CN110083778A - The figure convolutional neural networks construction method and device of study separation characterization - Google Patents

Abstract

The invention discloses a method and device for constructing a graph convolutional neural network that learns to separate representations, wherein the method includes: performing probability modeling on the formation process of an input graph, and generating probabilities that describe multiple potential factors that may lead to the formation of an edge Generate a model; use the derivable dynamic EM algorithm to reason in each convolution layer through the probability generation model, and obtain the factors corresponding to each neighbor of each node to separate the neighbor nodes; in each convolution layer, According to the neighbor nodes of different factors, a representation describing different aspects of each node is constructed. The method can generate representations that comprehensively and accurately describe multiple aspects of each data point in the graph, based on each factor.

Description

Method and device for constructing graph convolutional neural network for learning separation representation

technical field

The present invention relates to the technical field of social network analysis, in particular to a method and device for constructing a graph convolutional neural network for learning and separating representations.

Background technique

At present, the graph neural network represented by the graph convolutional network is a new generation of end-to-end deep learning technology for processing complex graph-structured data such as social networks and information networks. However, the existing graph neural network assumes that the formation of edges in the graph is driven by the same single factor, so it cannot capture the diverse causes behind the actual data.

SUMMARY OF THE INVENTION

The present invention aims to solve one of the technical problems in the related art at least to a certain extent.

Therefore, an object of the present invention is to propose a method for constructing a graph convolutional neural network that learns to separate representations, which can generate representations that can comprehensively and accurately describe multiple aspects of each data point in the graph.

Another object of the present invention is to propose a graph convolutional neural network construction device for learning separate representations.

In order to achieve the above purpose, an embodiment of the present invention proposes a method for constructing a graph convolutional neural network that learns to separate representations, including: probabilistically modeling the formation process of the input graph, and generating and describing multiple graphs that may lead to the formation of an edge. The probability generation model of latent factor; Use derivable dynamic EM algorithm (Expectation-Maximization, maximum expectation algorithm) in each convolutional layer by described probability generation model to carry out reasoning, obtain the factor corresponding to each neighbor of each node , so as to separate the neighbor nodes; in each of the convolutional layers, according to the neighbor nodes of different factors, a representation describing different aspects of each node is constructed.

The method for constructing a graph convolutional neural network for learning and separating representations in the embodiment of the present invention considers multiple factors behind the formation of a graph, separates these factors, obtains a more accurate and comprehensive representation, and retains the graph when separating each factor. The neural network supports the advantages of end-to-end learning and inductive learning. After separating each factor, it can generate representations that can comprehensively and accurately describe multiple aspects of each data point in the graph according to each factor.

In addition, the method for constructing a graph convolutional neural network for learning and separating representations according to the above-mentioned embodiments of the present invention may also have the following additional technical features:

Further, in an embodiment of the present invention, it also includes: stacking multiple convolutional layers, so as to utilize a preset high-order topology.

Further, in one embodiment of the present invention, each facet corresponds to a separated factor.

Further, in an embodiment of the present invention, the factors of the input graph are plural in number.

In order to achieve the above purpose, another embodiment of the present invention proposes a graph convolutional neural network construction device for learning to separate representations, including: a modeling module, which is used to perform probabilistic modeling on the formation process of the input graph, and generate multiple descriptions A probability generation model of potential factors that may lead to the formation of an edge; the reasoning module is used to use the derivable dynamic EM algorithm in each convolutional layer to reason through the probability generation model, and obtain all the neighbors of each node The corresponding factor is used to separate the neighbor nodes; the construction module is used to construct a representation describing different aspects of each node according to the neighbor nodes of different factors in each convolutional layer.

The graph convolutional neural network construction device for learning and separating representations in the embodiment of the present invention considers multiple factors behind the formation of a graph, separates these factors, obtains more accurate and comprehensive representations, and retains graphs when separating each factor. The neural network supports the advantages of end-to-end learning and inductive learning. After separating each factor, it can generate representations that can comprehensively and accurately describe multiple aspects of each data point in the graph according to each factor.

In addition, the device for constructing a graph convolutional neural network for learning and separating representations according to the above-mentioned embodiments of the present invention may also have the following additional technical features:

Further, in one embodiment of the present invention, it also includes: a stacking module, configured to stack a plurality of each of the convolutional layers, so as to utilize a preset high-order topology.

Further, in one embodiment of the present invention, each facet corresponds to a separated factor.

Further, in an embodiment of the present invention, the factors of the input graph are plural in number.

Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.

Description of drawings

The above and/or additional aspects and advantages of the present invention will become apparent and easy to understand from the following description of the embodiments in conjunction with the accompanying drawings, wherein:

Fig. 1 is a flowchart of a method for constructing a graph convolutional neural network for learning separation representations according to an embodiment of the present invention;

Fig. 2 is a flow chart of a method for constructing a graph convolutional neural network for learning separation representations according to a specific embodiment of the present invention;

FIG. 3 is a schematic structural diagram of a graph convolutional neural network construction device for learning separation representations according to an embodiment of the present invention.

Detailed ways

The following describes in detail the embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the figures are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

The method and device for constructing a graph convolutional neural network for learning and separating representations according to an embodiment of the present invention will be described below with reference to the accompanying drawings. First, the method for constructing a graph convolutional neural network for learning and separating representations according to an embodiment of the present invention will be described with reference to the accompanying drawings .

FIG. 1 is a flowchart of a method for constructing a graph convolutional neural network for learning separation representations according to an embodiment of the present invention.

As shown in Figure 1, the graph convolutional neural network construction method for learning separate representations includes the following steps:

In step S101, probabilistic modeling is performed on the formation process of the input graph, and a probability generation model describing multiple potential factors that may lead to the formation of an edge is generated.

It can be understood that, as shown in FIG. 2 , firstly, a probabilistic modeling is performed on the formation process of the input graph, and the established probabilistic generation model describes multiple potential factors that may lead to the formation of an edge.

Specifically, the reasoning module based on the probabilistic generative model, given a node and its neighbors in the graph, can unsupervisedly discover the potential factors that promote the formation of each edge, and classify the neighbors according to their corresponding factors or separation.

Wherein, in one embodiment of the present invention, the factors of the input graph are plural in number.

In step S102 , the probabilistic generation model is used to perform inference in each convolutional layer using a differentiable dynamic EM algorithm to obtain factors corresponding to each neighbor of each node, so as to separate the neighbor nodes.

It can be understood that, as shown in Figure 2, in each convolutional layer, according to the established probability generation model, the derivable dynamic EM algorithm is used for reasoning, and the factors corresponding to each neighbor of a node are reasoned, and the Neighbors separated.

In step S103, in each convolutional layer, a representation describing different aspects of each node is constructed according to neighboring nodes with different factors.

It can be understood that, as shown in Figure 2, in each convolutional layer, according to the neighbors corresponding to different factors obtained in the previous step, a representation describing the different sides of the node is constructed, and each side corresponds to a separated factor.

Specifically, the embodiment of the present invention proposes a new graph convolution layer applying factor separation technology, and the graph convolution layer can output to each node a representation that can accurately and comprehensively describe its multiple aspects. That is to say, the graph convolution layer in the embodiment of the present invention applies the factor separation technology, and after the factor separation, multiple graph convolution operations are applied to process information corresponding to each factor in parallel and independently.

Among them, factor separation technology is an unsupervised discovery of potential factors that promote the formation of each edge after a node and its neighbors are given in the graph, and the neighbors are classified/separated according to their corresponding factors. technology.

In specific applications, in the recommendation system, more comprehensive and accurate user portraits are automatically generated; and in the recommendation system, the interaction between users, items and other individuals naturally forms a picture, through the embodiment of the present invention The method can more accurately and comprehensively capture multiple interest points or demand points of users.

Further, in an embodiment of the present invention, the method of the embodiment of the present invention further includes: stacking multiple convolutional layers each, so as to utilize a preset high-order topology.

It can be understood that the embodiment of the present invention effectively utilizes the high-order topology structure in the figure by stacking multiple above-mentioned convolutional layers.

Specifically, an embodiment of the present invention proposes a graph convolutional neural network superimposed with multiple new graph convolution layers, which can further utilize additional information such as high-order topological structures in the graph. That is to say, the graph convolutional neural network of the embodiment of the present invention superimposes multiple new graph convolutional layers described above to further utilize additional information such as high-order topological structures in the graph.

To sum up, the embodiment of the present invention mainly aims at the challenges brought by trying to find and separate multiple factors when performing graph convolution, and proposes targeted measures, in the hope that the output of the improved graph convolutional neural network can be more accurate and comprehensive Describe the representation of a data point:

(1) Challenge 1: Graph data usually does not label the specific factors that promote the formation of an edge. For this purpose, the embodiment of the present invention proposes an unsupervised technology based on a probability generation model to infer the latent factor corresponding to each edge.

(2) Challenge 2: How to maintain the two advantages of the graph neural network while performing complex reasoning - supporting end-to-end learning and supporting inductive learning (extrapolating results to new data points that have not been seen). For this purpose, the embodiment of the present invention describes the probabilistic reasoning process as a differentiable (to support end-to-end) and dynamically executed (to support induction) EM algorithm.

According to the graph convolutional neural network construction method for learning separation representation proposed by the embodiment of the present invention, considering that there may be multiple factors that contribute to the formation of a graph, multiple potential factors can be deduced unsupervisedly and combined Separation, and after separating each factor, it can be used to generate a representation that can comprehensively and accurately describe multiple aspects of each data point in the graph, thus considering that the number of factors that promote the formation of a graph may be multiple, through the These different factors are separated when performing graph convolution, and then a representation that can more accurately and comprehensively describe multiple different aspects of each data point in the graph is obtained.

Next, a graph convolutional neural network construction device for learning separation representation proposed according to an embodiment of the present invention will be described with reference to the accompanying drawings.

Fig. 3 is a schematic structural diagram of a graph convolutional neural network construction device for learning separation representations according to an embodiment of the present invention.

As shown in FIG. 3 , the device 10 for constructing a graph convolutional neural network for learning separate representations includes: a modeling module 100 , an inference module 200 and a construction module 300 .

Wherein, the modeling module 100 is used to perform probability modeling on the formation process of the input graph, and generate a probability generation model describing multiple potential factors that may lead to the formation of an edge. The inference module 200 is used to perform inference using a differentiable dynamic EM algorithm in each convolutional layer through the probability generation model, and obtain factors corresponding to each neighbor of each node, so as to separate the neighbor nodes. The construction module 300 is used to construct representations describing different aspects of each node according to neighboring nodes of different factors in each convolutional layer. The device 10 in the embodiment of the present invention can generate representations that can comprehensively and accurately describe multiple aspects of each data point in the graph according to various factors.

Further, in an embodiment of the present invention, the device 10 of the embodiment of the present invention further includes: a superposition module. Among them, the stacking module is used to stack multiple convolutional layers to utilize a preset high-order topology.

Further, in one embodiment of the present invention, each facet corresponds to a separated factor.

Further, in one embodiment of the present invention, the factors of the input graph are plural in number.

It should be noted that the foregoing explanations for the embodiment of the method for constructing a graph convolutional neural network for learning separate representations are also applicable to the device for constructing a graph convolutional neural network for learning separate representations in this embodiment, and details are not repeated here.

According to the graph convolutional neural network construction device for learning separate representations proposed by the embodiment of the present invention, considering that there may be multiple factors that contribute to the formation of a graph, multiple potential factors can be deduced unsupervisedly and combined Separation, and after separating each factor, it can be used to generate a representation that can comprehensively and accurately describe multiple aspects of each data point in the graph, thus considering that the number of factors that promote the formation of a graph may be multiple, through the These different factors are separated when performing graph convolution, and then a representation that can more accurately and comprehensively describe multiple different aspects of each data point in the graph is obtained.

In describing the present invention, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", " Back", "Left", "Right", "Vertical", "Horizontal", "Top", "Bottom", "Inner", "Outer", "Clockwise", "Counterclockwise", "Axial", The orientation or positional relationship indicated by "radial", "circumferential", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying the referred device or element Must be in a particular orientation, be constructed in a particular orientation, and operate in a particular orientation, and therefore should not be construed as limiting the invention.

In addition, the terms "first" and "second" are only used for descriptive purposes, and should not be construed as indicating or implying relative importance or implying the number of indicated technical features. Thus, a feature delimited with "first", "second" may expressly or implicitly include at least one of that feature. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise expressly and specifically defined.

In the present invention, unless otherwise clearly specified and limited, terms such as "installation", "connection", "connection" and "fixation" should be understood in a broad sense, for example, it can be a fixed connection or a detachable connection , or integrated; it may be mechanically connected or electrically connected; it may be directly connected or indirectly connected through an intermediary, and it may be the internal communication of two components or the interaction relationship between two components, unless otherwise specified limit. Those of ordinary skill in the art can understand the specific meanings of the above terms in the present invention according to specific situations.

In the present invention, unless otherwise clearly specified and limited, the first feature may be in direct contact with the first feature or the first and second feature indirectly through an intermediary. touch. Moreover, "above", "above" and "above" the first feature on the second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. "Below", "beneath" and "beneath" the first feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is less horizontally than the second feature.

In the description of this specification, descriptions with reference to the terms "one embodiment", "some embodiments", "example", "specific examples", or "some examples" mean that specific features described in connection with the embodiment or example , structure, material or characteristic is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the described specific features, structures, materials or characteristics may be combined in any suitable manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine different embodiments or examples and features of different embodiments or examples described in this specification without conflicting with each other.

Although the embodiments of the present invention have been shown and described above, it should be understood that the above-mentioned embodiments are exemplary and should not be construed as limiting the present invention. Embodiments are subject to variations, modifications, substitutions and variations.

Claims (8)

1. A method for building a graph convolutional neural network for learning to separate representations, characterized in that it comprises: Probabilistically model the formation process of the input graph to generate a probabilistic generative model describing multiple potential factors that may lead to the formation of an edge; Using the derivable dynamic EM algorithm to perform inference in each convolutional layer through the probability generation model, and obtain the corresponding factors of each neighbor of each node to separate the neighbor nodes; In each convolutional layer, a representation describing different aspects of each node is constructed according to the neighbor nodes with different factors. 2. The graph convolutional neural network construction method of learning separation representation according to claim 1, is characterized in that, also comprises: A plurality of each of the convolutional layers is stacked to utilize a preset high-order topology. 3. The graph convolutional neural network construction method of learning separation representation according to claim 1, wherein each side corresponds to a separated factor. 4. The graph convolutional neural network construction method of learning separation representation according to claim 1, characterized in that, the factors of the input graph are complex numbers. 5. A graph convolutional neural network construction device for learning to separate representations, characterized in that it comprises: The modeling module is used for probabilistic modeling of the formation process of the input graph, and generates a probability generative model describing multiple potential factors that may lead to the formation of an edge; The reasoning module is used to use the derivable dynamic EM algorithm to reason in each convolution layer through the probability generation model, and obtain the factors corresponding to the neighbors of each node to separate the neighbor nodes; The construction module is used for constructing representations describing different aspects of each node according to the neighboring nodes of different factors in each of the convolutional layers. 6. The graph convolution neural network construction device of learning separation representation according to claim 5, is characterized in that, also comprises: A stacking module, configured to stack a plurality of each of the convolutional layers so as to utilize a preset high-order topology. 7. The graph convolutional neural network construction device for learning and separating representations according to claim 5, wherein each side corresponds to a factor that has been separated. 8. The device for constructing a graph convolutional neural network for learning and separating representations according to claim 5, wherein the factors of the input graph are plural in number.