CN111444394A

CN111444394A - Method, system and equipment for obtaining relation expression between entities and advertisement recalling system

Info

Publication number: CN111444394A
Application number: CN201910041466.9A
Authority: CN
Inventors: 陈怡然; 温世阳; 吴文金; 林伟; 朱晓宇
Original assignee: Alibaba Group Holding Ltd
Current assignee: Alibaba Group Holding Ltd
Priority date: 2019-01-16
Filing date: 2019-01-16
Publication date: 2020-07-24
Anticipated expiration: 2039-01-16
Also published as: CN111444394B; WO2020147594A1

Abstract

The invention discloses a method, a system and equipment for obtaining relationship expression between entities and an advertisement recalling system. The method comprises the following steps: dividing the heterogeneous graph into at least two heterogeneous subgraphs according to the meta-path, and acquiring sample data of one batch; learning the sample data according to the heterogeneous subgraph to obtain vector expression of nodes in the heterogeneous subgraph; aggregating vector expressions of the same node in different heterogeneous subgraphs based on sample data to obtain the same vector expression of the same node in different heterogeneous subgraphs; optimizing the parameters of the model based on the sample data and the same vector expression of the same node; and acquiring sample data of the next batch for learning until the sample data of all batches are learned, and obtaining a low-dimensional vector expression of each node in the abnormal graph. The method can realize the learning of complex abnormal composition, has high processing speed and high efficiency, and has higher matching degree of recalled advertisements when being used for searching advertisements.

Description

Method, system and equipment for obtaining relation expression between entities and advertisement recalling system

Technical Field

The invention relates to the technical field of data mining, in particular to a method, a system and equipment for obtaining relationship expression between entities and an advertisement recalling system.

Background

With the popularization of mobile terminals and application software, service providers in the fields of social contact, e-commerce, logistics, travel, take-out, marketing and the like deposit massive business data, and mining the relationship between different business entities (entities) becomes an important technical research direction in the field of data mining based on the massive business data. As machine processing power has increased, more and more technicians have begun to investigate how to mine by machine learning techniques.

The inventors of the present invention found that:

at present, learning mass business data through a machine learning technology to obtain a Graph (Graph) for expressing entities and relationships between the entities, that is, performing Graph learning on the mass business data becomes an optimal technical direction. It is simply understood that a graph is composed of nodes and edges, one node is used to represent one entity, and edges between nodes are used to represent relationships between nodes. A graph will typically include more than two nodes and more than one edge, and thus a graph may also be understood to be composed of a set of nodes and a set of edges, generally represented as: g (V, E), where G represents a graph, V represents a set of nodes in the graph G, and E is a set of edges in the graph G. The graph can be divided into a isomorphic graph and a heterogeneous graph, wherein the heterogeneous graph means that the types of nodes in one graph are different (the types of edges may be the same or different), or the types of edges in one graph are different (the types of nodes may be the same or different). Therefore, when the types of entities are more and need to be expressed by a plurality of types of nodes, or when the relationships between the entities do not only need to be expressed by a plurality of types of edges, it is preferable to express the entities and the relationships between the entities through a heterogeneous graph, and when the magnitude of the nodes and edges included in the heterogeneous graph is large, the heterogeneous graph is extremely complex and the data volume is extremely large, so that reducing the complexity and the data volume of the heterogeneous graph becomes a technical problem for those skilled in the art.

Disclosure of Invention

In view of the above, the present invention has been made to provide a method, system and apparatus, advertisement recall system for obtaining an expression of relationships between entities that overcomes or at least partially solves the above-mentioned problems.

The embodiment of the invention provides an advertisement recall system, which comprises a system for acquiring relation expression between entities and an advertisement recall matching system;

the system for obtaining the relation expression between the entities is used for constructing an abnormal graph for an advertisement search scene, and the node type in the abnormal graph comprises: the type of the edge comprises at least one of a click edge, a co-click edge, a collaborative filtering edge, a content semantic similar edge and an attribute similar edge;

dividing a pre-constructed heterogeneous graph into at least two heterogeneous subgraphs according to a predefined meta-path, wherein the meta-path is used for expressing the structure of the heterogeneous subgraphs and the node types and edge types included in the heterogeneous subgraphs;

acquiring sample data of a batch;

learning a batch of sample data by a preset graph volume model according to the heterogeneous subgraphs to obtain vector expression of nodes in the heterogeneous subgraphs, wherein one graph volume model corresponds to one heterogeneous subgraph;

the preset aggregation model aggregates vector expressions of the same node in different heterogeneous subgraphs based on sample data to obtain the same vector expression of the same node in different heterogeneous subgraphs;

a preset loss function optimizes the parameters of the model based on the sample data and the same vector expression of the same node;

continuously acquiring sample data of the next batch for learning until the sample data of all batches are learned, and obtaining low-dimensional vector expressions of advertisement nodes, commodity nodes and query word nodes in the heterogeneous graph, wherein one node in the heterogeneous graph corresponds to one entity in the sample data;

the advertisement recall matching system is used for determining the matching degree among the query term nodes, the commodity nodes and the search advertisement nodes by using the low-dimensional vector expressions of the query term nodes, the commodity nodes and the search advertisement nodes obtained by the system for obtaining the relation expression among the entities, and selecting the search advertisement of which the matching degree with the commodity and the query term meets the set requirement according to the matching degree.

In an optional embodiment, a meta path corresponds to a heterogeneous subgraph, where the meta path is used to express a structure of the heterogeneous subgraph and node types and edge types included in the heterogeneous subgraph specifically are: the meta path is used for expressing the structure of a heterogeneous subgraph and the node type and the edge type included by the heterogeneous subgraph;

the splitting the heterogeneous graph into at least two heterogeneous subgraphs according to the preset meta-path specifically includes:

and splitting the heterogeneous graph into at least two heterogeneous subgraphs according to at least two preset meta paths.

In an optional embodiment, the system for obtaining relationship expression between entities learns the sample data according to a heterogeneous subgraph by using a preset graph volume model to obtain vector expression of nodes in the heterogeneous subgraph, and specifically includes:

the preset graph volume model obtains vector expression of the nodes in the heterogeneous graph according to the attribute information of each node of the heterogeneous subgraph and the structure information and the attribute information of at least one-order neighbor node of each node in the heterogeneous subgraph.

In an optional embodiment, the system for obtaining relationship expression between entities aggregates vector expression of the same node in different heterogeneous subgraphs based on sample data through a preset aggregation model to obtain the same vector expression of the same node in different heterogeneous subgraphs, and specifically includes:

and the preset aggregation model aggregates vector expressions of the same node in different heterogeneous subgraphs by using attention mechanism aggregation learning or full-connection aggregation learning or weighted average aggregation learning based on the sample data to obtain the same vector expression of the same node in different heterogeneous subgraphs.

In an alternative embodiment, the advertisement recall matching system determines a degree of matching between the query term node, the goods node, and the search advertisement node, including:

converging the low-dimensional vector expression of the query word node and the low-dimensional vector expression of the commodity node pre-clicked by the user under the same query word by using an attention mechanism, a full-connection aggregation mechanism or a weighted average aggregation mechanism to obtain the low-dimensional vector expression of the virtual request node; the virtual request node is a virtual node constructed by a query term node and a commodity node clicked by a user in front under the query term;

and determining the matching degree among the query term nodes, the commodity nodes and the search advertisement nodes according to the low-dimensional vector expression of the virtual request nodes and the low-dimensional vector expression of the search advertisement nodes.

In an optional embodiment, the advertisement recall matching system selects a search advertisement which matches with the goods and the query term according to the matching degree, and includes:

and selecting the search advertisement with the distance meeting the set requirement according to the cosine distance between the low-dimensional vector expression of the virtual request node and the low-dimensional vector expression of the search advertisement node.

The embodiment of the invention also provides a method for obtaining the relation expression between the entities, which comprises the following steps:

acquiring sample data of a batch;

and continuously acquiring sample data of the next batch for learning until the sample data of all batches are learned, and obtaining a low-dimensional vector expression of each node in the heteromorphic image, wherein one node in the heteromorphic image corresponds to one entity in the sample data.

In an optional embodiment, the meta-path is used to express a structure of a heterogeneous subgraph and a node type and an edge type included in the heterogeneous subgraph, and specifically includes:

one element path comprises node types and edge types which are alternately arranged in sequence, wherein the node types are arranged at the first bit and the last bit, and the arrangement sequence of the node types and the edge types expresses the structure of the heterogeneous subgraph;

the splitting the heterogeneous graph into at least two heterogeneous subgraphs according to at least two preset meta-paths specifically includes:

for each meta-path in at least two preset meta-paths, acquiring nodes of corresponding types in the heterogeneous graph according to the node types included in the meta-paths; according to the type of the edge connecting each adjacent node, obtaining the edge meeting the requirement from the heteromorphic graph; and forming the heterogeneous subgraph corresponding to the element path by the acquired nodes of the corresponding types and the edges meeting the requirements.

In an optional embodiment, the learning of the sample data by the preset graph volume model according to the heterogeneous subgraph to obtain the vector expression of the node in the heterogeneous subgraph specifically includes:

and the preset graph volume model learns the sample data according to the attribute information of each node of the heterogeneous subgraph and the structural information and the attribute information of at least one-order neighbor node of each node in the heterogeneous subgraph to obtain the vector expression of the nodes in the heterogeneous subgraph.

In an optional embodiment, the learning, by the preset graph volume model, the sample data according to the attribute information of each node of the heterogeneous subgraph and the structural information and the attribute information of at least one-stage neighbor node of each node in the heterogeneous subgraph to obtain the vector expression of each node in the heterogeneous subgraph specifically includes:

traversing sample data, reading an entity recorded by the sample data according to the currently traversed sample data, and finding a node corresponding to the entity in the heterogeneous graph;

reading first-order to Nth-order neighbor nodes of the node from a heterogeneous subgraph comprising the node, wherein N is a preset positive integer;

and the preset graph volume model carries out N-layer convolution operation according to the attribute information of the node, the attribute information of the first-nth-order neighbor nodes and the structure information to obtain the vector expression of the node.

In an optional embodiment, the learning, by the preset graph volume model, the sample data according to the attribute information of each node of the heterogeneous subgraph and the structural information and the attribute information of at least one-order neighbor node of each node in the heterogeneous subgraph to obtain the vector expression of each node in the heterogeneous graph specifically includes:

sampling neighbor nodes of the same order from the first order to the Nth order of the node according to the weight of edges between the nodes and according to a preset number to obtain the sampled neighbor nodes of the first order to the Nth order;

and the preset graph volume model carries out N-layer convolution operation according to the attribute information of the node, the attribute information and the structure information of the first-Nth-order neighbor nodes after sampling, and obtains the vector expression of the node.

In an optional embodiment, the preset aggregation model aggregates vector expressions of the same node in different heterogeneous subgraphs based on sample data to obtain the same vector expression of the same node in different heterogeneous subgraphs, and specifically includes:

and the preset aggregation model aggregates vector expressions of the same node in different heterogeneous subgraphs by using an attention mechanism or a full-connection aggregation mechanism or a weighted average aggregation mechanism based on the sample data to obtain the same vector expression of the same node in different heterogeneous subgraphs.

The embodiment of the present invention further provides a system for obtaining relationship expression between entities, including: the device comprises a registration device, a storage device, a calculation device and a parameter exchange device;

the storage device is used for storing the data of the heterogeneous subgraph;

the computing device is used for acquiring the data of the heterogeneous subgraph from the storage device through the registration device, and learning the sample data based on the heterogeneous composition by adopting the method for acquiring the relation expression between the entities to obtain the low-dimensional vector expression of each node in the heterogeneous composition;

and the parameter exchange device is used for carrying out parameter interaction with the computing device.

The technical scheme provided by the embodiment of the invention has the beneficial effects that at least:

based on the heterogeneous subgraphs obtained by splitting the heterogeneous graph, a graph convolution model is used for learning sample data, vector expressions of the same nodes obtained by learning of the heterogeneous subgraphs are fused, parameters of a machine learning model are optimized according to fusion results of the vector expressions of the same nodes, the parameters are used for learning of samples of the next batch, iterative learning of the samples is achieved, low-dimensional vector expressions given to the nodes in the heterogeneous graph are finally obtained, data processing amount in the heterogeneous graph learning process is reduced, the problems that training parameters are increased explosively and the number of neighbor nodes are increased along with the exponential level of the number of layers in the heterogeneous graph processing process are solved, and speed and efficiency of the heterogeneous graph learning are improved. The heterogeneous graph learning method is used in an advertisement searching scene, entity relations in the advertisement searching scene are mined to realize accurate advertisement recall by using a large amount of information, the advertisement recall quality is improved, all advertisements are used as candidates, sufficient advertisements can be recalled at any flow rate, and advertisement rewriting and advertisement screening can be completed in one step in a vector mode.

Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention. In the drawings:

FIG. 1 is a flowchart illustrating a method for obtaining a relationship expression between entities according to an embodiment of the present invention;

FIG. 2 is a schematic diagram illustrating a method for obtaining an expression of a relationship between entities according to a second embodiment of the present invention;

FIG. 3 is a flowchart of a method for obtaining relationship expression between entities according to a second embodiment of the present invention;

FIG. 4a is an exemplary diagram of a heteromorphic graph constructed in accordance with a second embodiment of the present invention;

FIG. 4b is a diagram of another example of a heteromorphic image constructed in the second embodiment of the invention;

fig. 5 is an exemplary diagram of splitting a heterogeneous graph into heterogeneous subgraphs in the second embodiment of the present invention;

FIG. 6 is a diagram illustrating an example of a convolutional network model of a heterogeneous subgraph in accordance with a second embodiment of the present invention;

FIG. 7 is a diagram illustrating neighbor node sampling according to a second embodiment of the present invention;

FIG. 8 is a schematic structural diagram of a system for obtaining expression of relationships between entities according to an embodiment of the present invention;

fig. 9 is a schematic structural diagram of an advertisement recall system 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.

In order to solve the problem that the device cannot support operation with large magnitude order because training parameters exponentially increase and neighbor sampling exponentially increase along with the increase of the number of layers during heterogeneous image learning in the prior art, the embodiment of the invention provides a method, which can well solve the problem, effectively reduce data acquisition entity-to-entity relation expression throughput during heterogeneous image learning, and has high processing speed and high efficiency.

Graph learning has wide application in mining various data relationships in the real-world field, such as mining the correlation between a search request and an advertisement and the Click-Through Rate (CTR) in a search advertisement platform. Namely, the method can be used in the field of advertisement searching and is used for searching for recalls of advertisements. The search advertisement refers to an advertisement which is determined by an advertiser according to the content, characteristics and the like of the product or service of the advertiser, writes advertisement content, and puts the advertisement in a search result corresponding to the keyword in an independent pricing mode. Search advertisement recall refers to the selection of the most relevant advertisement from a large collection of advertisements through some algorithm or model.

The existing search advertisement recall technology screens the advertisement with high quality based on the matching degree of the query word and the advertiser bidding word (bid), the price of the advertiser bidding word and the statistical preference of the user to the advertisement; or adding the historical behavior data of each user to carry out personalized matching recall on the advertisement.

The inventor finds that the existing recall technology only emphasizes the matching degree of the advertisements and the query words or only emphasizes the improvement of the return advertisements in the prior art research, and an integration model is lacked to take both the advertisements and the query words into consideration. Because the quality of the advertisement recall is high and low and is important for searching advertisement income and user experience, the inventor provides a graph learning technology which is used for acquiring relationship expression between entities in the advertisement recall process and can obtain more advertisement recall sets with high quality and more interest of users.

The following describes in detail a method and system for obtaining relationship expression between entities and a specific implementation manner for an advertisement recall system by using specific embodiments.

Example one

An embodiment of the present invention provides a method for obtaining relationship expression between entities, where a flow of the method is shown in fig. 1, and the method includes the following steps:

step S101: dividing a pre-constructed heterogeneous graph into at least two heterogeneous subgraphs according to a predefined meta-path, wherein the meta-path is used for expressing the structure of the heterogeneous subgraph and the node type and the edge type included by the heterogeneous subgraph.

One meta-path corresponds to one heterogeneous subgraph, and the meta-path is used for expressing the structure of the heterogeneous subgraph and the node types and the edge types included in the heterogeneous subgraph specifically are as follows: a meta-path is used to express the structure of a heterogeneous subgraph and the node types and edge types included in the heterogeneous subgraph. Specifically, one element path includes node types and edge types which are alternately arranged in sequence, wherein the node types are arranged at the first bit and the last bit, and the arrangement sequence of the node types and the edge types expresses the structure of the heterogeneous subgraph.

Splitting the heterogeneous graph into at least two heterogeneous subgraphs according to a preset meta-path specifically comprises splitting the heterogeneous graph into at least two heterogeneous subgraphs according to at least two preset meta-paths. Specifically, for each meta-path of at least two preset meta-paths, a node of a corresponding type in the heterogeneous graph is obtained according to the node type included in the meta-path; according to the type of the edge connecting each adjacent node, obtaining the edge meeting the requirement from the heteromorphic graph; and forming the heterogeneous subgraph corresponding to the element path by the acquired nodes of the corresponding types and the edges meeting the requirements.

Step S102: sample data for a batch is obtained.

The sample data can be divided into a plurality of batches, and learning is carried out in batches based on heterogeneous subgraphs.

Step S103: the preset graph volume model learns sample data (or sample data set) of a batch according to the heterogeneous subgraph to obtain vector expression of nodes in the heterogeneous subgraph, and one graph volume model corresponds to one heterogeneous subgraph.

In the step, a preset graph volume model learns sample data according to the attribute information of each node of the heterogeneous subgraph and the structural information and the attribute information of at least one-stage neighbor node of each node in the heterogeneous subgraph to obtain vector expression of the nodes in the heterogeneous subgraph. There may be two alternative cases:

the method for learning the sample data based on all nodes in the heterogeneous subgraph comprises the following steps:

reading first-order to Nth-order neighbor nodes of the node from the heterogeneous subgraph comprising the node, wherein N is a preset positive integer;

and the preset graph volume model carries out N-layer convolution operation according to the attribute information of the node, the attribute information and the structure information of the first-nth-order neighbor nodes to obtain the vector expression of the node.

The method for learning the sample data of the sampling result of the node in the heterogeneous subgraph comprises the following steps:

Step S104: and the preset aggregation model aggregates vector expressions of the same node in different heterogeneous subgraphs based on sample data to obtain the same vector expression of the same node in different heterogeneous subgraphs.

And the preset aggregation model is based on sample data, and the vector expressions of the same node in different heterogeneous subgraphs are aggregated by using attention mechanism aggregation learning or full-connection aggregation learning or weighted average aggregation learning to obtain the same vector expression of the same node in different heterogeneous subgraphs.

Step S105: and the preset loss function optimizes the parameters of the model based on the sample data and the same vector expression of the same node.

After the same vector expression of the same node in different heterogeneous subgraphs is obtained, the vector expressions of the same node of at least two types are used for converging to obtain the low-dimensional vector expression of the virtual request node; the virtual request node is a virtual node constructed by at least two types of nodes with a certain incidence relation; and determining the association parameters among the nodes of at least three types according to the low-dimensional vector expression of the virtual request node and the low-dimensional vector expression of the node of another type, and optimizing the model parameters according to the association parameters.

Step S106: whether the sample data of all batches is acquired or not is finished, if not, the step 107 is executed; if yes, go to step S108.

Step 107: and acquiring sample data of the next batch, and returning to execute the step S103.

Therefore, the sample data of the next batch can be continuously acquired for learning until the sample data of all batches are completely learned.

Step S108: and obtaining a low-dimensional vector expression of each node in the heteromorphic graph. One node in the anomaly map corresponds to one entity in the sample data.

And after learning of all batches of samples is finished, obtaining a low-dimensional vector expression of each node in the composition, wherein the low-dimensional vector expression of each node in the composition is the same vector expression of the same node in different heterogeneous subgraphs obtained in the aggregation model after the last batch of samples are learned.

After the samples of all batches are learned, the matching degree between the nodes of different types can be obtained, and the matching degree is the correlation parameter between the nodes obtained by the loss function at the last time.

In the method of this embodiment, based on the heterogeneous subgraph after the heterogeneous graph is split, a machine learning model is used to learn sample data, vector expressions of the same nodes obtained by learning of each heterogeneous subgraph are fused, parameters of the machine learning model are optimized according to the fusion result of the vector expressions of the same nodes, the parameters are used for learning of samples of the next batch, iterative learning of the samples is achieved, and low-dimensional vector expressions for the nodes in the heterogeneous graph are finally obtained, so that the data processing amount in the heterogeneous graph learning process is reduced, the problems of explosive increase of training parameters and exponential increase of the number of neighbor nodes along with the number of layers in the heterogeneous graph processing process are avoided, and the speed and efficiency of the heterogeneous graph learning are improved.

Example two

A second embodiment of the present invention provides a specific implementation process of a method for obtaining relationship expression between entities, which is described by taking a process of implementing advertisement recall in an advertisement search scene as an example, an implementation principle of the method is shown in fig. 2, and a flow of the method is shown in fig. 3, and the method includes the following steps:

step S301: and constructing the abnormal pattern.

Taking an advertisement search scene as an example, according to a user log and related goods and advertisement data, a large-scale heterogeneous graph is constructed for a search recall scene, and is used as a rich search interaction graph of the advertisement search scene, and the constructed heterogeneous graph is used as subsequent graph data input, for example, graph data of the lowest heterogeneous graph in fig. 2.

An example of the constructed heterogeneous graph is shown in fig. 4a, the heterogeneous graph includes multiple types of nodes such as Query, Item, Ad, and the like to represent different entities in a search scene, and the heterogeneous graph includes multiple types of edges to represent multiple relationships between the entities. Wherein, the node type and its specific meaning can be as shown in the following table 1, and the edge type and its meaning can be as shown in the following table 2.

TABLE 1

Node type	Detailed description of the invention
		Item	All goods under advertisement search scenario
Ad	Search advertisement in advertisement search scenario
		Query	User query word in advertisement search scene

The Query node and the Item node are used as user intention nodes to depict the personalized search intention of the user, and the Ad node is an advertisement delivered by an advertiser.

TABLE 2

Wherein:

the user behavior edge represents the historical behavior preference of the user, for example, a click edge (click edge) can be established between a Query node and an Item node or between the Query node and an Ad node, and the click times are used as edge weights to represent the click between the Query node and the Item/Ad; for another example, a co-click edge (session edge) may be established, which represents an Item or Ad co-clicked with Query in the same session; as another example, collaborative filtering edges (cf edges) may also be established to represent collaborative filtering relationships between different nodes. In the advertisement searching scene, the user behavior describes a dynamically changing relationship. The hot nodes (such as the nodes of the high-frequency Query) have more displays and clicks, and further have denser edge relationships and larger edge weights, while the cold nodes and the new nodes have relatively sparse variable relationships and smaller edge weights, so that the user behavior edges can better depict the hot nodes.

Content similarity edges (semantic edges) are used for similarity between client nodes, for example: edges are created between Item nodes, using the text similarity of their titles as a weighting factor. The content similarity edge reflects a static relation between the nodes, is more stable, and can well depict the relation between the cold node and the new node.

Attribute-like edges (domain edges) represent overlapping contributions of domains between nodes, such as domains of brand, category, and the like.

FIG. 4b is a representation of a constructed heteromorphic graph, wherein nodes of the same shape represent nodes of the same type and edges of the same line shape represent edges of the same type.

Step S302: and dividing the constructed heterogeneous graph into at least two heterogeneous subgraphs according to a preset meta-path. The meta-path is used for expressing the structure of the heterogeneous subgraph and the node type and the edge type included in the heterogeneous subgraph.

The graph data to be learned in the application is a heterogeneous graph, and there may be many types of points and edges, the current graph convolutional neural network (GCN) is only suitable for homogeneous graphs, and learning by directly considering the heterogeneous graph as a homogeneous graph using the graph convolutional neural network cannot obtain effective low-dimensional vector expression. Therefore, in order to realize the learning of the heterogeneous graph, some significant meta-paths (meta-paths) are defined so as to divide the original large heterogeneous graph into a plurality of significant heterogeneous subgraphs for learning.

Taking an advertisement search scenario as an example, the meta path defined may be as shown in table 3 below.

TABLE 3

Numbering	Meta path
		a	Item/Ad node-co-click edge-Item/Ad node-attribute similarity edge-Item/Ad node
b	Item/Ad node-click edge-Query node-click edge-Item/Ad node
		c	Query node-click edge-Item/Ad node-co-click edge-Item/Ad node
d	Query node-collaborative filtering edge-Query node-semantic similar edge-Query node
		e	Query node-collaborative filtering edge-Item/Ad node
f	Query node-click edge-Item/Ad node-collaborative filtering edge-Query node

Based on the defined meta-path, the constructed heterogeneous graph is split, as shown in fig. 5, the heterogeneous graph shown in fig. 4b is split, taking 6 meta-paths as an example, and six heterogeneous subgraphs, i.e., subgraph a, subgraph b, subgraph c, subgraph d, subgraph e and subgraph f, can be split for meta-path a, meta-path b, meta-path c, meta-path d, meta-path e and meta-path f.

Taking meta path a as an example: and the meta path a comprises a node Item/Ad-co-click edge-node Item/Ad-attribute similar edge-node Item/Ad, when the subgraph a is constructed according to the meta path a, the node (Item and Ad) comprising the corresponding node type in the meta path a is obtained from the constructed heterogeneous graph, and the edge meeting the requirement is reserved to obtain the subgraph a. The construction of the heterogeneous subgraphs corresponding to other meta-paths is similar to the meta-path a, and is not described in detail here.

Referring to fig. 2, the bottom is a constructed singular map, and based on the singular map, an initial vector representation of each node is formed based on the characteristics of each node. For each designated node, defining a meta-path containing the designated node, and constructing a heterogeneous subgraph based on the defined meta-path, as shown in fig. 2, defining two meta-paths for searching advertisement nodes (Ad), and correspondingly splitting the two heterogeneous subgraphs; defining four element paths aiming at Query word nodes (Query), and correspondingly splitting four heterogeneous subgraphs; for k commodity (Item) nodes of 1, 2, … …, k and the like, each commodity node defines two meta paths, and two heterogeneous subgraphs are correspondingly split.

Step S303: sample data for a batch is obtained.

Sample data relevant to the advertisement search is extracted from the user log data. The sample data may be from a user historical behavior log, a merchandise basis attribute information table, an advertisement basis attribute information table, a query term basis attribute information table, and the like.

The extraction may be performed in multiple batches. And sequentially inputting the sample data of each batch into the machine learning model for training and learning, wherein the learning result of the previous batch can optimize the parameters of the model, and the optimized parameters are used for learning the sample data of the next batch, so that the iterative learning effect is achieved, and the final learning result is obtained.

Step S304: the preset graph volume model learns sample data of a batch according to the heterogeneous subgraphs to obtain vector expressions of nodes in the heterogeneous subgraphs, and one graph volume model corresponds to one heterogeneous subgraph.

When the sample data is learned based on the heterogeneous subgraph, a preset graph volume model is used, and the sample data is learned according to the attribute information of each node of the heterogeneous subgraph and the structural information and the attribute information of at least one-order neighbor nodes of each node in the heterogeneous subgraph to obtain the vector expression of the nodes in the heterogeneous subgraph.

Referring to fig. 2, each heterogeneous subgraph corresponds to a graph volume network model, for example, the leftmost group of two graph volume network models in fig. 2 corresponds to two split heterogeneous subgraphs of two meta-paths defined by search advertisement nodes (Ad), and the second group of four graph volume network models on the left corresponds to four split heterogeneous subgraphs of four meta-paths defined by Query nodes (Query); and the right 1, … … and k groups of graph convolution network models, wherein two graph convolution network models in each group respectively correspond to two heterogeneous subgraphs split by two meta-paths defined by one commodity (Item) node. And when the sample data is learned based on each heterogeneous subgraph, the sample data is used as input and corresponds to corresponding nodes in the heterogeneous subgraphs for learning. Ownership may be shared between the various convolutional network models shown in fig. 2.

Taking a heterogeneous subgraph as an example, traversing sample data, reading an entity recorded by the sample data for the currently traversed sample data, and finding a node corresponding to the entity in the heterogeneous subgraph; reading first-order to Nth-order neighbor nodes of the node from the heterogeneous subgraph comprising the node, wherein N is a preset positive integer; and the preset graph volume model carries out N-layer convolution operation according to the attribute information of the node, the attribute information and the structure information of the first-Nth-order neighbor nodes after sampling, and obtains the vector expression of the node.

The N-layer convolution operation specifically includes: acquiring an N-order neighbor node of a node in the heterogeneous subgraph, performing convolution operation in a layered mode, performing convolution operation on vector expressions of the N-order neighbor nodes connected with the N-1-order neighbor node for the N-1-order neighbor node to obtain neighbor low-dimensional vector expressions of the N-1-order neighbor node, and performing combined operation on the neighbor low-dimensional vector expressions of the N-1-order neighbor node and original low-dimensional vector expressions of the N-1-order neighbor node to obtain new low-dimensional vector expressions of the N-1-order neighbor node; by analogy, … …, performing convolution operation on the vector expression of the second-order neighbor node connected with the first-order neighbor node to obtain the neighbor low-dimensional vector expression of the first-order neighbor node, and performing combined operation on the neighbor low-dimensional vector expression of the first-order neighbor node and the original low-dimensional vector expression of the first-order neighbor node to obtain a new low-dimensional vector expression of the first-order neighbor node; performing convolution operation on the low-dimensional vector expression of each first-order neighbor node of the node to obtain the neighbor low-dimensional vector expression of the node, and performing combined operation on the neighbor low-dimensional vector expression of the node and the original low-dimensional vector expression of the node to obtain a new neighbor low-dimensional vector expression of the node.

Referring to fig. 6, a principle of learning sample data based on a heterogeneous subgraph, taking a subgraph a corresponding to a meta-path a as an example, for a node 1, a graph convolution network can be constructed as shown in fig. 6. For the convenience of observation, the two-layer convolution structure is taken as an example, and the practical extension can be to multiple layers. As shown in fig. 6, the first-order neighbor nodes of the node 1 in the sub-graph a have 2, 3, 4, and 6, and the second-order neighbor nodes have 1, 2, 3, 4, and 10. Second-

order neighbor nodes

1, 2, 3, 4 and 10 of the node 1 in the subgraph are subjected to graph convolution to obtain neighbor low-dimensional vector expressions of first-

order neighbor nodes

2, 3, 4 and 6, the neighbor low-dimensional vector expressions are spliced with the low-dimensional vector expressions of the

nodes

2, 3, 4 and 6 and subjected to nonlinear conversion to obtain final low-dimensional vector expressions of the

nodes

2, 3, 4 and 6, the final low-dimensional vector expressions are used as input and are subjected to graph convolution, the original low-dimensional vector expression of the node 1 is spliced, and the final low-dimensional vector expression of the second-order graph convolution network of the node 1 is obtained through conversion.

In the heterogeneous subgraph a, the final low-dimensional vector expression of other nodes is obtained in a manner similar to that of the node 1, and details are not repeated here. The isolated node 8 without a neighbor node retains the original vector representation. Based on a similar manner, the final low-dimensional vector expression of each node in each heterogeneous subgraph can be obtained.

The metapath (meta-path) -based graph volume advertisement recall scheme can effectively solve an advertisement recall scene by using a graph volume method, but still has a problem of computational complexity. Taking the heterogeneous subgraph shown in fig. 6 as an example, the number of neighbor nodes of a node increases exponentially with the increase of the number of graph convolution layers, and there are 3 first-order neighbors and 9 second-order neighbors of the node 1. In a real scene, the first-order neighbor nodes of the real nodes may be thousands of nodes, and the convolution result of directly calculating massive nodes can hardly be realized along with the increment of the number of layers. Thus, hierarchical neighbors may be sampled based on beam-search, with the neighbor space complexity being represented by O (n)^k) To O (kn).

In an optional embodiment, when the sample data is learned based on the heterogeneous subgraph, when there are many nodes in the heterogeneous subgraph, the neighbor nodes may be sampled, and convolution calculation is performed based on the sampled neighbor nodes. Taking a heterogeneous subgraph as an example, traversing sample data, reading an entity recorded by the sample data for the currently traversed sample data, and finding a node corresponding to the entity in the heterogeneous subgraph; reading first-order to Nth-order neighbor nodes of the node from the heterogeneous subgraph comprising the node, wherein N is a preset positive integer; sampling neighbor nodes of the same order from the first order to the Nth order of the node according to the weight of edges between the nodes and according to a preset number to obtain the sampled neighbor nodes of the first order to the Nth order; and the preset graph volume model carries out N-layer convolution operation according to the attribute information of the node, the attribute information and the structure information of the first-Nth-order neighbor nodes after sampling, and obtains the vector expression of the node.

Referring to the sub-differential graph shown in fig. 6, taking a two-layer convolution structure as an example, taking the sum of the edge weights of the neighboring nodes as a weight, and performing neighbor weighted sampling on the nodes. The principle of sampling based on edge weights is shown in fig. 7, the original convolution structure of the node 1 is shown in the left graph in fig. 7, and the weight of each edge is shown in the labeled number of each edge in the graph. If k is 2, namely, only two neighbor nodes are selected for convolution operation in each layer, for the node 1, the probability that the first-

order neighbor node

2 or 4 is selected is higher, because the weights of the node 2 or the node 4 are 3 and 4; if the first-order neighbor node selects the

node

2 or 4, the probability that the second-order neighbor node selects the

node

1 or 10 is high, because only the edge weights connected to the first-

order neighbor node

2 or 4 are calculated as the weights at this time, the weight of the node 1 is 3+4 — 7, the weight of the node 10 is 7, and the sampled probability is highest.

When sampling is carried out on the basis of the edge weight, k nodes with the highest weight are selected according to the weight, in order to prevent the result of neighbor sampling from excessively deviating to a small number of hot nodes, weighted sampling can be carried out on the basis of the weight w of the nodes to obtain k sampling nodes, and the weight w can be expressed as:

wherein the content of the first and second substances,

an edge weight representing the edge e is represented,

represents the current weight of node v at level L, and J represents the current weight of node v_iThe number of upper level nodes with edges, l represents the l-th level, and i and j are the serial numbers of the designated nodes.

The sampling of the layer node can reduce the growth trend of the neighbor node from exponential level to linear level on the basis of considering all connection relations of the neighbor node of the upper layer.

Step S305: and performing aggregation learning on the sample data by a preset aggregation model according to the vector expression of the nodes in the heterogeneous subgraph to obtain the same vector expression of the same node in different heterogeneous subgraphs.

The same node may exist in different heterogeneous subgraphs, for example, if the node 1 exists in subgraphs a, b, c, e and f, different convolutional neural networks of the heterogeneous subgraphs can obtain different node vector expressions, and the same node only needs one unique low-dimensional vector expression to perform subsequent recall work. Therefore, an attention mechanism or a full-connection aggregation mechanism or a weighted average aggregation mechanism is selected to aggregate vector expressions of the same node in different heterogeneous subgraphs to obtain the same vector expression of the same node in different heterogeneous subgraphs, namely, an aggregation weighting result is used as a final node low-dimensional vector expression (embedding) result.

The process of converging vector expressions of the same node in different heterogeneous graphs comprises the following steps:

calculating the weight of the vector expression of the node in each heterogeneous subgraph according to the vector expression of the node in each heterogeneous subgraph and the corresponding learning weight factor; taking the attention mechanism as an example, the weights are calculated

The formula of (1) is as follows:

wherein the content of the first and second substances,

representing multiple different vector representations of the same node derived from a heterogeneous graph,

representing a learning weight factor.

Using the calculated weight pairsCarrying out weighted summation on the vector expression of the node in each heterogeneous subgraph to obtain the aggregated low-dimensional vector expression of the node

Wherein, assuming that the type of the node v is p, then

A set of meta-paths (metapath) of type p representing the node at level L.

After adding neighbor samples, the convolution model can be adjusted as follows:

wherein the content of the first and second substances,

the node low-dimensional vector representation of the 0 th layer is shown,

low-dimensional vector representation of node v neighbor aggregation representing the l-th layer, WEIGHTEDMEAN representing weighted average, N representing satisfaction of metapath s^kThe neighborhood of node v, w represents the weight in the weighted average, CONCAT represents the direct concatenation of the two vectors,

representing nodes v of layer lAnd aggregating low-dimensional vector expressions of self information and neighbor information, wherein W represents the weight to be learned, and sigma represents nonlinear transformation.

Step S306: and the preset loss function optimizes the parameters of the model based on the sample data and the same vector expression of the same node.

Taking an advertisement search scene as an example, based on the above steps, low-dimensional vector expressions of advertisements, commodities and query words can be obtained. According to the extracted sample data, in order to implement personalized search recall, the current query word of user and advertisement or commodity clicked by user before are used as current search request of user, at the same time, the attention mechanism is used to express low-dimensional vector of query word (H)_Q) And low-dimensional vector representation (H) of multiple pre-clicks_1k、……、H_Ik) Aggregated into a final user search request vector. Vector (H) of user search requests_r) Low dimensional vector representation (H) with current advertisement_ad) Calculating cosine (cosine) distance, using click status as tag data (O)_label) And calculating the logistic regression (sigmoid) cross entropy as the final loss function of the model, and training the whole model.

Taking advertisement search as an example, when constructing a sample, regarding a clicked advertisement under a current request as a positive example, regarding an un-clicked advertisement as a negative example, obtaining a sample structure as follows: (request, ad, click-label), including request, search for ads, and click-through tags. Wherein, the request is (query, { realtimeverified items \ ads }_k) Including search ads and multiple real click goods.

Using sigmoid cross entropy as a loss function, the optimization objective of the entire model is expressed as:

wherein, y_iRepresenting tag data, p_iRepresenting the prior probability, v_request、v_adVector expressions, R (v), representing virtual requesting nodes and advertising nodes, respectively_request,v_ad) A distance metric function between vector representations representing the virtual requesting node and the advertising node.

Step S307: whether the sample data of all batches is acquired, if not, executing step 308; if yes, go to step S309.

Step S308: and acquiring sample data of the next batch, and returning to execute the step S204.

Step S309: and obtaining a low-dimensional vector expression of each node in the heteromorphic graph. One node in the anomaly map corresponds to one entity in the sample data.

And repeatedly training the sample data of all batches for preset times to obtain a low-dimensional vector expression of each node in the abnormal graph, wherein one node in the abnormal graph corresponds to one entity in the sample data.

The minimum edge is a schematic of an anomaly as shown in the system principle of fig. 2. Initializing the four rows of white small squares on the upper layer for the node vector in the heterogeneous graph to obtain the initial vector expression of each node, inputting the initial vector expression into the learning model corresponding to each heterogeneous subgraph, learning a batch of sample data through the learning model, updating the vector expression of each node in the heterogeneous subgraph according to the learning result, converging the vector expression of the same node in each heterogeneous subgraph to obtain a converged vector expression of the same node, such as the converged vector expression in fig. 2

To search the aggregated vector representation of the advertising node,

for the aggregated vector representation of the query term nodes,

and expressing the converged vectors of each commodity node. To pair

Is carried out to obtain

By

And

obtaining a loss function O_labelUsing O_labelOptimizing system parameters of each model, learning sample data of the next batch by using the system model after parameter optimization, updating vector expressions of each node in the heterogeneous subgraphs according to learning results, converging vector expressions of the same node in the heterogeneous subgraphs to obtain a converged vector expression of the same node, and further obtaining a new loss function O according to converging results_labelAnd after optimizing and updating the model parameters, continuing to use the sample data of the next batch until the sample data of all batches are completely learned, and obtaining a final vector expression of each node in the heterogeneous graph.

Based on the same inventive concept, the embodiment of the invention also provides a system for acquiring the relation expression between the entities, and the system can be arranged in network equipment, cloud equipment of a cloud end or server equipment, user side equipment and other equipment of the architecture in the network. The structure of the system is shown in fig. 8, and comprises: registration means 803, storage means 801, calculation means 802 and parameter exchange means 804.

A storage device 801, configured to store data of the heterogeneous subgraph;

the computing device 802 is configured to obtain data of the heterogeneous subgraph from the storage device 801 through the registration device 803, and learn sample data based on the heterogeneous graph by using the above method for obtaining the relationship expression between entities to obtain the low-dimensional vector expression of each node in the heterogeneous graph.

And a parameter exchanging device 804, configured to perform parameter interaction with the computing device.

The computing device 802 obtains data of each node and edge from the storage device through the registration device 803, and includes:

the computing device 802 sends a data query request to the registration device 803, wherein the data query request includes information of the heterogeneous subgraph to be queried; receiving a query result returned by the registration device 803, wherein the query result includes storage device information for storing the heterogeneous subgraph data; and acquiring data of the heterogeneous subgraph from the corresponding storage device 801 according to the storage device information.

Optionally, the storage device 801 may further store data and sample data of each node and edge in the abnormal graph.

The computing device 802 sends a data query request to the registration device 803, where the data query request includes information of nodes and edges to be queried; receiving a query result returned by the registration device 803, wherein the query result includes storage device information of data of storage nodes and edges; the data of each node and edge is acquired from the corresponding storage device 801 according to the storage device information.

Based on the same inventive concept, an advertisement recall system is further provided in the embodiments of the present invention, as shown in fig. 9, including a system 901 for obtaining relationship expression between entities and an advertisement recall matching system 902;

a system 901 for obtaining relationship expression between entities, configured to construct an abnormal graph for an advertisement search scenario, where the node types in the abnormal graph include: the type of the edge comprises at least one of a click edge, a co-click edge, a collaborative filtering edge, a content semantic similar edge and an attribute similar edge;

acquiring sample data of a batch;

and an advertisement recall matching system 902, configured to determine matching degrees among the query term nodes, the commodity nodes, and the search advertisement nodes by using the low-dimensional vector expressions of the query term nodes, the commodity nodes, and the search advertisement nodes obtained by the system for obtaining relationship expression among entities, and select a search advertisement whose matching degree with the commodity and the query term meets a set requirement according to the matching degrees.

In the system, meta-paths defined by the system for expressing relationships among entities are obtained, one meta-path corresponds to one heterogeneous subgraph, and the meta-path is used for expressing the structure of the heterogeneous subgraph and node types and edge types included in the heterogeneous subgraph specifically: the meta path is used for expressing the structure of a heterogeneous subgraph and the node type and the edge type included by the heterogeneous subgraph; the method specifically comprises the following steps: one element path comprises node types and edge types which are alternately arranged in sequence, wherein the node types are arranged at the first bit and the last bit, and the arrangement sequence of the node types and the edge types expresses the structure of the heterogeneous subgraph;

optionally, the step of splitting the heterogeneous graph into at least two heterogeneous subgraphs according to a preset meta-path by the system for obtaining the relationship expression between the entities specifically includes: splitting a heterogeneous graph into at least two heterogeneous subgraphs according to at least two preset meta-paths, specifically, for each meta-path of the at least two preset meta-paths, obtaining nodes of corresponding types in the heterogeneous graph according to the fact that the meta-path comprises node types; according to the type of the edge connecting each adjacent node, obtaining the edge meeting the requirement from the heteromorphic graph; and forming the heterogeneous subgraph corresponding to the element path by the acquired nodes of the corresponding types and the edges meeting the requirements.

Optionally, the system for obtaining the relationship expression between the entities learns the sample data according to the heterogeneous subgraph through a preset graph volume model to obtain the vector expression of the nodes in the heterogeneous subgraph, and specifically includes: the preset graph volume model obtains vector expression of the nodes in the heterogeneous graph according to the attribute information of each node of the heterogeneous subgraph and the structure information and the attribute information of at least one-order neighbor node of each node in the heterogeneous subgraph.

Optionally, the system for obtaining the relationship expression between entities obtains the vector expression of each node in the heterogeneous graph according to the attribute information of each node in the heterogeneous subgraph and the structural information and attribute information of at least one-stage neighbor node of each node in the heterogeneous subgraph through a preset graph volume model, and specifically includes:

Optionally, the system for obtaining the relationship expression between the entities aggregates the vector expressions of the same node in different heterogeneous subgraphs based on sample data through a preset aggregation model to obtain the same vector expression of the same node in different heterogeneous subgraphs, and specifically includes:

Optionally, the advertisement recall matching system determines the matching degree between the query term node, the commodity node and the search advertisement node, and includes:

Optionally, the advertisement recall matching system selects a search advertisement whose matching degree with the goods and the query term meets the set requirement according to the matching degree, and includes:

and selecting the search advertisement with the distance meeting the set requirement according to the cosine distance between the low-dimensional fusion information vector of the virtual request node and the low-dimensional fusion information vector of the search advertisement node.

The embodiment of the present invention further provides a computer-readable storage medium, on which computer instructions are stored, where the instructions, when executed by a processor, implement the method for obtaining a relationship expression between entities described above.

The embodiment of the present invention further provides a heterogeneous image learning apparatus, including: the system comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the program to realize the method for acquiring the relation expression between the entities.

With regard to the system in the above-described embodiment, the specific manner in which each device or module performs the operation has been described in detail in the embodiment related to the method, and will not be described in detail herein.

Unless specifically stated otherwise, terms such as processing, computing, calculating, determining, displaying, or the like, may refer to an action and/or process of one or more processing or computing systems or similar devices that manipulates and transforms data represented as physical (e.g., electronic) quantities within the processing system's registers and memories into other data similarly represented as physical quantities within the processing system's memories, registers or other such information storage, transmission or display devices. Information and signals may be represented using any of a variety of different technologies and techniques. For example, data, instructions, commands, information, signals, bits, symbols, and chips that may be referenced throughout the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or particles, optical fields or particles, or any combination thereof.

It should be understood that the specific order or hierarchy of steps in the processes disclosed is an example of exemplary approaches. Based upon design preferences, it is understood that the specific order or hierarchy of steps in the processes may be rearranged without departing from the scope of the present disclosure. The accompanying method claims present elements of the various steps in a sample order, and are not intended to be limited to the specific order or hierarchy presented.

In the foregoing detailed description, various features are grouped together in a single embodiment for the purpose of streamlining the disclosure. This method of disclosure is not to be interpreted as reflecting an intention that the claimed embodiments of the subject matter require more features than are expressly recited in each claim. Rather, as the following claims reflect, invention lies in less than all features of a single disclosed embodiment. Thus, the following claims are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

Those of skill would further appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. 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 disclosure.

The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium. Of course, the storage medium may also be integral to the processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. Of course, the processor and the storage medium may reside as discrete components in a user terminal.

For a software implementation, the techniques described herein may be implemented with modules (e.g., procedures, functions, and so on) that perform the functions described herein. The software codes may be stored in memory units and executed by processors. The memory unit may be implemented within the processor or external to the processor, in which case it can be communicatively coupled to the processor via various means as is known in the art.

What has been described above includes examples of one or more embodiments. It is, of course, not possible to describe every conceivable combination of components or methodologies for purposes of describing the aforementioned embodiments, but one of ordinary skill in the art may recognize that many further combinations and permutations of various embodiments are possible. Accordingly, the embodiments described herein are intended to embrace all such alterations, modifications and variations that fall within the scope of the appended claims. Furthermore, to the extent that the term "includes" is used in either the detailed description or the claims, such term is intended to be inclusive in a manner similar to the term "comprising" as "comprising" is interpreted when employed as a transitional word in a claim. Furthermore, any use of the term "or" in the specification of the claims is intended to mean a "non-exclusive or".

Claims

1. An advertisement recall system comprises a system for acquiring relational expression between entities and an advertisement recall matching system;

acquiring sample data of a batch;

2. The system of claim 1, wherein a meta-path corresponds to a heterogeneous subgraph, and the meta-path is used for expressing a structure of the heterogeneous subgraph and node types and edge types included in the heterogeneous subgraph specifically: the meta path is used for expressing the structure of a heterogeneous subgraph and the node type and the edge type included by the heterogeneous subgraph;

3. The system according to claim 1, wherein the system for obtaining the relationship expression between the entities learns the sample data according to the heterogeneous subgraph through a preset graph volume model to obtain the vector expression of the nodes in the heterogeneous subgraph, and specifically comprises:

4. The system according to claim 1, wherein the system for obtaining the relationship expression between the entities aggregates vector expressions of the same node in different heterogeneous subgraphs based on sample data through a preset aggregation model to obtain the same vector expression of the same node in different heterogeneous subgraphs, and specifically comprises:

5. The system of claim 1, wherein the advertisement recall matching system determines a degree of match between query term nodes, commodity nodes, and search advertisement nodes, comprising:

6. The system of claim 5, wherein the advertisement recall matching system selects a search advertisement matching a product or a query term according to the matching degree, and comprises:

7. A method for obtaining relational expressions between entities, comprising:

acquiring sample data of a batch;

8. The method of claim 7, wherein a meta-path corresponds to a heterogeneous subgraph, and the meta-path is used for expressing a structure of the heterogeneous subgraph and node types and edge types included in the heterogeneous subgraph specifically: the meta path is used for expressing the structure of a heterogeneous subgraph and the node type and the edge type included by the heterogeneous subgraph;

9. The method according to claim 8, wherein the meta-path is used for expressing a structure of a heterogeneous subgraph and a node type and an edge type included in the heterogeneous subgraph, and specifically comprises:

10. The method according to claim 7, wherein a preset graph volume model learns the sample data according to a heterogeneous subgraph to obtain vector expressions of nodes in the heterogeneous subgraph, and specifically comprises:

11. The method according to claim 10, wherein the learning of the sample data according to the preset graph volume model based on the attribute information of each node in the heterogeneous subgraph and the structural information and the attribute information of at least one-order neighbor node of each node in the heterogeneous subgraph to obtain the vector expression of each node in the heterogeneous subgraph specifically comprises:

12. The method according to claim 10, wherein the preset graph volume model learns the sample data according to the attribute information of each node of the heterogeneous subgraph and the structural information and the attribute information of at least one-order neighbor node of each node in the heterogeneous subgraph to obtain the vector expression of each node in the heterogeneous subgraph, and specifically comprises:

13. The method according to claim 7, wherein the preset aggregation model aggregates vector expressions of the same node in different heterogeneous subgraphs based on sample data to obtain the same vector expression of the same node in different heterogeneous subgraphs, and specifically comprises:

14. A system for obtaining relational expressions between entities, comprising: the device comprises a registration device, a storage device, a calculation device and a parameter exchange device;

the storage device is used for storing the data of the heterogeneous subgraph;

the computing device is used for acquiring data of the heterogeneous subgraph from the storage device through the registration device, and learning sample data based on the heterogeneous graph by adopting the method for acquiring the relation expression between the entities according to any one of claims 7 to 13 to obtain the low-dimensional vector expression of each node in the heterogeneous graph;