Data processing method and device and electronic equipment

Abstract

The invention provides a data processing method. The method comprises the steps of obtaining target input data; respectively performing first processing and second processing on the target input data to obtain first input data and second input data in response to the condition that the target input data meets the target condition; the information content of the first input data is different from that of the second input data; at least inputting the first input data into the first model; at least inputting the second input data into a second model; and based on the output of the first model and the second model, obtaining target output data responding to the target input data.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and more particularly to a data processing method, apparatus, device, medium, and program product. Background Art

[0002] In the current rapidly evolving technological environment, the demand for data processing and intelligent applications in intelligent devices is increasing day by day. Limited by the performance of the device itself and specific application scenarios, different types of models each have their own advantages and limitations, and it is difficult to meet all processing requirements alone. In order to improve the intelligence of interaction and the accuracy of response, more and more systems are starting to introduce multi-model collaborative architectures.

[0003] However, in existing multi-model architectures, the same data is often input into multiple models in a unified input manner for parallel processing, ignoring the differences in processing capabilities, reasoning mechanisms, and input adaptation among different models. Summary of the Invention

[0004] In view of the above problems, the present disclosure provides a data processing method, apparatus, device, medium, and program product.

[0005] According to a first aspect of the present disclosure, there is provided a data processing method, including: obtaining target input data; in response to the target input data satisfying a target condition, respectively performing a first process and a second process on the target input data to obtain first input data and second input data; the information content of the first input data is different from that of the second input data; inputting at least the first input data into a first model; inputting at least the second input data into a second model; and obtaining target output data in response to the target input data based on the outputs of the first model and the second model.

[0006] According to an embodiment of the present disclosure, obtaining target output data in response to the target input data based on the outputs of the first model and the second model includes: inputting at least first output data of the first model into the second model; obtaining target output data in response to the target input data according to second output data of the second model; or obtaining target output data in response to the target input data according to third output data of the first model and fourth output data of the second model.

[0007] According to an embodiment of the present disclosure, the first processing and the second processing of the target input data to obtain the first input data and the second input data include: inputting the target input data into a target intelligent engine, so that the target intelligent engine performs the first processing and the second processing on the target input data to output the first input data and the second input data; or respectively inputting the target input data into a first intelligent engine and a second intelligent engine, so that the first intelligent engine performs the first processing on the target input data to output the first input data, and the second intelligent engine performs the second processing on the target input data to output the second input data.

[0008] According to an embodiment of the present disclosure, the first processing is used to remove the target semantic expression containing individual features in the target input data; the second processing is used to expand the data content included in the target input data.

[0009] According to an embodiment of the present disclosure, the first processing is further used to expand the data content included in the target input data based on a target database; wherein the target database is used to store privatized data.

[0010] According to an embodiment of the present disclosure, the second processing is used to expand the data content included in the target input data based on the target database; wherein the data content expanded by the first processing based on the target database is abstract content; the data content expanded by the second processing based on the target database is specific content.

[0011] According to an embodiment of the present disclosure, in response to the target input data not meeting the target condition, the target input data is input into the second model to obtain the target output data in response to the target input data.

[0012] According to an embodiment of the present disclosure, the first model includes a cloud intelligent model, and the second model includes a terminal intelligent model.

[0013] A second aspect of the present disclosure provides a data processing system, including: an input data acquisition module, configured to acquire target input data; a processing module, configured to, in response to the target input data meeting a target condition, perform first processing and second processing on the target input data respectively to obtain first input data and second input data; the information content of the first input data is different from that of the second input data; a first input module, configured to input at least the first input data into a first model; a second input module, configured to input at least the second input data into a second model; and a response acquisition module, configured to obtain the target output data in response to the target input data based on the outputs of the first model and the second model.

[0014] A third aspect of the present disclosure provides an electronic device, including: a first application program running on the electronic device; the first application program is configured to: parse at least one task based on an input, call a target model to execute at least one task, at least for: receiving target input data; in response to the target input data meeting a target condition, performing first processing and second processing on the target input data respectively to obtain first input data and second input data; the information content of the first input data is different from that of the second input data; inputting at least the first input data into a first model; inputting at least the second input data into a second model; obtaining target output data in response to the target input data based on the outputs of the first model and the second model; the target model includes the first model and / or the second model.

[0015] A fourth aspect of the present disclosure further provides a computer program product, including a computer program or instruction, and when the computer program or instruction is executed by a processor, the steps of the above method are implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Through the following description of the embodiments of the present disclosure with reference to the drawings, the above content and other objects, features and advantages of the present disclosure will become clearer. In the drawings:

[0017] Figure 1 Schematically shows an application scenario diagram of a data processing method, device, equipment, medium and program product according to an embodiment of the present disclosure;

[0018] Figure 2 Schematically shows a flowchart of a data processing method according to an embodiment of the present disclosure;

[0019] Figure 3A Schematically shows a flowchart of a data processing method according to some embodiments of the present disclosure;

[0020] Figure 3B Schematically shows a flowchart of a data processing method according to other embodiments of the present disclosure;

[0021] Figure 4 Schematically shows a structural framework diagram of a data processing system based on a cloud model and a local model according to an embodiment of the present disclosure;

[0022] Figure 5 Schematically shows a structural block diagram of a data processing system according to an embodiment of the present disclosure; and

[0023] Figure 6 Schematically shows a block diagram of an electronic device suitable for implementing a data processing method according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the following detailed description, for the sake of explanation, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the present disclosure. However, it is obvious that one or more embodiments can be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts of the present disclosure.

[0025] The terms used herein are merely for describing specific embodiments and are not intended to limit the present disclosure. The terms "including", "comprising", etc. used herein indicate the presence of the described features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0026] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those of ordinary skill in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0027] In cases where expressions similar to "at least one of A, B, and C, etc." are used, generally, it should be interpreted according to the meaning commonly understood by those of ordinary skill in the art (for example, "a system having at least one of A, B, and C" should include, but not be limited to, a system having only A, only B, only C, having A and B, having A and C, having B and C, and / or having A, B, and C).

[0028] Some block diagrams and / or flowcharts are shown in the accompanying drawings. It should be understood that some of the blocks or combinations thereof in the block diagrams and / or flowcharts can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when executed by the processor, these instructions can create a device for implementing the functions / operations illustrated in these block diagrams and / or flowcharts.

[0029] Accordingly, the techniques of the present disclosure may be implemented in the form of hardware and / or software (including firmware, microcode, etc.). Additionally, the techniques of the present disclosure may take the form of a computer program product on a computer-readable medium storing instructions, which is available for use by or in conjunction with an instruction execution system. In the context of the present disclosure, a computer-readable medium may be any medium that can contain, store, transmit, propagate, or transport instructions. For example, a computer-readable medium may include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, components, or propagation media. Specific examples of computer-readable media include: magnetic storage devices such as magnetic tapes or hard disk drives (HDDs); optical storage devices such as compact discs (CD-ROMs); memories such as random access memories (RAMs) or flash memories; and / or wired / wireless communication links.

[0030] Embodiments of the present disclosure provide a data processing method, the method including: obtaining target input data; in response to the target input data satisfying a target condition, respectively performing first processing and second processing on the target input data to obtain first input data and second input data; information content of the first input data being different from that of the second input data; inputting at least the first input data into a first model; inputting at least the second input data into a second model; and obtaining target output data in response to the target input data based on outputs of the first model and the second model. By performing different processing on the target input data, first input data and second input data with different information contents are obtained, so that different models can learn according to different features, improving the adaptability to the input data. At the same time, by respectively processing different input data with at least two different models and obtaining the final target output data based on their outputs, different advantages of multiple models can be utilized to make the result more reliable and accurate.

[0031] Figure 1 Schematically shows an application scenario diagram of a data processing method, apparatus, device, medium, and program product according to an embodiment of the present disclosure.

[0032] As Figure 1 shown, an application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 is used to provide a medium for communication links between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0033] Users can interact with the server 105 through the network 104 using the first terminal device 101, the second terminal device 102, and the third terminal device 103 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (for example only).

[0034] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with a display screen and supporting web browsing, including but not limited to smartphones, tablets, laptop computers, desktop computers, and so on.

[0035] The server 105 can be a server providing various services, such as a background management server (for example only) that supports the websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server can analyze and process data such as received user requests, and feedback the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device. The server 105 can include a first server and a second server.

[0036] It should be noted that the data processing method provided by the embodiments of the present disclosure can generally be executed by the server 105. Correspondingly, the processing device for interest information provided by the embodiments of the present disclosure can generally be set in the server 105. The data processing method provided by the embodiments of the present disclosure can also be executed by a server or a server cluster different from the server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or the server 105. Correspondingly, the processing device for interest information provided by the embodiments of the present disclosure can also be set in a server or a server cluster different from the server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or the server 105.

[0037] It should be understood that Figure 1 the numbers of terminal devices, networks, and servers in

[0038] are merely illustrative. According to the implementation requirements, there can be any number of terminal devices, networks, and servers. Figure 1 The following will be based on Figures 2 to 3B the described scenario and will describe in detail the data processing method of the disclosed embodiments through

[0039] Figure 2 A flowchart of the data processing method according to the embodiments of the present disclosure is schematically shown.

[0040] As shown Figure 2 in the figure, the data processing method of this embodiment includes operations S210 to S250.

[0041] In operation S210, target input data is obtained.

[0042] In an embodiment of the present disclosure, the target input data may include original interaction information and an initial prompt initiated by a user through a mobile terminal. For example, it may include natural language texts such as voice-to-text instructions, manually input query statements, etc.; multimedia data such as images, videos, sensor signals, etc.; and context metadata such as timestamps, geographical locations, device status, etc. The target input data can be captured in real time through an input interface of the terminal device.

[0043] In operation S220, in response to the target input data satisfying a target condition, the target input data is respectively subjected to a first process and a second process to obtain first input data and second input data; the information content of the first input data is different from that of the second input data. Exemplarily, the data dimensions, data structures, abstraction levels, or semantic expressions included in the first input data and the second input data are different.

[0044] In an embodiment of the present disclosure, when the target input data satisfies a preset target condition, the system can initiate the first process and the second process. The setting of the target condition can be flexibly set based on factors such as the complexity of the task, response real-time performance, personalization degree, privacy sensitivity, etc.

[0045] Exemplarily, in a complex task scenario, the input data needs to be collaboratively processed by two models with complementary capabilities simultaneously.

[0046] Again, for example, in some embodiments, even if the target input data contains sensitive information, if the user has authorized external processing after desensitization, the system can also determine that it meets the preset conditions, and then enter the dual-processing path execution stage.

[0047] In an embodiment of the present disclosure, the first process and the second process can be embodied as a variety of different specific operations in practical applications, and are differentially processed according to the characteristics of the target input data and the differences in the capabilities of the models used to generate first input data and second input data with different information contents.

[0048] For example, in a multi-modal fusion application scenario, if the target input data includes an image and its accompanying natural language description, the first processing can extract visual elements suitable for an image recognition model, such as basic pixel structures like the edge features, texture information, and color block distribution of the image, and generate first input data adapted to the first model (such as a convolutional neural network) through image preprocessing (such as resolution unification, color space conversion, ROI extraction, etc.); while in the second processing, semantic analysis and structuring can be performed on the natural language part, such as extracting task descriptions, label pointers, or intent categories therein, and combining context information associated with the image (such as the user instruction "select the most suitable picture for the meeting cover") to generate second input data adapted to the second model (such as a Transformer model) for semantic understanding, target generation, or decision guidance.

[0049] For another example, in a device control task, the user can input via voice "Turn up the light and then also open the curtains in the living room". The first processing can perform standardized grammar parsing and structure extraction on this voice instruction to generate an instruction structure containing the operation object and action type, for example:

Device: light, Operation: turn up

Device: curtains, Operation: open

[0050] For another example, in a user recommendation system, the first processing can perform generalization feature modeling on the user input "What movies are recommended", including extracting general query elements such as time period and platform type, and constructing a feature template in combination with the full set of behavior data labels (such as "recently popular", "high score"); while the second processing can further extract personalized features from the user's historical behavior records, such as past viewing preferences, viewing frequency, social friend recommendation records, etc., and construct a relationship graph or temporal vector between the user and the content for in-depth recommendation optimization.

[0051] In operation S230, at least input the first input data into the first model.

[0052] In operation S240, at least input the second input data into the second model.

[0053] In the embodiments of the present disclosure, the processing capabilities of the first model and the second model are different. The differences between the first model and the second model can be reflected in multiple dimensions, including model scale, computational depth, knowledge coverage, response real-time performance, input sensitivity, reasoning method, etc. Through the differential processing of the target input data, the system can send different types of data into the most suitable model, give full play to their respective advantages, and improve the intelligence and adaptability of the overall response.

[0054] For example, in a multi-modal fusion application scenario, the first model can be a convolutional neural network dedicated to image content analysis, used to identify the basic structure and object types in the image; the second model can be a multi-layer Transformer model dedicated to semantic annotation and task interpretation, used to process natural language descriptions, user task intents, or label generation related to the image.

[0055] For another example, in a device control task, the first model can be a lightweight semantic parsing module, which is good at extracting the syntax of structured instructions; the second model can be a context understanding module, which focuses on identifying ellipses, context jumps, or reference relationships in continuous conversations.

[0056] For yet another example, in a user recommendation system, the first model can be a collaborative filtering-based interest matching model, which focuses on analyzing large-scale user behavior data and capturing common content preference patterns; the second model can be a relationship understanding model constructed by a graph neural network, which is good at processing complex graph structures such as social relationships, emotional connections, and interaction frequencies between users and content.

[0057] In operation S250, based on the outputs of the first model and the second model, target output data in response to the target input data is obtained.

[0058] In the embodiments of the present disclosure, this operation can flexibly combine the output results of different models, thereby improving the response quality and task adaptation ability of the system. Specifically, the system can integrate the outputs of the two models in various ways such as result weighting, content splicing, strategy selection, rule driving, attention fusion, modality collaboration, and context reconstruction.

[0059] In one embodiment, the first model and / or the second model can be a machine learning model that can identify natural language and / or other inputs (such as audio-visual, images, tables, etc.) input to the target model, and perform comprehensive language processing tasks such as semantic analysis and question answering, and then generate outputs related to the input and / or responses to the input.

[0060] This embodiment does not limit the form of the first model and / or the second model. The preset model can be a generative model or a generative language model (GLMs). For example, it can specifically include large language models (LLMs), GPT (Generative Pre-trained Transformer), etc. The models involved in the embodiments of the present disclosure can be general large models or expert large models obtained by fine-tuning based on requirements. The embodiments of the present disclosure do not limit this. The first model and / or the second model generally learn the features and rules of natural language by training a large amount of diverse data, so as to be able to understand and generate natural language. Usually, it can have model parameters ranging from hundreds of millions to hundreds of billions (model parameters are variables that control the behavior of the target model), and can capture complex relationships and patterns in natural language.

[0061] Figure 3A Schematically shows a flowchart of a data processing method according to some embodiments of the present disclosure.

[0062] As Figure 3A shown, the data processing method of this embodiment includes operations S310 to S320. Among them, operations S310 to S320 can be subsequent operations of operations S210 to S240, reflecting the processing path of "cascaded reasoning".

[0063] In operation S310, at least the first output data of the first model is input into the second model.

[0064] Exemplarily, the first model undertakes the tasks of preliminary information extraction or semantic modeling in this process, generating basic feature vectors, intermediate prediction labels or coarse-grained decision results; the second model further mines context associations, supplements missing information or completes refined discrimination based on this output. This structure is suitable for application scenarios where the first model is a lightweight and fast preprocessing module and the second model is a high-precision postprocessing module. For example, in a multi-turn dialogue system, the first model can be used to quickly identify the intent category of the user's current turn, while the second model further combines the historical dialogue records to complete the context and generate consistent response statements on this basis.

[0065] In operation S320, according to the second output data of the second model, target output data in response to the target input data is obtained. In this process, the target output data can be regarded as the product of a two-stage calculation process of "preprocessing - optimization processing", and has stronger consistency and semantic integrity.

[0066] Figure 3BA flowchart of a data processing method according to some other embodiments of the present disclosure is schematically shown.

[0067] As Figure 3B shown, the data processing method of this embodiment includes operation S410, where operation S410 may be a subsequent operation of operations S210 to S240.

[0068] In operation S410, target output data in response to the target input data is obtained according to the third output data of the first model and the fourth output data of the second model.

[0069] Unlike Figure 3A that, Figure 3B the processing paths in

[0070] do not form a sequential dependency relationship. The two models perform tasks independently. After completing their respective inference processes, the system then fuses their results uniformly. This approach is suitable for task types where the models have different focus points, can work independently but ultimately need to be aggregated. For example, in a joint task of event recognition and sentiment understanding, the first model is used to identify the core elements of an event, and the second model is used to analyze the emotional color contained in the statement. The system concatenates the outputs of the two to form an event record with an emotional label for subsequent applications.

[0071] In some embodiments, the target input data may be input to a target intelligent engine, and the engine performs the first processing and the second processing in parallel. The target intelligent engine may be a unified model integrating multitasking capabilities.

[0072] In some other embodiments, the target input data may be respectively input to a first intelligent engine and a second intelligent engine to perform the first processing and the second processing respectively. The first intelligent engine and the second intelligent engine may have different model architectures, optimization objectives, or semantic understanding methods, and are respectively adapted to the computational requirements of different processing directions, such as tasks oriented to general semantic transformation, context enhancement, information compression, structure completion, etc. Such a decoupled structure is suitable for system architecture designs of heterogeneous model combinations, modular deployment, and on-demand expansion.

[0073] For example, the data annotation process can follow hierarchical desensitization rules: identify the first-person subject (such as "I", "my") through semantic parsing, replace it with a generic reference label (such as "user") or directly delete it; generalize keywords containing personal status or emotions (such as "tired", "anxious"), for example, reconstruct "I had insomnia last night and am very tired now" into "The user's current state requires advice on adjusting the schedule"; further remove specific timestamps and geographical location details, and generalize precise information into fuzzy descriptions; and use social relationship labels to replace real names and private associations to completely strip individual identities.

[0074] Also, for example, the data annotation process can follow semantic enhancement rules: for a concise input without a subject (such as "tired"), supplement the user reference and emotional state to generate a complete expression "I am very tired now and need advice on rest"; add specific constraints to a fuzzy instruction based on the spatio-temporal data in the local database, expand "go swimming" to "Go to the swimming pool within 1 kilometer of the user this afternoon"; convert the generalized description into a precise action by parsing the social relationship graph; and at the same time integrate device sensor and environmental data, for example, upgrade "navigate home" to "Plan the optimal route according to the current traffic congestion situation (estimated time of 40 minutes)" to achieve real-time scene adaptation.

[0075] In addition, in some embodiments, the target intelligent engine can also use technologies such as federated learning and differential privacy to ensure that user privacy is fully protected during the data processing process. For example, during the personalized model training process, a decentralized training method can be adopted to ensure that user data is not directly uploaded to a centralized server, but the model is updated on the local device, and only the encrypted model parameters are shared, so as to continuously optimize the effect of personalized services while ensuring data security.

[0076] The optional target intelligent engine, first intelligent engine, and second intelligent engine can be a machine learning model that can identify natural language and / or other inputs (such as audio-video, images, tables, etc.) input into the target model, and perform comprehensive language processing tasks such as semantic analysis and answering questions, and then generate an output related to the input and / or respond to the input.

[0077] In the embodiments of the present disclosure, the first processing can be used to perform de-individualization processing on the semantic content in the target input data, with the focus on identifying and stripping the individual feature semantic expressions contained therein. Individual features can include but are not limited to user identity information, emotional state, behavior preferences, time and location descriptions, social relationships, etc.

[0078] For example, in the scenario of natural language input, this processing can identify the first-person subject, specific personal names, geographical locations, timestamps, and personal-related expression components through semantic parsing technology, and then abstract, replace, or delete them to generate generalized and de-sensitized first input data. This processing helps to improve the adaptability of the data across users and scenarios, making it more suitable for model inference with generalization capabilities.

[0079] In the embodiments of the present disclosure, the second processing can be used to expand the data content included in the target input data, enhancing the integrity of its expression, the clarity of the context, and the richness of the structure. This processing is applicable to situations where the input data has incomplete semantics, vague expressions, or missing context. The expansion methods can include natural language reconstruction, prompt completion, behavioral goal reasoning, parameter filling, etc.

[0080] For example, when the user inputs "want to go swimming", the system can infer the user's possible time arrangements, location restrictions, and preference types through the second processing, and expand and generate complete input content, such as "want to go to the nearby indoor swimming pool this afternoon and need to finish before 19:00".

[0081] Furthermore, in some embodiments, the first processing not only includes stripping the individual characteristics in the target input data, but can also, on the basis of basic desensitization, supplement and expand the data content based on the target database. In this embodiment, the first processing can call the structured or semi-structured information in the target database to fill the gaps that may occur during the semantic abstraction process of the target input data, ensuring that even after removing sensitive content, the generated first input data still has a certain context integrity and is suitable for stable input to subsequent models.

[0082] The target database can be a PKB (Personal Knowledge Base), which is used to store private data related to the user, including but not limited to personalized information such as the user's preference settings, behavior records, location information, schedule arrangements, and environmental status.

[0083] During the first processing, the system can generate extended content at the abstract level based on the data stored in the target database. The abstract content can be desensitized and generalized user feature information, which is used to support upstream processing links such as task recognition, semantic analysis, and intention modeling while protecting user privacy. For example, when the user inputs "book a restaurant", the first processing can combine the preference type (such as "Japanese cuisine"), time preference (such as "dinner"), and frequently active area recorded in the PKB to generate an abstract expression: "Request to recommend a Japanese restaurant in XX area, suitable for dinner", which can be used for standardized analysis by a general model or remote service call, avoiding the exposure of specific detailed information.

[0084] Accordingly, the second processing can also be based on the same target database to further extract the data content of the specific layer for generating more targeted second input data in the personalized inference stage. The specific content can include clear geographical locations, time arrangements, user status, historical behavior trajectories, etc. Continuing with the example of "booking a restaurant", the second processing can combine information such as real-time location, specific time, and the list of restaurants collected by the user to generate a personalized extended expression: "Before 18:00 today, near XX Building, book the XX Japanese restaurant collected last week". This expression is applicable to models with local inference capabilities or personalized service capabilities, supporting higher-precision processing and recommendations.

[0085] In practical applications, the data levels involved in the first processing and the second processing can also be flexibly selected through system policies. For example, for upstream tasks that only need to identify user intentions, generalization modeling can rely only on the abstract layer data; while for tasks that require specific decision support or content output, specific layer data needs to be introduced to improve the output quality.

[0086] It should be noted that the abstract layer extension and the specific layer extension in this embodiment are not independent of each other, but can be combined according to the scenario. It is possible to judge when to perform which type of processing through a rule engine or model adaptation logic, and the results of the two types of data processing can be separately or jointly input into different models for collaborative inference. For example, the abstract layer data can be used for the standardized model to make a preliminary judgment, and the specific layer data can be provided to the personalized model for result refinement, so as to achieve a comprehensive response with clear structure, complete semantics, and accurate results.

[0087] In the embodiments of the present disclosure, if the system determines that the target input data does not meet the target conditions, it can also choose to call only one of the models for processing.

[0088] For example, when the task complexity is low, the input data structure is clear, or within the scope of tasks covered by existing rule templates, the entire response process can be directly completed by a single model without differential processing. Typically, query requests within the whitelist, daily control instructions, etc., have clear processing logics and limited required computing resources, and calling a single model can meet the requirements of real-time response and accuracy.

[0089] Another example is that in scenarios involving sensitive privacy information, if the user sets restrictions on the data access rights of specific models, the system can select to use only trusted models for data processing according to the policy to ensure that sensitive information will not be used for external analysis without permission, reducing the potential risk of privacy leakage.

[0090] For another example, in an environment with limited communication conditions, such as when the network connection is unstable or the system detects that the latency exceeds a preset threshold, the system can also dynamically adjust the model selection strategy and preferentially use a model with local computing capabilities to respond, so as to ensure service continuity and interaction immediacy.

[0091] For another example, in some scenarios that require extensive knowledge coverage, cross-domain reasoning, or high-precision processing, the system can preferentially select a model with a wide coverage of capabilities and a high processing depth to execute the main task, ensuring that the final response has comprehensive knowledge and semantic accuracy.

[0092] In an embodiment of the present disclosure, the first model may be a cloud model. A cloud model is an AI model deployed on a remote server, which receives user requests and data through the Internet, completes calculations in the cloud, and returns the results to the terminal device. The cloud model can utilize high-performance computing clusters, large memory, and distributed computing resources to support complex models, and usually has stronger model accuracy and generalization ability, and can handle diverse scenarios.

[0093] Exemplarily, the cloud model receives requests from a mobile terminal through an Internet interface. Its operation relies on cloud high-performance computing resources, including GPU / TPU clusters, distributed storage systems, and large-scale memory pools, and can support complex model architectures with hundreds of billions of parameters. The training data of the cloud model covers multi-domain, multi-modal public datasets and industry-specific data, enabling it to demonstrate strong generalization ability in tasks such as natural language understanding, high-precision image recognition, and cross-domain knowledge reasoning.

[0094] In addition, the cloud model can support parallel access by multiple users and dynamically allocate computing resources based on a load balancing mechanism to ensure that the inference latency is within a reasonable range. For example, in an intelligent customer service scenario, the cloud model can simultaneously process user queries from different regions and provide accurate answers based on rich historical data and the latest knowledge base updates. In an industrial inspection application, the cloud model can analyze data uploaded from different sensors, identify equipment anomalies, and provide predictive maintenance suggestions. In addition, the cloud model can also continuously learn, update model parameters through periodic training, maintain adaptability to new data distributions, and improve prediction accuracy.

[0095] In an embodiment of the present disclosure, the second model may be a local model. A local model is an AI model deployed on a mobile terminal device (such as a mobile phone, tablet), which runs directly on the hardware resources of the device (such as CPU / GPU / NPU, memory), and can access local PKB data (Personal Knowledge Base, user personalized data, such as usage habits, historical behaviors, biometric features, etc.). The local model can directly utilize the locally stored PKB data without uploading sensitive data to the cloud, thus ensuring privacy.

[0096] The local model runs directly on the terminal hardware resources (such as NPU accelerators, GPU rendering units, and memory). The core advantage of the local model is the real-time access to the local PKB database of the device. The second input data can deeply integrate the PKB content in the second processing stage, for example, injecting the user's frequently selected route preferences in a navigation request and associating the allergy drug history in a health consultation, so as to generate a personalized inference context.

[0097] The local model can run in an offline environment and is suitable for application scenarios with high requirements for real-time performance and privacy protection. For example, in a smart voice assistant, the local model can directly process the user's voice commands and optimize them in combination with the locally stored voice habits to make the response more in line with the user's needs. In a medical and health management application, the local model can provide personalized health advice based on the user's health records and sensor data and still work properly even in a network-free environment.

[0098] Figure 4 A structural framework diagram of a data processing system based on a cloud model and a local model according to an embodiment of the present disclosure is schematically shown.

[0099] As Figure 4 shown, the structural framework of the data processing system according to an embodiment of the present disclosure schematically shows each functional module and its interaction method. This system combines the advantages of local computing and cloud computing. Among them, the core functional modules on the mobile user terminal include a standardization and extension module, a local computing module, and a PKB database. The local computing module includes a local model, and the cloud computing module includes a cloud model.

[0100] The main function of the standardization and extension module is to perform standardization processing and user intention extension on the target input data (such as an initial prompt) input by the user to generate a first input data (such as a standardized prompt) and a second input data (such as a user intention extension prompt). Among them, the first input data is uploaded to the cloud computing module to obtain a general result trained based on large-scale general knowledge data; the second input data is then passed to the local computing module to perform personalized inference and generate a personalized result. The standardization and extension module may include a target intelligent engine.

[0101] During the processing of the standardization and extension modules, the role of the PKB database is particularly important. During the user intention extension process, the system will query the PKB to retrieve information related to the user's historical habits in order to provide content that better meets personalized needs. For example, if the PKB record shows that the user has a habit of watching movies and often moves in a specific area, the input data after intention extension can be [I'm very tired, have a habit of watching movies, location information, regular activity area]. This enables the local AI model to reason based on the user's long-term preferences and then generate personalized recommendations.

[0102] During the standardization process, the system can also query the PKB to extract more generalized knowledge and use it as the basis for standardization processing. For example, based on the above input "I'm very tired", the system retrieves the user's regular activity area and relevant service data in the PKB and expands content such as [cinemas within the area, movies showing, plot summaries]. This process helps to standardize the user input into more general query content, enabling the remote AI model to more effectively understand the user's needs and provide more generalizable response results.

[0103] Furthermore, both user intention extension and standardization processing can be enhanced based on the PKB, and the two can be interrelated. User intention extension can expand personalized knowledge based on the PKB, while standardization processing can obtain more generalized general knowledge based on the PKB. For example, when the user inputs "I'm very tired", the system first identifies the intention of this input. If the user needs more comprehensive knowledge (such as recommended activities), the system executes the complete data processing path, including PKB expansion for both user intention extension and standardization processing, and combines it with the remote AI model for general knowledge calculation; if the user's needs are relatively simple, such as an instruction that the local device can directly execute, there is no need for standardization and general knowledge processing, but the instruction is directly executed.

[0104] Subsequently, during the reasoning process of the data processing system, by combining the personalized input information with intention extension and the generalized knowledge with standardization processing, the system can finally generate a target output that better meets the user's needs. For example, based on the two extended data sets [I'm very tired, have a habit of watching movies, location information, regular activity area] and [cinemas within the area, movies showing, plot summaries], the system can combine the knowledge base of the cloud AI model and finally generate a recommendation result, such as "The following movies are showing at the cinema in your current area: XXX, and the plot summary is as follows..." etc.

[0105] In an embodiment of the present disclosure, the system can also dynamically select a data processing path through a real-time intent recognition mechanism. For example, when a user enters a request, the system first performs semantic parsing on the input content through a local lightweight intent classification model to determine whether the task type belongs to "a high-complexity task that requires global knowledge support" or "a simple instruction that can be independently processed locally". If it is recognized as a simple instruction that can be covered locally (such as "turn on the bedroom lights" or "query today's schedule"), the system will directly call the local model of the terminal for real-time processing, bypassing the cloud collaborative processing path.

[0106] In addition, to enhance data privacy protection, the standardization and extension module can be split into two independent sub-modules, namely the user prompt standardization module and the user intent extension module, which can correspond to the first intelligent engine and the second intelligent engine. The user prompt standardization module focuses on standardizing the user input, removing sensitive information, and converting it into a more general query format without referring to PKB data. The user intent extension module is responsible for intent extension, enhancing the user input in a personalized manner by combining PKB data to ensure that the response content not only conforms to general logic but also meets personalized needs.

[0107] Overall, through the standardization and intent extension processing of the standardization and extension module in the data processing system of the present disclosure, the user prompt can efficiently flow between the cloud model and the local model, thereby achieving the data processing ability to meet both general needs and personalized needs. In addition, the system flexibly adjusts the data processing path according to the complexity of the user input intent, ensuring the efficient use of computing resources while protecting user privacy and security. This system can be widely applied in fields such as intelligent assistants, personalized recommendations, and intelligent Q&A to improve the human-computer interaction experience and the accuracy of data processing.

[0108] Figure 5 A structural block diagram of a data processing system according to an embodiment of the present disclosure is schematically shown.

[0109] As Figure 5 shown, the processing device 500 for interest information in this embodiment includes an input data acquisition module 510, a processing module 520, a first input module 530, a second input module 540, and a response acquisition module 550.

[0110] The input data acquisition module 510 can be used to acquire target input data. In one embodiment, the input data acquisition module 510 can be used to perform the operation S210 described above, which will not be elaborated here.

[0111] The processing module 520 can be used to perform a first processing and a second processing on the target input data respectively to obtain a first input data and a second input data in response to the target input data satisfying a target condition; the information content of the first input data is different from that of the second input data. In one embodiment, the processing module 520 can be used to execute the operation S220 described above, which will not be elaborated here.

[0112] The first input module 530 can be used to input at least the first input data into a first model. In one embodiment, the first input module 530 can be used to execute the operation S230 described above, which will not be elaborated here.

[0113] The second input module 540 can be used to input at least the second input data into a second model. In one embodiment, the second input module 540 can be used to execute the operation S240 described above, which will not be elaborated here.

[0114] The response acquisition module 550 can be used to obtain a target output data in response to the target input data based on the outputs of the first model and the second model. In one embodiment, the response acquisition module 550 can be used to execute the operation S250 described above, which will not be elaborated here.

[0115] According to an embodiment of the present disclosure, the processing module 520 can also be used to input the target input data into a target intelligent engine, so that the target intelligent engine performs a first processing and a second processing on the target input data to output a first input data and a second input data; or input the target input data into a first intelligent engine and a second intelligent engine respectively, so that the first intelligent engine performs a first processing on the target input data to output the first input data, and the second intelligent engine performs a second processing on the target input data to output the second input data.

[0116] According to an embodiment of the present disclosure, the processing module 520 may further include a first processing module and a second processing module.

[0117] According to an embodiment of the present disclosure, the first processing module is used to remove the target semantic expression containing individual features in the target input data. The first processing module is further used to expand the data content included in the target input data based on a target database; wherein the target database is used to store privatized data.

[0118] According to an embodiment of the present disclosure, the second processing module is used to expand the data content included in the target input data. The second processing module is further used to expand the data content included in the target input data based on the target database; wherein, the data content expanded by the first processing module based on the target database is abstract content; the data content expanded by the second processing module based on the target database is specific content.

[0119] According to an embodiment of the present disclosure, the response acquisition module 550 may further be used to input at least the first output data of the first model into the second model; obtain the target output data in response to the target input data according to the second output data of the second model; or obtain the target output data in response to the target input data according to the third output data of the first model and the fourth output data of the second model.

[0120] According to an embodiment of the present disclosure, any multiple of the input data acquisition module 510, the processing module 520, the first input module 530, the second input module 540, and the response acquisition module 550 may be combined and implemented in one module, or any one of them may be split into multiple modules. Or, at least part of the functions of one or more of these modules may be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the input data acquisition module 510, the processing module 520, the first input module 530, the second input module 540, and the response acquisition module 550 may be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or any other reasonable way of integrating or packaging circuits, etc., implemented by hardware or firmware, or implemented in any one of the three implementation manners of software, hardware, and firmware, or in an appropriate combination of any several of them. Or, at least one of the input data acquisition module 510, the processing module 520, the first input module 530, the second input module 540, and the response acquisition module 550 may be at least partially implemented as a computer program module, and when the computer program module runs, it can execute the corresponding functions.

[0121] An embodiment of the present disclosure further provides an electronic device, which includes at least one processor, a memory, and a first application program running on the electronic device. The first application program can execute data processing tasks, including tasks of parsing input data, calling a target model to execute tasks, and finally generating target output data.

[0122] In this embodiment, the first application is configured to receive target input data and parse the target input data to determine at least one task. For the target input data that meets the target conditions, the first application executes a preset data processing strategy, that is, performs a first processing and a second processing on the target input data to generate first input data and second input data respectively. The information content of the first input data and the second input data is different, ensuring that the data has different characteristics in different processing paths to meet the requirements of different computing models.

[0123] The first input data is at least input to the first model for calculation processing, while the second input data is at least input to the second model for calculation processing. The first model and the second model respectively perform inference calculations to generate their respective output data. The first application further obtains target output data through fusion processing based on the output data of the first model and the second model. The target output data is used to respond to the initial target input data to meet user needs or further support the execution of subsequent tasks.

[0124] In this embodiment, the target models called by the first application may include the first model and / or the second model, that is, the system can dynamically call one or more computing models according to requirements to perform data inference. For example, in some scenarios, the first model and the second model can perform collaborative calculations to achieve the combination of general inference and personalized inference; in other scenarios, the first model or the second model can be called separately to meet specific task requirements.

[0125] Optionally, the first application may be an agent deployed in an electronic device, which is an application of artificial intelligence technology and can be implemented based on a target model (such as a large language model). The behavior of the agent can be determined by the target model called by the agent according to the current state and external input, and can be used to specifically solve a certain type of problem. Among them, the target model provides a decision-making basis through learning and training with a large amount of data; the agent can also call tools, plugins, knowledge bases, etc. to provide reasoning, decision-making, and execution capabilities. Simply put, the target model can provide decision-making support, and the agent can continuously optimize the performance of its internal target model through the data fed back to the agent from actual applications or the information input by the user to the agent, such as a prompt. (The prompt can be the text corresponding to the user input information and / or a preset second text for guiding the target model to perform inference).

[0126] In a specific example, the agent can receive information input by the user and can also receive data fed back and sent by other applications (such as the intelligent assistant application in the device); afterwards, the agent can analyze the received input information, further call the target model for corresponding processing, and obtain the processing result of the target model. The agent can perform display or execute operations based on the processing result of the target model. Specifically, the agent can execute the method embodiments described above, perform intent matching for the received information to be matched, can also call the target model to execute the second screening method, determine the result intent information, and execute operations to implement the intent represented by the result intent information.

[0127] In addition, the electronic device can adopt various hardware architectures, including but not limited to smartphones, tablets, laptops, servers, etc. During implementation, the first application program can combine local computing resources and remote computing resources of the device to achieve efficient data processing. In particular, the first model and the second model can be deployed locally or in the cloud respectively, and data interaction can be carried out through network communication to optimize computing performance and response speed.

[0128] Figure 6 A block diagram of an electronic device suitable for implementing a data processing method according to an embodiment of the present disclosure is schematically shown.

[0129] As Figure 6 shown, the electronic device 600 according to an embodiment of the present disclosure includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage section 608 into a random access memory (RAM) 603. The processor 601 can include, for example, a general microprocessor (such as a CPU), an instruction set processor, and / or a related chipset, and / or a dedicated microprocessor (such as an application specific integrated circuit (ASIC)), etc. The processor 601 can also include on-board memory for caching purposes. The processor 601 can include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0130] In the RAM 603, various programs and data required for the operation of the electronic device 600 are stored. The processor 601, the ROM 602, and the RAM 603 are connected to each other through a bus 604. The processor 601 executes various operations of the method flow according to an embodiment of the present disclosure by executing the programs in the ROM 602 and / or the RAM 603. It should be noted that the program can also be stored in one or more memories other than the ROM 602 and the RAM 603. The processor 601 can also execute various operations of the method flow according to an embodiment of the present disclosure by executing the programs stored in the one or more memories.

[0131] According to an embodiment of the present disclosure, the electronic device 600 may further include an input / output (I / O) interface 605, and the input / output (I / O) interface 605 is also connected to the bus 604. The electronic device 600 may further include one or more of the following components connected to the input / output (I / O) interface 605: an input portion 606 including a keyboard, a mouse, etc.; an output portion 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc. and a speaker, etc.; a storage portion 608 including a hard disk, etc.; and a communication portion 609 including a network interface card such as a LAN card, a modem, etc. The communication portion 609 performs communication processing via a network such as the Internet. The drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 610 as needed so that a computer program read from it can be installed into the storage portion 608 as needed.

[0132] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist separately without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiments of the present disclosure is implemented.

[0133] According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, it may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program can be used by or combined with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present disclosure, the computer-readable storage medium may include the above-described ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603.

[0134] An embodiment of the present disclosure also includes a computer program product, which includes a computer program, and the computer program includes program code for executing the method shown in the flowchart. When the computer program product runs in a computer system, the program code is used to cause the computer system to implement the data processing method provided by the embodiments of the present disclosure.

[0135] When the computer program is executed by the processor 601, the above functions defined in the system / apparatus of the embodiments of the present disclosure are executed. According to the embodiments of the present disclosure, the above-described systems, apparatuses, modules, units, etc. can be implemented by computer program modules.

[0136] In one embodiment, the computer program can rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program can also be transmitted and distributed in the form of signals on a network medium, and be downloaded and installed through the communication part 609, and / or be installed from the removable medium 611. The program code included in the computer program can be transmitted by any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0137] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 609, and / or be installed from the removable medium 611. When the computer program is executed by the processor 601, the above functions defined in the system of the embodiments of the present disclosure are executed. According to the embodiments of the present disclosure, the above-described systems, devices, apparatuses, modules, units, etc. can be implemented by computer program modules.

[0138] According to the embodiments of the present disclosure, the program code for executing the computer program provided by the embodiments of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages include but are not limited to, such as Java, C++, python, the "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any kind of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, by using an Internet service provider to connect through the Internet).

[0139] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a portion of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that, in some alternative implementations, the functions noted in the blocks may occur in a different order than that noted in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, or they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, and combinations of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or by a combination of dedicated hardware and computer instructions.

[0140] Those skilled in the art will appreciate that the features described in the various embodiments of the present disclosure can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present disclosure. In particular, without departing from the spirit and teachings of the present disclosure, the features described in the various embodiments of the present disclosure can be combined and / or combined in various ways. All such combinations and / or combinations fall within the scope of the present disclosure.

[0141] The embodiments of the present disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although the embodiments have been described separately above, this does not mean that the measures in each embodiment cannot be used advantageously in combination. Without departing from the scope of the present disclosure, those skilled in the art can make various substitutions and modifications, and all such substitutions and modifications should fall within the scope of the present disclosure.

Claims

1. A data processing method, the method comprising: Obtaining target input data; In response to the target input data satisfying a target condition, performing a first process and a second process on the target input data respectively to obtain first input data and second input data; The information content of the first input data is different from that of the second input data; Inputting at least the first input data into a first model; Inputting at least the second input data into a second model; And Based on the outputs of the first model and the second model, obtaining target output data in response to the target input data.

2. The data processing method according to claim 1, wherein obtaining the target output data in response to the target input data based on the outputs of the first model and the second model comprises: Inputting at least the first output data of the first model into the second model; Obtaining the target output data in response to the target input data according to the second output data of the second model; Or Obtaining the target output data in response to the target input data according to the third output data of the first model and the fourth output data of the second model.

3. The data processing method according to claim 1, wherein performing the first process and the second process on the target input data respectively to obtain the first input data and the second input data comprises: Inputting the target input data into a target intelligent engine, so that the target intelligent engine performs the first process and the second process on the target input data to output the first input data and the second input data; Or Inputting the target input data into a first intelligent engine and a second intelligent engine respectively, so that the first intelligent engine performs the first process on the target input data to output the first input data, and the second intelligent engine performs the second process on the target input data to output the second input data.

4. The data processing method according to any one of claims 1 to 3, wherein the first process is used to remove the target semantic expression containing individual characteristics in the target input data; The second process is used to expand the data content included in the target input data.

5. The data processing method according to claim 3, wherein the first process is further used to expand the data content included in the target input data based on a target database; Among them, The target database is used to store privatized data.

6. The data processing method according to claim 5, wherein the second process is used to expand the data content included in the target input data based on the target database; Among them, The data content expanded by the first process based on the target database is abstract content; the data content expanded by the second process based on the target database is specific content.

7. The data processing method according to claim 1, in response to the target input data not satisfying the target condition, inputting the target input data into the second model to obtain the target output data in response to the target input data.

8. The data processing method according to claim 1, wherein the first model includes a cloud intelligent model, and the second model includes a terminal intelligent model.

9. A data processing system, comprising: An input data acquisition module for acquiring target input data; A processing module for, in response to the target input data satisfying a target condition, respectively performing a first processing and a second processing on the target input data to obtain first input data and second input data; The information content of the first input data is different from that of the second input data; A first input module for inputting at least the first input data into a first model; A second input module for inputting at least the second input data into a second model; And A response acquisition module for obtaining target output data in response to the target input data based on the outputs of the first model and the second model.

10. An electronic device, comprising: A first application program running on the electronic device; The first application program is configured to: parse at least one task based on an input, call a target model to execute at least one task, and at least be used to execute: Receive target input data; In response to the target input data satisfying a target condition, respectively perform a first processing and a second processing on the target input data to obtain first input data and second input data; The information content of the first input data is different from that of the second input data; Input at least the first input data into a first model; Input at least the second input data into a second model; Obtain target output data in response to the target input data based on the outputs of the first model and the second model; The target model includes the first model and / or the second model.

Images