Data processing methods, apparatus, computer equipment and computer-readable storage media

Abstract

This application provides a data processing method, apparatus, computer device, and computer-readable storage medium to improve the privacy protection of user data in voice interaction scenarios. The method includes: acquiring data to be processed; performing semantic recognition on the data to be processed to obtain a semantic feature vector; performing behavior clustering processing on the semantic feature vector to obtain a behavior pattern identifier; and uploading the behavior pattern identifier to a cloud server so that the cloud server can generate an intelligent response based on the behavior pattern identifier.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, specifically to a data processing method, apparatus, computer device, and computer-readable storage medium. Background Technology

[0002] Currently, intelligent voice assistants such as Siri and Baidu's Xiaodu are widely used. Their mainstream technical solution employs a centralized cloud server processing model: user devices collect raw voice data and upload it directly, with the cloud server performing complex calculations such as speech recognition and natural language understanding. However, this model has two significant drawbacks: first, a serious risk of privacy leakage, as the user's raw voice contains a large amount of sensitive content and is itself a biometric feature; uploading it to the cloud server exposes it completely to service providers, posing a risk of misuse or leakage; second, high operating costs, as cloud servers consume enormous computing and bandwidth resources to handle massive requests, and the service is heavily reliant on network connectivity.

[0003] To address these issues, the business offers data encryption transmission and localized semantic encoding solutions. However, both solutions still have their limitations. For instance, while data encryption protects data during transmission, the cloud server must decrypt the data to process it, thus the root cause of trust risks remains. Furthermore, the semantic information converted using localized semantic encoding remains highly sensitive and can be used for user tracking and profiling.

[0004] Therefore, how to improve the privacy protection of user data in voice interaction scenarios is an urgent problem to be solved. Summary of the Invention

[0005] This application provides a data processing method, apparatus, computer device, and computer-readable storage medium to improve the privacy protection of user data in voice interaction scenarios.

[0006] The technical solution adopted by this invention to solve the problem is as follows: Firstly, this application provides a data processing method, including: Obtain the data to be processed; Semantic recognition is performed on the data to be processed to obtain semantic feature vectors; The semantic feature vector is subjected to behavior clustering to obtain behavior pattern identifiers; The behavior pattern identifier is uploaded to the cloud server so that the cloud server can generate intelligent responses based on the behavior pattern identifier.

[0007] In some embodiments of this application, the semantic recognition of the data to be processed to obtain a semantic feature vector includes: The semantic encoder deployed locally is invoked to perform semantic recognition on the data to be processed in order to obtain the semantic feature vector, which is used to characterize the semantic information of the data to be processed.

[0008] In some embodiments of this application, the behavioral clustering processing of the semantic feature vector to obtain behavioral pattern identifiers includes: The autoencoder deployed locally is invoked to compress the semantic feature vector to obtain a low-dimensional semantic feature vector, which is used to represent features related to behavioral patterns. Clustering is performed on the low-dimensional semantic feature vector to obtain the behavioral pattern identifier.

[0009] In some embodiments of this application, uploading the behavior pattern identifier to a cloud server so that the cloud server can generate an intelligent response based on the behavior pattern identifier includes: The behavior pattern identifier was anonymized to obtain an anonymized behavior pattern identifier. The desensitized behavior pattern identifier is uploaded to the cloud server so that the cloud server can generate an intelligent response based on the desensitized behavior pattern identifier.

[0010] In some embodiments of this application, the desensitization process for the behavioral pattern identifier to obtain a desensitized behavioral pattern identifier includes: Obtain the privacy budget for this behavior pattern identifier; Based on the differential privacy model, the privacy budget is used to inject noise into the behavior pattern identifier to obtain the de-identified behavior pattern identifier.

[0011] In some embodiments of this application, the privacy budget for obtaining the behavioral pattern identifier includes: Obtain the frequency value of the behavior pattern corresponding to the behavior pattern identifier; When the frequency value is greater than the first threshold, a first privacy budget is allocated to the behavior pattern identifier; When the frequency value is less than or equal to the first threshold, a second privacy budget is allocated to the behavior pattern identifier, wherein the second privacy budget is greater than the first privacy budget.

[0012] In some embodiments of this application, the cloud server maps and decodes the de-identified behavior pattern identifier based on an abstract semantic library to obtain the behavior pattern corresponding to the data to be processed. The behavior pattern is used to generate the intelligent response. The abstract semantic library is built based on historical behavior patterns.

[0013] Secondly, this application provides a data processing apparatus, comprising: The acquisition module is used to acquire data to be processed. The processing module is used to perform semantic recognition on the data to be processed to obtain a semantic feature vector; and to perform behavior clustering processing on the semantic feature vector to obtain a behavior pattern identifier. The transceiver module is used to upload the behavior pattern identifier to the cloud server, so that the cloud server can generate a smart response based on the behavior pattern identifier.

[0014] Thirdly, this application also provides a computer device, which includes: One or more processors; Memory; and One or more applications, wherein the applications are stored in memory and configured to be executed by a processor to implement the data processing method of any of the first aspects.

[0015] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, the computer program being loaded by a processor to perform the steps of the data processing method of any one of the first aspects.

[0016] The beneficial effects of this invention are as follows: It introduces a semantic hierarchical abstraction scheme, whereby the input data is first semantically recognized to obtain a semantic vector; then, this semantic vector is abstracted and compressed to obtain a pattern ID. In the process of generating the pattern ID from the semantic vector, sensitive information in the input data is removed, thus partially de-identifying the input data and enhancing its privacy protection. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the system architecture provided in an embodiment of the present invention; Figure 2 This is a schematic flowchart of an embodiment of the data processing method provided by the present invention; Figure 3 This is a schematic diagram of another embodiment of the data processing method provided in this invention; Figure 4 This is a flowchart illustrating a data processing method provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of a software module of the data processing device provided in an embodiment of the present invention; Figure 6 A flowchart illustrating the behavior pattern identification mapping scheme provided in this embodiment of the invention; Figure 7 This is a schematic diagram of a specific embodiment of the data processing device provided in this invention. Figure 8 This is a schematic diagram of an embodiment of the computer device provided in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0020] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more features.

[0021] In this application, the term "exemplary" is used to mean "used as an example, illustration, or description." Any embodiment described as "exemplary" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes are not described in detail to avoid obscuring the description of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0022] It should be noted that since the method in this application embodiment is executed in a computer device, the processing objects of each computer device exist in the form of data or information, such as time, which is essentially time information. It is understood that if size, quantity, position, etc. are mentioned in subsequent embodiments, they are all corresponding data that exist so that the computer device can process them. Specific details will not be elaborated here.

[0023] With the rapid development of artificial intelligence and the Internet of Things (IoT) technologies, intelligent voice assistants (such as Apple's Siri, Baidu's "Xiaodu," and Amazon's Alexa) on platforms like smart speakers, smartphones, and in-vehicle systems have become important entry points for human-computer interaction. These devices receive users' voice commands to perform diverse functions such as information retrieval, device control, and schedule management, greatly enhancing the convenience of users' lives.

[0024] The core technical process of intelligent voice assistants typically includes speech recognition, natural language understanding, dialogue management, and task execution. To achieve high-precision speech recognition and deep semantic understanding, current mainstream solutions generally employ a centralized cloud server processing model. In this model, the user device is responsible for collecting the raw speech signal and then uploading it directly to the cloud server. Leveraging its powerful computing capabilities, the cloud server runs complex and large-scale artificial intelligence models to complete the core processing steps such as speech recognition and natural language understanding, and then returns the results to the user device.

[0025] However, existing solutions, such as centralized processing via cloud servers, have the following significant drawbacks: (1) Serious privacy breach risk: Users' voice data is highly sensitive personal information, potentially containing not only private details such as schedules, contacts, health status, and financial information, but also biometric data such as voiceprints. Uploading this raw, unprocessed data to cloud servers means that user privacy is completely exposed to service providers. Although service providers typically promise data security, data still faces the risk of theft, leakage, or misuse during transmission, storage, and processing, causing widespread privacy anxiety among users.

[0026] (2) High computing and bandwidth costs: Cloud servers need to handle massive concurrent user requests and process raw audio in real time, which consumes enormous computing resources and network bandwidth. For service providers, this represents a continuous and high operating cost. Furthermore, this model is heavily reliant on network connectivity; service interruptions or delays occur when the network is unstable or offline, impacting user experience.

[0027] To overcome the above shortcomings, the industry has proposed some improvement solutions, but each of them still has its limitations: (A) Data encryption transmission scheme: This solution encrypts voice data before it is uploaded and decrypts it on a cloud server before processing. While this can prevent data from being eavesdropped on by third parties during transmission to some extent, it does not solve the fundamental problem. Because the cloud server must possess the key to decrypt the data in order to process it, the data is still in a "plaintext" state when processed on the cloud server. Ultimately, the service provider retains control over user privacy, and the risk of mistrust remains.

[0028] (B) Localized semantic encoding scheme: This solution integrates lightweight speech recognition and natural language understanding models on the user's device, directly converting the user's voice commands into structured semantic information (such as intent, slot values, or feature vectors) locally. Only this semantic information is then uploaded to the cloud server for subsequent tasks. This approach avoids uploading raw voice data, representing a step forward in privacy protection. However, its drawback is: Semantic information remains sensitive: Structured semantic information (e.g., the instruction: "Schedule a meeting with Attorney Wang at 3 p.m."), although not the original speech, is still highly specific and sensitive in its content and can be used for precise user profiling and behavior tracking. The risk of privacy breaches has not been completely eliminated.

[0029] In conclusion, existing technologies, whether centralized processing on cloud servers or improvements such as encryption and localized encoding, have failed to effectively address how to better protect user privacy in intelligent voice interaction scenarios.

[0030] To address this technical problem, this application provides the following technical solution: acquiring data to be processed; performing semantic recognition on the data to obtain semantic feature vectors; performing behavioral clustering processing on the semantic feature vectors to obtain behavioral pattern identifiers; and uploading the behavioral pattern identifiers to a cloud server so that the cloud server can generate intelligent responses based on the behavioral pattern identifiers. This introduces a semantic hierarchical abstraction scheme, whereby the input data is first semantically recognized to obtain semantic vectors; then, these semantic vectors are abstracted and compressed to obtain pattern identifiers. In the process of generating pattern identifiers from semantic vectors, sensitive information in the input data is removed, resulting in partial de-identification of the input data and enhanced privacy protection.

[0031] This application provides a data processing method, apparatus, computer device, and computer-readable storage medium to improve user data privacy protection in voice interaction scenarios. The electronic device provided in this application can be implemented as various types of user terminals or as a server.

[0032] Electronic devices can improve the privacy protection of user data in voice interaction scenarios by running the data processing method provided in the embodiments of this application.

[0033] The above method can be applied to many intelligent voice interaction devices, such as smart TVs, smart speakers, smart air conditioners, etc.

[0034] In one exemplary solution, this data processing method can be applied to a smartphone's voice assistant service. For example, a user sets a command through the voice assistant, and the smartphone performs speech recognition on the user's command to obtain text content (such as "remind me to drink coffee every morning"). Then, the smartphone calls a locally deployed semantic recognition model to perform semantic recognition on the text content to obtain a high-dimensional semantic feature vector. Next, a locally deployed encoder compresses the semantic feature vector to obtain a behavior pattern identifier. Finally, the behavior pattern identifier is uploaded to the cloud server corresponding to the voice assistant. After receiving the behavior pattern identifier, the cloud server maps and encodes the behavior pattern identifier based on an abstract semantic library to obtain the behavior pattern corresponding to the command. Finally, the cloud server can set a specific execution plan for the behavior pattern on the smartphone and execute the execution plan at the corresponding time. For example, a user sets a "remind me to drink coffee every morning" through the voice assistant. The semantic encoder in the terminal device with the voice assistant deployed analyzes the sentence content, generates a feature vector [0.23, -0.17, 1.5], and then maps it to a pattern ID (such as "M102") through an autoencoder. Then, a privacy compression engine allocates low-noise injections based on high-frequency habits (daily reminders) to form a de-identified context code. This de-identified context code is uploaded to a cloud server, which then combines it with an abstract semantic library (such as a pattern dictionary) to map M102 to the general behavior "daily_coffee_reminder" and synchronously generate a response.

[0035] In another exemplary solution, this data processing method can be applied to a smart home scenario. For example, a user sets a command through the smart home system. The smart home system performs voice recognition on the user's command to obtain text content (e.g., "Lights off at 22:00"). Then, the smart home system calls a locally deployed semantic recognition model to perform semantic recognition on the text content to obtain a high-dimensional semantic feature vector. Next, a locally deployed encoder compresses the semantic feature vector to obtain a behavior pattern identifier. Finally, the behavior pattern identifier is uploaded to the cloud server corresponding to the smart home system. After receiving the behavior pattern identifier, the cloud server maps and encodes the behavior pattern identifier based on an abstract semantic library to obtain the behavior pattern corresponding to the command. Finally, the cloud server can set a specific execution plan for the behavior pattern on the smartphone and execute the execution plan at the corresponding time. For example, the user sets "Lights off at 22:00" through the smart home system. Smart home devices record habitual operations to generate pattern IDs (e.g., "H304") and inject Laplace noise to form de-identified data. The cloud server configures a pattern dictionary to identify high-frequency household behaviors and provide accurate energy-saving suggestions.

[0036] It should be understood that the above application scenarios are exemplary application scenarios for data processing, and there may be many other specific application scenarios, which are not limited here.

[0037] The data processing method provided in this application embodiment is applied to, for example, Figure 1 The system architecture diagram shown is for your reference. Figure 1 To support a data processing method, the terminal device 100 connects to the server 300 via network 200, and the server 300 connects to the database 400. Network 200 can be a wide area network (WAN), a local area network (LAN), or a combination of both. The client used to implement the data processing scheme is deployed on the terminal device 100, or it can run on the terminal device 100 as a standalone application. The specific form of the client is not limited here.

[0038] The server 300 involved in this application can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks, and big data and artificial intelligence platforms.

[0039] Terminal equipment 100, also known as user equipment (UE), mobile station (MS), mobile terminal (MT), customer premises equipment (CPE), etc., can be a device that includes both receiving and transmitting hardware, that is, a device with receiving and transmitting hardware capable of performing bidirectional communication on a bidirectional communication link. Such equipment can include cellular or other communication devices with single-line displays, multi-line displays, or no multi-line displays. Examples include handheld devices with wireless connectivity, vehicle-mounted devices, machine-type communication (MTC) terminals, etc. Currently, terminal devices 100 can include: mobile phones, tablets, laptops, PDAs, mobile internet devices (MIDs), wearable devices, virtual reality (VR) devices, augmented reality (AR) devices, wireless terminals in industrial control, wireless terminals in self-driving vehicles, wireless terminals in remote medical surgery, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities, or wireless terminals in smart homes, etc. For example, wireless terminals in self-driving vehicles can be drones, helicopters, or airplanes. For example, wireless terminals in vehicle-to-everything (V2X) systems can be in-vehicle equipment, vehicle-mounted equipment, in-vehicle modules, vehicles, or ships, etc. Wireless terminals in industrial control can be cameras, robots, or robotic arms, etc. Wireless terminals in smart homes can be televisions, air conditioners, robot vacuums, speakers, or set-top boxes, etc.

[0040] It should be noted that the terminal device 100 may be a device or apparatus with a chip, or a device or apparatus with integrated circuitry, or a chip, module, or control unit in the device or apparatus shown above; this application does not impose any specific limitations. The solution provided in this application can be implemented by the terminal device 100 and the server 300 working together.

[0041] In short, a database can be viewed as an electronic filing cabinet—a place to store electronic files, where users can perform operations such as adding, querying, updating, and deleting data. A "database" is a collection of data stored together in a certain way, shared by multiple users, with minimal redundancy, and independent of application programs. A Database Management System (DBMS) is a computer software system designed to manage databases, generally possessing basic functions such as storage, retrieval, security, and backup. DBMSs can be classified according to the database model they support, such as relational or Extensible Markup Language (XML); or according to the type of computer they support, such as server clusters or mobile phones; or according to the query language used, such as Structured Query Language (SQL) or XQuery; or according to performance priorities, such as maximum scale or maximum operating speed; or other classification methods. Regardless of the classification method used, some DBMSs can cross categories, for example, supporting multiple query languages ​​simultaneously. In this application, database 400 can be used to store data such as behavior pattern identifiers, de-identified behavior pattern identifiers, abstract semantic libraries, and encoders or models.

[0042] Those skilled in the art will understand that Figure 1 The system architecture diagram shown is one possible system architecture for this application and does not constitute a limitation on the system architecture of this application. Other system architectures may include more advanced architectures. Figure 1 The number of more or fewer terminal devices or servers shown, for example Figure 1 The diagram shows one server. It is understood that the system architecture may also include one or more other terminal devices or servers, which are not limited here.

[0043] It should be noted that, Figure 1 The system architecture shown is an example. The servers and scenarios described in the embodiments of this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided in the embodiments of this application. As those skilled in the art will know, with the evolution of servers and the emergence of new business scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0044] like Figure 2 The diagram shown is a flowchart of an embodiment of the data processing method in this application. The following description uses a terminal device and a cloud server as the execution entities, and the data processing method may include the following steps 201-205, as detailed below: 201. Obtain the data to be processed.

[0045] In a voice interaction scenario, the terminal device collects voice commands through a voice acquisition device (such as the main microphone array deployed on the terminal device), and then generates the data to be processed based on the voice commands.

[0046] In this embodiment, the terminal device can use the voice command as the data to be processed; or, the terminal device can perform speech recognition on the voice command to obtain the corresponding text content, and use the text content as the data to be processed. The specific implementation is not limited here, as long as the processing of the voice command can be achieved.

[0047] The data to be processed can refer to audio data blocks or text content that, after initial collection and boundary definition, has been confirmed to contain complete and valid user instructions and is ready to enter the semantic recognition stage. The audio data blocks or text content are the direct objects of subsequent processing steps. For example, the data to be processed could be the audio data "Remember to remind me of a meeting at 3 PM"; or it could be the text content "Remember to remind me of a meeting at 3 PM".

[0048] In this embodiment, the terminal device can acquire the voice command as follows: First, the terminal device can be a device with a built-in low-power wake word detection engine. During the operation of the terminal device, the wake word detection engine continuously monitors ambient sounds. When the user speaks a preset wake word (e.g., "Hello, assistant"), the wake word detection engine recognizes the keyword and triggers the main processor and related hardware modules to activate from sleep or standby state. When the terminal device is woken up, its voice acquisition device (e.g., main microphone array) is activated and begins to capture subsequent voice commands issued by the user, i.e., voice signals. The voice command is then sent to an analog-to-digital converter and processed according to a preset sampling rate (e.g., 16kHz) and bit depth (e.g., 16bit) to convert it into a continuous audio stream. In this way, the terminal device acquires a voice command and then uses the voice command as the data to be processed. Alternatively, the terminal device can perform speech recognition on the voice command to obtain text content and use the text content as the data to be processed.

[0049] Optionally, to improve the accuracy of subsequent processing, the acquired audio stream can undergo a series of preprocessing algorithms to obtain a clean audio stream. For example, acoustic echo cancellation can be used to filter out the sound played by the device itself, and noise suppression can be used to reduce the interference of ambient background noise.

[0050] When the terminal device collects the voice command, it can perform Voice Activity Detection (VAD) to analyze the input digital audio stream in real time, accurately determine the start and end points of the voice, and precisely separate and extract the part containing only the user's voice command to form a clearly defined data to be processed.

[0051] The VAD algorithm can be implemented as follows: First, the continuous audio stream is segmented into frames, i.e., the continuous audio stream is divided into a series of short audio frames (e.g., 20 milliseconds per frame). Then, feature extraction is performed on the short audio frames, i.e., for each audio frame, acoustic features that can distinguish speech from non-speech are extracted. Common features include, but are not limited to, short-term energy, short-term zero-crossing rate, and Mel-frequency cepstral coefficients (MFCCs). Speech frames usually have higher energy and specific spectral distributions. Then, based on these acoustic features, boundary judgment is performed on the audio data, for example, the algorithm maintains a state machine (e.g., silence state, speech state). When the acoustic feature values ​​of several consecutive frames exceed a preset speech threshold, the state machine switches from the silence state to the speech state and records the current time point as the speech start point. Conversely, when the feature values ​​of several consecutive frames are below the preset silence threshold, the state machine switches from the speech state back to the silence state and records the current time point as the speech end point. Finally, data is extracted from the audio stream based on the detected state transition times. That is, according to the detected start and end points of the speech, a segment is extracted from the middle of the original digital audio stream. This segment is the final generated voice command containing only the user's instructions.

[0052] Optionally, when the data to be processed is the text content after voice command speech recognition, the terminal device can also call the speech recognition module deployed locally to perform speech recognition on the collected voice command. The specific implementation process can be as follows: First, the terminal device's built-in speech recognition module extracts acoustic features from the voice command, dividing the input voice command into frames and extracting acoustic feature sequences from each frame. Then, the module decodes these acoustic feature sequences using an acoustic model, inputting the extracted acoustic feature sequences into a locally deployed, lightweight acoustic model (e.g., a model based on a recurrent neural network or a deep fully convolutional neural network). This model calculates the probability distribution between acoustic features and phonemes. Finally, the terminal device's built-in speech recognition module generates text based on this decoding result. This involves combining a local language model with decoding algorithms such as a weighted finite state converter or beam search to find the most probable target word sequence on the probability distribution output by the acoustic model, ultimately generating the command text.

[0053] Through the above process, the terminal device can isolate the user's voice commands from a complex acoustic environment and generate the data to be processed.

[0054] It should be understood that, in order to reduce the rate of user privacy leaks and reduce the computational load on terminal devices, the voice activity detection model and the voice recognition model involved can both be lightweight models deployed locally.

[0055] 202. Perform semantic recognition on the data to be processed to obtain semantic feature vectors.

[0056] In this embodiment, in order to ensure that the original text data containing highly sensitive information such as user intent, schedule, and contacts does not need to be uploaded to the cloud server, and to eliminate the risk of leakage or misuse during transmission and cloud server storage, thereby achieving source-level privacy protection, the terminal device can call a semantic encoder deployed locally to perform semantic recognition on the data to be processed, so as to obtain the semantic feature vector corresponding to the data to be processed. The semantic feature vector is used to characterize the high-dimensional semantic information of the data to be processed.

[0057] The semantic feature vector refers to a fixed-length, high-dimensional vector generated by a deep learning model, used to represent the semantic information of sentences in natural language. It is a mathematical and abstract representation of the core intent and semantic information contained in the user's instruction text. The position and direction of this vector in the multidimensional semantic space capture the semantic information of the instruction, rather than its literal composition. For example, "buy coffee" can be represented as a vector [0.23, -0.17, ...].

[0058] It should be understood that when the data to be processed is a voice command (i.e., an audio stream), the process of semantic recognition of the data to be processed by the terminal device may include two stages: the first stage is to call the voice recognition model to perform voice recognition on the voice command in order to output the text content; the second stage is to call the semantic encoder to perform semantic recognition on the text content in order to obtain the semantic feature vector.

[0059] When the data to be processed is text content of a voice command (i.e., when the voice recognition operation has been completed when the data to be processed is acquired), the terminal device can directly call the semantic encoder to perform semantic recognition on the text content in order to obtain the semantic feature vector.

[0060] Optionally, when the terminal device invokes the semantic encoder to perform semantic recognition on the text content, the specific process can be as follows: First, the terminal device needs to perform text preprocessing on the text content, that is, to perform necessary preprocessing on the instruction text generated by speech recognition, such as tokenization and normalization.

[0061] The terminal device then performs vectorization encoding on the preprocessed text. In one exemplary scheme, the terminal device inputs the preprocessed word sequence into a pre-trained semantic encoder model deployed locally for vectorization encoding. This semantic encoder is typically a lightweight model architecture, such as MobileBERT, DistilBERT, DistilBERT-base-uncased, or MiniLMv2-L6-H384, or a custom network designed specifically for on-device use.

[0062] In the aforementioned vectorization encoding process, the semantic encoder can capture the contextual dependencies and deep semantics in the text through its internal multi-layer self-attention mechanism and feedforward network. Finally, the semantic encoder outputs a fixed-dimensional semantic feature vector representing the meaning of the entire instruction text. For example, for the instruction "remind me to drink coffee every morning", the generated vector is [0.23, -0.17, 1.5].

[0063] During this process, the terminal device successfully converted the user's voice commands into a highly abstract feature vector rich in semantic information without relying on a cloud server, laying the foundation for subsequent privacy protection and efficient service. Simultaneously, the terminal device performs semantic understanding locally, avoiding network latency caused by communication with a cloud server, significantly reducing command response time, and improving the immediacy and smoothness of user interaction. Furthermore, it reduces the strong dependence on network connectivity, enhancing service availability in offline or weak network environments. Additionally, because compressed data is transmitted when uploading to the cloud server, it saves network bandwidth costs.

[0064] 203. Perform behavioral clustering on the semantic feature vector to obtain behavioral pattern identifiers.

[0065] In this embodiment, to remove sufficiently distinctive personalized details from the semantic feature vector, preventing the cloud server from inferring the user's refined preferences through data analysis and thus achieving more thorough privacy protection, the terminal device can invoke a locally deployed autoencoder to perform secondary semantic abstraction on the semantic feature vector. Specifically, it invokes the encoder part of a locally deployed, pre-trained autoencoder. This encoder is responsible for compressing the input semantic feature vector into a latent space representation and further mapping it to discrete identifiers through quantization or table lookup. The implementation process can be understood as follows: the terminal device invokes the locally deployed autoencoder to compress the semantic feature vector to obtain a low-dimensional semantic feature vector. During feature vector compression, the autoencoder removes redundant information from the semantic feature vector, retaining only features related to behavioral patterns to obtain the low-dimensional semantic feature vector. Then, the low-dimensional semantic feature vector is subjected to clustering and classification processing to obtain the behavioral pattern identifier corresponding to the data to be processed.

[0066] Here, the behavioral pattern identifier (also known as the habit pattern identifier) ​​refers to a discrete, predefined, and typically non-human-readable unique identifier (e.g., the string "M102" or the integer 102). It does not represent any single specific instruction, but rather an entire cluster of patterns—a set of semantically highly similar user behaviors. Therefore, it can also be an abstract identifier representing user behavioral habits, generated by an autoencoder clustering frequently similar behaviors. For example, the habit of shopping every Friday can be mapped to "M103".

[0067] The autoencoder is a common deep learning technique that consists of an encoder and a decoder, used to reduce data dimensionality, extract data features, or generate reconstructed representations of data.

[0068] The following example illustrates the process of performing secondary semantic abstraction on this terminal device using a specific application scenario: The terminal device loads a lightweight autoencoder that has been trained offline on a cloud server using massive amounts of anonymized data. Simultaneously, the terminal device uses semantic feature vectors (e.g., [0.23, -0.17, 1.5]) as input to the autoencoder. The autoencoder compresses the input semantic feature vectors through multiple layers of nonlinear transformations, removing redundant information and mapping it to a lower-dimensional latent space, generating a latent space representation (i.e., the low-dimensional semantic feature vector in this embodiment).

[0069] It should be understood that during the offline training phase, the latent space has been divided into several non-overlapping regions (e.g., pattern clusters) using clustering algorithms (such as K-Means or DBSCAN), and each region is assigned a unique behavioral pattern identifier. Therefore, the autoencoder quickly determines which pattern cluster the newly generated latent space representation falls into by calculating the distance between it and each predefined pattern cluster, or by using a pre-trained quantization function. Subsequently, the behavioral pattern identifier (e.g., "M102") corresponding to that pattern cluster is extracted as the final output of the autoencoder.

[0070] Optionally, during the offline training phase, when the terminal device performs clustering processing on the latent space (i.e., the low-dimensional semantic feature vector in this embodiment) using a clustering algorithm, it can also construct a pattern mapping table based on the clustering structure. In this pattern mapping table, each non-overlapping region (i.e., pattern cluster) of the latent space has a unique pattern cluster index; each index corresponds to a stable behavioral pattern identifier (a simplified number or code); and each behavioral pattern identifier is associated with a human-readable abstract semantic label (such as "check the weather" or "play music").

[0071] Similarly, during the offline training phase, an abstract semantic library (also known as a pattern dictionary) can be built based on this behavior pattern identifier. This abstract semantic library is responsible for maintaining the correspondence between all behavior pattern identifiers and high-level semantic labels.

[0072] Through the above process, the terminal device can transform a specific semantic feature vector that may reveal subtle privacy information into an abstract, standardized category identifier, which greatly improves the privacy and security level of the data and prepares for efficient and low-cost interaction with the cloud server in the future.

[0073] 204. Upload the behavior pattern identifier to the cloud server.

[0074] After obtaining the behavior pattern identifier, the terminal device can upload it to the cloud server. The upload process can utilize either a wireless network or a wired network; the specific method is not limited here.

[0075] 205. The cloud server generates intelligent responses based on this behavior pattern identifier.

[0076] After receiving the behavior pattern identifier, the cloud server maps and decodes the behavior pattern identifier based on the abstract semantic library to obtain the corresponding behavior pattern; then it performs an intelligent response based on the behavior pattern.

[0077] This intelligent response can interpret the behavioral pattern corresponding to the data to be processed and provide a corresponding reply. For example, if the user's voice command is "Set a reminder to drink coffee every morning," when the cloud server determines that the behavioral pattern is "Reminder Service," the corresponding application programming interface gateway of the cloud server will send the parsed structured command to the "Reminder Service" microservice. Then, the "Reminder Service" will create a new reminder task under the user's account based on the command content. After the reminder task is successfully executed, a confirmation response will be generated and sent back to the terminal device (such as "Okay, a reminder to drink coffee every morning has been set for you") for playback.

[0078] To further de-identify the data to be processed, the terminal device can also provide services such as... Figure 3 The data processing method shown. For example... Figure 3 The diagram shown is a flowchart of another embodiment of the data processing method in this application. The following description uses a terminal device as the execution subject and describes the data processing method, which may include the following steps 301-306, as detailed below: Steps 301 to 303 are the same as steps 201 to 203 mentioned above, and will not be repeated here.

[0079] 304. Desensitize the behavior pattern identifier to obtain a desensitized behavior pattern identifier.

[0080] In this embodiment, in order to ensure that no analyst can determine whether a specific user has performed a specific behavior, thereby providing solid protection for individual privacy, the terminal device can further desensitize the behavior pattern identifier to obtain a desensitized behavior pattern identifier.

[0081] Optionally, when desensitizing the behavior pattern identifier, the terminal device can introduce a differential privacy model to inject noise into the behavior pattern identifier. The process can be as follows: the terminal device obtains the privacy budget of the behavior pattern identifier; then, based on the differential privacy model, it uses the privacy budget to inject noise into the behavior pattern identifier to obtain the desensitized behavior pattern identifier.

[0082] The de-identified behavior pattern identifier refers to the probabilistic behavior pattern identifier after de-identification processing. This de-identified behavior pattern identifier has a high probability of being the same as the original behavior pattern identifier (to ensure service availability), but there is also a small probability that it will be replaced with another random identifier (to provide privacy guarantees).

[0083] Privacy budget is a non-negative real number parameter used to quantify the strength of differential privacy protection. A smaller privacy budget allows for more injected noise and a higher level of privacy protection, but at the cost of reduced data availability (accuracy); conversely, a larger privacy budget results in lower privacy protection and higher data availability. It is the core element in balancing privacy and utility.

[0084] Differential privacy (DP) is a formally defined privacy protection mechanism that protects data privacy by limiting the extent to which a single record can affect statistical results.

[0085] To achieve a strong desensitization of truly sensitive individual and abnormal behaviors without sacrificing the mainstream user experience, thus balancing service utility and privacy security, this embodiment also introduces a dynamic privacy budget mechanism. The implementation process is as follows: the terminal device obtains the frequency value of the behavior pattern corresponding to the behavior pattern identifier; when the frequency value is greater than a first threshold, a first privacy budget is allocated to the behavior pattern identifier; when the frequency value is less than or equal to the first threshold, a second privacy budget is allocated to the behavior pattern identifier, wherein the second privacy budget is greater than the first privacy budget.

[0086] Optionally, the terminal device can also incorporate sensitivity to behavioral patterns to allocate corresponding privacy budgets. For example, highly sensitive behaviors can be assigned higher noise levels, i.e., a smaller privacy budget; while low-sensitivity or routine behaviors can be assigned lower noise levels, i.e., a larger privacy budget.

[0087] This dynamic privacy budget mechanism allocates low noise to high-frequency, routine behaviors and high noise to low-frequency, sensitive behaviors, thereby achieving strong desensitization of truly individualized and abnormal behaviors that need protection, and achieving a balance between service utility and privacy security.

[0088] 305. Upload the desensitized behavior pattern identifier to the cloud server.

[0089] After obtaining the de-identified behavior pattern identifier, the terminal device can upload it to the cloud server. The upload process can utilize either a wireless network or a wired network; specific methods are not limited here.

[0090] 306. The cloud server generates intelligent responses based on the desensitized behavior pattern identifier.

[0091] After receiving the de-identified behavior pattern identifier, the cloud server maps and decodes the de-identified behavior pattern identifier based on the abstract semantic library to obtain the corresponding behavior pattern; then it performs an intelligent response based on the behavior pattern.

[0092] This intelligent response can interpret the behavioral pattern corresponding to the data to be processed and provide a corresponding reply. For example, if the user's voice command is "Set a reminder to drink coffee every morning," when the cloud server determines that the behavioral pattern is "Reminder Service," the corresponding application programming interface gateway of the cloud server will send the parsed structured command to the "Reminder Service" microservice. Then, the "Reminder Service" will create a new reminder task under the user's account based on the command content. After the reminder task is successfully executed, a confirmation response will be generated and sent back to the terminal device (such as "Okay, a reminder to drink coffee every morning has been set for you") for playback.

[0093] Based on the above description, the following will be based on... Figure 4 The flowchart shown illustrates the data processing method in this embodiment.

[0094] In this embodiment, the data processing method can be divided into two stages, where the first stage is implemented on a local terminal device and the second stage is implemented on a cloud server.

[0095] The modular architecture of the local terminal device and the cloud server can be as follows: Figure 5 As shown, the local endpoint includes a user input module, a semantic encoder, an autoencoder, and a privacy compression engine. The cloud server includes an input module, an abstract semantic library, a privacy budget manager, and an intelligent service module.

[0096] Combined with Figure 5 The module architecture shown and Figure 4 The flowchart shown illustrates the data processing method in this embodiment: The user input module of the local terminal device acquires the data to be processed.

[0097] This semantic encoder can use a deep learning model (such as MobileBERT) to encode the data to be processed (such as natural language input) obtained by the user input module, generating semantic feature vectors. Its implementation logic can be as follows: Input: User statement S (e.g., "Remind me to shop every Friday"); Output: Semantic vector V = [v1, v2, ..., vn].

[0098] This autoencoder includes an encoder for reducing the dimensionality of semantic vectors to extract features and a decoder for reconstructing high-dimensional targets to ensure no feature loss. The autoencoder is then used to cluster the semantic feature vectors output by the semantic encoder to obtain behavioral pattern identifiers.

[0099] The processing method of this autoencoder can be as follows: ID=Cluster(Enc(V)).

[0100] Wherein, ID is used to represent the behavior pattern identifier, V is the input semantic feature vector, Enc is the encoder function, and Cluster represents the clustering operation.

[0101] The privacy compression engine (also known as the differential privacy model in this embodiment) is used to add Laplace noise to the behavior pattern identifier output by the autoencoder, ultimately forming a desensitized behavior pattern identifier.

[0102] Based on the above module description, the following will be used as an example. Figure 6 The flowchart shown illustrates the mapping scheme for this desensitization behavior pattern identifier: exist Figure 6 In the mapping scheme shown, the local terminal device acquires natural language text input by the user; the semantic encoder performs semantic recognition on the natural language text to generate a high-dimensional semantic feature vector; then the encoder part of the autoencoder compresses the high-dimensional semantic feature vector to obtain a low-dimensional semantic feature vector; then, based on a clustering algorithm, the similarity between the low-dimensional semantic feature vector and the offline trained pattern cluster is calculated to obtain the pattern cluster in which the low-dimensional semantic feature vector belongs; then, based on the pattern mapping table and the pattern cluster, a lookup is performed to obtain the original behavioral pattern identifier of the data to be processed; then, the privacy compression engine is called to inject noise into the original behavioral pattern identifier based on the differential privacy model to obtain a de-identified behavioral pattern identifier (i.e., the protected final pattern identifier); finally, the de-identified behavioral pattern identifier is uploaded to the cloud.

[0103] After the privacy compression engine obtains the de-identified behavior pattern identifier, it uploads the de-identified behavior pattern identifier to the cloud server via network communication.

[0104] The cloud server's data input module is used to receive the de-identified behavior pattern identifier and input the de-identified behavior pattern identifier into the pattern dictionary.

[0105] This pattern dictionary is used to map de-identified behavioral pattern identifiers to specific high-level semantic patterns. This pattern dictionary can be represented as a predefined abstract semantic library. This abstract semantic library stores and maps patterns such as "M103" -> "Friday Shopping". Therefore, after obtaining a de-identified behavioral pattern identifier, it can be searched in this abstract semantic library to obtain the corresponding behavioral pattern.

[0106] Optionally, this abstract semantic library can be incrementally updated. That is, the abstract semantic library can be continuously improved based on newly added behavioral patterns.

[0107] Optionally, the privacy budget manager of the cloud server can statistically analyze the behavioral patterns processed each time and obtain the frequency value of the behavioral patterns; then, in subsequent applications, the privacy budget corresponding to the behavioral patterns can be dynamically adjusted based on the frequency value of the behavioral patterns. The implementation logic can be as follows: The frequency of interactions is counted based on voice interaction sessions. When the frequency of interactions is greater than a first threshold, a larger budget is allocated to reduce the noise injected into the behavior pattern. When the frequency of interactions is less than or equal to the first threshold, a smaller privacy budget is allocated to reduce the noise injected into the behavior pattern.

[0108] The cloud server's intelligent service generation module combines real-time semantic input with contextual memory to generate continuous semantic responses. For example, it receives contextual behavior pattern identifiers and the current de-identified behavior pattern identifier; it matches the contextual pattern with real-time semantics through a pattern dictionary, and outputs intelligent linkage suggestions or responses.

[0109] To better implement the data processing method in the embodiments of this application, based on the data processing method, the embodiments of this application also provide a data processing apparatus, such as... Figure 7 As shown, the data processing apparatus 700 includes: The acquisition module 701 is used to acquire data to be processed; The processing module 702 is used to perform semantic recognition on the data to be processed to obtain a semantic feature vector; and to perform behavior clustering processing on the semantic feature vector to obtain a behavior pattern identifier. The transceiver module 703 is used to upload the behavior pattern identifier to the cloud server so that the cloud server can generate an intelligent response based on the behavior pattern identifier.

[0110] In this embodiment, a semantic hierarchical abstraction scheme is introduced. First, the input data undergoes semantic recognition to obtain a semantic vector. Then, this semantic vector is abstracted and compressed to obtain a pattern identifier. During the process of generating the pattern identifier from the semantic vector, sensitive information in the input data is removed, thus partially de-identifying the input data and enhancing its privacy protection.

[0111] In some embodiments of this application, the processing module 702 is specifically used for: The semantic encoder deployed locally is invoked to perform semantic recognition on the data to be processed in order to obtain the semantic feature vector, which is used to characterize the semantic information of the data to be processed.

[0112] In this embodiment, a locally deployed semantic encoder performs semantic recognition on the data to be processed to obtain a high-dimensional semantic feature vector. This allows the text content recognized from the user's voice commands to be directly converted into an abstract semantic feature vector within the device. This localization of the process fundamentally ensures that the original text data containing highly sensitive information such as the user's specific intentions, schedules, and contacts does not need to be uploaded to a cloud server, eliminating the risk of leakage or misuse during transmission and cloud server storage, thus achieving source-level privacy protection. Secondly, since the crucial step of semantic understanding is completed instantly locally, network latency caused by communication with a cloud server is avoided, significantly reducing command response time and improving the immediacy and smoothness of user interaction. Furthermore, this solution reduces the strong dependence on network connectivity, enhances service availability in offline or weak network environments, and significantly saves network bandwidth costs by only needing to transmit compressed data.

[0113] In some embodiments of this application, the processing module 702 is specifically used for: The autoencoder deployed locally is invoked to compress the semantic feature vector to obtain a low-dimensional semantic feature vector, which is used to represent features related to behavioral patterns. Clustering is performed on the low-dimensional semantic feature vector to obtain the behavioral pattern identifier.

[0114] In this embodiment, the compression process of the autoencoder, as a second semantic abstraction, compresses a group of semantically similar vectors into a discrete, standardized pattern identifier. It actively removes detailed information from the vectors that could distinguish individual preferences, preventing the cloud server from inferring users' nuanced preferences through data analysis, thus achieving more thorough privacy protection. Simultaneously, compressing a complex floating-point vector into a minimal pattern identifier significantly reduces the volume of data to be uploaded, substantially lowers network bandwidth consumption and cloud server storage costs, and simplifies the processing logic of cloud server services.

[0115] In some embodiments of this application, the processing module 702 is specifically used to perform desensitization processing on the behavior pattern identifier to obtain a desensitized behavior pattern identifier; The transceiver module 703 is specifically used to upload the de-identified behavior pattern identifier to the cloud server, so that the cloud server can generate an intelligent response based on the de-identified behavior pattern identifier.

[0116] In this embodiment of the application, the behavior pattern identifier is further anonymized, which makes it impossible for any analyst to determine whether a specific user has performed a specific behavior, thus providing solid protection for individual privacy.

[0117] In some embodiments of this application, the processing module 702 is specifically used for: Obtain the privacy budget for this behavior pattern identifier; Based on the differential privacy model, the privacy budget is used to inject noise into the behavior pattern identifier to obtain the de-identified behavior pattern identifier.

[0118] In this embodiment, a differential privacy algorithm is introduced to de-identify behavioral pattern identifiers, fundamentally breaking the association between pattern identifiers and users. This makes it impossible for any analyst to determine whether a specific user has performed a particular behavior, thus providing robust protection for individual privacy.

[0119] In some embodiments of this application, the processing module 702 is specifically used for: Obtain the frequency value of the behavior pattern corresponding to the behavior pattern identifier; When the frequency value is greater than the first threshold, a first privacy budget is allocated to the behavior pattern identifier; When the frequency value is less than or equal to the first threshold, a second privacy budget is allocated to the behavior pattern identifier, wherein the second privacy budget is greater than the first privacy budget.

[0120] In this embodiment, by dynamically assigning low noise to high-frequency, routine behaviors and high noise to low-frequency, sensitive behaviors, powerful desensitization of truly individualized and abnormal behaviors that need protection is achieved without sacrificing the mainstream user experience, thus achieving a balance between service utility and privacy security.

[0121] In some embodiments of this application, the cloud server maps and decodes the de-identified behavior pattern identifier based on an abstract semantic library to obtain the behavior pattern corresponding to the data to be processed. The behavior pattern is used to generate the intelligent response. The abstract semantic library is built based on historical behavior patterns.

[0122] In this embodiment, the cloud server receives an abstract identifier with extremely low information entropy, which does not carry any specific user context or sensitive details. By querying a pre-built semantic library, the cloud server directly maps this identifier into standardized, executable, general task instructions. This process achieves complete separation between cloud server task execution and edge semantic understanding. The cloud server completes intelligent response services without knowing the specific data, eliminating the risk of cloud server misuse or leakage of user intent from an architectural perspective. Furthermore, querying the abstract semantic library has minimal computational overhead compared to running a large natural language understanding model on the cloud server in real time, thereby significantly reducing server costs, improving response speed, and making the entire system easier to maintain and expand.

[0123] This application also provides a computer device that integrates any of the data processing apparatuses provided in this application, the computer device comprising: One or more processors; Memory; and One or more applications, wherein the applications are stored in memory and configured to be executed by a processor from the steps of the data processing method in any of the embodiments described above.

[0124] This application also provides a computer device that integrates any of the data processing devices provided in this application. For example... Figure 8 As shown, it illustrates a structural schematic diagram of the computer device involved in the embodiments of this application, specifically: The computer device may include components such as a processor 801 with one or more processing cores, a memory 802 with one or more computer-readable storage media, a power supply 803, and an input unit 804. Those skilled in the art will understand that... Figure 8 The computer device structure shown does not constitute a limitation on the computer device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein: The processor 801 is the control center of the computer device. It connects various parts of the computer device via various interfaces and lines. By running or executing software programs and / or modules stored in the memory 802, and by calling data stored in the memory 802, it performs various functions of the computer device and processes data, thereby providing overall monitoring of the computer device. Optionally, the processor 801 may include one or more processing cores; preferably, the processor 801 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 801.

[0125] The memory 802 can be used to store software programs and modules. The processor 801 executes various functional applications and data processing by running the software programs and modules stored in the memory 802. The memory 802 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 802 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 802 may also include a memory controller to provide the processor 801 with access to the memory 802.

[0126] The computer device also includes a power supply 803 that supplies power to the various components. Preferably, the power supply 803 can be logically connected to the processor 801 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 803 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0127] The computer device may also include an input unit 804, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0128] Although not shown, the computer device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 801 in the computer device loads the executable files corresponding to the processes of one or more application programs into the memory 802 according to the following instructions, and the processor 801 runs the application programs stored in the memory 802 to realize various functions, as follows: Obtain the data to be processed; Semantic recognition is performed on the data to be processed to obtain semantic feature vectors; The semantic feature vector is subjected to behavior clustering to obtain behavior pattern identifiers; The behavior pattern identifier is uploaded to the cloud server so that the cloud server can generate intelligent responses based on the behavior pattern identifier.

[0129] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0130] Therefore, embodiments of this application provide a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk, etc. A computer program is stored thereon, and the computer program is loaded by a processor to execute the steps in any of the data processing methods provided in embodiments of this application. For example, the computer program loaded by the processor can execute the following steps: Obtain the data to be processed; Semantic recognition is performed on the data to be processed to obtain semantic feature vectors; The semantic feature vector is subjected to behavior clustering to obtain behavior pattern identifiers; The behavior pattern identifier is uploaded to the cloud server so that the cloud server can generate intelligent responses based on the behavior pattern identifier.

[0131] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the detailed descriptions of other embodiments above, which will not be repeated here.

[0132] In practice, each of the above units or structures can be implemented as an independent entity or can be arbitrarily combined to be implemented as the same or several entities. For the specific implementation of each of the above units or structures, please refer to the previous method embodiments, which will not be repeated here.

[0133] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0134] The data processing method, apparatus, computer device, and computer-readable storage medium provided in the embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A data processing method, characterized by, include: Obtain the data to be processed; The data to be processed is subjected to semantic recognition to obtain a semantic feature vector; The semantic feature vectors are subjected to behavior clustering to obtain behavior pattern identifiers; The behavior pattern identifier is uploaded to a cloud server so that the cloud server can generate a smart response based on the behavior pattern identifier.

2. The method of claim 1, wherein, The step of performing semantic recognition on the data to be processed to obtain a semantic feature vector includes: The semantic encoder deployed locally is invoked to perform semantic recognition on the data to be processed in order to obtain the semantic feature vector, which is used to characterize the semantic information of the data to be processed.

3. The method of claim 1, wherein, The step of performing behavior clustering processing on the semantic feature vector to obtain behavior pattern identifiers includes: The local autoencoder is invoked to compress the semantic feature vector to obtain a low-dimensional semantic feature vector, which is used to characterize features related to behavioral patterns. Clustering is performed on the low-dimensional semantic feature vectors to obtain the behavior pattern identifiers.

4. The method according to any one of claims 1 to 3, characterized in that, Uploading the behavior pattern identifier to a cloud server so that the cloud server can generate an intelligent response based on the behavior pattern identifier includes: The behavior pattern identifier is desensitized to obtain a desensitized behavior pattern identifier; The de-identified behavior pattern identifier is uploaded to the cloud server so that the cloud server can generate an intelligent response based on the de-identified behavior pattern identifier.

5. The method of claim 4, wherein, The process of de-identifying the behavior pattern identifier to obtain a de-identified behavior pattern identifier includes: Obtain the privacy budget for the behavioral pattern identifier; Based on the differential privacy model, the privacy budget is used to inject noise into the behavior pattern identifier to obtain the de-identified behavior pattern identifier.

6. The method of claim 5, wherein, The privacy budget for obtaining the behavior pattern identifier includes: Obtain the frequency value of the behavior pattern corresponding to the behavior pattern identifier; When the frequency value is greater than a first threshold, a first privacy budget is allocated to the behavior pattern identifier; When the frequency value is less than or equal to the first threshold, a second privacy budget is allocated to the behavior pattern identifier, wherein the second privacy budget is greater than the first privacy budget.

7. The method of claim 4, wherein, The cloud server maps and decodes the de-identified behavior pattern identifier based on the abstract semantic library to obtain the behavior pattern corresponding to the data to be processed. The behavior pattern is used to generate the intelligent response. The abstract semantic library is built based on historical behavior patterns.

8. A data processing apparatus, characterized by, include: The acquisition module is used to acquire data to be processed. The processing module is used to perform semantic recognition on the data to be processed in order to obtain a semantic feature vector; The semantic feature vectors are subjected to behavior clustering to obtain behavior pattern identifiers; The transceiver module is used to upload the behavior pattern identifier to the cloud server, so that the cloud server can generate an intelligent response based on the behavior pattern identifier.

9. A computer device, comprising: The computer device includes: One or more processors; Memory; and One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which is loaded by a processor to perform the steps of the method according to any one of claims 1 to 7.

Images