CN112860949A - Method and device for extracting map features - Google Patents
Method and device for extracting map features Download PDFInfo
- Publication number
- CN112860949A CN112860949A CN201911180676.2A CN201911180676A CN112860949A CN 112860949 A CN112860949 A CN 112860949A CN 201911180676 A CN201911180676 A CN 201911180676A CN 112860949 A CN112860949 A CN 112860949A
- Authority
- CN
- China
- Prior art keywords
- nodes
- graph
- node
- feature
- target
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 48
- 239000011159 matrix material Substances 0.000 claims abstract description 40
- 238000000605 extraction Methods 0.000 claims abstract description 20
- 230000008569 process Effects 0.000 claims description 12
- 238000001228 spectrum Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 4
- 230000003595 spectral effect Effects 0.000 claims 4
- 239000013598 vector Substances 0.000 claims 1
- 239000000284 extract Substances 0.000 abstract description 6
- 230000006870 function Effects 0.000 description 21
- 238000004891 communication Methods 0.000 description 6
- 230000004913 activation Effects 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000003287 optical effect Effects 0.000 description 4
- 238000012502 risk assessment Methods 0.000 description 4
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000003993 interaction Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/901—Indexing; Data structures therefor; Storage structures
- G06F16/9024—Graphs; Linked lists
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the invention provides a method and a device for extracting map features, relates to the technical field of computer software, and can extract the map features according to adjacent nodes and improve the accuracy of map feature extraction. The method comprises the following steps: extracting the characteristics of the nodes in the map to generate a characteristic set; acquiring adjacent nodes of a target node in a graph to generate a target adjacent graph; inquiring the characteristics of the adjacent nodes of the target node in the characteristic set according to the target neighbor graph to generate a characteristic matrix of the adjacent nodes; extracting edge feature sets of a target node and adjacent nodes; normalizing the edge feature set to generate an attention coefficient of an adjacent node; and carrying out weighted summation on the feature matrixes of the adjacent nodes by using the attention coefficients of the adjacent nodes, and calculating the feature code of the target node. The embodiment of the application is applied to extracting the map features of the nodes.
Description
Technical Field
The embodiment of the invention relates to the technical field of computer software, in particular to a method and a device for extracting map features.
Background
The traditional relational data expose the problems of modeling defects, horizontal scaling, etc., so that graph (graph) data with more powerful expression ability is greatly valued by the industry. A graph is an important data structure, which is composed of nodes (vertex) and edges (edge), and is generally denoted as G (V, E), wherein the nodes may also be called vertices and represent individuals; edges represent connections between individuals. In the graph, each node connection is different, and some nodes have three connections and some nodes have two connections and are irregular data structures.
In real life, a lot of data can be represented by graph data structures, for example, abstracted graphs of social relationships, electronic transaction relationships, molecular structures, and the like. For example, in a web page search, a web page may be taken as a node, and the hyperlink relationship between pages as an edge; in a social network, users may be used as nodes, and relationships established between the users are used as edges, for example, in a social network of WeChat, personal and public numbers are used as nodes, and attention and approval are used as edges to form a graph; for e-commerce, the trading network of the shopping mall is a map composed of nodes (individuals and commodities) and edges (purchasing and collecting).
The EdgeConv is a graph convolution network widely applied to graph feature extraction, and the graph feature extraction step mainly comprises the steps of extracting the feature of each edge, and then accumulating and summing the features of the edges to obtain the feature codes of the nodes to obtain the graph features of the nodes. The extracted profile features can be used for recommendation or anti-fraud. Firstly, the influence of the self characteristics of adjacent nodes on the nodes is ignored in a mode of extracting the map characteristics by the EdgeConv; secondly, because the neighboring nodes of different categories are far away and the neighboring nodes of the same category are near, the influence degree of each neighboring node on the central node is different, so that the method only considering the edge features can not extract the map features accurately enough, further, the matching degree between the recommended content and the recommended object is low when the map features are recommended, and the risk assessment is not accurate enough when the map features are applied for anti-fraud.
Disclosure of Invention
The embodiment of the invention provides a method and a device for extracting map features, which can extract the map features according to adjacent nodes and improve the accuracy of map feature extraction.
In a first aspect, a method for extracting features of a map is provided, which includes the following steps: extracting the characteristics of nodes in a map to generate a characteristic set, wherein the map comprises at least two nodes; the nodes in the graph comprise information in the e-commerce transaction process, the graph further comprises edges connecting the two nodes, and the edges are used for representing the relationship between the nodes at the two ends of the edges; acquiring adjacent nodes of a target node in a graph to generate a target adjacent graph; inquiring the characteristics of the adjacent nodes of the target node in the characteristic set according to the target neighbor graph to generate a characteristic matrix of the adjacent nodes; extracting edge feature sets of a target node and adjacent nodes; normalizing the edge feature set to generate an attention coefficient of an adjacent node; and carrying out weighted summation on the feature matrixes of the adjacent nodes by using the attention coefficients of the adjacent nodes, and calculating the feature code of the target node.
In the scheme, the characteristics of the nodes in the map are extracted to generate a characteristic set, wherein the map comprises at least two nodes; the nodes in the graph comprise information in the e-commerce transaction process, the graph further comprises edges connecting the two nodes, and the edges are used for representing the relationship between the nodes at the two ends of the edges; acquiring adjacent nodes of a target node in a graph to generate a target adjacent graph; inquiring the characteristics of the adjacent nodes of the target node in the characteristic set according to the target neighbor graph to generate a characteristic matrix of the adjacent nodes; extracting edge feature sets of a target node and adjacent nodes; normalizing the edge feature set to generate an attention coefficient of an adjacent node; and carrying out weighted summation on the feature matrixes of the adjacent nodes by using the attention coefficients of the adjacent nodes, and calculating the feature code of the target node. Therefore, the characteristics of the neighboring nodes of the target node are obtained firstly in the application; and then calculating the attention coefficient of each adjacent node, and finally calculating the feature code of the target node by using the attention coefficient of the adjacent node of the target node and the features of the adjacent node together, so that the influence of the adjacent node on the target node is not considered when the feature code of the target node is calculated by directly accumulating and summing the edge features of the target node and the adjacent node, the accuracy of map feature extraction is improved, the matching degree between recommended content and a recommended object is further improved when map features are applied for recommendation, and the accuracy of risk assessment can be improved when the map features are applied for anti-fraud.
Optionally, the weighted summation of the feature matrices of the neighboring nodes by using the attention coefficients of the neighboring nodes includes: and inputting the attention coefficient of the adjacent node and the feature matrix of the adjacent node into a graph convolution network, and calculating the feature code of the target node.
Optionally, inputting the attention coefficient of the neighboring node and the feature matrix of the neighboring node into a graph convolution network, and calculating the feature code of the target node, including: inputting attention coefficients of adjacent nodes and feature matrixes of adjacent nodes into a formulaObtaining the feature code of the target node, wherein piA signature code representing the target node,the attention coefficient of the neighboring node is represented,a feature that is representative of a neighboring node,the deviation of the graph convolution network is shown.
Optionally, extracting features of nodes in the graph to generate a feature set, including: and extracting the characteristics of the nodes in the map by using the multilayer perceptron to generate a characteristic set.
Optionally, obtaining neighboring nodes of the target node in the graph, and generating a target neighbor graph, includes: and acquiring adjacent nodes of the target node in the graph according to a K neighbor method to generate a target neighbor graph.
Optionally, the querying, according to the target neighbor graph, features of neighboring nodes of the target node in the feature set, and generating a feature matrix of the neighboring nodes, includes: using FIND NiThe function queries the characteristics of the adjacent nodes of the target node in the characteristic set according to the target neighbor graph to generate a characteristic matrix of the adjacent nodes, wherein N isiRepresenting a target neighbor graph.
Optionally, normalizing the edge feature set to generate an attention coefficient of a neighboring node, including: according to the formulaNormalizing the edge feature set to generate attention coefficients of neighboring nodes, wherein,the attention coefficient of the neighboring node is represented,representing edge features in the edge feature set.
In a second aspect, there is provided an apparatus for extracting features of a spectrum, including: the generating module is used for extracting the characteristics of the nodes in the map and generating a characteristic set, wherein the map comprises at least two nodes; the nodes in the graph comprise information in the e-commerce transaction process, the graph further comprises edges connecting the two nodes, and the edges are used for representing the relationship between the nodes at the two ends of the edges; the generation module is also used for acquiring adjacent nodes of the target node in the map and generating a target adjacent map; the query module is used for querying the characteristics of the adjacent nodes of the target node in the characteristic set according to the target neighbor graph generated by the generation module and generating a characteristic matrix of the adjacent nodes; the generating module is also used for extracting edge feature sets of the target node and the adjacent nodes; the generating module is further used for normalizing the edge feature set and generating an attention coefficient of the adjacent node; and the calculation module is used for performing weighted summation on the feature matrix of the adjacent node generated by the query module by using the attention coefficient of the adjacent node generated by the generation module, and calculating the feature code of the target node.
Optionally, the calculation module is specifically configured to input the attention coefficient of the neighboring node and the feature matrix of the neighboring node into the graph convolution network, and calculate the feature code of the target node.
Optionally, the calculation module is specifically configured to input the attention coefficient of the neighboring node and the feature matrix of the neighboring node into the formulaObtaining the feature code of the target node, wherein piA signature code representing the target node,the attention coefficient of the neighboring node is represented,a feature that is representative of a neighboring node,the deviation of the graph convolution network is shown.
Optionally, the generating module is specifically configured to extract features of nodes in the graph by using the multilayer perceptron, and generate a feature set.
Optionally, the generating module is specifically configured to obtain a neighboring node of the target node in the graph according to a K-nearest neighbor method, and generate the target-nearest neighbor graph.
Optionally, the query module is specifically configured to utilize FIND NiThe function queries the characteristics of the adjacent nodes of the target node in the characteristic set according to the target neighbor graph to generate a characteristic matrix of the adjacent nodes, wherein N isiRepresenting a target neighbor graph.
Optionally, the generating module is specifically configured to generate the formula Normalizing the set of edge features to generate neighboring nodesAttention is drawn to the force coefficient, where,the attention coefficient of the neighboring node is represented,representing edge features in the edge feature set.
In a third aspect, an apparatus for extracting a feature of a spectrum is provided, which includes a processor executing a computer to execute instructions to cause the apparatus for extracting a feature of a spectrum to perform the method for extracting a feature of a spectrum as described above.
In a fourth aspect, there is provided a computer storage medium comprising instructions which, when run on a computer, cause the computer to perform the method of extracting map features as described above.
In a fifth aspect, a computer program product is provided, which comprises instruction codes for executing the method for extracting the atlas feature as described above.
It should be understood that any one of the above-provided apparatus, computer storage medium, or computer program product for extracting a map feature is used to execute the method according to the first aspect provided above, and therefore, the beneficial effects achieved by the method according to the first aspect and the beneficial effects of the solutions in the following detailed description may be referred to, and are not repeated herein.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without creative efforts.
Fig. 1 is a schematic diagram of a transaction relationship map in an e-commerce scenario according to an embodiment of the present invention;
fig. 2 is a schematic flow chart of a method for extracting features of a map according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of an AttentionEdgeConv module according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of an apparatus for extracting features of a map provided in an embodiment of the present invention;
fig. 5 is a schematic structural diagram of an apparatus for extracting a feature of a spectrum according to another embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In real life, a lot of data can be represented by graph data structures, for example, abstracted graphs of social relationships, electronic transaction relationships, molecular structures, and the like. For example, in a web page search, a web page may be taken as a node, and the hyperlink relationship between pages as an edge; in a social network, users may be used as nodes, and relationships established between the users are used as edges, for example, in a social network of WeChat, personal and public numbers are used as nodes, and attention and approval are used as edges to form a graph; for e-commerce, the trading network of the shopping mall is a map composed of nodes (individuals and commodities) and edges (purchasing and collecting). For another example, as shown in fig. 1, the transaction relationship map in the e-commerce scene includes nodes such as a category 11, a commodity 12, a transaction network protocol (IP) 13, a transaction 14, a registration address 15, a user 16, and a receiving address 17, and then the node associated with the user 16 includes the registration address 15 and the receiving address 16; the nodes related to the transaction 14 comprise commodities 12, a transaction IP13, a registration address 15, a user 16 and a receiving address 17; the nodes associated with the items 12 have categories 11; the user 16 may also associate the product 12 with the score, and the score may be the edge between the user 16 and the product 12.
As can be seen from the above description, first, each node in the graph has its own feature information, for example, a wind control system is established for fig. 1, and whether the registered address 15, the transaction IP13, and the receiving address 16 of the user 16 in the transaction relationship graph in the e-commerce scene are consistent with the pre-stored node information associated with the user 16 is detected, and if these feature information do not match, the wind control system determines that the user 16 has a certain fraud risk. Second, each node in the graph has structural information. If a node has a large number of connected transaction nodes, i.e. a large number of edges extending from the node, at a certain time, the wind control system determines that the node is at risk.
EdgeConv is a graph convolution network widely used for graph feature extraction, and for a graph G, its node set is v ═ 1,2, … n }, and the corresponding directed edge setSuppose a distance node piNearest point pj1,pj2,…pjkThe directional edge is formed as<i,j1>,<i,j2>,…<i,jk>The edge of the directed edge is characterized byFirst, EdgeConv is based on the formulaComputing node piEdge feature ofWherein p isiRepresenting nodes in a graph, pjk-piRepresenting a node piAnd node pjkThe edge in between. Wherein h isΘCan be expressed asFor hiding the layer function, it is a nonlinear function composed of some learnable parameters weight and bias, specifically, hΘ(pi,pj-pi)=w×[pi||pj-pi]+ b, the activation function is ReLU, where the digital expression of the ReLU activation function is f (x) max (0, x), and | is the matrix join operation, representing the node piCharacteristic of (1) and edge pj-piThe characteristic of (a) is spliced point by point, and b is the deviation of the graph convolution network. Second, EdgeConv obtains edge featuresThen, according to the formulaPerforming channel-level aggregation operation on all AND nodes piAssociated edge featuresFusing, and taking the fused feature as a node piWherein the digital expression of the activation function σ ispi ′Is a node piCharacteristic code of (1), piRepresenting nodes in a graph, pjk-piRepresenting a node piAnd node pjkThe edge in between.
Firstly, the influence of the self characteristics of adjacent nodes on the nodes is ignored in a mode of extracting the map characteristics by the EdgeConv; secondly, because the neighboring nodes of different categories are far away and the neighboring nodes of the same category are near, the influence degree of each neighboring node on the central node is different, so that the method only considering the edge features can not extract the map features accurately enough, further, the matching degree between the recommended content and the recommended object is low when the map features are recommended, and the risk assessment is not accurate enough when the map features are applied for anti-fraud.
In order to solve the above problem, the present application provides a method for extracting a feature of a spectrum, which is shown in fig. 2 and specifically includes the following steps:
201. and extracting the characteristics of the nodes in the map to generate a characteristic set.
Firstly, the graph comprises at least two nodes; for example, as shown in fig. 1, when a user 16 can associate a commodity 12 by scoring in the transaction relationship map in the e-commerce scene, the scoring connects the user 16 and the commodity 12, which is the edge between the user 16 and the commodity 12.
Secondly, inputting the graph into an Artificial Neural Network (ANN) to extract the characteristics of the nodes in the graph and generate a characteristic set, specifically, the graph can be input into the ANN with a structure of { a }1,a2,…,anAnd (4) extracting the characteristics of the nodes in the map by using a multi-layer perceptron (MLP) to generate a characteristic set, wherein the MLP comprises an input layer, an output layer and a hidden layer and is used for extracting the characteristics of the nodes in the map.
Further, in another optional aspect of the present application, a Back Propagation (BP) neural network may also be used to extract features of nodes in the graph.
202. And acquiring adjacent nodes of the target node in the graph to generate a target adjacent graph.
Specifically, a target neighbor graph is generated by acquiring neighbor nodes of a target node in a graph according to a K-nearest neighbor (KNN) method, wherein the target neighbor graph includes a relationship between the target node and the neighbor nodes of the target node.
203. And querying the characteristics of the adjacent nodes of the target node in the characteristic set according to the target neighbor graph to generate a characteristic matrix of the adjacent nodes.
Specifically, the relationship between the target node and the neighboring node of the target node in the target neighbor graph is first utilized to query the neighboring node of the target node, and then the feature corresponding to the neighboring node of the target node in the feature set is obtained to generate the feature matrix of the neighboring node of the target node, specifically, the FIND N may be utilizediFunction queries neighbors of target nodes in feature set according to target neighbor graphFeatures of the near nodes, generating a feature matrix of the neighboring nodes, where NiRepresenting a target neighbor graph.
204. And extracting edge feature sets of the target node and the adjacent nodes.
Specifically, a structure of { a }can be used1,a2,…,anThe MLP of (1) extracts the set of edge features of the target node and neighboring nodes in the map, e.g., in MLP, hΘAs a function of the hidden layer in MLP, in particular according to the formulaExtracting a target node piEdge feature ofWherein p isiRepresenting target nodes in a graph, pjk-piRepresenting a target node piNeighboring node p with the target nodejkThe edge in between. Wherein h isΘIs a non-linear function composed of some learnable parameters weight and bias. h isΘ(pi,pj-pi)=w×[pi||pj-pi]+ b, the activation function is ReLU, where the digital expression of the ReLU activation function is f (x) max (0, x), and | is the matrix join operation, representing the target node piCharacteristic of (1) and edge pj-piThe characteristic of (a) is spliced point by point, and b is the deviation of the graph convolution network.
205. And normalizing the edge feature set to generate the attention coefficient of the adjacent node.
Specifically, since the hidden layer function for extracting the edge feature set is a nonlinear function composed of learnable parameters weight and bias, the features in the edge feature set of the target node and the neighboring nodes extracted by MLP indicate the importance degree of the neighboring nodes to the target node, so that the attention coefficient can be easily compared between different neighboring nodes according to a formulaThe set of normalized edge features is then normalized,attention coefficients for neighboring nodes are generated, wherein,the attention coefficient of the neighboring node is represented,representing edge features in the edge feature set.
Further, in another optional scheme of the present application, the edge feature set of the target node and the neighboring node may also be directly used as the attention coefficient of the neighboring node.
206. And carrying out weighted summation on the feature matrixes of the adjacent nodes by using the attention coefficients of the adjacent nodes, and calculating the feature code of the target node.
Specifically, the attention coefficients of the adjacent nodes and the feature matrix of the adjacent nodes are input into a graph convolution network, and the feature code of the target node is calculated, wherein the graph convolution network is used for performing weighted summation on the attention coefficients of the adjacent nodes and the feature matrix of the adjacent nodes, and calculating the feature code of the target node. Further, the attention coefficient of the adjacent node and the characteristic matrix of the adjacent node are input into the convolution formula of the graph convolution networkObtaining a feature code of the target node, wherein piA signature code representing the target node,the attention coefficient of the neighboring node is represented,a feature that is representative of a neighboring node,the deviation of the graph convolution network is shown.
Further, in another optional scheme of the present application, the attention coefficient of the neighboring node may also be compared with the neighboring nodeConvolution formula of characteristic matrix input graph convolution network of node Obtaining the feature code of the target node, wherein piA signature code representing the target node,the attention coefficient of the neighboring node is represented,representing characteristics of neighboring nodes.
For example, the map feature extraction device of the present application may include an AttentionEdgeConv module, as shown in fig. 3, where the AttentionEdgeConv module includes a KNN module 31, an MLP module 32, a softmax module 33, a FIND module 34, and a weighted summation module 35, where the KNN module 31 is configured to acquire neighboring nodes of a target node in a map, and generate a target neighbor map; the MLP module 32 is configured to extract features of nodes in the map, generate a feature set, and extract edge feature sets of a target node and neighboring nodes; the softmax module 33 is used for normalizing the edge feature set and generating an attention coefficient of a neighboring node; the FIND module 34 is configured to query, according to the target neighbor graph, features of neighboring nodes of the target node in the feature set, and generate a feature matrix of the neighboring nodes; the weighted summation module 35 is configured to perform weighted summation on the feature matrix of the neighboring node by using the attention coefficient of the neighboring node, and calculate a feature code of the target node to obtain a feature of the target node.
In the scheme, the characteristics of the nodes in the map are extracted to generate a characteristic set, wherein the map comprises at least two nodes; the nodes in the graph comprise information in the e-commerce transaction process, the graph further comprises edges connecting the two nodes, and the edges are used for representing the relationship between the nodes at the two ends of the edges; acquiring adjacent nodes of a target node in a graph to generate a target adjacent graph; inquiring the characteristics of the adjacent nodes of the target node in the characteristic set according to the target neighbor graph to generate a characteristic matrix of the adjacent nodes; extracting edge feature sets of a target node and adjacent nodes; normalizing the edge feature set to generate an attention coefficient of an adjacent node; and carrying out weighted summation on the feature matrixes of the adjacent nodes by using the attention coefficients of the adjacent nodes, and calculating the feature code of the target node. Therefore, the characteristics of the neighboring nodes of the target node are obtained firstly in the application; and then calculating the attention coefficient of each adjacent node, and finally calculating the feature code of the target node by using the attention coefficient of the adjacent node of the target node and the features of the adjacent node together, so that the influence of the adjacent node on the target node is not considered when the feature code of the target node is calculated by directly accumulating and summing the edge features of the target node and the adjacent node, the accuracy of map feature extraction is improved, the matching degree between recommended content and a recommended object is further improved when map features are applied for recommendation, and the accuracy of risk assessment can be improved when the map features are applied for anti-fraud.
In the embodiment of the present invention, the function modules of the device for extracting feature of a map may be divided according to the above method embodiment, for example, each function module may be divided corresponding to each function, or two or more functions may be integrated into one processing module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode. It should be noted that, the division of the modules in the embodiment of the present invention is schematic, and is only a logic function division, and there may be another division manner in actual implementation.
Referring to fig. 4, the present application provides an apparatus for extracting a feature of a spectrum, including: a generating module 41, configured to extract features of nodes in a graph, and generate a feature set, where the graph includes at least two nodes; the nodes in the graph comprise information in the e-commerce transaction process, the graph further comprises an edge connecting the two nodes, and the edge is used for representing the relationship between the nodes at the two ends of the edge; the generating module 41 is further configured to obtain neighboring nodes of the target node in the graph, and generate a target neighbor graph; a query module 42, configured to query features of neighboring nodes of the target node in the feature set according to the target neighbor graph generated by the generation module 41, and generate a feature matrix of the neighboring nodes; the generating module 41 is further configured to extract a set of edge features of the target node and the neighboring nodes; the generating module 41 is further configured to normalize the edge feature set, and generate an attention coefficient of the neighboring node; a calculating module 43, configured to perform weighted summation on the feature matrix of the neighboring node generated by the querying module 42 by using the attention coefficient of the neighboring node generated by the generating module 41, and calculate a feature code of the target node.
Optionally, the calculating module 43 is specifically configured to input the attention coefficient of the neighboring node and the feature matrix of the neighboring node into a graph convolution network, and calculate the feature code of the target node.
Optionally, the calculating module 43 is specifically configured to input the attention coefficient of the neighboring node and the feature matrix of the neighboring node into a formulaAcquiring the feature code of the target node, wherein p isiA signature code representing the target node,an attention coefficient representing the neighboring node,a characteristic of the neighboring node is represented,representing the deviation of the graph convolution network.
Optionally, the generating module 41 is specifically configured to extract features of nodes in the graph by using a multi-layer perceptron, and generate a feature set.
Optionally, the generating module 41 is specifically configured to obtain neighboring nodes of the target node in the graph according to a K-nearest neighbor method, and generate the target-nearest neighbor graph.
Optionally, the query module 42 is specifically configured to utilize FIND NiThe function queries the characteristics of the adjacent nodes of the target node in the characteristic set according to the target neighbor graph to generate a characteristic matrix of the adjacent nodes, wherein N isiRepresenting the target neighbor graph.
Optionally, the generating module 41 is specifically configured to generate the formula Normalizing the set of edge features to generate attention coefficients for the neighboring nodes, wherein,an attention coefficient representing the neighboring node,representing edge features in the edge feature set.
In the case of using an integrated module, the extraction means of the map features comprises: the device comprises a storage unit, a processing unit and an interface unit. The processing unit is used for controlling and managing the action of the map feature extraction device. And the interface unit is responsible for information interaction between the extraction device of the map features and other equipment. And the storage unit is used for storing program codes and data of the extraction device of the map features.
Wherein, the processing unit may be a processor, the storage unit may be a memory, and the interface unit may be a communication interface.
The device for extracting the atlas feature is shown in fig. 5, and includes a processor 502, where the processor 502 is configured to execute application program code, so as to implement the method described in the embodiment of the present application.
The processor 502 may be a general-purpose Central Processing Unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more ics for controlling the execution of programs in accordance with the present disclosure.
As shown in fig. 5, the device for extracting feature map may further include a memory 503.
The memory 503 may be, but is not limited to, a read-only memory (ROM) or other type of static storage device that may store static information and instructions, a Random Access Memory (RAM) or other type of dynamic storage device that may store information and instructions, an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compact disc, laser disc, optical disc, digital versatile disc, blu-ray disc, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. The memory may be self-contained and coupled to the processor via a bus. The memory may also be integral to the processor.
The memory 503 is used for storing application program codes for executing the scheme of the application, and the processor 502 controls the execution. As shown in fig. 5, the extraction device of the atlas feature may further include a communication interface 501. The communication interface 501, the processor 502, and the memory 503 may be coupled to each other, for example, by a bus 504.
The communication interface 501 is used for information interaction with other devices, for example, information interaction between the extraction apparatus supporting the feature of the map and other devices, for example, data acquisition from other devices or data transmission to other devices.
Further, a computing storage medium (or medium) is also provided, which includes instructions that when executed perform the operations of the extraction method of the feature map in the above-described embodiments. Additionally, a computer program product is also provided, comprising the above-described computing storage medium (or media).
All relevant contents of each step related to the above method embodiment may be referred to the functional description of the corresponding functional module, and the function thereof is not described herein again.
It should be understood that, in various embodiments of the present invention, the sequence numbers of the above-mentioned processes do not mean the execution sequence, and the execution sequence of each process should be determined by its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.
Those of ordinary skill in the art would appreciate that the various illustrative modules, elements, and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus, and method may be implemented in other ways. For example, the device embodiments described above are merely illustrative, e.g., multiple units or components may be combined or may be integrated into another system, or some features may be omitted, or not implemented. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U disk, a removable hard disk, a read-only memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
The above description is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present invention, and all the changes or substitutions should be covered within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the appended claims.
Claims (17)
1. A method for extracting map features is characterized in that,
extracting the characteristics of nodes in a map to generate a characteristic set, wherein the map comprises at least two nodes; the nodes in the graph comprise information in the e-commerce transaction process, the graph further comprises an edge connecting the two nodes, and the edge is used for representing the relationship between the nodes at the two ends of the edge;
acquiring adjacent nodes of a target node in a graph to generate a target adjacent graph;
inquiring the characteristics of the adjacent nodes of the target node in the characteristic set according to the target neighbor graph to generate a characteristic matrix of the adjacent nodes;
extracting edge feature sets of the target node and the adjacent nodes;
normalizing the edge feature set to generate an attention coefficient of the adjacent node;
and carrying out weighted summation on the feature matrixes of the adjacent nodes by using the attention coefficients of the adjacent nodes, and calculating the feature code of the target node.
2. The method for extracting feature vectors as claimed in claim 1, wherein the weighted summation of the feature matrices of the neighboring nodes using the attention coefficients of the neighboring nodes comprises:
and inputting the attention coefficient of the adjacent node and the feature matrix of the adjacent node into a graph convolution network, and calculating the feature code of the target node.
3. The method for extracting features of a graph according to claim 2, wherein inputting the attention coefficients of the neighboring nodes and the feature matrix of the neighboring nodes into a graph convolution network, and calculating the feature code of the target node comprises:
inputting the attention coefficient of the adjacent node and the feature matrix of the adjacent node into a formulaAcquiring the feature code of the target node, wherein p isiA signature code representing the target node,an attention coefficient representing the neighboring node,represents the aboveThe characteristics of the adjacent nodes are such that,representing the deviation of the graph convolution network.
4. The method for extracting features of a graph according to claim 1, wherein the extracting features of nodes in the graph and generating a feature set comprise:
and extracting the characteristics of the nodes in the map by using the multilayer perceptron to generate a characteristic set.
5. The method for extracting graph features according to claim 1, wherein the obtaining neighboring nodes of a target node in a graph and generating a target neighbor graph comprises:
and acquiring adjacent nodes of the target node in the graph according to a K neighbor method to generate a target neighbor graph.
6. The method for extracting feature of a graph according to claim 1, wherein the querying features of neighboring nodes of the target node in the feature set according to the target neighbor graph to generate a feature matrix of the neighboring nodes comprises:
using FIND NiThe function queries the characteristics of the adjacent nodes of the target node in the characteristic set according to the target neighbor graph to generate a characteristic matrix of the adjacent nodes, wherein N isiRepresenting the target neighbor graph.
7. The method for extracting features of a graph according to claim 1, wherein the normalizing the edge feature set to generate the attention coefficient of the neighboring node comprises:
8. An apparatus for extracting features of a spectrum, comprising:
the generating module is used for extracting the characteristics of nodes in a map and generating a characteristic set, wherein the map comprises at least two nodes; the nodes in the graph comprise information in the e-commerce transaction process, the graph further comprises an edge connecting the two nodes, and the edge is used for representing the relationship between the nodes at the two ends of the edge;
the generation module is further used for acquiring neighboring nodes of the target node in the graph and generating a target neighboring graph;
the query module is used for querying the characteristics of the adjacent nodes of the target node in the characteristic set according to the target neighbor graph generated by the generation module to generate a characteristic matrix of the adjacent nodes;
the generating module is further used for extracting edge feature sets of the target node and the adjacent nodes;
the generating module is further configured to normalize the edge feature set, and generate an attention coefficient of the neighboring node;
and the calculation module is used for performing weighted summation on the feature matrix of the adjacent node generated by the query module by using the attention coefficient of the adjacent node generated by the generation module, and calculating the feature code of the target node.
9. The apparatus for extracting features of an atlas of claim 8,
the calculation module is specifically configured to input the attention coefficient of the neighboring node and the feature matrix of the neighboring node into a graph convolution network, and calculate the feature code of the target node.
10. The apparatus for extracting features of an atlas of claim 9,
the calculation module is specifically configured to input the attention coefficient of the neighboring node and the feature matrix of the neighboring node into a formulaAcquiring the feature code of the target node, wherein p isiA signature code representing the target node,an attention coefficient representing the neighboring node,a characteristic of the neighboring node is represented,representing the deviation of the graph convolution network.
11. The apparatus for extracting features of an atlas of claim 8,
the generating module is specifically configured to extract features of nodes in the graph by using the multilayer perceptron, and generate a feature set.
12. The apparatus for extracting features of an atlas of claim 8,
the generation module is specifically configured to acquire a neighboring node of a target node in the graph according to a K-nearest neighbor method, and generate a target-nearest neighbor graph.
13. The apparatus for extracting features of an atlas of claim 8,
the query module is specifically configured to utilize FIND NiFunction queries the feature from the target neighbor graphCentralizing the characteristics of the neighboring nodes of the target node to generate a characteristic matrix of the neighboring nodes, wherein N isiRepresenting the target neighbor graph.
14. The apparatus for extracting features of an atlas of claim 8,
15. An extraction apparatus of spectral features, characterized by comprising a processor, when the extraction apparatus of spectral features is in operation, the processor executes computer-executable instructions to cause the extraction apparatus of spectral features to perform the extraction method of spectral features according to any one of claims 1 to 7.
16. A computer storage medium comprising instructions that, when executed on a computer, cause the computer to perform a method of extracting atlas feature of any of claims 1-7.
17. A computer program product, characterized in that it comprises instruction code for performing the method of extraction of atlas features of any of claims 1-7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911180676.2A CN112860949A (en) | 2019-11-27 | 2019-11-27 | Method and device for extracting map features |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911180676.2A CN112860949A (en) | 2019-11-27 | 2019-11-27 | Method and device for extracting map features |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112860949A true CN112860949A (en) | 2021-05-28 |
Family
ID=75985572
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911180676.2A Pending CN112860949A (en) | 2019-11-27 | 2019-11-27 | Method and device for extracting map features |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112860949A (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9760619B1 (en) * | 2014-04-29 | 2017-09-12 | Google Inc. | Generating weighted clustering coefficients for a social network graph |
CN108959328A (en) * | 2017-05-27 | 2018-12-07 | 株式会社理光 | Processing method, device and the electronic equipment of knowledge mapping |
CN110009474A (en) * | 2019-01-31 | 2019-07-12 | 阿里巴巴集团控股有限公司 | A kind of assessing credit risks method, apparatus and electronic equipment |
WO2019137050A1 (en) * | 2018-01-12 | 2019-07-18 | 阳光财产保险股份有限公司 | Real-time fraud detection method and device under internet credit scene, and server |
CN110188168A (en) * | 2019-05-24 | 2019-08-30 | 北京邮电大学 | Semantic relation recognition methods and device |
-
2019
- 2019-11-27 CN CN201911180676.2A patent/CN112860949A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9760619B1 (en) * | 2014-04-29 | 2017-09-12 | Google Inc. | Generating weighted clustering coefficients for a social network graph |
CN108959328A (en) * | 2017-05-27 | 2018-12-07 | 株式会社理光 | Processing method, device and the electronic equipment of knowledge mapping |
WO2019137050A1 (en) * | 2018-01-12 | 2019-07-18 | 阳光财产保险股份有限公司 | Real-time fraud detection method and device under internet credit scene, and server |
CN110009474A (en) * | 2019-01-31 | 2019-07-12 | 阿里巴巴集团控股有限公司 | A kind of assessing credit risks method, apparatus and electronic equipment |
CN110188168A (en) * | 2019-05-24 | 2019-08-30 | 北京邮电大学 | Semantic relation recognition methods and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11080637B2 (en) | Universal model scoring engine | |
TWI776894B (en) | Merchant evaluation method and system | |
US9563921B2 (en) | System and method for detecting merchant points of compromise using network analysis and modeling | |
CN110009174A (en) | Risk identification model training method, device and server | |
JP2019525309A (en) | Method and apparatus for controlling data risk | |
US11983203B1 (en) | Systems and methods of establishing correlative relationships between geospatial data features in feature vectors representing property locations | |
CN112396189A (en) | Method and device for multi-party construction of federal learning model | |
US11909749B2 (en) | Fraud detection based on analysis of frequency-domain data | |
CN113364748B (en) | Ether house phishing node detection method and system based on transaction subgraph network | |
WO2016127632A1 (en) | Method, system, and computer device for electronic payment behavior-based data processing | |
CN111008335B (en) | Information processing method, device, equipment and storage medium | |
CN109559230B (en) | Bank transaction group discovery method and system based on overlapping community discovery algorithm | |
US11762819B2 (en) | Clustering model analysis for big data environments | |
CN111340612A (en) | Account risk identification method and device and electronic equipment | |
KR102459466B1 (en) | Integrated management method for global e-commerce based on metabus and nft and integrated management system for the same | |
CN106600360B (en) | Method and device for sorting recommended objects | |
US10713538B2 (en) | System and method for learning from the images of raw data | |
CN103678548A (en) | Failure service replacement and recommendation method based on composite patterns | |
CN117422553A (en) | Transaction processing method, device, equipment, medium and product of blockchain network | |
CN112860949A (en) | Method and device for extracting map features | |
EP3912117A1 (en) | Real time user matching using purchasing behavior | |
CN113706173B (en) | Information management method and device, electronic equipment and storage medium | |
CN114491086A (en) | Clothing personalized matching recommendation method and system, electronic equipment and storage medium | |
CN113706279A (en) | Fraud analysis method and device, electronic equipment and storage medium | |
CN113032694A (en) | Scene-based query method and device, storage medium and computer equipment |
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 |