CN112785350B - Product vector determining method and device - Google Patents
Product vector determining method and device Download PDFInfo
- Publication number
- CN112785350B CN112785350B CN202110206095.2A CN202110206095A CN112785350B CN 112785350 B CN112785350 B CN 112785350B CN 202110206095 A CN202110206095 A CN 202110206095A CN 112785350 B CN112785350 B CN 112785350B
- Authority
- CN
- China
- Prior art keywords
- product
- node
- vector
- target
- preset
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 239000013598 vector Substances 0.000 title claims abstract description 270
- 238000000034 method Methods 0.000 title claims abstract description 45
- 230000002776 aggregation Effects 0.000 claims abstract description 91
- 238000004220 aggregation Methods 0.000 claims abstract description 91
- 239000011159 matrix material Substances 0.000 claims abstract description 77
- 238000011176 pooling Methods 0.000 claims abstract description 62
- 239000013604 expression vector Substances 0.000 claims description 40
- 238000005070 sampling Methods 0.000 claims description 9
- 230000004931 aggregating effect Effects 0.000 claims description 6
- 238000009826 distribution Methods 0.000 claims description 4
- 238000012935 Averaging Methods 0.000 claims 1
- 230000006870 function Effects 0.000 description 40
- 238000003860 storage Methods 0.000 description 9
- 201000010099 disease Diseases 0.000 description 8
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 8
- 230000009466 transformation Effects 0.000 description 8
- 238000013528 artificial neural network Methods 0.000 description 6
- 238000004590 computer program Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 238000013519 translation Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
- G06F16/355—Class or cluster creation or modification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0203—Market surveys; Market polls
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- Entrepreneurship & Innovation (AREA)
- General Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Mathematical Physics (AREA)
- Computational Mathematics (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Economics (AREA)
- Marketing (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Animal Behavior & Ethology (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computing Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Probability & Statistics with Applications (AREA)
- Evolutionary Biology (AREA)
- Algebra (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method and a device for determining a product vector, which are used for determining a first initial vector of each product node and a second initial vector of each attribute node in a preset product knowledge graph. And determining a first representation vector of each product node at least according to the first initial vector of each product node, the second initial vector of the target attribute node which has a first target relation with each product node respectively and the preset attribute average value aggregation weight matrix. And determining a second representation vector of each product node at least according to the first representation vector of each product node, the first representation vector of the target product node which has a second target relation with each product node respectively and the preset product pooling aggregation parameter. According to the invention, on the basis of the initial vector of the product node and the initial vector of the attribute node in the product knowledge graph, the product node can be effectively and accurately vectorized and expressed by combining the preset attribute mean value aggregation weight matrix and the preset product pooling aggregation parameter.
Description
Technical Field
The present invention relates to the field of data processing technologies, and in particular, to a method and an apparatus for determining a product vector.
Background
The Knowledge Graph (knowledgegraph) is a series of different graphs showing the Knowledge development process and the structural relationship, and the Knowledge resource and the carrier thereof are described by using the visualization technology, and Knowledge and the interrelationship between the Knowledge resource and the carrier thereof are mined, analyzed, constructed, drawn and displayed.
Along with the application of the knowledge graph in each field, the knowledge graph constructed according to the products in each field is more and more frequently used in the aspects of product recommendation, product search, product data analysis and the like. Therefore, in order to make efficient and accurate use of various product data in the product knowledge graph, it is necessary to vectorize the expression of the nodes in the component product knowledge graph.
Disclosure of Invention
In view of the above problems, the present invention provides a method and an apparatus for determining a product vector, which overcome or at least partially solve the above problems, and the technical solution is as follows:
a method of product vector determination, comprising:
determining a first initial vector of each product node and a second initial vector of each attribute node in a preset product knowledge graph, wherein the preset product knowledge graph is formed by connecting at least one product node and at least one attribute node;
Determining a first representation vector of each product node at least according to the first initial vector of each product node, the second initial vector of a target attribute node having a first target relationship with each product node and a preset attribute mean value aggregation weight matrix;
and determining a second representation vector of each product node at least according to the first representation vector of each product node, the first representation vector of a target product node which has a second target relationship with each product node respectively and a preset product pooling aggregation parameter.
Optionally, the determining a first initial vector of each product node and a second initial vector of each attribute node in the preset product knowledge graph includes:
vectorizing the product name of each product node by using a first preset bag-of-word model to obtain a first initial vector of each product node;
and/or vectorizing the product name of each attribute node by using a second preset bag-of-word model to obtain a second initial vector of each attribute node.
Optionally, the determining the first expression vector of each product node at least according to the first initial vector of each product node, the second initial vector of the target attribute node having the first target relationship with each product node, and a preset attribute average value aggregation weight matrix includes:
Determining a first aggregate vector for each of the product nodes based at least on the first initial vector for each of the product nodes and the second initial vector for a target attribute node having a first target relationship with each of the product nodes, respectively;
and respectively inputting the first aggregate vector of each product node and a preset attribute mean value aggregate weight matrix into a first preset formula, and calculating to obtain a first expression vector of each product node.
Optionally, the determining the first aggregate vector of each product node at least according to the first initial vector of each product node and the second initial vector of the target attribute node having the first target relationship with each product node respectively includes:
determining a first number of the product nodes in the preset product knowledge graph;
respectively determining a second number of product nodes connected with each target attribute node;
determining the inverse degree value corresponding to each target attribute node according to the first quantity and the second quantity;
for any one of the product nodes: inputting the first initial vector of the product node, the second initial vector of the target attribute node with a first target relation with the product node and the inverse output value corresponding to the target attribute node into a second preset formula, and calculating to obtain a first aggregate vector of the product node.
Optionally, the preset product pooling aggregation parameter includes a preset product pooling aggregation matrix and a preset offset, and the determining the second expression vector of each product node at least according to the first expression vector of each product node, the first expression vector of a target product node having a second target relationship with each product node, and the preset product pooling aggregation parameter includes:
inputting the first expression vector of each product node, the first expression vector of a target product node which has a second target relation with each product node, the preset product pooling aggregation weight matrix and the preset offset into a third preset formula respectively, and calculating to obtain a second aggregation vector of each product node;
and respectively inputting the second aggregate vector of each product node and the preset product pooling aggregate weight matrix into a fourth preset formula, and calculating to obtain a second expression vector of each product node.
Optionally, the first target relationship is a one-hop neighbor, and/or the second target relationship is a two-hop neighbor.
Optionally, the second preset formula is:
Wherein AGGREGATEmean is the first aggregate vector, MEAN is matrix average function, CONCATE is matrix splicing function, v is the product node, i is the number of the target attribute node, u i For the target attribute node numbered i having a first target relationship with the product node,for said first initial vector of the product node v,/i>The target attribute node u with the number of i i Is not less than the second primary stage of (1)Initial vector, V p For said first quantity,/a>The target attribute node u with the number of i i A second number of connected product nodes, +.>The target attribute node u with the number of i i The corresponding inverse degree value, N (v), is a first connection node function of the target attribute node having a first target relationship with the product node v;
the first preset formula is:
wherein ,for the first representation vector of product node v, RELU is a nonlinear function, W mean And aggregating a weight matrix for the preset attribute mean value.
Optionally, the third preset formula is:
therein, AGGREGATE pool For the second aggregate vector, CONCATE is a matrix splicing function, v is the product node,for the first representation vector of product node v, max is the max pooling operation, RELU is a nonlinear function, W pool Pooling the aggregation weight matrix for the preset product, j being the number of the target product node, w j For said target product node numbered j having said second target relationship with said product node, ->The target product node w with the number j j B is the preset offset, and M (v) is the target product node w having a second target relationship with the product node v j Is a second connection node function of (a);
the fourth preset formula is:
wherein ,is a second representation vector of the product node v.
Optionally, the method further comprises:
determining the preset attribute mean value aggregation weight matrix through a preset loss function, and/or pooling aggregation parameters of the preset products, wherein the preset loss function is as follows:
wherein ,Jg For the representation of the preset loss function, v is the product node in the preset product knowledge graph, and z v For a third representation vector of the product node v, σ is a sigmoid function, T is a transposed matrix symbol, u is the target product node having the second target relationship with the product node v, z u A third representation vector for the target product node u n For the negative sampled product node for which the second target relationship does not exist with the product node v,for the negative sampling product node u n Q is the absence of said product node vThe negative sampling product node u of the second target relationship n Is the number of (E) is the desired, P n (u) is the negative sampling product node u having no second target relationship with the product node v n Is a distribution of (a).
A product vector determination apparatus comprising: an initial vector determination unit, a first expression vector determination unit, and a second expression vector determination unit,
the initial vector determining unit is used for determining a first initial vector of each product node and a second initial vector of each attribute node in a preset product knowledge graph, wherein the preset product knowledge graph is formed by connecting at least one product node and at least one attribute node;
the first representation vector determining unit is configured to determine a first representation vector of each product node according to at least the first initial vector of each product node, the second initial vector of a target attribute node having a first target relationship with each product node, and a preset attribute average value aggregation weight matrix;
The second representation vector determining unit is configured to determine a second representation vector of each product node according to at least the first representation vector of each product node, the first representation vector of a target product node having a second target relationship with each product node, and a preset product pooling aggregation parameter.
By means of the technical scheme, the method and the device for determining the product vector can determine the first initial vector of each product node and the second initial vector of each attribute node in the preset product knowledge graph, wherein the preset product knowledge graph is formed by connecting at least one product node and at least one attribute node; determining a first expression vector of each product node at least according to the first initial vector of each product node, the second initial vector of the target attribute node which has a first target relation with each product node respectively and a preset attribute average value aggregation weight matrix; and determining a second representation vector of each product node at least according to the first representation vector of each product node, the first representation vector of the target product node which has a second target relation with each product node respectively and the preset product pooling aggregation parameter. According to the embodiment of the invention, on the basis of the initial vector of the product node and the initial vector of the attribute node in the product knowledge graph, the product node can be effectively and accurately vectorized by combining the preset attribute mean value aggregation weight matrix and the preset product pooling aggregation parameter.
The foregoing description is only an overview of the present invention, and is intended to be implemented in accordance with the teachings of the present invention in order that the same may be more clearly understood and to make the same and other objects, features and advantages of the present invention more readily apparent.
Drawings
Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to designate like parts throughout the figures. In the drawings:
fig. 1 is a schematic flow chart of a method for determining a product vector according to an embodiment of the present invention;
FIG. 2 is a flow chart of another method for determining a product vector according to an embodiment of the present invention;
FIG. 3 is a flow chart illustrating another method for determining a product vector according to an embodiment of the present invention;
FIG. 4 is a flow chart illustrating another method for determining a product vector according to an embodiment of the present invention;
fig. 5 shows a schematic structural diagram of a product vector determining device according to an embodiment of the present invention;
Fig. 6 shows a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
As shown in fig. 1, a method for determining a product vector according to an embodiment of the present invention includes:
s100, determining a first initial vector of each product node and a second initial vector of each attribute node in a preset product knowledge graph.
The preset product knowledge graph is formed by connecting at least one product node and at least one attribute node.
It will be appreciated that a Knowledge Graph (knowledgegraph) is a Graph-based data structure, consisting of nodes (points) and edges (edges). In the knowledge graph, each node represents an ontology, and each edge represents a relationship between two connected ontologies. Alternatively, in embodiments of the present invention, the product may be embodied as an insurance product. The product nodes may correspond to insurance product bodies. The attribute nodes may correspond to insurance attribute ontologies. For example: the insurance attribute ontology corresponding to the attribute node may be any ontology of critical diseases, moderate diseases, mild diseases, specific diseases, exemption terms, feature labels, insurance ages, interval periods, guarantee periods, and insurance companies.
Optionally, in the preset product knowledge graph, any two product nodes are connected through at least one attribute node.
Optionally, the embodiment of the present invention may use a first preset bag-of-word model to vectorize the product name of each product node, so as to obtain a first initial vector of each product node.
Optionally, the embodiment of the present invention may use a second preset bag-of-word model to vectorize the product name of each attribute node, so as to obtain a second initial vector of each attribute node.
Alternatively, the first preset bag-of-words model and the second preset bag-of-words model may be the same.
Optionally, in the case that a plurality of different names exist in a node, the embodiment of the invention can splice the plurality of names, and then vectorize the spliced names by using a word bag model to obtain an initial vector of the node.
S200, determining a first expression vector of each product node at least according to the first initial vector of each product node, the second initial vector of each target attribute node having a first target relation with each product node and a preset attribute mean value aggregation weight matrix.
Optionally, the first target relationship is a one-hop neighbor. For example: other nodes directly connected to any node through an edge are one-hop neighbors to that node. The embodiment of the invention can take the attribute node directly connected with any product node as the target attribute node of the product node.
Optionally, the embodiment of the present invention may be applied to any product node: and aggregating a second initial vector of the target attribute node with a first target relation with the product node, splicing the vector obtained after aggregation with the first initial vector of the product node, performing matrix average value calculation on the spliced vector through a preset attribute average value aggregation weight matrix, and obtaining a first representation vector of the product node through nonlinear transformation on the calculated vector.
It should be noted that the target attribute nodes of different product nodes may be the same or different. The number of target attribute nodes of different product nodes may be the same or different.
Optionally, based on the method shown in fig. 1, as shown in fig. 2, another method for determining a product vector according to an embodiment of the present invention, step S200 may include:
S210, determining a first aggregate vector of each product node at least according to the first initial vector of each product node and the second initial vector of a target attribute node which has a first target relation with each product node respectively.
Because the important programs of the target attribute nodes on the product nodes are different, the embodiment of the invention can determine the inverse output value of the target attribute nodes according to the number of the product nodes in the preset product knowledge graph and the number of the product nodes connected with the target attribute nodes, and further calculate the first aggregate vector of the product nodes according to the inverse output value.
Specifically, based on the method shown in fig. 2, as shown in fig. 3, another method for determining a product vector according to an embodiment of the present invention, step S210 may include:
s211, determining a first number of product nodes in the preset product knowledge graph.
It is understood that the first number is the total number of product nodes included in the preset product knowledge graph.
S212, respectively determining the second number of the product nodes connected with each target attribute node.
For ease of understanding, this is illustrated by way of example: and a target attribute node a with a first target relation with the product node A, wherein the one-hop neighbor connected with the target attribute node a comprises the product node A, the product node B and the product node C, and the second number of the product nodes connected with the target attribute node is 3.
S213, determining the inverse degree value corresponding to each target attribute node according to the first quantity and the second quantity.
Alternatively, the inverse value corresponding to the target attribute node may be a value obtained by dividing the first number by the second number.
S214, for any product node: inputting the first initial vector of the product node, the second initial vector of the target attribute node with a first target relation with the product node and the inverse output value corresponding to the target attribute node into a second preset formula, and calculating to obtain a first aggregate vector of the product node.
Optionally, the second preset formula is:
wherein AGGREGATEmean is the first aggregate vector, MEAN is matrix average function, CONCATE is matrix splicing function, v is the product node, i is the number of the target attribute node, u i For the target attribute node numbered i having a first target relationship with the product node,for said first initial vector of the product node v,/i>The target attribute node u with the number of i i V of the first initial vector of (a) p For said first quantity,/a>The target attribute node u with the number of i i A second number of connected product nodes, +.>The target attribute node u with the number of i i And the corresponding inverse degree value N (v) is a first connection node function of the target attribute node with a first target relation with the product node v.
S220, respectively inputting the first aggregate vector of each product node and a preset attribute mean value aggregate weight matrix into a first preset formula, and calculating to obtain a first expression vector of each product node.
Optionally, the first preset formula is:
wherein ,for the first representation vector of product node v, RELU is a nonlinear function, W mean And aggregating a weight matrix for a preset attribute mean value.
S300, determining a second expression vector of each product node at least according to the first expression vector of each product node, the first expression vector of a target product node which has a second target relation with each product node respectively and a preset product pooling aggregation parameter.
Optionally, the second target relationship is a two-hop neighbor. According to the embodiment of the invention, other product nodes indirectly connected with any product node through one attribute node can be used as target product nodes of the product node.
Optionally, the embodiment of the present invention may be applied to any product node: and carrying out nonlinear transformation on a first expression vector of a target product node with a second target relation with the product node, carrying out maximum Pooling (Max-Pooling) operation on the vector after nonlinear transformation through a preset product Pooling aggregation parameter to obtain a characteristic maximum value of the target product node, splicing the characteristic maximum value of the target product node with the first expression vector of the target product node, and carrying out nonlinear transformation on the spliced vector to obtain the second expression vector of the target product node.
Optionally, based on the method shown in fig. 1, as shown in fig. 4, in another method for determining a product vector according to an embodiment of the present invention, the preset product pooling aggregation parameter includes a preset product pooling aggregation matrix and a preset offset, and step S300 may include:
s310, respectively inputting the first expression vector of each product node, the first expression vector of a target product node which has a second target relation with each product node, the preset product pooling aggregation weight matrix and the preset offset into a third preset formula, and calculating to obtain a second aggregation vector of each product node.
The bias quantity (bias) is also called a neural network bias term, and plays a role in increasing the translation capacity of the neural network and enhancing the generalization effect of the neural network.
Optionally, the third preset formula is:
therein, AGGREGATE pool For the second aggregate vector, CONCATE is a matrix splicing function, v is the product node,for the first representation vector of product node v, max is the max pooling operation, RELU is a nonlinear function, W pool Pooling the aggregation weight matrix for the preset product, j being the number of the target product node, w j For the target product node numbered j having the second target relationship with the product node v, +.>The target product node w with the number j j B is the preset offset, and M (v) is the target product node w having a second target relationship with the product node v j Is provided for the second connection node function of (a).
S320, respectively inputting the second aggregate vector of each product node and the preset product pooling aggregate weight matrix into a fourth preset formula, and calculating to obtain a second expression vector of each product node.
Optionally, the fourth preset formula is:
wherein ,is a second representation vector of the product node v.
The method for determining the product vector can determine the first initial vector of each product node and the second initial vector of each attribute node in the preset product knowledge graph, wherein the preset product knowledge graph is formed by connecting at least one product node and at least one attribute node; determining a first expression vector of each product node at least according to the first initial vector of each product node, the second initial vector of the target attribute node which has a first target relation with each product node respectively and a preset attribute average value aggregation weight matrix; and determining a second representation vector of each product node at least according to the first representation vector of each product node, the first representation vector of the target product node which has a second target relation with each product node respectively and the preset product pooling aggregation parameter. According to the embodiment of the invention, on the basis of the initial vector of the product node and the initial vector of the attribute node in the product knowledge graph, the product node can be effectively and accurately vectorized by combining the preset attribute mean value aggregation weight matrix and the preset product pooling aggregation parameter.
Optionally, the embodiment of the invention can be applied to a preset neural network model. The neural network model includes an attribute mean aggregation layer and a product pooling aggregation layer. The attribute mean aggregation layer is a network structure preconfigured with at least the first configuration parameters to perform step S200, and the product pooling aggregation layer is a network structure preconfigured with at least the second configuration parameters to perform step S300. The first configuration parameter comprises a preset attribute mean aggregation weight matrix. The second configuration parameters include preset product pooling aggregation parameters.
Optionally, the embodiment of the present invention may determine the preset attribute mean aggregation weight matrix through a preset loss function, and/or the preset product pooling aggregation parameter.
Wherein, the preset loss function is:
wherein ,Jg For the representation of the preset loss function, v is the yield in the preset product knowledge graphProduct node, z v For a third representation vector of the product node v, σ is a sigmoid function, T is a transposed matrix symbol, u is the target product node having the second target relationship with the product node v, z u A third representation vector for the target product node u n Z for the negatively sampled product node for which the second target relationship does not exist with the product node v un For the negative sampling product node u n Q is the negative sampled product node u having no second target relationship with the product node v n Is the number of (E) is the desired, P n (u) is the negative sampling product node u having no second target relationship with the product node v n Is a distribution of (a).
Optionally, the embodiment of the invention can aim at minimizing the loss function, and simultaneously, perform iterative training on the initialized attribute mean value aggregation weight matrix and the initialized product pooling aggregation parameter, so as to determine the preset attribute mean value aggregation weight matrix and the preset product pooling aggregation parameter. The initialization product pooling aggregation weight matrix in the initialization attribute mean aggregation weight matrix and the initialization product pooling aggregation parameter is an initial random matrix.
It can be understood that, in the embodiment of the present invention, the initialization attribute mean aggregation weight matrix and the initialization product pooling aggregation parameter may be iteratively trained by referring to the above steps and formulas, and the third expression vector of the product node is obtained by forward propagation. Wherein the third representative vector of the product node may be a final output representative vector under the conditions of the attribute mean aggregation weight matrix and the product pooling aggregation parameter used in each iteration. When the value of the loss function is minimal, the second representation vector of the product node is the same as the third representation vector.
According to the embodiment of the invention, through the preset loss function in the Adam optimizer, the attribute mean value aggregation weight matrix and the product pooling aggregation parameter are optimized by back propagation in iterative training, so that the attribute mean value aggregation weight matrix and the product pooling aggregation parameter are determined when the preset loss function is minimum.
According to the embodiment of the invention, the preset attribute average value aggregation weight matrix and the preset product pooling aggregation parameter are determined through the preset loss function, so that the representation vector of the product node determined according to the preset attribute average value aggregation weight matrix and the preset product pooling aggregation parameter can more accurately express the characteristics of the product body corresponding to the product node, and a guarantee is provided for the effective use of the product data in the follow-up process.
Corresponding to the above method embodiment, the embodiment of the present invention further provides a product vector determining device, where the structure of the product vector determining device is shown in fig. 5, and the product vector determining device may include: an initial vector determination unit 100, a first representation vector determination unit 200, and a second representation vector determination unit 300.
The initial vector determining unit 100 is configured to determine a first initial vector of each product node and a second initial vector of each attribute node in a preset product knowledge graph.
The preset product knowledge graph is formed by connecting at least one product node and at least one attribute node.
It will be appreciated that a Knowledge Graph (knowledgegraph) is a Graph-based data structure, consisting of nodes (points) and edges (edges). In the knowledge graph, each node represents an ontology, and each edge represents a relationship between two connected ontologies. Alternatively, in embodiments of the present invention, the product may be embodied as an insurance product. The product nodes may correspond to insurance product bodies. The attribute nodes may correspond to insurance attribute ontologies. For example: the insurance attribute ontology corresponding to the attribute node may be any ontology of critical diseases, moderate diseases, mild diseases, specific diseases, exemption terms, feature labels, insurance ages, interval periods, guarantee periods, and insurance companies.
Optionally, in the preset product knowledge graph, any two product nodes are connected through at least one attribute node.
Alternatively, the initial vector determination unit 100 may include: the first initial vector determination subunit, and/or the second initial vector determination subunit.
The first initial vector determining subunit is configured to use a first preset bag-of-word model to vectorize a product name of each product node, so as to obtain a first initial vector of each product node.
And the second initial vector determining subunit is configured to use a second preset bag-of-word model to vectorize the product name of each attribute node, so as to obtain a second initial vector of each attribute node.
Alternatively, the first preset bag-of-words model and the second preset bag-of-words model may be the same.
Alternatively, in the case that a plurality of different names exist in a node, the initial vector determining unit 100 may splice the plurality of names, and then vectorize the spliced names by using the bag-of-word model to obtain the initial vector of the node.
The first representation vector determining unit 200 is configured to determine a first representation vector of each product node according to at least the first initial vector of each product node, the second initial vector of a target attribute node having a first target relationship with each product node, and a preset attribute mean aggregation weight matrix.
Optionally, the first target relationship is a one-hop neighbor.
Alternatively, the first representation vector determination unit 200 may, for any product node: and aggregating a second initial vector of the target attribute node with a first target relation with the product node, splicing the vector obtained after aggregation with the first initial vector of the product node, performing matrix average value calculation on the spliced vector through a preset attribute average value aggregation weight matrix, and obtaining a first representation vector of the product node through nonlinear transformation on the calculated vector.
It should be noted that the target attribute nodes of different product nodes may be the same or different. The number of target attribute nodes of different product nodes may be the same or different.
Alternatively, the first representation vector determination unit 200 may include: a first aggregate vector determination subunit and a first representation vector determination subunit.
The first aggregate vector determining subunit is configured to determine a first aggregate vector of each product node according to at least the first initial vector of each product node and the second initial vector of a target attribute node that has a first target relationship with each product node.
Optionally, the first aggregate vector determining subunit may be specifically configured to determine the first number of product nodes in the preset product knowledge graph. And respectively determining a second number of the product nodes connected with each target attribute node. And determining the inverse degree value corresponding to each target attribute node according to the first quantity and the second quantity. For any one of the product nodes: inputting the first initial vector of the product node, the second initial vector of the target attribute node with a first target relation with the product node and the inverse output value corresponding to the target attribute node into a second preset formula, and calculating to obtain a first aggregate vector of the product node.
Optionally, the second preset formula is:
wherein AGGREGATEmean is the first aggregate vector, MEAN is matrix average function, CONCATE is matrix splicing function, v is the product node, i is the number of the target attribute node, u i For the target attribute node numbered i having a first target relationship with the product node,for said first initial vector of the product node v,/i>The target attribute node u with the number of i i V of the first initial vector of (a) p For said first quantity,/a>The target attribute node u with the number of i i A second number of connected product nodes, +.>The target attribute node u with the number of i i And the corresponding inverse degree value N (v) is a first connection node function of the target attribute node with a first target relation with the product node v.
The first expression vector determining subunit is configured to input the first aggregate vector of each product node and a preset attribute average aggregate weight matrix into a first preset formula, and calculate to obtain a first expression vector of each product node.
Optionally, the first preset formula is:
wherein ,for the first representation vector of product node v, RELU is a nonlinear function, W mean And aggregating a weight matrix for the preset attribute mean value.
The second representation vector determining unit 300 is configured to determine a second representation vector of each product node according to at least the first representation vector of each product node, the first representation vector of a target product node having a second target relationship with each product node, and a preset product pooling aggregation parameter.
Optionally, the second target relationship is a two-hop neighbor.
Alternatively, the second representation vector determination unit 300 may, for any product node: and carrying out nonlinear transformation on a first expression vector of a target product node with a second target relation with the product node, carrying out maximum Pooling (Max-Pooling) operation on the vector after nonlinear transformation through a preset product Pooling aggregation parameter to obtain a characteristic maximum value of the target product node, splicing the characteristic maximum value of the target product node with the first expression vector of the target product node, and carrying out nonlinear transformation on the spliced vector to obtain the second expression vector of the target product node.
Optionally, the preset product pooling aggregation parameter includes a preset product pooling aggregation matrix and a preset offset, and the second representation vector determining unit 300 may include: a second aggregate vector determination subunit and a second representation vector determination subunit.
The second aggregate vector determining subunit is configured to input the first representative vector of each product node, the first representative vector of a target product node having a second target relationship with each product node, the preset product pooling aggregate weight matrix, and the preset bias amount to a third preset formula, respectively, and calculate to obtain a second aggregate vector of each product node.
The bias quantity (bias) is also called a neural network bias term, and plays a role in increasing the translation capacity of the neural network and enhancing the generalization effect of the neural network.
Optionally, the third preset formula is:
therein, AGGREGATE pool For the second aggregate vector, CONCATE is a matrix splicing function, v is the product node,for the first representation vector of product node v, max is the max pooling operation, RELU is a nonlinear function, W pool Pooling the aggregation weight matrix for the preset product, j being the number of the target product node, w j For said target product node numbered j having said second target relationship with said product node, ->The target product node w with the number j j B is the preset offset, and M (v) is the target product node w having a second target relationship with the product node v j Is provided for the second connection node function of (a).
The second expression vector determining subunit is configured to input the second aggregate vector of each product node and the preset product pooling aggregate weight matrix into a fourth preset formula, and calculate to obtain a second expression vector of each product node.
Optionally, the fourth preset formula is:
wherein ,is a second representation vector of the product node v.
The product vector determining device provided by the invention can determine the first initial vector of each product node and the second initial vector of each attribute node in the preset product knowledge graph, wherein the preset product knowledge graph is formed by connecting at least one product node and at least one attribute node; determining a first expression vector of each product node at least according to the first initial vector of each product node, the second initial vector of the target attribute node which has a first target relation with each product node respectively and a preset attribute average value aggregation weight matrix; and determining a second representation vector of each product node at least according to the first representation vector of each product node, the first representation vector of the target product node which has a second target relation with each product node respectively and the preset product pooling aggregation parameter. According to the embodiment of the invention, on the basis of the initial vector of the product node and the initial vector of the attribute node in the product knowledge graph, the product node can be effectively and accurately vectorized by combining the preset attribute mean value aggregation weight matrix and the preset product pooling aggregation parameter.
Optionally, another product vector determining device provided by the embodiment of the present invention may further include a parameter optimizing unit.
The parameter optimization unit is configured to determine the preset attribute mean value aggregation weight matrix through a preset loss function, and/or the preset product pooling aggregation parameter, where the preset loss function is:
wherein ,Jg For the representation of the preset loss function, v is the product node in the preset product knowledge graph, and z v For a third representation vector of the product node v, σ is a sigmoid function, T is a transposed matrix symbol, u is the target product node having the second target relationship with the product node v, z u A third representation vector for the target product node u n Z for the negatively sampled product node for which the second target relationship does not exist with the product node v un For the negative sampling product node u n Q is the negative sampled product node u having no second target relationship with the product node v n Is the number of (E) is the desired, P n (u) is the negative sampling product node u having no second target relationship with the product node v n Is a distribution of (a).
The product vector determination device includes a processor and a memory, the initial vector determination unit 100, the first expression vector determination unit 200, the second expression vector determination unit 300, and the like are stored in the memory as program units, and the processor executes the program units stored in the memory to realize corresponding functions.
The processor includes a kernel, and the kernel fetches the corresponding program unit from the memory. The kernel can be provided with one or more than one kernel, and the kernel parameters are adjusted to effectively and accurately vector the product nodes by combining a preset attribute mean value aggregation weight matrix and a preset product pooling aggregation parameter on the basis of the initial vectors of the product nodes and the initial vectors of the attribute nodes in the product knowledge graph.
The embodiment of the invention provides a storage medium, on which a program is stored, which when executed by a processor implements the product vector determination method.
The embodiment of the invention provides a processor which is used for running a program, wherein the program runs to execute the product vector determining method.
As shown in fig. 6, an embodiment of the present invention provides an electronic device 400, where the electronic device 400 includes at least one processor 401, and at least one memory 402 and a bus 403 connected to the processor 401; wherein, the processor 401 and the memory 402 complete the communication with each other through the bus 403; the processor 401 is arranged to call program instructions in the memory 402 to perform the product vector determination method described above. The electronic device 400 herein may be a server, PC, PAD, cell phone, etc.
The application also provides a computer program product adapted to perform a program initialized with the steps of the product vector determination method when executed on an electronic device.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus, electronic devices (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In one typical configuration, the electronic device includes one or more processors (CPUs), memory, and a bus. The electronic device may also include input/output interfaces, network interfaces, and the like.
The memory may include volatile memory, random Access Memory (RAM), and/or nonvolatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM), among other forms in computer readable media, the memory including at least one memory chip. Memory is an example of a computer-readable medium.
Computer readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device. Computer-readable media, as defined herein, does not include transitory computer-readable media (transmission media), such as modulated data signals and carrier waves.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises an element.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The foregoing is merely exemplary of the present application and is not intended to limit the present application. Various modifications and variations of the present application will be apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. which come within the spirit and principles of the application are to be included in the scope of the claims of the present application.
Claims (10)
1. A method of product vector determination, comprising:
determining a first initial vector of each product node and a second initial vector of each attribute node in a preset product knowledge graph, wherein the preset product knowledge graph is formed by connecting at least one product node and at least one attribute node;
Determining a first representation vector of each product node at least according to the first initial vector of each product node, the second initial vector of a target attribute node having a first target relationship with each product node and a preset attribute mean value aggregation weight matrix;
and determining a second representation vector of each product node at least according to the first representation vector of each product node, the first representation vector of a target product node which has a second target relationship with each product node respectively and a preset product pooling aggregation parameter.
2. The method of claim 1, wherein determining a first initial vector for each product node and a second initial vector for each attribute node in a preset product knowledge-graph comprises:
vectorizing the product name of each product node by using a first preset bag-of-word model to obtain a first initial vector of each product node;
and/or vectorizing the product name of each attribute node by using a second preset bag-of-word model to obtain a second initial vector of each attribute node.
3. The method of claim 1, wherein the determining the first representation vector for each of the product nodes based at least on the first initial vector for each of the product nodes, the second initial vector for each of the product nodes for which a first target relationship exists for the target attribute node, and a preset attribute mean aggregation weight matrix comprises:
Determining a first aggregate vector for each of the product nodes based at least on the first initial vector for each of the product nodes and the second initial vector for a target attribute node having a first target relationship with each of the product nodes, respectively;
and respectively inputting the first aggregate vector of each product node and a preset attribute mean value aggregate weight matrix into a first preset formula, and calculating to obtain a first expression vector of each product node.
4. A method according to claim 3, wherein said determining a first aggregate vector for each of said product nodes based at least on said first initial vector for each of said product nodes and said second initial vector for a target attribute node having a first target relationship with each of said product nodes, respectively, comprises:
determining a first number of the product nodes in the preset product knowledge graph;
respectively determining a second number of product nodes connected with each target attribute node;
determining the inverse degree value corresponding to each target attribute node according to the first quantity and the second quantity;
for any one of the product nodes: inputting the first initial vector of the product node, the second initial vector of the target attribute node with a first target relation with the product node and the inverse output value corresponding to the target attribute node into a second preset formula, and calculating to obtain a first aggregate vector of the product node.
5. The method of claim 1, wherein the predetermined product pooling aggregation parameters include a predetermined product pooling aggregation matrix and a predetermined offset, wherein the determining the second representation vector for each of the product nodes based at least on the first representation vector for each of the product nodes, the first representation vector for the target product node having a second target relationship with each of the product nodes, respectively, and the predetermined product pooling aggregation parameters comprises:
inputting the first expression vector of each product node, the first expression vector of a target product node which has a second target relation with each product node, the preset product pooling aggregation weight matrix and the preset offset into a third preset formula respectively, and calculating to obtain a second aggregation vector of each product node;
and respectively inputting the second aggregate vector of each product node and the preset product pooling aggregate weight matrix into a fourth preset formula, and calculating to obtain a second expression vector of each product node.
6. The method of claim 1, wherein the first target relationship is a one-hop neighbor and/or the second target relationship is a two-hop neighbor.
7. The method of claim 4, wherein the second predetermined formula is:
therein, AGGREGATE mean For the first aggregate vector, MEAN is a matrix averaging function, CONCATE is a matrix splicing function, v is the product node, i is the number of the target attribute node, u i For the target attribute node numbered i having a first target relationship with the product node,for said first initial vector of the product node v,/i>The target attribute node u with the number of i i V of the first initial vector of (a) p For said first quantity,/a>The target attribute node u with the number of i i A second number of connected product nodes, +.>The target attribute node u with the number of i i The corresponding inverse degree value, N (v), is a first connection node function of the target attribute node having a first target relationship with the product node v;
the first preset formula is:
wherein ,for the first representation vector of product node v, RELU is a nonlinear function, W mean And aggregating a weight matrix for the preset attribute mean value.
8. The method of claim 5, wherein the third predetermined formula is:
therein, AGGREGATE pool For the second aggregate vector, CONCATE is a matrix splicing function, v is the product node,for the first representation vector of product node v, max is the max pooling operation, RELU is a nonlinear function, W pool Pooling the aggregation weight matrix for the preset product, j being the number of the target product node, w j For said target product node numbered j having said second target relationship with said product node, ->The target product node w with the number j j B is the preset offset, and M (v) is the target product node w having a second target relationship with the product node v j Is a second connection node function of (a);
the fourth preset formula is:
wherein ,is a second representation vector of the product node v.
9. The method as recited in claim 1, further comprising:
determining the preset attribute mean value aggregation weight matrix through a preset loss function, and/or pooling aggregation parameters of the preset products, wherein the preset loss function is as follows:
wherein ,Jg For the representation of the preset loss function, v is the product node in the preset product knowledge graph, and z v For a third representation vector of the product node v, σ is a sigmoid function, T is a transposed matrix symbol, u is the target product node having the second target relationship with the product node v, z u A third representation vector for the target product node u n For a negatively sampled product node for which said second target relationship does not exist with said product node v,for the negative sampling product node u n Q is the negative sampled product node u having no second target relationship with the product node v n Is the number of (E) is the desired, P n (u) is the negative sampling product node u having no second target relationship with the product node v n Is a distribution of (a).
10. A product vector determination apparatus, comprising: an initial vector determination unit, a first expression vector determination unit, and a second expression vector determination unit,
the initial vector determining unit is used for determining a first initial vector of each product node and a second initial vector of each attribute node in a preset product knowledge graph, wherein the preset product knowledge graph is formed by connecting at least one product node and at least one attribute node;
the first representation vector determining unit is configured to determine a first representation vector of each product node according to at least the first initial vector of each product node, the second initial vector of a target attribute node having a first target relationship with each product node, and a preset attribute average value aggregation weight matrix;
The second representation vector determining unit is configured to determine a second representation vector of each product node according to at least the first representation vector of each product node, the first representation vector of a target product node having a second target relationship with each product node, and a preset product pooling aggregation parameter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110206095.2A CN112785350B (en) | 2021-02-24 | 2021-02-24 | Product vector determining method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110206095.2A CN112785350B (en) | 2021-02-24 | 2021-02-24 | Product vector determining method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112785350A CN112785350A (en) | 2021-05-11 |
| CN112785350B true CN112785350B (en) | 2023-09-19 |
Family
ID=75761898
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110206095.2A Active CN112785350B (en) | 2021-02-24 | 2021-02-24 | Product vector determining method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112785350B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108805291A (en) * | 2017-04-27 | 2018-11-13 | 清华大学 | A kind of training method and server of network representation learning model |
| CN109597856A (en) * | 2018-12-05 | 2019-04-09 | 北京知道创宇信息技术有限公司 | A kind of data processing method, device, electronic equipment and storage medium |
| CN110084296A (en) * | 2019-04-22 | 2019-08-02 | 中山大学 | A kind of figure expression learning framework and its multi-tag classification method based on certain semantic |
| CN111046185A (en) * | 2019-12-16 | 2020-04-21 | 重庆邮电大学 | Method, device and terminal for extracting knowledge graph relation of text information |
| CN111460171A (en) * | 2020-03-30 | 2020-07-28 | 支付宝(杭州)信息技术有限公司 | Target user identification method and device for server |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10984045B2 (en) * | 2017-05-24 | 2021-04-20 | International Business Machines Corporation | Neural bit embeddings for graphs |
| CN111461301B (en) * | 2020-03-30 | 2021-05-25 | 北京沃东天骏信息技术有限公司 | Serialized data processing method and device, and text processing method and device |
-
2021
- 2021-02-24 CN CN202110206095.2A patent/CN112785350B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108805291A (en) * | 2017-04-27 | 2018-11-13 | 清华大学 | A kind of training method and server of network representation learning model |
| CN109597856A (en) * | 2018-12-05 | 2019-04-09 | 北京知道创宇信息技术有限公司 | A kind of data processing method, device, electronic equipment and storage medium |
| CN110084296A (en) * | 2019-04-22 | 2019-08-02 | 中山大学 | A kind of figure expression learning framework and its multi-tag classification method based on certain semantic |
| CN111046185A (en) * | 2019-12-16 | 2020-04-21 | 重庆邮电大学 | Method, device and terminal for extracting knowledge graph relation of text information |
| CN111460171A (en) * | 2020-03-30 | 2020-07-28 | 支付宝(杭州)信息技术有限公司 | Target user identification method and device for server |
Non-Patent Citations (2)
| Title |
|---|
| Self-supervised Hierarchical Graph Neural Network for Graph Representation;Sambaran Bandyopadhyay等;2020 IEEE International Conference on Big Data (Big Data);603-608 * |
| 自然语言驱动的地理知识图谱构建方法研究;王曙;中国博士学位论文全文数据库基础科学辑(第2期);A002-16 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112785350A (en) | 2021-05-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10762163B2 (en) | Probabilistic matrix factorization for automated machine learning | |
| CN111241412B (en) | Method, system and device for determining map for information recommendation | |
| CN104809139B (en) | Code file querying method and device | |
| CN113516246A (en) | Parameter optimization method, quantum chip control method and device | |
| CN111626311A (en) | Heterogeneous graph data processing method and device | |
| CN115699058A (en) | Feature interaction through edge search | |
| CN109597982A (en) | Summary texts recognition methods and device | |
| Abdi et al. | A globally convergent BFGS method for pseudo-monotone variational inequality problems | |
| CN112785350B (en) | Product vector determining method and device | |
| CN112232360A (en) | Image retrieval model optimization method, image retrieval device and storage medium | |
| CN112001748A (en) | Data expansion method and equipment based on label propagation | |
| CN112200666A (en) | Feature vector processing method and related device | |
| CN112364181B (en) | Insurance product matching degree determining method and apparatus | |
| CN116257798A (en) | Click rate prediction model training and click rate prediction method, system and equipment | |
| Garouani et al. | Scalable meta-bayesian based hyperparameters optimization for machine learning | |
| CN111475641B (en) | Data extraction method and device, storage medium and equipment | |
| CN112613544A (en) | Target detection method, device, electronic equipment and computer readable medium | |
| Toscano-Palmerin et al. | Stratified bayesian optimization | |
| Yang et al. | Autoregressive graph Volterra models and applications | |
| Sharma et al. | Exponential adaptive strategy in spider monkey optimization algorithm | |
| CN115035349B (en) | Point representation learning method, representation method and device of graph data and storage medium | |
| Surapunt et al. | MIC for Analyzing Attributes Associated with Thai Agricultural Products | |
| CN116644802A (en) | Model training method and device | |
| CN117726402A (en) | Commodity ordering information prediction method and device and electronic equipment | |
| CN113868428A (en) | Service recommendation method and device, electronic equipment and computer readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20231215 Address after: B3, No.1 Jinyun East Third Lane, Chengdu High tech Zone, Chengdu City, Sichuan Province, 610000 Patentee after: Huize (Chengdu) Network Technology Co.,Ltd. Address before: 1201-1207, Jingxing Haihai building, 91 guiwan Third Road, Nanshan street, Qianhai Shenzhen Hong Kong cooperation zone, Shenzhen, Guangdong 518000 Patentee before: SHENZHEN HUIZE TIMES TECHNOLOGY CO.,LTD. |
|
| TR01 | Transfer of patent right |