Multi-device distributed natural language processing method and system and vehicle

Abstract

The invention relates to a multi-device distributed natural language processing method and system and a vehicle. The method comprises the following steps: receiving an input question based on a target device; determining a plurality of processing links for processing the input question to obtain an answer to the question and an association relationship among the processing links; and allocating the plurality of processing links to the target device and at least one cooperative device, and enabling the target device and the cooperative device to execute the plurality of processing links in a distributed and cooperative manner according to the association relationship to obtain an answer to the question. According to the method, the computing power consumption of the target equipment can be reduced, the target equipment is ensured to have more computing power for executing the core task, input problem processing can be carried out under the condition that execution of the core task is not influenced, and meanwhile it is ensured that privacy data in the knowledge base cannot be leaked.

Description

Technical Field

[0001] The present disclosure relates to the fields of vehicles and natural language processing, and in particular to a method, system and vehicle for multi-device distributed natural language processing. Background Art

[0002] With the rapid development of artificial intelligence (AI) technology, deep learning, reinforcement learning technology, adversarial learning technology, etc. are constantly developing and iterating, and the application of AI models in natural language understanding, intelligent dialogue, knowledge question and answer, intelligent customer service, image processing and other fields is becoming more and more extensive.

[0003] Currently in the field of natural language processing, common models include: the Transformer-based pre-trained language model BERT (Bidirectional Encoder Representations from Transformers) model, the GPT model (Generative Pre-training Transformer, including a multi-layer Transformer decoder), the large language model LLM (Large Language Model) model, etc. These models can perform natural language understanding and generative processing of answers to questions. Summary of the invention

[0004] During the research and development, it was found that the relevant technologies have the following technical problems: In the relevant technologies, when performing natural language processing, most of the language processing models are deployed based on the current device or the corresponding server, so that the device can process the corresponding language processing tasks; however, in many scenarios, the computing power of some devices is limited, and the language processing task to be performed is only one of many tasks, so the computing power allocated for language processing is limited. In this scenario, due to the limited computing power allocation, the language processing task cannot be successfully executed, or the processing efficiency is low, or the language processing task occupies more computing power, resulting in adverse effects on other major tasks.

[0005] In order to overcome the problems existing in the related art, the present disclosure provides a method, system and vehicle for multi-device distributed natural language processing.

[0006] According to the first aspect of the embodiment of the present disclosure, a method for multi-device distributed natural language processing is provided. The method includes: receiving an input question based on a target device; determining multiple processing links for processing the input question to obtain an answer to the question and the association relationship between the processing links; allocating the multiple processing links to the target device and at least one collaborative device, and allowing the target device and the collaborative device to distribute and collaboratively execute the multiple processing links according to the association relationship to obtain an answer to the question.

[0007] In some embodiments, the above-mentioned multiple processing links are allocated to the above-mentioned target device and at least one collaborative device, including: according to the deployment information of the above-mentioned target device and the above-mentioned collaborative device, the above-mentioned multiple processing links are allocated to the above-mentioned target device and at least one collaborative device; wherein the above-mentioned deployment information includes information on the pre-deployment of the knowledge base or language processing model of the above-mentioned target device and the above-mentioned collaborative device.

[0008] In some embodiments, the corresponding relationship of the allocation of the above-mentioned multiple processing links includes one or more of the following:

[0009] A processing link is assigned to an execution subject, and the execution subject is one or more of the target device or the at least one collaborative device; or,

[0010] A processing link is assigned to an execution subject, and the execution subject is one of the target device or the at least one collaborative device; or,

[0011] One processing link is assigned to two execution subjects and the two execution subjects have a progressive execution relationship, a parallel execution relationship or a cross execution relationship when jointly completing the above processing link; the above two execution subjects are the above target device and the above collaborative device.

[0012] In some embodiments, the above-mentioned processing link includes: retrieving relevant content for the above-mentioned user question. Wherein, in the process that the above-mentioned target device and the above-mentioned collaborative device distribute and collaboratively execute the above-mentioned multiple processing links according to the above-mentioned association relationship, the processing link of retrieving relevant content for the above-mentioned user question is executed, including: retrieving relevant content for the above-mentioned user question from the knowledge base deployed by the first execution subject, and obtaining relevant content corresponding to the above-mentioned user question; wherein the above-mentioned first execution subject is the above-mentioned target device or the above-mentioned collaborative device. Wherein, the above-mentioned knowledge base satisfies at least one of the following conditions: the computing power consumption required for the retrieval of the above-mentioned knowledge base, or the computing power consumption required for storage and retrieval, respectively matches the disposable computing power resources of the above-mentioned first execution subject; or, the privacy of the stored data of the above-mentioned knowledge base requires that only the above-mentioned first execution subject has access to the above-mentioned knowledge base, and other external devices do not have direct access rights and the content indirectly accessed through the above-mentioned first execution subject is restricted.

[0013] In some embodiments, the processing link further includes: generating answers based on the retrieved relevant content and the user questions. In the process of the target device and the collaborative device distributing and collaboratively executing the above multiple processing links according to the above association relationship, the processing link of generating answers based on the retrieved relevant content and the user questions is executed, including: inputting the retrieved relevant content and the user questions into the language processing model deployed by the second execution subject, and outputting the answers to the user questions after referring to the relevant content; wherein the second execution subject is the target device or the collaborative device. In the above language processing model, at least one of the following conditions is met: the computing power consumption required for the language processing model to generate answers, or the computing power consumption required for deployment and answer generation are respectively matched with the available computing power resources of the second execution subject; or, the privacy of the stored data of the language processing model requires that only the second execution subject has access to the language processing model, and other external devices do not have direct access rights and the content indirectly accessed through the second execution subject is restricted.

[0014] In some embodiments, the above-mentioned multiple processing links include a combination of the following basic processing links, or a combination of the following basic processing links and one or more supplementary processing links:

[0015] The first basic processing link: search for relevant content for user questions;

[0016] The second basic processing step: generating answers based on the retrieved relevant content and the above user questions;

[0017] The first supplementary processing step: classifying the user questions to obtain question categories; wherein the above-mentioned searching for relevant content for the user questions includes: searching for relevant content according to the user questions and the question categories;

[0018] Second supplementary processing step: screening and enhancing the retrieved relevant content according to the relevance between the retrieved relevant content and the user question to obtain accurate relevant content; wherein, generating an answer based on the retrieved relevant content and the user question includes: generating an answer based on the accurate relevant content and the user question;

[0019] The third supplementary processing link: verify the generated answers to the questions and output them after they pass the verification.

[0020] In some embodiments, the target device and the collaborative device jointly share the task of processing user questions to obtain answers to the questions, wherein the target device itself also undertakes core tasks, and the importance of the processing tasks is lower than that of the core tasks.

[0021] The target device and the collaborative device are one of the following:

[0022] The target device is an on-board device in a vehicle, and the cooperative device is a terminal device in the vehicle; or,

[0023] The target device is an on-board device in a vehicle, and the collaborative device is a cloud device; or,

[0024] The target device is a first device in the target space, and the collaborative device is one or more second devices in the target space; or,

[0025] The target device is a target drone in a drone group, and the coordinated device is one or more drones other than the target drone in the drone group;

[0026] The target device is a robot that supports question answering, and the collaborative device is an electronic device deployed with a knowledge base or a language processing model; or,

[0027] The target device is a warehousing and logistics device that supports question answering, and the collaborative device is an electronic device deployed with a knowledge base or a language processing model; or,

[0028] The target device is an exploration device that supports question and answer, and the collaborative device is an electronic device deployed with a knowledge base or a language processing model.

[0029] In some embodiments, the target device and the collaborative device communicate based on a data distribution service protocol. The target device, as a publisher, publishes the processing links shared by the collaborative device as sub-tasks, and the collaborative device, as a subscriber, receives the corresponding sub-tasks and feeds back the processing results of the sub-tasks to the target device.

[0030] According to the second aspect of the embodiment of the present disclosure, a system for distributed natural language processing of multiple devices is provided. The above-mentioned system includes a target device and at least one collaborative device. A first question receiving module is arranged in the above-mentioned target device, and is used to receive an input question. A first processing link determination module is arranged in the above-mentioned target device or the above-mentioned collaborative device, and is used to determine a plurality of processing links for processing the above-mentioned input question to obtain an answer to the question and the association relationship between each processing link. A first task allocation module is arranged in the above-mentioned target device or the above-mentioned collaborative device, and is used to allocate the above-mentioned plurality of processing links to the above-mentioned target device and the above-mentioned at least one collaborative device. A first task execution module is arranged in the above-mentioned target device, and is used to execute the corresponding assigned processing link. A second task execution module is arranged in the above-mentioned collaborative device, and is used to execute the corresponding assigned processing link. Among them, the above-mentioned first task execution module and the above-mentioned second task execution module are used to distribute and collaboratively execute the above-mentioned plurality of processing links according to the above-mentioned association relationship to obtain an answer to the question.

[0031] According to a third aspect of an embodiment of the present disclosure, a vehicle is provided. The vehicle has an on-board device and at least one terminal device. A second question receiving module is provided in the on-board device and is used to receive input questions. A second processing link determination module is provided in the on-board device or the terminal device and is used to determine a plurality of processing links for processing the input questions to obtain answers to the questions and the association relationship between the processing links. A second task allocation module is provided in the on-board device or the terminal device and is used to allocate the plurality of processing links to the on-board device and the at least one terminal device. A third task execution module is provided in the on-board device and is used to execute the corresponding assigned processing links. A fourth task execution module is provided in the terminal device and is used to execute the corresponding assigned processing links. The third task execution module and the fourth task execution module are used to execute the plurality of processing links in a distributed and collaborative manner according to the association relationship to obtain answers to the questions.

[0032] The technical solution provided by the embodiments of the present disclosure may have the following beneficial effects:

[0033] By determining multiple processing links for processing input questions to obtain answers to questions and the association between each processing link; allocating the above multiple processing links to the above target device and at least one collaborative device, and making the above target device and the above collaborative device distribute and collaboratively execute the above multiple processing links according to the above association to obtain answers to questions, the dispersion of processing computing power and the distributed and collaborative execution of processing tasks by multiple devices are realized. On the one hand, multiple processing links are distributed in the target device and the collaborative device for distributed processing, which can reduce the computing power consumption of the target device, ensure that the target device has more computing power to perform core tasks, and can process input questions without affecting the execution of core tasks, avoiding the problems of unable to successfully execute language processing tasks or low processing efficiency due to limited computing power allocation in related technologies, or adversely affecting other main tasks due to the large computing power occupied by processing language processing tasks; on the other hand, for some scenarios where knowledge bases have data confidentiality requirements, these knowledge bases can only be deployed on specific devices; by distributing processing computing power, knowledge bases with confidentiality or privacy requirements can be deployed on designated devices, and these designated devices as collaborative devices bear part of the computing power, while realizing the generation of answers to questions, ensuring that privacy data will not be leaked.

[0034] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0036] Figure 1 The present invention is a flowchart of a method for multi-device distributed natural language processing according to an exemplary embodiment.

[0037] Figure 2 It is a schematic diagram of an application scenario of a method for multi-device distributed natural language processing according to an exemplary embodiment.

[0038] Figure 3 It is a schematic diagram of an application scenario of a method for multi-device distributed natural language processing according to another exemplary embodiment.

[0039] Figure 4 The invention is a block diagram of a system for distributed natural language processing on multiple devices according to an exemplary embodiment.

[0040] Figure 5 is a block diagram of a vehicle according to an exemplary embodiment. DETAILED DESCRIPTION

[0041] Exemplary embodiments will be described in detail below with reference to the accompanying drawings.

[0042] It should be pointed out that the relevant embodiments and drawings are only for describing exemplary embodiments provided by the present disclosure, rather than all embodiments of the present disclosure, and it should not be understood that the present disclosure is limited to the relevant exemplary embodiments.

[0043] It should be noted that the terms "first", "second", etc. used in the present disclosure are only used to distinguish different steps, devices or modules, etc. The related terms neither represent any specific technical meanings nor indicate the order or interdependence between them.

[0044] It should be noted that the modifications of the terms "one", "multiple", and "at least one" used in the present disclosure are illustrative rather than restrictive. Unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0045] It should be noted that the term "and / or" used in this disclosure is used to describe the association relationship between associated objects, and generally indicates that there are at least three association relationships. For example, A and / or B can at least indicate the existence of three association relationships: A exists alone, A and B exist at the same time, and B exists alone.

[0046] It should be noted that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. Unless otherwise specified, the scope of the present disclosure is not limited by the order of description of the steps in the relevant embodiments.

[0047] It should be noted that all actions of acquiring signals, information or data in the present disclosure are carried out in compliance with the relevant data protection laws and policies of the country where the device is located and with the authorization given by the owner of the corresponding device.

[0048] During the research and development, it was found that the relevant technologies have the following technical problems: in many scenarios, the computing power of some devices is limited, and the language processing task to be performed is only one of many tasks. Therefore, the computing power allocated for language processing is limited. For example, taking vehicles as an example, the main computing tasks corresponding to safe driving (such as road condition recognition, driving behavior planning and route planning, obstacle avoidance, etc.) are performed, and the computing resources allocated for natural language processing tasks involved in user entertainment interaction are limited. In this scenario, due to the limited computing power allocation, the language processing task cannot be successfully executed, or due to the allocation of computing power and the priority scheduling of computing power to perform more important safe driving computing tasks, the processing efficiency of natural language processing tasks is low, or the processing of language processing tasks takes up more computing power, resulting in adverse effects on other major tasks.

[0049] In addition, in some cases, for some professional questions or when users provide less information in their questions, the results generated by problem processing based on the generative model may have large deviations or even be inconsistent with the common sense in the relevant field, thus affecting the user experience.

[0050] In addition, it is also found that: in the solution of applying knowledge bases to assist in answer generation, the confidentiality or privacy requirements of the data in these knowledge bases are not considered. For some scenarios where knowledge bases have data confidentiality requirements, these knowledge bases can only be deployed on specific devices. Therefore, the embodiment of the present disclosure provides a solution for dispersing computing power and dividing natural language processing tasks to multiple devices for distributed and collaborative execution. This can ensure the confidentiality of the data in the knowledge base and improve the professionalism and accuracy of the answers generated after problem processing.

[0051] In view of this, the embodiments of the present disclosure provide a method, system and vehicle for multi-device distributed natural language processing, which allocate multiple processing links involved in processing input problems to the target device and collaborative devices to share the processing, thereby achieving the dispersion of processing power and the distributed and collaborative execution of processing tasks by multiple devices.

[0052] In some embodiments, RAG (Retrieval-Augmented Generation, a technology that combines knowledge base retrieval and natural language generation) technology is applied to the process of processing input questions. First, relevant content related to the input question is retrieved from the knowledge base (for example, including knowledge extracted from journal articles, question-answer pairs, and knowledge documents), and then a generation model (for example, the language processing model described in the embodiment of the present disclosure) is used to generate answers based on the input question and related content. Specifically, multiple processing links are distributed execution processes of RAG, and multiple processing links are distributed and collaboratively executed by target devices and collaborative devices, which can not only improve the accuracy, professionalism and richness of the generated content, but also ensure that some knowledge bases with data privacy requirements are stored in specific devices (for example, collaborative devices), and in the process of participating in the execution of the task of processing the input question, the data access in the corresponding knowledge base is controlled by itself and the access of other external devices to the database is restricted, thereby ensuring that private data will not be leaked.

[0053] On the one hand, distributing multiple processing links on the target device and collaborative devices for distributed processing can reduce the computing power consumption of the target device, ensure that the target device has more computing power to perform core tasks, and can process input questions without affecting the execution of core tasks, avoiding the problems of failure to successfully perform language processing tasks or low processing efficiency due to limited computing power allocation in related technologies, or adverse effects on other major tasks due to the large amount of computing power occupied by language processing tasks. On the other hand, for some scenarios where knowledge bases have data confidentiality requirements, these knowledge bases can only be deployed on specific devices; by dispersing processing computing power, knowledge bases with confidentiality or privacy requirements can be deployed on designated devices, and these designated devices can bear part of the computing power as collaborative devices, ensuring that private data will not be leaked while achieving the generation of answers to questions.

[0054] The above-mentioned multi-device distributed natural language processing method, system and vehicle are introduced in detail below in conjunction with specific embodiments.

[0055] Exemplary Methods

[0056] Figure 1 The present invention is a flowchart of a method for multi-device distributed natural language processing according to an exemplary embodiment.

[0057] Reference Figure 1 As shown, the method of the embodiment of the present disclosure can be applied to a system composed of a target device and a cooperative device, and includes the following steps: S110, S120 and S130.

[0058] In step S110, an input question is received based on the target device.

[0059] In some embodiments, the input question may be a question asked by the user in voice form, a question input in text form, or a question obtained by the user after uploading an image with text and converting it and performing text recognition.

[0060] Accordingly, the target device may receive input questions in the following ways: receiving questions based on user voice input, receiving questions based on user text input, or receiving images with text uploaded by the user to identify questions contained in the text.

[0061] In different application scenarios, the input question can be expressed in the form of a question, or in the form of a sentence containing the meanings of inquiry, discussion, request for recommendation, request for help, etc.

[0062] The target device refers to the device that receives the input question, and the collaborative device described later refers to other devices that work together with the target device to complete the processing of the input question.

[0063] In some embodiments, the target device and the collaborative device jointly share the task of processing the user's question to obtain an answer to the question.

[0064] Among them, the above-mentioned target device itself also undertakes core tasks, and the importance of the above-mentioned processing tasks is lower than the importance of the above-mentioned core tasks. In some embodiments, in the above-mentioned target device, the importance of the task corresponds to the allocation priority of the computing resources. The higher the importance of the task, the higher the corresponding allocation priority. For example, in the scenario where the target device is an on-board device in a vehicle, the core tasks of the on-board device are tasks related to autonomous driving or assisted driving, such as adaptive cruise control (ACC), autonomous driving or assisted driving, road condition monitoring, path navigation planning, obstacle recognition and obstacle avoidance processing, etc.

[0065] The above-mentioned collaborative device itself can run other tasks or not. In some scenarios, the collaborative device can allocate sufficient computing power to participate in part of the processing links of the above-mentioned processing tasks; in other scenarios, certain designated devices that are allowed to store knowledge bases with data privacy requirements are used as collaborative devices. This collaborative device can ensure that the corresponding computing power participates in part of the processing links of the processing tasks while also ensuring the privacy of data access within the corresponding knowledge base. The knowledge base is accessed by the collaborative device, while other external devices (such as target devices) do not have direct access to the data. The query results obtained through the collaborative device correspond to the fragmented content obtained after the knowledge base query, which is a restricted indirect result viewing, and can ensure the data privacy of the knowledge base.

[0066] Figure 2 It is a schematic diagram of an application scenario of a method for multi-device distributed natural language processing according to an exemplary embodiment.

[0067] In some application scenarios, refer to Figure 2 As shown, the above system may include: 1 target device, 1 or more collaborative devices, Figure 2 In the example, a target device T and three collaborative devices are used, and the three collaborative devices are collaborative device A, collaborative device B and collaborative device C. In this embodiment, RAG technology is applied to the process of processing input questions. First, relevant content related to the input question is retrieved from a knowledge base (for example, including knowledge extracted from journal articles, question-answer pairs, and knowledge documents), and then the language processing model is used to generate answers based on the input question and related content. Specifically, multiple processing links are distributed execution of RAG, and multiple processing links are distributed and collaboratively executed by the target device and the collaborative device.

[0068] In some embodiments, the target device and the collaborative device are pre-deployed with a knowledge base or a language processing model. Deployment means that the knowledge base or the language processing model may be stored on the device, or the device may be able to access the knowledge base or the language processing model.

[0069] A knowledge base or a language processing model can be deployed in the target device; a knowledge base or a language processing model can be deployed in the collaborative device. For the same system, the knowledge base and the language processing model are deployed in a distributed manner in the target device and the collaborative device. For example, if a certain knowledge base is deployed in the target device, other knowledge bases and language processing models are deployed in multiple collaborative devices; if a language processing model is deployed in the target device, multiple categories of knowledge bases are deployed in multiple collaborative devices. Figure 2 As shown, as an example, a lightweight knowledge base KZ for problem classification is deployed in the target device T, an encyclopedia knowledge base KX is deployed in the collaborative device A, a language processing model M is deployed in the collaborative device B, and an audio-visual knowledge base KY is deployed in the collaborative device C.

[0070] It is understandable that a larger number of collaborative devices and corresponding knowledge bases can be expanded, or if the existing collaborative devices meet the computing power and data privacy requirements, more knowledge bases or information or databases required for the corresponding processing links can be deployed in the existing collaborative devices.

[0071] In some embodiments, a knowledge base or language processing model is deployed on a device, and a corresponding deployment allocation relationship is constructed. The allocation principle of this deployment allocation relationship mainly considers at least one of the following: whether the computing power of the device can store the corresponding knowledge base or whether it supports access to a certain knowledge base (if the computing power does not support, it cannot be stored in the corresponding device); or whether a certain device meets the requirements of a certain knowledge base for storing privacy data.

[0072] Therefore, in some embodiments, the target device and the collaborative device are pre-deployed with a knowledge base or a language processing model according to at least one of computing power or data privacy to form deployment information of the target device and the collaborative device. This deployment information describes which knowledge bases or models are deployed on the target device and which knowledge bases or models are deployed on the collaborative device.

[0073] The users of the target device may be: persons who use the target device, and the corresponding input questions of these persons may involve but are not limited to the following fields: daily life (weather, clothing suggestions, news information, daily chat, etc.), leisure and entertainment (music, video, short plays, e-books, online shopping, restaurant recommendations, parking lot recommendations), professional fields (such as mathematical calculations, multi-language conversion and translation, ancient poetry, homework questions, software development, system operation and maintenance, image processing, office software use, electrical appliance repair, etc.), etc.

[0074] In an exemplary application scenario, the target device is an on-board device in a vehicle, and the collaborative device is a terminal device in the vehicle.

[0075] In an example application scenario, the target device is an on-board device in a vehicle, and the collaborative device is a cloud device. For example, the cloud device is a cloud server, or other devices that are deployed with a knowledge base or can access a knowledge base to which the on-board device is connected via the Internet; the communication mechanism can be based on various communication methods, such as short-range communication technology (Bluetooth, Zigbee, etc.), distributed communication technology (such as DDS protocol), Internet communication technology, etc. (4G, 5G, etc.).

[0076] In an exemplary application scenario, the target device is a first device in a target space, and the collaborative device is one or more second devices in the target space.

[0077] In an exemplary application scenario, the target device is a target drone in a drone group, and the cooperative device is one or more drones other than the target drone in the drone group. In some scenarios, in a drone group, the target drone and other drones are in joint operation and have division of labor.

[0078] In an example application scenario, the target device is a robot that supports question answering, and the collaborative device is an electronic device deployed with a knowledge base or language processing model. For example, the robot can be, but is not limited to: some robots with simple functions (such as voice chat robots, intelligent search robots, intelligent customer service robots, sweeping robots, etc.), embodied robots, etc. In some scenarios, the robot and the electronic device can be located in the same preset space.

[0079] In an example application scenario, the target device is a warehousing and logistics device that supports question and answer, and the collaborative device is an electronic device deployed with a knowledge base or a language processing model.

[0080] In an example application scenario, the target device is an exploration device that supports question and answer, and the collaborative device is an electronic device deployed with a knowledge base or a language processing model.

[0081] In step S120, multiple processing steps for processing the input question to obtain the answer to the question and the correlation relationship between the processing steps are determined.

[0082] In some embodiments, the above-mentioned step S120 specifically includes: obtaining a pre-configured question answering processing flow from the configuration file of the above-mentioned target device; parsing the above-mentioned question answering processing flow to obtain multiple processing links for processing the above-mentioned user questions to obtain answers to the questions and the correlation relationship between each processing link.

[0083] In some embodiments, the execution subject of step S120 may be a target device. This embodiment takes the target device as the execution subject for example.

[0084] In other embodiments, the execution subject of step S120 may also be a collaborative device. When the execution subject is a collaborative device, the collaborative device obtains the configuration file of the target device; obtains the pre-configured question answering process from the configuration file; parses the question answering process to obtain multiple processing links for processing the user question to obtain the answer and the relationship between each processing link; or, the collaborative device requests the question answering process from the target device, and the target device searches for the question answering process from its own configuration file and feeds it back to the collaborative device; then the collaborative device parses the question answering process to obtain multiple processing links for processing the user question to obtain the answer and the relationship between each processing link.

[0085] The above-mentioned association relationship may include a sequential execution relationship, an association between input and output, etc., for example, including: the execution order between various processing links, the output of a processing link as the input of another processing link, etc.

[0086] In some embodiments, the above-mentioned multiple processing links include a combination of the following basic processing links: a first basic processing link: retrieving relevant content for user questions; a second basic processing link: generating answers based on the retrieved relevant content and the above-mentioned user questions; that is, the above-mentioned processing links include: retrieving relevant content for the above-mentioned user questions, and generating answers based on the retrieved relevant content and the above-mentioned user questions.

[0087] Taking into account that in some cases, for some professional questions or users' questions, the results generated by problem processing based on the generative model have large deviations and even do not conform to the common sense in the relevant field, affecting the user experience, the embodiment of the present disclosure uses multiple processing links as a distributed execution RAG (retrieve relevant information + refer to relevant information to generate answers) process, and multiple processing links are distributed and collaboratively executed by the target device and the collaborative device. This can not only improve the accuracy, professionalism and richness of the generated content, but also ensure that some knowledge bases with data privacy requirements are stored in specific devices (such as collaborative devices), and in the process of participating in the execution of the input problem processing task, the data access in the corresponding knowledge base is controlled by itself and the access to the database by other external devices is restricted, thereby ensuring that privacy data will not be leaked.

[0088] In other embodiments, the above-mentioned multiple processing steps include a combination of the following basic processing steps and one or more supplementary processing steps:

[0089] The first basic processing link: search for relevant content for user questions;

[0090] The second basic processing step: generating answers based on the retrieved relevant content and the above user questions;

[0091] The first supplementary processing step: classifying the user questions to obtain question categories; wherein the above-mentioned searching for relevant content for the user questions includes: searching for relevant content according to the user questions and the question categories;

[0092] Second supplementary processing step: screening and enhancing the retrieved relevant content according to the relevance between the retrieved relevant content and the user question to obtain accurate relevant content; wherein, generating an answer based on the retrieved relevant content and the user question includes: generating an answer based on the accurate relevant content and the user question;

[0093] The third supplementary processing link: verify the generated answers to the questions and output them after they pass the verification.

[0094] It should be noted that the contents of the first to third supplementary processing links are only examples. In fact, the question answering processing flow can be updated in the configuration file of the target device according to actual needs, so as to adjust and update the overall processing logic. The above update includes: deleting some processing links, adding new processing links, changing the content of processing links, or changing the relationship between processing links.

[0095] In step S130, the above-mentioned multiple processing links are allocated to the above-mentioned target device and at least one collaborative device, and the above-mentioned target device and the above-mentioned collaborative device are made to distribute and collaboratively execute the above-mentioned multiple processing links according to the above-mentioned association relationship to obtain an answer to the question.

[0096] The above-mentioned multiple processing links are allocated between the target device and the collaborative device, and when the target device and the collaborative device each execute the corresponding allocated processing link, they are collaboratively executed according to the above-mentioned association relationship.

[0097] In some embodiments, the above-mentioned multiple processing links are allocated to the above-mentioned target device and at least one collaborative device, including: according to the deployment information of the above-mentioned target device and the above-mentioned collaborative device, the above-mentioned multiple processing links are allocated to the above-mentioned target device and at least one collaborative device; wherein the above-mentioned deployment information includes information on the pre-deployment of the knowledge base or language processing model of the above-mentioned target device and the above-mentioned collaborative device.

[0098] For example, the target device and the collaborative device deploy the knowledge base or language processing model in advance according to at least one of computing power or data privacy. The result after deployment is as follows: a collaborative device is deployed with a corresponding knowledge base according to semantic classification, for example, Figure 2 As shown, collaborative device A is deployed with an encyclopedia knowledge base KX, and collaborative device C is deployed with an audio-visual knowledge base KY.

[0099] Reference Figure 2 As shown, the implementation process of steps S110 to S130 is introduced by taking the above-mentioned multiple processing links including the first basic processing link, the second basic processing link and the first supplementary processing link as an example. The target device T receives the input question Q1, classifies the question based on the knowledge base KZ deployed by itself, and obtains the question category L1 corresponding to the input question Q1. The target device T publishes the retrieval subtask corresponding to the above-mentioned user question Q1 and the above-mentioned question category L1 based on the DDS protocol. Collaborative devices A to C receive the corresponding retrieval subtask based on the subscription mechanism. Since the above-mentioned retrieval requires a knowledge base, the language processing model is deployed in collaborative device B instead of the knowledge base, so collaborative device B does not perform the retrieval subtask; assuming that the question category L1 corresponding to the above-mentioned input question Q1 meets the processing categories of the encyclopedia knowledge base KX of collaborative device A and the audio-visual knowledge base KY of collaborative device C, then collaborative devices A and collaborative devices C respectively perform the above-mentioned retrieval subtask based on their respective knowledge bases, and obtain the relevant content R1 and relevant content R2 corresponding to the input question Q1. Collaborative devices A and C can synchronize the search results to target device T based on the DDS protocol, such as Figure 2As shown, collaborative device A synchronizes relevant content R1 to target device T, and collaborative device C synchronizes relevant content R2 to target device T. Target device T publishes the answer generation subtask containing input question Q1, relevant content R1, and relevant content R2. Collaborative devices A to C receive the above subtasks according to the subject of the subscribed subtask. Since collaborative devices A and C have not yet deployed a model capable of answer generation, the answer generation subtask is not executed. Collaborative device B executes the above answer generation subtask based on the deployed language processing model M and obtains the question answer D1 corresponding to the input question Q1.

[0100] In some embodiments, the corresponding relationship of the allocation of the above-mentioned multiple processing links includes one or more of the following:

[0101] A processing link is assigned to an execution subject, and the execution subject is one or more of the target device or the at least one collaborative device; or,

[0102] A processing link is assigned to an execution subject, and the execution subject is one of the target device or the at least one collaborative device; or,

[0103] One processing link is assigned to two execution subjects and the two execution subjects have a progressive execution relationship, a parallel execution relationship or a cross execution relationship when jointly completing the above processing link; the above two execution subjects are the above target device and the above collaborative device.

[0104] In the embodiments of the present disclosure, the progressive execution relationship is that one execution subject is executed first and triggers the execution of another execution subject when the execution result meets a certain condition. For example, the target device is deployed with a lightweight knowledge base: a fault identification knowledge base. In the processing link of retrieving relevant content for user problems, the target device first retrieves the relevant content and identifies the relevance. If the relevance of the relevant content retrieved in the lightweight knowledge base is lower than the preset value, the relevant content will continue to be retrieved from the collaborative device.

[0105] In the embodiments of the present disclosure, the parallel execution relationship is that two execution entities execute in parallel. For example, the target device is deployed with a lightweight knowledge base: a problem classification database. The problem classification database adopts a multi-level classification system. When searching based on the multi-level classification, the search is performed step by step from a hierarchical and gradually refined perspective. The granularity of the first-level search is coarser, and the second-level search is to continue to refine the search under the branch of the first-level search. By analogy, a total of two or more levels of search catalogs can be included, corresponding to the most refined problem classification. As an example: the first-level classification includes: vehicles, drones, airplanes, and ships. The second-level classification corresponding to vehicles includes: cars, bicycles, and motorcycles; the third-level classification corresponding to cars includes: sedans, SUVs, MPVs, sports cars, vans, etc. Collaborative device A is deployed with a paper knowledge base. In the processing link of classifying user questions, the above processing links are executed in parallel by the target device and collaborative device A respectively.

[0106] In the embodiments of the present disclosure, the cross-execution relationship may include at least one of mutual execution between multiple devices of the same execution subject and mutual execution between multiple devices of different execution subjects; for example, the target device is deployed with a lightweight knowledge base: an answer verification knowledge base; and the collaborative device B is deployed with a language processing model (for example, a large language model). In the processing link of answer generation, the target device A first sends the relevant content retrieved from the knowledge base to the collaborative device B, and then the collaborative device B obtains a generated answer based on the language processing model, sends the generated answer to the target device, and then the target device verifies the generated answer based on the answer verification knowledge base, and outputs it as the answer to the question after the verification passes.

[0107] In some scenarios, the collaborative device may output the answer to the question. In other scenarios, the collaborative device may synchronize the obtained answer to the question with the target device, and the target device may output the answer to the question. The output form of the answer to the question includes but is not limited to one of the following forms: text form, voice form.

[0108] Figure 3 It is a schematic diagram of an application scenario of a method for multi-device distributed natural language processing according to another exemplary embodiment.

[0109] exist Figure 3 In FIG. 1 , a processing flow is drawn from the perspective of processing an input question. The corresponding execution subject can be understood according to the description of the previous embodiment, which respectively illustrates the first-level indexing process (a static search process) of searching for problem classification matching based on the deployed classification knowledge base KZ in the target device, the second-level indexing process of searching for relevant content matching in the knowledge base corresponding to each collaborative device based on the problem category after problem classification, and the third-level indexing process of classifying and matching the found relevant content with the problem category. Figure 3The classification matching process of the classifier is used as an illustration. If the relevant content found matches the question category, positive feedback is output; otherwise, negative feedback is input, so that the top N (most relevant) relevant content matching the question category of the input question can be output.

[0110] As an example, the secondary indexing process and the tertiary indexing process can be executed in a collaborative device deployed with the corresponding knowledge base (the specific execution subject can be changed or adjusted according to the previous description of the deployment information). Since the data in the knowledge base has confidentiality requirements, these knowledge bases (for example, the knowledge bases containing technical document 1, technical document 2, and technical document 3 all need to be kept confidential) are deployed in designated devices (knowledge base 1 corresponding to technical document 1 is deployed on collaborative device A, knowledge base 2 corresponding to technical document 2 is deployed on collaborative device B, and knowledge base 3 corresponding to the technical document is deployed on collaborative device C) and external devices (for example, target devices) cannot directly access the knowledge base. The designated device serves as the execution subject of the collaborative device to search for relevant content corresponding to the input question. The retrieved relevant content is processed through a preset encryption algorithm and then synchronized to the external device, for example, to the target device T. After the target device T decrypts the encrypted topN related contents, it generates a prompt word based on the topN related contents. The prompt word includes the input question and the topN related contents, or the prompt word includes the input question and the precise related contents after the topN related contents are screened and enhanced; the prompt word is input into the generation model (as an example of the language processing model, the generation model can be LLM, GPT, BERT, etc.) deployed on the target device (here the target device is used as an example of the deployment of the language processing model. The language processing model can be deployed on other execution entities, which can be understood by reference) and the answer to the input question is output. For example, for the input question Q of the example: How to optimize the jitter performance of DDS during the development process. This problem involves many confidential files or R&D data within the enterprise. These internal data have data privacy requirements and cannot be publicly accessed or external access is restricted. The restricted access methods include: only some data can be viewed, users or devices within a specific range can view, and encrypted transmission is used for restriction. By decentralizing the privacy of processing power, knowledge bases with confidentiality or privacy requirements can be deployed on designated devices, and these designated devices can act as collaborative devices to bear part of the computing power, ensuring that data will not be leaked while generating answers to questions. Figure 3 As shown, the output answer to question A is as follows: Solution 1: xxxx, Solution 2: yyyyy.

[0111] In an embodiment including steps S110 to S130, multiple processing links for processing an input question to obtain an answer to the question and the relationship between each processing link are determined; the multiple processing links are allocated to the target device and at least one collaborative device, and the target device and the collaborative device are made to distribute and collaboratively execute the multiple processing links according to the relationship to obtain an answer to the question, thereby achieving the dispersion of processing power and the distributed and collaborative execution of processing tasks by multiple devices. On the one hand, multiple processing links are distributed in the target device and the collaborative device for distributed processing, which can reduce the computing power consumption of the target device, ensure that more computing power is used to execute core tasks, and can be used without affecting the execution of core tasks. Perform input problem processing (for example, the priority of processing tasks for input problems is lower than that of core tasks) to avoid problems in related technologies such as failure to successfully execute language processing tasks or low processing efficiency due to limited computing power allocation, or adverse effects on other main tasks due to the large amount of computing power occupied by language processing tasks; on the other hand, for scenarios where some knowledge bases have data confidentiality requirements, these knowledge bases can only be deployed on specific devices; by dispersing processing computing power, knowledge bases with confidentiality or privacy requirements can be deployed on designated devices, and these designated devices can assume part of the computing power as collaborative devices, thereby ensuring that private data will not be leaked while generating answers to questions.

[0112] In some embodiments, in the process of the target device and the collaborative device distributing and collaboratively executing the multiple processing steps according to the association relationship, the processing step of retrieving relevant content for the user question includes:

[0113] Relevant content is retrieved for the user question from a knowledge base deployed by a first execution subject to obtain relevant content corresponding to the user question; wherein the first execution subject is the target device or the collaborative device.

[0114] The above knowledge base satisfies at least one of the following conditions:

[0115] The computing power consumption required for searching the knowledge base, or the computing power consumption required for storage and retrieval, respectively matches the computing power resources available to the first execution subject; or,

[0116] The privacy of the stored data of the above-mentioned knowledge base requires that only the above-mentioned first execution subject has access rights to the above-mentioned knowledge base, and other external devices do not have direct access rights and the content indirectly accessed through the above-mentioned first execution subject is restricted.

[0117] In some embodiments, in the process of the target device and the collaborative device distributing and collaboratively executing the multiple processing steps according to the association relationship, the processing step of generating an answer based on the retrieved relevant content and the user question includes:

[0118] The retrieved relevant content and the above-mentioned user question are input into the language processing model deployed by the second execution entity, and the answer to the above-mentioned user question is output after referring to the above-mentioned relevant content; wherein the above-mentioned second execution entity is the above-mentioned target device or the above-mentioned collaborative device.

[0119] The above language processing model satisfies at least one of the following conditions:

[0120] The computing power consumption required for answer generation, or the computing power consumption required for deployment and answer generation of the above-mentioned language processing model matches the computing power resources available to the above-mentioned second execution subject; or,

[0121] The privacy of the stored data of the above-mentioned language processing model requires that only the above-mentioned second execution subject has access rights to the above-mentioned language processing model, and other external devices do not have direct access rights and the content indirectly accessed through the above-mentioned second execution subject is restricted.

[0122] In some embodiments, the target device and the collaborative device communicate based on the data distribution service protocol (DDS), and the target device, as a publisher, publishes the processing links shared by the collaborative device as sub-tasks, and the collaborative device, as a subscriber, receives the corresponding sub-tasks and feeds back the processing results of the sub-tasks to the target device.

[0123] In the system composed of target devices and collaborative devices, communication is carried out between the target devices and the collaborative devices based on the DDS protocol. Based on the subscription mechanism, the target device as the publisher only needs to publish subtasks, and does not need to establish a one-to-one session link with one or more collaborative devices in the system. While the target device is processing its own core tasks and the corresponding assigned subtasks, other collaborative devices in the system can receive and execute subtasks based on the DDS communication mechanism according to their own subscription content (for example, subscribing to the subtask topic, the specific subtask content is not required, if a collaborative device subscribes to a subtask that is not assigned to itself, the corresponding subtask does not need to be executed). The communication is safe and reliable, and network resources are saved. If a new collaborative device needs to be added to the system, it only needs to subscribe the new collaborative device to the subtask topic, which is flexible to expand.

[0124] In some embodiments, the prerequisite of the above method may also include: the step of pre-building a knowledge base. For example, the original file carrying knowledge, question-answer pairs or related information is processed and converted into data that can be recognized by the database. For example, for document files, the original document can be processed by segmenting the original document, converting it into a word vector and storing it in a database. For image files, OCR (optical character recognition) recognition can be used to convert it into text information, segment the text, convert it into a word vector and store it in a database. Other methods will not be described here, and any method that can convert an original file into a knowledge base is within the protection scope of this disclosure.

[0125] It should be noted that the relevant content corresponding to the input question mentioned in the embodiments of the present disclosure refers to the content related to the user's question. These contents can be questions, knowledge or answers related to the user's question, etc. that can prompt more information, so as to facilitate the subsequent answer generation process of the language processing model (which can be a generative model) to provide more relevant information and more dimensional reference information, which helps to improve the accuracy and richness of the generated answers to the questions.

[0126] Exemplary Systems

[0127] Figure 4 The invention is a block diagram of a system for distributed natural language processing on multiple devices according to an exemplary embodiment.

[0128] Reference Figure 4 As shown, the multi-device distributed natural language processing system 400 includes: a target device 410 and at least one collaborative device 420.

[0129] The first question receiving module 411 is disposed in the target device 410 and is used to receive an input question.

[0130] The first processing link determination module 412 is provided in the target device or the collaborative device, and is used to determine multiple processing links for processing the input question to obtain the answer to the question and the relationship between the processing links. Figure 4 The first processing step determining module 412 is indicated by a dotted box. In this embodiment, the first processing step determining module 412 is set in the target device 410 as an example.

[0131] The first task allocation module 413 is provided in the target device or the collaborative device, and is used to allocate the plurality of processing steps to the target device and the at least one collaborative device. Figure 4The first task assignment module 413 is indicated by a dotted box in FIG. 4 . In this embodiment, the first task assignment module 413 is set in the target device 410 as an example. In other embodiments, the first processing link determination module and the first task assignment module can also be set in the collaborative device 420.

[0132] The first task execution module 414 is disposed in the target device 410 and is used to execute the corresponding assigned processing link.

[0133] The second task execution module 421 is arranged in the above-mentioned collaborative device 420 and is used to execute the corresponding assigned processing link.

[0134] Among them, the above-mentioned first task execution module 414 and the above-mentioned second task execution module 421 are used to distribute and coordinately execute the above-mentioned multiple processing links according to the above-mentioned association relationship to obtain the answer to the question.

[0135] The execution details, interaction details, etc. of the target device 410 and the collaborative device 410 in this system can refer to the relevant description of the first embodiment and will not be repeated here.

[0136] This system distributes the processing power and the processing tasks among multiple devices for distributed and collaborative execution. On the one hand, multiple processing links are distributed among the target device and the collaborative device for distributed processing, which can reduce the computing power consumption of the target device, ensure that the target device has more computing power to execute the core tasks, and can process the input questions without affecting the execution of the core tasks, thereby avoiding the problems in related technologies such as the inability to successfully execute the language processing tasks or low processing efficiency due to limited computing power allocation, or the adverse effects on other major tasks due to the large computing power occupied by the language processing tasks; on the other hand, for some scenarios where knowledge bases have data confidentiality requirements, these knowledge bases can only be deployed on specific devices; by dispersing the processing power, knowledge bases with confidentiality or privacy requirements can be deployed on designated devices, and these designated devices can bear part of the computing power as collaborative devices, thereby ensuring that private data will not be leaked while realizing the generation of answers to questions.

[0137] In the system 400 of this embodiment, data communication can be carried out between the target device 410 and one or more collaborative devices 420 based on various communication methods such as: 3G, 4G, 5G, short-range communication protocols (such as Bluetooth, Zigbee, etc.), local Internet WiFi, DDS communication protocols, etc.

[0138] In this embodiment, the target device and the collaborative device communicate based on the DDS protocol. The target device, as a publisher, publishes the processing links shared by the collaborative device as sub-tasks, and the collaborative device, as a subscriber, receives the corresponding sub-tasks and feeds back the processing results of the sub-tasks to the target device.

[0139] In the system composed of target devices and collaborative devices, communication is carried out between the target devices and the collaborative devices based on the DDS protocol. Based on the subscription mechanism, the target device as the publisher only needs to publish subtasks, and does not need to establish a one-to-one session link with one or more collaborative devices in the system. While the target device is processing its own core tasks and the corresponding assigned subtasks, other collaborative devices in the system can receive and execute subtasks based on the DDS communication mechanism according to their own subscription content (for example, subscribing to the subtask topic, the specific subtask content is not required, if a collaborative device subscribes to a subtask that is not assigned to itself, the corresponding subtask does not need to be executed). The communication is safe and reliable, and network resources are saved. If a new collaborative device needs to be added to the system, it only needs to subscribe the new collaborative device to the subtask topic, which is flexible to expand.

[0140] Example Vehicles

[0141] Figure 5 is a block diagram of a vehicle according to an exemplary embodiment.

[0142] The vehicle 500 may be a fuel vehicle, a hybrid vehicle, an electric vehicle, a fuel cell vehicle, or other types of vehicles, such as a vehicle that supports automatic driving or assisted driving.

[0143] The vehicle 500 has an on-board device 510 and at least one terminal device 520 .

[0144] The second question receiving module 501 is provided in the vehicle-mounted device 510 and is used to receive input questions.

[0145] The second processing link determination module 502 is arranged in the above-mentioned vehicle-mounted device or terminal device, and is used to determine multiple processing links for processing the above-mentioned input question to obtain the answer to the question and the correlation relationship between each processing link.

[0146] The second task allocation module 503 is arranged in the vehicle-mounted device or the terminal device, and is used to allocate the multiple processing links to the vehicle-mounted device and the at least one terminal device.

[0147] exist Figure 5The second processing link determination module 502 and the second task assignment module 503 are respectively indicated by dashed boxes. In this embodiment, the second processing link determination module 502 and the second task assignment module 503 are set in the vehicle-mounted device 510 as an example. In other embodiments, the second processing link determination module and the second task assignment module can also be set in one terminal device 510 or dispersedly set in multiple terminal devices 520.

[0148] The third task execution module 504 is arranged in the vehicle-mounted device 510 and is used to execute the corresponding assigned processing link.

[0149] The fourth task execution module 505 is provided in the terminal device 520 and is used to execute the corresponding assigned processing steps. The third task execution module 504 and the fourth task execution module 505 are used to execute the above multiple processing steps in a distributed and coordinated manner according to the above association relationship to obtain the answer to the question.

[0150] The above-mentioned terminal devices may be, but are not limited to: mobile phones, tablet computers, smart bracelets, smart watches, etc.

[0151] In some implementation scenarios, a mobile phone is used as an example of a terminal device 520, and different types of knowledge bases are pre-set and initialized in different mobile phones. The mobile phone broadcasts the current knowledge base classification. Figure 3 As shown in the figure, distributed multi-level index matching is performed on the vehicle side and the mobile phone side: First-level indexing: After receiving the input question on the vehicle side, the vehicle side performs question classification matching and search, obtains the question category, and converts it into the corresponding word embedding vector. After that, the mobile phone side performs secondary and tertiary indexing based on its own knowledge base, retrieves the topN related content that matches the input question and the corresponding question category, encrypts it and returns it to the vehicle side. The vehicle side generates prompt words based on the topN related content and inputs them into the large language model for answer generation.

[0152] In this embodiment, in the vehicle driving scenario, by dispersing the processing power, and distributing the processing tasks for generating answers to input questions to the on-board devices and terminal devices in the vehicle for distributed and collaborative execution, the storage space occupied by the vehicle system in the on-board devices and the consumption of computing power on the vehicle side are reduced. In addition, the distributed execution of the computing power execution subject helps to improve the security of privacy data, ensure that privacy data is stored in the knowledge base corresponding to a specific device, and access to the knowledge base is restricted. Retrieval operations are only performed by specific devices (authorized devices, such as terminal devices acting as collaborative device roles). The retrieval results are fragmented and segmented results, and other devices (for example, the vehicle side can only obtain fragmented results) will not leak privacy data.

[0153] For more details and other beneficial effects of this embodiment, reference may be made to the related descriptions of the first and second embodiments, which will not be repeated here.

[0154] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the present disclosure. This application is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The specification and examples are to be considered as exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0155] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A method for multi-device distributed natural language processing, characterized in that: include: Receive input issues based on the target device; Determine multiple processing links for processing the input question to obtain an answer to the question and the correlation relationship between the processing links; The multiple processing steps are allocated to the target device and at least one collaborative device, and the target device and the collaborative device are enabled to execute the multiple processing steps in a distributed and collaborative manner according to the association relationship to obtain an answer to the question.

2. The method according to claim 1, characterized in that Allocating the multiple processing links to the target device and at least one collaborative device includes: According to the deployment information of the target device and the collaborative device, the multiple processing links are allocated to the target device and at least one collaborative device; wherein the deployment information includes information on the pre-deployment of the knowledge base or language processing model of the target device and the collaborative device.

3. The method according to claim 2, characterized in that The corresponding relationship of the allocation of the multiple processing links includes one or more of the following: A processing link is assigned to an execution subject, and the execution subject is one or more of the target device or the at least one collaborative device; or, A processing link is assigned to an execution subject, and the execution subject is one of the target device or the at least one collaborative device; or, One processing link is allocated to two execution entities, and the two execution entities have a progressive execution relationship, a parallel execution relationship or a cross execution relationship when jointly completing the processing link; the two execution entities are the target device and the collaborative device.

4. The method according to claim 1, characterized in that: The processing steps include: searching for relevant content for the user question; Among them, in the process that the target device and the collaborative device distribute and collaboratively execute the multiple processing links according to the association relationship, the processing link of retrieving relevant content for the user question is executed, including: retrieving relevant content for the user question from a knowledge base deployed by a first execution subject, and obtaining relevant content corresponding to the user question; wherein the first execution subject is the target device or the collaborative device; The knowledge base satisfies at least one of the following conditions: The computing power consumption required for searching the knowledge base, or the computing power consumption required for storage and retrieval, respectively matches the computing power resources available to the first execution entity; or, The privacy of the stored data of the knowledge base requires that only the first execution subject has access rights to the knowledge base, and other external devices do not have direct access rights and the content indirectly accessed through the first execution subject is restricted.

5. The method according to claim 4, characterized in that The processing step also includes: generating answers based on the retrieved relevant content and the user question; Wherein, in the process of the target device and the collaborative device distributing and collaboratively executing the multiple processing links according to the association relationship, a processing link for generating an answer based on the retrieved relevant content and the user question is executed, including: inputting the retrieved relevant content and the user question into a language processing model deployed by a second execution subject, and outputting an answer to the user question after referring to the relevant content; wherein the second execution subject is the target device or the collaborative device; The language processing model satisfies at least one of the following conditions: The computing power consumption required by the language processing model for answer generation, or the computing power consumption required for deployment and answer generation, respectively matches the computing power resources available to the second execution entity; or, The privacy of the stored data of the language processing model requires that only the second execution subject has access rights to the language processing model, and other external devices do not have direct access rights and the content indirectly accessed through the second execution subject is restricted.

6. The method according to claim 1, characterized in that The multiple processing steps include a combination of the following basic processing steps, or a combination of the following basic processing steps and one or more supplementary processing steps: The first basic processing link: search for relevant content for user questions; The second basic processing step: generating answers based on the retrieved relevant content and the user's question; The first supplementary processing link: classifying the user question to obtain a question category; wherein the retrieving relevant content from the user question includes: retrieving relevant content based on the user question and the question category; Second supplementary processing link: screening and enhancing the retrieved relevant content according to the relevance between the retrieved relevant content and the user question to obtain accurate relevant content; wherein generating an answer based on the retrieved relevant content and the user question includes: generating an answer based on the accurate relevant content and the user question; The third supplementary processing link: verify the generated answers to the questions and output them after they pass the verification.

7. The method according to any one of claims 1 to 6, characterized in that The target device and the collaborative device jointly share the processing link of the processing task of processing the user question to obtain the answer to the question, wherein the target device itself also undertakes the core task, and the importance of the processing task is lower than the importance of the core task; wherein the target device and the collaborative device are one of the following situations: The target device is an on-board device in a vehicle, and the collaborative device is a terminal device in the vehicle; or, The target device is an on-board device in a vehicle, and the collaborative device is a cloud device; or, The target device is a first device in a target space, and the collaborative device is one or more second devices in the target space; or, The target device is a target drone in a drone group, and the coordinated device is one or more drones other than the target drone in the drone group; The target device is a robot that supports question answering, and the collaborative device is an electronic device deployed with a knowledge base or a language processing model; or, The target device is a warehousing and logistics device that supports question answering, and the collaborative device is an electronic device deployed with a knowledge base or a language processing model; or, The target device is an exploration device that supports question and answer, and the collaborative device is an electronic device deployed with a knowledge base or a language processing model.

8. The method according to any one of claims 1 to 6, characterized in that The target device and the collaborative device communicate based on the data distribution service protocol. The target device, as a publisher, publishes the processing links shared by the collaborative device as sub-tasks, and the collaborative device, as a subscriber, receives the corresponding sub-tasks and feeds back the processing results of the sub-tasks to the target device.

9. A multi-device distributed natural language processing system, characterized in that: include: a target device and at least one cooperating device; A first question receiving module, provided in the target device, for receiving an input question; A first processing link determination module, which is disposed in the target device or the collaborative device, and is used to determine a plurality of processing links for processing the input question to obtain an answer to the question and the correlation relationship between the processing links; A first task allocation module, arranged in the target device or the collaborative device, for allocating the plurality of processing links to the target device and the at least one collaborative device; A first task execution module, arranged in the target device, for executing a corresponding assigned processing link; The second task execution module is arranged in the collaborative device and is used to execute the corresponding assigned processing links; wherein, the first task execution module and the second task execution module are used to distribute and collaboratively execute the multiple processing links according to the association relationship to obtain the answer to the question.

10. A vehicle, characterized in that: The vehicle has an on-board device and at least one terminal device; A second question receiving module, provided in the vehicle-mounted device, for receiving an input question; A second processing link determination module is provided in the vehicle-mounted device or the terminal device, and is used to determine a plurality of processing links for processing the input question to obtain an answer to the question and the correlation relationship between the processing links; A second task allocation module, provided in the vehicle-mounted device or the terminal device, for allocating the plurality of processing links to the vehicle-mounted device and the at least one terminal device; A third task execution module is provided in the vehicle-mounted device and is used to execute the corresponding assigned processing link; The fourth task execution module is arranged in the terminal device and is used to execute the corresponding assigned processing links; wherein, the third task execution module and the fourth task execution module are used to distribute and coordinately execute the multiple processing links according to the association relationship to obtain the answer to the question.

Images