A method for local search based on intelligent devices, learning machine and recording medium

Abstract

This invention discloses a localized search method, learning machine, and recording medium based on smart devices. The method includes: acquiring application information of various applications installed on the smart device, including preset static resource information and dynamically generated dynamic resource information at runtime; based on the application information, analyzing the functions supported by each application and the corresponding function entry paths through a locally deployed language model to generate function description information; receiving a search request input by the user, and determining the matching target function and its function entry path based on the function description information; providing a guidance entry point for the target function on the display interface, the guidance entry point containing the function entry path, and automatically jumping to the page where the target function is located in response to a trigger operation. This invention achieves fine-grained function-level search and guidance without network access through a combination of static and dynamic information collection and a multi-path fusion mechanism, and has the beneficial effects of real-time updates, personalized adaptation, and privacy security.

Description

Technical Field

[0001] This invention relates to the field of intelligent education technology, specifically providing a localized search method, learning machine, and recording medium based on intelligent devices. Background Technology

[0002] Currently, smart devices such as learning machines, smartphones, and tablets have become important tools in people's daily lives and work. These devices typically have dozens or even hundreds of applications installed, each containing numerous functional modules. With the increasing richness and iterative updates of these applications, users, especially new or elderly users, often find it difficult to quickly locate the specific entry point for the function they need. For example, in the context of learning machines, if a user wants to find specific content such as "children's English picture books" or "mental arithmetic exercises," they often need to click through menus in multiple applications, a cumbersome and inefficient process.

[0003] To solve the above problems, existing technologies mainly adopt the following two methods: One approach is based on static help documentation or user manuals. This method presents the application's features to users in a hierarchical categorization format by pre-installing help documentation on the device or obtaining it online. However, this approach has significant technical drawbacks: First, the help documentation is rather coarse-grained, typically only guiding users to the application level or core functions, failing to delve into second- or third-level pages or specific functional modules within the application. Second, the help documentation's updates lag far behind application feature iterations; when the application version is updated or new features are added, the help documentation often fails to keep pace, leaving users without the latest feature guidance information.

[0004] Another approach is a search solution based on cloud-based voice assistants. This solution uploads the user's voice request to a cloud server, where semantic analysis is performed and search results are returned. However, this solution also has limitations: firstly, cloud-based voice assistants typically can only search publicly available information on the internet or pre-configured skill libraries, and cannot access the internal page structure of the device's local applications or dynamically generated content in real time; secondly, this solution heavily relies on a network connection, cannot be used offline, and poses a security risk of uploading user privacy data. Furthermore, cloud-based voice assistants have limited ability to recognize internal application functions, making it difficult to accurately guide users to specific function pages.

[0005] In view of this, this invention patent is hereby proposed. Summary of the Invention

[0006] To address the shortcomings of the existing technologies, this invention proposes a localized search method, learning machine, and recording medium based on smart devices. By collecting static resource information and runtime dynamic resource information of locally installed applications on the device, and using locally deployed language models to analyze the functions and entry paths of each application, it achieves localized function search and guidance that is offline, fine-grained, and updated in real time, thereby improving user operating efficiency and device usage experience.

[0007] Specifically, the following technical solution was adopted: In a first aspect, the present invention provides a localized search method based on smart devices, comprising: Obtain application information of various applications installed on a smart device, including preset static resource information and dynamically generated dynamic resource information at runtime; Based on the application information, the functions supported by each application and the corresponding function entry paths are analyzed by a language model deployed locally, and function description information is generated. Receive a search request input by the user, and determine the target function and its entry path that match the search request based on the function description information; The smart device provides a guide entry point for the target function on its display interface. The guide entry point contains the function entry path and automatically redirects to the page containing the target function in response to the user's trigger operation.

[0008] As an optional embodiment of the present invention, in a localized search method based on a smart device, the step of obtaining application information of various applications installed on the smart device includes: Scan the application list in the smart device using system permissions to obtain the identification information of each application; Based on the identification information, the static resource information is extracted from the system installation package directory, resource folder, or application sandbox. The static resource information includes at least one of the following: application name, version number, permission declaration, interface layout file, preset text, and menu hierarchy structure. The dynamic resource information is obtained by monitoring the application's runtime behavior data, intercepting system logs, or reading the runtime database. The dynamic resource information includes at least one of the following: user behavior trajectory, dynamically downloaded content resources, real-time generated interface elements, and temporary cache data.

[0009] As an optional embodiment of the present invention, in a localized search method based on smart devices, based on the application information, the functions supported by each application and the corresponding function entry paths are analyzed by a locally deployed language model to generate function description information, including: Based on the static resource information, the preset interface hierarchy and function entry points of each application are parsed, and a first set of function paths is extracted. The first set of function paths reflects the standard access paths at the application design level. Based on the dynamic resource information, the operation sequence and frequency distribution of users accessing each function during actual use are identified, and a second function path set is extracted. The second function path set reflects the actual access path at the user's habit level. The first functional path set and the second functional path set are merged, multiple candidate paths are associated with the same function, and a confidence weight is assigned to each candidate path, wherein the path corresponding to the dynamic resource information has a higher initial confidence. The fused functional path information is input into the language model to generate functional description information that includes functional name, functional description, multiple candidate paths and their confidence weights.

[0010] As an optional embodiment of the present invention, in a localized search method based on a smart device, determining the target function and its entry path matching the search request based on the function description information includes: The search request is semantically parsed and intent is identified to generate a query vector; Calculate the semantic similarity between the query vector and each function in the function description information, and determine one or more candidate functions with the highest similarity; For each candidate function, the path with the highest confidence score is selected from its multiple associated candidate paths based on the confidence score weight as the recommended entry path for that function. The candidate functions are ranked comprehensively based on their similarity scores and the confidence weights of the recommended entry paths. The function with the highest ranking is selected as the target function and its corresponding recommended entry path.

[0011] As an optional embodiment of the present invention, in a localized search method based on smart devices, when the target function matching the search request cannot be determined, or the similarity scores of all candidate functions are lower than a preset threshold, fuzzy matching suggestions are provided to the user or a list of alternative functions generated by the language model is displayed. Record user feedback behavior on the list of alternative functions, including user clicks to select or abandon, and use the feedback behavior as training data for subsequent optimization and adjustment of the language model, as well as for adjusting the confidence weights of the corresponding functions.

[0012] As an optional embodiment of the present invention, in a localized search method based on a smart device, the application information further includes the user's historical behavior data and preference settings, and the localized search method includes: Based on the historical behavior data and preference settings, the confidence weights of each path in the second functional path set are dynamically adjusted, specifically including: Get the user's historical usage frequency for each path fi and most recent usage time ti ; According to the preset time decay factor λ and frequency weighting coefficient k Calculate the adjustment factor , where Δ ti The time interval between the current time and the most recent usage time; The initial confidence weight of the path w 0 i Multiply by the adjustment factor αi The adjusted confidence weights are obtained. wi = w 0 i · αi ; When merging the first functional path set and the second functional path set, the adjusted confidence weights are used. wi As the confidence weight of each path in the second functional path set, the overall confidence of each candidate path is recalculated.

[0013] As an optional embodiment of the present invention, a localized search method based on smart devices includes: Real-time monitoring of changes in the status of applications installed on smart devices, including applications being added, uninstalled, or updated. When a change in the application status is detected, the application information of the application is re-collected and the function description information is updated to maintain the real-time performance of the local function index library.

[0014] As an optional embodiment of the present invention, in a localized search method based on a smart device, a guidance entry for the target function is provided on the display interface of the smart device, including: The target function is displayed in the form of a card, which includes the name, brief description, and jump button. The card is either displayed floating on the top layer of the current interface or embedded in the search results list. In response to the user's triggering of the jump button, the system simulates click operations or sends Intent commands sequentially according to the recommended entry path corresponding to the target function, and automatically navigates to the in-app page where the target function is located.

[0015] In a second aspect, the present invention provides a learning machine employing the localized search method, comprising: The application information acquisition module acquires the application information of each installed application, including preset static resource information and dynamically generated dynamic resource information at runtime. The function description information generation module, based on the application information, analyzes the functions supported by each application and the corresponding function entry paths through a language model deployed locally, and generates function description information. The search request receiving module receives the search request input by the user and, based on the function description information, determines the target function and its function entry path that match the search request. The guidance entry module provides a guidance entry for the target function on the display interface of the smart device. The guidance entry includes the function entry path and automatically jumps to the page where the target function is located in response to the user's trigger operation.

[0016] In a third aspect, the present invention provides a computer-readable recording medium storing a computer-executable program, characterized in that, when the computer-executable program is executed, it implements the localized search method based on a smart device.

[0017] The present invention achieves the following beneficial technical effects through the above technical solution: (a) Fine-grained localization functions are directly accessible.

[0018] This invention, by collecting static resource information (such as interface layout files and menu hierarchy) and dynamic resource information (such as user behavior patterns and real-time generated interface elements) of locally installed applications, can deeply analyze specific functional pages at the second, third, and even deeper levels within the application, rather than simply remaining at the application or core function level. After a user enters a search request, the system can directly guide them to the specific page containing the target function, achieving a breakthrough in granularity from "application-level search" to "function-level search," significantly improving the efficiency and accuracy of users reaching their target functions.

[0019] (ii) A multi-path fusion mechanism that combines dynamic and static approaches.

[0020] This invention innovatively integrates the "standard path at the design level" extracted from static resource information with the "actual path at the user habit level" identified from dynamic resource information, associating multiple candidate paths for the same function and assigning confidence weights. This mechanism ensures both basic coverage of function entry points (static paths) and reflects actual user habits (dynamic paths). Especially when the application interface is redesigned or function entry points change, dynamic paths can effectively compensate for the lag of static paths, ensuring the accuracy and usability of the guidance entry points.

[0021] (iii) Personalized dynamic adjustment of confidence level.

[0022] This invention further incorporates user historical behavior data and preference settings, using a time decay function. The confidence weights of dynamic paths are adjusted individually. This mechanism assigns higher weights to frequently used and recently used paths, while the weights of long-unused paths gradually decrease, thus achieving adaptive and personalized search result ranking. As users accumulate usage time, the system can predict user intent more and more accurately, providing a personalized user experience.

[0023] (iv) Real-time self-learning ability.

[0024] This invention achieves dynamic updates to the functional index and continuous optimization of the language model by real-time monitoring of application status changes (addition, uninstallation, version updates) and recording user feedback behavior. When the application version iterates or users exhibit new usage habits, the system can automatically trigger information re-collection and confidence weight adjustment, maintaining the timeliness and accuracy of guidance information without manual intervention. Simultaneously, user clicks and selections can serve as training data to feed back into the model, forming a virtuous cycle of "use-feedback-optimization".

[0025] (v) Offline availability and privacy security.

[0026] In this invention, all information collection, semantic parsing, path matching, and redirection guidance are completed locally on the device, without the need for an internet connection. This completely eliminates dependence on a network environment, ensuring that users can still access complete functional search services offline. Furthermore, since user behavior data and search requests do not need to be uploaded to cloud servers, the risk of privacy leaks is effectively avoided, aligning with current technological trends in user data security protection.

[0027] (vi) Reduce maintenance costs and improve equipment intelligence.

[0028] For dedicated smart devices such as learning machines, this invention eliminates the need to configure separate help documents or voice skills for each application. Instead, it provides a unified, intelligent function guidance service through system-level automatic data collection and analysis capabilities. This not only significantly reduces the manpower costs of application development and maintenance but also enables the device to automatically adapt to unknown applications or new versions, fundamentally improving the device's intelligence level and user experience. Attached Figure Description

[0029] Figure 1 A flowchart of a localized search method based on smart devices according to an embodiment of the present invention; Figure 2 A schematic diagram of the structure of the electronic device according to an embodiment of the present invention; Figure 3 A schematic diagram of a computer-readable recording medium according to an embodiment of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0031] Therefore, the following detailed description of embodiments of the present invention is not intended to limit the scope of the claimed invention, but merely illustrates some embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0032] It should be noted that, unless otherwise specified, the embodiments and features and technical solutions in the present invention can be combined with each other.

[0033] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0034] In the description of this invention, it should be noted that the terms "upper," "lower," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of this invention is in use, or the orientation or positional relationship commonly understood by those skilled in the art. These terms are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention. In addition, the terms "first," "second," etc., are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0035] First, it should be noted that the smart devices involved in this invention, such as learning machines, smartphones, and tablets, typically have operating systems (e.g., Android, iOS, HarmonyOS) and can install and run multiple applications. These applications include pre-installed static resources (e.g., layout files, configuration files, pre-installed text in the installation package) and dynamically generated resources at runtime (e.g., user behavior data, real-time downloaded content, cached data). This invention utilizes system permissions to collect this information, analyzes it through a locally deployed language model, and provides users with intelligent function search and guidance services. Example

[0036] like Figure 1 As shown, this embodiment provides a localized search method based on smart devices, including the following steps: Step S101: Obtain application information of each application installed in the smart device. The application information includes preset static resource information and dynamically generated dynamic resource information at runtime.

[0037] Specifically, when a smart device is first started or during routine maintenance, the system scans all installed applications on the device using system permissions to obtain the identification information of each application (such as package name, application name, version number, etc.). Based on this identification information, the system extracts static resource information from the system installation package directory, resource folder, or application sandbox, including but not limited to: application name, version number, permission declaration, interface layout files (such as XML layout files in Android, XIB or Storyboard files in iOS), pre-defined text (such as text content in strings.xml or Localizable.strings), and menu hierarchy (such as navigation drawers, tab bars, list menus, etc.). This static resource information reflects the application's pre-defined functional entry points and interface structure at the design level.

[0038] Meanwhile, the system obtains dynamic resource information by monitoring application runtime behavior data, intercepting system logs, or reading runtime databases. Specific implementation methods include: utilizing the operating system's AccessibilityService to monitor user clicks, swipes, and other actions; intercepting event distribution within the application using Hook technology; or periodically reading database files (such as SQLite databases) and cache files generated during application runtime. Dynamic resource information includes at least one of the following: user behavior trajectories (such as the complete sequence of actions a user takes from the homepage to a specific function), dynamically downloaded content resources (such as online courses and e-books), real-time generated interface elements (such as dynamically generated recommendation cards based on user profiles), and temporary cached data (such as recent browsing history). This dynamic resource information reflects the user's actual access paths and behavioral habits during actual use.

[0039] Step S102: Based on the application information, analyze the functions supported by each application and the corresponding function entry paths through a language model deployed locally, and generate function description information.

[0040] This step is one of the core components of the present invention. The system inputs the collected application information into a locally deployed language model for analysis. Specifically, this step further includes: First, based on static resource information, the system analyzes the preset interface hierarchy and functional entry points of each application to extract the first set of functional paths. For example, for a reading application, by analyzing its interface layout file, the system can discover preset hierarchical paths such as "Bookshelf" → "Categories" → "Science Fiction" → "The Three-Body Problem," and identify the Activity or ViewController corresponding to each level as the functional entry point. The first set of functional paths reflects the standard access paths at the application design level.

[0041] Secondly, based on dynamic resource information, the system identifies the operation sequences and frequency distribution of users accessing various functions during actual use, and extracts a set of secondary function paths. For example, by analyzing the operation logs of a large number of users, the system may find that most users actually enter the reading page of "The Three-Body Problem" by clicking "search box" → "enter 'The Three-Body Problem'" → "click search results," rather than clicking through the preset menu hierarchy. The set of secondary function paths reflects the actual access paths at the user's habitual level.

[0042] Then, the first set of functional paths and the second set of functional paths are merged, and multiple candidate paths are associated with the same function, with each candidate path assigned a confidence weight. The fusion strategy can employ weighted averaging, voting mechanisms, or machine learning methods. This embodiment uses a heuristic fusion strategy: for the same function, if the static path and the dynamic path point to the same entry point, a higher confidence weight is assigned; if there are differences, the path corresponding to the dynamic resource information has a higher initial confidence weight because it reflects the user's actual usage habits. For example, the preset path "Settings" → "Account and Security" → "Change Password" might be assigned an initial confidence weight of 0.6, while the dynamic path "Personal Center" → "Security Center" → "Change Password" (a path that users actually use frequently) might be assigned an initial confidence weight of 0.9.

[0043] Finally, the fused functional path information is input into the language model to generate functional description information including functional name, functional description, multiple candidate paths and their confidence weights. The language model can be a lightweight pre-trained model, such as BERT-Tiny, MobileBERT, or TinyBERT, deployed in the local inference engine of the smart device (such as TFLite, CoreML, MNN, etc.), and can periodically obtain model parameter updates through differential updates. The differential updates only transmit parameter data that has changed compared to the previous version, in order to save network traffic and storage space.

[0044] Step S103: Receive the user's search request and, based on the function description information, determine the target function and its entry path that match the search request.

[0045] Users can input search requests in various ways, including but not limited to: receiving keywords or natural language queries through the system search box; receiving voice commands through the voice input interface and converting the voice commands into text search requests; and receiving user intent regarding the current interface through screenshot recognition or image recognition and converting the intent into search requests. For example, if a user long-presses on a piece of text on the reading interface, the system recognizes that the user may want to search for the definition or related knowledge points of that text and automatically converts it into a search request such as "search for the meaning of XX word".

[0046] After receiving a search request, the system performs semantic parsing and intent recognition to generate a query vector. Specifically, this can be achieved by inputting the search request text into the language model to obtain its semantic vector representation. Then, the system calculates the semantic similarity (e.g., cosine similarity, Euclidean distance, or dot product) between the query vector and each function in the function description information, identifying one or more candidate functions with the highest similarity.

[0047] For each candidate function, the path with the highest confidence score among its associated candidate paths is selected as the recommended entry path based on the confidence score weight. For example, for the "Change Password" function, if candidate path A has a confidence score of 0.6 and candidate path B has a confidence score of 0.9, then candidate path B is selected as the recommended entry path.

[0048] The candidate functions are ranked comprehensively based on their similarity scores and the confidence weights of the recommended entry paths. The highest-ranked function and its corresponding recommended entry path are selected. The comprehensive ranking can be calculated as follows: Comprehensive Score = α × Semantic Similarity + β × Path Confidence, where α and β are preset weight coefficients (e.g., α = 0.6, β = 0.4), which can be adjusted according to the actual application scenario.

[0049] Step S104: Provide a guide entry for the target function on the display interface of the smart device. The guide entry includes the function entry path and automatically jumps to the page where the target function is located in response to the user's trigger operation.

[0050] The specific implementation method is as follows: Figure 3 As shown: The target function is displayed in card format, including its name, brief description, and jump button. The card can be displayed floating on top of the current interface (e.g., a floating window) or embedded in the search results list (e.g., a list item). The card design should be concise and clear, including the function name (e.g., "Change Password"), a brief function description (e.g., "Go to Security Center to change your login password"), and a prominent jump button (e.g., "Go Now").

[0051] In response to the user's triggering of the jump button, the system simulates click operations or sends Intent commands sequentially according to the recommended entry path corresponding to the target function, automatically navigating to the in-app page where the target function is located. For example, in the Android system, an Intent containing the target Activity can be constructed, and the startActivity() method can be used to jump directly; if the target function is located in a deeper page within the application and requires simulating multiple clicks, accessibility services can be used to simulate click events, triggering interface jumps sequentially according to the path sequence. The entire process is seamless for the user, as if the system directly opens the required function for the user. Example

[0052] This embodiment further refines and expands the steps for obtaining application information based on Embodiment 1. Step S101 specifically includes: Step S201: Scan the application list in the smart device using system permissions to obtain the identification information of each application. The system calls system APIs such as PackageManager (Android) or LSApplicationWorkspace (iOS) to obtain information such as package name, application name, version number, and installation time of all installed applications.

[0053] Step S202: Based on the identification information, extract the static resource information from the system installation package directory, resource folder, or application sandbox. Specifically: In the Android system, you can obtain the application's resource path through PackageManager, and then parse the AndroidManifest.xml file in the APK file to obtain information such as permission declarations and Activity declarations; you can also obtain string resources, layout resources, etc. through the AAPT tool or by directly parsing the resources.arsc file.

[0054] In iOS, you can use NSBundle to get the application's resource directory, parse the Info.plist file to get permission declarations, read the Localizable.strings file to get localized text, and parse the NIB or Storyboard file to get the UI structure.

[0055] Static resource information includes, but is not limited to, at least one of the following: application name, version number, permission declaration, interface layout file, preset text, and menu hierarchy structure.

[0056] Step S203 involves obtaining the dynamic resource information by monitoring application runtime behavior data, intercepting system logs, or reading the runtime database. Specific implementation methods include: Utilize accessibility services: Register an AccessibilityService, listen for window state changes and gesture operations, record the user's operation sequence (such as click, swipe, long press, etc.), and associate it with the corresponding interface elements and functions.

[0057] Utilize logging systems: Intercept application log output through Logcat (Android) or system logs (iOS) to extract key behavioral information.

[0058] Reading runtime databases: Many applications store user data in SQLite databases. With authorization, the system can read these database files to obtain information such as the user's browsing history, favorites, and recent usage.

[0059] Dynamic resource information includes, but is not limited to, at least one of the following: user behavior trajectories (such as the click sequence of "Homepage → Categories → Science Fiction → The Three-Body Problem"), dynamically downloaded content resources (such as offline courses and e-books downloaded by users), real-time generated interface elements (such as personalized cards generated according to recommendation algorithms), and temporary cached data (such as recent search records).

[0060] By combining static and dynamic information collection methods, the system can comprehensively grasp the functional layout of the application and the user's usage habits, laying a solid data foundation for subsequent intelligent analysis. Example

[0061] This embodiment, based on Embodiment 1, further refines and expands the process of generating functional description information. Step S102 specifically includes: Step S301: Based on the static resource information, parse the preset interface hierarchy and functional entry points of each application, and extract the first set of functional paths. The system constructs the application's interface tree by analyzing the application's interface layout file and navigation structure, and identifies the functional labels and jump relationships of each interface node. For example, for a news application, the system may identify a preset path such as "Homepage" → "News Channel" → "Technology News" → "Details Page", and record the interface class name and jump Intent corresponding to each level.

[0062] Step S302: Based on the dynamic resource information, identify the operation sequence and frequency distribution of users accessing various functions during actual use, and extract the second function path set. The system performs cluster analysis on a large number of user operation logs to discover high-frequency operation path patterns. For example, the system may find that 70% of users enter the details page through the method of "search box" → "enter keywords" → "click search results", instead of following the preset path of "homepage → channel list → news list → details page". The system uses these high-frequency paths as the second function path set and counts the usage frequency of each path.

[0063] Step S303: Merge the first functional path set and the second functional path set, associating multiple candidate paths for the same function, and assigning a confidence weight to each candidate path. The fusion strategy can be implemented in the following ways: For paths in the first functional path set, assign a basic weight W_static (e.g., 0.5). For paths in the second function path set, a weight W_dynamic = base_weight × (1 + log(frequency)) is assigned based on the frequency of use, where base_weight is the preset base weight (e.g., 0.3) and frequency is the frequency of use of the path. For paths that appear in both sets, the weights are combined: W_combined = W_static + W_dynamic; For multiple candidate paths with the same function, normalization is performed so that the sum of the confidence weights of all paths is 1.

[0064] In this embodiment, the path corresponding to the dynamic resource information has a higher initial confidence level because it reflects the user's actual usage habits and can better guarantee the actual availability of the guidance entry.

[0065] Step S304: Input the fused functional path information into the language model to generate functional description information including a functional name, functional description, multiple candidate paths, and their confidence weights. The language model can adopt a sequence-to-sequence (Seq2Seq) generation architecture, with input being functional path information (such as path node sequences and weights) and output being functional description text in natural language form. For example, inputting ["Settings", "Security Center", "Change Password", weight 0.9] and ["Personal Center", "Account and Security", "Change Password", weight 0.6], the model may generate the functional name "Change Password", the functional description "You can go to the Security Center or the Account and Security page to change your login password", and associate two candidate paths with their confidence weights. Example

[0066] This embodiment, based on Embodiment 3, further refines and expands the process of determining the target function and its entry path. Step S103 specifically includes: Step S401: Semantic parsing and intent recognition are performed on the search request to generate a query vector. The system inputs the user's search request text (such as "how to change password") into the language model to obtain its semantic vector representation V_query. This vector is typically a high-dimensional floating-point vector (such as 256-dimensional or 512-dimensional) that can capture the semantic information of the query.

[0067] Step S402: Calculate the semantic similarity between the query vector and each function in the function description information, and determine one or more candidate functions with the highest similarity. The system pre-generates a function vector V_func for each function (which can be obtained by inputting the function name and function description into a language model). Calculate the cosine similarity between V_query and each V_func: similarity = cos(V_query, V_func) = (V_query·V_func) / (|V_query|×|V_func|). Select functions with similarity exceeding a preset threshold (e.g., 0.7) or ranking in the top K (e.g., K=5) as candidate functions.

[0068] Step S403: For each candidate function, select the path with the highest confidence score from its associated candidate paths based on the confidence score weight as the recommended entry path for that function. For example, for the candidate function "Change Password", its associated candidate paths include: path A (confidence score 0.9), path B (confidence score 0.6), and path C (confidence score 0.4), then path A is selected as the recommended entry path.

[0069] Step S404: Based on the similarity scores of candidate functions and the confidence weights of recommended entry paths, a comprehensive ranking is performed, and the highest-ranked function and its corresponding recommended entry path are selected. The comprehensive ranking can be calculated using a weighted product or a weighted sum: Method 1 (weighted sum): score = λ × similarity + μ × confidence, where λ + μ = 1; Method 2 (weighted product): score = similarity^α × confidence^β, where α and β are exponential parameters.

[0070] In this embodiment, the preferred method is to set λ=0.7 and μ=0.3, which focuses on semantic matching while also taking into account the reliability of the path. Example

[0071] This embodiment adds a user feedback mechanism to the previous one, and after step S103, it also includes: Step S501: When a target function matching the search request cannot be determined, or the similarity score of all candidate functions is lower than a preset threshold (e.g., 0.5), the system provides the user with fuzzy matching suggestions or displays a list of alternative functions generated by the language model. For example, the system can display relevant function suggestions such as "Do you want to find the following functions:", "Change Password", "Forgot Password", "Reset Password".

[0072] Step S502: Record the user's feedback behavior on the list of alternative functions, including the user's click to select or abandon the function, and use the feedback behavior as training data for subsequent optimization and adjustment of the language model, as well as for adjusting the confidence weight of the corresponding function.

[0073] Specifically, if a user selects a function from the list of options, it means that the function is relevant to the user's query intent. The system can increase the association strength between the function and the current query keywords and appropriately increase the confidence weight of the path corresponding to the function. If the user abandons the operation (such as closing the list of options or returning), it means that the suggested options are irrelevant. The system can record this query as a "missed" case for subsequent model improvement.

[0074] Through this feedback mechanism, the system can continuously learn and optimize, forming a virtuous cycle of "use-feedback-optimization". As usage time accumulates, search accuracy and user satisfaction will continue to improve. Example

[0075] This embodiment, based on Embodiment 3, introduces a user personalization mechanism to dynamically adjust the confidence weight. After step S102, it further includes: Step S601: Obtain the user's historical behavior data and preference settings. The system analyzes the user's operation records over a period of time (e.g., 30 days) to statistically analyze the user's behavioral characteristics such as the frequency, duration, and time of use of each function. Simultaneously, it reads the user's explicitly specified preference settings in the system settings (e.g., shortcuts to "frequently used functions," theme preferences, etc.).

[0076] Step S602: Based on the historical behavior data and preference settings, dynamically adjust the confidence weight of each path in the second functional path set. The specific adjustment method is as follows: Get the user's historical usage frequency for each path fi and most recent usage time ti ; According to the preset time decay factor λ and frequency weighting coefficient k Calculate the adjustment factor , where Δ ti The time interval (in days) between the current time and the most recent time of use. The initial confidence weight of the path w 0 i Multiply by the adjustment factor αi The adjusted confidence weights are obtained. wi = w 0 i · αi .

[0077] Example parameter settings: Time decay factor λ=0.1, meaning the weight decays by approximately 63% every 10 days; Frequency weight coefficient k=0.05, meaning the weight increases by 5% each time it is used. This ensures the priority of high-frequency paths while also reflecting changes in usage habits over time.

[0078] Step S603: When merging the first functional path set and the second functional path set, the adjusted confidence weights are used. wi As the confidence weight of each path in the second functional path set, the overall confidence of each candidate path is recalculated.

[0079] Through these personalized adjustments, the system can provide differentiated guidance services for different users. For example, for users who prefer to use the search function to access content, the search path will receive higher weight; for users who are used to accessing content through menus, the menu path will maintain a higher recommendation priority. Example

[0080] This embodiment, based on Embodiment 1, adds a real-time update mechanism and also includes: Step S701: Monitor the status changes of installed applications on the smart device in real time, including the addition, uninstallation, or version update of applications. The system monitors application status changes by registering application installation / uninstallation broadcasts (such as ACTION_PACKAGE_ADDED, ACTION_PACKAGE_REMOVED, ACTION_PACKAGE_REPLACED in Android) or by periodically scanning the application list.

[0081] Step S702: When a change in the application status is detected, the application information of the application is re-collected and the function description information is updated to maintain the real-time performance of the local function index library.

[0082] Specifically: When a new application is installed, the system immediately collects information and analyzes its functions, adding its function information to the local function index library. When an application is uninstalled, the system removes the application's related functional information from the index. When an application is updated, the system re-collects the application's static resource information (because version updates may cause changes in the interface structure), while retaining the original dynamic resource information (unless user data is cleared), and performs a second fusion analysis to update the function description information.

[0083] Through this real-time update mechanism, the system can always reflect the latest application status on the device, ensuring the accuracy and timeliness of the guidance entry. Example

[0084] This embodiment, based on embodiment four, further refines and expands the method of providing the guidance entry point. Step S104 specifically includes: Step S801: Display the name, brief description, and jump button of the target function in the form of a card. The card is displayed floating on the top layer of the current interface or embedded in the search results list.

[0085] Card design can employ the following technical solutions: Floating Cards: These cards are created at the top level of the system using WindowManager (Android) or UIWindow (iOS), overlaying the current application interface. Each card contains a function icon, a function name (e.g., "Change Password"), a brief description (e.g., "Go to Security Center to change your login password"), and a prominent jump button (e.g., a green "Go Now" button). Cards can be dragged and closed; users can move them anywhere on the screen or hide them by clicking the close button.

[0086] Embedded Card: In the system search application's search results list, the feature's entry point is displayed as a special search result, alongside other application search results. The card style is consistent with the search results list style, including the feature name, the name of the application it belongs to, and a jump button.

[0087] Step S802: In response to the user's triggering operation on the jump button, simulate click operations or send Intent commands in sequence according to the recommended entry path corresponding to the target function, and automatically navigate to the in-application page where the target function is located.

[0088] Specific implementation methods include: Intent redirection: If the entry path corresponding to the target function can be directly reached via an Intent (e.g., an Activity has a public Intent Filter), the system constructs the corresponding Intent and calls the startActivity() method to directly redirect. For example, for the "Change Password" function in the "Settings" app, you can send Intent.ACTION_SETTINGS with the corresponding parameters.

[0089] Accessibility Simulation: If the target function is located deep within the application and there is no publicly available Intent entry point, the system uses accessibility services to simulate user actions. The specific process is as follows: The target application is launched, and based on the node information in the path sequence, clickable elements on the interface are searched sequentially (using attributes such as View ID, text content, or ContentDescription), and then click events are simulated. After each step, the system waits for the interface to load completely (by listening for changes in window state) before proceeding to the next step, until the target function page is reached. Example

[0090] This embodiment provides a learning machine that employs the localized search method described in any one of embodiments one to eight. The learning machine includes: The application information acquisition module is used to acquire application information of various applications installed in the learning machine. This application information includes preset static resource information and dynamically generated resource information at runtime. For specific implementation details of this module, please refer to step S101 in Embodiment 1 and Embodiment 2.

[0091] The function description information generation module is used to generate function description information based on the application information, by analyzing the functions supported by each application and the corresponding function entry paths through a locally deployed language model. For specific implementation of this module, please refer to step S102 in Embodiment 1 and Embodiment 3.

[0092] The search request receiving module is used to receive search requests input by the user and, based on the function description information, determine the target function matching the search request and its function entry path. For a specific implementation of this module, please refer to step S103 in Embodiment 1 and Embodiments 4 to 6.

[0093] The guidance entry module is used to provide a guidance entry for the target function on the display interface of the learning machine. The guidance entry includes the function entry path and automatically jumps to the page where the target function is located in response to the user's trigger operation. For the specific implementation of this module, please refer to step S104 in Embodiment 1 and Embodiment 8.

[0094] The update module is used to monitor the status changes of applications installed on the learning machine in real time, including the addition, uninstallation, or version update of applications; when a change in the application status is detected, it triggers the re-collection of application information and the updating of function description information of the application. See Example 7 for a detailed implementation of this module.

[0095] The personalized adjustment module is used to dynamically adjust the confidence weight of each path in the second set of functional paths based on the user's historical behavior data and preference settings. See Example Six for a detailed implementation of this module.

[0096] The feedback optimization module records user feedback on the list of alternative functions and uses this feedback as training data for subsequent optimization and adjustment of the language model, as well as for adjusting the confidence weights of corresponding functions. See Example 5 for a detailed implementation of this module.

[0097] The modules described above can be integrated into the learning machine's operating system and run as system-level services; alternatively, they can exist as independent applications, accessing necessary information through system permissions. The modules communicate and exchange data with each other through defined interfaces. Example

[0098] Figure 2 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. The electronic device includes a processor and a memory. The memory is used to store a computer-executable program. When the computer program is executed by the processor, the processor executes the localized search method described in any one of embodiments one to eight.

[0099] like Figure 2 As shown, the electronic device is embodied in the form of a general-purpose computing device. There can be one or more processors working collaboratively. This invention also does not preclude distributed processing, meaning that processors can be distributed across different physical devices. The electronic device of this invention is not limited to a single entity, but can also be the sum of multiple physical devices.

[0100] The memory stores a computer-executable program, typically machine-readable code. The computer-readable program can be executed by the processor to enable the electronic device to perform the method of the present invention, or at least some steps of the method.

[0101] The memory includes volatile memory, such as random access memory (RAM) and / or cache memory, and may also be non-volatile memory, such as read-only memory (ROM).

[0102] Optionally, in this embodiment, the electronic device further includes an I / O interface for exchanging data with external devices. The I / O interface can represent one or more of several bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0103] It should be understood that Figure 2The electronic device shown is merely one example of the present invention, and the electronic device of the present invention may also include elements or components not shown in the above examples. For example, some electronic devices also include display units such as displays, and some electronic devices also include human-computer interaction elements such as buttons and keyboards. Any electronic device capable of executing a computer-readable program in memory to implement the method of the present invention or at least some steps of the method can be considered as an electronic device covered by the present invention.

[0104] Figure 3 This is a schematic diagram of a computer-readable recording medium according to an embodiment of the present invention. Figure 3 As shown, a computer-readable recording medium stores a computer-executable program, which, when executed, implements the localized search method described in any one of embodiments one through eight of the present invention. The computer-readable recording medium may include a data signal propagated in baseband or as part of a carrier wave, carrying readable program code. This propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The readable recording medium may also be any readable medium other than a readable recording medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the readable recording medium can be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0105] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0106] From the above description of the embodiments, those skilled in the art will readily understand that the present invention can be implemented by hardware capable of executing specific computer programs, such as the system of the present invention, and the electronic processing unit, server, client, mobile phone, control unit, processor, etc. included in the system. The present invention can also be implemented by computer software executing the methods of the present invention, for example, by control software executed by a microprocessor, electronic control unit, client, server, etc. However, it should be noted that the computer software executing the methods of the present invention is not limited to execution in one or a specific set of hardware entities; it can also be implemented in a distributed manner by unspecified hardware. For computer software, the software product can be stored on a computer-readable recording medium (such as a CD-ROM, USB flash drive, portable hard disk, etc.) or distributed across a network, as long as it enables electronic devices to execute the methods according to the present invention.

[0107] In summary, this invention solves the problems of coarse search granularity, delayed updates, network dependence, and privacy risks in existing technologies by using dynamic and static information collection, multi-path fusion analysis, personalized weight adjustment, and real-time update feedback. It achieves localized functional search and guidance that is fine-grained, personalized, updated in real time, available offline, and secure in terms of privacy, and has significant technological progress and practical value.

[0108] The above embodiments are only used to illustrate the present invention and are not intended to limit the technical solutions described herein. Although the present invention has been described in detail with reference to the above embodiments, the present invention is not limited to the specific embodiments described above. Therefore, any modifications or equivalent substitutions to the present invention, as well as all technical solutions and improvements that do not depart from the spirit and scope of the invention, are covered within the scope of the claims of the present invention.

Claims

1. A localized search method based on smart devices, characterized in that, include: Obtain application information of each application installed in the smart device, the application information including preset static resource information and dynamic resource information generated at runtime; Based on the application information, the functions supported by each application and the corresponding function entry paths are analyzed by a language model deployed locally, and function description information is generated. Receive a search request input by the user, and determine the target function and its entry path that match the search request based on the function description information; The smart device provides a guide entry point for the target function on its display interface. The guide entry point contains the function entry path and automatically redirects to the page containing the target function in response to the user's trigger operation.

2. The localized search method based on smart devices according to claim 1, characterized in that, The process of obtaining application information for each application installed on the smart device includes: Scan the application list in the smart device using system permissions to obtain the identification information of each application; Based on the identification information, the static resource information is extracted from the system installation package directory, resource folder, or application sandbox. The static resource information includes at least one of the following: application name, version number, permission declaration, interface layout file, preset text, and menu hierarchy structure. The dynamic resource information is obtained by monitoring the application's runtime behavior data, intercepting system logs, or reading the runtime database. The dynamic resource information includes at least one of the following: user behavior trajectory, dynamically downloaded content resources, real-time generated interface elements, and temporary cache data.

3. The localized search method based on smart devices according to claim 1, characterized in that, Based on the application information, the functions supported by each application and their corresponding entry paths are analyzed using a locally deployed language model to generate function description information, including: Based on the static resource information, the preset interface hierarchy and function entry points of each application are parsed, and a first set of function paths is extracted. The first set of function paths reflects the standard access paths at the application design level. Based on the dynamic resource information, the operation sequence and frequency distribution of users accessing each function during actual use are identified, and a second function path set is extracted. The second function path set reflects the actual access path at the user's habit level. The first set of functional paths and the second set of functional paths are merged, multiple candidate paths are associated with the same function, and a confidence weight is assigned to each candidate path, wherein the path corresponding to the dynamic resource information has a higher initial confidence. The fused functional path information is input into the language model to generate functional description information that includes functional name, functional description, multiple candidate paths and their confidence weights.

4. The localized search method based on smart devices according to claim 3, characterized in that, Based on the functional description information, determine the target function and its entry path that match the search request, including: The search request is semantically parsed and intent is identified to generate a query vector; Calculate the semantic similarity between the query vector and each function in the function description information, and determine one or more candidate functions with the highest similarity; For each candidate function, the path with the highest confidence score is selected from its multiple associated candidate paths based on the confidence score weight as the recommended entry path for that function. The candidate functions are ranked comprehensively based on their similarity scores and the confidence weights of the recommended entry paths. The function with the highest ranking is selected as the target function and its corresponding recommended entry path.

5. The localized search method based on smart devices according to claim 4, characterized in that, When the target function that matches the search request cannot be determined, or when the similarity score of all candidate functions is lower than a preset threshold, the user is provided with fuzzy matching suggestions or a list of alternative functions generated by the language model is displayed. Record user feedback behavior on the list of alternative functions, including user clicks to select or abandon, and use the feedback behavior as training data for subsequent optimization and adjustment of the language model, as well as for adjusting the confidence weights of the corresponding functions.

6. The localized search method based on smart devices according to claim 3, characterized in that, The application information also includes the user's historical behavior data and preference settings, and the localized search method includes: Based on the historical behavior data and preference settings, the confidence weights of each path in the second functional path set are dynamically adjusted, specifically including: Get the user's historical usage frequency for each path fi and most recent usage time ti ; According to the preset time decay factor λ and frequency weighting coefficient k Calculate the adjustment coefficient , where Δ ti The time interval between the current time and the most recent usage time; The initial confidence weight of the path w 0 i Multiply by the adjustment factor αi The adjusted confidence weights are obtained. wi = w 0 i · αi ; When merging the first functional path set and the second functional path set, the adjusted confidence weights are used. wi As the confidence weight of each path in the second functional path set, the overall confidence of each candidate path is recalculated.

7. The localized search method based on smart devices according to claim 1, characterized in that, include: Real-time monitoring of changes in the status of applications installed on smart devices, including applications being added, uninstalled, or updated. When a change in the application status is detected, the application information of the application is re-collected and the function description information is updated to maintain the real-time performance of the local function index library.

8. The localized search method based on smart devices according to claim 1, characterized in that, The smart device provides a guide entry point for the target function on its display interface, including: The target function is displayed in the form of a card, which is either floating on the top layer of the current interface or embedded in the search results list. In response to the user's triggering of the jump button, the system simulates click operations or sends Intent commands sequentially according to the recommended entry path corresponding to the target function, and automatically navigates to the in-app page where the target function is located.

9. A learning machine employing the localized search method as described in any one of claims 1-8, characterized in that, include: The application information acquisition module acquires the application information of each installed application, including preset static resource information and dynamically generated dynamic resource information at runtime. The function description information generation module, based on the application information, analyzes the functions supported by each application and the corresponding function entry paths through a language model deployed locally, and generates function description information. The search request receiving module receives the search request input by the user and, based on the function description information, determines the target function and its function entry path that match the search request. The guidance entry module provides a guidance entry for the target function on the display interface of the smart device. The guidance entry includes the function entry path and automatically jumps to the page where the target function is located in response to the user's trigger operation.

10. A computer-readable recording medium storing a computer-executable program, characterized in that, When the computer executable program is executed, it implements a localized search method based on a smart device as described in any one of claims 1-8.

Images