysis unit configured to analyze the service request to determine the occupant's request intent; a determination unit configured to determine at least one data domain involved based on the request intent; a fusion processing unit configured to extract relevant data from a corresponding data domain of the vehicle based on the at least one data domain and perform fusion processing to generate a fused data table; and an execution unit configured to perform a corresponding in-vehicle application operation based on the fused data table and according to the request intent in response to the service request, wherein each unit of the service provisioning apparatus preferably performs a step of the service provisioning method provided in the embodiments of the first aspect described above.
[0014] According to a third aspect of the present invention, a computer program product is provided, comprising computer program instructions, wherein, when executed by a processor, the computer program instructions implement the steps of the service provision method provided in the embodiments of the first aspect described above.
[0015] According to a fourth aspect of the present invention, a vehicle is provided, comprising the service providing apparatus provided in the embodiments of the second aspect above and / or the computer program product provided in the embodiments of the third aspect above.
[0016] Through certain embodiments of the present invention, multi-source heterogeneous data generated during vehicle use is fully utilized. Cross-domain fusion is used to form a unified data table, enabling the system to intelligently respond to complex query requests from passengers. The large table structure after data fusion facilitates structured retrieval and correlation analysis, providing data support for subsequent in-vehicle application operations. This simplifies human-computer interaction and enhances the user experience.
[0017] According to certain embodiments of the present invention, by utilizing multi-source heterogeneous data generated during vehicle use and its fusion, prompt words can be automatically generated to perform semantic queries using a large language model. By leveraging the reasoning capabilities of the large language model, more intelligent understanding and execution of user instructions / service requests involving cross-domain data, containing contextual information, and being more complex and / or closer to natural language can be achieved. Attached Figure Description
[0018] The invention will now be described in more detail with reference to the accompanying drawings, which will provide a better understanding of its principles, features, and advantages. The drawings include: Figure 1 A schematic diagram of a vehicle according to an embodiment of the present invention is shown; Figure 2 A schematic diagram of a scenario according to an embodiment of the present invention is shown; Figure 3 A schematic diagram of a fused data table according to an embodiment of the present invention is shown; Figure 4 A schematic diagram of a fused data table according to another embodiment of the present invention is shown; Figure 5 A schematic flowchart of a service provision method for a vehicle according to an embodiment of the present invention is shown; Figure 6 A schematic flowchart illustrating some steps of a service provision method for a vehicle according to an embodiment of the present invention is shown; Figure 7 A schematic flowchart illustrating some steps of a service provision method for a vehicle according to an embodiment of the present invention is shown; Figure 8 A schematic flowchart illustrating some steps of a service provision method for a vehicle according to an embodiment of the present invention is shown; Figure 9 A schematic flowchart illustrating some steps of a service provision method for a vehicle according to an embodiment of the present invention is shown; Figure 10 A schematic structural framework diagram of a service provision device for a vehicle according to an embodiment of the present invention is shown; and Figure 11 A structural framework diagram of a computer system according to an embodiment of the present invention is shown. Detailed Implementation
[0019] To make the technical problems to be solved, the technical solutions, and the beneficial technical effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and several exemplary embodiments. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the scope of protection of the present invention. Various embodiments may share the same view or multiple views for description, but not all features appearing in the same view should be interpreted as features that must be present in an embodiment.
[0020] Exemplary embodiments of the present invention will now be described with reference to the accompanying drawings.
[0021] Figure 1A schematic diagram of a vehicle 2000 according to an embodiment of the present invention is shown; Figure 2 A schematic diagram of a scenario according to an embodiment of the present invention is shown; Figure 3 A schematic diagram of a fusion data table T according to an embodiment of the present invention is shown; Figure 4 A schematic diagram of a fusion data table T according to another embodiment of the present invention is shown; Figure 5 A schematic flowchart of a service provision method 1000 for a vehicle 2000 according to an embodiment of the present invention is shown.
[0022] like Figure 5 Combination Figures 1 to 4 As shown, the service provision method 1000 for vehicle 2000 exemplarily includes steps S100, S200, S300, S400 and S500.
[0023] In step S100, the service request R of the occupant 2001 of vehicle 2000 is obtained.
[0024] In step S200, the service request R is analyzed to determine the request intent of occupant 2001.
[0025] In step S300, at least one data field is determined based on the request intent.
[0026] In step S400, relevant data is extracted from the corresponding data field of vehicle 2000 according to at least one data field and fused to generate fused data table T.
[0027] In step S500, based on the fused data table T and according to the request intent, the corresponding in-vehicle application operation is performed to respond to the service request R.
[0028] Specifically, such as Figure 2As shown, occupant 2001 generates a large amount of data during the use of vehicle 2000. This data may include, for example, trip data 10, entertainment application data 20, vehicle status data 30, environmental data 40, and in-vehicle call data (not shown), as well as various other possible data types related to vehicle 2000. Trip data 10 may include, for example, occupant 2001's travel records, such as origin, destination, departure time, arrival time, and trip distance. For example, a trip record from home to work may include origin GPS coordinates, destination GPS coordinates, departure time "2023-11-01 10:00", arrival time "2023-11-01 10:30", origin label "home", destination label "work", and distance "10KM". Entertainment application data 20 may include, for example, records of occupant 2001's use of the entertainment system within vehicle 2000, such as music listened to, videos watched, and radio stations used. For example, music playback records may include information such as song title "Song X", artist "Artist A", playback start time "2023-11-01 10:05", and playback end time "2023-11-01 10:10". Vehicle status data 30 may include records related to the status and operation of the vehicle 2000, such as refueling, charging, and maintenance. For example, refueling records may include information such as refueling time and amount of fuel. Environmental data 40 may include records related to the usage environment of the vehicle 2000, such as interior and exterior temperatures, road conditions, and weather. For example, temperature records may include information such as time, interior temperature, and exterior temperature. This data is typically collected by various sensors of the vehicle 2000 and stored in different data tables on the vehicle terminal and / or uploaded to the cloud. Within the scope of this invention, these different types of data can be referred to as different "data domains", each corresponding to a specific type of data.
[0029] Based on this, the exemplary process of service provision method 1000 can be implemented as follows: First, for example, the voice request of the occupant 2001 in the vehicle is obtained through the vehicle microphone of the vehicle 2000 as a service request R, and the voice request is converted into text form.
[0030] Then, the service request R is sent to the analysis unit on the vehicle or in the cloud. For example, through... Figure 2 The example shows a language model M that has been pre-trained and fine-tuned for in-vehicle scenarios. It analyzes the service request R, extracts key information, and determines the true request intent of passenger 2001.
[0031] Next, determine the target data domain to be queried based on the request intent, preferably through methods such as... Figure 2The intent recognition function G, as illustrated below, can be implemented through traditional Natural Language Understanding (NLU) or by using a Large Language Model (LLM). This allows us to determine that the occupant's request intent may involve trip data 10, entertainment application data 20, vehicle status data 30, environmental data 40, and in-vehicle communication data.
[0032] Subsequently, the required relevant data is extracted from the target data domain of vehicle 2000. Here, the various data domains can be merged into a wide table, namely the merged data table T, based on correlation attributes such as timestamps. Each row of this merged data table T can, for example, represent a specific point in time or all cross-domain data related to an event. The intent recognition function G can also be used in the merged data table T, as well as... Figure 2 As illustrated by the dashed arrow, the data in the fusion data table T is filtered or extracted based on the actual request intent of the occupant 2001.
[0033] Finally, based on all the data presented in the fusion data table T, and according to the request intent determined by the service request R, the corresponding operation instructions are generated and sent to the execution unit of the vehicle 2000 to control the in-vehicle applications such as navigation, audio and video, and cabin environment to perform corresponding operations, thereby realizing the response to the occupant's service request R.
[0034] As can be seen, the service provision method for vehicles provided in this embodiment of the invention, by introducing intent analysis and data domain fusion processing, enables the following: First, accurately understanding the true intent expressed in the passenger's service request; then, determining the relevant data domains to be invoked based on this intent, which may include navigation history, media usage records, vehicle status, etc.; next, extracting relevant data from these data domains and performing cross-domain fusion to form a highly correlated fused data table; finally, performing corresponding operations based on this context-rich fused data to provide passengers with accurate service responses. Therefore, this service provision method overcomes the data silo limitations of traditional in-vehicle systems, enabling in-vehicle service systems to understand and respond to complex requests involving cross-domain data, significantly improving the intelligence level and user experience of in-vehicle services.
[0035] Figure 6 A schematic flowchart illustrating some steps of a service provision method 1000 for a vehicle 2000 according to an embodiment of the present invention is shown. Figure 7 A schematic flowchart illustrating some steps of a service provision method 1000 for a vehicle 2000 according to an embodiment of the present invention is shown. Figure 8A schematic flowchart illustrating some steps of a service provision method 1000 for a vehicle 2000 according to an embodiment of the present invention is shown. Figure 9 A schematic flowchart illustrating some steps of a service provision method 1000 for a vehicle 2000 according to an embodiment of the present invention is shown.
[0036] Combination Figures 1 to 5 Especially Figure 2 As shown by the solid arrow in the image, the exemplary flow of service provision method 1000 can be further implemented as follows: In step S100, the service request R of the occupant 2001 is still obtained, for example, through the vehicle microphone or other input device of the vehicle 2000. Here, the service request R is usually in the form of natural language, such as "Navigate to the court where I played ball last Wednesday night" or "I listened to a song by singer A on my way to work yesterday, can you play it for me?" etc.
[0037] In step S200, the service request R is analyzed to determine the request intent of occupant 2001. Specifically, as follows... Figure 6 The diagram includes steps S210 and S220.
[0038] In optional step S210, a language model M is provided, specifically a trained large language model.
[0039] In step S220, the language model M is used to perform natural language understanding on the service request R in order to obtain a query statement containing extraction conditions as the request intent.
[0040] In a preferred embodiment, the language model M can be a large language model based on the Transformer architecture, such as the BERT model or the LLaMA model. The language model M should be trained on a large amount of text data to possess strong natural language understanding capabilities, thereby accurately extracting intent and key information from the natural language requests of occupant 2001. In practical applications, a language model of appropriate size can be selected based on the vehicle's computing resources. For example, a model with 7B parameters (i.e., containing 7 billion trainable parameters) can be selected. This size of model already possesses good natural language understanding capabilities, while its computing resource requirements are relatively moderate, making it suitable for deployment in an in-vehicle environment. In another practical application, the language model can also be deployed on a cloud server, with the vehicle calling it via a network API; or a hybrid deployment mode can be used, where a small language model is deployed locally in the vehicle to handle basic requests, and a large model in the cloud is called when more powerful processing capabilities are needed, thus balancing response speed and processing power.
[0041] In another embodiment, natural language understanding of the service request R can be performed using traditional Natural Language Understanding (NLU) methods, which mainly include rule engines, statistical learning methods, and tree-based methods. However, these methods require the manual design of a large number of features and rules, and the manual annotation of a large amount of training data. Therefore, compared with traditional NLU methods, using Large Language Models (LLM) for natural language understanding not only has stronger generalization ability, such as the ability to understand multiple unlabeled expressions, but also better contextual understanding ability, and can handle sentences containing referential and implicit information, such as handling information transmission in multi-turn dialogues, such as continuous dialogues like "Who is the most popular member of Deyun Society now?" or "Play one of his crosstalk performances".
[0042] In step S300, at least one data field is determined based on the request intent. Specifically, this may be as follows: Figure 7 The diagram includes step S310.
[0043] In step S310, when the extraction conditions involve multiple data fields, the corresponding N data fields of vehicle 2000 are determined as target data fields, where N is an integer greater than 1.
[0044] For example, for the service request R, "I listened to a song by artist A on my way to work yesterday, could you play it again for me?", the system can identify that the request involves, for example, Figure 2 The two data fields shown are: Trip Data Field 10 (containing trip information from the origin to the company labeled "company") and Entertainment Application Data Field 20 (containing music playback records).
[0045] In step S400, based on at least one data field, relevant data is extracted from the corresponding data field of the vehicle 2000 and fused to generate a fused data table T. Specifically, as follows... Figure 7 The diagram includes steps S410 and S420.
[0046] In step S410, relevant data that meets the extraction conditions is extracted from one of the N data domains and used as the base record K0. For example, one of the N data domains is selected as the main domain, and data that meets the conditions is extracted as the base record K0. Typically, the principle for selecting the base record K0 is to choose the data domain most directly related to the user request. For example, for navigation requests, the trip data domain is usually the main domain; for media playback requests, the entertainment application data domain is usually the main domain. However, in order to better associate different data domains in the next step S420, it is preferable to use the data domain that is closely related to time (or the corresponding attribute that can play a correlation role in the service request R) as the main domain. For example, for the service request R "Yesterday on my way to the company, I listened to a song by singer A, can you play it back for me?", the trip data domain 10 should be used as the main domain, while for the service request R "Yesterday when I called Xiao C, I listened to a song by singer A, can you play it back for me?", the in-vehicle call data domain should be used as the main domain.
[0047] In step S420, relevant data that meets the extraction conditions is extracted from the remaining data domains among the N data domains, and merged with the base record K0 from step S410 according to timestamps or other association attributes to form a fused data table T. Here, for example, the attributes used to associate data from different data domains are first determined. As mentioned above, the most commonly used association attribute is timestamps, because most vehicle data contains time information. In other cases, other association attributes, such as geographic location, can also be used.
[0048] Preferably, data from the same data domain is vertically expanded in the fused data table T to generate fused data columns that allow query statements to locate, while data from different data domains is horizontally expanded in the fused data table T to generate fused data rows that allow query statements to locate, so that each row in the fused data table T contains cross-domain data that occurred in the same time period or related events. Figure 3 The base record K0 is illustrated by a dashed box, and its vertical extension K1 and horizontal extension K2 are illustrated by dashed boxes. That is, when a base record K0 is associated with multiple records from other fields, the base record K0 (e.g., ...) needs to be linked to other records in the same field. Figure 3 The itinerary data shown in the example is copied multiple times, each copy associated with a single record (such as...). Figure 3 (Example audio playback data) combinations. For example, if multiple songs by artist A were played on "yesterday on my way to work," such as... Figure 3As shown in the example, if songs X and Y by artist A are played, then the merged data table T will have two rows of merged data (song Z by artist B is not considered here). For data from different data domains, the system creates new columns in the merged data table T, each column corresponding to a field of a data domain. For example, the merged data table T may contain columns related to the trip (such as origin, destination, trip time, etc.) and columns related to music playback (such as song name, artist, playback time, etc.).
[0049] Here, Figure 3 and Figure 4 The correspondence between the letters and their meanings in the fusion data table T, as illustrated schematically, can be found in Table 1 below: Table 1
[0050] Figure 4 An example shows a ratio Figure 3 Another more complex fusion data table, T. For example, when constructing fusion data table T, if passenger 2001 listened to multiple songs during a trip, this trip record can be copied multiple times and associated with each song record; and if passenger 2001 also answered a phone call while listening to a song, a call record column is added horizontally to the corresponding row. For example, for the service request R, "Yesterday on my way to the company, I called Xiao D and listened to a song by artist A. Can you play it again for me?", the trip data field 10 can first be used as the main field, and relevant data can be extracted as the base record K0 (and... Figure 3 (Similarly), then expand horizontally and vertically accordingly, and... Figure 3 The difference is that, since this service request R involves three data domains: trip data domain 10, entertainment application data domain 20, and in-vehicle call data domain, the fused data table T is further expanded with parts K3, K4, K5, and K6, taking into account time correlation. Although this expansion introduces some data redundancy, it significantly simplifies subsequent query operations, eliminating the need for complex table joins.
[0051] like Figure 3 and Figure 4 As illustrated, the fused data table T integrates data from multiple data domains. For historical data with a long time span (e.g., vehicle 2000 has been used for a year), the system can trim the data according to the request intent, for example, extracting only data records related to "yesterday" or "last Wednesday night," instead of loading all historical data. Furthermore, the system can filter data based on key events in the request (e.g., "on the way to the company" or "playing ball"), which helps improve processing efficiency and reduce resource consumption.
[0052] On the one hand, time-based information extraction refers to filtering data records within a specific time range based on the time constraints specified in the service request. This time information can be expressed in various ways, including absolute time (specific dates and times, such as "July 15, 2023," "10:00 AM," etc.), relative time (expressions relative to the current time, such as "yesterday," "last Wednesday," "half an hour ago," etc.), and fuzzy time (imprecise time ranges, such as "recently," "last time," etc.). Regardless of the time expression, the system needs to convert these expressions into specific time ranges before using them for data filtering. For example, "yesterday" should be converted to the time range from 0:00 to 23:59:59 of the previous day, and "last Wednesday evening" should be converted to the time range from 18:00 to 23:59:59 of last Wednesday.
[0053] On the other hand, event information extraction refers to filtering relevant data records based on specific events or activities mentioned in a user request. This event information can be expressed in various ways, including behavioral activities (such as "playing ball," "eating," "shopping"), location associations (such as "on the way to work," "on the way home"), person associations (such as "with Zhang San," "picking up Xiaoming from school"), and status descriptions (such as "rainy day," "traffic jam"). Regardless of the event expression, the system needs to identify this event information and map it to corresponding data features. For example, "on the way to work" can be mapped to a travel record with home (or another location) as the starting point and the company as the destination; "playing ball" may require inference based on the occupant's historical behavior or location features (such as a sports field); and "with Zhang San" could involve functions such as facial recognition via an in-vehicle camera.
[0054] Specifically, the system can extract and correlate data from various data domains using a mechanism similar to SQL statements. For example, when passenger 2001 inquires about the itinerary for July 2nd, the system can generate an SQL statement to query the itinerary data for July 2nd and present the results as a structured data table.
[0055] Having already constructed the fused data columns and rows that allow query statements to locate data based on step S420, in step S500, the system can generate operation instructions to execute corresponding in-vehicle application operations (such as informing passengers of specific travel information via voice playback) based on the specific fused data row and column directly located in the fused data table T by the query statement (e.g., directly locating the travel information for July 2nd). To save resources, the fused data table T is preferably used only for the current request and is not permanently saved.
[0056] In another embodiment, step S500 may be as follows: Figure 8 The steps shown are S510, S520, S530 and S540.
[0057] In step S510, prompt words containing contextual information are constructed for the fused data table T based on the query statement.
[0058] In step S520, the prompt word is formatted into a structure suitable for processing by the language model M.
[0059] In step S530, the structured prompt words are sent to the language model M for processing to obtain the processing results.
[0060] In step S540, operation instructions are generated based on the processing results to execute the corresponding vehicle application operations.
[0061] Here, the system can construct a prompt word P containing rich contextual information based on the original service request from passenger 2001 and the fused data table, enabling the language model M, especially the large language model, to understand the background of the original service request and the available information resources. This step... Figure 2 Examples are illustrated with arrows. In other words, prompt word formatting refers to converting the constructed prompt words into a format suitable for processing by the target language model. Different language models may have different input requirements and optimization methods. For example, some language models perform better with JSON format input, while others are more suitable for natural language input. In step S520, prompt word formatting can preferably take into account the following aspects: such as model characteristics, context window, information priority, instruction clarity, and / or example guidance. Considering model characteristics is because different language models have different requirements and preferences for input formats, requiring targeted design. Considering the context window means taking into account the context window limitations of the language model to ensure that important information is within the effective range. Considering information priority requires ensuring that the most important information is placed in a key position of the prompt word, such as at the beginning or end. Considering instruction clarity requires using clear and specific instruction language to avoid ambiguity or vagueness. More preferably, necessary examples can guide the language model to generate the expected format of output. Then, in step S530, the formatted prompt words are sent to the language model (especially a large language model) for processing. The language model generates the corresponding processing results based on the task description, data content, and question in the prompt words. For example, for a service request like "Please tell me my total mileage on June 15th," the system's suggested prompts might include: the passenger's question, a description of the data table's structure (such as the meaning of rows and columns), and the trip data for June 15th. The large language model then receives these prompts and, combined with contextual information, provides an accurate answer.
[0062] It should be understood that the processing results of the large language model need to be further converted into specific operation instructions in step S540. This conversion process can be completed by the vehicle's execution unit, for example. The execution unit can parse the processing results, extract key information, and generate operation instructions that meet the requirements of the in-vehicle application interface. For example, it can convert JSON format song information into playback commands for a media playback system, or location information into navigation commands for a navigation system.
[0063] In a further preferred embodiment, the service provision method 1000 (particularly in step S220 and / or step S540) may be as follows: Figure 9 The diagram includes steps S10 and S20.
[0064] In step S10, when the processing result of the language model M contains multiple possible options and / or uncertain information, a dialogue with the occupant 2001 is initiated to obtain the occupant 2001's confirmation.
[0065] In step S20, based on the feedback from the occupant 2001 in the dialogue, the original data in the data domain of the vehicle 2000 is updated, preferably by adding tags at specific locations, correcting data associations, and / or supplementing missing information. This allows for continuous enrichment of data tags through interaction with the occupant. When the dialogue involves unmarked locations or events, corresponding type tags can be added based on the occupant's feedback. For example, the system can automatically identify service requests or unmarked location references appearing in the dialogue and learn their semantic tags through the occupant's dialogue feedback, establishing a personalized location knowledge base. For instance, if the occupant's service request is "navigate to mom's house" but the system cannot confirm the specific location of "mom's house" in the trip data domain or other data domains, it initiates a dialogue to inquire and record the address provided by the occupant, associating it with the semantic tag "mom's house." This enables the system to understand and respond to highly personalized location expressions without requiring the occupant to memorize and input specific addresses, significantly improving the convenience and personalization of navigation services. Furthermore, the system can learn and associate multi-dimensional tags such as location function type, emotional association, and usage scenario, enriching the semantic representation of the location. For example, in addition to labeling a location as a "coffee shop," it might also be associated with tags such as "workplace," "frequently visited," or "favorite place," enabling the system to understand and respond to complex requests such as "navigate to my favorite coffee shop" or "find a place similar to my favorite restaurant." This rich semantic tagging greatly expands the system's location understanding dimensions, supporting more natural and diverse ways of referencing locations.
[0066] Furthermore, the dialogue strategy and / or execution method of the language model M can be dynamically adjusted based on the historical interaction behavior of the occupant 2001. For example, for occupants who require detailed confirmation of preferences, the system can provide more options and information; while for occupants who prefer direct execution, the system can reduce the confirmation step and directly provide the result.
[0067] The following two specific workflow examples illustrate the implementation of this invention in detail: For the service request from passenger 2001 in vehicle 2000, "I listened to a song by singer A on my way to work yesterday, could you play it again for me?", the processing flow for R can be as follows: First, obtain, for example, a service request R presented via voice information: "Yesterday on my way to the company, I listened to a song by singer A. Could you play it back for me?"
[0068] Then, analyzing the service request R, it was determined that the request intent of passenger 2001 involved two target data fields: trip data 10 and entertainment application data 20.
[0069] Next, the relevant data from these two data domains will be merged into a merged data table T in the form of a wide table, which contains information such as song, artist, playback time, start point, and end point.
[0070] Subsequently, based on the fused data table T and the request intent of the service request R, a prompt word is constructed, such as: "Based on the data in the table, it is inferred that 'I listened to a song by artist A on my way to the company yesterday, could you play it for me again?'", and the result is returned in JSON format. {Artist: xx, Song Name: xxx}.
[0071] Therefore, the language model M is used to process the prompt words and return the result: "{Singer: Singer A, Song Name: Song X}".
[0072] Finally, based on the returned results, play artist A's song X.
[0073] In another implementation, the processing flow for the service request R made by occupant 2001 in vehicle 2000 to "navigate to the court where I played golf last Wednesday night" can be as follows: First, obtain, for example, a service request R presented via voice information: "Navigate to the court where I played golf last Wednesday night".
[0074] Then, analyzing the service request R, it was determined that the request intent of passenger 2001 involved trip data 10 as the target data field.
[0075] Next, extract the itinerary record data from last Wednesday night from the itinerary data field 10 and form a fused data table T.
[0076] Subsequently, based on the fused data table T and the request intent of the service request R, a prompt word is constructed, such as: "Based on the data in the table, infer 'Please determine which location is likely to be the stadium mentioned by the crew, and return it in JSON format: {Stadium coordinates: [longitude, latitude], degree of certainty: high / medium / low}.
[0077] After processing by language model M, it was found that there are two trip records with adjacent times and the same intermediate location, but neither is explicitly marked as "ball court". Therefore, it cannot be determined whether they are the ball court mentioned by the passenger. The result is as follows: { "Court coordinates": [116.4886322, 40.0070648] Degree of certainty: Medium } Based on the certainty level of the processing result ("medium") and the historical interaction behavior of occupant 2001, the system determines that occupant 2001 prefers detailed confirmation. Therefore, the system can initiate a dialogue: "You stayed for about 1.5 hours last Wednesday night in xxx Street, Haidian District, Beijing. Is this the stadium you mentioned? Do you need to confirm it on the map?" Occupant 2001, for example, confirms: "Yes, that's the stadium." The system then updates the database, marking the GPS coordinates as "sports stadium" in the trip record, and confirms the navigation intention again: "Okay, the location has been marked as a sports stadium. Do you need to navigate there now?" Occupant 2001, for example, confirms: "Yes, please navigate." The execution unit generates navigation instructions based on the confirmed information, starts the in-vehicle navigation system, and sets the destination to [116.4886322, 40.0070648]. Once occupant 2001 mentions "sports stadium" again, the system can directly recognize it.
[0078] Therefore, it is evident that language models, especially trained large language models, can not only identify which data domains a service request involves, but also more accurately understand the contextual information and implicit logic within the request. Furthermore, large language models can handle information transmission in continuous dialogues, such as dealing with pronoun reference and contextual relationships.
[0079] The technical solution proposed in this invention enables the in-vehicle system to understand and process user requests with complex contexts, significantly improving the naturalness and efficiency of human-computer interaction. With increased usage time, the system can continuously learn and optimize through a feedback mechanism, making in-vehicle services more personalized and precise.
[0080] Figure 10 A schematic structural framework diagram of a service provisioning device 3000 for a vehicle 2000 according to an embodiment of the present invention is shown. Figure 10As shown, and in conjunction with the above embodiments, the service providing apparatus 3000 for a vehicle 2000 includes an acquisition unit 3100, an analysis unit 3200, a determination unit 3300, a fusion processing unit 3400, and an execution unit 3500. Here, the acquisition unit 3100 is configured to acquire a service request R from an occupant 2001 of the vehicle 2000; the analysis unit 3200 is configured to analyze the service request R to determine the occupant 2001's request intent; the determination unit 3300 is configured to determine at least one data domain involved based on the request intent; the fusion processing unit 3400 is configured to extract relevant data from the corresponding data domain of the vehicle 2000 based on at least one data domain and perform fusion processing to generate a fused data table T; the execution unit 3500 is configured to perform corresponding in-vehicle application operations based on the fused data table T and according to the request intent to respond to the service request R. Here, each unit of the service providing apparatus 3000 preferably performs each step of the service providing method 1000 of the above embodiments.
[0081] Figure 11 A structural framework diagram of a computer system according to an embodiment of the present invention is shown. This computer system is configured to perform various steps of the service provision method 1000 provided according to the above embodiments of the present invention. Figure 11 As shown, the computer system includes a memory 1, a processor 2, a communication interface 3, and a bus 4. Here, the memory 1, processor 2, and communication interface 3 are interconnected via the bus 4. According to one aspect of the invention, a computer program product is also provided, comprising computer program instructions, wherein when executed by a processor, particularly by the processor 2 of the aforementioned computer system, the computer program instructions implement the various steps of the service provision method 1000 provided according to the above embodiments of the invention.
[0082] Although specific embodiments of the invention have been described in detail herein, they are given for illustrative purposes only and should not be construed as limiting the scope of the invention. Various substitutions, alterations, and modifications can be conceived without departing from the spirit and scope of the invention.
Claims
1. A service providing method (1000) for a vehicle (2000), wherein, the service providing method (1000) comprises the following steps: S100: obtaining a service request (R) of an occupant (2001) of the vehicle (2000); S200: analyzing the service request (R) to determine a request intention of the occupant (2001); S300: determining at least one data domain involved according to the request intention; S400: extracting relevant data from corresponding data domains of the vehicle (2000) according to the at least one data domain for fusion processing to generate a fusion data table (T); S500: executing a corresponding vehicle application operation based on the fusion data table (T) and according to the request intention to respond to the service request (R).
2. The service providing method (1000) according to claim 1, wherein, the service providing method (1000) comprises the following steps: S220: using a language model (M) to perform natural language understanding on the service request (R) to obtain a query statement containing an extraction condition as the request intention.
3. The service providing method (1000) according to claim 2, wherein, the service providing method (1000) comprises the following steps: S310: in the case where the extraction condition involves multiple data domains, determining N data domains corresponding to the vehicle (2000) as target data domains, N being an integer greater than 1; S410: extracting relevant data satisfying the extraction condition from one data domain of the N data domains as a base record (K0); S420: extracting relevant data satisfying the extraction condition from the remaining data domains of the N data domains and merging with the base record (K0) according to a timestamp or other association attribute to form the fusion data table (T), wherein the extraction condition preferably contains time information and / or event information identified from the service request (R) as a condition for limiting the extraction range of the relevant data.
4. The service providing method (1000) according to claim 3, wherein, in step S420, data in the same data domain is expanded longitudinally in the fusion data table (T) to generate a fusion data column allowing the query statement to be located, and data in different data domains is expanded horizontally in the fusion data table (T) to generate a fusion data row allowing the query statement to be located, wherein each row in the fusion data table (T) contains cross-domain data occurring in the same time period or associated events; in step S500, operation instructions are generated to execute a corresponding vehicle application operation according to the fusion data row and fusion data column located by the query statement in the fusion data table (T).
5. The service providing method (1000) according to any one of claims 2 to 4, wherein, the service providing method (1000) comprises the following steps: S510: constructing a prompt word containing context information for the fusion data table (T) based on the query statement; S520: format the prompt word into a structure suitable for processing by the language model (M); S530: send the structured prompt word to the language model (M) for processing to obtain a processing result; S540: generate an operation instruction according to the processing result to execute a corresponding vehicle-mounted application operation.
6. The service providing method (1000) according to claim 2 or 5, wherein, the service providing method (1000) comprises the following steps: S10: when the processing result of the language model (M) contains multiple possible options and / or uncertain information, initiating a dialogue with the occupant (2001) to obtain the confirmation of the occupant (2001); S20: updating the data of the vehicle (2000) according to the feedback of the occupant (2001) in the dialogue, the updating preferably including adding labels for specific locations, correcting data associations and / or supplementing missing information.
7. The service providing method (1000) according to claim 6, wherein, in step S10, dynamically adjusting the dialogue strategy of the language model (M) and / or the execution mode of the processing result according to the historical interaction behavior of the occupant (2001); and / or in step S20, when unmarked locations and / or unmarked events are involved in the dialogue, adding type labels to the unmarked locations and / or the unmarked events according to the feedback of the occupant (2001).
8. A service providing apparatus (3000) for a vehicle (2000), wherein, the service providing apparatus (3000) comprises: an acquisition unit (3100) configured to acquire a service request (R) of an occupant (2001) of the vehicle (2000); an analysis unit (3200) configured to analyze the service request (R) to determine the request intention of the occupant (2001); a determination unit (3300) configured to determine at least one data domain involved according to the request intention; a fusion processing unit (3400) configured to extract relevant data from corresponding data domains of the vehicle (2000) according to the at least one data domain for fusion processing to generate a fusion data table (T); and an execution unit (3500) configured to execute a corresponding vehicle-mounted application operation based on the fusion data table (T) and according to the request intention to respond to the service request (R), wherein the units of the service providing apparatus (3000) preferably correspond to the execution of the steps of the service providing method (1000) according to any one of claims 1 to 7.
9. A computer program product comprising computer program instructions, wherein, The computer program instructions, when executed by a processor, implement the steps of the service providing method (1000) according to any one of claims 1 to 7.
10. A vehicle (2000) comprising the service providing apparatus (3000) according to claim 8 and / or the computer program product according to claim 9.
