CN111459990B

CN111459990B - Object processing method, system, computer readable storage medium and computer device

Info

Publication number: CN111459990B
Application number: CN202010243704.7A
Authority: CN
Inventors: 毛铁峥; 李子健; 赵子元; 颜强
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd; Guangzhou Tencent Technology Co Ltd
Priority date: 2020-03-31
Filing date: 2020-03-31
Publication date: 2021-07-06
Anticipated expiration: 2040-03-31
Also published as: CN111459990A

Abstract

The embodiment of the invention discloses an object processing method, an object processing system, a computer readable storage medium and computer equipment, which are applied to the technical field of information processing based on artificial intelligence. Dividing a plurality of objects into a plurality of object sets according to object behavior information applied to a target function, respectively extracting overall characteristic information of each object set, when at least two object sets in the object sets have a common object, updating initial characteristic information of the common object of any one object set in the at least two object sets, enabling the updated characteristic information of the common object of any one object set to contain semantic characteristics of the common object in the other object sets, and then classifying or searching the objects similarly after obtaining final characteristic information of each object according to the updated overall characteristic information of the plurality of object sets. The initial characteristic information of the common object obtained in different object sets is fused with each other, so that the final characteristic information of the common object is more accurate.

Description

Object processing method, system, computer readable storage medium and computer device

Technical Field

The present invention relates to the field of artificial intelligence information processing technologies, and in particular, to an object processing method, an object processing system, a computer-readable storage medium, and a computer device.

Background

In the existing information management system, a user interface is provided, so that a user can query information in the information management system through the user interface, for example, in an information search system, a corresponding search can be performed in a database through a search object input by the user and displayed to the user. The information management system can acquire the characteristic information of the corresponding object of the user behavior and perform processing such as similar recommendation or object classification based on the characteristic information.

The existing method for acquiring the characteristic information of the object comprises the following steps: and acquiring a heterogeneous graph for representing each object, and extracting the characteristic information of the object represented by each node in the heterogeneous graph according to a certain machine learning model. If the user has different types of behaviors, but the behaviors are related, the existing method cannot well represent the characteristic information of the objects, and further the result of information processing based on the characteristic information of the objects is not very accurate.

Disclosure of Invention

The embodiment of the invention provides an object processing method, an object processing system, a computer readable storage medium and computer equipment, which realize object processing after updating initial characteristic information of some objects.

An embodiment of the present invention provides an object processing method, including:

acquiring object behavior information aiming at target function application; the object behavior information includes: information of a plurality of objects;

dividing the plurality of objects into a plurality of object sets according to the object behavior information, wherein at least two object sets in the plurality of object sets have a common object;

respectively extracting overall characteristic information of each object set, wherein the overall characteristic information of each object set comprises initial characteristic information corresponding to each object in each object set;

respectively updating initial feature information of a common object in any object set of the at least two object sets to obtain updated overall feature information of the any object set, wherein the updated feature information of the common object in any object set has semantic features of the common object in other object sets, and the other object sets are object sets except the any object set in the at least two object sets;

and acquiring final characteristic information of the plurality of objects according to the updated overall characteristic information of the plurality of object sets so as to classify the plurality of objects or search similar processing.

Another aspect of an embodiment of the present invention provides an object processing system, including:

an information acquisition unit configured to acquire object behavior information for a target function application; the object behavior information includes: information of a plurality of objects;

the dividing unit is used for dividing the plurality of objects into a plurality of object sets according to the object behavior information, and at least two object sets in the plurality of object sets have a common object;

the characteristic extraction unit is used for respectively extracting overall characteristic information of each object set, and the overall characteristic information of each object set comprises initial characteristic information corresponding to each object in each object set;

a feature updating unit, configured to update initial feature information of a common object in any one of the at least two object sets, respectively, to obtain updated overall feature information of the any one object set, where the updated feature information of the common object in any one object set has semantic features of common objects in other object sets, and the other object sets are object sets other than the any one object set in the at least two object sets;

and the object processing unit is used for acquiring the final characteristic information of the objects according to the updated overall characteristic information of the object sets so as to classify the objects or search similar processing.

Another aspect of the embodiments of the present invention provides a computer-readable storage medium storing a plurality of computer programs, the computer programs being adapted to be loaded by a processor and to perform an object processing method according to the embodiments of the present invention.

Another aspect of an embodiment of the present invention provides a computer device, including a processor and a memory;

the memory is used for storing a plurality of computer programs, and the computer programs are used for being loaded by the processor and executing the object processing method according to the embodiment of the invention; the processor is configured to implement each of the plurality of computer programs.

It can be seen that, in the method of this embodiment, the object processing system divides a plurality of objects into a plurality of object sets according to object behavior information applied to the target function, and extracts overall feature information of each object set, if a common object exists between at least two object sets in an object set, initial feature information of the common object in any one of the at least two object sets needs to be updated, so that updated overall feature information of the common object in any one of the at least two object sets includes semantic features of the common object in the other object sets, and then performs object classification or similarity search after obtaining final feature information of each object according to updated overall feature information of the plurality of object sets. Therefore, initial characteristic information obtained by some objects (namely common objects) in different object sets is fused with each other, so that the final characteristic information of the common objects is more accurate, and classification of the objects according to the final characteristic information of the objects is more accurate or the searched similarity is more accurate.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings 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 for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.

Fig. 1 is a schematic diagram of an object processing method according to an embodiment of the present invention;

FIG. 2 is a flow chart of a method for object processing according to an embodiment of the invention;

FIG. 3 is a diagram illustrating updating initial feature information of a common node in one embodiment of the invention;

FIG. 4 is a diagram illustrating obtaining final feature information for each node in an embodiment of the invention;

FIG. 5 is a flow chart of a method for object processing according to an embodiment of the invention;

FIG. 6 is a flow diagram of a method of training a feature extraction model and a translation reconstruction model in one embodiment of the invention;

FIG. 7 is a flow chart of a method for training a feature extraction model and a translation reconstruction model in an embodiment of the invention;

FIG. 8 is a schematic structural diagram of a feature extraction model and a translation reconstruction model in an embodiment of the present invention;

FIG. 9 is a diagram illustrating an object processing method according to an embodiment of the present invention;

FIG. 10 is a diagram illustrating a distributed system to which an object processing method is applied in another embodiment of the present invention;

FIG. 11 is a block diagram illustrating an exemplary block structure according to another embodiment of the present invention;

FIG. 12 is a schematic structural diagram of an object processing system according to an embodiment of the present invention;

fig. 13 is a schematic structural diagram of a computer device according to an 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.

The terms "first," "second," "third," "fourth," and the like in the description and in the claims, as well as in the drawings, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are, for example, capable of operation in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

An embodiment of the present invention provides an object processing method, which is a method for performing object processing according to feature information of an object involved in an object operation process applied to a target function, and as shown in fig. 1, an object processing system may implement the method according to the embodiment of the present invention by using the following steps:

acquiring object behavior information aiming at target function application; the object behavior information includes: information of a plurality of objects; dividing the plurality of objects into a plurality of object sets (illustrated in fig. 1 by taking n object sets as an example) according to the object behavior information, wherein at least two object sets in the plurality of object sets have a common object therebetween; respectively extracting overall characteristic information of each object set, wherein the overall characteristic information of each object set comprises initial characteristic information corresponding to each object in each object set; respectively updating initial feature information of common objects in any object set of the at least two object sets to obtain updated overall feature information of the any object set, wherein the updated feature information of the common objects in any object set has semantic features of the common objects in other object sets (the integration of the semantic features is represented by a dotted arrow in the figure), and the other object sets are object sets except the any object set in the at least two object sets; and acquiring final characteristic information of the plurality of objects according to the updated overall characteristic information of the plurality of object sets so as to classify the plurality of objects or search similar processing.

In the actual application process, only the object processing client can be included in the object processing system, and the object processing client executes the steps; in another case, an object processing client and an object processing server may be included in the object processing system, and the steps described above are performed by the object processing server.

And when the overall characteristic information of each object set is extracted, the overall characteristic information can be extracted by a preset characteristic extraction model, and when the initial characteristic information of a common object in any one of the at least two object sets is updated, the initial characteristic information can be updated by a preset translation reconstruction model, and the characteristic extraction model and the translation reconstruction model are machine learning models based on artificial intelligence respectively. Among them, Artificial Intelligence (AI) is a theory, method, technique and application system that simulates, extends and expands human Intelligence using a digital computer or a machine controlled by a digital computer, senses the environment, acquires knowledge and uses the knowledge to obtain the best result. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making.

The artificial intelligence technology is a comprehensive subject and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like.

Machine Learning (ML) is a multi-domain cross subject, and relates to multiple subjects such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and the like. The special research on how the computer simulates or realizes the learning behavior of human beings so as to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve the performance of the computer. Machine learning is the core of artificial intelligence, is the fundamental approach for computers to have intelligence, and is applied to all fields of artificial intelligence. Machine learning and deep learning generally include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and formal education learning.

Therefore, initial characteristic information obtained by some objects (namely common objects) in different object sets is fused with each other, so that the final characteristic information of the common objects is more accurate, and the classification of the objects according to the final characteristic information of the objects is more accurate or the searched similarity is more accurate.

An embodiment of the present invention provides an object processing method, which is a method executed by an object processing system, and a flowchart is shown in fig. 2, where the method includes:

step 101, obtaining object behavior information for a target function application, where the object behavior information includes information of a plurality of objects.

It is understood that any functional application (e.g. target functional application) provides a user interface, so that a user can perform a user operation through the user interface, and the target functional application performs corresponding processing according to the user operation, specifically how the processing is related to an actual application.

In the embodiment of the present invention, the object processing system may count object behavior information of the target function application in a period of time, and initiate the process of the embodiment, where the object behavior information refers to operation behavior information of any object in a corresponding processing process of the target function application according to a user operation, and any object may be any type of referring object, such as a commodity, a search word, or news.

Specifically, the object behavior information may include information of objects, relationship information between objects, and the like, where the information of the objects may include information such as names or identifications of the objects, and the relationship information between the objects may include: the method includes the steps of obtaining information such as whether the objects have relationship identification information and relationship types of the relationships between the related objects, and specifically, applying a target function to operation behavior types of any two objects as the relationship types between the objects.

For example, when a user inputs a word to be searched, the target function application performs corresponding search according to the word to be searched, wherein the word to be searched is an object; or, if the user clicks a news link, the target function application displays the corresponding news to the user, and the news is the object; or the user purchases two commodities through the interface provided by the target function application, the two commodities are the objects, the two objects are considered to have a relationship, and the relationship type is purchase and the like.

And 102, dividing the plurality of objects into a plurality of object sets according to the object behavior information, wherein at least two object sets in the plurality of object sets have a common object.

Specifically, the object processing system may divide the plurality of objects according to the relationship types between the objects in the object behavior information, wherein if the relationship types between the plurality of objects belong to the same type, the object processing system may divide the plurality of objects into the same object set, so that some object sets may have the same object, that is, a common object, among the object sets obtained through division; alternatively, the information processing system may randomly divide a plurality of objects having a relationship into one object set; or the information processing system takes one object as a center, divides a part of objects related to the object into one object set, and divides another part of objects related to the object into another object set, and the like.

For example, objects 1 to 5 are divided into 3 object sets, object set 1 includes object 1, object 2, and object 3, object set 2 includes object 2, object 3, and object 4, and object set 3 includes object 4 and object 5, where objects common between object set 1 and object set 2 are object 2 and object 3, and objects common between object set 2 and object set 3 are object 4.

And 103, respectively extracting the overall characteristic information of each object set, wherein the overall characteristic information of each object set comprises initial characteristic information corresponding to each object in each object set.

Specifically, the object processing system may call a preset feature extraction model, and the feature extraction model extracts the overall feature information of each object set according to the information of each object in each object set and the relationship information between the objects.

The preset feature extraction model may be obtained by training through a certain training method, and then the running logic of the preset feature extraction model is stored in the object processing system, and specifically may be a graph symbol (graph) model, a Convolutional Neural Network (CNN) model, a deep path (deep walk) model, or the like, which is not limited herein.

And 104, respectively updating the initial characteristic information of the common object in any object set of the at least two object sets to obtain updated overall characteristic information of any object set, wherein the updated characteristic information of the common object in any object set has semantic characteristics of the common object in other object sets, and the other object sets are object sets except any object set in the at least two object sets.

Specifically, the object processing system may first determine at least two sets of objects having common objects; then the object processing system can call a preset translation reconstruction model, and the preset translation reconstruction model respectively updates the initial characteristic information of the common object in any one object set of at least two object sets, wherein the translation reconstruction model can be obtained by training through a certain method, and then the operation logic of the translation reconstruction model is stored in the object processing system, and specifically can be a Cross view (Cross view) model, a CNN (computer network) model, a full connection layer, a transformation (transformer) neural network and the like.

Specifically, the translation reconstruction model translates initial characteristic information of common objects in any object set respectively to enable translated characteristics of the common objects between the object sets to be similar, specifically, the translation reconstruction model multiplies a matrix formed by the initial characteristic information of the common objects in any object set by a conversion matrix, and the obtained translated characteristics of the common objects in any object set are aligned with the initial characteristic information of the common objects in other object sets, for example, the translated characteristics have the same size and the like; and then, the translation reconstruction model reconstructs the translated features of the common objects in any object set, so that the reconstructed features of the common objects in different object sets can keep the features of the common objects in different object sets, the features are similar to the initial feature information of the common objects in the same object set, overfitting of the translation process to the initial feature information is prevented, specifically, a matrix formed by the translated features of the common objects in any object set is subjected to a conversion neural network, the obtained reconstructed features of the common objects in any object set are aligned with the initial feature information of the common objects in the same object set, for example, the reconstructed features of the common objects in any object set are the updated feature information of the common objects in any object set.

In this embodiment, the initial feature information of the common object in each of the at least two object sets may be analogized to the linguistic expression of the same semantic word in at least two different environments, and after the initial feature information of the common object in each of the at least two object sets is subjected to the translation reconstruction model, the updated feature information of the common object in each of the at least two object sets is similar to each other.

Note that, in this step 104, only the initial feature information of the common object obtained in the above step 103 is updated, and thus the updated overall feature information of some object sets obtained in this way includes the updated feature information of the common object and the initial feature information of the non-common object.

For example, as shown in fig. 3, the overall feature information of the object set 1 includes initial feature information 11 of the object 1, initial feature information 12 of the object 2, initial feature information 13 of the object 3, and initial feature information 14 of the object 4; and the global feature information of the object set 2 includes the initial feature information 23 of the object 3, the initial feature information 24 of the object 4, the initial feature information 25 of the object 5, and the initial feature information 26 of the object 6, the object 3 and the object 4 are common objects. In this step, it is necessary to update the initial feature information 13 of the object 3 and the initial feature information 14 of the object 4 in the object set 1, and the initial feature information 23 of the object 3 and the initial feature information 24 of the object 4 in the object set 2, respectively, and the updated feature 13 'of the object 3 in the object set 1 obtained separately includes the semantic feature of the initial feature information 23 (the integration of the semantic feature is indicated by a dotted arrow in the figure), while the updated feature 14' of the object 4 in the object set 1 includes the semantic feature of the initial feature information 24, the updated feature 23 'of the object 3 in the object set 2 includes the semantic feature of the initial feature information 13, and the updated feature 24' of the object 4 in the object set 2 includes the semantic feature of the initial feature information 14.

And 105, acquiring final feature information of the plurality of objects according to the updated overall feature information of the plurality of object sets so as to classify the plurality of objects or search similar processing.

It is understood that the updated global feature information of any one object set of the at least two object sets obtained in step 105 includes: and when the object processing system acquires the final feature information of each object in any object set according to the updated overall feature information of any object set, the object processing system mainly takes the average feature of the updated feature information of the common objects in the at least two object sets as the final feature information of the common object and takes the initial feature information of the non-common object in any object set as the final feature information of the non-common object.

For example, as shown in fig. 4, if the updated global feature information of the object set 1 includes the initial feature information 11 of the object 1, the updated feature information 12 of the object 2, and the initial feature information 13 of the object 3, and the updated global feature information of the object set 2 includes the initial feature information 24 of the object 4, the updated feature information 22 of the object 2, and the initial feature information 25 of the object 5, the final feature information of the object 2 may be the average feature of the updated feature information 12 and the updated feature information 22, and the final feature information of other objects may be the initial feature information of these objects, respectively.

Further, after obtaining the final feature information of the plurality of objects in the object set, the object processing system may classify the corresponding object according to the final feature information of each object, or search for information similar to the corresponding object.

Another embodiment of the present invention provides an object processing method, which is a method executed by an object processing system, in this embodiment, the object processing system converts acquired object behavior information into an abnormal composition form for processing, and a flowchart is shown in fig. 5, and includes:

in step 201, object behavior information for a target function application is obtained.

Specifically, the object processing system may count object behavior information of the target function application in a period of time, and initiate the process of this embodiment, where the object behavior information refers to operation behavior information of any object in a corresponding processing process of the target function application according to a user operation, and any object may be any type of referring object, such as a commodity, a search term, or news. The object behavior information may include information of objects, relationship information between objects, and the like, and an operation behavior type of applying a target function to any two objects may be used as a relationship type between the objects.

Step 202, generating a heterogeneous graph according to the object behavior information, where the heterogeneous graph includes a plurality of nodes and edges between the plurality of nodes, any node represents an object, and an edge between any two nodes is used to represent a relationship between objects represented by any two nodes.

When determining whether any two nodes in the abnormal graph have edges, the determination may be performed according to relationship information between objects represented by any two nodes, and if two objects have a relationship, the corresponding nodes have edges. The relationship between two objects is not necessarily an obvious directional relationship, but may also have an implicit relationship, for example, if the time interval between two object behaviors is less than a preset value, it may be considered that the objects respectively related to the two object behaviors have an implicit relationship; if one object behavior is executed on an interface displayed after the other object behavior is executed, the objects respectively related to the two object behaviors have obvious pointing relationships.

And 203, dividing the heterogeneous graph to obtain a heterogeneous sub-graph set, wherein at least two heterogeneous sub-graphs in the heterogeneous sub-graph set have a common node. Wherein the heterogeneous subgraph set comprises a plurality of heterogeneous subgraphs.

Specifically, the object processing system may divide the heterogeneous graph according to types of edges between nodes included in the heterogeneous graph, where a type of an edge between any two nodes in the heterogeneous graph specifically refers to a relationship type between objects represented by two nodes, and if relationship types between objects represented by some nodes in the heterogeneous graph belong to the same type, the object processing system may divide the nodes into the same heterogeneous subgraph, so that some heterogeneous subgraphs have the same node, that is, a common node, in the heterogeneous subgraph set obtained through division.

And 204, respectively extracting integral characteristic information of each heterogeneous subgraph, wherein the integral characteristic information of each heterogeneous subgraph comprises initial characteristic information corresponding to each node in each heterogeneous subgraph.

Specifically, the object processing system may call a preset feature extraction model, and the feature extraction model extracts the overall feature information of each heterogeneous subgraph according to the object information represented by each node in each heterogeneous subgraph and the relationship information between the objects. Here, the preset feature extraction model may be a Graphsage model, a CNN model, a DeepWalk model, or the like, which is not limited herein.

And step 205, respectively updating the initial feature information of the common node in any one of the at least two heterogeneous subgraphs to obtain updated overall feature information of any one of the heterogeneous subgraphs, wherein the updated feature information of the common node in any one of the heterogeneous subgraphs has semantic features of the common node in other heterogeneous subgraphs, and the other heterogeneous subgraphs are heterogeneous subgraphs of the at least two heterogeneous subgraphs except any one of the heterogeneous subgraphs.

Specifically, the object processing system determines at least two heterogeneous subgraphs having a common node; then, the object processing system can call a preset translation reconstruction model, and the preset translation reconstruction model respectively updates the initial feature information of a common node in any one of the at least two heterogeneous subgraphs, wherein the translation reconstruction model can be a Cross view model, a CNN model, a full connection layer, a transform neural network and the like.

Specifically, the translation reconstruction model multiplies a matrix formed by initial feature information of a common node in any heterogeneous subgraph by a conversion matrix, and the obtained translated feature of the common node in any heterogeneous subgraph is aligned with the initial feature information of the common node in other heterogeneous subgraphs, for example, the translated feature and the initial feature information have the same size; and then, the translation reconstruction model passes a matrix formed by the translated features of the common nodes in any heterogeneous subgraph through a conversion neural network, and the obtained reconstructed features of the common nodes in any heterogeneous subgraph are aligned with the initial feature information of the common nodes in the same heterogeneous subgraph, for example, the reconstructed features have the same size and the like, so that the reconstructed features of the common nodes in any heterogeneous subgraph are the updated feature information of the common nodes in any heterogeneous subgraph.

It should be noted that, in this step 105, only the initial feature information of the common node obtained in the above step 104 is updated, so that the obtained updated overall feature information of some heterogeneous subgraphs includes the updated feature information of the common node and the initial feature information of the non-common node.

And step 206, acquiring final feature information of the objects respectively represented by the nodes according to the updated overall feature information of the heterogeneous subgraphs, so as to perform classification processing or similar searching processing on the objects represented by the nodes.

It can be understood that the updated overall characteristic information of any one of the at least two heterogeneous subgraphs obtained in step 205 includes: and when the object processing system acquires the final feature information of the object represented by each node in any heterogeneous subgraph according to the updated overall feature information of any heterogeneous subgraph, the average feature of the updated feature information of the common node in at least two heterogeneous subgraphs is mainly used as the final feature information of the common node, and the initial feature information of the non-common node in any heterogeneous subgraph is used as the final feature information of the non-common node.

Further, after the final feature information of the objects represented by the multiple nodes in the heterogeneous subgraph is obtained, the object processing system may classify the corresponding objects according to the final feature information of the objects represented by the nodes, or search for information similar to the corresponding objects.

It can be seen that, in the method of this embodiment, the object processing system generates the heterogeneous graph according to the object behavior information applied to the target function, then divides the heterogeneous graph to obtain heterogeneous sub-graph sets, and extracts the overall feature information of each heterogeneous sub-graph respectively, if at least two heterogeneous sub-graphs in the heterogeneous sub-graph sets have a common node, the initial feature information of the common node of any one of the at least two heterogeneous sub-graphs needs to be updated, so that the updated overall feature information of the common node of any one of the heterogeneous sub-graphs includes semantic features of the common node in other heterogeneous sub-graphs, and then performs classification or search for similar objects after obtaining the final feature information of the object represented by each node according to the updated overall feature information of the plurality of heterogeneous sub-graphs. Therefore, initial characteristic information obtained by some nodes (namely common nodes) in different heterogeneous subgraphs is fused with each other, so that the final characteristic information of the object represented by the common nodes is more accurate, and classification of each object according to the final characteristic information of the object represented by each node is more accurate or the searched similarity is more accurate.

In a specific embodiment, the above feature extraction model and the translation reconstruction model may be trained as follows, and a flowchart is shown in fig. 6, and includes:

step 301, determining a feature extraction initial model and a translation reconstruction initial model, wherein an output end of the feature extraction initial model is connected to an input end of the translation reconstruction initial model.

It is understood that when determining the initial feature extraction model and the initial translation reconstruction model, the object processing system determines the initial values of the parameters in the multilayer structure and each layer mechanism included in the initial feature extraction model and the initial translation reconstruction model, respectively.

Specifically, the feature extraction initial model is mainly used for extracting overall feature information of each heterogeneous subgraph, and may also be used for extracting feature information of each object having a relationship in an object set. The translation reconstruction initial model is used for respectively updating the feature information of the common node in any heterogeneous subgraph for two heterogeneous subgraphs with the common node, so that the updated feature information of the common node in any heterogeneous subgraph contains the semantic feature of the common node in the other heterogeneous subgraph; the method and the device can also be used for updating the feature information of the common object in any one set respectively for two object sets with the common object, so that the updated feature information of the common object in any one object set comprises the semantic feature of the common object in the other object set. Specifically, the translation reconstruction initial model may be a Cross view model, or a CNN model, or a full link layer, or a transform, etc.

In a specific application, the above translation reconstruction initial model may include: the translation module is used for multiplying a matrix formed by the characteristic information of the common node in any heterogeneous subgraph by a conversion matrix to obtain translated characteristics aligned with the characteristic information of the common node in another heterogeneous subgraph; the reconstruction module is used for enabling a matrix formed by translated features of the common node in any heterogeneous subgraph to pass through a conversion neural network, and the obtained reconstructed features are aligned with feature information of the common node in the same heterogeneous subgraph, so that the reconstructed features of the common node in any heterogeneous subgraph are updated feature information of the common node in any heterogeneous subgraph.

Or the translation module is used for multiplying a matrix formed by the characteristic information of the common objects in any object set by a conversion matrix to obtain translated characteristics aligned with the characteristic information of the common objects in another object set; the reconstruction module is used for enabling a matrix formed by the translated features of the common objects in any object set to pass through a conversion neural network, and enabling the obtained reconstructed features to be aligned with the feature information of the common objects in the same object set, so that the reconstructed features of the common objects in any object set are updated feature information of the common objects in any object set.

The parameters of the initial feature extraction model and the initial translation reconstruction model are fixed parameters used in the calculation process of each layer structure in the initial feature extraction model and the initial translation reconstruction model, and do not need to be assigned at any time, such as parameters of parameter scale, network layer number, user vector length and the like.

Step 302, determining a training sample, where the training sample includes information of a plurality of first sample object sets and information of a plurality of second sample object sets, and any first sample object set and another second sample object set have a common object, and the information of any sample object set (the first sample object set and the second sample object set) includes information of each sample object in any sample object set and relationship information between each sample object, that is, a heterogeneous subgraph can be generated according to each sample object set.

That is, there is a one-to-one correspondence between a plurality of first sample object sets and a plurality of second sample object sets, and one first sample object set corresponds to one second sample object set with a common object therebetween.

Step 303, respectively determining feature information of each sample object in each sample object set through the feature extraction initial model, respectively updating the feature information of the common object in any sample object set through the translation reconstruction initial model, and obtaining updated feature information of the common object in any sample object set, wherein the updated feature information of the common object in any sample object set comprises semantic features of the common object in another sample object set.

Specifically, after extracting feature information of each sample object in each sample object set (including a first sample object set and a second sample object set), the feature extraction initial model is transmitted to a translation reconstruction initial model; the translation reconstruction initial model respectively updates the feature information of the common object in any sample object set extracted by the feature extraction initial model.

And 304, adjusting parameter values in the feature extraction initial model and the translation reconstruction initial model to obtain a final feature extraction model and a final translation reconstruction model.

Specifically, the object processing system may adjust the feature extraction initial model according to feature information of each sample object obtained by the feature extraction initial model; and adjusting parameter values in the translation reconstruction initial model according to updated characteristic information of the common object in any sample object set obtained by translating the reconstruction model.

The object processing system may first calculate a first loss function associated with the feature extraction initial model according to the feature information obtained from the feature extraction initial model in step 302, where the first loss function is used to represent a feature difference, such as a difference between feature information of adjacent sample objects having a relationship, and then adjust a parameter value in the feature extraction initial model through the first loss function. In addition, the object processing system calculates a second loss function related to the translation reconstruction initial model according to the updated feature information obtained by the translation reconstruction initial model, the second loss function is used for representing another feature difference value, and then parameter values in the translation reconstruction initial model are adjusted through the second loss function. Wherein, the setting of the first loss function and the second loss function is the setting mode of unsupervised training.

Further, if the translation reconstruction initial model includes a translation model and a reconstruction model, the second loss function may be divided into a loss function related to the translation model and a function calculation value of the loss function related to the reconstruction model, where the loss function related to the translation model may represent a difference value between the updated feature information of the common object in any one first sample object set and the updated feature information of the common object in another second sample object set, and the like; and the loss function associated with the reconstructed model may represent a difference between the updated feature information of the common object in any sample object set and the un-updated feature information of the common object in the sample object set, and so on.

The training process of the feature extraction model and the translation reconstruction model needs to reduce the value of the feature difference as much as possible, and the training process continuously optimizes the parameter values of the parameters in the feature extraction initial model and the translation reconstruction initial model determined in the step 301 through a series of mathematical optimization means such as back propagation derivation and gradient reduction, so as to respectively minimize the calculated values of the first loss function and the second loss function.

It should be noted that, in the above steps 303 to 304, the feature information of each sample object detected by the feature extraction initial model and the updated feature information of the common object in any sample object set obtained by translating the reconstructed model are respectively adjusted once for the parameter values in the feature extraction initial model and the translation reconstructed initial model, and in practical applications, the above steps 303 to 304 need to be executed continuously and circularly until the adjustment for the parameter values meets a certain stop condition.

Therefore, after the steps 301 to 304 of the above embodiment are executed, the object processing system further needs to determine whether the current adjustment on the parameter values meets the preset stop condition, and if so, the feature extraction initial model and the translation reconstruction initial model after the parameter values are adjusted in the step 304 are the trained feature extraction initial model and the translation reconstruction initial model, and the process is ended; if not, the initial model is extracted according to the characteristics after the parameter values are adjusted and the initial model is reconstructed according to the translation after the parameters are adjusted, and the steps 303 to 304 are executed. Wherein the preset stop condition includes but is not limited to any one of the following conditions: the difference value between the current adjusted parameter value and the last adjusted parameter value is smaller than a threshold value, namely the adjusted parameter value reaches convergence; and the adjustment times of the parameter values are equal to the preset times, and the like.

In addition, it should be noted that the training in the above process is an unsupervised training process, and in another specific embodiment, the feature extraction model and the translation reconstruction model may also be trained by a supervised training method, and specifically, when the object processing system executes the above step 301, it may further be determined that an output end of the translation reconstruction initial model is connected to a classification module, where the classification module is configured to classify each sample object according to the updated feature information of each sample object in each sample object set output by the translation reconstruction initial model. And in executing step 302, the type marking information of each sample object in each sample object set (including the first sample object set and the second sample object set) is included in the determined training sample.

Further, when the step 303 is executed, after the translation reconstruction model updates the feature information of the common object in any sample object set, the updated feature information of each sample object in each sample object set is output to the classification module, and the classification model determines the type of each sample object according to the updated feature information of each sample object in each sample object set output by the translation reconstruction model.

In this case, when the step 304 is executed, the adjustment may be performed in two ways including, but not limited to:

(1) adjusting by adopting supervised training mode

And calculating a comprehensive loss function related to the characteristic extraction initial model and the translation reconstruction initial model according to the result obtained by the classification module and the type marking information in the training sample, wherein the comprehensive loss function is used for indicating the type of each sample object jointly predicted by the characteristic extraction initial model, the translation reconstruction initial model and the classification module and the error between the type of each sample object and the actual type (obtained according to the type marking information in the training sample) of each sample object, and further adjusting the parameter values in the characteristic extraction initial model and the translation reconstruction initial model through the comprehensive loss function. The setting of the comprehensive loss function is a setting mode with supervision training, namely, the results output by the feature extraction initial model, the translation reconstruction initial model and the classification module need to be supervised by type labeling information in a training sample.

In this case, the training process of the feature extraction model and the translation reconstruction model is to reduce the error value as much as possible, and the training process is to continuously optimize the parameter values of the parameters in the feature extraction initial model and the translation reconstruction initial model determined in the step 301 by a series of mathematical optimization means such as back propagation derivation and gradient descent, and to minimize the calculated value of the comprehensive loss function.

(2) Training mode adjustment combining unsupervised mode and supervised mode

The object processing system calculates a first loss function associated with the feature extraction initial model according to the feature information obtained from the feature extraction initial model in step 303, where the first loss function is used to represent a feature difference, such as a difference between feature information of adjacent sample objects having a relationship, and then adjusts a parameter value in the feature extraction initial model through the first loss function. And the object processing system calculates another loss function related to the translation reconstruction initial model according to the result obtained by the classification module and the type marking information in the training sample, the another loss function is used for indicating the type of each sample object jointly predicted by the translation reconstruction initial model and the classification module and the error of the actual type (obtained according to the type marking information in the training sample) of each sample object, and then the parameter value in the translation reconstruction initial model is adjusted through the another loss function. The setting corresponding to the first loss function is the setting of unsupervised training, and the setting corresponding to the other loss function is the setting mode of supervised training, namely, the translation reconstruction initial model and the result output by the classification module need to be supervised by type marking information in the training sample.

In this case, the training process of the feature extraction model and the translation reconstruction model is to reduce the feature difference and the error as much as possible, and the training process is to continuously optimize the parameter values of the parameters in the feature extraction initial model and the translation reconstruction initial model determined in the step 301 by a series of mathematical optimization means such as back propagation derivation and gradient descent, so as to respectively minimize the calculated values of the first loss function and the other loss function.

In the following, a specific application example is used to describe the object processing method in the present invention, in this embodiment, each object may be represented in the form of an abnormal composition, and the method of this embodiment may specifically include the following two parts:

(1) as shown in fig. 7, when the feature extraction model and the translation reconstruction model are trained through the following steps, specifically:

step 401, determining the feature extraction initial model as a graphsage model, determining that the translation reconstruction initial model adopts a Cross view model, and initializing the feature extraction initial model and the translation reconstruction initial model, that is, determining initial values of parameters in the initial models.

As shown in FIG. 8, the Cross view model includes a Translation module (Translation Tasks) and a Reconstruction module (Reconstruction Tasks), an output of the graph view model is connected to the Translation module, and an output of the Translation module is connected to the Reconstruction module. The graph merge model can adopt a 3-layer average pooling convolution structure and a last layer full-connection output layer, information of nodes and neighbor nodes in a heterogeneous subgraph can be aggregated into self node information in each layer of structure, then the aggregated information is input into a next layer, and a result is output by the last layer.

Step 402, determining a training sample, including a plurality of first sample object sets and a plurality of second sample object sets, where information of any sample object set includes relationship information between sample objects in any sample object set, and any first sample object set and another second sample object set have a common object. In this way, any sample object set can form a heterogeneous subgraph (Paired-Subview), each node in the heterogeneous subgraph represents a sample object, and edges between the nodes represent the relationship between the sample objects.

In this embodiment, each sample object may be represented by a heterogeneous graph, where the heterogeneous graph is a graph network, the graph network is composed of several nodes and links connecting the nodes, the links between two nodes may be called edges to represent objects and their interconnections, and the heterogeneous graph is a graph network with multiple types of nodes or multiple types of edges.

In this embodiment, a sample object is represented by a node, if there is a relationship between two sample objects, there are edges between the nodes represented by the two sample objects, and each sample object set can be represented by a heterogeneous subgraph, where any heterogeneous subgraph represented by a first sample object set has a common node with another heterogeneous subgraph represented by a second sample object set, and the common node is used to represent a common object.

Step 403, extracting feature information of each sample object by using the graphsage model determined in step 401, and then updating the feature information of the common object in any sample object set by using the translation module and the reconstruction module to obtain updated feature information of the common object in each sample object set.

Specifically, for each node in the heterogeneous subgraph formed by each sample object set, information of sample objects represented by each node and its neighboring nodes needs to be input into the graph, the sample objects represented by each node and its neighboring nodes have a relationship, and the graph outputs feature information, that is, embedding (embedding) features, of each node in each heterogeneous subgraph.

If the information of the sample object input into the graph sage model is a text, the text is encoded and then input into the graph sage model; if digital, the numbers can be directly entered into the graphsage model.

In all feature information output by the Cross view model, a translation module in the Cross view model obtains translated feature information of a common node in any heterogeneous subgraph aiming at feature information of a common object represented by a common node between any two heterogeneous subgraphs, so that the translated feature information of the common node in any heterogeneous subgraph is aligned with the translated feature information of the corresponding common node in the other heterogeneous subgraph, namely the translated feature information of the same common node in different heterogeneous subgraphs is similar, and the information of the common node in different heterogeneous subgraphs is interacted.

A reconstruction module in the Cross view model outputs reconstructed feature information of a common node in any heterogeneous subgraph aiming at the translated feature information of the common node in any heterogeneous subgraph, wherein the reconstructed feature information is updated feature information, so that the reconstructed feature information of the common node in any heterogeneous subgraph is aligned with the reconstructed feature information of the corresponding common node in the heterogeneous subgraph, namely the reconstructed feature information of each common node needs to be similar to the feature information of the corresponding common node output by the previous graphsage model, the self features of the heterogeneous subgraphs are maintained, meanwhile, the translation module is enabled to have consistency, and the overfitting of the translation module is prevented.

Step 404, calculating a first loss function related to the graph sage model according to the feature information extracted by the graph sage model, and adjusting a parameter value in the graph sage model according to the first loss function; and calculating a second loss function related to the Cross view model according to the updated characteristic information output by the Cross view model, and adjusting the parameter value in the Cross view model according to the second loss function.

Specifically, the first loss function is used to represent a difference between feature information of each node and a neighboring node in any one of the heterogeneous subgraphs extracted by the graph sage model, and after parameter values in the graph sage model are adjusted, the first loss function is converged, that is, the first loss function is minimized. The second loss function may be divided into two parts, namely a loss function associated with the translation module and a loss function associated with the reconstruction module, wherein:

the loss function associated with the translation module is composed of function computed values of the loss function represented by the following equations 1 and 2, i represents a common node in any one of the heterogeneous subgraphs, j represents a corresponding common node in the other heterogeneous subgraph, A_iRepresenting the difference, λ, between the characteristic information of one common node in any one of the heterogeneous subgraphs and the corresponding characteristic information of the common node in the other heterogeneous subgraph_iRepresenting the number of common nodes in any heterogeneous subgraph, see that_i→j(λ_i) The expression translation module obtains the error of the translated characteristic information of the common node in any heterogeneous subgraph through L_j→i(λ_j) Representing translation module to obtain translation of common node in another heterogeneous subgraphError of post-feature information.

The loss function associated with the reconstruction module is composed of function computed values of the loss function represented by the following equations 3 and 4, i represents a common node in any one of the heterogeneous subgraphs, j represents a corresponding common node in the other heterogeneous subgraph, and the loss function L is used for calculating the loss function_i→j→i(λ_i) The representation reconstruction module processes the translated characteristic information of the common node in any heterogeneous subgraph to obtain the error of the reconstructed characteristic information, and the error is processed through a loss function L_j→i→j(λ_j) And the representation translation module processes the translated characteristic information of the common node in the other heterogeneous subgraph to obtain the error of the reconstructed characteristic information.

And a second loss function L_crossThe value may be calculated for the function of the loss function calculated by the above equations 1 to 4, and may be expressed by, for example, the following equation 5:

through the multiple cycles of the

steps

403 and 404, more appropriate parameter values in the graph model and the Cross view model can be obtained, and finally the trained graph model and the Cross view model are obtained.

(2) When the target function application is a search application, the object behavior information of the target function application may include search information input by a user, information clicked by the user, and the like, and the final feature information of an object involved in the search process of the target function application may be obtained through the trained graphpage model and Cross view model, which may be implemented by the following steps, as shown in fig. 8 and 9:

step 501, counting object behavior information of the target function application in a period of time, which may include search information input by a user, information clicked by the user, and the like, generating a heteromorphic graph according to the object behavior information, wherein each object represents a node, and if two objects have a relationship, an edge exists between corresponding nodes.

The target function application can be a search-and-search application, can search friend circles, articles, public numbers, novels, music, expressions and the like according to keywords, can also comprise a vertical search (namely vertical search) application, can be a subdivision and extension of a search engine aiming at a professional search engine in a certain industry, integrates certain special information in a library once, extracts required data from directed subsections, processes the extracted data and returns the processed data to a user in a certain form. The first search is specific to a certain type of result search, such as a public number search, an applet search, and the like.

Step 502, dividing the heterogeneous composition according to the type of the edge between each node in the obtained heterogeneous composition, that is, the relationship type between the objects represented by the nodes, to obtain a heterogeneous subgraph set, where some heterogeneous subgraphs in the heterogeneous subgraph set have a common node.

Step 503, inputting the information of the objects represented by the nodes in each heterogeneous subgraph and the relationship information between the objects into the trained graph sage model, so that the graph sage model can output the initial feature information of each node in each heterogeneous subgraph.

Step 504, for any two heterogeneous subgraphs with a common node, random walk (random walk) is performed in each of the two heterogeneous subgraphs, a random walk path of the common node is reserved, and the random walk path of the common node in each heterogeneous subgraph is input into a Cross view model.

And 505, updating the initial characteristic information of the common node in each heterogeneous subgraph output by the graph view model according to the random walk path of the common node in each heterogeneous subgraph by the Cross view model, wherein the updated characteristic information of the common node in each heterogeneous subgraph is output by the Cross view model.

Step 506, the average feature of the updated feature information of the corresponding common node in each heterogeneous subgraph is used as the final feature information of the common nodes, and the initial feature information of the non-common nodes in each heterogeneous subgraph is the final feature information of the non-common nodes.

And 507, obtaining final feature information of each object represented by each node according to the final feature information of each node in the heterogeneous subgraph, and classifying each object according to the final feature information of each object, or searching information similar to the objects.

After practice, the method of the present embodiment and the existing methods, such as Node2vec model and Node2vec, are adopted¹As shown in table 1 below, it can be known that both the F1 and the AUC values of the method according to the present embodiment are the highest values, which indicates that the characteristic information of each object determined by the method according to the present embodiment is more accurate, and that the characteristic information of each object is more accurate, when the method according to the present embodiment is used for determining the characteristic information of each object, and calculating two evaluation parameters under multiple methods, such as F1 and the area under the Receiver Operating Characteristic (ROC) curve (AUC), where F1 is used for representing the harmonic mean of precision (precision) and recall (recall) of the model, and the ROC curve has vertical coordinates of True Positive Rate (TPR) and horizontal coordinates of False Positive Rate (FPR):

TABLE 1

It can be seen that, in this embodiment, for any two heterogeneous subgraphs having a common node, the initial feature information of the common node in any one heterogeneous subgraph is fused with the initial feature information of the common node in another heterogeneous subgraph through the Crossview model, and the information of each heterogeneous subgraph can be combined nonlinearly, so that the initial feature information of the object with the same semantic expressed by two languages can be aligned, the actual meaning of the object is well reflected, and tasks such as classification or similar search of each object are more accurate.

In the following, another specific application example is used to describe the object processing method in the present invention, the object processing system in the embodiment of the present invention is mainly a distributed system 100, the distributed system may include a client 300 and a plurality of nodes 200 (any form of computing devices in an access network, such as servers and user terminals), and the client 300 and the nodes 200 are connected through a network communication form.

Taking a distributed system as an example of a blockchain system, referring To fig. 10, which is an optional structural schematic diagram of the distributed system 100 applied To the blockchain system provided in the embodiment of the present invention, the system is formed by a plurality of nodes 200 (computing devices in any form in an access network, such as servers and user terminals) and clients 300, a Peer-To-Peer (P2P, Peer To Peer) network is formed between the nodes, and the P2P Protocol is an application layer Protocol operating on a Transmission Control Protocol (TCP). In a distributed system, any machine, such as a server or a terminal, can join to become a node, and the node comprises a hardware layer, a middle layer, an operating system layer and an application layer.

Referring to the functions of each node in the blockchain system shown in fig. 10, the functions involved include:

1) routing, a basic function that a node has, is used to support communication between nodes.

Besides the routing function, the node may also have the following functions:

2) the application is used for being deployed in a block chain, realizing specific services according to actual service requirements, recording data related to the realization function to form recording data, carrying a digital signature in the recording data to represent a source of task data, and sending the recording data to other nodes in the block chain system, so that the other nodes add the recording data to a temporary block when the source and integrity of the recording data are verified successfully.

For example, the services implemented by the application include:

the wallet is used for providing functions of conducting transactions of electronic money, and comprises the functions of initiating transactions (namely, sending transaction records of current transactions to other nodes in a blockchain system, and storing the record data of the transactions into a temporary block of the blockchain as a response for confirming that the transactions are valid after the other nodes are successfully verified, of course, the wallet also supports the inquiry of the electronic money remaining in an electronic money address, a shared book for providing functions of storing, inquiring, modifying and the like of account data, sending the record data of the operation on the account data to other nodes in the blockchain system, and storing the record data into the temporary block as a response for confirming that the account data are valid after the other nodes are verified to be valid, and also can send confirmation to the node initiating the operation, and can also comprise intelligent contracts and computerized protocols, and can execute terms of a certain contract, the method is realized by codes which are deployed on a shared account and are used for executing when certain conditions are met, and the codes are used for completing automated transaction according to actual business requirements, such as inquiring the logistics state of goods purchased by a buyer and transferring the electronic money of the buyer to the address of a merchant after the buyer signs the goods; of course, smart contracts are not limited to executing contracts for trading, but may also execute contracts that process received information.

In this embodiment, the application in the node further includes a code for implementing an object processing function, where the object processing function mainly includes:

acquiring object behavior information aiming at target function application; the object behavior information includes: information of a plurality of objects; dividing the plurality of objects into a plurality of object sets according to the object behavior information, wherein at least two object sets in the plurality of object sets have a common object; respectively extracting overall characteristic information of each object set, wherein the overall characteristic information of each object set comprises initial characteristic information corresponding to each object in each object set; respectively updating initial feature information of a common object in any object set of the at least two object sets to obtain updated overall feature information of the any object set, wherein the updated feature information of the common object in any object set has semantic features of the common object in other object sets, and the other object sets are object sets except the any object set in the at least two object sets; and acquiring final characteristic information of the plurality of objects according to the updated overall characteristic information of the plurality of object sets so as to classify the plurality of objects or search similar processing.

3) And the Block chain comprises a series of blocks (blocks) which are mutually connected according to the generated chronological order, new blocks cannot be removed once being added into the Block chain, and recorded data submitted by nodes in the Block chain system are recorded in the blocks.

Referring to fig. 11, an optional schematic diagram of a Block Structure (Block Structure) provided in the embodiment of the present invention is shown, where each Block includes a hash value of a transaction record stored in the Block (hash value of the Block) and a hash value of a previous Block, and the blocks are connected by the hash values to form a Block chain. The block may include information such as a time stamp at the time of block generation. A block chain (Blockchain), which is essentially a decentralized database, is a string of data blocks associated by using cryptography, and each data block contains related information for verifying the validity (anti-counterfeiting) of the information and generating a next block.

An embodiment of the present invention further provides an object processing system, a schematic structural diagram of which is shown in fig. 12, and the object processing system may specifically include:

an information acquisition unit 10 for acquiring object behavior information for a target function application; the object behavior information includes: information of a plurality of objects.

A dividing unit 11, configured to divide the multiple objects into multiple object sets according to the object behavior information acquired by the information acquiring unit 10, where at least two object sets in the multiple object sets have a common object therebetween.

A feature extraction unit 12, configured to extract overall feature information of each object set obtained by the dividing unit 11, where the overall feature information of each object set includes initial feature information corresponding to each object in each object set.

The feature extraction unit 12 is specifically configured to invoke a preset feature extraction model; and the preset feature extraction model respectively extracts the overall feature information of each object set according to the information of each object in each object set and the relationship information among the objects.

A feature updating unit 13, configured to update initial feature information of a common object in any one of the at least two object sets extracted by the feature extracting unit 12, respectively, to obtain updated overall feature information of the any one object set, where the updated feature information of the common object in any one object set has semantic features of a common object in other object sets, and the other object sets are object sets other than the any one object set in the at least two object sets.

The feature updating unit 13 is specifically configured to invoke a preset translation reconstruction model; the preset translation reconstruction model multiplies a matrix formed by initial characteristic information of common objects in any object set of the at least two object sets by a conversion matrix to obtain translated characteristics of the common objects in any object set, and the translated characteristics of the common objects in the other object sets are aligned with the initial characteristic information of the common objects in the other object sets; the preset translation reconstruction model enables a matrix formed by translated features of common objects in any object set to pass through a conversion neural network, and the obtained reconstructed features of the common objects in any object set are aligned with the initial feature information of the common objects in any object set; the reconstructed feature of the common object in any one of the object sets is updated feature information of the common object in any one of the object sets.

And an object processing unit 14, configured to obtain final feature information of the multiple objects according to the updated overall feature information of the multiple object sets obtained by the feature updating unit 13, so as to perform classification processing or search for similar processing on the multiple objects.

It should be noted that, when the updated overall feature information of any one object set in the at least two object sets obtained by the feature updating unit 13 includes: updated feature information of the common object and initial feature information of the non-common object; the object processing unit 14 is specifically configured to use an average feature of the updated feature information of the common object in the at least two object sets as the final feature information of the common object when acquiring the final feature information of each object in any object set according to the updated overall feature information of any object set; and taking the initial characteristic information of the non-common object in any object set as the final characteristic information of the non-common object.

Further, the object processing system in this embodiment further includes: the heterogeneous graph generating unit 15 is configured to generate a heterogeneous graph according to the object behavior information acquired by the information acquiring unit 10, where the heterogeneous graph includes a plurality of nodes and edges between the plurality of nodes, the nodes represent an object, and the edges between any two nodes are used to represent a relationship between the objects represented by any two nodes. The dividing unit 11 is specifically configured to divide the heterogeneous graph generated by the heterogeneous graph unit 15 to obtain heterogeneous sub-graph sets, where at least two heterogeneous sub-graph sets in the heterogeneous sub-graph set have a common node. When obtaining the heterogeneous sub-graph set, the dividing unit 11 divides some nodes into the same heterogeneous sub-graph when the relationship types between the objects represented by some nodes in the heterogeneous sub-graph belong to the same type.

In this case, the feature extraction unit 12 is specifically configured to invoke a preset feature extraction model; and the preset feature extraction model respectively extracts the overall feature information of each heterogeneous subgraph according to the object information represented by each node in each heterogeneous subgraph and the relationship information between the objects. The feature updating unit 13 is specifically configured to invoke a preset translation reconstruction model; the preset translation reconstruction model multiplies a matrix formed by initial characteristic information of a common node in any one heterogeneous subgraph of the at least two heterogeneous subgraphs by a conversion matrix to obtain translated characteristics of the common node in any one heterogeneous subgraph, and the translated characteristics of the common node in any one heterogeneous subgraph are aligned with the initial characteristic information of the common node in other heterogeneous subgraphs; the preset translation reconstruction model enables a matrix formed by translated features of common nodes in any heterogeneous subgraph to pass through a conversion neural network, and the obtained reconstructed features of the common nodes in any heterogeneous subgraph are aligned with the initial feature information of the common nodes in any heterogeneous subgraph; and the reconstructed characteristic of the common node in any heterogeneous subgraph is the updated characteristic information of the common node in any heterogeneous subgraph.

Further, the object processing system in this embodiment further includes: the model training unit 16 is used for determining a feature extraction initial model and a translation reconstruction initial model, wherein the output end of the feature extraction initial model is connected to the input end of the translation reconstruction initial model; determining training samples, wherein the training samples comprise information of a plurality of first sample object sets and information of a plurality of second sample object sets, the information of any sample object set comprises information of all sample objects in any sample object set and relation information among all sample objects, and a common object is arranged between any first sample object set and another second sample object set; respectively determining the characteristic information of each sample object in each sample object set through the characteristic extraction initial model, respectively updating the characteristic information of a common object in any sample object set through the translation reconstruction initial model, and obtaining the updated characteristic information of the common object in any sample object set, wherein the updated characteristic information contains the semantic characteristics of the common object in another sample object set; adjusting the initial feature extraction model according to the feature information of each sample object obtained by the initial feature extraction model to obtain a final feature extraction model as the preset feature extraction model; and adjusting the parameter value in the translation reconstruction initial model according to the updated characteristic information of the common object in any sample object set obtained by the translation reconstruction model so as to obtain a final translation reconstruction model as the preset translation reconstruction model.

Or, the model training unit 16 is configured to determine a feature extraction initial model and a translation reconstruction initial model, where an output end of the feature extraction initial model is connected to an input end of the translation reconstruction initial model, and an output end of the translation reconstruction initial model is connected to the classification module; determining training samples, wherein the training samples comprise information of a plurality of first sample object sets, information of a plurality of second sample object sets and type marking information of each sample object in each sample object set, the information of any sample object set comprises information of each sample object in any sample object set and relation information among the sample objects, and a common object is arranged between any first sample object set and the other second sample object set; respectively determining the characteristic information of each sample object in each sample object set through the characteristic extraction initial model, respectively updating the characteristic information of a common object in any sample object set through the translation reconstruction initial model, obtaining the characteristic information of the common object in any sample object set, wherein the characteristic information of the common object in any sample object set comprises the semantic characteristics of the common object in another sample object set, and determining the type of each sample object through the classification module; and adjusting parameter values in the feature extraction initial model and the translation reconstruction initial model according to the type of each sample object obtained by the classification module and the type standard information of each sample object in the training sample to obtain a final feature extraction model and a final translation reconstruction model which are respectively the preset feature extraction model and the preset translation reconstruction model.

Further, the model training unit 16 is further configured to stop adjusting the fixed parameter value when the number of times of adjusting the parameter value is equal to a preset number of times, or if a difference between the currently adjusted fixed parameter value and the last adjusted fixed parameter value is smaller than a threshold.

It can be seen that, in the object processing system of this embodiment, the dividing unit 11 divides a plurality of objects into a plurality of object sets according to object behavior information applied to a target function, the feature extracting unit 12 extracts overall feature information of each object set, if a common object exists between at least two object sets in an object set, the feature updating unit 13 needs to update initial feature information of the common object in any one of the at least two object sets, so that updated overall feature information of the common object in any one object set includes semantic features of the common object in other object sets, and then the object processing unit 14 performs object classification or search for similarity after obtaining final feature information of each object according to updated overall feature information of the plurality of object sets. Therefore, initial characteristic information obtained by some objects (namely common objects) in different object sets is fused with each other, so that the final characteristic information of the common objects is more accurate, and classification of the objects according to the final characteristic information of the objects is more accurate or the searched similarity is more accurate.

An embodiment of the present invention further provides a computer device, which is schematically shown in fig. 13, and the computer device may generate a relatively large difference due to different configurations or performances, and may include one or more Central Processing Units (CPUs) 20 (e.g., one or more processors) and a memory 21, and one or more storage media 22 (e.g., one or more mass storage devices) for storing the application programs 221 or the data 222. Wherein the memory 21 and the storage medium 22 may be a transient storage or a persistent storage. The program stored on the storage medium 22 may include one or more modules (not shown), each of which may include a series of instructions operating on a computer device. Still further, the central processor 20 may be configured to communicate with the storage medium 22 to execute a series of instruction operations in the storage medium 22 on a computer device.

Specifically, the application 221 stored in the storage medium 22 includes an application for object processing, and the program may include the information obtaining unit 10, the dividing unit 11, the feature extracting unit 12, the feature updating unit 13, the object processing unit 14, the heterogeneous map unit 15, and the model training unit 16 in the object processing system, which is not described herein again. Further, the central processor 20 may be configured to communicate with the storage medium 22, and execute a series of operations corresponding to the application program of the object processing stored in the storage medium 22 on the computer device.

The computer device may also include one or more power supplies 23, one or more wired or wireless network interfaces 24, and/or one or more operating systems 223, such as Windows Server, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, and the like.

The steps performed by the object processing system described above in the method embodiment described above may be based on the structure of the computer device shown in fig. 13.

Embodiments of the present invention further provide a computer-readable storage medium, which stores a plurality of computer programs, where the computer programs are adapted to be loaded by a processor and execute the object processing method performed by the object processing system.

The embodiment of the invention also provides computer equipment, which comprises a processor and a memory; the memory is used for storing a plurality of computer programs which are used for being loaded by the processor and executing the object processing method executed by the object processing system; the processor is configured to implement each of the plurality of computer programs.

Those skilled in the art will appreciate that all or part of the steps in the methods of the above embodiments may be implemented by associated hardware instructed by a program, which may be stored in a computer-readable storage medium, and the storage medium may include: read Only Memory (ROM), Random Access Memory (RAM), magnetic or optical disks, and the like.

The object processing method, the object processing system, the computer readable storage medium and the computer device provided by the embodiments of the present invention are described in detail above, and a specific example is applied in the present disclosure to explain the principle and the implementation of the present invention, and the description of the above embodiments is only used to help understanding the method and the core idea of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.

Claims

1. An object processing method, comprising:

respectively updating initial feature information of a common object in any object set of at least two object sets to obtain updated overall feature information of the object set, wherein the updated feature information of the common object in any object set has semantic features of the common object in other object sets, and the other object sets are object sets except the object set in any object set in at least two object sets;

2. The method of claim 1, wherein after obtaining the object behavior information for the target functionality application, the method further comprises:

generating a heterogeneous graph according to the object behavior information, wherein the heterogeneous graph comprises a plurality of nodes and edges among the nodes, the nodes represent an object, and the edges among any two nodes are used for representing the relationship among the objects represented by any two nodes;

dividing the plurality of objects into a plurality of object sets according to the object behavior information, specifically including: and dividing the heterogeneous graph to obtain heterogeneous sub-graph sets, wherein at least two heterogeneous sub-graph sets in the heterogeneous sub-graph sets have a common node.

3. The method according to claim 2, wherein the dividing the heterogeneous composition to obtain a heterogeneous sub-graph set specifically comprises: and when the relationship types among the objects represented by certain nodes in the heterogeneous graph belong to the same type, dividing the certain nodes into the same heterogeneous subgraph.

4. The method of claim 2, wherein the extracting the global feature information of each object set respectively comprises:

calling a preset feature extraction model;

and the preset feature extraction model respectively extracts the overall feature information of each heterogeneous subgraph according to the object information represented by each node in each heterogeneous subgraph and the relationship information between the objects.

5. The method according to claim 4, wherein the updating the initial feature information of the common object in any one of the at least two object sets respectively comprises:

calling a preset translation reconstruction model;

the preset translation reconstruction model multiplies a matrix formed by initial characteristic information of a common node in any one heterogeneous subgraph of the at least two heterogeneous subgraphs by a conversion matrix to obtain translated characteristics of the common node in any one heterogeneous subgraph, and the translated characteristics of the common node in any one heterogeneous subgraph are aligned with the initial characteristic information of the common node in other heterogeneous subgraphs;

the preset translation reconstruction model enables a matrix formed by translated features of common nodes in any heterogeneous subgraph to pass through a conversion neural network, and the obtained reconstructed features of the common nodes in any heterogeneous subgraph are aligned with the initial feature information of the common nodes in any heterogeneous subgraph;

and the reconstructed characteristic of the common node in any heterogeneous subgraph is the updated characteristic information of the common node in any heterogeneous subgraph.

6. The method of claim 5, wherein the method further comprises:

determining a feature extraction initial model and a translation reconstruction initial model, wherein the output end of the feature extraction initial model is connected to the input end of the translation reconstruction initial model;

determining training samples, wherein the training samples comprise information of a plurality of first sample object sets and information of a plurality of second sample object sets, the information of any sample object set comprises information of all sample objects in any sample object set and relation information among all sample objects, and a common object is arranged between any first sample object set and another second sample object set;

respectively determining the characteristic information of each sample object in each sample object set through the characteristic extraction initial model, respectively updating the characteristic information of a common object in any sample object set through the translation reconstruction initial model, and obtaining the updated characteristic information of the common object in any sample object set, wherein the updated characteristic information contains the semantic characteristics of the common object in another sample object set;

adjusting the initial feature extraction model according to the feature information of each sample object obtained by the initial feature extraction model to obtain a final feature extraction model as the preset feature extraction model; and adjusting the parameter value in the translation reconstruction initial model according to the updated characteristic information of the common object in any sample object set obtained by the translation reconstruction model so as to obtain a final translation reconstruction model as the preset translation reconstruction model.

7. The method of claim 5, wherein the method further comprises:

determining a feature extraction initial model and a translation reconstruction initial model, wherein the output end of the feature extraction initial model is connected to the input end of the translation reconstruction initial model, and the output end of the translation reconstruction initial model is connected to a classification module;

determining training samples, wherein the training samples comprise information of a plurality of first sample object sets, information of a plurality of second sample object sets and type marking information of each sample object in each sample object set, the information of any sample object set comprises information of each sample object in any sample object set and relation information among the sample objects, and a common object is arranged between any first sample object set and the other second sample object set;

respectively determining the characteristic information of each sample object in each sample object set through the characteristic extraction initial model, respectively updating the characteristic information of a common object in any sample object set through the translation reconstruction initial model, obtaining the characteristic information of the common object in any sample object set, wherein the characteristic information of the common object in any sample object set comprises the semantic characteristics of the common object in another sample object set, and determining the type of each sample object through the classification module;

and adjusting parameter values in the feature extraction initial model and the translation reconstruction initial model according to the type of each sample object obtained by the classification module and the type standard information of each sample object in the training sample to obtain a final feature extraction model and a final translation reconstruction model which are respectively the preset feature extraction model and the preset translation reconstruction model.

8. The method of claim 7, wherein the adjusting of the fixed parameter value is stopped when the number of times of adjustment to the parameter value is equal to a preset number of times or if a difference between a currently adjusted fixed parameter value and a last adjusted fixed parameter value is less than a threshold value.

9. The method of any of claims 1 to 8, wherein the updated global feature information for any of the at least two sets of objects comprises: updated feature information of the common object and initial feature information of the non-common object;

acquiring final feature information of each object in any object set according to the updated overall feature information of any object set, wherein the method specifically comprises the following steps:

taking the average characteristic of the updated characteristic information of the common object in the at least two object sets as the final characteristic information of the common object;

and taking the initial characteristic information of the non-common object in any object set as the final characteristic information of the non-common object.

10. An object handling system, comprising:

the characteristic updating unit is used for respectively updating initial characteristic information of a common object in any object set of at least two object sets to obtain updated overall characteristic information of the object set, wherein the updated characteristic information of the common object in any object set has semantic characteristics of the common object in other object sets, and the other object sets are object sets except the object set in any object set in at least two object sets;

11. A computer-readable storage medium, characterized in that it stores a plurality of computer programs adapted to be loaded by a processor and to execute the object processing method according to any one of claims 1 to 9.

12. A computer device comprising a processor and a memory;

the memory is used for storing a plurality of computer programs for being loaded by the processor and executing the object processing method according to any one of claims 1 to 9; the processor is configured to implement each of the plurality of computer programs.