Small sample knowledge reasoning method and device based on three partial sequence structures and medium

Abstract

The invention discloses a small sample knowledge reasoning method and device based on a three-branch partial sequence structure and a medium. According to the method, three partial sequence structure diagrams are generated according to data to be analyzed, particle splitting operation, particle recombination operation and particle isomorphism operation of attribute particles are carried out layer by layer on the basis of the three partial sequence structure diagrams, and then a new object mode is generated and summarized and output. According to the embodiment of the invention, data augmentation can be realized through deductive reasoning from the perspective of symbolic representation, and compared with an existing vector-based small sample reasoning method, the method is more suitable for processing small sample problems with higher interpretability requirements, and can be widely applied to small sample reasoning learning processes requiring high interpretability, such as syndrome differentiation and treatment in the field of traditional Chinese medicine.

Description

Technical Field

[0001] The present invention relates to the technical field of small sample knowledge reasoning, and in particular to a small sample knowledge reasoning method, device and medium based on a three-branch partial order structure. Background Art

[0002] Few-shot learning (FSL), as a learning model for small sample data problems, is an important and challenging problem in the field of artificial intelligence. The massive data generated in the big data era has promoted the rapid development of artificial intelligence technology, which has led to the situation that most artificial intelligence models can only rely on a large number of samples to fully learn. However, in some specific fields, such as medical research, face recognition, and anomaly detection, it is difficult to obtain a large number of samples. In response to this situation, scholars began to think about whether the knowledge learned from small samples can be used to solve this problem. This conjecture has aroused widespread research interest among scholars, and thus the idea of ​​small-shot learning has gradually emerged.

[0003] At present, some explorations on small-sample knowledge reasoning have achieved preliminary results, such as: small-sample link prediction on knowledge graphs through meta-learning, small-sample node classification on attribute networks, and small-sample relationship learning on knowledge graphs through hybrid attention mechanisms.

[0004] However, most of the above-mentioned results focus on small-sample reasoning from the perspective of vector representation, and their interpretability is relatively low, so they cannot be effectively applied in fields that require high interpretability. Summary of the invention

[0005] In view of this, the embodiments of the present invention provide a small sample knowledge reasoning method, device and medium based on a three-branch partial order structure, which is based on the concept recognition learning method of the three-branch partial order structure.

[0006] A first aspect of the present invention provides a small sample knowledge reasoning method based on a three-branch partial order structure, comprising the following steps:

[0007] Acquire data to be analyzed, and generate a three-branch partial order structure graph according to the data to be analyzed;

[0008] The first layer from top to bottom of the three-branch partial order structure graph is used as the target layer;

[0009] Get the attribute particles of the target layer;

[0010] When the attribute particles of the target layer meet the particle splitting conditions, the particle splitting operation is performed on the attribute particles of the target layer to split one attribute particle into multiple new attribute particles;

[0011] When the attribute particles of the target layer meet the particle reorganization conditions, the particle reorganization operation is performed on the attribute particles of the target layer to combine multiple attribute particles into a new attribute particle;

[0012] When the attribute granules of the target layer meet the particle isomorphism condition, the particle isomorphism operation is performed on the attribute granules of the target layer, multiple attribute granules are isomorphized into new attribute granules and a new object mode is generated;

[0013] Determine whether a new object mode is generated in the particle isomorphism operation; when a new object mode is generated, return to the step of performing a particle reorganization operation on the attribute particles of the target layer when the attribute particles of the target layer meet the particle reorganization conditions, and when the attribute particles of the target layer meet the particle reorganization conditions, perform a particle reorganization operation and a particle isomorphism operation on the attribute particles of the target layer until no new object mode is generated;

[0014] When no new object pattern is generated, the next layer is added to the target layer, and the step of obtaining the attribute particles of the target layer is returned. When the attribute particles of the target layer meet the particle splitting conditions, the attribute particles of the target layer are subjected to particle splitting operations, particle reorganization operations, and particle isomorphism operations until all levels in the three-branch partial order structure diagram are processed;

[0015] All newly generated object patterns are obtained from the three partial order structure graphs and summarized and output.

[0016] Further, the data to be analyzed is represented by triple data, specifically including an object set, an attribute set, and an object-attribute correspondence relationship; the object set has multiple object modes, the attribute set has multiple attribute elements; the object-attribute correspondence relationship refers to the correspondence relationship between the object mode and the attribute element;

[0017] Generating a three-branch partial order structure diagram according to the data to be analyzed specifically includes the following steps:

[0018] Expand the attribute element with the most object-attribute correspondences as the root node, and generate a tree diagram with each object pattern corresponding to the root node as a branch;

[0019] On each branch, the attribute elements corresponding to the object pattern are arranged as nodes from top to bottom in a preset attribute element order;

[0020] Each group of nodes extending downward from the root node is considered as a level;

[0021] Check the attribute elements of the nodes in each level. If the same attribute element appears in nodes of multiple levels, merge the node into the node of the level where the attribute element first appears on the corresponding branch, so that after the merger, the node of the branch at this level has multiple attribute elements.

[0022] Merge the nodes with the same extension path in different branches expanded by the root node;

[0023] In the three-branch partial order structure diagram, each node forms an attribute granule; the attribute granule is represented by triple data, which are the core attribute of the node corresponding to the attribute granule, the attribute element set of the line from the node corresponding to the attribute granule to the root node, and the object mode set corresponding to the node corresponding to the attribute granule; the core attribute refers to one or more attribute elements possessed by the node corresponding to the attribute granule.

[0024] Furthermore, before the step of taking the first layer as the target layer, the following steps are also included:

[0025] Determine whether the root node of the three-branch partial order structure graph can perform a particle splitting operation;

[0026] When the root node of the three-branch partial order structure graph can be subjected to a particle splitting operation, the particle splitting operation is performed on the root node of the three-branch partial order structure graph, and the new attribute particle formed by the splitting is used as a new root node, and the original root node is used as a child node of the new root node.

[0027] Furthermore, the particle splitting operation is implemented by a preset particle splitting operator, and specifically includes the following steps:

[0028] Establish a first set, a second set and a third set; each target layer attribute particle is an element in the first set, and the second set and the third set are empty sets;

[0029] Traverse each attribute grain in the first set in turn, and take the first attribute grain visited as the target attribute grain;

[0030] Determine the core attribute of the target attribute particle;

[0031] If the target attribute particle has more than one core attribute, a preset particle splitting operator is used to perform a particle splitting operation on the target attribute particle to generate multiple new attribute particles;

[0032] Adding the new attribute particle obtained by the particle splitting operation as an element to the third set;

[0033] Traversing each new attribute particle in the third set in turn, and comparing each new attribute particle in the third set with the union of the first set and the second set;

[0034] If there is an attribute grain whose core attribute and parent node are exactly the same as the new attribute grain in the third set in the union of the first set and the second set, then the attribute element set of the new attribute grain is added to the attribute element set of the target attribute grain, and the new attribute grain is deleted from the third set;

[0035] Add the elements in the third set to the second set, and clear the third set;

[0036] The next attribute granule to be visited is taken as the target attribute granule, and the process returns to the step of determining the core attribute of the target attribute granule until each attribute granule in the first set is visited;

[0037] Each of the new attribute particles in the second set is taken as a node, added to the level corresponding to the target attribute particle in the three-branch partial order structure graph, and connected to the parent node of the target attribute particle corresponding to each new attribute particle.

[0038] Furthermore, the particle reorganization operation is implemented by a preset particle reorganization operator, and specifically includes the following steps:

[0039] Establishing a fourth set, a fifth set and a sixth set; in the fourth set, each target layer attribute particle is an element, and the fifth set and the sixth set are empty sets;

[0040] Arbitrarily combining each target layer attribute particle in the fourth set to form a plurality of attribute particle pairs;

[0041] Traversing the plurality of attribute particle pairs in sequence, and taking the first pair of attribute particles visited as the target attribute particle pair;

[0042] Determine whether the target attribute particle pair meets the particle recombination condition;

[0043] If the target attribute particle pair meets the particle recombination condition, a preset particle recombination operator is used to perform a particle recombination operation on the target attribute particle pair to combine them into a new attribute particle;

[0044] Adding new attribute particles obtained by the particle reorganization operation as elements to the sixth set;

[0045] Traversing each new attribute particle in the sixth set in turn, and comparing each new attribute particle in the sixth set with the union of the fourth set and the fifth set;

[0046] If there is an attribute grain whose core attribute and parent node are exactly the same as the new attribute grain in the sixth set in the union of the fourth set and the fifth set, then the new attribute grain is deleted from the sixth set;

[0047] Add the elements in the sixth set to the fifth set, and clear the sixth set;

[0048] The next pair of attribute grains to be visited is taken as the target attribute grain pair, and the process returns to the step of determining whether the target attribute grain pair satisfies the grain reorganization condition, until every pair of attribute grains in the fourth set is visited;

[0049] Each of the new attribute particles in the fifth set is taken as a node, added to the level corresponding to the target attribute particle in the three-branch partial order structure graph, and connected to the parent node of the target attribute particle corresponding to each new attribute particle.

[0050] Furthermore, the particle recombination conditions specifically include:

[0051] The corresponding nodes of each attribute particle in the target attribute particle pair have a common parent node;

[0052] The corresponding nodes of each attribute particle in the target attribute particle pair are located at the same level;

[0053] The corresponding nodes of each attribute particle in the target attribute particle pair correspond to different object modes;

[0054] When the target attribute particle pair satisfies the above conditions at the same time, it is considered that the target attribute particle pair satisfies the particle recombination condition.

[0055] Furthermore, the particle isomorphism operation is implemented by a preset particle isomorphism operator, and specifically includes the following steps:

[0056] Establishing a seventh set, an eighth set and a ninth set; in the seventh set, each target layer attribute particle is an element, and the eighth set and the ninth set are empty sets;

[0057] Arbitrarily combining each target layer attribute particle in the seventh set to form a plurality of attribute particle pairs;

[0058] Traversing the plurality of attribute grain pairs in sequence, taking the first pair of attribute grains visited as the target attribute grain pair; the target attribute grain pair includes a first target attribute grain and a second target attribute grain;

[0059] Determining whether the target attribute particle pair satisfies a particle isomorphism condition;

[0060] If the target attribute particle pair meets the particle recombination condition, the following steps are performed on the first target attribute particle: a particle isomorphism operation is performed on the first target attribute particle using the second target attribute particle based on the first target attribute particle using a preset particle recombination operator to obtain a first new attribute particle; the first new attribute particle is added as an element to the eighth set; it is determined whether the core attribute of each first new attribute particle in the eighth set is an empty set; if so, a node corresponding to the first new attribute particle is added to the next layer of the level corresponding to the first target attribute particle in the three-branch partial order structure diagram, and the node corresponding to the first new attribute particle is treated as a child node of the node corresponding to the first target attribute particle; the branch corresponding to the first new attribute particle is generated as a new object pattern;

[0061] The following steps are performed on the second target attribute particle: a preset particle recombination operator is used to perform a particle isomorphism operation on the first target attribute particle based on the second target attribute particle to obtain a second new attribute particle; the second new attribute particle is added as an element to the ninth set; it is determined whether the core attribute of each second new attribute particle in the ninth set is an empty set; if so, a node corresponding to the second new attribute particle is added to the next layer of the level corresponding to the second target attribute particle in the three-branch partial order structure diagram, and the node corresponding to the second new attribute particle is treated as a child node of the node corresponding to the second target attribute particle; the branch corresponding to the second new attribute particle is generated as a new object pattern;

[0062] Clearing the eighth set and the ninth set;

[0063] The next pair of attribute particles to be visited is taken as the target attribute particle pair, and the process returns to the step of determining whether the target attribute particle pair satisfies the particle isomorphism condition, until every pair of attribute particles in the seventh set is visited;

[0064] The new object patterns generated by the particle isomorphism operation are summarized and output.

[0065] Furthermore, the particle isomorphism conditions specifically include:

[0066] The corresponding nodes of each attribute particle in the target attribute particle pair are located at the same level;

[0067] The intersection of the core attributes of each attribute particle in the target attribute particle pair is a non-empty set;

[0068] When the target attribute particle pair satisfies the above conditions at the same time, it is considered that the target attribute particle pair satisfies the particle isomorphism condition.

[0069] A second aspect of the present invention discloses an electronic device, comprising a processor and a memory;

[0070] The memory is used to store programs;

[0071] The processor executes the program to implement the small sample knowledge reasoning method based on a three-branch partial order structure.

[0072] The third aspect of the present invention discloses a computer-readable storage medium, characterized in that the storage medium stores a program, and the program is executed by a processor to implement the small sample knowledge reasoning method based on a three-branch partial order structure.

[0073] The embodiment of the present invention also discloses a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device can read the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the above method.

[0074] The embodiments of the present invention have the following beneficial effects: the small sample knowledge reasoning method, device and medium based on the three-branch partial order structure proposed by the present invention can realize data augmentation through deductive reasoning from the perspective of symbolic representation, and obtain new knowledge from the original data through particle splitting, combination and isomorphism operations. Therefore, compared with the existing vector-based small sample reasoning method, the present invention is more suitable for processing small sample problems with higher interpretability requirements, and is widely used in the small sample reasoning learning process that requires high interpretability, such as syndrome differentiation and treatment in the field of traditional Chinese medicine.

[0075] Additional aspects and advantages of the present invention will be given in part in the following description, and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0077] Figure 1 It is the main implementation flow chart of a small sample knowledge reasoning method based on a three-branch partial order structure of the present invention.

[0078] Figure 2 It is a schematic diagram of a three-branch partial order structure generated by the present invention based on the data to be analyzed in an embodiment of the present invention.

[0079] Figure 3 It is a detailed implementation flow chart of another small sample knowledge reasoning method based on a three-branch partial order structure of the present invention.

[0080] Figure 4 This is an incomplete picture of an unprocessed prescription such as Guizhi Decoction.

[0081] Figure 5 The present invention discloses a Guizhi Decoction-type prescription completion diagram processed by a small sample knowledge reasoning method based on a three-branch partial order structure.

[0082] Figure 6 It is a structural schematic diagram of an electronic device of the present invention.

[0083] Figure 7 It is a schematic diagram of the structure of a computer-readable storage medium of the present invention. DETAILED DESCRIPTION

[0084] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0085] Compared with vector-based representation, symbolic representation has relatively high interpretability. Therefore, it is necessary to explore small sample reasoning from the perspective of symbolic representation, such as concept-based small sample reasoning. To this end, the embodiments of the present invention provide a small sample knowledge reasoning method, device and medium based on a three-branch partial order structure from the perspective of symbolic representation and based on the concept cognition learning method of the three-branch partial order structure.

[0086] like Figure 1 As shown, the first embodiment of the present invention provides a small sample knowledge reasoning method based on a three-branch partial order structure, comprising the following steps:

[0087] S1. Obtain the data to be analyzed, and generate a three-branch partial order structure diagram based on the data to be analyzed.

[0088] S2. The first layer of the three-branch partial order structure graph from top to bottom is taken as the target layer;

[0089] S3. Get the attribute particles of the target layer;

[0090] S4. When the attribute particles of the target layer meet the particle splitting conditions, a particle splitting operation is performed on the attribute particles of the target layer to split one attribute particle into multiple new attribute particles;

[0091] S5. When the attribute particles of the target layer meet the particle reorganization conditions, a particle reorganization operation is performed on the attribute particles of the target layer to combine multiple attribute particles into a new attribute particle;

[0092] S6. When the attribute particles of the target layer meet the particle isomorphism condition, a particle isomorphism operation is performed on the attribute particles of the target layer, multiple attribute particles are isomorphized into new attribute particles and a new object mode is generated;

[0093] S7. Determine whether a new object pattern is generated in the particle isomorphism operation; when a new object pattern is generated, return to the step of performing a particle reorganization operation on the attribute particles of the target layer when the attribute particles of the target layer meet the particle reorganization conditions, and when the attribute particles of the target layer meet the particle reorganization conditions, perform a particle reorganization operation and a particle isomorphism operation on the attribute particles of the target layer until no new object pattern is generated;

[0094] When no new object pattern is generated, the next layer is added to the target layer, and the step of obtaining the attribute particles of the target layer is returned. When the attribute particles of the target layer meet the particle splitting conditions, the attribute particles of the target layer are subjected to particle splitting, particle reorganization, and particle isomorphism operations until all levels in the three-branch partial order structure diagram are processed;

[0095] S8. Get all newly generated object patterns from the three-branch partial order structure graph and summarize the output.

[0096] The present invention proposes a small sample knowledge reasoning method based on a three-branch partial order structure, which can realize data augmentation through deductive reasoning from the perspective of symbolic representation, and can be applied to small sample reasoning learning processes that require high interpretability, such as syndrome differentiation and treatment in the field of traditional Chinese medicine.

[0097] As a further embodiment, the implementation process of the main steps of the present invention is specifically discussed below:

[0098] S1. Obtain the data to be analyzed, and generate a three-branch partial order structure diagram based on the data to be analyzed.

[0099] In the embodiment of the present invention, the data to be analyzed is represented by triple data K={U,M,I}, which specifically includes an object set U, an attribute set M, and an object-attribute correspondence I. The object set U consists of multiple object patterns {1,2,3,...}, the attribute set M consists of multiple attribute elements {a,b,c,...}, and the object-attribute correspondence I is a binary correspondence between the object set U and the attribute set M, that is,

[0100] Table 1 Example of data to be analyzed

[0101]

[0102] Table 1 shown above is an exemplary diagram of the data to be analyzed. The data in the first column on the left side of Table 1 is the object set U = {1, 2, 3, 4, 5, 6, 7, 8}, the data in the first row on the upper side is the attribute set M = {a, b, c, d, e, f, g, h}, and the data of the intersection unit of the first row and the first column is the object-attribute correspondence relationship In the intersection unit, the number "1" indicates that the object mode of the corresponding row has the attribute element of the corresponding column, and there is a binary correspondence between the two. If it is empty, it means that there is no binary correspondence.

[0103] Taking the data in Table 1 as an example, the process of generating the three-branch partial order structure diagram in step S1 of the embodiment of the present invention is described:

[0104] S1-1. Expand the attribute element with the most object-attribute correspondence as the root node, and generate a tree diagram with each object pattern corresponding to the root node as a branch. In Table 1, the attribute element with the most object-attribute correspondence is "a", so Figure 2 In the tree diagram, a is used as the root node. Since the object patterns corresponding to attribute element a include {1,2,3,4,5,6,7,8}, 8 branches need to be expanded from node a.

[0105] In some embodiments, there may be object patterns that have no corresponding relationship with the attribute elements with the most object-attribute correspondences. For these object patterns, it is necessary to find the attribute elements with the most object-attribute correspondences in the data table to be analyzed among the attribute elements with the object-attribute correspondences with these object patterns, and establish multiple three-branch partial order structure graphs for processing. The establishment process of each three-branch partial order structure graph is consistent with the establishment process of the three-branch partial order structure graph described in the embodiment of the present invention, and will not be repeated here.

[0106] S1-2. On each branch, the attribute elements corresponding to the object pattern are arranged from top to bottom as nodes according to the preset attribute element order. According to the data in Table 1, the node order of each branch is [a,b,g], [a,b,g,h], [a,b,c,g,h], [a,c,g,h,i], [a,b,d,f], [a,b,c,d,f], [a,c,d,e] and [a,c,d,f].

[0107] S1-3. Treat each group of nodes extending downward from the root node as a level. According to the order of the attribute elements in step S1-2, the attribute element set of the first level can be arranged as [b,c], the attribute element set of the second level can be arranged as [g,c,d], the attribute element set of the third level can be arranged as [h,g,f,e], and the attribute element set of the fourth level can be arranged as [h,i,f].

[0108] S1-4. Check the attribute elements of the nodes in each level. If the same attribute element appears in nodes of multiple levels, merge the node into the node of the level where the attribute element first appears on the corresponding branch, so that after the merger, the node of the branch at this level has multiple attribute elements.

[0109] According to the attribute element set of each layer obtained in step S1-3, it can be found that the attribute element [c] of the first layer also appears in the second layer, and the attribute element [c] that appears in the second layer belongs to object mode 3 and object mode 6. Therefore, it is necessary to merge the attribute nodes of the first and second layers in object mode 3 and object mode 6 to form a composite node. That is, the node order of the corresponding branch of object mode 3 is [a, bc, g, h]; the node order of the corresponding branch of object mode 6 is [a, bc, d, f].

[0110] After completing the merging of attribute elements, the attribute element set of the first layer of the three-branch partial order structure graph generated based on Table 1 is [b, c, bc], the attribute element set of the second layer is [d, g], the attribute element set of the third layer is [e, f, h], and the attribute element of the fourth layer is [i]. Check again whether the attribute elements of the nodes in each level appear in more than one level, and find that each attribute element appears in only one level, so the processing of step S1-4 is considered to be completed.

[0111] S1-5. Merge the nodes with the same extension path in different branches expanded by the root node.

[0112] In the embodiment of the present invention, nodes with the same departure node and arrival node are regarded as nodes with the same extension path. For example, the branch node of object mode 1 is [a, b, g], and the branch node of object mode 2 is [a, b, g, h]. They have the same extension path [a, b, g]. Therefore, it is necessary to merge the same part [a, b, g] of the branch corresponding to object mode 1 and the corresponding branch of object mode 2.

[0113] The three-branch partial order structure diagram finally obtained based on the data in Table 1 is as follows Figure 2 shown. Figure 2 The three-branch partial order structure in has 4 levels and 8 branches, each branch corresponding to an object pattern in Table 1.

[0114] S2. Take the first layer of the three-branch partial order structure graph from top to bottom as the target layer.

[0115] After the construction of the three-branch partial order structure graph is completed, the nodes are divided into multiple levels starting from the root node of the three-branch partial order structure graph, and the attribute particle processing operation is performed layer by layer in a progressive manner. The division method of the levels is the same as the division method of step S1-3, which will not be repeated here.

[0116] In some embodiments, before determining the target layer in step S2, the following steps are also included:

[0117] S1-01. Determine whether the root node of the three-branch partial order structure graph can perform a particle splitting operation;

[0118] S2-02. When the root node of the three-branch partial order structure graph is capable of a particle splitting operation, the root node of the three-branch partial order structure graph is subjected to a particle splitting operation. The particle splitting operation on the root node is applicable to the case where there is more than one core attribute of the root node, that is, the attribute elements appear in a combined form in the attribute element set M of the data to be analyzed. Since the particle splitting, particle reorganization and particle isomorphism operations on the nodes in the embodiment of the present invention start from the first layer of the three-branch partial order structure graph, it is easy to omit the knowledge expansion of the root node. However, the root node itself does not meet the respective conditions of the particle reorganization and particle isomorphism operations. Therefore, in this part of the embodiments, the particle splitting operation on the root node is supplemented to better acquire knowledge from the data to be analyzed. The implementation process of the method of this part of the embodiments is as follows. Figure 3 shown.

[0119] S3. Obtain the attribute particles of the target layer.

[0120] The following is based on Figure 2 The concepts of attribute granules and object patterns involved in the embodiments of the present invention are described as follows:

[0121] In the three-branch partial order structure graph, each node forms an attribute granule (AG), which is represented by the triple data AG = {K, O, A}. K represents the core attribute of the node corresponding to the attribute granule. For example, the three nodes AG in the first layer 11 AG 12 and AG 13 The core attributes of the attribute granules formed are [b], [c] and [bc] respectively. O represents the object mode of the node corresponding to the attribute granule. Since the node merging operation is performed in step S1-5, the attribute granule at the top level may have more than one object mode. For example, the object modes of the attribute granules formed by the three nodes in the first layer are 125, 478 and 36 respectively. This means that object mode 1, object mode 2 and object mode 5 pass through node AG. 11 ; Object mode 4, object mode 7 and object mode 8 pass through node AG 12 ; Object mode 3 and object mode 6 pass through node AG 13 A represents the attribute element set of the line from the attribute particle corresponding to the node to the root node. For example, the attribute element set of the attribute particle formed by the three nodes in the first layer is [ab], [ac] and [abc]. In summary, Figure 2 The attribute particles of the three nodes in the first layer can be expressed as: AG 11 =(b,125,ab),AG 12 =(c,478,ac),AG 13 =(bc,36,abc).

[0122] In the three-branch partial order structure diagram, each branch from top to bottom corresponds to an object pattern (ObjectPattern, OP). The object pattern can also be represented by a two-tuple data OP=(O,A). O represents the object set element of the object pattern, for example, the object set element of object pattern 1 (OP1) is 1, and the object set element of object pattern 2 (OP2) is 2. A represents the attribute element that has an object-attribute correspondence relationship with the object pattern, for example, the attribute element of object pattern 1 (OP1) is [abdf], and the attribute element of object pattern 2 (OP2) is [abgh]. In summary, Figure 2 The object pattern of each branch can be described as follows: OP1 = (5, abdf), OP2 = (2, abgh), OP3 = (1, abg), OP4 = (8, acdf), OP5 = (7, acde), OP6 = (4, acghi), OP7 = (6, abcdf), OP8 = (3, abcgh).

[0123] S4. When the attribute particles of the target layer meet the particle splitting conditions, a particle splitting operation is performed on the attribute particles of the target layer to split one attribute particle into multiple new attribute particles.

[0124] In the embodiment of the present invention, for a given attribute particle AG={K,O,A} in a three-branch attribute partial order structure, if the core attribute K={a1,a2,…,a k} contains multiple elements (|K|>1), then theoretically, the attribute particle AG has the ability to split into new particles.

[0125] The specific process of particle splitting operation of the present invention is described below:

[0126] S4-1. Establish the first set AGL, the second set AGR and the third set AGS; in the first set, each target layer attribute particle is an element, that is, AGL = {AG}, the second set and the third set are empty sets

[0127] S4-2. Traverse each attribute granule AG in the first set in turn, and use the first attribute granule AG1 to be visited as the target attribute granule. As a more specific embodiment, the attribute granules AG in the first set AGL can be sorted in descending order according to a certain arrangement order of core elements |AG.K|, and then each attribute granule is visited in turn according to the order.

[0128] S4-3. Determine the core attribute of the target attribute particle AG1; if the target attribute particle AG1 has more than one core attribute AG1.K, that is, |AG1.K|>1, use the preset particle splitting operator gs to perform a particle splitting operation on the target attribute particle AG1 to generate multiple new attribute particles AGi .

[0129] Assume 2 K is the power set of K, then the particle splitting operator gs can be described as follows:

[0130]

[0131] As shown in Formula 1, for each new attribute particle AG generated by particle splitting gs i , its core attribute set AG i .K and concept connotation AG i .A are subsets of the target attribute particles K and A respectively; while its concept extension AGi.O has a certain degree of uncertainty, and the uncertainty part {?} is exactly what the entire knowledge reasoning process needs to fill.

[0132] S4-4. The new attribute particle AG obtained by the particle splitting operation i Added as an element to the third set AGS, that is, AGS = gs(AG1) = {AG i}.

[0133] S4-5. Traverse each new attribute particle AG in the third set AGS in turn i , each new attribute particle AG in the third set AGS i Compare in the union of the first set and the second set AGL∪AGR; if there is a core attribute AG in the union of the first set and the second set AGL∪AGR i .K and its parent node are exactly the same as the new attribute grain in the third set, then the attribute element set of the new attribute grain is added to the attribute element set of the target attribute grain, that is, AG1.A = AGi.A∪AG1.A; at the same time, the new attribute grain AG i Deleted from the third set AGS.

[0134] S4-6. Add the element AG in the third set AGS i Add to the second set AGR, clear the third set AGS;

[0135] S4-7. The next attribute grain to be accessed is used as the target attribute grain AG n , return to the step of determining the core attributes of the target attribute particle, until each attribute particle AG in the first set AGL is visited;

[0136] S4-8. The new attribute particle AG in the second set AGR i Each is added as a node to the three-branch partial order structure graph and the target attribute particle AG n In the corresponding level, and connect each new attribute particle AG iThe corresponding target attribute particle AG n The parent node of .

[0137] As a further explanation, according to the principle of granular computing, for the particle splitting operator, Theorem 1 can be obtained: for an attribute particle AG in a given attribute partial order structure, if particle AG is capable of splitting into new particles, then through the particle splitting operator gs, all newly generated particles gs(AG) must be located at the same layer as AG in the corresponding particle structure; at the same time, gs(AG) and AG have the same parent particle. Therefore, in the embodiment of the present invention, the new attribute particle AG in the second set AGR is i Each is added as a node to the three-branch partial order structure graph and the target attribute particle AG n In the corresponding level, and connect each new attribute particle AG i The corresponding target attribute particle AG n The parent node of .

[0138] S5. When the attribute particles of the target layer meet the particle reorganization conditions, a particle reorganization operation is performed on the attribute particles of the target layer to combine multiple attribute particles into a new attribute particle.

[0139] In the embodiment of the present invention, for two attribute particles AG1 and AG2 in a given attribute partial order structure, if they satisfy: a) the two particles are located in the same layer; b) the two particles each correspond to an object mode; c) the two particles have a common parent particle; then theoretically, particles AG1 and AG2 can be combined to form a new particle. Therefore, the particle recombination conditions of the embodiment of the present invention specifically include:

[0140] The corresponding nodes of each attribute particle AG in the target attribute particle pair have a common parent node;

[0141] The corresponding nodes of each attribute particle AG in the target attribute particle pair are all located at the same level;

[0142] The corresponding nodes of each attribute particle AG in the target attribute particle pair correspond to different object modes;

[0143] When the target attribute particle pair satisfies the above conditions at the same time, it is considered that the target attribute particle pair satisfies the particle recombination condition.

[0144] The specific process of particle recombination operation of the present invention is described below:

[0145] S5-1. Establish the fourth set AGL, the fifth set AGR and the sixth set AGS; the fourth set AGL takes each target layer attribute particle as an element, that is, AGL = {AG}; the fifth set and the sixth set are empty sets

[0146] S5-2. Arbitrarily combine each target layer attribute particle AG in the fourth set AGL to form multiple attribute particle pairs [AG m , A.G. n The embodiment of the present invention provides a particle reorganization description of an attribute particle pair composed of two target layer attribute particles AG. For a target attribute particle pair composed of more than two target layer attribute particles AG, the particle reorganization is performed with reference to the process of the attribute particle pair composed of two target layer attribute particles AG.

[0147] S5-3. Traverse multiple attribute particle pairs in sequence [AG m , A.G. n ], the first pair of attribute particles to be accessed is taken as the target attribute particle pair [AG1, AG2]. As a more specific embodiment, multiple target attribute particle pairs [AG m , A.G. n ] are sorted in descending order, and then each pair of target attribute particles is visited in turn according to the sorting.

[0148] S5-4. Determine whether the target attribute particle pair [AG1, AG2] meets the particle reorganization condition; if the target attribute particle pair [AG1, AG2] meets the particle reorganization condition, use the preset particle reorganization operator to perform a particle reorganization operation on the target attribute particle pair to combine into a new attribute particle AG i .

[0149] The particle recombination operator gc can be described as follows:

[0150] gc(AG1,AG2)={AG i |AG i .K=AG1.K∪AG2.K,AG i .A=AG1.A∪AG2.A,AG i .O=?,i=1}(Formula

[0151] 2).

[0152] As shown in Formula 2, the particle recombination operator gc satisfies the commutative law: gc(AG1,AG2)=gc(AG2,AG1); for each new particle AG generated by gc i , its core attribute set AG i .K and concept connotation AG i .A is the union of K and A of the two original particles; and its conceptual extension AG i .O is completely uncertain and needs to be filled in during the process of knowledge reasoning.

[0153] S5-5. Reorganize the particles to obtain the new attribute particles AG iAdd as an element to the sixth set AGS, that is, AGS = gc (AG1, AG2) = {AG i}.

[0154] S5-6. Sequentially traverse each new attribute particle AG in the sixth set AGS i , each new attribute particle AG in the sixth set AGS i Compare in the union of the fourth set and the fifth set AGL∪AGR; if there is a core attribute AG in the union of the fourth set and the fifth set AGL∪AGR i .K and the parent node and the new attribute particle AG in the sixth set i If the attribute particles AG are exactly the same, the new attribute particles AG i Deleted from the sixth set AGS.

[0155] S5-7. Add the elements in the sixth set AGS to the fifth set AGR, and clear the sixth set AGS.

[0156] S5-8. The next pair of attribute particles to be accessed [AG m+1 AG n+1 ] as the target attribute particle pair, and return to the step of judging whether the target attribute particle pair meets the particle recombination condition, until each pair of attribute particles in the fourth set [AG m AG n ] were visited.

[0157] S5-9. Add the new attribute particles in the fifth set to AG i Each of them is added as a node to the three-branch partial order structure graph with the target attribute particle [AG m AG n ] corresponding level, and connect each new attribute particle AG i The corresponding target attribute particle [AG m AG n ]'s parent node.

[0158] As a further explanation, according to the principle of granular computing, for the particle recombination operator, we can obtain Theorem 2:

[0159] For a target attribute particle pair [AG1, AG2] in a given attribute partial order structure, if they can be combined to form a new particle AG i , then through the particle reorganization operator gc, the newly generated particle gc(AG1,AG2) must be located in the same layer as the original particle [AG1,AG2] and have the same parent particle. i Each of them is added as a node to the three-branch partial order structure graph with the target attribute particle [AG m AGn ] corresponding level, and connect each new attribute particle AG i The corresponding target attribute particle [AG m AG n ]'s parent node.

[0160] S6. When the attribute granules of the target layer meet the particle isomorphism condition, a particle isomorphism operation is performed on the attribute granules of the target layer, multiple attribute granules are isomorphized into new attribute granules and a new object mode is generated.

[0161] In the embodiment of the present invention, for two attribute particles AG1 and AG2 in a given attribute partial order structure, if they satisfy: a) the two particles are located in the same layer; b) Theoretically, the attribute particles AG1 and AG2 can respectively expand their own sub-particles to form new particles based on the core attributes of the other sub-particles through isomorphic operations. Therefore, the particle isomorphism conditions of the embodiment of the present invention specifically include:

[0162] The corresponding nodes of each attribute particle in the target attribute particle pair are located at the same level;

[0163] The intersection of the core attributes of each attribute particle in the target attribute particle pair is a non-empty set;

[0164] When the target attribute particle pair satisfies the above conditions at the same time, it is considered that the target attribute particle pair satisfies the particle isomorphism condition.

[0165] The specific process of particle isomorphism operation of the present invention is described below:

[0166] S6-1. Establish the seventh set AGL, the eighth set AGS1 and the ninth set AGS2; in the seventh set, each target layer attribute particle is an element, that is, AGL = {AG}; the eighth set and the ninth set are empty sets

[0167] S6-2. Arbitrarily combine the attribute particles AG of each target layer in the seventh set AGL to form multiple attribute particle pairs [AG m , A.G. n The embodiment of the present invention provides a particle isomorphism description of an attribute particle pair composed of two target layer attribute particles AG. For a target attribute particle pair composed of more than two target layer attribute particles AG, particle isomorphism is performed with reference to the process of an attribute particle pair composed of two target layer attribute particles AG.

[0168] S6-3. Traverse multiple attribute particle pairs in sequence [AG m , A.G. n], the first pair of attribute particles to be accessed is taken as the target attribute particle pair; the target attribute particle pair includes the first target attribute particle and the second target attribute particle. As a more specific embodiment, multiple target attribute particle pairs [AG m , A.G. n ] are sorted in descending order, and then each pair of target attribute particles is visited in turn according to the sorting.

[0169] S6-4. Determine whether the target attribute particle pair satisfies the particle isomorphism condition; if the target attribute particle pair satisfies the particle recombination condition, perform the following steps on the first target attribute particle:

[0170] S6-4-a1. Use the preset particle recombination operator to perform particle isomorphism operation on the basis of the first target attribute particle using the second target attribute particle to obtain the first new attribute particle gi (AG1, AG2).

[0171] S6-4-a2. Add the first new attribute particle as an element to the eighth set, ie, AGS1 = gi (AG1, AG2) = {AG i}.

[0172] S6-4-a3. Determine whether the core attribute of each first new attribute particle in the eighth set is an empty set, that is, If so, add the first new attribute particle corresponding node to the next layer of the first target attribute particle corresponding level in the three-branch partial order structure diagram, and treat the first new attribute particle corresponding node as a child node of the first target attribute particle corresponding node; treat the first new attribute particle corresponding branch as a new object pattern OP=(AG i .O,AG i .A)Generate.

[0173] Perform the following steps for the second target attribute particle:

[0174] S6-4-b1. Use the preset particle reorganization operator to perform particle isomorphism operation on the basis of the second target attribute particle using the first target attribute particle to obtain the second new attribute particle gi (AG2, AG1);

[0175] S6-4-b2. Add the second new attribute particle as an element to the ninth set, ie AGS2 = gi (AG2, AG1) = {AG i}.

[0176] S6-4-b3. Determine whether the core attribute of each second new attribute particle in the ninth set is an empty set, that is If so, then add the second new attribute particle corresponding node to the next layer of the second target attribute particle corresponding level in the three-branch partial order structure diagram, and treat the second new attribute particle corresponding node as a child node of the second target attribute particle corresponding node; treat the second new attribute particle corresponding branch as a new object pattern OP=(AG i .O,AG i .A)Generate.

[0177] Specifically, assuming that {CG1} and {CG2} are child nodes of AG1 and AG2 respectively, the particle isomorphism operator gi can be described as follows:

[0178]

[0179] (Formula 3).

[0180] As shown in Formula 3, the particle isomorphism operator gi does not satisfy the commutative law: gi(AG1,AG2)≠gi(AG2,AG1); for each new attribute particle AG generated by gi(AG1,AG2) i , its core attribute set AG i .K must be the same as one of the attribute particles of the child node of AG2 and not appear in the original child node of AG1; at the same time, its conceptual connotation AG i .A is AG i .K is the union of AG2.A; and its conceptual extension AG i .O is completely uncertain and needs to be filled in during the process of knowledge reasoning. For each new attribute particle AG generated by gi(AG2,AG1) i , the situation is similar.

[0181] According to the principle of particle computing, for particle isomorphic operators, we can obtain Theorem 3:

[0182] For two attribute particles AG1 and AG2 in a given attribute partial order structure, if they can be isomorphically formed into new attribute particles, then through the particle isomorphism operator gi, the newly generated attribute particles gi(AG1,AG2) and gi(AG2,AG1) must be located in the next layer of the original particles AG1 and AG2 respectively, and are the sub-particles of AG1 and AG2 respectively.

[0183] In addition, for AG i ∈{gi(AG1,AG2)∪gi(AG2,AG1)}, if Then AG i is a possible object model and can be described as: OP = (AG i .O,AG i .A).

[0184] S6-5. Clear the eighth set AGS1 and the ninth set AGS2.

[0185] S6-6. The next pair of attribute particles to be visited [AG m+1 AG n+1 ] as the target attribute particle pair, and return to the step of judging whether the target attribute particle pair satisfies the particle isomorphism condition, until each pair of attribute particles [AG m AG n ] were visited.

[0186] S6-7. Summarize and output the new object pattern OP generated by the particle isomorphism operation.

[0187] It can be seen that the embodiment of the present invention can obtain new object patterns through particle operations and learn the new object patterns as new knowledge.

[0188] S7. Determine whether a new object pattern is generated in the particle isomorphism operation; when a new object pattern is generated, return to the step of performing a particle reorganization operation on the attribute particles of the target layer when the attribute particles of the target layer meet the particle reorganization conditions, and when the attribute particles of the target layer meet the particle reorganization conditions, perform a particle reorganization operation and a particle isomorphism operation on the attribute particles of the target layer until no new object pattern is generated;

[0189] When no new object pattern is generated, the next layer is added to the target layer, and the step of obtaining the attribute particles of the target layer is returned. When the attribute particles of the target layer meet the particle splitting conditions, the attribute particles of the target layer are subjected to particle splitting operations, particle recombination operations, and particle isomorphism operations until all levels in the three-branch partial order structure diagram are processed.

[0190] In an embodiment of the present invention, after completing the processing of the current target layer, the next level in the three-branch partial order structure diagram is added to the target layer, and attempts are made to learn new knowledge from nodes of different levels, different core attributes and different object modes through particle splitting operations, particle recombination operations and particle isomorphism operations, and small sample knowledge reasoning is performed layer by layer from top to bottom using granular computing thinking.

[0191] S8. Get all newly generated object patterns from the three-branch partial order structure graph and summarize the output.

[0192] "Treatise on Febrile Diseases" is a clinical work of traditional Chinese medicine written by Zhang Zhongjing in the Han Dynasty. It contains 113 traditional Chinese medicine prescriptions for treating various exogenous diseases and miscellaneous diseases. Among these 113 prescriptions, "Guizhi Decoction" and its similar prescriptions are very famous and widely used in clinical Chinese medicine. Therefore, as a specific example, Figure 4 , 5The prescriptions of "Guizhi Decoction" in the ancient book "Treatise on Febrile Diseases" are given based on the knowledge reasoning of the method of the embodiment of the present invention. There are about 21 prescriptions of Guizhi Decoction in "Treatise on Febrile Diseases". Figure 4 What is shown is a three-branch partial order structure diagram generated based on the Chinese medicinal composition of 9 prescriptions.

[0193] After the operations of particle splitting, particle recombination, particle isomorphism, etc. in the embodiment of the present invention, it can be inferred that Figure 5 Knowledge shown:

[0194] At the top level, the composite node "Licorice-Ginger" can generate two independent nodes "Licorice" and "Ginger" at the same level after particle splitting; and through particle isomorphism, the two independent nodes "Licorice" and "Ginger" can have similar sub-node links as the composite node; at the bottom level, the composite node "Poria-Atractylodes" can generate two independent nodes "Poria" and "Atractylodes" at the same level after particle splitting; and through particle recombination, the two independent nodes "Atractylodes" and "Aconite" can generate the composite node "Aconite-Atractylodes" at the same level;

[0195] From the top layer to the bottom layer, new knowledge can be formed through continuous iteration of particle splitting, combination, and isomorphism.

[0196] After granular computing knowledge reasoning based on partial order structure, Figure 5 It not only completes Figure 4 We can also infer a lot of other knowledge from the Guizhi Decoction-like prescriptions in Treatise on Febrile Diseases that are missing. Figure 5 It also presents some knowledge that can be verified in other ancient books (such as the Chinese medicine composition model of "Baoyuan Dacheng Decoction" in the book "Taiping Huimin Hejijufang").

[0197] The first embodiment of the present invention proposes a small sample knowledge reasoning method based on a three-branch partial order structure, which can realize data augmentation through deductive reasoning from the perspective of symbolic representation, and obtain new knowledge from the original data through particle splitting, combination and isomorphism operations. Therefore, compared with the existing vector-based small sample reasoning method, the present invention is more suitable for processing small sample problems with higher interpretability requirements, and is widely used in the small sample reasoning learning process that requires high interpretability, such as syndrome differentiation and treatment in the field of traditional Chinese medicine.

[0198] Figure 6It is a structural diagram of the electronic device proposed in the second embodiment of the present invention. The memory of this embodiment stores program instructions for implementing the small sample knowledge reasoning method based on the three-branch partial order structure of any of the above embodiments. The processor is used to execute the program instructions stored in the memory to perform small sample knowledge reasoning based on the three-branch partial order structure. Among them, the processor can also be called CPU (Central Processing Unit). The processor may be an integrated circuit chip with signal processing capabilities. The processor can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0199] The contents of the method in the first embodiment of the present invention are all applicable to the electronic device embodiment. The functions specifically implemented by the electronic device embodiment are the same as those of the above method embodiment, and the beneficial effects achieved are also the same as those achieved by the above method.

[0200] Figure 7 It is a schematic diagram of the structure of the computer-readable storage medium of the third embodiment of the present invention. The computer-readable storage medium of the fourth embodiment of the present invention stores program instructions that can implement the above-mentioned small sample knowledge reasoning method based on the three-branch partial order structure, wherein the program instructions can be stored in the above-mentioned storage medium in the form of a software product, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned computer-readable storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes, or terminal devices such as computers, servers, mobile phones, tablets, etc.

[0201] The contents of the method in the first embodiment of the present invention are all applicable to the computer-readable storage medium embodiment. The functions specifically implemented by the computer-readable storage medium embodiment are the same as those of the above method embodiment, and the beneficial effects achieved are also the same as those achieved by the above method.

[0202] This embodiment also provides a computer program product. When the computer program product runs on a computer, it enables the computer to execute the above-mentioned related steps to implement the small sample knowledge reasoning method based on the three-branch partial order structure provided by the above embodiment.

[0203] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of the present invention are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0204] Those skilled in the art will appreciate that the modules in the devices in the embodiments of the present invention may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments of the present invention may be combined into one module or unit or component, and in addition they may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including corresponding claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including corresponding claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0205] Through the description of the above implementation methods, technical personnel in the relevant field can understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0206] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection portion with one or more wirings (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0207] In addition, each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. In particular, for embodiments such as devices and equipment, since they are basically similar to method embodiments, the relevant parts can refer to the partial description of the method embodiments. The embodiments of the devices and equipment described above are merely schematic, wherein the modules, units, etc. described as separate components may or may not be physically separated, that is, they may be located in one place, or they may be distributed in multiple places, such as nodes of a system network. Specifically, some or all of the modules and units may be selected according to actual needs to achieve the purpose of the above-mentioned embodiment scheme. Those skilled in the art can understand and implement it without paying creative labor.

[0208] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0209] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0210] In addition, the terms "first", "second", etc. used in the embodiments of the present invention are only used for descriptive purposes and should not be understood as indicating or implying relative importance, or implicitly indicating the number of technical features indicated in the present embodiment. Therefore, the features defined by the terms "first", "second", etc. in the embodiments of the present invention can explicitly or implicitly indicate that the embodiment includes at least one of the features. In the description of the present invention, the word "multiple" means at least two or two or more, such as two, three, four, etc., unless otherwise clearly and specifically defined in the embodiments.

[0211] In the embodiments of the present invention, the terms "include", "comprises" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element. In addition, components, features, and elements with the same name in different embodiments of the present invention may have the same meaning or different meanings, and their specific meanings need to be determined by their explanation in the specific embodiment or further combined with the context in the specific embodiment.

[0212] Although embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and cannot be construed as limitations of the present invention, and those of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present invention. Those skilled in the art will readily come to think of other embodiments of the present invention after considering the specification and practicing the present invention. This application is intended to cover any variations, uses or adaptations of the present invention, which follow the general principles of the present invention and include common knowledge or customary technical means in the art that are not disclosed in the present invention. The specification and embodiments are intended to be exemplary only, and the true scope and spirit of the present invention are indicated by the following claims.

Claims

1. A small sample knowledge reasoning method based on a three-branch partial order structure, characterized in that: The following steps are involved: Acquire data to be analyzed, and generate a three-branch partial order structure graph according to the data to be analyzed; The first layer from top to bottom of the three-branch partial order structure graph is used as the target layer; Get the attribute particles of the target layer; When the attribute particles of the target layer meet the particle splitting conditions, the particle splitting operation is performed on the attribute particles of the target layer to split one attribute particle into multiple new attribute particles; When the attribute particles of the target layer meet the particle reorganization conditions, the particle reorganization operation is performed on the attribute particles of the target layer to combine multiple attribute particles into a new attribute particle; When the attribute granules of the target layer meet the particle isomorphism condition, the particle isomorphism operation is performed on the attribute granules of the target layer, multiple attribute granules are isomorphized into new attribute granules and a new object mode is generated; Determine whether a new object mode is generated in the particle isomorphism operation; when a new object mode is generated, return to the step of performing a particle reorganization operation on the attribute particles of the target layer when the attribute particles of the target layer meet the particle reorganization conditions, and when the attribute particles of the target layer meet the particle reorganization conditions, perform a particle reorganization operation and a particle isomorphism operation on the attribute particles of the target layer until no new object mode is generated; When no new object pattern is generated, the next layer is added to the target layer, and the step of obtaining the attribute particles of the target layer is returned. When the attribute particles of the target layer meet the particle splitting conditions, the attribute particles of the target layer are subjected to particle splitting operations, particle reorganization operations, and particle isomorphism operations until all levels in the three-branch partial order structure diagram are processed; All newly generated object patterns are obtained from the three partial order structure graphs and summarized and output.

2. According to claim 1, a small sample knowledge reasoning method based on a three-branch partial order structure is characterized in that: The data to be analyzed is represented by triple data, specifically including an object set, an attribute set and an object-attribute correspondence relationship; the object set has multiple object patterns, the attribute set has multiple attribute elements; the object-attribute correspondence relationship refers to the correspondence relationship between the object pattern and the attribute element; Generating a three-branch partial order structure diagram according to the data to be analyzed specifically includes the following steps: Expand the attribute element with the most object-attribute correspondences as the root node, and generate a tree diagram with each object pattern corresponding to the root node as a branch; On each branch, the attribute elements corresponding to the object pattern are arranged as nodes from top to bottom in a preset attribute element order; Each group of nodes extending downward from the root node is considered as a level; Check the attribute elements of the nodes in each level. If the same attribute element appears in nodes of multiple levels, merge the node into the node of the level where the attribute element first appears on the corresponding branch, so that after the merger, the node of the branch at this level has multiple attribute elements. Merge the nodes with the same extension path in different branches expanded by the root node; In the three-branch partial order structure diagram, each node forms an attribute granule; the attribute granule is represented by triple data, which are the core attribute of the node corresponding to the attribute granule, the attribute element set of the line from the node corresponding to the attribute granule to the root node, and the object mode set corresponding to the node corresponding to the attribute granule; the core attribute refers to one or more attribute elements possessed by the node corresponding to the attribute granule.

3. According to claim 1, a small sample knowledge reasoning method based on a three-branch partial order structure is characterized in that: Before the step of taking the first layer as the target layer, the following steps are also included: Determine whether the root node of the three-branch partial order structure graph can perform a particle splitting operation; When the root node of the three-branch partial order structure graph can be subjected to a particle splitting operation, the particle splitting operation is performed on the root node of the three-branch partial order structure graph, and the new attribute particle formed by the splitting is used as a new root node, and the original root node is used as a child node of the new root node.

4. According to claim 1, a small sample knowledge reasoning method based on a three-branch partial order structure is characterized in that: The particle splitting operation is implemented by a preset particle splitting operator, and specifically includes the following steps: Establish a first set, a second set and a third set; each target layer attribute particle is an element in the first set, and the second set and the third set are empty sets; Traverse each attribute grain in the first set in turn, and take the first attribute grain visited as the target attribute grain; Determine the core attribute of the target attribute particle; If the target attribute particle has more than one core attribute, a preset particle splitting operator is used to perform a particle splitting operation on the target attribute particle to generate multiple new attribute particles; Adding the new attribute particle obtained by the particle splitting operation as an element to the third set; Traversing each new attribute particle in the third set in turn, and comparing each new attribute particle in the third set with the union of the first set and the second set; If there is an attribute grain whose core attribute and parent node are exactly the same as the new attribute grain in the third set in the union of the first set and the second set, then the attribute element set of the new attribute grain is added to the attribute element set of the target attribute grain, and the new attribute grain is deleted from the third set; Add the elements in the third set to the second set, and clear the third set; The next attribute granule to be visited is taken as the target attribute granule, and the process returns to the step of determining the core attribute of the target attribute granule until each attribute granule in the first set is visited; Each of the new attribute particles in the second set is taken as a node, added to the level corresponding to the target attribute particle in the three-branch partial order structure graph, and connected to the parent node of the target attribute particle corresponding to each new attribute particle.

5. According to claim 1, a small sample knowledge reasoning method based on a three-branch partial order structure is characterized in that: The particle reorganization operation is implemented by a preset particle reorganization operator, and specifically includes the following steps: Establishing a fourth set, a fifth set and a sixth set; in the fourth set, each target layer attribute particle is an element, and the fifth set and the sixth set are empty sets; Arbitrarily combining each target layer attribute particle in the fourth set to form a plurality of attribute particle pairs; Traversing the plurality of attribute particle pairs in sequence, and taking the first pair of attribute particles visited as the target attribute particle pair; Determine whether the target attribute particle pair meets the particle recombination condition; If the target attribute particle pair meets the particle recombination condition, a preset particle recombination operator is used to perform a particle recombination operation on the target attribute particle pair to combine them into a new attribute particle; Adding new attribute particles obtained by the particle reorganization operation as elements to the sixth set; Traversing each new attribute particle in the sixth set in turn, and comparing each new attribute particle in the sixth set with the union of the fourth set and the fifth set; If there is an attribute grain whose core attribute and parent node are exactly the same as the new attribute grain in the sixth set in the union of the fourth set and the fifth set, then the new attribute grain is deleted from the sixth set; Add the elements in the sixth set to the fifth set, and clear the sixth set; The next pair of attribute grains to be visited is taken as the target attribute grain pair, and the process returns to the step of determining whether the target attribute grain pair satisfies the grain reorganization condition, until every pair of attribute grains in the fourth set is visited; Each of the new attribute particles in the fifth set is taken as a node, added to the level corresponding to the target attribute particle in the three-branch partial order structure graph, and connected to the parent node of the target attribute particle corresponding to each new attribute particle.

6. According to claim 5, a small sample knowledge reasoning method based on a three-branch partial order structure is characterized in that: The particle recombination conditions specifically include: The corresponding nodes of each attribute particle in the target attribute particle pair have a common parent node; The corresponding nodes of each attribute particle in the target attribute particle pair are located at the same level; The corresponding nodes of each attribute particle in the target attribute particle pair correspond to different object modes; When the target attribute particle pair satisfies the above conditions at the same time, it is considered that the target attribute particle pair satisfies the particle recombination condition.

7. According to claim 1, a small sample knowledge reasoning method based on a three-branch partial order structure is characterized in that: The particle isomorphism operation is implemented by a preset particle isomorphism operator, and specifically includes the following steps: Establishing a seventh set, an eighth set and a ninth set; in the seventh set, each target layer attribute particle is an element, and the eighth set and the ninth set are empty sets; Arbitrarily combining each target layer attribute particle in the seventh set to form a plurality of attribute particle pairs; Traversing the plurality of attribute grain pairs in sequence, taking the first pair of attribute grains visited as the target attribute grain pair; the target attribute grain pair includes a first target attribute grain and a second target attribute grain; Determining whether the target attribute particle pair satisfies a particle isomorphism condition; If the target attribute particle pair meets the particle recombination condition, the following steps are performed on the first target attribute particle: a particle isomorphism operation is performed on the first target attribute particle using the second target attribute particle based on the first target attribute particle using a preset particle recombination operator to obtain a first new attribute particle; the first new attribute particle is added as an element to the eighth set; it is determined whether the core attribute of each first new attribute particle in the eighth set is an empty set; if so, a node corresponding to the first new attribute particle is added to the next layer of the level corresponding to the first target attribute particle in the three-branch partial order structure diagram, and the node corresponding to the first new attribute particle is treated as a child node of the node corresponding to the first target attribute particle; the branch corresponding to the first new attribute particle is generated as a new object pattern; The following steps are performed on the second target attribute particle: a preset particle recombination operator is used to perform a particle isomorphism operation on the first target attribute particle based on the second target attribute particle to obtain a second new attribute particle; the second new attribute particle is added as an element to the ninth set; it is determined whether the core attribute of each second new attribute particle in the ninth set is an empty set; if so, a node corresponding to the second new attribute particle is added to the next layer of the level corresponding to the second target attribute particle in the three-branch partial order structure diagram, and the node corresponding to the second new attribute particle is treated as a child node of the node corresponding to the second target attribute particle; the branch corresponding to the second new attribute particle is generated as a new object pattern; Clearing the eighth set and the ninth set; The next pair of attribute particles to be visited is taken as the target attribute particle pair, and the process returns to the step of determining whether the target attribute particle pair satisfies the particle isomorphism condition, until every pair of attribute particles in the seventh set is visited; The new object patterns generated by the particle isomorphism operation are summarized and output.

8. The small sample knowledge reasoning method based on a three-branch partial order structure according to claim 7 is characterized in that: The particle isomorphism conditions specifically include: The corresponding nodes of each attribute particle in the target attribute particle pair are located at the same level; The intersection of the core attributes of each attribute particle in the target attribute particle pair is a non-empty set; When the target attribute particle pair satisfies the above conditions at the same time, it is considered that the target attribute particle pair satisfies the particle isomorphism condition.

9. An electronic device, characterized in that: including a processor and a memory; The memory is used to store programs; The processor executes the program to implement a small sample knowledge reasoning method based on a three-branch partial order structure as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: The storage medium stores a program, and the program is executed by a processor to implement a small sample knowledge reasoning method based on a three-branch partial order structure as described in any one of claims 1-8.

Images