Intelligent application recommendation and aggregation method and device based on scene perception

Abstract

The invention discloses an intelligent application recommendation and aggregation method and device based on scene perception, and relates to the technical field of mobile computing and human-computer interaction. The method comprises the steps of inputting multi-dimensional data of a current user scene into a pre-trained intention recognition model, and outputting a current task type of a user and an associated application function demand; screening application components meeting all application function requirements; constructing a cross-application API bus, and generating a data flow diagram; generating a dynamic aggregation interface through an adaptive layout engine; in the interaction process, data field change is monitored in real time, and preloading and data updating of downstream application components are triggered through an API bus; if the user leaves the current task scene, unused application components are automatically unloaded. By adopting the technical scheme provided by the invention, the task completion steps can be reduced by more than half; the cross-application data synchronization delay is maintained at a hundred millisecond level; and a multi-mode input triggering aggregation service is supported.

Description

Technical Field

[0001] This invention relates to the fields of mobile computing and human-computer interaction technology, and in particular to a method and apparatus for intelligent application recommendation and aggregation based on scene awareness. Background Technology

[0002] Currently, a wide variety of feature-rich smart applications have not only brought convenience to people's work and life, but have also led to significant dependence on them. To better utilize and manage these smart applications, people often use application folders for categorization—users manually create folders and group similar applications (such as "social" or "tools") together for easy access. Sometimes, shortcuts and widgets are created on the desktop to save space. Mini-programs or lightweight applications are also used, running lightweight sub-applications through a super app (such as WeChat). However, regardless of the method, each smart application operates independently, unable to dynamically respond to changes in the scenario, and unable to achieve functional aggregation when multiple applications are involved in a scenario. Their data also exists in isolation; cross-application data cannot flow automatically (e.g., manually copying the address to a map app after booking a hotel). This results in a significant disconnect between smart applications, hindering their interoperability and aggregation, leading to considerable inconvenience for users.

[0003] Therefore, how to achieve automatic data flow between related intelligent applications in specific application scenarios, and how to aggregate multiple related intelligent applications to improve user efficiency and satisfaction across applications, is an important issue that can meet social needs. Summary of the Invention

[0004] In order to solve the problems existing in the prior art, the present invention provides the following technical solution.

[0005] The first aspect of this invention provides a method for intelligent application recommendation and aggregation, comprising:

[0006] S101 acquires multi-dimensional data of the current user scenario through multi-source sensors and user behavior logs, including geographical location, timestamps, calendar events, and application usage sequences;

[0007] S102, input multi-dimensional data into the pre-trained intent recognition model, and output the user's current task type and associated application function requirements;

[0008] S103, filter application components that meet all application function requirements from the local application library and cloud service directory, and extract the input data field D_in and output data field D_out required by each component;

[0009] S104, Construct a cross-application API bus, map the input data field D_in and the output data field D_out to a unified data format, and generate a data flow graph, where the vertices of the data flow graph represent data fields and the edges of the data flow graph represent data transfer relationships between application components;

[0010] S105, based on the data flow diagram and device screen parameters, generates a dynamic aggregation interface through an adaptive layout engine, in which the relevant application components are sorted according to the operation flow topology and the data dependencies are marked.

[0011] S106: During the interaction between the user and the dynamic aggregation interface, the changes in data fields are monitored in real time, and the preloading and data updates of downstream application components are triggered through the API bus.

[0012] S107 If it is detected that the user leaves the current task scenario, unused application components will be automatically uninstalled and resources will be released.

[0013] Preferably, the intent recognition model adopts a multimodal Transformer architecture, including:

[0014] A text encoder for processing natural language data, including calendar events and / or SMS content;

[0015] A spatiotemporal encoder is used to analyze the movement trajectory and time period of a geographical location;

[0016] Apply graph networks to construct Markov transition probabilities for application switching;

[0017] A fusion processor is used to generate a probability distribution of task types through a cross-attention mechanism.

[0018] Preferably, the step of selecting application components from the local application library and cloud service catalog that meet all application functional requirements includes:

[0019] Extract the input and output interfaces of application components based on application permission declarations;

[0020] The semantic similarity calculation is used to match the application's functional requirements with the descriptions of its components.

[0021] Preferably, the generation of the data flow graph includes:

[0022] Create a virtual node for each data field;

[0023] If the D_out of the first application component satisfies the D_in of the second application component, then add an edge pointing from the first application component to the second application component.

[0024] Use topological sorting algorithms to eliminate circular dependencies.

[0025] Preferably, generating the dynamic aggregation interface through the adaptive layout engine includes:

[0026] Calculate the available area matrix based on the device screen size and orientation;

[0027] Decompose the data flow graph into linear, tree, or mesh layout templates;

[0028] The location of application components is optimized using a genetic algorithm to minimize the visual jump distance.

[0029] Preferably, S106 includes:

[0030] When the input data field D_in of an application component is updated, the application component is triggered to execute.

[0031] An incremental update strategy is adopted to redraw the affected interface areas.

[0032] A second aspect of the present invention provides an intelligent application recommendation and aggregation device, comprising:

[0033] The multi-source data acquisition module is used to acquire multi-dimensional data of the current user scenario through multi-source sensors and user behavior logs, including geographical location, timestamps, calendar events and application usage sequences;

[0034] The intent recognition module is used to input multi-dimensional data into a pre-trained intent recognition model and output the user's current task type and associated application function requirements.

[0035] The application component filtering module is used to filter application components that meet all application functional requirements from the local application library and cloud service directory, and extract the input data field D_in and output data field D_out required by each component;

[0036] The data flow graph generation module is used to build a cross-application API bus, mapping the input data field D_in and the output data field D_out to a unified data format and generating a data flow graph. The vertices of the data flow graph represent data fields, and the edges of the data flow graph represent data transfer relationships between application components.

[0037] The dynamic interface rendering module is used to generate a dynamic aggregated interface based on the data flow diagram and device screen parameters through an adaptive layout engine. The relevant application components are sorted according to the operation flow topology and the data dependencies are marked.

[0038] The data monitoring module is used to monitor changes in data fields in real time during user interaction with the dynamic aggregation interface, and trigger the preloading and data update of downstream application components through the API bus.

[0039] The maintenance module is used to automatically uninstall unused application components and release resources when it detects that a user has left the current task scenario.

[0040] Preferably, in the multi-source data acquisition module, differential privacy technology is used to process sensitive information, including:

[0041] Add Laplace noise to the geographic location;

[0042] The application uses k-anonymization of the sequence.

[0043] A third aspect of the present invention provides a memory that stores a plurality of instructions for implementing the intelligent application recommendation and aggregation method as described in the first aspect.

[0044] A fourth aspect of the present invention provides an electronic device, including a processor and a memory connected to the processor, the memory storing a plurality of instructions which can be loaded and executed by the processor to enable the processor to perform the intelligent application recommendation and aggregation method as described in Embodiment 1.

[0045] The beneficial effects of this invention are as follows: This invention provides a scene-aware intelligent application recommendation and aggregation method and device. Through a technical chain of scene awareness, intelligent aggregation, and dynamic collaboration, it solves the core problem of cross-application collaboration. Specifically, it achieves intelligent aggregation of application functions through multimodal intent recognition, cross-application data bus, and dynamic interface generation technology. Dynamic scene awareness technology can identify user intent (such as travel planning and medical consultation) in real time and associate it with relevant applications; cross-application data interoperability technology can break down data barriers between applications and achieve field-level information synchronization (such as time, location, and amount); and dynamic interface generation technology can automatically adapt the interactive layout according to device form (mobile phone, foldable screen, in-vehicle system). Using the technical solution provided by this invention, the number of task completion steps can be reduced by more than half; cross-application data synchronization latency is maintained at the level of hundreds of milliseconds; and multimodal input (voice, gesture, location) triggers aggregation services. It can be widely applied in mobile office, smart healthcare, industrial control, and other fields. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the intelligent application recommendation and aggregation method described in this invention;

[0047] Figure 2 This is a functional structure diagram of the intelligent application recommendation and aggregation device described in this invention. Detailed Implementation

[0048] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0049] The method provided by this invention can be implemented in a terminal environment that may include one or more of the following components: a processor, a memory, and a display screen. The memory stores at least one instruction, which is loaded and executed by the processor to implement the method described in the following embodiments.

[0050] A processor may include one or more processing cores. The processor uses various interfaces and lines to connect various parts of the terminal, and performs various functions and processes data by running or executing instructions, programs, code sets or instruction sets stored in memory, and by calling data stored in memory.

[0051] Memory can include random access memory (RAM) or read-only memory (ROM). Memory can be used to store instructions, programs, code, code sets, or instructions.

[0052] The display screen is used to show the user interface of each application.

[0053] In addition, those skilled in the art will understand that the structure of the terminal described above does not constitute a limitation on the terminal. The terminal may include more or fewer components, or combine certain components, or have different component arrangements. For example, the terminal may also include radio frequency circuits, input units, sensors, audio circuits, power supplies, and other components, which will not be described in detail here.

[0054] Example 1

[0055] like Figure 1 As shown, this embodiment of the invention provides an intelligent application recommendation and aggregation method, including the following steps S101~S107:

[0056] S101 acquires multi-dimensional data of the current user scenario through multi-source sensors and user behavior logs, including geographical location, timestamps, calendar events, and application usage sequences;

[0057] S102, input multi-dimensional data into the pre-trained intent recognition model, and output the user's current task type and associated application function requirements;

[0058] S103, filter application components that meet all application function requirements from the local application library and cloud service directory, and extract the input data field D_in and output data field D_out required by each component;

[0059] S104, Construct a cross-application API bus, map the input data field D_in and the output data field D_out to a unified data format, and generate a data flow graph, where the vertices of the data flow graph represent data fields and the edges of the data flow graph represent data transfer relationships between application components;

[0060] S105, based on the data flow diagram and device screen parameters, generates a dynamic aggregation interface through an adaptive layout engine, in which the relevant application components are sorted according to the operation flow topology and the data dependencies are marked.

[0061] S106: During the interaction between the user and the dynamic aggregation interface, the changes in data fields are monitored in real time, and the preloading and data updates of downstream application components are triggered through the API bus.

[0062] S107 If it is detected that the user leaves the current task scenario, unused application components will be automatically uninstalled and resources will be released.

[0063] In step S101, the multi-source sensors may include a GPS module for acquiring the user's geographical location data, sampling every 5 seconds with an accuracy ≤10 meters. They may also include inertial sensors equipped with an MPU-6050 six-axis gyroscope and accelerometer for detecting the user's device posture (e.g., switching between portrait and landscape modes). Application logs can be hooked into the system API to record application startup or exit times, foreground dwell time, and touch hotspots. In this embodiment, the acquired multi-dimensional data of the current user scenario is preprocessed, including spatiotemporal data alignment and privacy protection. Spatiotemporal data alignment can use the NTP protocol to synchronize the timestamps of each sensor and perform coordinate transformation based on GPS coordinates. Privacy protection can include geographic location fuzzing, such as adding Laplacian noise (ε=0.5) to latitude and longitude to generate fuzzy coordinates like (31.23±0.02°N, 121.47±0.02°E). Sensitive fields (such as ID card numbers) can be transmitted using homomorphic encryption. All data, after anonymization, can be stored in a secure non-pool. This also includes anonymizing the application's usage sequence, which involves hashing the application package name and retaining only the usage duration and switching order.

[0064] In step S102, the intent recognition model may employ a multimodal Transformer architecture, and may include:

[0065] A text encoder is used to process natural language data, including calendar events and / or SMS content; specifically, the text encoder can use the BERT-base model to embed 768 dimensions into text content such as calendar events and SMS messages.

[0066] A spatiotemporal encoder is used to analyze the movement trajectory and time period of a geographic location; specifically, an LSTM network can be used to process location sequences and output 128-dimensional spatiotemporal features. Temporal feature extraction includes converting timestamps into periodic codes.

[0067] The application components utilize graph networks to construct Markov transition probabilities for application switching;

[0068] A fusion processor is used to generate a probability distribution of task types through a cross-attention mechanism.

[0069] During training, the intent recognition model can employ cross-entropy loss and contrastive loss (aimed at enhancing clustering of similar scenes). Data augmentation can involve adding random perturbations to location sequences or performing synonym replacements on the text. As an example, consider the multimodal features of the current user scenario: travel from point A to point B. The text encoder embeds the travel information to point B; the spatiotemporal encoder analyzes the movement trend vector and time period from point A to point B; the application graph network frequently switches between map and browser applications, weighting the "navigation" intent; and the fusion processor outputs a task type of "travel planning" (92% confidence), with the required function set F = {hotel booking, transportation ticketing, attraction navigation}.

[0070] In step S103, application components that meet all application functional requirements are selected from the local application library and cloud service directory, and the input data field D_in and output data field D_out required by each component are extracted, including: extracting the input and output interfaces of the application components according to the application permission declaration; and matching the application functional requirements with the application component descriptions through semantic similarity calculation.

[0071] The application components can include local components and cloud components. Local components are used to parse "AndroidManifest.xml" (a very powerful file on the Android platform that facilitates describing the application's functionality and requirements to Android) and extract the "Activity" (component) data. <intent-filter>"statement( <intent-filter>It's an important part of the AndroidManifest.xml file, defining the types of Intents that an application can respond to. An Intent is a way for components in the Android system to communicate, such as starting one Activity from another, or starting a Service from an Activity. Proper configuration in AndroidManifest.xml... <intent-filter>This ensures that application components can be correctly discovered and invoked. Cloud components are used to obtain RESTful interface parameters through OpenAPI description files (Swagger). In the semantic matching algorithm, BERT similarity can be used to calculate the cosine similarity between functional requirements and component descriptions. IO matching degree is used to verify the compatibility of input and output fields.

[0072] In step S104, a cross-application API bus is constructed, mapping the input data field D_in and the output data field D_out to a unified data format, generating a data flow graph. The vertices of the data flow graph represent data fields, and the edges represent data transfer relationships between application components. Generating the data flow graph may include: creating virtual nodes for each data field; adding an edge pointing from the first application component to the second application component if the D_out of the first application component satisfies the D_in of the second application component; and using a topological sorting algorithm to eliminate circular dependencies. In implementation, the data format can be unified first to eliminate differences in data formats between different applications (such as date, currency, coordinates, etc.). Then, automated field mapping is performed, automatically associating input and output fields based on semantics and data types. Dynamic data flow management is then implemented to respond to data changes in real time, ensuring cross-application data synchronization. Finally, visualization and debugging are completed, generating an interpretable data flow graph that supports developer monitoring and optimization.

[0073] In step S105, a dynamic aggregation interface is generated using an adaptive layout engine based on the data flow diagram and device screen parameters. The relevant application components are sorted according to the operation flow topology, and data dependencies are labeled. Specifically, this can be implemented through the following steps: calculating the available area matrix based on the device screen size and orientation; decomposing the data flow diagram into linear, tree, or mesh layout templates; and optimizing the positions of application components using a genetic algorithm to minimize visual jump distances.

[0074] To integrate multiple applications into a single app, saving desktop space on mobile devices, and enabling these applications to intelligently and dynamically adjust their layout based on the current scenario, an adaptive layout engine can be used. This engine automatically optimizes the arrangement of interface elements based on different screen parameters and the complexity of the data flow graph.

[0075] First, data flow diagrams may contain data dependencies between different applications. For example, in a travel planning scenario, hotel booking, ticket purchase, and attraction navigation require the exchange of information such as location and time. Device screen parameters include screen size, resolution, and orientation (landscape or portrait). The impact of these screen parameters must be considered when implementing the display of the aforementioned application components.

[0076] Adaptive layout engines can automatically arrange application components based on screen space and the node relationships in the data flow graph using layout algorithms, ensuring a smooth user experience and clear data flow visibility. Responsive design dynamically adjusts the layout for different screen sizes and orientations, such as different display methods on mobile phones in portrait or landscape mode, or on tablets. Performance optimization ensures that layout calculations do not cause interface lag, especially when dealing with complex data flow graphs. User interaction design allows user actions in the dynamic interface to be reflected in the data flow graph, such as dragging components and resizing.

[0077] Common layout algorithms include flow layout, grid layout, and force-directed layout. This invention uses the force-directed layout algorithm, which can naturally display the relationships between nodes in a data flow graph. By optimizing the force-directed layout, it can be adapted to different screen sizes and performance requirements of mobile devices.

[0078] In this invention, platform-provided APIs, such as Android's "DisplayMetrics" or iOS's "UIScreen", can be used to obtain information such as the width, height, and pixel density of the device screen in real time, and the size and position of components can be adjusted based on this information.

[0079] When generating a dynamic aggregation interface, the layout engine arranges the positions of components based on the data dependencies between them, while ensuring the visualization of data flow and the intuitiveness of the transmission path. For example, the output of the first component is passed to the second component as its input.

[0080] This invention employs various layout strategies. For example, the main flow uses a flow layout to ensure components are arranged in the order of operation, while complex data dependencies are visualized using a force-directed layout. Simultaneously, hardware acceleration and asynchronous computing are utilized to optimize performance. This approach achieves the following technical effects: when the data flow graph is very complex, it avoids an overly cluttered interface while maintaining user convenience; it enables cross-platform adaptation when different operating systems and devices have different limitations and requirements for layout implementation; and the layout can respond quickly when the user switches device orientation or adjusts window size, avoiding delays.

[0081] The method described above in this invention, through testing on different devices and screen sizes, ensures the stability and adaptability of the layout engine. By closely collaborating the layout engine with the data flow graph construction module, and receiving data flow changes in real time and adjusting the layout accordingly, the correct transmission of data flow and dynamic updates of the interface are ensured.

[0082] In the above method of the present invention, the adaptive layout engine effectively generates a dynamic aggregation interface by comprehensively considering layout algorithms, device adaptation, performance optimization and user interaction, as well as being tightly integrated with the data flow management module, thereby improving the user experience.

[0083] In step S106, during user interaction with the dynamic aggregation interface, changes in data fields are monitored in real time, and preloading and data updates of downstream application components are triggered via the API bus. Specifically, this may include: triggering the execution of an application component when its input data field D_in is updated; and redrawing the affected interface area using an incremental update strategy. In the specific implementation, an observer can be registered for each data field to trigger a notification when the field value changes. The version number can be incremented each time data changes, and downstream components can determine whether an update is needed based on the version number. A change summary can be used to record the changed fields and their old or new values, reducing data transmission volume. In the API bus construction, topics can be categorized according to data fields or component functions. A precise routing strategy can be adopted, that is, determining downstream components based on the edges (E) of the data flow graph. The downstream component preloading strategy can adopt a context prediction model based on user behavior, that is, predicting the next component that may be used by analyzing the current operation sequence. For example, if a user is comparing hotels, the next component may be transportation or attractions. Machine learning models can also be used, such as using RNN or Transformer to predict component loading priority. While asynchronous preloading is in progress, component code packages and static resources (such as icons and templates) can be downloaded in advance. Data preloading is also possible, which involves initiating API requests based on predictions and then caching the results. When updating data, an incremental update mechanism is used, comparing the differences between old and new data and updating only the changed parts. For UI refreshes, a partial refresh strategy can be employed, i.e., component-level updates, re-rendering only the affected component subtrees.

[0084] In step S107, if it is detected that the user has left the current task scenario, unused application components are automatically uninstalled and resources are released. In specific applications, scenario detection can be performed first, and then the decision on whether to uninstall components can be made based on the detection results. Scenario detection can be implemented from two dimensions: location and time. For example, if it is detected that the user has moved 10 kilometers outside the target city (e.g., location B), or the current time has exceeded the task end time, it is considered that the user has left the current task scenario. When uninstalling components, Android's "finishAndRemoveTask()" can be called to destroy the "Activity". After component uninstallation, memory cleanup can be performed, using the mark-and-sweep algorithm to release memory for useless data fields. For example (travel planning scenario): after the user arrives at the hotel in location B (the target location), and the GPS location is stable for more than 1 hour, the system will automatically uninstall the ticketing component (considering the current task scenario is complete) and load the local attraction guide component.

[0085] The technical solution provided by this invention can solve the core problem of cross-application collaboration through a technical chain of scene perception, intelligent aggregation and dynamic collaboration, and can be widely used in mobile office, smart healthcare, industrial control and other fields.

[0086] Among these technologies, multimodal intent recognition, cross-application data bus, and dynamic interface generation enable the intelligent aggregation of application functions. The following example, a travel planning scenario, illustrates this. Example scenario: A user plans to travel to location A over the weekend and needs to complete tasks such as hotel booking, ticket purchase, and route planning.

[0087] Step 1: The device's GPS module displays that the user is located in location A, there is a "travel to location B" event in the calendar, and recent search records include "homestay C". Finally, a multi-dimensional scene data matrix is ​​output.

[0088] Step 2: Input the multi-dimensional scene data obtained in Step 1 into the intent recognition model, and output the task type as "travel planning" with functional requirements F = {hotel booking, transportation ticket purchase, attraction navigation}.

[0089] Step 3, filter components: Hotel API for the first application (travel application) (requires input of date and location), ticketing API for the second application (transportation ticketing application) (requires departure point and destination), and navigation component for the third application (map navigation application) (requires POI name).

[0090] Step 4: Construct a data flow diagram: The "Location B" field is passed from the calendar to the hotel component (first application) and the ticket component (second application); the hotel address is output to the map component (third application).

[0091] Step 5: Generate a vertical flow layout (dynamic aggregation interface): for example, the top is for hotel selection, the middle is for ticket booking, and the bottom is for map navigation. Data fields are visualized by connecting them with lines.

[0092] Step 6: When a user changes their departure date, the ticket search criteria will be automatically updated and the results list will be refreshed.

[0093] Step 7: After the user arrives at location B, the system detects the location change, automatically uninstalls the ticket component from the dynamic aggregation interface, and enhances the local tour guide function.

[0094] Example 2

[0095] like Figure 2 As shown, another aspect of the present invention also includes a functional module architecture that is completely consistent with the aforementioned method flow. That is, the embodiments of the present invention also provide an intelligent application recommendation and aggregation device, including:

[0096] The multi-source data acquisition module 201 is used to acquire multi-dimensional data of the current user scenario through multi-source sensors and user behavior logs, including geographical location, timestamps, calendar events and application usage sequences;

[0097] The intent recognition module 202 is used to input multi-dimensional data into a pre-trained intent recognition model and output the user's current task type and associated application function requirements.

[0098] The application component filtering module 203 is used to filter application components that meet all application function requirements from the local application library and cloud service directory, and extract the input data field D_in and output data field D_out required by each component;

[0099] The data flow graph generation module 204 is used to build a cross-application API bus, map the input data field D_in and the output data field D_out to a unified data format, and generate a data flow graph, where the vertices of the data flow graph represent data fields and the edges of the data flow graph represent data transfer relationships between application components.

[0100] The dynamic interface rendering module 205 is used to generate a dynamic aggregated interface through an adaptive layout engine based on the data flow diagram and device screen parameters. The relevant application components are sorted according to the operation flow topology and the data dependencies are marked.

[0101] The data monitoring module 206 is used to monitor changes in data fields in real time during user interaction with the dynamic aggregation interface, and trigger the preloading and data update of downstream application components through the API bus.

[0102] Maintenance module 207 is used to automatically uninstall unused application components and release resources when it detects that a user has left the current task scenario.

[0103] Furthermore, in the multi-source data acquisition module, differential privacy technology is used to process sensitive information, including:

[0104] Add Laplace noise to the geographic location;

[0105] The application uses k-anonymization of the sequence.

[0106] It's important to note that k-anonymization is a privacy-preserving technique designed to prevent the re-identification of individuals within data using quasi-identifiers. The core objective of this technique is to prevent re-identification, ensuring that each record in the dataset has at least k-1 other records with the exact same attribute value for its quasi-identifier, making it impossible for attackers to pinpoint specific individuals based on background knowledge. This can be achieved through generalization or suppression. For example, generalization can replace data with more ambiguous representations (e.g., replacing age 35 with "30-40 years old", or zip code "10025" with "100**"). Hierarchical generalization (e.g., replacing date with year) or range segmentation (e.g., income range) can also be employed. Suppression involves removing rare or unique attribute values ​​(e.g., removing "occupation: astronaut" which appears only once). In this invention, based on k-anonymization, the diversity of sensitive attributes within each group of data is ensured. Simultaneously, stronger privacy guarantees are provided by adding noise.

[0107] This device can be implemented using the intelligent application recommendation and aggregation method provided in Embodiment 1 above. For the specific implementation method, please refer to the description in Embodiment 1, which will not be repeated here.

[0108] The present invention also provides a memory that stores multiple instructions for implementing the intelligent application recommendation and aggregation method as described in Embodiment 1.

[0109] The present invention also provides an electronic device, including a processor and a memory connected to the processor, the memory storing a plurality of instructions which can be loaded and executed by the processor to enable the processor to perform the intelligent application recommendation and aggregation method as described in Embodiment 1.

[0110] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A method for intelligent application recommendation and aggregation, characterized in that, include: S101 acquires multi-dimensional data of the current user scenario through multi-source sensors and user behavior logs, including geographical location, timestamps, calendar events, and application usage sequences; S102, input multi-dimensional data into the pre-trained intent recognition model, and output the user's current task type and associated application function requirements; S103, filter application components that meet all application function requirements from the local application library and cloud service directory, and extract the input data field D_in and output data field D_out required by each component; S104, Construct a cross-application API bus, map the input data field D_in and the output data field D_out to a unified data format, and generate a data flow graph, where the vertices of the data flow graph represent data fields and the edges of the data flow graph represent data transfer relationships between application components; S105, based on the data flow diagram and device screen parameters, generates a dynamic aggregation interface through an adaptive layout engine, in which the relevant application components are sorted according to the operation flow topology and the data dependencies are marked. S106: During the interaction between the user and the dynamic aggregation interface, the changes in data fields are monitored in real time, and the preloading and data updates of downstream application components are triggered through the API bus. S107 If it is detected that the user leaves the current task scenario, unused application components will be automatically uninstalled and resources will be released.

2. The intelligent application recommendation and aggregation method according to claim 1, characterized in that, The intent recognition model adopts a multimodal Transformer architecture, including: A text encoder for processing natural language data, including calendar events and / or SMS content; A spatiotemporal encoder is used to analyze the movement trajectory and time period of a geographical location; Apply graph networks to construct Markov transition probabilities for application switching; A fusion processor is used to generate a probability distribution of task types through a cross-attention mechanism.

3. The intelligent application recommendation and aggregation method according to claim 1, characterized in that, The application components selected from the local application library and cloud service catalog to meet all application functional requirements include: Extract the input and output interfaces of application components based on application permission declarations; The semantic similarity calculation is used to match the application's functional requirements with the descriptions of its components.

4. The intelligent application recommendation and aggregation method according to claim 1, characterized in that, The generated data flow graph includes: Create a virtual node for each data field; If the D_out of the first application component satisfies the D_in of the second application component, then add an edge pointing from the first application component to the second application component. Use topological sorting algorithms to eliminate circular dependencies.

5. The intelligent application recommendation and aggregation method according to claim 1, characterized in that, The process of generating a dynamic aggregation interface using an adaptive layout engine includes: Calculate the available area matrix based on the device screen size and orientation; Decompose the data flow graph into linear, tree, or mesh layout templates; The location of application components is optimized using a genetic algorithm to minimize the visual jump distance.

6. The intelligent application recommendation and aggregation method according to claim 1, characterized in that, S106 includes: When the input data field D_in of an application component is updated, the application component is triggered to execute. An incremental update strategy is adopted to redraw the affected interface areas.

7. A smart application recommendation and aggregation device, characterized in that, include: The multi-source data acquisition module is used to acquire multi-dimensional data of the current user scenario through multi-source sensors and user behavior logs, including geographical location, timestamps, calendar events and application usage sequences; The intent recognition module is used to input multi-dimensional data into a pre-trained intent recognition model and output the user's current task type and associated application function requirements. The application component filtering module is used to filter application components that meet all application functional requirements from the local application library and cloud service directory, and extract the input data field D_in and output data field D_out required by each component; The data flow graph generation module is used to build a cross-application API bus, mapping the input data field D_in and the output data field D_out to a unified data format and generating a data flow graph. The vertices of the data flow graph represent data fields, and the edges of the data flow graph represent data transfer relationships between application components. The dynamic interface rendering module is used to generate a dynamic aggregated interface based on the data flow diagram and device screen parameters through an adaptive layout engine. The relevant application components are sorted according to the operation flow topology and the data dependencies are marked. The data monitoring module is used to monitor changes in data fields in real time during user interaction with the dynamic aggregation interface, and trigger the preloading and data update of downstream application components through the API bus. The maintenance module is used to automatically uninstall unused application components and release resources when it detects that a user has left the current task scenario.

8. The intelligent application recommendation and aggregation device according to claim 7, characterized in that, In the multi-source data acquisition module, differential privacy technology is used to process sensitive information, including: Add Laplace noise to the geographic location; The application uses k-anonymization of the sequence.

9. A memory, characterized in that, The system stores multiple instructions for implementing the intelligent application recommendation and aggregation method as described in any one of claims 1-6.

10. An electronic device, characterized in that, The system includes a processor and a memory connected to the processor, the memory storing multiple instructions that can be loaded and executed by the processor to enable the processor to perform the intelligent application recommendation and aggregation method as described in any one of claims 1-6.

Images