A blockchain data retrieval method for improving retrieval efficiency

By constructing feature models and index relationship lists and searching based on blockchain data features, the problem of inefficient multidimensional data retrieval is solved, efficient retrieval and adaptability are achieved, and retrieval efficiency and user experience are significantly improved.

CN115080637BActive Publication Date: 2024-12-06CHINA SHIP SCIENTIFIC RESEARCH CENTER +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210833019.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-07-14
Publication Date
2024-12-06
Estimated Expiration
2042-07-14

AI Technical Summary

Technical Problem

Existing blockchains are inefficient in multidimensional data retrieval. The traditional composite key indexing method needs to be searched one by one, which is inefficient and cannot directly retrieve data object sets that meet specific features.

Method used

By constructing a feature model and index relationship list based on historical search records, using feature models to build data groupings, providing data features based on data features, narrowing the query scope, and improving retrieval efficiency.

Benefits of technology

It achieves more than 90% of the search time savings, improves the efficiency of multi-dimensional feature query, breaks through the search efficiency limitations of the traditional blockchain K-V storage model, supports efficient fuzzy queries, and can adapt to changes in user habits.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115080637B_ABST
    Figure CN115080637B_ABST
Patent Text Reader

Abstract

The present application discloses a blockchain data retrieval method for improving retrieval efficiency, which relates to the field of blockchain technology. The method constructs a feature model that reflects user usage habits through a large number of historical retrieval records, and constructs an index relationship list based on the feature model. Multidimensional business data is associated with each data group in the index relationship list, and each data group has a corresponding feature index array. Therefore, by locating the data group that meets the conditions through data feature retrieval, it is possible to associate it with all multidimensional business data under the group. The retrieval conditions can be directly implemented through the index relationship list to realize retrieval based on data features, without performing retrieval matching in the data content, which can greatly save retrieval time and improve retrieval efficiency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of blockchain technology, and in particular to a blockchain data retrieval method for improving retrieval efficiency. Background Art

[0002] With the development of cloud storage and cloud computing, the scale and complexity of data storage have gradually increased, and the requirements for confidentiality and security of data storage have become increasingly higher. Blockchain technology can achieve objective data recording, data privacy protection, data tamper-proofing and other characteristics, and is a high-quality data security storage solution. In many application fields, blockchain storage is gradually replacing the traditional centralized data storage model, thereby improving data storage security. Among them, in application fields such as geographic information, smart cities, and manufacturing model design, while ensuring data security, the demand for supporting efficient multi-dimensional data chaining and multi-dimensional data retrieval is becoming increasingly clear.

[0003] Based on the traditional relational storage model, the multi-dimensional data storage design is simpler and more direct. Because blockchain uses a multi-party consensus strategy and complex cryptographic mechanisms, to achieve both security and performance, a more precise and detailed storage and retrieval management strategy is required. How to achieve efficient storage and retrieval of multi-dimensional data based on blockchain is one of the research focuses of blockchain for massive storage applications.

[0004] The existing blockchain mainly supports the KV storage mode. K is divided into ordinary index and composite key index. The composite key index has the ability to query in order. The ordinary index method requires retrieval one by one, which is inefficient. Based on the existing composite key index scheme, multidimensional data is designed as a composite key and data. Select certain features of the multidimensional data and splice them into a composite key K in a fixed order. The multidimensional data is encrypted as the V value and inserted into the blockchain account book in KV mode. However, when retrieving data, feature retrieval can only be performed in this fixed order. For example, if two-dimensional data has two-dimensional features A and B, then the composite key K will be assembled into A and B. During the retrieval operation, only the data object set that meets the A condition can be retrieved first, and the data object set that meets the B feature cannot be directly retrieved. Since repeated polling is required and all data must be traversed for feature comparison each time, the retrieval efficiency of the conventional composite key query mechanism is low. Summary of the invention

[0005] In response to the above problems and technical requirements, the applicant has proposed a blockchain data retrieval method that improves retrieval efficiency. The technical solution of this application is as follows:

[0006] A blockchain data retrieval method for improving retrieval efficiency, the method comprising:

[0007] Building a feature model based on historical search records, the feature model includes several data features used as search conditions in the historical search records;

[0008] An index relationship list is constructed according to the feature model and stored in the blockchain account book, wherein the index relationship list includes several data groups, and each data group has a corresponding feature index array, each data group includes several multidimensional business data, each feature index array includes several data features in the feature model, and the data content of each multidimensional business data in a data group includes all data features in the corresponding feature index array; several multidimensional business data in the same data group have feature association, and multidimensional business data in different data groups do not have feature association;

[0009] The index relationship list is used to provide a data retrieval function, and the retrieval result is obtained based on the data grouping corresponding to the feature index array where the data features contained in the retrieval request sent by the client are located.

[0010] Its further technical solution is that the method further comprises:

[0011] At predetermined time intervals, the feature model is updated according to the most recent N historical retrieval records, and the index relationship list is updated using the updated feature model.

[0012] A further technical solution is that the data features included in the feature model are data features determined based on historical retrieval records and are used to reflect retrieval habit trends.

[0013] Its further technical solution is that the blockchain account book stores index relationship lists corresponding to at least two different users, the index relationship list corresponding to each user is constructed by the user's feature model, and the feature model of each user is constructed by the user's historical search records, and the feature models of different users are used to reflect the search habit trends of different users; when the index relationship list in the blockchain account book is used to provide a data retrieval function, the index relationship list corresponding to the user who sends the retrieval request is used to respond to the retrieval request.

[0014] A further technical solution is to construct a feature model based on historical search records, including:

[0015] Summarize all data features that appear as search conditions in all historical search records, and determine the influence value of each data feature according to the historical search records where each data feature is located. The higher the influence value of a data feature, the higher the bias of the search habit trend represented by it.

[0016] The K data features with the largest influence values ​​are selected to construct a feature model. The K data features with the largest influence values ​​are the data features that best reflect the retrieval habit trend.

[0017] A further technical solution is to determine the impact value of each data feature according to the historical search records where each data feature is located, including:

[0018] Determine the impact value of each data feature according to the total number of times each data feature appears in the historical search records and the generation time of the historical search records containing the data feature;

[0019] Among them, the more total times a data feature appears in the historical search records and the closer the generation time of the historical search records containing the data feature is to the current time, the greater the influence value of the data feature.

[0020] A further technical solution is to determine the influence value of each data feature, including determining the influence value S of each data feature according to the following formula:

[0021]

[0022] Among them, A and A0 are constant coefficients. The time interval covered by the generation time of all historical retrieval records used to construct the feature model is [T B , T E ], T i is the generation time of the i-th historical search record containing the data feature, and set I is the set consisting of all historical search records containing the data feature.

[0023] A further technical solution is to use the index relationship list to provide data retrieval function, including:

[0024] Obtaining a search request input by a client, the search request including typical features and atypical features, wherein the typical features are data features included in the feature model, and the atypical features are data features not included in the feature model;

[0025] Obtain all data groups corresponding to the feature index array containing typical features as candidate groups;

[0026] The atypical features are used to search and compare the multi-dimensional business data contained in all candidate groups one by one, and finally the search results are screened out.

[0027] Its further technical solution is that the method further comprises:

[0028] The multi-dimensional business data contained in the data group that is relevant to the search result is actively pushed to the client that sent the search request.

[0029] A further technical solution is to construct an index relationship list according to the feature model, including:

[0030] The feature matching matrix is ​​constructed with the data features in the feature model as rows and all the multidimensional business data in the blockchain account book as columns. When the data content of the multidimensional business data corresponding to the jth column contains the data features of the ith row, the element of the ith row and jth column in the feature matching matrix is ​​1; when the data content of the multidimensional business data corresponding to the jth column does not contain the data features of the ith row, the element of the ith row and jth column in the feature matching matrix is ​​0;

[0031] A local sensitive hashing algorithm is used to calculate the similarity of each multidimensional business data based on a feature matching matrix, and a number of data groups are obtained according to the similarity of each multidimensional business data. Each data group corresponds to a similarity range and the similarity ranges of different data groups do not overlap with each other. Each data group contains a number of multidimensional business data whose similarities fall within the corresponding similarity range. It is determined that the data groups whose differences in similarity ranges are less than a difference threshold have usage correlation.

[0032] The beneficial technical effects of this application are:

[0033] The present application discloses a blockchain data retrieval method for improving retrieval efficiency. The method constructs a feature model that reflects user usage habits through a large number of historical retrieval records, and constructs an index relationship list based on the feature model. Multidimensional business data is associated with each group in the index relationship list. Therefore, by locating a group that meets the conditions through data feature retrieval, all multidimensional business data under the group can be associated. The retrieval conditions can be directly implemented based on data features through the index relationship list without performing retrieval matching in the data content. This can save more than 90% of the retrieval time and improve retrieval efficiency.

[0034] This application breaks through the limitations of traditional KV-type blockchain ledgers in terms of retrieval scope and retrieval efficiency, and can realize efficient multi-dimensional feature fuzzy query. For the traditional blockchain KV storage mode, to realize fuzzy query, you can only traverse all data and compare the data content one by one, and the query efficiency is very low. If multi-dimensional features are queried at the same time, repeated polling is required, and the efficiency is even lower. This application uses an index relationship list to convert retrieval queries based on data content into retrieval queries based on data features, which greatly reduces the scope of data queries; at the same time, it breaks through the query limitations of multi-dimensional features, so that multi-dimensional feature queries are the same as single-dimensional feature queries, and only need to perform one coverage calculation match.

[0035] Moreover, the feature model and index relationship list of the present application are not fixed, but are automatically updated periodically, and the present application sets a time-weighted parameter when calculating the impact value of data features during the model construction process, so as to fully absorb the impact of recent frequent feature retrieval operations on the overall model, realize adaptive tracking and matching of changes in user habits, and maintain the matching of feature models with user habits. The feature model is a general expression of user retrieval habits. Accurately constructing the feature model can build the optimal and efficient retrieval path in advance, so that more than 80% of the user's retrieval operations can be accurately matched and hit, shortening the search path. Self-optimization also reduces the system maintenance work of technical personnel and avoids the impact of human errors on the correctness of system operation.

[0036] Considering that the process of obtaining information is an orderly and associated process, if the relationship between account book data can be mastered, customer data needs can be accurately predicted, thereby converting passive retrieval into active push. This application accurately quantifies the characteristic correlation between multi-dimensional business data through local sensitivity analysis, and reflects it in the index relationship list. In the index relationship list, the usage correlation of adjacent groups is strong, and the farther the group distance, the weaker the usage correlation. For the retrieval requirements input by the client, the index relationship list can not only quickly locate the required retrieval results, but also quickly locate its adjacent data, thereby accurately predicting the customer's next data retrieval needs. That is, this application can also predict the data usage route and realize the fixed-point push of expected data, which not only saves the user's time to obtain data, but also ensures the integrity of the acquired valuable data. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a method flow chart of a blockchain data retrieval method in one embodiment of the present application.

[0038] Figure 2 It is a flowchart of constructing an index relationship list based on a feature model in an embodiment of the present application.

[0039] Figure 3 It is a flowchart of obtaining search results by using an index relationship list to provide a data search function in one embodiment of the present application.

[0040] Figure 4 It is a flowchart of returning search results to a client and actively pushing multi-dimensional business data with usage relevance in one embodiment of the present application.

[0041] Figure 5 It is a flowchart of updating a feature model and an index relationship list in one embodiment of the present application. DETAILED DESCRIPTION

[0042] The specific implementation of the present application is further described below in conjunction with the accompanying drawings.

[0043] This application discloses a blockchain data retrieval method that improves retrieval efficiency. The method includes the following steps, please refer to Figure 1 :

[0044] Step S1, constructing a feature model based on historical retrieval records.

[0045] The historical retrieval records used to build the feature model are generally the N historical retrieval records closest to the current moment. The value of N can be customized, and is generally larger, such as 20000. Each time the client executes the retrieval function, the smart contract will generate and save a historical retrieval record in the blockchain account book for this retrieval operation. The historical retrieval record includes the retrieval conditions, the generation time of the historical retrieval record, and the user ID of the user who performed the retrieval operation. The retrieval conditions include data features. For example, in one instance, a historical retrieval record is {"character":"hot end component","date":"2021.3.18","user":"wangfeng0064"}, then the data feature included in the retrieval condition is "hot end component", the generation time of the historical retrieval record is 2021.3.18, and the user ID of the user who performed the retrieval operation is wangfeng0064. The generation time in this application is measured in days as the time granularity, and it can actually be measured in hours, weeks, or other time granularities.

[0046] The feature model constructed based on several historical search records includes several data features, which are data features included in the historical search records as search conditions. In one embodiment, the feature model includes all data features that appear in all historical search records. However, considering that there may be occasional searches or atypical searches in reality, and these search experiences are not very helpful for improving the overall search efficiency, if all these data features are included in the data model, it will result in too many atypical data features and too high data redundancy. Therefore, in another embodiment, the feature model includes some data features that appear in the historical search records. Considering that the feature model is constructed in this application to better use the historical search experience, the data features included in the feature model are data features that are determined based on the historical search records to reflect the search habit trends, so that the feature model can well reflect the user's search habit trends.

[0047] In order to screen out data features that reflect retrieval habit trends to construct a feature model, an embodiment provides a method that is: summarize all data features that appear as retrieval conditions in all historical retrieval records, and determine the influence value of the data feature according to each historical retrieval record where each data feature is located. A data feature may exist in only one historical retrieval record or in multiple historical retrieval records, and the generation time of each historical retrieval record where the same data feature is located may be the same or different, and the user identifiers contained in each historical retrieval record where the same data feature is located may be the same or different. The higher the influence value of a data feature, the higher the bias of the retrieval habit trend represented by the data feature, that is, the data feature is more inclined to be used. After determining the influence value of each data feature, the K data features with the largest influence values ​​are selected to construct a feature model. The K data features with the largest influence values ​​are the data features that can best reflect the retrieval habit trend.

[0048] For example, for the data feature "hot end component", it appears in the 200001st historical search record, the 210000th historical search record, and the 280000th historical search record. These three historical search records are as follows:

[0049] {"character":"hot end component","date":"2021.3.18","user":"wangfeng0064"};

[0050] {"character":"hot end component","date":"2021.5.8","user":"zhouhai0238"};

[0051] {"character":"hot end component","date":"2021.8.19","user":"liming0576"};

[0052] The impact value of the data feature "hot end component" is determined based on the above three historical retrieval records.

[0053] In one embodiment, when determining the impact value of a data feature based on each historical search record in which the data feature is located, the impact value of each data feature is determined based on the total number of times each data feature appears in the historical search record and the generation time of the historical search record containing the data feature. The more times a data feature appears in the historical search record, the higher the frequency with which users use the data feature to search among all data features, and the greater the impact value of the data feature. The closer the generation time of the historical search record containing the data feature is to the current moment, the higher the frequency with which users use the data feature to search recently, and the greater the impact value of the data feature.

[0054] In one embodiment, the influence value S of each data feature can be determined according to the following formula:

[0055]

[0056] Among them, A and A0 are constant coefficients, which are generally pre-specified based on empirical values. The time interval covered by the generation time of all historical retrieval records used to construct the feature model is [T B , T E ], T B 、T E The time granularity of is the same as the time granularity of the generation time of each historical search record mentioned above. In an example, the time granularity is day. i is the generation time of the i-th historical search record containing the data feature, and set I is the set consisting of all historical search records containing the data feature.

[0057] Based on the above method, a feature model can be constructed according to the historical search records. The feature model includes several data features, and can also include the impact value corresponding to each data feature. For example, in an example, when the data features are arranged in order from large to small in terms of impact value, the form of some data features included in the feature model is as follows: [{"character":"hot end component","power":23.86},{"character":"East China Third Design Institute","power":23.36},{"character":"diesel power generation","power":23.13}]. In this example, the feature model includes three data features, namely "hot end component", "East China Third Design Institute" and "diesel power generation", the impact value of "hot end component" is 23.86, the impact value of "East China Third Design Institute" is 23.36, and the impact value of "diesel power generation" is 23.13.

[0058] Step S2: construct an index relationship list based on the feature model and store it in the blockchain account book.

[0059] In one embodiment, the method for constructing an index relationship list includes the following steps, please refer to Figure 2 :

[0060] 1. The feature matching matrix is ​​constructed with the data features in the feature model as rows and all the multidimensional business data in the blockchain account book as columns. When the data content of the multidimensional business data corresponding to the jth column contains the data features of the ith row, the element of the ith row and jth column in the feature matching matrix is ​​1; when the data content of the multidimensional business data corresponding to the jth column does not contain the data features of the ith row, the element of the ith row and jth column in the feature matching matrix is ​​0.

[0061] 2. The local sensitive hashing algorithm is used to calculate the similarity of each multi-dimensional business data based on the feature matching matrix.

[0062] 3. According to the similarity of each multidimensional business data, several data groups are obtained. Each data group corresponds to a similarity range and the similarity ranges of different data groups do not overlap. Each data group contains several multidimensional business data whose similarity falls within the corresponding similarity range, thereby obtaining several data groups and grouping the multidimensional business data according to the similarity. In practical applications, the similarity of the multidimensional business data contained in each data group can be equal, or unequal but with a small difference and belong to the same similarity range.

[0063] Thus, the index relationship list constructed includes several data groups, each of which includes several multi-dimensional business data, which actually generally includes the data addresses of several multi-dimensional business data. Moreover, the similarities of several multi-dimensional business data in the same data group are close and have characteristic relevance, while the similarities of multi-dimensional business data in different data groups are greatly different and do not have characteristic relevance.

[0064] Each data group has a corresponding feature index array, each feature index array contains several data features in the feature model, and the data content of each multi-dimensional business data in a data group contains all the data features in the corresponding feature index array. The same data feature in the feature model can be included in the feature index arrays corresponding to one or more different data groups.

[0065] The index relationship list also records the unique grouping identifier of each data group. Optionally, the index relationship list also records the similarity range of each data group. For example, in one example, the content related to each data group recorded in the index relationship list is:

[0066] {"groupId":"DF93FDSA0","reference":["hot end components","East China Third Design Institute","diesel power generation","Zhang Qiang",……],"similarity":1363,"values":["DFW982FJS09WFJ09F","VR498NVOWIU94U090D",……]}.

[0067] In the above example, the addresses of the multi-dimensional business data included in the data group with the group identifier "DF93FDSA0" are "DFW982FJS09WFJ09F", "VR498NVOWIU94U090D", ..., and the addresses of all the multi-dimensional business data are not fully recorded in this example. The similarity range corresponding to the data group is a single point value of 1363, that is, the similarity of all the multi-dimensional business data in the data group is 1363. The data features included in the feature index array reference of the data group are "hot end components", "East China Third Design Institute", "diesel power generation", "Zhang Qiang", ..., and this example does not show all the data features in detail. The data content of each multi-dimensional business data includes all the data features in the corresponding feature index array, that is, for example, the data content of the multi-dimensional business data with the address "DFW982FJS09WFJ09F" includes all the data features in the feature index array reference. Similarly, the data content of the multi-dimensional business data with the address "VR498NVOWIU94U090D" also includes all the data features in the feature index array reference. For example, the data feature "hot end component" can be included in the feature index array corresponding to the data group with the group identifier "DF93FDSA0" and can also be included in the feature index array corresponding to other data groups.

[0068] Step S3, using the index relationship list to provide a data retrieval function, based on the data grouping corresponding to the feature index array where the data feature contained in the retrieval request sent by the client is located, obtain the retrieval result, please refer to Figure 3 Flowchart shown.

[0069] The search request sent by the user through the client includes one or more data features, and the search request generally also includes the user identifier that sent the search request. After receiving the search request, the blockchain splits it into various data features. The data features included in the search request are more likely to be included in the feature model, so the search request at least includes typical features, and the typical features are data features included in the feature model. In one embodiment, the search request only includes typical features. In this embodiment, the multi-dimensional business data in all data groups corresponding to the feature index array containing the typical feature in the index relationship list is directly obtained as the search result. Since the same data feature can be included in one or more feature index arrays, in this embodiment, one or more data groups can be obtained accordingly.

[0070] In another embodiment, the search request includes not only typical features but also atypical features, which are data features not included in the feature model. First, all data groups corresponding to the feature index array containing typical features in the index relationship list are obtained as candidate groups, and then the multi-dimensional business data contained in all candidate groups are searched and compared one by one using atypical features, and finally the search results are screened out. Similarly, in this embodiment, one or more candidate groups can be obtained first.

[0071] Regardless of which of the above embodiments is used, the result generated after the smart contract executes the endorsement is returned to the client by the blockchain SDK, and the returned content includes the above-mentioned search result.

[0072] As described above, several multidimensional business data in the same data group have characteristic correlations. Although the multidimensional business data in different data groups do not have characteristic correlations, the multidimensional business data in different data groups will have usage correlations. Usage correlation represents the application value of data series between different data groups. For example, when a user uses data A, it means that data A is valuable to the user, then data B that has usage correlation with data A can also generate value for the user. The closer the similarity range of the data groups, the stronger the usage correlation between the data groups. That is, the usage correlation between data groups with close similarity ranges and small differences is often strong, while the usage correlation between data groups with far similarity ranges and large differences is weaker.

[0073] Therefore, in one embodiment, please refer to Figure 4 In addition to generating and returning the above search structure, the multi-dimensional business data contained in the data groupings that are relevant to the search results will be actively pushed to the client that sent the search request, and the data groups whose differences in the similarity range are less than the difference threshold have usage relevance. That is, the content returned to the client by the blockchain SDK includes not only the search results, but also the multi-dimensional business data contained in the data groupings that are relevant to the search results. In this way, when the user uses data A, not only the data A required by the user can be returned to the user, but also the data B that is relevant to the A data and can generate value for the user can be actively pushed to the user. The active push function is provided, which not only saves the time of users to obtain data, but also ensures the integrity of the obtained valuable data.

[0074] In actual application, after the index relationship list is constructed and stored in the blockchain ledger, the index relationship list in the blockchain ledger is used directly in subsequent use. However, this index relationship list does not always remain unchanged. Similarly, the feature model does not remain unchanged. At predetermined time intervals, the feature model is updated according to the above method based on the most recent N historical retrieval records, so that the feature model can achieve adaptive tracking and matching of changes in user habits. After the feature model is updated, the blockchain smart contract is triggered to use the updated feature model to update the index relationship list in the blockchain ledger, and then use the updated index relationship list to provide retrieval function. In other words, Figure 5 As shown in the figure, if the time interval from the last update of the feature model does not reach the predetermined time interval, the index relationship list stored in the current blockchain account book will continue to be used to respond to the retrieval request. If the time interval from the last update of the feature model reaches the predetermined time interval, the feature model will be rebuilt according to the historical retrieval records, and the index relationship list in the blockchain account book will be updated accordingly, and the updated index relationship list will be used to respond to the retrieval request.

[0075] In one embodiment, the historical search records generated by all users are summarized according to the above method to generate the same feature model. The feature model obtained at this time is used to reflect the search habit trends of all users. In this embodiment, a unique index relationship list is constructed accordingly and stored in the blockchain account book. When the blockchain provides the data retrieval function, when receiving a retrieval request sent by any user through the client, the unique index relationship list is used to respond to the retrieval request and provide the data retrieval function according to the above method. Similarly, when updating the feature model, the most recent N historical search records generated by all users are used to update the feature model according to the above method and then update the index relationship list.

[0076] However, considering that different users may have different search habits and trends, in another embodiment, the historical search records of different users are summarized and the feature models of each user are generated according to the above method, and the feature models of different users are used to reflect the search habits and trends of different users. In this embodiment, the index relationship list corresponding to each user is constructed according to the feature models of each user, so that at least two index relationship lists corresponding to different users are stored in the blockchain account book. When the blockchain provides the data retrieval function, when receiving the retrieval request sent by the user through the client, the index relationship list corresponding to the user who sent the retrieval request is selected from the multiple index relationship lists stored in the blockchain account book to respond to the retrieval request and provide the data retrieval function, thereby using different index relationship lists to respond to the retrieval requests of different users. In this embodiment, the feature models of each user are updated according to the above method according to the most recent N historical search records generated by each user, and then the corresponding index relationship lists are updated. In addition, in this embodiment, the update frequency of the feature models of all users is the same, or the update frequency of the feature models of two users is different.

[0077] The above is only a preferred embodiment of the present application, and the present application is not limited to the above embodiments. It is understood that other improvements and changes directly derived or associated by those skilled in the art without departing from the spirit and concept of the present application should be considered to be included in the protection scope of the present application.

Claims

1. A blockchain data retrieval method for improving retrieval efficiency, characterized in that: The method comprises: Building a feature model based on the historical search records, wherein the feature model includes a plurality of data features used as search conditions in the historical search records; An index relationship list is constructed according to the feature model and stored in the blockchain account book, wherein the index relationship list includes a plurality of data groups, and each data group has a corresponding feature index array, each data group includes a plurality of multidimensional business data, each feature index array includes a plurality of data features in the feature model, and the data content of each multidimensional business data in a data group includes all data features in the corresponding feature index array; the plurality of multidimensional business data in the same data group have feature association, and the multidimensional business data in different data groups do not have feature association; The index relationship list is used to provide a data retrieval function, and a retrieval result is obtained based on the data grouping corresponding to the feature index array where the data feature contained in the retrieval request sent by the client is located; The constructing of the index relationship list according to the feature model includes: constructing a feature matching matrix with data features in the feature model as rows and all multidimensional business data in the blockchain account book as columns, when the data content of the multidimensional business data corresponding to the jth column includes the data feature of the ith row, the element of the ith row and jth column in the feature matching matrix is ​​1; when the data content of the multidimensional business data corresponding to the jth column does not include the data feature of the ith row, the element of the ith row and jth column in the feature matching matrix is ​​0; using a local sensitive hashing algorithm to calculate the similarity of each multidimensional business data based on the feature matching matrix, and obtaining a plurality of data groups according to the similarity of each multidimensional business data, each data group corresponds to a similarity range and the similarity ranges of different data groups do not overlap, and each data group contains a plurality of multidimensional business data whose similarity falls within the corresponding similarity range; determining that the data groups whose difference in similarity range is less than a difference threshold have usage correlation.

2. The method according to claim 1, characterized in that The method further comprises: At predetermined time intervals, the feature model is updated according to the most recent N historical retrieval records, and the index relationship list is updated using the updated feature model.

3. The method according to claim 1, characterized in that The data features included in the feature model are data features determined based on historical search records and used to reflect search habit trends.

4. The method according to claim 3, characterized in that The blockchain account book stores index relationship lists corresponding to at least two different users, the index relationship list corresponding to each user is constructed by a feature model of the user, the feature model of each user is constructed by the historical search record of the user, and the feature models of different users are used to reflect the search habit trends of different users; when the index relationship list in the blockchain account book is used to provide a data retrieval function, the index relationship list corresponding to the user who sends the retrieval request is used to respond to the retrieval request.

5. The method according to claim 3, characterized in that: The step of constructing a feature model based on historical search records includes: Summarize all data features that appear as search conditions in all historical search records, and determine the influence value of each data feature according to the historical search records where the data feature is located. The higher the influence value of the data feature, the higher the bias of the search habit trend represented by it; The feature model is constructed by selecting K data features with the largest influence values, and the K data features with the largest influence values ​​are the data features that best reflect the retrieval habit trend.

6. The method according to claim 5, characterized in that The determining the impact value of each data feature according to the historical search record where the data feature is located includes: Determine the impact value of each data feature according to the total number of times each data feature appears in the historical search records and the generation time of the historical search records containing the data feature; The more times a data feature appears in the historical search records, and the closer the generation time of the historical search records containing the data feature is to the current time, the greater the influence value of the data feature.

7. The method according to claim 6, characterized in that Determining the influence value of each data feature includes determining the influence value of each data feature according to the following formula for: ; in, are all constant coefficients. The time interval covered by the generation time of all historical retrieval records used to construct the feature model is , is the first The time when the historical search records were generated, collection It is a set consisting of all historical search records containing the data feature.

8. The method according to claim 1, characterized in that The providing of data retrieval function by using the index relationship list includes: Obtaining a search request input by a client, wherein the search request includes typical features and atypical features, wherein the typical features are data features included in the feature model, and the atypical features are data features not included in the feature model; Obtain all data groups corresponding to the feature index array containing typical features as candidate groups; The atypical features are used to search and compare the multi-dimensional business data contained in all candidate groups one by one, and finally the search results are screened out.

9. The method according to claim 1 or 8, characterized in that: The method further comprises: The multi-dimensional business data contained in the data group having usage relevance to the search result is actively pushed to the client that sent the search request.