Data retrieval method and device

By creating multi-attribute data storage tensors and concept hierarchies, data is converted into tensors of different granularity levels, the heterogeneity and privacy issues in data retrieval are solved, and efficient data retrieval and privacy protection are achieved.

CN119066101APending Publication Date: 2024-12-03BEIJING JINGDONG QIANSHITECHNOLOGY CO LTD

Patent Information

Application Number
CN202310638435.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-31
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The prior art faces heterogeneity and privacy challenges in data retrieval. The storage and analysis of heterogeneous data are inefficient, the privacy protection process is cumbersome and the computing volume is large.

Method used

By creating multi-attribute data storage tensors and establishing a concept hierarchy for each attribute, converting the data into tensors of different granularity levels, obtaining the target granularity tensors according to the query statement for calculating the query results, using aggregation, decomposition, etc. The function gets the query result.

Benefits of technology

It realizes the protection of user privacy while occupying less computing resources, ensures data security, and effectively solves the problem of data heterogeneity and improves data retrieval efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066101A_ABST
    Figure CN119066101A_ABST
Patent Text Reader

Abstract

The invention discloses a data retrieval method and device, and relates to the technical field of computers. A specific embodiment of the method comprises the steps of creating a data storage tensor containing a plurality of attributes based on to-be-stored data; creating a concept hierarchical structure containing a plurality of concept levels for each attribute, and converting the data storage tensor into granularity level tensors in one-to-one correspondence with the concept levels according to the concept hierarchical structure of each attribute; in response to a received query statement, obtaining a query attribute and / or a query attribute value from the query statement, and determining a granularity level tensor corresponding to a concept level of the query attribute and / or the query attribute value in the corresponding concept hierarchical structure as a target granularity tensor; creating a result tensor corresponding to the query statement based on the target granularity tensor; and executing operation based on a preset function on the result tensor to obtain a query result corresponding to the query statement. According to the embodiment, the privacy problem in data retrieval can be solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a data retrieval method and device. Background Art

[0002] Data retrieval can provide basic information for subsequent data analysis and mining, and is an effective way to analyze data. At present, data retrieval faces two major challenges: heterogeneity and privacy. The heterogeneity challenge refers to the fact that heterogeneous data such as text data, audio data, and image data must be operated in a manageable form, and the privacy challenge refers to the fact that business needs in different fields and with different permissions have special privacy protection requirements.

[0003] In the prior art, each type of heterogeneous data is generally stored separately in its original format, for example, text data is stored as a text file, audio data is stored as an audio file, and video data is stored as a video file. Data analysis based on this original format is extremely inefficient. Regarding privacy protection, the prior art generally locates the query results first, and then uses user permissions and privacy protection rules to mask the query results. The data processing process is relatively cumbersome and requires a large amount of computation. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides a data retrieval method and device, which can solve the privacy query problem through a data storage tensor and a granularity level tensor corresponding to different concept levels of each attribute.

[0005] To achieve the above objective, according to one aspect of the present invention, a data retrieval method is provided.

[0006] The data retrieval method of an embodiment of the present invention includes: creating a data storage tensor containing multiple attributes based on the data to be stored; creating a concept hierarchy containing multiple concept levels for each attribute, and converting the data storage tensor into a granularity level tensor corresponding to the concept level one by one according to the concept hierarchy of each attribute; in response to receiving a query statement, obtaining query attributes and / or query attribute values ​​from the query statement, and determining the granularity level tensor corresponding to the concept level of the query attributes and / or query attribute values ​​in the corresponding concept hierarchy as a target granularity tensor; creating a result tensor corresponding to the query statement based on the target granularity tensor; and performing operations based on preset functions on the result tensor to obtain a query result corresponding to the query statement.

[0007] Optionally, the data to be stored is a variety of heterogeneous data; and the creating a data storage tensor containing multiple attributes based on the data to be stored includes: extracting key information from the variety of heterogeneous data, and creating the data storage tensor according to the extracted key information.

[0008] Optionally, the data storage tensor is a third-order tensor based on three basic features: time, indicator, and user; when a new user's data is added to the data storage tensor, the dimension of the user's basic features increases, and the time basic features and the indicator basic features remain unchanged; when an existing user's data is added to the data storage tensor, the dimension of the time basic features increases, and the indicator basic features and the user basic features remain unchanged.

[0009] Optionally, the method further comprises: after acquiring the query attributes and / or query attribute values ​​from the query statement, creating a result list according to the acquired query attributes and / or query attribute values.

[0010] Optionally, creating a result tensor corresponding to the query statement based on the target granularity tensor includes: when the result list contains the query attribute but does not contain the query attribute value, selecting the front slicing matrix of the target granularity tensor corresponding to the query attribute contained in the result list to form the result tensor; wherein the front slicing matrix is ​​generated by slicing the target granularity tensor along the indicator basic feature; when the result list contains the query attribute value but does not contain the query attribute: if all the query attribute values ​​contained in the result list are in different front slicing matrices of the target granularity tensor, then the different front slicing matrices are combined into the result tensor; if some or all of the query attribute values ​​contained in the result list are in the same front slicing matrix of the target granularity tensor, then the column fibers where the query attribute value is located and the front slicing matrix composed of the column fibers are combined into the result tensor. The result tensor is composed of a matrix; wherein the column fibers are generated by slicing the front slice matrix along the time-based features; when the result list contains both query attributes and query attribute values: the front slice matrix of the target granularity tensor corresponding to the query attributes contained in the result list is selected to form a first intermediate tensor; if all the query attribute values ​​contained in the result list are in different front slice matrices of the target granularity tensor, the different front slice matrices are composed of a second intermediate tensor, and the first intermediate tensor and the second intermediate tensor are combined to form the result tensor; if some or all of the query attribute values ​​contained in the result list are in the same front slice matrix of the target granularity tensor, the column fibers where the query attribute values ​​are located and the front slice matrix composed of the column fibers are composed of a third intermediate tensor, and the first intermediate tensor and the third intermediate tensor are combined to form the result tensor.

[0011] Optionally, the heterogeneous data includes at least one of the following: text data, report record data, image data, audio data, video data; the concept hierarchy is a tree structure; the function includes at least one of the following: aggregation function, decomposition function, clustering function.

[0012] To achieve the above objective, according to another aspect of the present invention, a data retrieval device is provided.

[0013] The data retrieval device of an embodiment of the present invention may include: a storage tensor creation unit, which is used to create a data storage tensor containing multiple attributes based on the data to be stored; a granularity tensor creation unit, which is used to create a concept hierarchy containing multiple concept levels for each attribute, and convert the data storage tensor into a granularity level tensor corresponding to the concept level one by one according to the concept hierarchy of each attribute; a query preparation unit, which is used to: in response to receiving a query statement, obtain query attributes and / or query attribute values ​​from the query statement, determine the granularity level tensor corresponding to the concept level of the query attribute and / or query attribute value in the corresponding concept hierarchy as a target granularity tensor; create a result tensor corresponding to the query statement based on the target granularity tensor; and a query execution unit, which is used to perform operations based on preset functions on the result tensor to obtain query results corresponding to the query statement.

[0014] Optionally, the data to be stored is a variety of heterogeneous data, and the heterogeneous data includes at least one of the following: text data, report record data, image data, audio data, and video data; the data storage tensor is a third-order tensor based on three basic features of time, indicators, and users, and the function includes at least one of the following: aggregation function, decomposition function, and clustering function.

[0015] To achieve the above objective, according to another aspect of the present invention, an electronic device is provided.

[0016] An electronic device of the present invention comprises: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the data retrieval method provided by the present invention.

[0017] To achieve the above objective, according to yet another aspect of the present invention, a computer-readable storage medium is provided.

[0018] A computer-readable storage medium of the present invention stores a computer program, which implements the data retrieval method provided by the present invention when executed by a processor.

[0019] According to the technical solution of the present invention, the embodiments of the above invention have the following advantages or beneficial effects:

[0020] In the data preparation stage, firstly, a data storage tensor containing multiple attributes is created based on the various heterogeneous data to be stored, then a concept hierarchy containing multiple concept levels is created for each attribute, and the data storage tensor is converted into a granularity level tensor corresponding to each concept level according to the concept hierarchy of each attribute. In the data retrieval stage, after receiving the query statement, the query attribute and / or query attribute value is first obtained from the query statement, the granularity level tensor corresponding to the concept level of the query attribute and / or query attribute value is determined as the target granularity tensor, and the result tensor corresponding to the query statement is created based on the target granularity tensor, and finally the query result can be obtained by performing decomposition, aggregation and other function operations on the result tensor. In this way, data retrieval is performed based on the granularity level tensor corresponding to different attribute granularities, and the query result of the corresponding granularity can be returned according to the requirements of the query statement, thereby protecting user privacy and ensuring data security while occupying less computing resources. In addition, the third-order data storage tensor is used to effectively store various heterogeneous data including text, images, and videos, solving the problem of data heterogeneity in data retrieval.

[0021] The further effects of the above-mentioned non-conventional optional manner will be described below in conjunction with the specific implementation manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention.

[0023] Figure 1 Schematic diagram of the main steps of the data retrieval method in an embodiment of the present invention;

[0024] Figure 2 is a schematic diagram of a data retrieval algorithm flow in an embodiment of the present invention;

[0025] Figure 3 is a schematic diagram of a conceptual hierarchical structure of an embodiment of the present invention;

[0026] Figure 4 is a schematic diagram of mapping granularity level tensors and concept levels in an embodiment of the present invention;

[0027] Figure 5 It is a schematic diagram of components of a data retrieval device in an embodiment of the present invention;

[0028] Figure 6 is an exemplary system architecture diagram in which embodiments of the present invention may be applied;

[0029] Figure 7 It is a schematic diagram of the structure of an electronic device used to implement the data retrieval method in the embodiment of the present invention. DETAILED DESCRIPTION

[0030] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0031] It should be pointed out that the embodiments of the present invention and the technical features therein may be combined with each other without conflict.

[0032] Figure 1 Schematic diagram of main steps of the data retrieval method according to an embodiment of the present invention.

[0033] like Figure 1 As shown, the data retrieval method of the embodiment of the present invention can be specifically performed according to the following steps:

[0034] Step S101: Create a data storage tensor containing multiple attributes based on the data to be stored.

[0035] In an embodiment of the present invention, the above data to be stored may be a variety of heterogeneous data, such as text data, report record data, image data, audio data, video data, etc. In this step, the server may extract key information from a variety of heterogeneous data, and create a data storage tensor based on the extracted key information. Preferably, the above data storage tensor may be a third-order tensor based on three basic features: time, index, and user. By slicing the data storage tensor along the index basic feature, multiple positive slice matrices may be obtained; by slicing each positive slice matrix along the time basic feature, multiple column fibers may be generated; by slicing each column fiber along the user basic feature, multiple attributes may be generated.

[0036] In actual scenarios, the above data storage tensors can be used to store a variety of heterogeneous data, thereby solving the data heterogeneity problem in the current data retrieval process. Generally, when adding a new user's data to the data storage tensor, the dimension of the user's basic features increases, and the time-based features and indicator-based features remain unchanged; when adding an existing user's data to the data storage tensor, the dimension of the time-based features increases, and the indicator-based features and user-based features remain unchanged. Through the above steps, a variety of heterogeneous data such as text, report records, images, audio, and video can be uniformly integrated and stored in the data storage tensor for subsequent data retrieval and data analysis, replacing the current heterogeneous data processing method based on separate storage of the original format, and improving the efficiency of subsequent data analysis.

[0037] Step S102: Create a concept hierarchy structure containing multiple concept levels for each attribute, and convert the data storage tensor into a granularity level tensor corresponding to the concept level one by one according to the concept hierarchy structure of each attribute.

[0038] The concept hierarchy in this step can be a tree structure, see Figure 3 .for Figure 3 The tree of the age attribute on the left and the tree of the address attribute on the right are both coarse-grained and fine-grained from the root node downwards, and each level is a concept level (also called a granularity level). In this step, the server can combine the concept levels of each attribute in the corresponding concept hierarchy and convert the data storage tensor created in step S101 into a granularity level tensor corresponding to the concept level. For specific mapping relationships, see Figure 4 In this way, the creation of specific granularity tensors (i.e., granularity level tensors) for each attribute and each concept level is achieved, which is helpful for the implementation of subsequent data retrieval solutions based on granular query methods that meet privacy protection requirements.

[0039] Step S103: In response to receiving a query statement, query attributes and / or query attribute values ​​are obtained from the query statement, and a granularity level tensor corresponding to the concept level of the query attributes and / or query attribute values ​​in the corresponding concept hierarchy is determined as a target granularity tensor; and a result tensor corresponding to the query statement is created based on the target granularity tensor.

[0040] In this step, the server can determine the target granularity tensor required for this query from the pre-created multiple granularity level tensors by using the query attributes and / or query attribute values ​​determined from the query statement, that is, the granularity level tensor corresponding to the concept level of the query attribute and / or query attribute value in the corresponding concept hierarchy is determined as the target granularity tensor. Thereafter, the server can determine the result tensor corresponding to the query statement based on the target granularity tensor and the query statement information, and the result tensor contains all the attributes and their attribute values ​​involved in the query statement. After determining the result tensor corresponding to the query statement, the required query result can be obtained by performing function operations in the result tensor.

[0041] In practical applications, after obtaining query attributes and / or query attribute values ​​from the query statement, the server can create a result list based on the obtained query attributes and / or query attribute values. Thereafter, the server performs corresponding steps in three cases to obtain a result tensor.

[0042] The first is that when the result list contains the query attribute but does not contain the query attribute value, the positive slice matrix of the target granularity tensor corresponding to the query attribute contained in the result list is selected to form the result tensor. It can be understood that the above positive slice matrix is ​​generated by slicing the target granularity tensor along the indicator basic feature.

[0043] The second type is that when the result list contains the query attribute value and does not contain the query attribute, if all the query attribute values ​​contained in the result list are in different frontal slice matrices of the target granularity tensor, the server will form the result tensor with different frontal slice matrices; if some or all of the query attribute values ​​contained in the result list are in the same frontal slice matrix of the target granularity tensor, the server will form the result tensor with the column fiber where the query attribute value is located and the frontal slice matrix composed of the column fiber. It can be understood that the above column fiber is generated by slicing the frontal slice matrix along the time-based feature.

[0044] The third type, when the result list contains both query attributes and query attribute values: the server selects the front slice matrix of the target granularity tensor corresponding to the query attributes contained in the result list to form the first intermediate tensor. If all the query attribute values ​​contained in the result list are in different front slice matrices of the target granularity tensor, the server combines the different front slice matrices into a second intermediate tensor, and combines the first intermediate tensor and the second intermediate tensor to form a result tensor. If all or part of the query attribute values ​​contained in the result list are in the same front slice matrix of the target granularity tensor, the server combines the column fibers where the query attribute values ​​are located and the front slice matrix composed of the column fibers into a third intermediate tensor, and combines the first intermediate tensor and the third intermediate tensor to form a result tensor.

[0045] Through the above steps, a matching target granularity tensor is determined for the query statement and an accurate result tensor is formed for different query situations based on the target granularity tensor, which helps to achieve fast response, high accuracy and high recall rate for subsequent query operations.

[0046] Step S104: performing an operation based on a preset function on the result tensor to obtain a query result corresponding to the query statement.

[0047] In this step, the server can select the corresponding function to perform calculation according to the query statement to obtain the query result. The above functions can be: aggregation function, decomposition function, clustering function, etc. So far, the embodiment of the present invention has achieved the purpose of protecting data privacy by using granular computing through the granularity level processing of the data to be stored. The data to be stored forms various granularity levels based on the conceptual hierarchy. When executing a query, the user can only see part of the data at the corresponding granularity level that meets the privacy protection requirements, thereby successfully coping with the privacy challenge in the data retrieval process.

[0048] A specific embodiment of the present invention is described below.

[0049] Using data retrieval to provide information for subsequent analysis and mining is an effective way to analyze data, which requires facing three major challenges: heterogeneity, privacy, and abstraction. Some heterogeneous data, such as sensor data, scanned images, and data reports must be operated in a manageable form. In addition, some data contains important information that needs to be protected to avoid privacy violations. At the same time, any business requires simple and effective information that is only related to their respective business scope. Therefore, this embodiment adopts the concept of granular computing, analyzes and manages heterogeneous data with a tensor model, and maps the original tensor into granular tensors of different abstract levels according to the conceptual hierarchy of the attributes, thereby establishing the basis for query processing.

[0050] Granularity is the degree of scale or level of detail in a set of data units. It provides an abstraction of the underlying details of things in order to highlight the focus required for analysis and can be applied to extract important and necessary attributes. Granular computing is a technique for extracting information from different and finite abstract objects, so that each computational problem can be solved at different levels of granularity by removing irrelevant details from the available data set. It uses the concepts of many existing theories, such as fuzzy sets, set theory, rough sets, divide-and-conquer algorithms, etc., to solve the computational problems encountered. Cloud computing and quantum computing provide a powerful platform for big data processing, while granular computing captures the essence of big data processing by using granularity such as clusters, intervals, and subsets to control the size of big data.

[0051] The technical problem of this embodiment can be described as follows: assuming that there is a batch of data, such as relevant illustrations, report records, and sensor data, these data need to be analyzed and mined to assist decision-making and understand user needs. However, the existing data has heterogeneous characteristics. At the same time, business needs in different fields have challenges of abstractness and privacy. For example, organizations pay more attention to macro information, while individual users pay more attention to personal information.

[0052] The solution of this embodiment is to propose a data retrieval algorithm based on the combination of tensors and granular computing to solve the problems of heterogeneity, privacy and abstraction faced by data analysis and mining. This method uses tensors to manage heterogeneous data information, uses the concept of granular computing to abstract data information into different levels of granularity, and uses the concept of coarse-grainedness to maintain user privacy issues.

[0053] The data retrieval algorithm based on tensor and granular computing (DrTGc) proposed in this embodiment is divided into two parts: Figure 2As shown in the figure. In the first part, keyword information including text, data reports, and scanned images is efficiently stored in a third-order tensor called DataStorage Tensor DsTr. The data is then processed in DsTr using the conceptual hierarchy of granular computing to create various processed tensors that span different levels of granularity. The second part uses the generated granular tensor for efficient query processing and result formation, using keyword matching to identify each attribute required for the final query result and the granularity level of the attribute.

[0054] The following first explains the concept of data storage tensor. Big data comes from various sources and in different formats, such as text data, report records, scanned images, audio data, and video data. The main challenge at this stage is how to manipulate such heterogeneous data in a manageable form so that it can be processed efficiently later. To achieve this goal, a data storage tensor (Data storage tensor DsTr) is proposed to manage this data.

[0055] In the whole retrieval process, all information is stored in raw format, i.e. text as text files, sound files as audio and video files as video. These files are stored in their respective databases. However, this method of analyzing data in raw format is extremely inefficient in data analysis. In order to make tensors effective for the information granularization step, this paper uses known keyword extraction methods to store keyword information of various types of data into tensor slices.

[0056] The data storage tensor includes three basic features: time, index, and user. A data storage tensor can be represented as a third-order tensor. I P where I represents the user, T Indicates time, I M Indicates the index, which is the specific attribute value of each attribute. DsTr can be expressed as follows:

[0057] DsTr=[H1,H2,H3,H4,……,H m ]

[0058] Among them, H i It is a slice matrix generated by slicing the data storage tensor along the indicator field. Each indicator can include various attributes, so the data storage tensor can be further sliced ​​into attributes. ij ) and attributes (a ijk ) can be further written as shown below.

[0059]

[0060] When a new user registers in the system, its related data is added to the DsTr, this operation is called data addition. On the other hand, adding new data entries about existing users in the DsTr is called data append. These two operations help in automatically updating the DsTr over time.

[0061] When adding data, given a new user d, add it to the data storage tensor In this case, the time and index order of the data storage tensor remain unchanged, and the user dimension increases. P =I P +1,I T =I T , I M =I M .

[0062] When data is appended, for user j who already has data storage tensor, add data d. If I T =t j , add one to the time order of user j. P =I P , I M =I M ,

[0063] The following explains the granularization of data information. Once various types of data are effectively stored in the data storage tensor, they must be abstracted into particles of different sizes as required to form different granularity levels. A framework mapping is needed to process tensors and the concept of particles in granular computing to create tensors at different abstraction levels. To achieve this goal, a corresponding concept hierarchy can be created for each attribute in the tensor, and then each attribute can be mapped one by one to the concept hierarchy level to create tensors of different granularity. Therefore, each attribute is divided into multiple levels of concept hierarchy, such as Figure 3 As shown in the concept hierarchy tree in , a concept hierarchy tree is constructed for two attributes, age and address. Similarly, a concept hierarchy is created for all attributes that appear in the dataset. Each concept hierarchy has multiple levels, and these hierarchies are different for all attributes depending on the required classification level. For example Figure 3 The attributes age and address appear. The attribute age has three hierarchical levels, while the attribute address has five hierarchical levels.

[0064] Regarding the concept hierarchy, it is defined as, given an attribute value "a" of an arbitrary attribute and its sequence of attribute values ​​from specific to general, where the most specific attribute value consists of the most precise "a", and the most general attribute value consists of other values ​​composed of a, such as 'ANY', then the concept hierarchy of "a" is defined by the partial order relationship of "a" from the most specific to the most general.

[0065] The concept hierarchy of each attribute is created by forming a specific-to-general sequence of attribute values. For example, for an attribute value of 2 for the age attribute, a specific-to-general sorting sequence of "2, infant, child, ANY" is given, with the most specific value being 2 and the most general value being ANY. When such a sorting sequence is created for each value in the attribute domain, the conceptual level of the value can be determined through a one-to-one mapping with the concept hierarchy tree.

[0066] The concept level is defined as follows: the concept level of the attribute value "a" is determined by the sorting sequence of "a". Assume that l a Indicates the concept level of "a", when the most specific attribute value of "a" exists a =0, l a As the sorting sequence increases from specific to general, the number of sequences increases from specific to general, l a Plus one.

[0067] There are two conceptual levels a With l' a , these two concept levels contain the same attribute value "a", l a =i at the same time l' a =j, if and only if i>j and l a ≥l' a Established at that time.

[0068] The definition of a concept hierarchy tree is that the concept hierarchy tree of any attribute value "a" is composed of a finite set of nodes: (1) The tree contains a root node whose concept level is l r >l c , c is any node other than the root node. (2) The remaining nodes of the tree form an ordered set consisting of zero or more disjoint trees. The concept level of each node "d" is l d = l p -1, where l p is the concept level of the parent node of node "d". (3) Leaf nodes are at concept level 0.

[0069] The conceptual hierarchy formed in this way is used to create tensors at different granularity levels. Each synthesized tensor is called a granular level tensor TanGLei (Tensor at Granular Level i).

[0070] The definition of the granularity level tensor is, TanGLe0=DsTR, TanGLe i By TanGLe i-1 produce, Where "i" is The maximum value of x is the sum of the number of attributes. The tensor TanGLe at each granularity level i Formed through a one-to-one mapping with the concept hierarchy, e.g. Figure 4 As shown in Figure 1, a one-to-one mapping of the dimensions and conceptual levels of low-level tensors will produce high-level tensors with more refined information.

[0071] Therefore, a hierarchical tensor is formed from the finest tensor to the coarsest tensor, where the finest tensor is at the lowest conceptual level and the coarsest tensor is at the highest conceptual level.

[0072] Mathematically, the tensor TanGLe at the i-level of granularity i It can be written in frontal slices using the following formula and further can be expressed in column fibers and properties by using equations.

[0073]

[0074] It can be seen that combining tensor information representation with the concept of conceptual hierarchy can effectively store large amounts of information. Depending on the type of query that arrives at the system, information can be mined at different levels of granularity.

[0075] The following explains query processing and query results. After efficient storage of tensors of different granularities, the most important step is to determine how to process the incoming query statement to generate the correct results. To achieve this goal, the query statement is first keyword analyzed to identify the attributes required in the query results. Here, a data storage query result attribute called resultant tensor (Resultant Tensor Rtr) can be used. This tensor consists of all the queried attributes and their values. The order and size of the result tensor can be different from the granular level tensor and the data storage tensor, depending on the number of attributes required in the query. Finally, the query results can be easily retrieved from the result tensor. The benefits of choosing to create a result tensor instead of solving the query directly are two-fold. First, the complexity of the data is reduced to only the attributes required, while eliminating unnecessary information. Second, the probability that a new query is related to the previous query is usually high, so the same RTr can be used to solve the next query, thereby shortening the query parsing time.

[0076] From the above, it can be noted that there are three steps in query processing. The first step identifies the required attributes, and the second step creates the RTr. The third step simply applies functions such as decomposition and aggregation on the RTr to obtain the result. These steps are explained in detail below.

[0077] Step 1: Determine the required attributes. To this end, keywords are extracted from the query using a known keyword extraction algorithm, and the selected keywords correspond to the attributes or attribute values ​​in the tensor.

[0078] Given the query statement: "How many hospitals in Xi'an have been treating diabetic patients aged 20-30?", it can be seen that the keywords include "Xi'an", "hospital", "20-30 years old", "diabetic patients", the keyword "Xi'an" belongs to the address attribute, "hospital" and "diabetic patients" belong to the medical attribute, and "20-30 years old" belongs to the age attribute. When the extracted keyword corresponds to the value of a certain attribute, the concept hierarchy tree is used to identify the corresponding attribute. For example, for the keyword "Xi'an", the attribute address is identified.

[0079] The above process will generate a result list L of the attributes and indicator values ​​required in the query. In addition, a concept hierarchy tree is used to identify the concept hierarchy of each attribute and indicator value in L. For example, in the query example in step 1, the concept level of age is 0, and the concept level of address is 1. Through the mapping relationship of the concept hierarchy tree, the relevant dimension matrix of the tensor of the granularity level where the relevant attributes are located is found. Therefore, the list L contains all the required attributes and indicators, as well as the concept level of each attribute and indicator.

[0080] The list L is represented by the following formula:

[0081] If exists when at the same time hour

[0082] Step 2: Create the result tensor RTr. The construction of the result tensor is divided into the following three cases according to the attributes to be queried:

[0083] (a) Assumptions That is, if L consists only of the indicator matrix, the result tensor is formed by selecting the positive slices of the granularity tensor corresponding to the indices in "L".

[0084] (b) Assumptions That is, if L consists only of attribute values, there are two cases:

[0085] The first case: All attribute values ​​are on different front slice matrices. and represents two different attribute values, and p≠x, then the resulting tensor is

[0086] The second case: some attribute values ​​are on the same front slice matrix. and represents two different attribute values, and p = x, then the resulting tensor is created by the columnar fibers where the element is located and the frontal slice matrix composed of these fibers. And s={1,2,...,I T}but The slice matrix is ​​H p =[h p1 h p2 …h pIT ]. Once all the slice matrices are obtained, the resulting tensor can be obtained by step 2(a).

[0087] (c) Assumptions That is, L contains not only the indicator matrix but also the attribute values, and the resulting tensor is obtained using both step 1 and step 2.

[0088] Step 3: Final Result. Once the RTr is created, the third stage is to apply aggregation, decomposition, and clustering functions to the RTr to obtain the result. Therefore, it can be seen that RTr provides a very flexible and efficient way to solve any type of query that arrives at the system.

[0089] The following is an explanation of the algorithm flow framework of this embodiment. The entire algorithm flow is a process of combining the data storage tensor granularization with the query processing tensor. The detailed flow is as follows:

[0090] The input of the data retrieval algorithm based on tensor and granular computing is various heterogeneous data and information query statements, and the output is the query results. The steps are:

[0091] 1. Perform keyword extraction on various types of heterogeneous data, extract key information to create data storage tensors, and establish semantic subnets based on keyword information.

[0092] 2. Use the partitioning concept of granular computing to create a corresponding concept hierarchy for each indicator and attribute in the tensor, thereby forming a corresponding concept hierarchy tree.

[0093] 3. Use the mapping method of the concept hierarchy tree to convert the data storage tensor into granularity-level tensors of different granularity levels, and the tensors of different granularity levels are mapped one by one with the concept hierarchy tree.

[0094] 4. Identify and extract keywords in the query statement and use the mapping relationship to create a result list L.

[0095] 5. Create the result tensor RTr.

[0096] 6. Apply the aggregation function on the result tensor RTr to obtain the query results.

[0097] Another embodiment of the present invention is described below. In this embodiment, the previously proposed DrTGc algorithm is combined and compared with the classic PARAFAC2 algorithm and Tucker algorithm. The performance of the DrTGc algorithm is evaluated by comparing the retrieval response time, query accuracy, and recall rate.

[0098] The first part is the preparation.

[0099] (1) Dataset

[0100] In order to demonstrate the impact of dataset size on algorithm precision, accuracy, and time spent, two datasets are used in the execution process. The first dataset used for DsTr algorithm performance evaluation is extracted from student information. This dataset has 13 different feature values, contains 230,318 instance records, and related activities of 115 independent students. The dataset includes students' learning records, as well as activity, exercise, and exam records. The second dataset is the student information dataset. This dataset contains 30 different attributes and more than 6,498 instances.

[0101] (2) Execution process

[0102] For the first student data set, the processed tensor is divided into multiple tensors according to the information granularity level of 9 features: age, grade, subject, grade, exam, exercise, material, study time, and activity time. The PARAFAC2 algorithm and the DrTGc algorithm are used for comparison in the medical data set. For the second student performance data set, information granularity levels are created for school, age, address, family size, study time, travel time, etiology, and health. Tucker decomposition and DrTGc algorithm are used for comparison. During the 60-minute retrieval process, the retrieval system sends 10 queries per minute. This embodiment creates a set of 400 different types of queries for the medical data set and 100 different types of queries for the student performance data set. The following table shows the different query types and their probability of occurrence for the two data sets. A normal query is a simple query with only one attribute. A personal query is to retrieve information about the query user. An advanced query is a query that contains multi-dimensional attributes in the query statement. A complex query refers to a query statement with more than 5 dimensions of attributes. The MATLAB code has been deployed as a Java package, and this embodiment randomly selects queries from these query sets according to the probability distribution shown in the following table. The computing platform is Windows 10, the programming language is Java, the processor is Intel(R) Core(TM) i5-7300HQ CPU@2.50GHz, the memory size is 16GB RAM, and the hard disk size is 1TB.

[0103]

[0104] The second part is the result analysis.

[0105] To evaluate the DrTGc algorithm, a 60-minute retrieval process was conducted in which the PARAFAC2 decomposition, Tucker decomposition, and DrTGc algorithms received the same number and type of queries from the created query pool.

[0106] (1) Unanswered queries. Due to the created granularity and concept level, most queries can be effectively answered after adding the algorithm proposed in this embodiment, while the number of unanswered queries of the original PARAFAC2 decomposition and original Tucker decomposition systems gradually increases with the increase of query statements. The total number of unanswered queries of the PARAFAC2 decomposition and Tucker decomposition methods increases linearly over time, while the number of unanswered queries of the DrTGc algorithm is stable at about 20. Because the data set is small, the number of unanswered queries of the DrTGc algorithm is stable at about 8 in the second student data set.

[0107] (2) Accuracy and recall. Accuracy and recall are very important indicators for any query system, and both can explain the relevance of the query. The accuracy of the DrTGc algorithm is above 85% from the beginning, and after a 10-minute retrieval process, its accuracy is between 94% and 99%. On the other hand, the accuracy of the retrieval using the PARAFAC2 algorithm is about 60% at the beginning, and the maximum accuracy can reach 90% by the end of the retrieval. The accuracy of Tucker decomposition starts with an accuracy of about 60% and reaches 87% at the end. The difference in accuracy between the DrTGc algorithm and the original Tucker algorithm is not that large after 40 minutes of retrieval, which is due to the small size of the student score data set.

[0108] The DrTGc algorithm provides a recall rate of more than 85% throughout the retrieval process, while the PARAFAC2 algorithm has a recall rate of up to 70%. The Tucker algorithm has a recall rate between 60% and 80% throughout the retrieval process. This proves that the DrTGc algorithm not only provides more answers to queries, but also the answers provided by the system are relevant.

[0109] (3) Average response time. The initial response time of the DrTGc algorithm is longer than that of the original PARAFAC2 and the original Tucker because DrTGc must create a result tensor for each query. However, when the execution time of PARAFAC2 is 32 minutes and the execution time of Tucker is 28 minutes, the response time of the system is reduced by 50% due to the cache storage of the result tensor. It can be seen that the response time of PARAFAC2 and Tucker increases linearly because the number and complexity of query statements increase linearly over time throughout the retrieval.

[0110] (4) Privacy protection. In order to study the security query problem of the algorithm proposed in this embodiment in terms of user privacy, it can be tested using the personal information queries listed in the table above. The DrTGc algorithm hardly answers any queries related to user personal privacy. This is because these query results belong to the fine-grained level of the concept hierarchy, and the retrieved results show the coarse-grained level of the concept hierarchy. However, traditional tensor decomposition methods PARAFAC2 decomposition and Tucker decomposition cannot distinguish personal queries from all other queries, which makes them unsuitable for data mining that does not meet privacy requirements. Therefore, the algorithm proposed in this embodiment can provide privacy and security functions for personal data.

[0111] The above results show that the DrTGc algorithm provides a unique method for data management and retrieval. Compared with the traditional PARAFAC2 and Tucker methods, the DrTGc algorithm has the advantages of fast calculation speed, short response time, good privacy, and strong relevance.

[0112] Big data retrieval research is a hot area in the field of data mining. This embodiment proposes a data retrieval method based on the combination of tensor and granular computing. The algorithm extracts data information in the form of information granules. At the same time, the original data information is stored in the form of a multidimensional matrix, and the multidimensional matrix is ​​mapped at different abstract levels using the conceptual hierarchy of each attribute. After analysis, compared with traditional methods such as PARAFAC2 decomposition and Tucker decomposition, the performance of the DrTGc algorithm has been greatly improved.

[0113] In the technical solution of the embodiment of the present invention, data retrieval is performed based on granularity level tensors corresponding to different attribute granularities, and query results of corresponding granularities can be returned according to the requirements of the query statement, thereby maintaining user privacy and ensuring data security. In addition, a variety of heterogeneous data including text, images, and videos are effectively stored through a third-order data storage tensor, solving the problem of data heterogeneity in data retrieval.

[0114] It should be noted that the collection, collection, updating, analysis, processing, use, transmission, storage and other aspects of user personal information involved in the technical solution of the present invention are in compliance with the provisions of relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken for user personal information to prevent illegal access to user personal information data and maintain the security of user personal information, network security and national security.

[0115] For the above-mentioned method embodiments, for the convenience of description, they are expressed as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the order of the actions described, and some steps can actually be performed in other orders or simultaneously. In addition, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily necessary to implement the present invention.

[0116] In order to better implement the above-mentioned solution of the embodiment of the present invention, relevant devices for implementing the above-mentioned solution are also provided below.

[0117] See also Figure 5 As shown, the data retrieval device 500 provided by the embodiment of the present invention may include: a storage tensor creation unit 501 , a granular tensor creation unit 502 , a query preparation unit 503 and a query execution unit 504 .

[0118] Among them, the storage tensor creation unit 501 can be used to create a data storage tensor containing multiple attributes based on the data to be stored; the granularity tensor creation unit 502 can be used to create a concept hierarchy containing multiple concept levels for each attribute, and convert the data storage tensor into a granularity level tensor corresponding to the concept level one by one according to the concept hierarchy of each attribute; the query preparation unit 503 can be used to: in response to receiving a query statement, obtain query attributes and / or query attribute values ​​from the query statement, and determine the granularity level tensor corresponding to the concept level of the query attribute and / or query attribute value in the corresponding concept hierarchy as the target granularity tensor; create a result tensor corresponding to the query statement based on the target granularity tensor; the query execution unit 504 can be used to perform operations based on preset functions on the result tensor to obtain query results corresponding to the query statement.

[0119] In an embodiment of the present invention, the data to be stored is a variety of heterogeneous data, and the heterogeneous data includes at least one of the following: text data, report record data, image data, audio data, and video data; the data storage tensor is a third-order tensor based on three basic features of time, indicators, and users, and the function includes at least one of the following: aggregation function, decomposition function, and clustering function.

[0120] In a specific application, the storage tensor creation unit 501 can be further used to: extract key information from the multiple heterogeneous data, and create the data storage tensor according to the extracted key information.

[0121] In actual applications, when adding the data of a new user to the data storage tensor, the dimension of the user's basic features increases, and the time basic features and the indicator basic features remain unchanged; when adding the data of an existing user to the data storage tensor, the dimension of the time basic features increases, and the indicator basic features and the user basic features remain unchanged.

[0122] As a preferred solution, the query preparation unit 503 may be further configured to: after acquiring the query attributes and / or query attribute values ​​from the query statement, create a result list according to the acquired query attributes and / or query attribute values.

[0123] Preferably, the query preparation unit 503 can be further used for: when the result list contains the query attribute but does not contain the query attribute value, selecting the front slicing matrix of the target granularity tensor corresponding to the query attribute contained in the result list to form the result tensor; wherein the front slicing matrix is ​​generated by slicing the target granularity tensor along the indicator basic feature; when the result list contains the query attribute value but does not contain the query attribute: if all the query attribute values ​​contained in the result list are in different front slicing matrices of the target granularity tensor, then the different front slicing matrices are combined into the result tensor; if some or all of the query attribute values ​​contained in the result list are in the same front slicing matrix of the target granularity tensor, then the column fibers where the query attribute value is located and the front slicing matrix composed of the column fibers are combined into the result. tensor; wherein, the column fibers are generated by slicing the frontal slice matrix along the time-based features; when the result list contains both query attributes and query attribute values: the frontal slice matrix of the target granularity tensor corresponding to the query attributes contained in the result list is selected to form a first intermediate tensor; if all the query attribute values ​​contained in the result list are in different frontal slice matrices of the target granularity tensor, the different frontal slice matrices are combined into a second intermediate tensor, and the first intermediate tensor and the second intermediate tensor are combined to form the result tensor; if some or all of the query attribute values ​​contained in the result list are in the same frontal slice matrix of the target granularity tensor, the column fibers where the query attribute values ​​are located and the frontal slice matrix composed of the column fibers are combined into a third intermediate tensor, and the first intermediate tensor and the third intermediate tensor are combined to form the result tensor.

[0124] Furthermore, in the embodiment of the present invention, the concept hierarchy is a tree structure.

[0125] According to the technical solution of the embodiment of the present invention, data retrieval is performed based on granularity level tensors corresponding to different attribute granularities, and query results of corresponding granularity can be returned according to the requirements of the query statement, thereby maintaining user privacy and ensuring data security. In addition, a variety of heterogeneous data including text, images, and videos are effectively stored through a third-order data storage tensor, solving the problem of data heterogeneity in data retrieval.

[0126] Figure 6 An exemplary system architecture 600 is shown to which the data retrieval method or data retrieval device according to the embodiment of the present invention can be applied.

[0127] like Figure 6 As shown, the system architecture 600 may include terminal devices 601, 602, 603, a network 604 and a server 605 (this architecture is only an example, and the components included in the specific architecture may be adjusted according to the specific application). The network 604 is used to provide a medium for a communication link between the terminal devices 601, 602, 603 and the server 605. The network 604 may include various connection types, such as wired, wireless communication links or optical fiber cables.

[0128] The user can use the terminal devices 601, 602, 603 to interact with the server 605 through the network 604 to receive or send messages, etc. Various client applications can be installed on the terminal devices 601, 602, 603, such as a data query application (only an example).

[0129] The terminal devices 601 , 602 , and 603 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers, etc.

[0130] The server 605 may be a server that provides various services, such as a backend server (for example only) that supports data query applications operated by users using the terminal devices 601, 602, and 603. The backend server may process the received query request and feed back the processing result (for example, the query result - for example only) to the terminal devices 601, 602, and 603.

[0131] It should be noted that the data retrieval method provided in the embodiment of the present invention is generally executed by the server 605 , and accordingly, the data retrieval device is generally disposed in the server 605 .

[0132] It should be understood that Figure 6 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0133] The present invention also provides an electronic device. The electronic device of the embodiment of the present invention comprises: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the data retrieval method provided by the present invention.

[0134] Reference below Figure 7 , which shows a schematic diagram of the structure of a computer system 700 of an electronic device suitable for implementing an embodiment of the present invention. Figure 7 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0135] like Figure 7 As shown, the computer system 700 includes a central processing unit (CPU) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage part 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the computer system 700 are also stored. The CPU 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0136] The following components are connected to the I / O interface 705: an input section 706 including a keyboard, a mouse, etc.; an output section 707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, a modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 710 as needed so that a computer program read therefrom is installed into the storage section 708 as needed.

[0137] In particular, according to the embodiments disclosed in the present invention, the process described in the main step diagram above can be implemented as a computer software program. For example, the embodiments of the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the main step diagram. In the above embodiment, the computer program can be downloaded and installed from the network through the communication part 709, and / or installed from the removable medium 711. When the computer program is executed by the central processing unit 701, the above functions defined in the system of the present invention are executed.

[0138] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer readable signal media may also be any computer readable medium other than computer readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0139] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the above-mentioned module, program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0140] The units involved in the embodiments of the present invention may be implemented in software or hardware. The units described may also be provided in a processor, for example, may be described as: a processor including a storage tensor creation unit, a granular tensor creation unit, a query preparation unit, and a query execution unit. The names of these units do not, in some cases, constitute limitations on the units themselves, for example, a storage tensor creation unit may also be described as a "unit that provides a data storage tensor to a granular tensor creation unit".

[0141] As another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiment; or it may exist independently and not be assembled into the device. The above computer-readable medium carries one or more programs. When the above one or more programs are executed by the device, the steps executed by the device include: creating a data storage tensor containing multiple attributes based on the data to be stored; creating a concept hierarchy containing multiple concept levels for each attribute, and converting the data storage tensor into a granularity level tensor corresponding to the concept level one by one according to the concept hierarchy of each attribute; in response to receiving a query statement, obtaining the query attribute and / or the query attribute value from the query statement, and determining the granularity level tensor corresponding to the concept level of the query attribute and / or the query attribute value in the corresponding concept hierarchy as the target granularity tensor; creating a result tensor corresponding to the query statement based on the target granularity tensor; performing an operation based on a preset function on the result tensor to obtain the query result corresponding to the query statement.

[0142] In the technical solution of the embodiment of the present invention, data retrieval is performed based on granularity level tensors corresponding to different attribute granularities, and query results of corresponding granularities can be returned according to the requirements of the query statement, thereby maintaining user privacy and ensuring data security. In addition, a variety of heterogeneous data including text, images, and videos are effectively stored through a third-order data storage tensor, solving the problem of data heterogeneity in data retrieval.

[0143] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A data retrieval method, It is characterized in that include: Create a data storage tensor with multiple attributes based on the data to be stored; Creating a concept hierarchy structure containing a plurality of concept levels for each attribute, and converting the data storage tensor into a granularity level tensor corresponding to the concept level one by one according to the concept hierarchy structure of each attribute; In response to receiving a query statement, obtaining a query attribute and / or a query attribute value from the query statement, and determining a granularity level tensor corresponding to a concept level of the query attribute and / or the query attribute value in a corresponding concept hierarchy as a target granularity tensor; Creating a result tensor corresponding to the query statement based on the target granularity tensor; An operation based on a preset function is performed on the result tensor to obtain a query result corresponding to the query statement.

2. The method according to claim 1, It is characterized in that The data to be stored is a plurality of heterogeneous data; and the step of creating a data storage tensor containing a plurality of attributes based on the data to be stored includes: Key information is extracted from the multiple heterogeneous data, and the data storage tensor is created according to the extracted key information.

3. The method according to claim 1, It is characterized in that The data storage tensor is a third-order tensor based on three basic features: time, index, and user; When adding the data of a new user to the data storage tensor, the dimension of the user basic features increases, and the time basic features and the indicator basic features remain unchanged; When the data of an existing user is added to the data storage tensor, the dimension of the time-based feature increases, and the indicator-based feature and the user-based feature remain unchanged.

4. The method according to claim 3, It is characterized in that The method further comprises: After obtaining the query attributes and / or query attribute values ​​from the query statement, a result list is created according to the obtained query attributes and / or query attribute values.

5. The method according to claim 4, It is characterized in that The creating a result tensor corresponding to the query statement based on the target granularity tensor includes: In the case where the result list contains the query attribute but does not contain the query attribute value, a positive slice matrix of the target granularity tensor corresponding to the query attribute contained in the result list is selected to form the result tensor; wherein the positive slice matrix is ​​generated by slicing the target granularity tensor along the indicator basic feature; In the case where the result list contains query attribute values ​​and does not contain query attributes: if all query attribute values ​​contained in the result list are in different frontal slice matrices of the target granularity tensor, the different frontal slice matrices are composed into the result tensor; if some or all query attribute values ​​contained in the result list are in the same frontal slice matrix of the target granularity tensor, the column fibers where the query attribute values ​​are located and the frontal slice matrix composed of the column fibers are composed into the result tensor; wherein the column fibers are generated by slicing the frontal slice matrix along the time-based feature; In the case where the result list contains both query attributes and query attribute values: the front slicing matrix of the target granularity tensor corresponding to the query attributes contained in the result list is selected to form a first intermediate tensor; if all the query attribute values ​​contained in the result list are in different front slicing matrices of the target granularity tensor, the different front slicing matrices are combined into a second intermediate tensor, and the first intermediate tensor and the second intermediate tensor are combined to form the result tensor; if some or all of the query attribute values ​​contained in the result list are in the same front slicing matrix of the target granularity tensor, the column fibers where the query attribute values ​​are located and the front slicing matrix composed of the column fibers are combined into a third intermediate tensor, and the first intermediate tensor and the third intermediate tensor are combined to form the result tensor.

6. The method according to claim 2, It is characterized in that The heterogeneous data includes at least one of the following: text data, report record data, image data, audio data, and video data; The concept hierarchy is a tree structure; The function includes at least one of the following: an aggregation function, a decomposition function, and a clustering function.

7. A data retrieval device, It is characterized in that include: A storage tensor creation unit, used to create a data storage tensor containing multiple attributes based on the data to be stored; A granularity tensor creation unit, configured to create a concept hierarchy structure containing a plurality of concept levels for each attribute, and convert the data storage tensor into a granularity level tensor corresponding to each concept level according to the concept hierarchy structure of each attribute; A query preparation unit, configured to: in response to receiving a query statement, obtain a query attribute and / or a query attribute value from the query statement, and determine a granularity level tensor corresponding to a concept level of the query attribute and / or the query attribute value in a corresponding concept hierarchy as a target granularity tensor; Creating a result tensor corresponding to the query statement based on the target granularity tensor; The query execution unit is used to perform an operation based on a preset function on the result tensor to obtain a query result corresponding to the query statement.

8. The device according to claim 7, It is characterized in that The data to be stored is a plurality of heterogeneous data, and the heterogeneous data includes at least one of the following: text data, report record data, image data, audio data, and video data; The data storage tensor is a third-order tensor based on three basic features: time, index, and user. The function includes at least one of the following: an aggregation function, a decomposition function, and a clustering function.

9. An electronic device, It is characterized in that include: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Citation Information

Patent Citations

  • Information retrieval method based on voice semantics and related equipment thereof

    CN111782763A

  • Statistical machine learning and business process models systems and methods

    US9092802B1

Cited By

  • Query response method and related device

    WO2026179627A1