Multi-attribute secret state data efficient matching and self-adaptive task allocation method and system supporting dynamic updating

Abstract

The invention relates to a multi-attribute secret state data efficient matching and self-adaptive task allocation method and system supporting dynamic updating, and the method comprises the steps: a data outsourcing step: a crowdsourcing platform constructs an index tree fusing multiple attributes of space, time and keywords, encrypts the index tree, and outsources the encrypted index tree to a cloud server; a privacy protection query step: a worker submits an encrypted query token, and the double cloud servers cooperatively traverse the index tree in a ciphertext state by executing a security comparison protocol and a security subset query protocol; and a dynamic updating step, wherein the double cloud servers support operations such as task insertion, task deletion and node splitting on the index tree in a ciphertext state. According to the crowdsourcing task allocation method, comprehensive privacy protection on the task data and the query request is realized in an untrusted cloud environment, meanwhile, efficient multi-attribute joint retrieval and dynamic data updating are supported, and the problems of single privacy protection dimension, low calculation efficiency and insufficient dynamic updating support in an existing crowdsourcing task allocation scheme are effectively solved.

Description

TECHNICAL FIELD

[0002] The application belongs to the field of information technology, and particularly relates to a multi-attribute encrypted data efficient matching and adaptive task allocation method and system supporting dynamic update. BACKGROUND

[0004] With the rapid development of mobile devices and wireless network technology, crowdsourcing application as an innovative distributed service mode has been widely concerned. By 2025, the global mobile Internet users will break through 5 billion, which provides huge development potential for crowdsourcing application to fully utilize extensive distributed resources for large-scale task processing. In crowdsourcing application, task publishers publish various tasks to ensure the tasks can be completed smoothly by describing the space, time and keyword information required by the tasks. For example, Taobao flash shopping is a platform specially designed for part-time workers. The platform allocates tasks through the spatial data and interest keywords of the workers. After completing the task, the worker arrives at the designated location to pick up the package and obtains the corresponding remuneration. Figure 1 As shown in the figure, it is assumed that the task set is . The worker queries the tasks in the area [(17, 23), (30, 32)], the expected work date is , and the keyword is "photographer". The worker submits the query to the crowdsourcing platform, the platform performs task matching according to the query, and returns the related tasks. Finally, the worker is allocated the task .With the popularity of crowdsourcing services, the sharp increase in task volume puts a huge performance and cost pressure on the platform. Therefore, many crowdsourcing platforms choose to outsource task data and matching computation to cloud servers to improve platform efficiency and reduce operating costs. Currently, existing research work mainly faces three key problems: data privacy leakage, multi-attribute task processing, and inability to efficiently cope with dynamic task updates. Since the cloud server is not fully trusted, protecting the privacy of participants is particularly important in task allocation, especially the sensitive data of workers such as spatial information, social relationships, and interests. Although a variety of privacy protection technologies have been proposed to ensure the security of crowdsourcing platforms, these solutions mostly focus on protecting spatial data, while ignoring other attributes of tasks. For example, task allocation efficiency may be affected by the mismatch between workers' interests and availability. To address the above challenges, existing spatial keyword search techniques support keyword-based spatial data outsourcing. However, these solutions are often limited to processing single attributes or time information, limiting the flexibility of task allocation and making it difficult to adapt to dynamic task environments. In addition to privacy protection and multi-attribute task allocation, crowdsourcing platforms also need to meet the demand for dynamic task data updates. Many existing spatial keyword search methods focus on static data and lack support for dynamic task updates. To improve platform efficiency and reduce operating costs, many platforms choose to outsource task data and computation to cloud servers, but the trustworthiness of cloud servers makes the protection of personal data of task publishers and workers particularly critical. Although a variety of privacy protection technologies have been applied to secure crowdsourcing, there are still many technical problems that need to be addressed.

[0005] In the field of privacy protection, researchers have proposed a variety of technical methods to ensure data security, which can be divided into the following three methods:

[0006] (1) Differential privacy technology protects data privacy by introducing noise, ensuring that even if an external attacker obtains part of the data, it is still impossible to infer the specific information of an individual. The core idea is to add noise to the spatial data of participants and tasks when submitting to a trusted server, masking the accuracy of the data, thereby preventing accurate identification of individual spatial data; when querying or analyzing data, add certain noise to the output results, thereby breaking the influence of individual data items on the results. For example, in crowdsourcing tasks, if the location information of task participants is processed with differential privacy, it can be ensured that even if the cloud server processes a large amount of task data, an attacker will be difficult to infer the specific location of an individual by analyzing the query results, thereby protecting the privacy of workers. However, differential privacy technology, while effectively preventing data leakage, often faces a trade-off between privacy protection and query accuracy in practical applications.

[0007] (2) Spatial obfuscation technology enhances privacy by mapping workers' spatial data to a larger obfuscated area. This technology expands the scope of location information and obfuscates specific location details, making it difficult for external attackers to deduce the accurate physical location even if the data is leaked. However, while spatial obfuscation can effectively increase privacy and security, it usually leads to a decrease in task allocation efficiency. For example, if a task is assigned to a large virtual area, multiple workers may meet the task requirements, but to ensure privacy, the task matching process must take into account the intersection of these virtual areas, resulting in a significant increase in computational load and matching complexity. Therefore, how to balance privacy protection and task allocation efficiency has become the main challenge in the application of this technology.

[0008] (3) In encryption-based methods, encryption technology transforms data into ciphertext, ensuring that even if the data is acquired by a third party, sensitive information cannot be read. Task privacy is ensured by performing task matching and data processing in the ciphertext domain. For example, spatial data of tasks and workers can be transmitted after homomorphic encryption. Even without decryption, the cloud server can still perform distance calculations and task matching, thus supporting privacy-preserving task allocation. However, the main problem with encryption-based schemes is the high computational cost, especially when dealing with large-scale task and worker data. Ensuring efficient computation and low-latency response becomes a major challenge in practical applications. In the field of privacy-preserving crowdsourcing task allocation, Kamel et al. proposed a dynamic spatial data retrieval scheme based on HR trees, specifically for range queries in encrypted cloud databases. This scheme successfully ensures data confidentiality by applying spatial transformation and order-preserving encryption techniques. However, while order-preserving encryption maintains the orderliness of the data, it also exposes the sequential relationships between records. Furthermore, this scheme does not consider task allocation based on multiple attributes, thus having certain limitations when dealing with complex tasks. Yan et al. designed another privacy-preserving scheme based on the Counting Bloom Filter (CBF) technique, supporting efficient fuzzy queries and multi-keyword searches, and dynamically updating the dataset. This scheme performs well when handling dynamic data and effectively solves the privacy protection problem in multi-keyword queries. Xia et al. proposed a multi-keyword ranking query scheme, which organizes records into a binary tree structure to maintain keyword balance and enhances data security. This method improves accuracy and strengthens privacy protection during data retrieval.

[0009] Although existing privacy protection schemes have made some progress, they still have some significant limitations in complex crowdsourcing scenarios. (1) First, many schemes rely on symmetric encryption technology, which faces challenges in key sharing and access control in environments with multiple publishers, potentially leading to security vulnerabilities. (2) Second, current technologies focus on processing single-attribute data (such as spatial data, time, or keywords), while providing insufficient support for multi-attribute task allocation. Especially in practical applications, task allocation usually involves multiple dimensions of attributes, such as space, time, and keywords, requiring schemes to effectively handle complex multi-attribute data. (3) Finally, although some technologies have made progress in privacy protection, there is still a gap between them and the needs of practical applications. For example, how to achieve real-time task allocation and dynamic data updates while ensuring privacy is a problem that many existing schemes have not fully solved. Therefore, designing a privacy protection scheme that combines symmetric and asymmetric encryption technologies, can efficiently support the processing and updating of multi-attribute data, and meet the privacy protection needs in practical crowdsourcing scenarios remains a major challenge in current research. Summary of the Invention

[0011] To address the problems existing in the prior art, the purpose of this invention is to provide a method and system for efficient matching and adaptive task allocation of dynamically updated multi-attribute dense data, which is specifically achieved through the following technical solutions:

[0012] A method for efficient matching and adaptive task allocation of dynamically updated multi-attribute dense data is provided. This method is applied to a system including a crowdsourcing platform, workers, a first cloud server, and a second cloud server. The method includes the following steps:

[0013] 1) The data outsourcing step is executed by the crowdsourcing platform;

[0014] A Hilbert multi-attribute R-tree is constructed based on the task dataset, wherein each task data in the task dataset contains spatial location, time, and keyword set; the leaf nodes of the Hilbert multi-attribute R-tree store the Hilbert spatial code value, time, and keyword set of the corresponding task data, and its internal nodes store the maximum Hilbert value, latest time, and count Bloom filter generated based on all keyword sets in the subtree of the task data in its subtree.

[0015] The attribute information stored in each node of the Hilbert multi-attribute R-tree is encrypted using a homomorphic encryption algorithm to generate a dense Hilbert multi-attribute R-tree, which is then outsourced to the first cloud server.

[0016] The master key for the homomorphic encryption algorithm is sent to the second cloud server for storage.

[0017] 2) Privacy protection query steps. These are executed collaboratively by the worker and the first and second cloud servers.

[0018] Workers generate encrypted query tokens based on query requests. These tokens contain a set of homomorphically encrypted query space range Hilbert values, query time, and a Bloom filter corresponding to the query keyword set. The first and second cloud servers collaboratively traverse the encrypted Hilbert multi-attribute R-tree based on the encrypted query tokens. By executing a secure comparison protocol and a secure subset query protocol, they perform multi-attribute matching and pruning in the encrypted state, filtering out encrypted task data that meets the query conditions and returning it as the query result. 3) Dynamic update step: This is collaboratively executed by the crowdsourcing platform, the first cloud server, and the second cloud server. In response to the task data update request, the first and second cloud servers collaboratively locate the target node on the encrypted Hilbert multi-attribute R-tree based on the encrypted task data to be updated. They then insert or delete the node in the encrypted state and update the encrypted aggregate information of the relevant internal nodes.

[0019] Furthermore, the construction of the Hilbert multi-attribute R-tree specifically involves: encoding the two-dimensional spatial position of each task data into a one-dimensional Hilbert value using the Hilbert space-filling curve; sorting all task data according to their Hilbert values ​​and dividing and aggregating them based on a preset node capacity threshold to construct a tree-like index structure; wherein, a counting Bloom filter is constructed for each internal node to aggregate the set of keywords representing all task data in its subtree.

[0020] Furthermore, a homomorphic encryption algorithm is used to encrypt the attribute information stored in each node of the Hilbert multi-attribute R-tree. Specifically, for each task data in the leaf node, the Hilbert value and time are encrypted using a homomorphic encryption algorithm, the key set is mapped to a Bloom filter and then homomorphically encrypted, and the detailed content is encrypted using a symmetric encryption algorithm. For internal nodes, the maximum Hilbert value, the latest time, and the count Bloom filter stored in them are encrypted using a homomorphic encryption algorithm.

[0021] Furthermore, the secure subset query protocol is used to determine whether the query keyword set is a subset of the task data keyword set in the encrypted state, and its execution process includes:

[0022] The encrypted task data keyword Bloom filter is converted into a ciphertext vector in the format of a simulated ternary content-addressable memory, where each bit of the vector is an encrypted "0", "1" or wildcard "*".

[0023] Use an encrypted query keyword Bloom filter as the query vector;

[0024] A secure ternary matching protocol is executed to compare the ciphertext vector with the query vector bit by bit, and output a ciphertext result indicating whether the subset relationship is valid.

[0025] Furthermore, the collaborative traversal of the dense Hilbert multi-attribute R-tree and the performance of multi-attribute matching and pruning specifically include:

[0026] Node pruning: When traversing internal nodes, the first cloud server and the second cloud server work together to perform the following:

[0027] 5-1) Based on the minimum Hilbert value ciphertext of the query spatial range set and the maximum Hilbert value ciphertext of the internal node, perform a safe less than comparison to determine whether the spatial ranges may intersect.

[0028] 5-2) Based on the ciphertext of the query time and the latest ciphertext of the internal node, perform a safe less than comparison to determine whether the time condition can be met;

[0029] 5-3) Calculate the inner product of the count Bloom filter ciphertext of the internal node and the query keyword Bloom filter ciphertext, and perform a secure comparison to determine whether the keyword set may contain it;

[0030] Only when the results of the judgments under the ciphertext for conditions 5-1), 5-2), and 5-3) are all possible to be true, will the child nodes of the node continue to be traversed;

[0031] Leaf node verification: For each dense-state task data in the leaf nodes reached during traversal, the first cloud server and the second cloud server jointly perform the following:

[0032] 5-4) Verify whether its Hilbert value ciphertext is equal to the ciphertext of a value in the query Hilbert value set;

[0033] 5-5) Verify whether the ciphertext of the query time is less than its ciphertext of the time;

[0034] 5-6) Verify whether the set of query keywords is a subset of its set of keywords using the security subset query protocol.

[0035] The encrypted details of the task data that simultaneously meet the three conditions 5-4), 5-5), and 5-6) will be added to the query results.

[0036] Furthermore, in the dynamic update step, the insertion or deletion of nodes is completed in the ciphertext state, specifically including: when the insertion of new task data ciphertext causes the leaf node to overflow, the ciphertext task data in the node is split according to the Hilbert value to form two new leaf nodes; and the maximum Hilbert value ciphertext, the latest time ciphertext, and the count Bloom filter ciphertext of all internal nodes on the path from the affected leaf node to the root node are updated from bottom to top.

[0037] A system for efficient matching and adaptive task allocation of dynamically updated multi-attribute dense-state data includes a crowdsourcing platform, workers, a first cloud server, and a second cloud server. The crowdsourcing platform comprises: a data processing module for acquiring a task dataset, mapping the spatial location attributes of each task data to Hilbert values ​​based on Hilbert curves, and constructing a Hilbert multi-attribute R-tree based on the Hilbert values; a data encryption module for encrypting the Hilbert multi-attribute R-tree using a homomorphic encryption algorithm to generate a dense-state Hilbert multi-attribute R-tree; and a key management module for generating the key for the homomorphic encryption algorithm and storing the dense-state Hilbert multi-attribute R-tree. The query request is sent to the first cloud server, and the master key is sent to the second cloud server. The worker includes: a query generation module, used to generate an encrypted query token based on the query request, the encrypted query token containing an encrypted spatial range Hilbert value set, a time condition, and a Bloom filter for the query keyword set; a result decryption module, used to decrypt the received query result; the first cloud server is used to store the encrypted Hilbert multi-attribute R-tree; the second cloud server is used to store the master key; wherein, the first cloud server and the second cloud server are configured to work together to execute the query step and the dynamic update step in the above method.

[0038] The technical solution of this invention combines multiple attributes such as space constraints, time, and keywords, while ensuring the privacy of multiple attributes and supporting high-performance adaptive task allocation in crowdsourcing platforms. This paper proposes an innovative Hilbert multi-attribute R-tree (HRAT) that supports dynamic task updates by encoding spatial data using Hilbert curves and mapping keywords to Bloom filters and counted Bloom filters. By introducing an improved symmetric homomorphic encryption scheme (iSHE) to encrypt spatial, temporal, and keyword attributes, a unified encrypted index structure is established, ensuring privacy while improving computational and communication efficiency. Furthermore, by mapping keywords to Bloom filters and counted Bloom filters, an innovative method is proposed to transform subset queries into Bloom filter inner product operations and simulated ternary content-addressable memory (TCAM) matching, significantly improving encrypted query efficiency. A secure ternary matching protocol is designed by integrating the iSHE encryption comparison protocol and simulated TCAM, effectively ensuring privacy during Bloom filter matching. Finally, a secure subset query scheme that guarantees data and query privacy is constructed by integrating the ternary matching protocol with a Bloom filter-based subset query protocol. Based on the Hilbert multi-attribute R-tree, ternary matching protocol, and secure subset query scheme, an efficient privacy-preserving dynamic multi-attribute task search algorithm is designed. This invention improves the efficiency of task search while providing privacy protection. Through performance evaluation and security analysis, its high efficiency in practical applications has been verified, and the designed scheme has been proven to meet the security requirements against selection query attacks. Attached Figure Description

[0040] Figure 1 This is a diagram illustrating crowdsourcing in existing technologies.

[0041] Figure 2 This is a system model diagram of the present invention;

[0042] Figure 3 This is a system architecture diagram of the present invention;

[0043] Figure 4 This is a schematic diagram of the HR tree structure of the present invention. Detailed Implementation

[0045] The present invention will be further described below with reference to specific embodiments in order to better understand the technical solution.

[0046] Example 1

[0047] The system architecture of this invention is as follows: Figure 2 As shown, a typical multi-attribute dense-state data matching and task allocation model was designed, which includes the following three main participants:

[0048] (1) Crowdsourcing platform (CP): responsible for managing a large number of task records Each task record includes spatial location, time, and keyword attributes for the task, specifically represented as follows: ,in Indicates the spatial location of the task. The time required to complete the task. This is the set of keywords for the task. Due to the limited computing and storage capabilities of crowdsourcing platforms, task data is outsourced to cloud servers, which provide query services for workers. Meanwhile, the CP also acts as a trusted third party for key generation and distribution.

[0049] (2) Workers: The group of workers Query requests can be submitted based on the task's spatial and temporal information, as well as relevant keywords. The query request format submitted by the worker is as follows: ,in The defined rectangular area represents the spatial range for the query. and These represent the coordinates of the lower left and upper right corners of the area, respectively. For workers' preferred time, This is a set of keywords representing workers' interests. The task query returns all tasks that meet the following criteria: 1) That is, the task location is within the query area; 2) That is, the task completion time is later than the worker's preferred time; 3) That is, the set of keywords for the task contains keywords that the workers are interested in.

[0050] (3) Cloud Server (CS): including and These two servers possess highly efficient storage and computing capabilities. Store the dense task dataset and index provided by CP, and The server is responsible for storing the decryption key. The two servers collaborate to provide workers and content providers (CPs) with multi-attribute task query services and dynamic task update services. This model utilizes a dual-cloud architecture to enhance privacy protection. In this design, encrypted data and keys are stored on two separate cloud servers, ensuring that neither server can independently reconstruct private information.

[0051] In the system model of this invention, the data owner, namely the crowdsourcing platform CP, is regarded as a trusted entity because it possesses the complete dataset and lacks the motivation to violate privacy. The worker nodes are assumed to be honest but curious, that is, they perform the assigned tasks according to the protocol, but may try to infer additional information by analyzing the query results. Similarly, cloud servers CS1 and CS2 follow the assumption of being honest but curious: they correctly perform task query and update operations, but may try to access sensitive information, such as task data, query content or results, etc. It is further assumed that there is no collusion relationship between CS1 and CS2. This assumption is widely used in the field of secure computing because collusion between servers can seriously damage the credibility and reputation of the system. The goal of this invention is to build an efficient, privacy-secure, dynamically updatable crowdsourcing task allocation scheme that supports multi-attribute queries. Its core design goals include: (1) Privacy protection: ensuring that cloud servers CS1 and CS2 cannot obtain or infer any sensitive information, including task details, query content and results, thereby preventing privacy leakage. (2) Multi-attribute and multi-keyword search: The task allocation process should comprehensively consider multi-dimensional attributes such as space, time and keywords to improve the matching accuracy between tasks and workers, while supporting multi-keyword queries to improve the flexibility and adaptability of the system. (3) Efficiency: Under the premise of ensuring privacy and security, the system should balance computational efficiency and communication overhead to ensure availability and real-time response capabilities in actual crowdsourcing environments. (4) Dynamic updability: The system needs to support dynamic changes in task data, such as the addition, modification and deletion of tasks, to ensure stable and efficient query performance in actual scenarios where tasks change frequently.

[0052] The system consists of the following 9 modules, and the system architecture diagram is as follows: Figure 3 As shown:

[0053] (1) Pre-processing module

[0054] This module includes Hilbert curves, Hilbert R-trees, Bloom filters, count Bloom filters, ternary content-addressable memory, and the iSHE homomorphic encryption scheme.

[0055] 1) Hilbert curve: The Hilbert curve (H curve) is a technique for mapping multidimensional space to a one-dimensional sequence. In the grid, The Hilbert curve assigns a unique number to each grid cell, generating values ​​ranging from 0 to 1. The Hilbert value of a Hilbert curve is uniquely determined by parameters such as its starting point, direction, granularity, turning rules, and order. Once specified, these parameters define the curve's structure. Granularity, in particular, determines the number of points with the same Hilbert value within each grid cell. Higher-order curves are constructed by applying transformations such as rotation and reflection to lower-order curves. The Hilbert R-tree (HR tree) is an indexing method that combines the features of R-trees and B+ trees, enabling efficient querying of high-dimensional data and adapting to dynamic operations. Its structural diagram is shown below. Figure 4 As shown. The structure and node arrangement of the HR tree follow the order defined by the H curve; the construction process of the HR tree is summarized as follows: a) High-dimensional data points are converted into one-dimensional H values ​​using the H curve, thereby ensuring that the spatial relationship between adjacent points is preserved; b) These H values ​​are sorted, which can be done using algorithms such as quicksort; c) The sorted data is divided into child nodes according to the granularity of the H curve, and the nodes are gradually merged until the root node is formed; d) In the HR tree, nodes are divided into internal nodes. Leaf nodes Each Store the maximum H value in its subtree ( ) and pointers to its child nodes, leaf nodes Save the H value for each data point;

[0056] 2) Bloom filters and counting Bloom filters. A Bloom filter (BF) is a space-efficient probabilistic structure used to check the membership of a set. Each element is mapped to a Bit array Initially, ,and Through application hash functions , All elements are hashed to a specific position, corresponding to the bit. ,in For all elements, although the BF hash function has high memory utilization, it may produce false positives, incorrectly classifying an element as belonging to the set. The probability of such false positives is given by the following formula: The CBF (Cellular Bloom Filter) enhances the standard Bloom filter (BF) by replacing each bit with a counter. When an element is added, the counter at the corresponding hash position increments. To query elements Does it belong to Only need to verify for ,all Whether this is true, if any counter equals 0, then ,otherwise, The confidence level is very high;

[0057] 3) Tri-state Content-Addressable Memory (TCAM): TCAM is an associative memory designed to support the use of binary values ​​(0 and 1) as well as wildcards. ) Perform storage and search operations. Wildcards It can represent binary states or other wildcards. Given a hash function... It will collect Mapped to A binary vector. Let express The Bit. Furthermore, definition. ,in It is the first A ternary value, when hour, ;otherwise, For sets , It is A vector of position 1, whose first position 2 Position Given. Represents the insertion set. The standard BF is obtained after considering all elements. Similarly, It is Bit Vector, where the first Bit set to ,if Otherwise, set it to 0. Consider a set J and a query set Y. First, compute the bit vector. For J, the corresponding... And stored in the simulated TCAM. Using Query TCAM if and only if Matching is triggered on time, that is: .

[0058] (2) System initialization module

[0059] During the system startup phase, the crowdsourcing platform needs to complete the initialization process of key generation and distribution. The specific steps are as follows:

[0060] 1) Parameter setting and hash function initialization: The platform first inputs the security parameters of iSHE. And randomly select γ hash functions Subsequently, the AES algorithm is used to generate a symmetric key. This key is used for rapid encryption of subsequent data blocks and communication protection;

[0061] 2) Key generation and ciphertext initialization, based on the input security parameters. The system generates the iSHE master key. The initial ciphertext set is obtained according to the iSHE encryption algorithm. This is used for the encryption and verification of data in subsequent tasks.

[0062] 3) Key distribution and secure transmission: Through a trusted communication channel, the crowdsourcing platform will... Securely sent to cloud server CS1 for storage and computing operations; simultaneously, The key is sent to CS2 to support the decryption service; finally, the platform distributes the public key to all workers. This ensures that all participants can perform encrypted query and verification operations within the encrypted domain.

[0063] (3) Data organization module

[0064] After system initialization, the crowdsourcing platform needs to organize and build an index structure for the task data to support efficient spatial and multi-attribute queries. The specific steps are as follows:

[0065] 1) Spatial mapping and grid encoding: First, map areas are mapped... Divided into Each equal-scale sub-unit is used to establish... Hilbert space-filling curve for each task Geographical location It is mapped to the corresponding grid cell, and its spatial index, i.e., the Hilbert value, is calculated. This encoding process can maintain spatial proximity, so that spatially similar tasks also have high clustering in the index, thereby optimizing the performance of subsequent queries and updates.

[0066] 2) Index structure generation and data sorting: Based on the obtained Hilbert sequence, sort the task set... Tasks in According to Hilbert value Sort the data in ascending order; based on this, construct a Hilbert multi-attribute R-tree, denoted as […]. During the node insertion phase, the time attribute of each task is... and keyword set Store them together in the corresponding index node, so that It can simultaneously support joint queries based on spatial, temporal, and semantic keywords. Compared with traditional R-trees, this structure not only maintains spatial continuity but also enhances the ability to express multidimensional attributes;

[0067] 3) Leaf node data storage: At the leaf node level of the index tree, the system records the set of all task space codes within the corresponding region. Its corresponding time attribute and keyword set Each leaf node stores at most The task records are used to control node load and ensure retrieval efficiency. These nodes are the basic operation units for subsequent task search and dynamic updates.

[0068] 4) Internal node aggregation information maintenance in Hilbert multi-attribute R-trees Within the internal nodes, the system stores statistical and aggregated information about the subtrees, including: the maximum Hilbert value in the subtree. Used for spatial range determination, the most recent task timestamp Supports time-constrained filtering and the union of all child node keyword sets. This is used for fast keyword matching and pre-screening. This summary information can be used for efficient pruning when traversing the index, reducing access to irrelevant nodes and thus significantly improving query performance.

[0069] (4) Data encryption module

[0070] The crowdsourcing platform performs the following operations to encrypt a Hilbert multi-attribute R-tree:

[0071] 1) Encrypt the Hilbert value and time data of each task in the leaf nodes using an improved symmetric homomorphic encryption scheme to obtain... and ;

[0072] 2) Use Each hash function will assign a key set to each task. Mapped to a Bloom filter, the key Bloom filter is encrypted using an improved symmetric homomorphic encryption scheme. generate ;

[0073] 3) Use AES to encrypt task information ,generate Finally, each dense state task in the Hilbert multi-attribute R leaf node is obtained. = ;

[0074] 4) Encrypt the maximum Hilbert value in the internal nodes using an improved symmetric homomorphic encryption scheme. Recent temporary data Generate ciphertext and ;

[0075] 5) Map all keys in the subtree to the CBF and encrypt them to generate ciphertext. ,in This represents the number of tasks within the subtree;

[0076] 6) Store dense leaf nodes To Hilbert multi-attribute R-tree, where For node capacity, store dense internal nodes. From the Hilbert multi-attribute R-tree, generate the ciphertext table of the Hilbert multi-attribute R-tree. And outsourced to .

[0077] (5) Query token generation module

[0078] The worker performs the following operations to generate a query token:

[0079] 1) Query range Mapping onto a Hilbert curve generates a set Each Hilbert value is stored in a Hilbert set of spatial data. middle;

[0080] 2) Run the symmetric homomorphic encryption scheme to generate the encryption algorithm. and , set of keywords Mapped to BF, the keyword Bloom filter is encrypted using a symmetric homomorphic encryption scheme to generate encrypted data. ;

[0081] 3) Select a session key The session key is encrypted using a symmetric homomorphic encryption scheme. Encrypt and generate ciphertext. ;

[0082] 4) Send query token Give cloud server .

[0083] (6) Ciphertext Subset Judgment Module

[0084] cloud server With cloud servers Collaborative execution of secure subset queries, verification Is it true or false?

[0085] 1) Consider and for Vector, definition ;

[0086] 2) To Each member jointly executes the ciphertext equality protocol based on a symmetric homomorphic encryption scheme. get Each digit is represented as ,in , It is an encrypted wildcard in TCAM ;

[0087] 3) Secure ternary matching protocol ,enter: and Output: ):

[0088] a) Initialization ;

[0089] b) For Each bit in the ciphertext is equal if the ciphertext equality protocol based on the symmetric homomorphic encryption scheme satisfies ,show and The The capital and Correspondingly, it means the first If the positions match, continue iterating backwards;

[0090] c) If ,express match or If so, continue searching backwards;

[0091] d) Otherwise set ;

[0092] e) Return Give to cloud server;

[0093] 4) Jointly execute a secure ternary matching protocol ,in ,if The output is ,show Then output Otherwise, output .

[0094] (7) Multi-attribute task search module

[0095] cloud server With cloud servers To collaboratively complete a secure query, follow these steps to obtain the result set. :

[0096] 1) Jointly decrypt the messages sent by the user Obtain the session key ;

[0097] 2) Starting from the root node, perform a collaborative traversal of the encrypted Hilbert multi-attribute R-tree until a leaf node is encountered. Then it contains encrypted task information. Add to candidate task set In the middle, encountering internal nodes If the following conditions are met, the branch is retained and the traversal continues downward:

[0098] a) Jointly implement a ciphertext-less-than-comparison protocol based on a symmetric homomorphic encryption scheme. ,mean If any of the child nodes contain tasks that fall within the worker's spatial query range, then output... ;

[0099] b) Jointly implement a ciphertext-to-comparison protocol based on a symmetric homomorphic encryption scheme. This indicates that the worker's expected working time is earlier than the task completion time, so the output is... ;

[0100] c) Jointly implement a ciphertext-less-than-comparison protocol based on a symmetric homomorphic encryption scheme. ,in The calculation steps are as follows:

[0101] i) Right and Perform the inner product operation to obtain ,in Indicate the lengths of BF and CBF;

[0102] ii) Jointly implement the ciphertext-less-than-comparison protocol based on a symmetric homomorphic encryption scheme. ,show If the child nodes of a function may contain the keywords that the worker expects, then the output will be... ;

[0103] 3) Generate candidate set ,in express The number of tasks in the process;

[0104] 4) Collaboratively verify candidate records Medium task Does the attribute meet the query criteria? :

[0105] a) Execute a ciphertext-equals comparison protocol based on a symmetric homomorphic encryption scheme. If the protocol output This indicates that the scope of space accepted by workers includes tasks. If the spatial data is not available, then the verification process continues.

[0106] b) Execute the ciphertext-less-comparison protocol based on a symmetric homomorphic encryption scheme. If the protocol output is This indicates that workers can meet [the requirements]. If the completion time is not specified, the verification process will continue.

[0107] c) Execute the ciphertext equals comparison protocol based on a symmetric homomorphic encryption scheme. ciphertext subset judgment module If the ciphertext subset judgment module The output is This indicates that a task exists. Keywords that can cover workers If so, then this task is set as a match and added to the result set. ;

[0108] 5) Execute the AES encryption algorithm respectively. And generate an encrypted result set. ;

[0109] 6) Send the encrypted result set For workers.

[0110] (8) Query result recovery module

[0111] Workers received Then, the negotiated key decryption algorithm and AES decryption algorithm are used to decrypt and recover the task information. .

[0112] (9) Task dynamic update module

[0113] Crowdsourcing platform issues update request task The dual cloud servers encrypt the spatial data ciphertext of the encryption task. Perform the following steps on the encrypted Hilbert multi-attribute R-tree Perform task insertion and task deletion / update separately:

[0114] Task: Insert / Update, Input: Tree Index and secret mission information Output: Updated tree

[0115] 1) Input dense state task information According to dense space data The multi-attribute task search module is executed to determine the affected leaf nodes. In dense Hilbert multi-attribute R-trees The position in the middle, from the root node Start locating the affected leaf nodes Select the layer that satisfies And has the smallest Continue traversing the subtree until the affected leaf node is reached. ;

[0116] 2) Compare leaf nodes Number of tasks With node capacity limit ,if Insert dense task information To leaf node In the middle, update the leaf nodes The ciphertext of the maximum Hilbert value in its parent node Ciphertext of the latest task completion time Modify each The impact Bit Iterate from bottom to top, layer by layer until the root node is reached;

[0117] 3) Compare leaf nodes Number of tasks With node capacity limit ,if Split leaf nodes The dense-state tasks are stored in two new leaf nodes. Inserting encrypted task information To the new leaf node Perform similar operations to update the tree from bottom to top until the root node;

[0118] Task: Delete / Update; Input: Tree Index and dense state tasks Output: Updated tree

[0119] 1) Input dense state task information According to dense space data The multi-attribute task search module is executed to determine the affected leaf nodes. In dense Hilbert multi-attribute R-trees The position in the middle, from the root node Start locating the affected leaf nodes Select the layer that satisfies And has the smallest Continue traversing the subtree until the affected leaf node is reached. ;

[0120] 2) Delete leaf nodes Secret state task information Update leaf nodes The ciphertext of the maximum Hilbert value in its parent node Ciphertext of the latest task completion time Modify each The impact Bit Iterates from bottom to top, layer by layer until the root node is reached.

[0121] Technical features of the system of this invention:

[0122] (1) Based on the construction of spatial multi-attribute index based on Hilbert curve, this invention first uses the Hilbert space filling curve to map the original two-dimensional coordinates into a one-dimensional sequence to maintain spatial proximity and reduce dimensional distortion. At the same time, the keyword information of the task is mapped to the Bloom filter (BF) and count Bloom filter (CBF) structures, thereby unifying spatial location, temporal attributes and semantic features under the same index system. The resulting Hilbert multi-attribute R-tree (HR-Tree) not only supports efficient storage of multi-dimensional attributes, but also allows dynamic insertion, deletion and updating of task nodes, ensuring the high scalability and adaptability of the system in the real-time crowdsourcing environment.

[0123] (2) Unified encrypted index structure and improved homomorphic encryption mechanism. This invention uses an improved symmetric homomorphic encryption algorithm (iSHE) to uniformly encrypt the spatial, temporal and keyword attributes of the task. Unlike traditional homomorphic schemes, the improved iSHE model optimizes the computational complexity and ciphertext length, which can significantly reduce communication and computation costs while maintaining the additivity of operations. This unified encrypted index structure allows comparison, matching and range judgment operations to be performed directly in the ciphertext field, realizing secure computation without decryption, thus taking into account both privacy protection and high query efficiency.

[0124] (3) Ciphertext subset query and inner product calculation fusion mechanism: In order to improve the performance of encrypted query, this invention embeds the keyword set into the binary vector space of Bloom filter and count Bloom filter, and designs a mechanism to transform subset query into ciphertext inner product operation. The server can calculate the keyword matching degree without decryption, and at the same time use the comparison logic of simulated ternary content addressable memory (TCAM) to complete fast filtering. This ciphertext-level subset matching strategy effectively reduces the number of interactions and computational overhead, making the query process both secure and real-time.

[0125] (4) Based on the design of a security protocol for ternary content matching, this invention combines the improved iSHE encryption comparison protocol with the simulated TCAM structure to construct a secure ternary matching protocol. This protocol can determine three logical relationships (matching, not matching, and fuzzy matching) in the encrypted state, supports high-concurrency queries under multiple keyword conditions, and can complete the comparison without leaking any internal state of the Bloom filter, thereby ensuring the privacy and security of the query process.

[0126] (5) Multi-level secure subset query mechanism: Based on the secure ternary matching protocol, this invention further integrates the encrypted subset query algorithm based on Bloom filter to construct a multi-level privacy protection scheme. This scheme not only prevents a single server from obtaining plaintext keywords or matching relationships, but also achieves minimal exposure of queries in a dual-server collaborative environment, ensuring bidirectional privacy and security of data and queries. Compared with traditional schemes, this mechanism has significantly improved in terms of security and scalability.

[0127] (6) Privacy-preserving multi-attribute task search algorithm for dynamic environments. Combining the above mechanisms, this invention finally forms an efficient privacy-preserving multi-attribute task search algorithm that supports dynamic updates. Under the framework of Hilbert multi-attribute R-tree, this algorithm combines protocols such as ciphertext comparison, ternary matching and secure subset verification, and can maintain a fast response in environments where tasks change frequently. Experimental and theoretical analysis show that the algorithm can ensure data and query privacy while having high query accuracy and low latency, meeting the dual requirements of privacy security and running performance in actual crowdsourcing scenarios.

Claims

1. A method for efficient matching and adaptive task allocation of dynamically updated multi-attribute dense-state data, characterized in that, This method is applied to a system including a crowdsourcing platform, workers, a first cloud server, and a second cloud server, and the method includes the following steps: 1) The data outsourcing step is executed by the crowdsourcing platform; A Hilbert multi-attribute R-tree is constructed based on the task dataset, wherein each task data in the task dataset contains spatial location, time, and keyword set; the leaf nodes of the Hilbert multi-attribute R-tree store the Hilbert spatial code value, time, and keyword set of the corresponding task data, and its internal nodes store the maximum Hilbert value, latest time, and count Bloom filter generated based on all keyword sets in the subtree of the task data in its subtree. The attribute information stored in each node of the Hilbert multi-attribute R-tree is encrypted using a homomorphic encryption algorithm to generate a dense Hilbert multi-attribute R-tree, which is then outsourced to the first cloud server. The master key for the homomorphic encryption algorithm is sent to the second cloud server for storage. 2) Privacy protection query steps. These are executed collaboratively by the worker and the first and second cloud servers. Workers generate encrypted query tokens based on query requests. The encrypted query tokens contain a set of Hilbert values ​​for the query space range that are homomorphically encrypted, the query time, and a Bloom filter corresponding to the set of query keywords. The first cloud server and the second cloud server collaboratively traverse the encrypted Hilbert multi-attribute R-tree based on the encrypted query token. By executing the secure comparison protocol and the secure subset query protocol, they perform multi-attribute matching and pruning in the encrypted state, and filter out the encrypted task data that meets the query conditions as the query result. 3) Dynamic update steps: executed collaboratively by the crowdsourcing platform, the first cloud server, and the second cloud server; In response to the task data update request, the first cloud server and the second cloud server work together to locate the target node on the ciphertext of the task data to be updated, and complete the insertion or deletion of the node in the ciphertext state, while updating the aggregate information ciphertext of the relevant internal nodes.

2. The method for efficient matching and adaptive task allocation of multi-attribute dense-state data supporting dynamic updates as described in claim 1, characterized in that, The construction of the Hilbert multi-attribute R-tree is specifically as follows: The two-dimensional spatial location of each task data is encoded into a one-dimensional Hilbert value using the Hilbert space-filling curve; Sort all task data according to Hilbert values, and divide and aggregate them according to preset node capacity thresholds to build a tree index structure; Specifically, a counting Bloom filter is constructed for each internal node to aggregate the set of keywords representing all task data in its subtree.

3. The method for efficient matching and adaptive task allocation of multi-attribute dense-state data supporting dynamic updates as described in claim 1, characterized in that, The attribute information stored in each node of the Hilbert multi-attribute R-tree is encrypted using a homomorphic encryption algorithm, specifically including: For each task data in a leaf node, its Hilbert value and time are encrypted using a homomorphic encryption algorithm, its key set is mapped to a Bloom filter and then homomorphically encrypted, and its detailed content is encrypted using a symmetric encryption algorithm; for internal nodes, its stored maximum Hilbert value, latest time, and count Bloom filter are encrypted using a homomorphic encryption algorithm.

4. The method for efficient matching and adaptive task allocation of multi-attribute dense-state data supporting dynamic updates as described in claim 1, characterized in that, The secure subset query protocol is used to determine, in encrypted form, whether the set of query keywords is a subset of the set of task data keywords. Its execution process includes: The encrypted task data keyword Bloom filter is converted into a ciphertext vector in the format of a simulated ternary content-addressable memory, where each bit of the vector is an encrypted "0", "1" or wildcard "*". Use an encrypted query keyword Bloom filter as the query vector; A secure ternary matching protocol is executed to compare the ciphertext vector with the query vector bit by bit, and output a ciphertext result indicating whether the subset relationship is valid.

5. The method for efficient matching and adaptive task allocation of multi-attribute dense-state data supporting dynamic updates as described in claim 1, characterized in that, The collaborative traversal of the dense Hilbert multi-attribute R-tree and the performance of multi-attribute matching and pruning specifically include: Node pruning: When traversing internal nodes, the first cloud server and the second cloud server work together to perform the following: 5-1) Based on the minimum Hilbert value ciphertext of the query spatial range set and the maximum Hilbert value ciphertext of the internal node, perform a safe less than comparison to determine whether the spatial ranges may intersect. 5-2) Based on the ciphertext of the query time and the latest ciphertext of the internal node, perform a safe less than comparison to determine whether the time condition can be met; 5-3) Calculate the inner product of the count Bloom filter ciphertext of the internal node and the query keyword Bloom filter ciphertext, and perform a secure comparison to determine whether the keyword set may contain it; Only when the results of the judgments under the ciphertext for conditions 5-1), 5-2), and 5-3) are all possible to be true, will the child nodes of the node continue to be traversed; Leaf node verification: For each dense-state task data in the leaf nodes reached during traversal, the first cloud server and the second cloud server jointly perform the following: 5-4) Verify whether its Hilbert value ciphertext is equal to the ciphertext of a value in the query Hilbert value set; 5-5) Verify whether the ciphertext of the query time is less than its ciphertext of the time; 5-6) Verify whether the set of query keywords is a subset of its set of keywords using the security subset query protocol. The encrypted details of the task data that simultaneously meet the three conditions 5-4), 5-5), and 5-6) will be added to the query results.

6. The method for efficient matching and adaptive task allocation of multi-attribute dense-state data supporting dynamic updates as described in claim 1, characterized in that, In the dynamic update step, the insertion or deletion of nodes is performed in the encrypted state, specifically including: When inserting new task data ciphertext causes a leaf node to overflow, the ciphertext task data within the node is split according to the Hilbert value to form two new leaf nodes. Update the maximum Hilbert value ciphertext, latest time ciphertext, and count Bloom filter ciphertext of all internal nodes on the path from the affected leaf node to the root node from bottom to top.

7. A system for efficient matching and adaptive task allocation of dynamically updated multi-attribute dense-state data, characterized in that, The system includes a crowdsourcing platform, workers, a first cloud server, and a second cloud server. The crowdsourcing platform includes: The data processing module is used to acquire task datasets, map the spatial location attributes of each task data to Hilbert values ​​based on Hilbert curves, and construct a Hilbert multi-attribute R-tree based on the Hilbert values. The data encryption module is used to encrypt the Hilbert multi-attribute R-tree using a homomorphic encryption algorithm to generate a dense Hilbert multi-attribute R-tree; The key management module is used to generate the key for the homomorphic encryption algorithm, send the encrypted Hilbert multi-attribute R-tree to the first cloud server, and send the master key to the second cloud server; The workers include: The query generation module is used to generate an encrypted query token based on the query request. The encrypted query token includes an encrypted set of Hilbert values ​​for spatial range, a time condition, and a Bloom filter for the set of query keywords. The result decryption module is used to decrypt the received query results; The first cloud server is used to store the dense Hilbert multi-attribute R-tree; The second cloud server is used to store the master key; The first cloud server and the second cloud server are configured to work together to perform the query step and the dynamic update step in the method as described in any one of claims 1 to 6.

Images