A cross-application data retrieval method, apparatus, terminal device and medium

By using field-level access control and content fingerprinting technology, the problems of coarse permission control and data redundancy in cross-application retrieval are solved, enabling secure and efficient cross-application data retrieval and improving the accuracy and cleanliness of retrieval results.

CN123115239APending Publication Date: 2026-11-17TONGCHENG NETWORK TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611290810.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-25
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing cross-application search methods suffer from coarse-grained permission control, leading to high privacy and security risks, and redundant multi-source data results in low accuracy of search results.

Method used

By using field-level fine-grained permission control for secure retrieval and content fingerprint-based cross-application deduplication, the system can desensitize or remove candidate data, and use content fingerprint to determine the source and deduplicate candidate data from different target applications.

Benefits of technology

While protecting user data privacy, it improves the accuracy and cleanliness of search results, reduces the risk of excessive collection or accidental leakage of privacy data, and improves the efficiency of information acquisition.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115239A_ABST
    Figure CN123115239A_ABST
Patent Text Reader

Abstract

This application provides a cross-application data retrieval method, apparatus, terminal device, and medium, relating to the field of computer data processing technology. The method includes: in response to receiving a retrieval request, distributing retrieval tasks to multiple target applications to obtain candidate data; performing desensitization or removal processing on restricted fields in the candidate data based on field-level access control policies; extracting content fingerprints from the processed candidate data, and accordingly performing source homology determination and deduplication processing on candidate data from different target applications to obtain aggregated retrieval results; and outputting the aggregated retrieval results. This application achieves fine-grained field-level privacy protection and cross-application source homology data deduplication, improving retrieval security and result accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of computer data processing technology, and more specifically, to a cross-application data retrieval method, apparatus, terminal device, and medium. Background Technology

[0002] As the number of applications installed on terminal devices continues to increase, users often need to search for historical information or business data across multiple applications. While existing system-level search solutions support cross-application retrieval, their access control mechanisms typically remain at the application level. This means that once a user authorizes an application to participate in the search, all data fields within that application may be retrieved and displayed, lacking fine-grained control over sensitive fields and easily leading to privacy risks. Furthermore, when multiple similar applications (such as multiple e-commerce platforms or social media apps) return data pointing to the same entity, existing solutions lack effective cross-application source identification mechanisms, reducing information retrieval efficiency. Summary of the Invention

[0003] To address the issues of high privacy and security risks caused by coarse-grained permission control in cross-application retrieval and low accuracy of retrieval results due to redundancy of multi-source data in existing technologies, this application provides a cross-application data retrieval method, apparatus, terminal device, and storage medium. This achieves secure retrieval based on fine-grained field-level permission control and cross-application same-source deduplication based on content fingerprints, thereby improving the accuracy and cleanliness of retrieval results while protecting user data privacy.

[0004] Firstly, a cross-application data retrieval method is provided, which may include: In response to receiving a retrieval request, a retrieval task is distributed to multiple target applications to obtain candidate data from multiple target applications; Based on field-level access control policies, restricted fields in the candidate data are de-identified or removed to obtain processed candidate data. Content fingerprints are extracted from the processed candidate data, and the candidate data from different target applications are subjected to homology determination and deduplication based on the content fingerprints to obtain aggregated retrieval results. Output the aggregated search results.

[0005] In one possible implementation, based on a field-level access control policy, de-identification or removal processing is performed on restricted fields in the candidate data, including: Obtain the field permission configuration information corresponding to each of the target applications; Based on the field permission configuration information, the data fields in the candidate data are divided into searchable fields and restricted fields; Perform data desensitization or field removal operations on the restricted fields.

[0006] In one possible implementation, content fingerprints are extracted based on the processed candidate data, and the candidate data from different target applications are subjected to homology determination and deduplication based on the content fingerprints, including: Extract key attribute features from the processed candidate data to generate the content fingerprint; Calculate the fingerprint similarity between the processed candidate data from different target applications; When the fingerprint similarity meets the preset similarity condition and the corresponding timestamp difference is within the preset time window, the corresponding processed candidate data is determined to be source data, and deduplication is performed on the source data.

[0007] In one possible implementation, performing data desensitization or field removal operations on the restricted field includes at least one of the following: Masking and replacing sensitive characters in the restricted field; The restricted fields are hashed. Set the field value of the restricted field to null; The method further includes: In response to receiving a fine-grained authorization instruction for the target application, the field permission configuration information is dynamically updated to adjust the scope of the searchable fields and the restricted fields.

[0008] In one possible implementation, performing a deduplication operation on the homogeneous data includes: Calculate the relevance score of each candidate data in the source data; The candidate data with the highest relevance score is retained as the target display data; Fold the candidate data in the same source data, excluding the target display data, to generate a folded display identifier; The output of the aggregated search results includes: The target data is displayed on the search results page, and the collapsed display icon is displayed in association. In response to a trigger operation on the collapsed display icon, the other candidate data that has been collapsed is expanded and displayed.

[0009] In one possible implementation, retrieval tasks are distributed to multiple target applications to obtain candidate data from multiple target applications, including: The retrieval task is distributed to multiple target applications using a standardized data extraction adapter; Receive the raw application data returned by the multiple target applications; The candidate data is obtained by converting the original application data into a unified data structure using the standardized data extraction adapter. The unified data structure includes a content identifier field, a source identifier field, and a content summary field.

[0010] In one possible implementation, the unified data structure also includes page positioning parameters; The method further includes: In response to a trigger operation on a target result item in the aggregated search results, the page positioning parameters corresponding to the target result item are obtained; By using a redirect protocol adapter, the page positioning parameters are converted into a dedicated redirect protocol for the source target application corresponding to the target result item; Based on the exclusive redirection protocol, the control terminal device redirects to the content details page in the source target application that corresponds to the page positioning parameters.

[0011] Secondly, a cross-application data retrieval device is provided, which may include: The data acquisition module is used to respond to a received retrieval request by distributing retrieval tasks to multiple target applications in order to obtain candidate data from multiple target applications; The field processing module is used to perform desensitization or removal processing on restricted fields in the candidate data based on field-level access control policies to obtain processed candidate data. The fingerprint deduplication module is used to extract content fingerprints based on the processed candidate data, and to perform homology determination and deduplication processing on candidate data from different target applications based on the content fingerprints to obtain aggregated retrieval results. The result output module is used to output the aggregated search results.

[0012] Thirdly, a terminal device includes a memory and a processor, the memory storing a computer program, characterized in that the processor executes any of the steps of the method described in the first aspect above.

[0013] Fourthly, a computer-readable storage medium is provided, wherein a computer program is stored therein, and when executed by a processor, the computer program implements the steps of any of the methods described in the first aspect above.

[0014] This application refines privacy protection granularity from the application level to the field level by performing anonymization or removal of restricted fields in candidate data based on field-level access control policies. This allows the system to automatically filter or mask sensitive information based on granular permission configurations when searching using data from multiple applications, fundamentally reducing the risk of excessive collection or accidental leakage of user privacy data without sacrificing cross-application search capabilities. Simultaneously, by extracting content fingerprints from the processed candidate data and combining fingerprint similarity and time windows to determine the source and remove duplicates from data from different target applications, redundant data scattered across multiple applications but essentially pointing to the same entity can be effectively identified. This avoids the repeated stacking of similar information in search results, significantly improving the information density and reading efficiency of the results page. Furthermore, through the cooperation of standardized data extraction adapters and jump protocol adapters, unified processing and accurate backtracking of heterogeneous application data are achieved. Combined with a dynamic sorting mechanism based on user behavior feedback, the cross-application search system possesses good scalability and personalized service capabilities while ensuring security and cleanliness. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A system architecture diagram for cross-application unified search provided in the embodiments of this application; Figure 2 A flowchart illustrating a cross-application data retrieval method provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a cross-application data retrieval device provided in an embodiment of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0018] As the number of applications installed on terminal devices continues to increase, users often need to search for historical information or business data across multiple applications. While existing system-level search solutions support cross-application retrieval, their access control mechanisms typically remain at the application level. This means that once a user authorizes an application to participate in the search, all data fields within that application may be retrieved and displayed, lacking fine-grained control over sensitive fields and easily leading to privacy risks. Furthermore, when multiple similar applications (such as multiple e-commerce platforms or social media apps) return data pointing to the same entity, existing solutions lack effective cross-application source identification mechanisms, reducing information retrieval efficiency.

[0019] Therefore, in view of the problems of high privacy and security risks caused by coarse-grained permission control in cross-application retrieval and low accuracy of retrieval results due to redundancy of multi-source data in existing technologies, this application realizes secure retrieval based on field-level fine-grained permission control and cross-application same-source deduplication based on content fingerprints through a cross-application data retrieval method, thereby improving the accuracy and cleanliness of retrieval results while protecting user data privacy.

[0020] The preferred embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit this application. Furthermore, the embodiments and features in the embodiments of this application can be combined with each other without conflict.

[0021] Figure 2 This is a flowchart illustrating a cross-application data retrieval method provided in an embodiment of this application. Figure 2 As shown, this method can be applied to search agent modules or processors in terminal devices. This method achieves cross-application information aggregation while ensuring data security. Specifically, the method includes the following steps: Step S210: In response to receiving a retrieval request, a retrieval task is distributed to multiple target applications to obtain candidate data from multiple target applications.

[0022] Specifically, the unified search entry point is located on the mobile terminal's system desktop, drop-down search bar, or settings menu. Users can access the unified search interface by clicking on it. This entry point supports multiple input methods for initiating search requests: users can directly enter text keywords, input voice commands via voice-to-text conversion, perform image recognition search by taking a photo or selecting an image from the album, or quickly reuse previously used search keywords by clicking on historical input records. When a user initiates a search using any of these methods, the system generates a search request containing a search intent identifier. To shorten user waiting time, this embodiment employs a parallel distribution mechanism, whereby the search agent module simultaneously sends search tasks to multiple authorized target applications, rather than sequentially calling them. For example, when the search intent is identified as "product query," the system can simultaneously initiate search calls to the first e-commerce application, the second e-commerce application, and the third e-commerce application. After receiving the task, each target application returns matching data based on local indexes, cloud interfaces, or historical records; this raw returned data constitutes the candidate data. It should be understood that although this embodiment preferably uses parallel distribution to improve search response speed, in certain resource-constrained or specific business scenarios, grouped parallel or priority sequential methods can also be used, as long as multi-source data acquisition can be achieved. Through the parallel distribution mechanism, the system can control the total time of cross-application retrieval to within the response time of the slowest single application, significantly improving the efficiency of multi-source data acquisition.

[0023] Before distributing retrieval tasks to multiple target applications, intent recognition and candidate application screening are required for the keywords in the retrieval request. Specifically, the keywords entered by the user are first segmented into independent semantic units. Then, based on preset intent classification rules, the segmentation results are type-determined to identify the content category that the keyword may correspond to, including but not limited to products, orders, chat logs, favorites, image content, notes, and documents. After completing intent recognition, the search agent module determines the scope of target applications to be prioritized for retrieval based on the semantic features of the keywords, that is, it executes candidate application selection rules: when the keywords contain shopping-related semantic units such as "buy," "order," "product," "logistics," and "shopping cart," shopping applications are identified as priority retrieval targets; when the keywords contain social semantic units such as "chat," "send me," "friends say," "message," and "group chat," social applications are identified as priority retrieval targets; when the keywords contain image semantic units such as "photo," "image," "screenshot," and "album," album applications are identified as priority retrieval targets. Through the aforementioned intent recognition and candidate application selection rules, the search agent module can dynamically determine the target application list based on the search intent, avoiding the waste of computing resources and response delays caused by distributing search tasks to irrelevant applications. This enables accurate and efficient task distribution across multiple application scenarios. It should be understood that the aforementioned intent classification rules and keyword mapping relationships are merely illustrative examples. In practical applications, the rules can be expanded or adjusted based on changes in the application ecosystem and the accumulation of user behavior data, as long as the goal of filtering candidate applications based on keyword semantics is achieved.

[0024] Step S220: Based on the field-level access control policy, perform desensitization or removal processing on the restricted fields in the candidate data to obtain the processed candidate data.

[0025] Specifically, the field-level access control policy exists as a logical processing unit. Its input is the raw candidate data obtained in step S210, and its output is the processed candidate data after security cleaning. The core of this strategy lies in sinking the granularity of permission control from the traditional application level to the field level. That is, before the data enters the subsequent retrieval process, it first identifies which fields are restricted fields (such as personal privacy information, sensitive transaction details, etc.) according to preset rules, and performs corresponding security processing on them. It is important to emphasize that this step must be performed before content fingerprint extraction to ensure that sensitive data is not included in the subsequent fingerprint calculation and comparison, thereby blocking the risk of privacy leakage at the source. The specific field division method and the detailed implementation of the de-identification operation will be further elaborated in subsequent embodiments. Here, it is only defined as a necessary processing pipeline connecting raw data and secure data. Through this pre-processing mechanism, the system can effectively avoid compliance and security risks caused by excessive data exposure while providing retrieval services using multi-application data.

[0026] Step S230: Extract content fingerprints based on the processed candidate data, and perform homology determination and deduplication on candidate data from different target applications based on the content fingerprints to obtain aggregated search results.

[0027] Specifically, since multiple target applications may store or display information pointing to the same entity (e.g., the same product is sold on different e-commerce platforms, or the same article is reprinted by multiple news applications), directly displaying this information would lead to redundancy in the results page. Content fingerprinting technology is introduced to solve this cross-application same-origin determination problem. Content fingerprints are unique or near-unique identifiers generated based on the key attribute features of processed candidate data, specifically targeting data from different target applications. By comparing the fingerprint similarity of data from different sources, it is determined whether they substantially point to the same content entity. For cross-application data determined to be of the same origin, the system performs deduplication, merging or filtering them to ultimately form clean aggregated search results. This fingerprint-based cross-application deduplication mechanism, unlike simple text deduplication within a single application, effectively addresses the semantic alignment challenges between heterogeneous data sources, significantly improving the information density and reading experience of the search results.

[0028] Step S240: Output the aggregated search results.

[0029] Specifically, after completing security filtering and deduplication, the aggregated search results are rendered onto a unified search results page for users to view. The output should include at least the deduplicated target data and its corresponding source identifier. It should be understood that the output action here is a broad concept, encompassing both visual display on the terminal screen and returning structured data to the upper-level API. Within the basic framework, the output focuses on presenting a clean and secure set of results processed through the aforementioned steps.

[0030] Through the steps outlined above, a complete cross-application data retrieval framework has been constructed. This framework addresses the efficiency issues of multi-source acquisition through parallel distribution, the coarse-grained privacy protection issues through field-level access control strategies, and the redundancy of cross-application results through content fingerprinting technology. These three core mechanisms work together to support a cross-application retrieval system that is both efficient and secure, and provides accurate results.

[0031] This paper further elaborates on the specific implementation process of processing candidate data based on field-level access control policies. As one implementation method, the process of de-identifying or removing restricted fields in candidate data based on field-level access control policies includes: obtaining field permission configuration information corresponding to each target application; dividing the data fields in the candidate data into searchable fields and restricted fields based on the field permission configuration information; and performing data de-identification or field removal operations on the restricted fields.

[0032] Specifically, field permission configuration information is the foundation for implementing fine-grained field-level control, and its acquisition methods are diverse. In one implementation scenario, this configuration information can be pre-installed in the local storage of the terminal device as the system's default security baseline. For example, a configuration file containing the hierarchical definitions of mainstream application fields can be loaded during operating system installation or search agent module initialization. In another implementation scenario, to cope with real-time adjustments to application version iterations or privacy compliance policies, this configuration information can also be dynamically distributed by a cloud server. The search agent module periodically or before each search task starts, pulls the latest configuration version from the cloud. Regardless of the acquisition method, the attribute determination of each data field in the candidate data is performed based on the field identifier and security level mapping relationship contained in the configuration information. Fields marked by the configuration information as allowed to participate in retrieval, display, or indexing are classified as searchable fields; while fields marked as containing personal privacy, trade secrets, or other sensitive information are classified as restricted fields. This configuration-driven classification mechanism makes permission control no longer dependent on hard-coded logic, but can flexibly adapt to the differences in data structures of different target applications by updating the configuration file, thus decoupling security policies from business logic.

[0033] After identifying the restricted fields, appropriate security measures must be taken. One implementation method is to perform data anonymization or field removal operations on the restricted fields, including at least one of the following: masking and replacing sensitive characters in the restricted fields; hashing the restricted fields; or setting the field value of the restricted fields to null.

[0034] Specifically, the system employs differentiated de-identification methods to balance security and data usability for different types of sensitive data. For example, for fields with fixed formats, such as mobile phone numbers and ID card numbers, where some information still has identifiable value, a mask replacement method is used, retaining the first or last few characters of plaintext while replacing the middle part with asterisks. This prevents complete privacy leakage while preserving the user's ability to verify the data source. For absolutely sensitive fields such as passwords, keys, or biometric digests that do not need to be displayed in search results, hashing or direct nullification is used. Hashing is suitable for scenarios requiring verification of data existence or non-plaintext comparison, while nullifying field values ​​is suitable for scenarios where no information needs to be retained at all. The field is directly removed from the data structure or filled with an empty value to minimize the data exposure surface. It should be understood that the above de-identification operations are only illustrative examples. In practical applications, other technical means such as encryption, generalization, and perturbation can be used according to the data sensitivity level, as long as the goal of blocking the plaintext output of sensitive information can be achieved. Through this diversified de-identification mechanism, the system can maximize the preservation of data retrieval value and contextual integrity while ensuring privacy and security.

[0035] Furthermore, to empower users with autonomy over their personal data, this application also supports dynamic adjustment of permission configurations. As one implementation, the method further includes: dynamically updating field permission configuration information in response to receiving a fine-grained authorization instruction for a target application, to adjust the scope of the division between searchable fields and restricted fields.

[0036] Specifically, fine-grained authorization commands refer to explicit authorization or revocation operations issued by users in the terminal settings interface or search results page, targeting specific fields of a specific application. For example, a user can choose to allow browsing history of the first e-commerce application to participate in the search in the unified search settings panel, but prohibit its order delivery address field from being indexed. When the search agent module receives such a command, it will immediately modify the field permission configuration information in memory or local persistent storage, so that the new classification rules take effect immediately when the next search task is distributed. This dynamic update mechanism breaks the rigid mode of traditional static authorization, allowing field-level access control policies to evolve in real time according to user wishes, enhancing the trust and transparency of human-computer interaction.

[0037] As a further enhancement to the field-level access control policy, this application also introduces a privacy authorization token mechanism. Specifically, when a user authorizes a specified data field of a target application to participate in a unified search for the first time, the permission control module generates a privacy authorization token corresponding to the target application and the scope of authorized fields. The privacy authorization token is an access credential issued by the system, internally encoded with information such as the target application identifier, the list of authorized fields, the authorization effective time, the authorization expiration time, and the token signature. When the search proxy module issues a search task to the application data connector of the target application, it must pass a valid privacy authorization token as a request parameter. Before executing the search, the application data connector first verifies the signature validity of the token and the scope of authorized fields. Only when the token verification passes and the field to be searched falls within the token's authorization scope will the actual data matching and return operation be performed. For fields outside the token's authorization scope, the connector automatically performs de-identification or removal processing. When a user adjusts the field permission configuration through fine-grained authorization commands, the permission control module synchronously updates or revokes the corresponding privacy authorization token, ensuring that the new authorization scope takes effect immediately in the next search. Privacy authorization tokens employ a short-term validity issuance strategy, automatically expiring upon default. Users must re-authorize the tokens, or the system can automatically renew them based on continued user activity, thus preventing the misuse of long-term valid static credentials. Through this privacy authorization token mechanism, the system upgrades field-level access control from simple configuration comparison to an access control system with credential verification. This ensures that every cross-application data access has traceable authorization evidence, further enhancing privacy and security in cross-application retrieval scenarios.

[0038] It is particularly important to emphasize that the above-mentioned field-level access control policy-based processing must be strictly scheduled before the content fingerprint extraction step. This is because content fingerprint calculation typically relies on key attribute features of the data. If restricted fields are not anonymized or removed beforehand, sensitive data is highly likely to be included in the input parameters of the fingerprint generation algorithm, resulting in the generated fingerprint itself carrying privacy information, which could lead to secondary leakage risks in subsequent cross-application comparisons. By implementing security processing upfront, it ensures that all data entering the fingerprint calculation stage is cleaned and processed candidate data, building a privacy protection firewall at the source of data flow and completely blocking the possibility of sensitive information infiltrating downstream processing links.

[0039] This paper further elaborates on the specific implementation process of cross-application source identification and deduplication based on content fingerprints, as well as the display strategy of aggregated search results after deduplication. As one implementation method, content fingerprints are extracted from processed candidate data, and source identification and deduplication are performed on candidate data from different target applications based on these content fingerprints. This includes: extracting key attribute features from the processed candidate data to generate content fingerprints; calculating the fingerprint similarity between processed candidate data from different target applications; and when the fingerprint similarity meets a preset similarity condition and the corresponding timestamp difference is within a preset time window, the corresponding processed candidate data is identified as source data, and deduplication is performed on the source data.

[0040] Specifically, in step S230, key attribute features are first extracted from the processed candidate data to generate content fingerprints. These key attribute features refer to data fields that can uniquely identify an entity and remain relatively stable across application scenarios. Their selection depends on the business type of the target application. For example, for e-commerce applications, key attribute features typically include a combination of product title, SKU code, specifications, and price; for news and information applications, they may include a combination of article title, text summary, author identifier, and publication time; and for social chat applications, they may include a combination of message content hash, sender ID, and session identifier. It should be understood that the emphasis on using "processed candidate data" as the input for fingerprint extraction is to form a logical closed loop with the field-level access control policy. Since restricted fields have been desensitized or removed in the previous step, the extracted features only include secure and compliant searchable fields. This avoids sensitive information (such as user phone numbers and shipping addresses) being included in the fingerprint calculation algorithm, preventing the risk of secondary leakage of privacy data during fingerprint comparison and avoiding the problem of fingerprint misalignment due to personalized differences in sensitive fields.

[0041] After generating content fingerprints, the fingerprint similarity between processed candidate data from different target applications is calculated. A dual verification mechanism of fingerprint similarity and time window can be used to determine same-origin data, rather than relying solely on single text similarity matching. Specifically, only when the fingerprint similarity of two candidate data from different applications exceeds a preset threshold (e.g., SimHash Hamming distance less than 3), and the difference in their timestamps falls within a preset time window (e.g., 24 hours or 7 days), will they be determined as same-origin data pointing to the same entity. The necessity of introducing time window verification is to eliminate interference from "historically similar but not the same entity". For example, if a user purchased a "blue down jacket" last year and searches for a new "blue down jacket" with the same name but from a different year this year, relying solely on the text similarity of the title and specifications could easily lead to misjudgment as a duplicate result. However, through time window constraints, the system can accurately distinguish independent transaction records or product information under different time periods, significantly reducing the false positive rate of cross-application deduplication and ensuring the accuracy of aggregated search results.

[0042] After determining the source data, the system needs to perform deduplication on the identified source data groups to eliminate redundancy. As one implementation method, deduplication of the source data includes: calculating the relevance score of each candidate data item in the source data; retaining the candidate data with the highest relevance score as the target display data; and folding the other candidate data in the source data (excluding the target display data) to generate a folded display identifier.

[0043] Specifically, the relevance score is not calculated based on a single dimension of matching degree, but rather through a weighted evaluation of multiple factors. These factors include at least keyword matching degree, content completeness, user preference weight of the source application, and time freshness. The relevance score calculation formula is: Score = A × TitleHit + B × SummaryHit + C × TimeWeight + D × AppPreference + E × ContentCompleteness, where A, B, C, D, and E are the corresponding weight parameters; TitleHit indicates whether the title matches; SummaryHit indicates whether the summary matches; TimeWeight indicates time relevance; AppPreference indicates user preference weight; and ContentCompleteness indicates content completeness. For example, if a user has frequently used the first e-commerce application in the past, and that application returns the most complete data fields, their score may be higher than the same source data returned by the second e-commerce application. The system will determine the candidate data with the highest score as the target display data, serving as the sole representative item of that same source data group in the search results list. For the remaining low-scoring candidate data within the group, the system does not physically delete them directly. Instead, it performs a collapse process, generating a collapse display icon associated with the target display data. This icon can be displayed on the interface as an interactive control such as "View 3 other sources" or "Expand to more platforms".

[0044] Correspondingly, when outputting aggregated search results, the target data is displayed on the search results page, along with a collapsed display indicator. In response to a trigger action on the collapsed display indicator, other collapsed candidate data is displayed. This strategy of retaining the best data and collapsing the rest essentially deeply integrates data cleaning logic with the user experience. From a technical perspective, on the one hand, it significantly improves the information density and reading efficiency of the search results page by hiding redundant information with low relevance, preventing users from getting lost in a large amount of repetitive content; on the other hand, it retains the entry point for users to trace other data sources, meeting users' deeper needs in specific scenarios such as cross-application price comparison and multi-source verification. Compared to a simple and crude direct deletion strategy, this collapsing mechanism ensures the cleanliness of the results while taking into account the integrity of information and the user's right to choose, demonstrating the collaborative optimization of the technical solution at the data processing and human-computer interaction levels.

[0045] This section further elaborates on the standardized processing mechanism during the data acquisition stage of cross-application retrieval and the precise jumpback implementation method during the result output stage. For example... Figure 1 or Figure 3 As shown, in order to solve the retrieval difficulties and jump breakage problems caused by heterogeneous data structures between different target applications, this application introduces a technical means that combines adapters with unified data structures.

[0046] To improve the response speed of cross-application retrieval, this application also includes a content indexing module in the system architecture. This module pre-generates local or cloud indexes for the searchable content of each target application. Specifically, during idle periods on the terminal device or after user authorization, the content indexing module extracts searchable content from each target application through the application data connector and generates structured index data according to preset index field specifications. Index fields include at least the content title, content summary, time, source application, content category, and user identifier. The content title is used for quickly matching search keywords; the content summary provides a result preview; the time field supports filtering by time range and sorting by time freshness; the source application identifies the data origin and serves as the basis for subsequent field-level access control and redirection protocol matching; the content category distinguishes different data types such as products, orders, chat, and images; and the user identifier isolates index data from different users in multi-account scenarios. The content indexing module supports two deployment methods: local indexes are stored in the terminal device's non-volatile memory, suitable for frequently accessed personal data; cloud indexes are stored on a remote server, suitable for application content that is synchronized across devices or has a large data volume. When the search agent module receives a retrieval request, it can prioritize querying the index data in the content index module. Only when the index is not found or real-time data is required will it call the application data connector for real-time querying, thereby significantly shortening the retrieval response time. The content index module also supports an incremental update mechanism, that is, when new or changed content is detected in the target application, only the changed parts of the index are updated, rather than a full rebuild, to reduce the computational overhead and energy consumption of index maintenance.

[0047] As one implementation method, a retrieval task is distributed to multiple target applications to obtain candidate data from multiple target applications, including: distributing the retrieval task to multiple target applications through a standardized data extraction adapter; receiving the original application data returned by multiple target applications; and converting the original application data into a unified data structure through the standardized data extraction adapter to obtain candidate data; wherein the unified data structure includes a content identifier field, a source identifier field, and a content summary field.

[0048] Specifically, the standardized data extraction adapter acts as a universal translation layer between the search agent module and various heterogeneous target applications. Unlike the traditional approach of developing hard-coded connectors for each application, this adapter adopts a configuration-driven mapping mechanism. When the system needs to integrate a new target application, developers or operations personnel only need to provide a field mapping configuration file, declaring the correspondence between the application's original data fields and the fields in the unified data structure, without modifying the adapter's core code logic. For example, the configuration file can specify that the "item_id" field of the first e-commerce application is mapped to the "content identifier field" in the unified structure, "shop_name" is mapped to the "source identifier field," and "product_desc" is mapped to the "content summary field." This design greatly reduces the development cost and maintenance difficulty of integrating new applications, enabling the cross-application retrieval system to have good horizontal scalability.

[0049] In the data transformation process, the unified data structure plays a crucial role in bridging the gap between the preceding and following steps. The content identifier field uniquely identifies a candidate piece of data within the system, typically consisting of the primary key ID of the source application or a hashed feature value, ensuring the accuracy of subsequent deduplication and indexing operations. The source identifier field explicitly records the original source of the data, such as the application package name, application ID, or service provider name. This is not only the basis for displaying the source icon on the front end but also the basis for implementing field-level access control policies and redirection protocol matching. The content summary field provides a concise summary of the original content, used to offer a quick preview in the search results list, avoiding the direct display of lengthy original details. It should be understood that although the above lists three core fields, in practical applications, the unified data structure can be expanded to include other general fields such as timestamps, data type tags, and initial relevance values, as long as they can support subsequent aggregation processing and display. As a further extension of the unified data structure, it also includes content type fields and user identifier fields. The content type field is used to label the business category to which the candidate data belongs, including but not limited to products, orders, chat logs, favorite content, images, notes, and documents. This field provides a classification basis for intent recognition matching, same-origin grouping, and result classification display. The user identifier field records the user account to which the data belongs. In scenarios where multiple users share a terminal or multiple accounts are switched within the application, the user identifier field ensures that the search results correspond only to the data of the currently active user, avoiding the confusion of index data from different users. The above fields, together with the content identifier field, source identifier field, content summary field, page positioning parameters, timestamp, and initial relevance value, constitute a complete and unified data structure, providing a standardized data foundation for subsequent field-level desensitization, fingerprint deduplication, adaptive sorting, and precise jumpback. Through the transformation of the standardized data extraction adapter, the originally diverse original application data is regularized into a semantically consistent and structurally unified candidate data stream, providing a standardized data foundation for field-level desensitization and fingerprint deduplication, thereby ensuring the consistency and stability of the entire retrieval chain processing logic.

[0050] To further reduce the development cost of new application integration, this application integrates the aforementioned standardized data extraction adapter and jump protocol adapter into a general connector development framework. This framework clearly defines standardized data extraction interfaces, field mapping rules, and jump protocol adaptation specifications. The standardized data extraction interface specifies the set of methods that connectors must implement, including methods for receiving retrieval requests, acquiring raw data, converting field mappings, and returning standardized results. Any newly integrated application connector that implements this interface can be uniformly scheduled by the search proxy module. Field mapping rules are defined in a configurable manner. When a new application integrates, it only needs to provide a field mapping configuration file, declaring the correspondence between the application's raw data fields and unified data structure fields (e.g., declaring which raw fields correspond to content titles, which correspond to content summaries, and which correspond to page positioning parameters), as well as the application's deep-link protocol format. The framework can then automatically complete the data conversion based on the configuration file, eliminating the need to write customized connector code for each application. The jump protocol adaptation specification defines a general conversion interface from page positioning parameters to application-specific jump protocols. New applications only need to declare their deep-link protocol template in the configuration file, and the framework can automatically generate the corresponding jump protocol adapter. By using a general connector development framework, the integration of new applications is transformed from code development to configuration declaration, reducing the workload of integrating a single application from several days of coding development to several hours of configuration and debugging, which significantly improves the horizontal scalability and maintainability of the cross-application retrieval system.

[0051] After completing data retrieval and aggregation display, users often need to jump back to the source application from the unified search results page to view the full details. As one implementation method, the unified data structure also includes page positioning parameters; the method further includes: in response to a trigger operation on a target result item in the aggregated search results, obtaining the page positioning parameters corresponding to the target result item; converting the page positioning parameters into a dedicated jump protocol for the source target application corresponding to the target result item through a jump protocol adapter; and, based on the dedicated jump protocol, controlling the terminal device to jump to the content details page in the source target application corresponding to the page positioning parameters.

[0052] Specifically, page location parameters are deep link information embedded in a unified data structure. Their specific form depends on the source application's technical implementation and can be a DeepLink URL, a custom schema URI, an in-application routing path, or a specific combination of page parameters. For example, for an e-commerce application, the page location parameter might be "app: / / product / detail?id=12345"; while for a social application, it might be "social: / / chat / session / abcde". Since different target applications use different redirection protocols, directly calling the original parameters from the unified search entry point can easily lead to redirection failures or only opening the application's homepage without reaching the specific content page. Therefore, a redirection protocol adapter is introduced. This adapter maintains a protocol conversion rule library that can identify the application type of the target result item based on the source identifier field and automatically load the corresponding conversion plugin or rule. When a user clicks on a target result item on the search results page, the page location parameter is first extracted from the unified data structure of that result item and then passed to the redirection protocol adapter. The adapter encapsulates or converts the general or original page positioning parameters into a jump instruction specific to the target application that can be recognized by the current terminal environment (such as an Intent object in Android or an OpenURL request in iOS). Finally, the terminal device's operating system responds to this specific jump protocol, wakes up the corresponding target application, and passes in the page positioning parameters, enabling the target application to directly render the content details page that precisely corresponds to the search results.

[0053] Through the collaborative work of the standardized data extraction adapter and the jump protocol adapter described above, a complete technical closed loop is constructed from heterogeneous data acquisition to seamless jumpback. For example... Figure 1 or Figure 3 As shown, the data flow path clearly demonstrates how the original application data is transformed into a unified structure via the adapter, and then, during user interaction, is reversed and restored to the source application's executable jump instructions via the protocol adapter. This mechanism not only solves the two major engineering challenges of incomprehensible data and inability to jump between pages in cross-application searches, but more importantly, it decouples data standardization and jump capabilities into independent adaptation components. This allows the system to quickly adapt to new application types or new jump protocols that may emerge in the future by updating configurations or plugins, without having to reconstruct the core search logic, significantly improving the lifecycle and practical value of the technical solution.

[0054] This paper further elaborates on an adaptive ranking mechanism based on user behavior feedback to address the technical problem that static ranking rules are difficult to adapt to changes in users' personalized needs. As one implementation method, the method further includes: acquiring user behavior feedback data regarding historical aggregated search results; dynamically adjusting the weight parameters in the relevance scoring model based on the behavior feedback data; and outputting aggregated search results, including: ranking the aggregated search results based on the adjusted weight parameters and outputting the ranked aggregated search results.

[0055] Specifically, behavioral feedback data is a multi-dimensional set of signals characterizing users' true satisfaction with search results and their intent matching degree. It encompasses both explicit interaction and implicit behavior. Explicit interaction data includes, but is not limited to, users' clicks, favorites, or shares of specific items in aggregated search results; this data directly reflects users' recognition of the relevance of the content. Implicit behavioral data includes the duration of time users spend on the results details page, the depth of page scrolling, and the behavior of quickly swiping over or ignoring unclicked results. In particular, the system also includes users' secondary search behavior or keyword modification behavior in the feedback data. When a user initiates a similar or modified search request shortly after viewing the current result, the system identifies this as a negative feedback signal to the previous ranking result. By integrating multimodal behavioral feedback data, the system can capture users' true intent and preference shifts more accurately than a single click-through rate, providing high-confidence training samples for optimizing the ranking model.

[0056] After acquiring behavioral feedback data, instead of simply accumulating statistical values, the weight parameters in the relevance scoring model are dynamically adjusted through a specific algorithm. Specifically, an incremental update strategy based on time decay is used to perform weight adjustments. A weight vector corresponding to each scoring factor in the relevance scoring model (such as keyword matching degree, time freshness, source application preference, etc.) is maintained. Whenever new behavioral feedback data is collected, the system calculates a gradient update value based on the feedback type (positive or negative) and feedback intensity (such as dwell time). Simultaneously, a time decay factor is introduced, ensuring that recent behavioral feedback has a greater impact on weight adjustments than historical data from a long period, thus giving the model a rapid response capability to changes in user interests. For example, if a user frequently clicks on results from the first target application and spends a long time on the page recently, the system will increase the weight value of the first target application in the source application preference factor by a preset step size, and decrease the weight values ​​of other applications by a decay coefficient. It should be understood that, in addition to the incremental update strategy, offline reinforcement learning or periodic batch regression can also be used for weight optimization in other embodiments, as long as the weight parameters can dynamically evolve with the feedback data. This dynamic adjustment mechanism upgrades the ranking model from a fixed rule-driven system to a data-driven adaptive system, effectively overcoming the shortcomings of traditional solutions where weight parameters are fixed once set and cannot evolve with user habits.

[0057] After adjusting the weight parameters, the system directly uses these adjusted parameters to calculate and rank the relevance scores of the current candidate data when outputting aggregated search results. This means that even with the same search keywords and candidate datasets, the final ranking results will exhibit personalized differences at different times or under different user accounts, because the weight parameters have evolved differently based on their historical behavioral feedback. For example, for users who prefer in-depth content, the weight of the "content completeness" factor may be automatically increased, resulting in longer articles being prioritized in the ranking; while for users who prioritize timeliness, the weight of the "freshness" factor will dominate. In this way, the system transforms general search ranking into personalized intelligent recommendation ranking, significantly improving the efficiency of users obtaining target information.

[0058] It is particularly important to emphasize that, to ensure real-time response performance for cross-application searches, the acquisition of behavioral feedback data and the dynamic adjustment of weight parameters are designed to be executed asynchronously or offline. Specifically, user interactions on the search results page are only lightweightly recorded in the local cache or asynchronous reporting queue, without blocking any part of the main search chain. The calculation and updating of weight parameters are performed during idle periods of the terminal device, while it is charging, or in a low-priority background thread. The updated weight parameters are then written to the shared configuration storage area. When a user initiates a new search request, the ranking module only needs to read the latest weight parameters from the configuration storage area and substitute them into the scoring formula, without waiting for the feedback processing to complete. This read-write separation and asynchronous decoupling architecture ensures that the adaptive ranking mechanism, while giving the system intelligent capabilities, does not negatively impact the end-to-end response latency of search requests, achieving a dual guarantee of personalized experience and system performance.

[0059] like Figure 3 As shown, corresponding to the above method, this embodiment provides a cross-application data retrieval device. This device can be deployed locally on a terminal device or can be constructed by the terminal device and a cloud server working together. The device adopts a modular architecture, achieving full-process cross-application retrieval through data flow between modules. Specifically, the device includes a data acquisition module, a field processing module, a fingerprint deduplication module, and a result output module. It should be understood that these modules are virtual units based on functional logic, and their physical implementation can be independent software components, code segments integrated in the same processor, or service processes distributed across different hardware entities, as long as they can achieve the following functions.

[0060] The data acquisition module 310, in response to a received retrieval request, distributes retrieval tasks to multiple target applications to obtain candidate data from these applications. Specifically, the data acquisition module corresponds to the functional set of step S210 and the standardized data extraction adapter. When a user-initiated retrieval request is received, the module first parses the retrieval intent, determines the list of target applications to be invoked, and then distributes retrieval tasks to these target applications in parallel. After receiving the raw application data returned by each target application, the data acquisition module is also responsible for converting it into a unified data structure, thereby generating a standardized candidate data stream for use by downstream modules. This design allows the data acquisition module to not only assume the responsibility of communication scheduling but also shield the data differences of the underlying heterogeneous applications, providing a standardized input basis for subsequent processing.

[0061] The field processing module 320 is used to perform desensitization or removal processing on restricted fields in candidate data based on field-level access control policies, obtaining processed candidate data. Specifically, the field processing module corresponds to a privacy protection mechanism. This module is located after the data acquisition module and before the fingerprint deduplication module in the data flow path. This arrangement is crucial because it ensures that all data entering subsequent analysis stages has undergone security cleaning. The field processing module internally maintains or dynamically loads field permission configuration information corresponding to each target application. Based on this configuration information, it identifies restricted fields in the candidate data and performs desensitization operations such as mask replacement, hashing, or setting them to null. Through this pre-emptive field-level processing, this module blocks the risk of sensitive information spreading in the cross-application retrieval chain from the source.

[0062] The fingerprint deduplication module 330 is used to extract content fingerprints from processed candidate data and perform source identification and deduplication on candidate data from different target applications based on the content fingerprints to obtain aggregated retrieval results. Specifically, the fingerprint deduplication module corresponds to step S230 and the source identification logic. This module receives processed candidate data from the field processing module as input and extracts key attribute features to generate content fingerprints. Subsequently, the module calculates the fingerprint similarity between data from different target applications and performs double verification in conjunction with time window conditions, identifying data that meets preset conditions as source data. For the identified source data groups, the fingerprint deduplication module further performs deduplication operations, such as retaining the target display data with the highest relevance score and folding other redundant data. This modular deduplication mechanism effectively solves the problem of multi-source data redundancy and significantly improves the information density and cleanliness of the aggregated retrieval results.

[0063] The result output module 340 is used to output the aggregated search results. Specifically, the result output module corresponds to step S240 and the adaptive sorting and display strategy. This module receives the deduplicated aggregated search results, sorts them according to the adjusted weight parameters, and finally renders the sorted results to the unified search results page or returns them to the upper-level calling interface. At the display level, the result output module also supports associated display collapse display icons to respond to user trigger operations to expand collapsed same-source data, thereby maintaining the simplicity of the results while retaining the information traceability entry point. In addition, this module can also work with the jump protocol adapter to convert the user's click operation on the target result item into the source application's exclusive jump protocol, realizing accurate jump back from the unified search results to the specific content details page.

[0064] Through the collaborative work of the aforementioned data acquisition module, field processing module, fingerprint deduplication module, and result output module, the cross-application data retrieval device provided in this embodiment fully reproduces the core technical solution of the method embodiment. The modules are coupled through clearly defined data interfaces. This modular device architecture not only provides a corresponding product protection carrier for the method patent, facilitating functional comparison and infringement evidence collection in actual rights protection, but also provides clear boundaries for the subsequent expansion and maintenance of the system. For example, when a new target application needs to be integrated, only the adapter configuration in the data acquisition module needs to be updated, without modifying the core field processing or fingerprint deduplication logic, demonstrating good engineering scalability.

[0065] This embodiment provides a terminal device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method of the aforementioned embodiment.

[0066] Specifically, the terminal device constitutes the physical operating carrier of the aforementioned cross-application data retrieval method. The memory can be high-speed random access memory or non-volatile memory, such as one or more disk storage devices, flash memory devices, or other volatile solid-state storage devices. It is used to persistently or temporarily store program instructions and related configuration data that implement logic such as field-level access control policies, content fingerprint extraction, homogeneity determination and deduplication, and adaptive sorting. The processor can communicate with the memory via a bus, responsible for reading and executing the computer program from the memory, thereby driving the terminal device to complete a series of operations such as distributing retrieval tasks to multiple target applications, and performing secure cleaning and aggregation output of candidate data. It should be understood that this application does not limit the specific product form of the terminal device. For example, the terminal device can be a smartphone, tablet, laptop, or other mobile computing device; it can also be an in-vehicle device such as an in-vehicle infotainment system or smart cockpit terminal; it can also be a wearable device such as a smartwatch or smart glasses; or even a fixed interactive terminal such as a smart TV or smart home control screen. Any device capable of installing or connecting to multiple target applications, and possessing sufficient computing and storage resources to run the aforementioned retrieval program, can be considered an object of protection under this embodiment. This overarching generalization of terminal form ensures that the technical solution can cover diverse application scenarios, from personal mobile communications to the Internet of Things and the Internet of Vehicles, avoiding an overly narrow scope of protection due to limitation to a single device type.

[0067] In addition, this embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method of the foregoing embodiment.

[0068] Specifically, the computer-readable storage medium serves as the software distribution and circulation carrier for the aforementioned method technology. When the storage medium is loaded onto a terminal device or connected to a terminal device in communication, the computer program stored on it can be recognized and executed by the processor of the terminal device, thereby enabling the terminal device to perform cross-application data retrieval. The types of storage media include, but are not limited to, read-only memory, random access memory, optical disc read-only memory, digital multifunction optical disc, magnetic tape, floppy disk, optical data storage device, and portable flash memory devices such as USB flash drives and SD cards. In practical applications, the storage medium can be sold as a standalone software product, or it can be distributed online or pre-installed as part of an operating system, search agent module installation package, or application update package. By solidifying the method flow into program code on the storage medium, this embodiment achieves independent protection of the technical solution in the upstream software distribution stage of the industry chain. Even without selling the complete hardware, rights can be asserted against the act of providing software programs or firmware upgrade packages separately, thus constructing a complete patent protection system covering the method, apparatus, hardware device, and storage medium.

[0069] This embodiment provides a specific application scenario for cross-e-commerce platform product retrieval, visually demonstrating how the dispersed technical features in the aforementioned embodiments work together to solve complex technical problems such as privacy leakage, result redundancy, and redirection breaks. It should be understood that although this embodiment uses e-commerce shopping as an example, it is illustrative rather than restrictive. In other embodiments, this technical solution is also applicable to scenarios such as message retrieval across social applications, content aggregation across information platforms, or document searching across office systems, as long as it involves unified retrieval and secure processing of multi-source heterogeneous data.

[0070] Specifically, suppose a user enters the keyword "blue down jacket" into the unified search portal on their terminal device and initiates a search request. In response, the system first uses the search agent module to parse the search intent as "product query," and then distributes the search task in parallel to the authorized first, second, and third e-commerce applications. At this stage, the system does not sequentially wait for one application to return before calling the next; instead, it simultaneously triggers multiple standardized data extraction adapters, each connecting to the local index or cloud interface of each e-commerce application. This reduces the overall search time to the response time of the slowest single application, significantly improving the efficiency of multi-source data acquisition.

[0071] After each e-commerce application returns the original candidate data, the system immediately executes security processing based on field-level access control policies. For example, a historical order data returned by the first e-commerce application contains a "recipient's mobile phone number" field, which is marked as sensitive privacy information in the pre-configured field permission information. Based on this, the system classifies it as a restricted field and performs a mask replacement operation, replacing the middle four digits with asterisks, retaining only the first three and last four digits for user identification of the source, while searchable fields such as product name, specifications, and price remain unchanged. Through this fine-grained field-level filtering, the system, while providing search services using order data, completely blocks the risk of leakage of the complete mobile phone number in subsequent fingerprint calculation, result display, and caching storage stages, achieving a balance between privacy protection and data availability.

[0072] Subsequently, the system performs cross-application source identification on the anonymized candidate data. Assume that the first, second, and third e-commerce applications all return data entries pointing to the same product, "2023 Men's Blue Down Jacket." The system extracts key attribute features (such as SKU code and title keyword combination) from these three data entries to generate content fingerprints and calculates the fingerprint similarity between them. When the fingerprint similarity of all three exceeds a preset threshold, and the difference in order timestamps is within a preset 7-day time window, the system determines that these three data entries are source data pointing to the same entity. Based on this, the system further calculates the relevance score for each data entry. Because the user has frequently used the first e-commerce application historically and this data entry has the highest matching degree, its score is higher than the other two. Therefore, the system retains this data entry from the first e-commerce application as the target display data, while the source data from the second and third e-commerce applications is folded, generating a folded display identifier associated with the target display data, such as "View the sources of the other 2 platforms." This approach avoids overwhelming search results pages with duplicate product information, thus increasing information density, while also preserving user interaction points for comparing prices across multiple platforms or tracing the source of products, thus balancing the cleanliness and completeness of the results.

[0073] Furthermore, the search process incorporates an adaptive ranking mechanism. If the system records that a user has consistently prioritized clicking on results from the First E-commerce app in multiple similar searches in the past, and has spent a considerable amount of time on the details page, the behavioral feedback data will trigger a dynamic increase in the "source app preference" weight parameter in the relevance scoring model. This means that the next time a user searches for similar products, even if the text matching degree of the data returned by each platform is the same, the results from the First E-commerce app will automatically rank higher due to its higher weight. This continuous optimization based on real user behavior feedback allows the search system to evolve in a personalized way as user habits change, rather than remaining at the level of mechanical matching of static rules.

[0074] In summary, this embodiment, through a complete cross-e-commerce retrieval process, verifies the synergistic effect of field-level access control, content fingerprint deduplication, precise bounce, and adaptive sorting technologies in a real-world business scenario. These mechanisms are not isolated but interdependent and mutually supportive: field-level anonymization provides a secure and clean data foundation for fingerprint calculation; fingerprint deduplication reduces noise interference for adaptive sorting; and precise bounce completes the functional loop of the entire retrieval experience. It is this systematic integration of technologies that enables this application to provide efficient, accurate, and personalized cross-application data retrieval services while ensuring user privacy and security.

[0075] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected," "coupled," or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0076] Although preferred embodiments have been described in this application, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the embodiments in this application are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments in this application.

[0077] Obviously, those skilled in the art can make various modifications and variations to the embodiments of this application without departing from the spirit and scope of the embodiments of this application. Therefore, if these modifications and variations to the embodiments of this application fall within the scope of the embodiments of this application and their equivalents, then these modifications and variations are also intended to be included in the embodiments of this application.

Claims

1. A cross-application data retrieval method, characterized in that, The method includes: In response to receiving a retrieval request, a retrieval task is distributed to multiple target applications to obtain candidate data from multiple target applications; Based on field-level access control policies, restricted fields in the candidate data are de-identified or removed to obtain processed candidate data. Content fingerprints are extracted from the processed candidate data, and the candidate data from different target applications are subjected to homology determination and deduplication based on the content fingerprints to obtain aggregated retrieval results. Output the aggregated search results.

2. The method as described in claim 1, characterized in that, Based on field-level access control policies, the restricted fields in the candidate data are de-identified or removed, including: Obtain the field permission configuration information corresponding to each of the target applications; Based on the field permission configuration information, the data fields in the candidate data are divided into searchable fields and restricted fields; Perform data desensitization or field removal operations on the restricted fields.

3. The method as described in claim 1, characterized in that, Content fingerprints are extracted from the processed candidate data, and the candidate data from different target applications are subjected to homology determination and deduplication based on the content fingerprints, including: Extract key attribute features from the processed candidate data to generate the content fingerprint; Calculate the fingerprint similarity between the processed candidate data from different target applications; When the fingerprint similarity meets the preset similarity condition and the corresponding timestamp difference is within the preset time window, the corresponding processed candidate data is determined to be source data, and deduplication is performed on the source data.

4. The method as described in claim 2, characterized in that, Performing data desensitization or field removal operations on the restricted fields includes at least one of the following: Masking and replacing sensitive characters in the restricted field; The restricted fields are hashed. Set the field value of the restricted field to null; The method further includes: In response to receiving a fine-grained authorization instruction for the target application, the field permission configuration information is dynamically updated to adjust the scope of the searchable fields and the restricted fields.

5. The method as described in claim 3, characterized in that, Performing deduplication on the aforementioned homogeneous data includes: Calculate the relevance score of each candidate data in the source data; The candidate data with the highest relevance score is retained as the target display data; Fold the candidate data in the same source data, excluding the target display data, to generate a folded display identifier; The output of the aggregated search results includes: The target data is displayed on the search results page, and the collapsed display icon is displayed in association. In response to a trigger operation on the collapsed display icon, the other candidate data that has been collapsed is expanded and displayed.

6. The method as described in claim 1, characterized in that, Distribute retrieval tasks to multiple target applications to obtain candidate data from multiple target applications, including: The retrieval task is distributed to multiple target applications using a standardized data extraction adapter; Receive the raw application data returned by the multiple target applications; The candidate data is obtained by converting the original application data into a unified data structure using the standardized data extraction adapter. The unified data structure includes a content identifier field, a source identifier field, and a content summary field.

7. The method as described in claim 6, characterized in that, The unified data structure also includes page positioning parameters; The method further includes: In response to a trigger operation on a target result item in the aggregated search results, the page positioning parameters corresponding to the target result item are obtained; By using a redirect protocol adapter, the page positioning parameters are converted into a dedicated redirect protocol for the source target application corresponding to the target result item; Based on the exclusive redirection protocol, the control terminal device redirects to the content details page in the source target application that corresponds to the page positioning parameters.

8. A cross-application data retrieval device, characterized in that, include: The data acquisition module is used to respond to a received retrieval request by distributing retrieval tasks to multiple target applications in order to obtain candidate data from multiple target applications; The field processing module is used to perform desensitization or removal processing on restricted fields in the candidate data based on field-level access control policies to obtain processed candidate data. The fingerprint deduplication module is used to extract content fingerprints based on the processed candidate data, and to perform homology determination and deduplication processing on candidate data from different target applications based on the content fingerprints to obtain aggregated retrieval results. The result output module is used to output the aggregated search results.

9. A terminal device, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.