Data evaluation method and training method of quantum graph convolutional neural network model

Abstract

The present invention provides a data evaluation method and a training method for a quantum graph convolutional neural network model, which obtains a user evaluation data set to be processed; processes the user evaluation data set to be processed to obtain a quantum state to be processed corresponding to the user evaluation data set to be processed; inputs the quantum state to be processed into a pre-trained quantum graph convolutional neural network model, and outputs an evaluation result of the user evaluation data set to be processed. This method can process the user evaluation data set to be processed into a corresponding quantum state to be processed, and output the corresponding evaluation result through the quantum graph convolutional neural network model. Compared with the classical machine learning method, the method using the quantum graph convolutional neural network model has stronger processing capabilities.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a data evaluation method and a training method for a quantum graph convolutional neural network model. Background Art

[0002] At present, e-commerce is developing in full swing. Thousands of people will choose to shop online to varying degrees. The development of e-commerce platforms has been fully integrated into all areas of production and life in my country, becoming an important force in improving people's quality of life and promoting economic and social development. When people buy a product, they often check the comments of previous buyers on the product after purchase. However, some individual comments are valuable comments, while some are deceptive comments that induce buyers to buy the product. For example, the same person may use different accounts to comment. Connecting users who comment on the product forms a product user evaluation network graph, that is, the nodes in the graph represent product evaluators. How to distinguish whether the product evaluators in the network graph are benign individuals or fraudulent individuals is of great significance to product buyers. At present, the product user evaluation network is mainly processed by classical machine learning, which has limited processing capabilities. Summary of the invention

[0003] The purpose of the present invention is to provide a data evaluation method and a training method for a quantum graph convolutional neural network model to improve the processing capability of a user evaluation data set to be processed.

[0004] The present invention provides a data evaluation method, which includes: obtaining a user evaluation data set to be processed; processing the user evaluation data set to be processed to obtain a quantum state to be processed corresponding to the user evaluation data set to be processed; inputting the quantum state to be processed into a pre-trained quantum graph convolutional neural network model, and outputting an evaluation result of the user evaluation data set to be processed.

[0005] Furthermore, the step of processing the user evaluation data set to be processed to obtain the quantum state to be processed corresponding to the user evaluation data set to be processed includes: determining the user evaluation network graph and the first multi-dimensional node feature vector based on the user evaluation data set to be processed; constructing an initial adjacency matrix and a degree matrix based on the user evaluation network graph; normalizing the initial adjacency matrix based on the degree matrix to obtain a standardized adjacency matrix; and determining the quantum state to be processed corresponding to the user evaluation data set to be processed based on the standardized adjacency matrix.

[0006] Furthermore, based on the standardized adjacency matrix, the step of determining the quantum state to be processed corresponding to the user evaluation data set to be processed includes: based on the standardized adjacency matrix, processing the first multidimensional node feature vector to obtain the second multidimensional node feature vector; performing dimensionality reduction processing on the second multidimensional node feature vector to obtain the third multidimensional node feature vector; encoding processing on the third multidimensional node feature vector to obtain the quantum state to be processed corresponding to the user evaluation data set to be processed.

[0007] Furthermore, the step of encoding the third multi-dimensional node feature vector to obtain the quantum state to be processed corresponding to the user evaluation data set to be processed includes: for each node feature vector in the third multi-dimensional node feature vector, taking the arcsine of the node feature vector to obtain the quantum gate rotation angle corresponding to the node feature vector; encoding the multiple quantum gate rotation angles corresponding to the third multi-dimensional node feature vector through a preset rotating quantum gate to obtain the quantum state to be processed corresponding to the user evaluation data set to be processed.

[0008] Furthermore, the quantum graph convolutional neural network model includes multiple layers of graph convolution modules, measurement modules and fully connected layers connected in sequence; the first layer of graph convolution module is used to perform convolution processing on the received standardized adjacency matrix, the first multi-dimensional node feature vector and the quantum bits corresponding to the quantum state to be processed, and output the first aggregate node feature vector; for each layer of graph convolution module except the first layer of graph convolution module, the graph convolution module of this layer is used to perform convolution processing on the received standardized adjacency matrix, the aggregate node feature vector output by the previous layer of graph convolution module and the quantum bits corresponding to the quantum state to be processed, and output the second aggregate node feature vector; the measurement module is used to measure the expected value of each quantum bit based on the second aggregate node feature vector output by the last layer of graph convolution module; the fully connected layer is used to perform fully connected processing on the expected value of each quantum bit, and output the evaluation result of the user evaluation data set to be processed.

[0009] Furthermore, each layer of graph convolution module includes multiple unitary modules; each unitary module is composed of preset rotation quantum gates.

[0010] The present invention provides a training method for a quantum graph convolutional neural network model, which includes: obtaining a preset data set; wherein the preset data set includes: a training sample set and a test sample set; each user evaluation data training sample in the training sample set carries a first evaluation label; inputting the preset data set into an initial quantum graph convolutional neural network model, and outputting a first training result corresponding to the preset data set; selecting a second training result corresponding to the training sample set from the first training result; based on a preset loss function and the first evaluation label, calculating a loss value of a prediction result corresponding to the second training result, and training the initial quantum graph convolutional neural network model based on the loss value to obtain a trained quantum graph convolutional neural network model.

[0011] Furthermore, each user evaluation data test sample in the test sample set carries a second evaluation label, and the method also includes: inputting a preset data set into the quantum graph convolutional neural network model, and outputting a first test result corresponding to the preset data set; selecting a second test result corresponding to the test sample set from the first test result; for each user evaluation data test sample in the test sample set, comparing the second evaluation label corresponding to the user evaluation data test sample with the corresponding second test result to obtain a comparison result; and determining the accuracy of the quantum graph convolutional neural network model based on the comparison result corresponding to each user evaluation data test sample.

[0012] The present invention provides a data evaluation device, which includes: a first acquisition module, used to acquire a user evaluation data set to be processed; a processing module, used to process the user evaluation data set to be processed to obtain a quantum state to be processed corresponding to the user evaluation data set to be processed; a first output module, used to input the quantum state to be processed into a pre-trained quantum graph convolutional neural network model, and output an evaluation result of the user evaluation data set to be processed.

[0013] The present invention provides a training device for a quantum graph convolutional neural network model, which includes: a second acquisition module, used to acquire a preset data set; wherein the preset data set includes: a training sample set and a test sample set; each user evaluation data training sample in the training sample set carries a first evaluation label; an input module, used to input the preset data set into the initial quantum graph convolutional neural network model, and output a first training result corresponding to the preset data set; a selection module, used to select a second training result corresponding to the training sample set from the first training result; a calculation module, used to calculate the loss value of the prediction result corresponding to the second training result based on a preset loss function and the first evaluation label, and train the initial quantum graph convolutional neural network model based on the loss value to obtain a trained quantum graph convolutional neural network model.

[0014] An electronic device provided by the present invention includes a processor and a memory, the memory stores machine executable instructions that can be executed by the processor, and the processor executes the machine executable instructions to implement any of the above-mentioned data evaluation methods, or any of the above-mentioned quantum graph convolutional neural network model training methods.

[0015] The present invention provides a machine-readable storage medium, which stores machine-executable instructions. When the machine-executable instructions are called and executed by a processor, the machine-executable instructions prompt the processor to implement any of the above-mentioned data evaluation methods, or any of the above-mentioned quantum graph convolutional neural network model training methods.

[0016] The data evaluation method and the training method of the quantum graph convolutional neural network model provided by the present invention obtain a user evaluation data set to be processed; process the user evaluation data set to be processed to obtain a quantum state to be processed corresponding to the user evaluation data set to be processed; input the quantum state to be processed into a pre-trained quantum graph convolutional neural network model, and output the evaluation result of the user evaluation data set to be processed. This method can process the user evaluation data set to be processed into the corresponding quantum state to be processed, and output the corresponding evaluation result through the quantum graph convolutional neural network model. Compared with the classical machine learning method, the method using the quantum graph convolutional neural network model has stronger processing capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 A flow chart of a data evaluation method provided by an embodiment of the present invention;

[0019] Figure 2 A schematic diagram of a user evaluation network diagram provided by an embodiment of the present invention;

[0020] Figure 3 A schematic diagram of a quantum circuit provided by an embodiment of the present invention;

[0021] Figure 4 A schematic diagram of the overall architecture of a quantum graph convolutional neural network model provided by an embodiment of the present invention;

[0022] Figure 5 A schematic diagram of the structure of a unitary module provided by an embodiment of the present invention;

[0023] Figure 6 A flowchart of a training method for a quantum graph convolutional neural network model provided by an embodiment of the present invention;

[0024] Figure 7 A schematic diagram of the structure of a data evaluation device provided by an embodiment of the present invention;

[0025] Figure 8 A schematic diagram of the structure of a training device for a quantum graph convolutional neural network model provided by an embodiment of the present invention;

[0026] Fig. 9 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] At present, unstructured data (such as citation networks, etc.) are mainly processed by classical machine learning, such as graph convolutional neural networks. At present, there are some quantum graph convolutional neural networks for processing image data, which mainly express the topological structure of the input graph data directly with quantum circuits. The existing quantum graph convolutional neural network model mainly processes structured data in Euclidean space, and there is little research on how quantum graph convolutional neural networks process data in non-Euclidean space. When processing image data, the current quantum graph convolutional neural network uses amplitude coding. Amplitude coding requires a large number of quantum gates, which is difficult to apply to the current NISQ (noisy intermediate-scale quantum, containing noisy medium-scale quantum computer) equipment. And currently, the commodity user evaluation network is processed by classical machine learning, which has limited processing power, insufficient feature learning, and slow training speed; and the learning of the commodity user evaluation network by the quantum graph convolutional neural network has not yet been involved. Based on this, the embodiment of the present invention provides a data evaluation method and a training method for a quantum graph convolutional neural network model, which can be applied to applications that require processing of commodity user evaluation networks.

[0029] To facilitate understanding of this embodiment, a data evaluation method disclosed in an embodiment of the present invention is first introduced. Figure 1 As shown, the method comprises the following steps:

[0030] Step S102, obtaining a user evaluation data set to be processed;

[0031] The above-mentioned user evaluation data set to be processed can be understood as the relevant evaluation information of the purchased goods by the users after purchasing the goods; the user evaluation data set to be processed usually includes the evaluation information of the goods by different users, and the evaluation information may be valuable information or worthless and deceptive information. In actual implementation, the user evaluation data set to be processed can be obtained first to evaluate whether the user evaluation data set to be processed is valuable.

[0032] Step S104, processing the user evaluation data set to be processed to obtain a quantum state to be processed corresponding to the user evaluation data set to be processed;

[0033] A quantum state is a complex mathematical object that can represent the superposition and entanglement of multiple states. In actual implementation, the user evaluation data set to be processed obtained above can be standardized, dimensionally reduced, encoded, etc. to convert the user evaluation data set to be processed into the corresponding quantum state to be processed.

[0034] Step S106, input the quantum state to be processed into a pre-trained quantum graph convolutional neural network model, and output the evaluation result of the user evaluation data set to be processed.

[0035] The above-mentioned quantum graph convolutional neural network model usually includes a graph convolution module, a fully connected layer, etc.; in actual implementation, after the user evaluation data set to be processed is converted into the corresponding quantum state to be processed, the quantum state to be processed can be input into a pre-trained quantum graph convolutional neural network model, and the evaluation results of the user evaluation data set to be processed are output through the quantum graph convolutional neural network model, wherein the evaluation results may include: valuable comments, deceptive comments, etc.

[0036] The above data evaluation method obtains a user evaluation data set to be processed; processes the user evaluation data set to be processed to obtain a quantum state to be processed corresponding to the user evaluation data set to be processed; inputs the quantum state to be processed into a pre-trained quantum graph convolutional neural network model, and outputs the evaluation result of the user evaluation data set to be processed. This method can process the user evaluation data set to be processed into the corresponding quantum state to be processed, and output the corresponding evaluation result through the quantum graph convolutional neural network model. Compared with the classical machine learning method, the method using the quantum graph convolutional neural network model has stronger processing capabilities.

[0037] The embodiment of the present invention further provides another data evaluation method, which is implemented on the basis of the method in the above embodiment, and comprises the following steps:

[0038] Step 1: Obtain the user evaluation data set to be processed;

[0039] Step 2: determining a user evaluation network graph and a first multi-dimensional node feature vector based on the user evaluation data set to be processed;

[0040] The above-mentioned user evaluation network graph is usually an undirected and unweighted graph, through which the correlation relationship between multiple product evaluator nodes can be represented; the dimension of the above-mentioned first multi-dimensional node feature vector can be set according to actual needs, such as 25 dimensions; in actual implementation, the corresponding user evaluation network graph can be generated according to the obtained user evaluation data set to be processed, as well as the first multi-dimensional node feature vector corresponding to each product evaluator node in the user evaluation network graph, such as the 25-dimensional node feature vector corresponding to each product evaluator node.

[0041] Step 3: Based on the user evaluation network graph, construct the initial adjacency matrix and degree matrix;

[0042] The above-mentioned initial adjacency matrix can represent the correlation between the nearest neighbor product evaluator nodes in the user evaluation network graph, and also includes the correlation between the product evaluator nodes themselves; the initial adjacency matrix can be constructed according to the above-mentioned user evaluation network graph; and then the corresponding degree matrix can be determined according to the initial adjacency matrix. For details, please refer to the relevant technology and will not be repeated here.

[0043] For example, take the user evaluation network diagram of four product users as an example. Figure 2 The diagram of a user evaluation network graph is shown in Figure 1. This is an undirected and unweighted graph. The corresponding adjacency matrix of the graph composed of four product evaluator nodes is as follows:

[0044]

[0045] The adjacency matrix A represents the correlation between the nearest neighbor product reviewer nodes in the user evaluation network graph. Here, the correlation between the product reviewer node itself and itself is added to obtain a new adjacency matrix Where I represents the identity matrix of the same dimension as the adjacency matrix A. The new adjacency matrix That corresponds to the above initial adjacency matrix.

[0046] The corresponding degree matrix can be obtained from the initial adjacency matrix:

[0047]

[0048] Step 4: normalize the initial adjacency matrix based on the degree matrix to obtain a standardized adjacency matrix;

[0049] The above initial adjacency matrix is ​​transformed into Standardize and get the standardized adjacency matrix

[0050] Step 5: Based on the standardized adjacency matrix, determine the quantum state to be processed corresponding to the user evaluation data set to be processed.

[0051] The step five can be implemented by the following steps 50 to 52:

[0052] Step 50: Process the first multi-dimensional node feature vector based on the standardized adjacency matrix to obtain a second multi-dimensional node feature vector;

[0053] The above standardized adjacency matrix The first multi-dimensional node feature vector corresponding to each product evaluator node in the user evaluation network diagram is acted on to aggregate the information of neighboring product evaluator nodes to obtain a second multi-dimensional node feature vector; for example, the first multi-dimensional node feature vector corresponding to each product evaluator node is a 25-dimensional node feature vector, and through the action of the standardized adjacency matrix, the second multi-dimensional node feature vector obtained is usually a new 25-dimensional node feature vector.

[0054] Step 51, performing dimensionality reduction processing on the second multi-dimensional node feature vector to obtain a third multi-dimensional node feature vector;

[0055] The second multi-dimensional node feature vector can be reduced to a specified dimension, such as 20 dimensions, through PCA (Principal Component Analysis) to obtain a third multi-dimensional node feature vector. The dimensionality reduction process can make subsequent node feature vector data encoding more suitable for NISQ devices.

[0056] Step 52: Encode the third multi-dimensional node feature vector to obtain the quantum state to be processed corresponding to the user evaluation data set to be processed.

[0057] The step 52 can be implemented by the following steps 520 to 521:

[0058] Step 520: for each node feature vector in the third multi-dimensional node feature vector, take the arcsine of the node feature vector to obtain the quantum gate rotation angle corresponding to the node feature vector;

[0059] For example, taking the third multi-dimensional node feature vector as having 20 dimensions, the 20-dimensional node feature vector can be expressed as Taking the inverse sine of each element to obtain the rotation angle of the quantum gate is as follows:

[0060] θ i =arcsin(x i );

[0061] Among them, x i represents the i-th characteristic attribute of the product evaluator node, where i represents the i-th characteristic attribute of the i-th rotating quantum gate acceptor node. The above 20-dimensional node feature vectors are 20 feature vectors of the same product evaluator node, and each product evaluator node has its own corresponding 20 feature vectors.

[0062] Step 521, encoding multiple quantum gate rotation angles corresponding to the third multi-dimensional node feature vector through a preset rotation quantum gate, and obtaining a quantum state to be processed corresponding to the user evaluation data set to be processed.

[0063] The above preset rotation quantum gate may include R y (θ) and R x (θ), R y (θ) and R x (θ) The corresponding expressions can refer to the relevant technology; in actual implementation, such as Figure 3 The schematic diagram of a quantum circuit shown in FIG. 1 is a schematic diagram of a quantum circuit based on which a predetermined rotating quantum gate R can be used. y (θ) and R x The angle encoding method of (θ) encodes the third multi-dimensional node feature vector into a quantum state. For a 20-dimensional node feature vector, 10 quantum bits are required. The 20-dimensional node feature vector can be input into Figure 3 The quantum circuit in outputs the corresponding quantum state to be processed.

[0064] Step six, input the quantum state to be processed into the quantum graph convolutional neural network model, and output the evaluation results of the user evaluation data set to be processed.

[0065] The quantum graph convolutional neural network model includes multiple layers of graph convolution modules, measurement modules and fully connected layers connected in sequence; the number of layers of the above graph convolution modules is usually multiple layers.

[0066] The first-layer graph convolution module is used to perform convolution processing on the received standardized adjacency matrix, the first multi-dimensional node feature vector, and the quantum bits corresponding to the quantum state to be processed, and output a first aggregated node feature vector;

[0067] For each layer of graph convolution modules except the first layer of graph convolution modules, the graph convolution modules of this layer are used to perform convolution processing on the received standardized adjacency matrix, the aggregation node feature vector output by the previous layer of graph convolution modules, and the quantum bits corresponding to the quantum state to be processed, and output a second aggregation node feature vector;

[0068] The quantum graph convolutional neural network can use the information of the edges between products provided in the user evaluation network graph to aggregate the product evaluator node information of the product to generate a new node representation and obtain the hidden feature representation of each product evaluator node. For the convenience of explanation, taking the two-layer graph convolution module as an example, the node feature vector output by the first layer of the graph convolution module is Where X 1 is the first multi-dimensional node feature vector, s is the Sigmoid activation function, W 1 A quantum neural network that operates on node feature vectors that integrate graph information, such as Figure 4The overall architecture diagram of a quantum graph convolutional neural network model is shown in the figure. Here, 10 quantum bits are used. It is mainly realized by a unitary module U composed of basic quantum gates acting alternately on two quantum bits. The dotted part in the figure can be repeated multiple times according to the limitations of the computing device used to increase the trainable parameters of the model. Here, it is repeated twice.

[0069] Each layer of graph convolution module includes multiple unitary modules U; each unitary module U is composed of preset rotation quantum gates, such as Figure 5 The schematic diagram of the structure of a unitary module is shown in FIG. x , R y , R z Rotating quantum gate. The node feature vector output by the similar second-layer graph convolution module is Where s is the Sigmoid activation function, and W 2 With W 1 It is also a quantum neural network that operates on the node feature vector of the previous layer that contains graph information.

[0070] The measurement module is used to measure the expected value of each quantum bit based on the second aggregation node feature vector output by the last layer of graph convolution module;

[0071] The fully connected layer is used to perform full connection processing on the expected value of each quantum bit and output the evaluation results of the user evaluation data set to be processed.

[0072] After passing through two layers of graph convolution modules, a node feature vector that aggregates the features of two-order neighbor nodes can be obtained. The PAULI-Z expected values ​​of all quantum bits can be measured through the measurement module, and the obtained results are passed to the fully connected layer. Here, according to the number of categories of the data set nodes is 2, the output dimension of the fully connected layer is selected as 2, and the node feature vector of the commodity user evaluation network is obtained. The numerical information of different sizes of each category belonging to the input quantum graph convolutional neural network model can be determined through this numerical information. The evaluation results of the user evaluation data set to be processed can be determined, such as valuable comments or deceptive comments.

[0073] The above data evaluation method first obtains an initial adjacency matrix according to the user evaluation data set to be processed, and then uses the degree matrix to standardize the initial adjacency matrix, and applies the standardized adjacency matrix to the node feature vector of the commodity user evaluation network graph to aggregate the node neighboring node information to obtain a new node feature vector, and then reduces the node feature dimension through PCA, and then encodes the classical information of the node feature vector into a quantum state through angle encoding for the node feature vector after the dimension reduction, and then processes the quantum state encoding the node feature vector information through the quantum graph convolutional neural network model, and then measures the quantum bits of the quantum system to obtain classical information, and the above steps can be repeated multiple times, indicating the farthest distance that the node feature in the graph can reach or the order of the corresponding node in the graph to aggregate its neighboring nodes. Then the classical information obtained by measuring the quantum bits is passed into a single-layer fully connected layer designed according to the number of categories of the data set nodes, and the different sizes of numerical information belonging to each category output by the commodity user evaluation network node feature vector input quantum graph convolutional neural network model are obtained. This method can process the user evaluation data set to be processed into the corresponding quantum state to be processed, and output the corresponding evaluation results through the quantum graph convolutional neural network model. Compared with the classical machine learning method, the method using the quantum graph convolutional neural network model has stronger processing capabilities and can effectively predict the evaluation categories in the network.

[0074] The embodiment of the present invention provides a training method for a quantum graph convolutional neural network model, such as Figure 6 As shown, the method comprises the following steps:

[0075] Step S602, obtaining a preset data set; wherein the preset data set includes: a training sample set and a test sample set; each user evaluation data training sample in the training sample set carries a first evaluation label;

[0076] The preset data set may be a Fraud product user evaluation network data set, 90% of which may be used as a training sample set, and 10% as a test sample set. Each user evaluation data training sample in the training sample set is pre-labeled with a first evaluation label corresponding to each user evaluation data, which may specifically be a valuable review, a deceptive review, etc.

[0077] Step S604, inputting a preset data set into the initial quantum graph convolutional neural network model, and outputting a first training result corresponding to the preset data set;

[0078] Step S606, selecting a second training result corresponding to the training sample set from the first training result;

[0079] The above-mentioned initial quantum graph convolutional neural network model can be understood as an untrained quantum graph convolutional neural network model; the preset data set can be input into the initial quantum graph convolutional neural network model to obtain a first training result corresponding to each sample in the preset data set, and then the second training result corresponding to the training sample set is selected from the first training result.

[0080] Step S608, based on the preset loss function and the first evaluation label, calculate the loss value of the prediction result corresponding to the second training result, train the initial quantum graph convolutional neural network model based on the loss value, and obtain a trained quantum graph convolutional neural network model.

[0081] After inputting the preset data set into the above-mentioned initial quantum graph convolution module, the first training result corresponding to the preset data set can be obtained, for example, the numerical information of different sizes of the two categories to which each product evaluator node belongs can be obtained. Then, the numerical information of the two categories of all the above-mentioned product evaluator nodes is indexed according to the divided training sample set, and the numerical information of the two categories corresponding to each training sample set node output by the above-mentioned initial quantum graph convolution neural network model module is obtained, and the cross entropy loss function loss is performed on it and the one-hot encoded label information corresponding to the training sample set node. The cross entropy loss function loss is expressed as follows:

[0082]

[0083] Where i represents the category to which the node belongs. For the i-th category, the average value of the above cross entropy loss function of the nodes in all training sample sets is calculated as the loss function of the entire model.

[0084] First, a measurement operator In the parameterized quantum circuit U(θ i ) can be expressed as:

[0085]

[0086] Among them, U(θ i ) represents the above W1 or W2, such as Figure 4 As shown in; U + (θ i ) represents U(θ i ) is the transposed complex conjugate; |0> represents the initial state of the quantum system; θ i Represents the parameters in a quantum graph convolutional neural network that operates on node feature vectors that incorporate graph information.

[0087] Then the expected value function f(θ i ) About the parameterized quantum circuit parameter θ i The gradient of can be expressed as

[0088]

[0089] The above method is called the parameter transfer rule for the parameter analysis gradient of the parameterized quantum circuit operator expectation value. For details, please refer to the relevant technology and will not be repeated here. Based on the parameter transfer rule, the cross entropy loss function obtained in the previous module is calculated. Figure 4 The analytical gradients of each parameter of the quantum graph convolutional neural network model are obtained, and then the parameters of the quantum circuit and the last single-layer fully connected layer are updated using a classical computer. Finally, the product user evaluation network dataset is used to train for multiple epochs, and the loss function is stopped after convergence.

[0090] The training method of the above quantum graph convolutional neural network model obtains a preset data set; wherein the preset data set includes: a training sample set and a test sample set; each user evaluation data training sample in the training sample set carries a first evaluation label; the preset data set is input into the initial quantum graph convolutional neural network model, and the first training result corresponding to the preset data set is output; the second training result corresponding to the training sample set is selected from the first training result; based on the preset loss function and the first evaluation label, the loss value of the prediction result corresponding to the second training result is calculated, and the initial quantum graph convolutional neural network model is trained based on the loss value to obtain a trained quantum graph convolutional neural network model. This method inputs the preset data set into the initial quantum graph convolutional neural network model, selects the second training result corresponding to the training sample set from the first training result, and then trains the initial quantum graph convolutional neural network model based on the loss function and the first evaluation label, so that the trained quantum graph convolutional neural network model has stronger processing capabilities.

[0091] The embodiment of the present invention provides a training method for a quantum graph convolutional neural network model. The method is implemented on the basis of the method of the above embodiment. In the method, each user evaluation data test sample in the test sample set carries a second evaluation label. The method includes the following steps:

[0092] Step 10, obtaining a preset data set; wherein the preset data set includes: a training sample set and a test sample set; each user evaluation data training sample in the training sample set carries a first evaluation label;

[0093] Step 11, inputting the preset data set into the initial quantum graph convolutional neural network model, and outputting the first training result corresponding to the preset data set;

[0094] Step 12: Select a second training result corresponding to the training sample set from the first training result;

[0095] Step thirteen, based on the preset loss function and the first evaluation label, calculate the loss value of the prediction result corresponding to the second training result, train the initial quantum graph convolutional neural network model based on the loss value, and obtain the trained quantum graph convolutional neural network model.

[0096] Step 14: input the preset data set into the quantum graph convolutional neural network model, and output the first test result corresponding to the preset data set;

[0097] Step 15, selecting a second test result corresponding to the test sample set from the first test result;

[0098] Step 16: for each user evaluation data test sample in the test sample set, compare the second evaluation label corresponding to the user evaluation data test sample with the corresponding second test result to obtain a comparison result;

[0099] Step seventeen, determine the accuracy of the quantum graph convolutional neural network model based on the comparison results corresponding to each user evaluation data test sample.

[0100] The preset data set can be input into the trained quantum graph convolutional neural network model to obtain the first test result corresponding to each sample in the preset data set, and then the numerical information part corresponding to the two categories of the test sample set is taken out from the first test result, and then compared with the true category label of each node corresponding to the test sample set (corresponding to the second evaluation label mentioned above), and finally the accuracy of the prediction of the quantum graph convolutional neural network model is output.

[0101] In summary, the data evaluation method in this embodiment may include a data preprocessing module, a standardized initial adjacency matrix module, a feature dimension reduction module, a data encoding module, a quantum graph convolutional neural network model, a loss function construction module, a quantum circuit parameter update module, and a model verification module.

[0102] The data preprocessing module is used to: first download the Fraud dataset, and then divide the product user evaluation network dataset into a training sample set and a test sample set. Then, the degree matrix is ​​obtained based on the graph information of the edges connecting the product user evaluation network and the nodes, and then the degree matrix is ​​used to standardize the initial adjacency matrix, and the standardized standard adjacency matrix is ​​applied to the node feature vector of the product user evaluation network graph, and the node neighboring node information is aggregated to obtain a new node feature vector (corresponding to the second multi-dimensional node feature vector mentioned above).

[0103] The feature dimension reduction module is used to reduce the new node feature vector obtained in the previous module to a specified dimension through PCA to obtain a commodity user evaluation network node feature vector with a lower dimension (corresponding to the third multi-dimensional node feature vector mentioned above), so that the subsequent node feature vector data encoding is more suitable for NISQ equipment.

[0104] The data encoding module is used to: first take the arcsine of the node feature vector information after dimension reduction in the previous module to obtain the rotation angle of the quantum gate, and then use the angle encoding method to pass R y (θ) and R x (θ) The rotation quantum gate acts on the quantum bit to encode the node characteristic data into a quantum state.

[0105] The quantum graph convolutional neural network model is used to: include R through parameterized quantum gates x (θ), R y (θ), R z (θ) Rotation quantum gate, as well as the basic quantum gate CNOT gate supported by quantum computers, are used to construct a quantum neural network to process the node feature vector data of the commodity user evaluation network. The encoding module encodes the quantum state of the node feature data, and finally measures the quantum bits to obtain classical information, thereby obtaining the node feature vector of this layer that collects the feature information of neighboring nodes. The above process can be repeated many times to realize the propagation of node feature information in the commodity user evaluation network, and then the classical information obtained by measuring the quantum bits in the last layer is passed into a single-layer fully connected layer designed according to the number of categories of the data set nodes to obtain the numerical information of different sizes belonging to each category output by the quantum graph convolutional neural network model of the node feature vector of the commodity user evaluation network.

[0106] The loss function module is constructed to: after all nodes of the entire user evaluation network graph pass through the above-mentioned quantum graph convolutional neural network model, the output dimension of each node is the number of categories of the product user evaluation network data set, and then the predicted category numerical information of all the above nodes is indexed according to the divided training sample set, and the predicted category numerical information corresponding to the training sample set nodes is obtained. The cross entropy loss function is performed with the corresponding labels of the nodes in the training sample set to obtain the loss function used to characterize the model performance.

[0107] The module for updating quantum circuit parameters is used to: based on the existing parameterized circuit movement rules, calculate the analytical gradient of the loss function of the previous module with respect to each parameter of the quantum circuit, and then use a classical computer to update the parameters of the quantum circuit. Finally, the product user evaluation network data set is used to train multiple epochs so that the data evaluation method and the training method of the quantum graph convolutional neural network model proposed in this application stop after the loss function converges.

[0108] Finally, the model verification module is used to: input the feature vectors of all nodes in the product user evaluation network diagram into the quantum graph convolutional neural network model trained in the previous module, extract the category prediction output part corresponding to the test sample set in the model output, and then compare it with the true category label of each node corresponding to the test sample set, and finally output the accuracy of the model prediction.

[0109] The training method of the quantum graph convolutional neural network model first calculates the cross entropy loss function of the output prediction result corresponding to the training sample set and the corresponding first evaluation label, then updates the quantum circuit parameters to minimize the cross entropy loss function until the loss function reaches the minimum value, and finally obtains the quantum graph convolutional neural network model for the commodity user evaluation network, and then uses the trained quantum graph convolutional neural network model for the test sample set to test the model performance. The reliability and effectiveness of the quantum graph convolutional neural network model can be guaranteed.

[0110] This method mainly uses a quantum graph convolutional neural network model to process the node feature vectors that collect graph information. Before encoding the node feature vector data of the commodity user evaluation network, the node feature vector is reduced to a specified dimension through the PCA method to obtain a lower-dimensional commodity user evaluation network node feature vector. This method is particularly useful for network graphs with large node feature dimensions, making the subsequent node feature vector data angle encoding method more suitable for current NISQ equipment. It is also used for classification learning of commodity user evaluation networks without waiting for future large-scale fault-tolerant quantum computers, so that the currently developed quantum computers can be more fully utilized in the real field to solve some problems in this field that classical computers cannot or are difficult to solve.

[0111] The quantum graph convolutional neural network proposed in this application adopts a new computing model based on the basic principles of quantum mechanics, namely quantum computing. Since the quantum graph convolutional neural network model has strong parallelism and non-local characteristics, it is better than the classical neural network in extracting the global features of the commodity user evaluation network information and the convergence speed, and the quantum graph convolutional neural network parameters used are far less than the classical neural network.

[0112] The embodiment of the present invention provides a data evaluation device, such as Figure 7 As shown, the device includes: a first acquisition module 70, used to obtain a user evaluation data set to be processed; a processing module 71, used to process the user evaluation data set to be processed, and obtain a quantum state to be processed corresponding to the user evaluation data set to be processed; a first output module 72, used to input the quantum state to be processed into a pre-trained quantum graph convolutional neural network model, and output an evaluation result of the user evaluation data set to be processed.

[0113] The above-mentioned data evaluation device can process the user evaluation data set to be processed into the corresponding quantum state to be processed, and output the corresponding evaluation result through the quantum graph convolutional neural network model. Compared with the classical machine learning method, the method using the quantum graph convolutional neural network model has stronger processing capabilities.

[0114] Furthermore, the processing module is also used to: determine the user evaluation network graph and the first multi-dimensional node feature vector based on the user evaluation data set to be processed; construct an initial adjacency matrix and a degree matrix based on the user evaluation network graph; standardize the initial adjacency matrix based on the degree matrix to obtain a standardized adjacency matrix; determine the quantum state to be processed corresponding to the user evaluation data set to be processed based on the standardized adjacency matrix.

[0115] Furthermore, the processing module is also used to: process the first multidimensional node feature vector based on the standardized adjacency matrix to obtain a second multidimensional node feature vector; perform dimensionality reduction processing on the second multidimensional node feature vector to obtain a third multidimensional node feature vector; perform encoding processing on the third multidimensional node feature vector to obtain a quantum state to be processed corresponding to the user evaluation data set to be processed.

[0116] Furthermore, the processing module is also used to: for each node feature vector in the third multi-dimensional node feature vector, take the arcsine of the node feature vector to obtain the quantum gate rotation angle corresponding to the node feature vector; encode the multiple quantum gate rotation angles corresponding to the third multi-dimensional node feature vector through a preset rotating quantum gate to obtain the quantum state to be processed corresponding to the user evaluation data set to be processed.

[0117] Furthermore, the quantum graph convolutional neural network model includes multiple layers of graph convolution modules, measurement modules and fully connected layers connected in sequence; the first layer of graph convolution module is used to perform convolution processing on the received standardized adjacency matrix, the first multi-dimensional node feature vector and the quantum bits corresponding to the quantum state to be processed, and output the first aggregate node feature vector; for each layer of graph convolution module except the first layer of graph convolution module, the graph convolution module of this layer is used to perform convolution processing on the received standardized adjacency matrix, the aggregate node feature vector output by the previous layer of graph convolution module and the quantum bits corresponding to the quantum state to be processed, and output the second aggregate node feature vector; the measurement module is used to measure the expected value of each quantum bit based on the second aggregate node feature vector output by the last layer of graph convolution module; the fully connected layer is used to perform fully connected processing on the expected value of each quantum bit, and output the evaluation result of the user evaluation data set to be processed.

[0118] Furthermore, each layer of graph convolution module includes multiple unitary modules; each unitary module is composed of preset rotation quantum gates.

[0119] The data evaluation device provided in the embodiment of the present invention has the same implementation principle and technical effects as those of the aforementioned data evaluation method embodiment. For the sake of brief description, for matters not mentioned in the data evaluation device embodiment, reference may be made to the corresponding contents in the aforementioned data evaluation method embodiment.

[0120] The embodiment of the present invention provides a training device for a quantum graph convolutional neural network model, such as Figure 8 As shown, the device comprises:

[0121] The second acquisition module 80 is used to obtain a preset data set; wherein the preset data set includes: a training sample set and a test sample set; each user evaluation data training sample in the training sample set carries a first evaluation label; the input module 81 is used to input the preset data set into the initial quantum graph convolutional neural network model, and output the first training result corresponding to the preset data set; the selection module 82 is used to select the second training result corresponding to the training sample set from the first training result; the calculation module 83 is used to calculate the loss value of the prediction result corresponding to the second training result based on the preset loss function and the first evaluation label, and train the initial quantum graph convolutional neural network model based on the loss value to obtain a trained quantum graph convolutional neural network model.

[0122] The training device of the above-mentioned quantum graph convolutional neural network model inputs a preset data set into the initial quantum graph convolutional neural network model, selects a second training result corresponding to the training sample set from the first training result, and then trains the initial quantum graph convolutional neural network model based on the loss function and the first evaluation label, so that the trained quantum graph convolutional neural network model has stronger processing capabilities.

[0123] Furthermore, each user evaluation data test sample in the test sample set carries a second evaluation label, and the device is also used to: input a preset data set into the quantum graph convolutional neural network model, and output a first test result corresponding to the preset data set; select a second test result corresponding to the test sample set from the first test result; for each user evaluation data test sample in the test sample set, compare the second evaluation label corresponding to the user evaluation data test sample with the corresponding second test result to obtain a comparison result; and determine the accuracy of the quantum graph convolutional neural network model based on the comparison result corresponding to each user evaluation data test sample.

[0124] The training device for the quantum graph convolutional neural network model provided in the embodiment of the present invention has the same implementation principle and technical effects as those in the aforementioned training method embodiment of the quantum graph convolutional neural network model. For the sake of brief description, for matters not mentioned in the embodiment of the training device for the quantum graph convolutional neural network model, reference may be made to the corresponding contents in the aforementioned training method embodiment of the quantum graph convolutional neural network model.

[0125] The embodiment of the present invention further provides an electronic device, see Fig. 9 As shown, the electronic device includes a processor 130 and a memory 131, wherein the memory 131 stores machine executable instructions that can be executed by the processor 130, and the processor 130 executes the machine executable instructions to implement the above-mentioned data evaluation method and the training method of the quantum graph convolutional neural network model.

[0126] Further, Fig. 9 The electronic device shown further includes a bus 132 and a communication interface 133 , and the processor 130 , the communication interface 133 and the memory 131 are connected via the bus 132 .

[0127] The memory 131 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk storage. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 133 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used. The bus 132 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig. 9 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0128] The processor 130 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 130. The above processor 130 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in conjunction with the embodiments of the present invention can be directly embodied as a hardware decoding processor for execution, or a combination of hardware and software modules in the decoding processor for execution. The software module may be located in a storage medium mature in the art, such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 131, and the processor 130 reads the information in the memory 131 and completes the steps of the method of the above embodiment in combination with its hardware.

[0129] An embodiment of the present invention also provides a machine-readable storage medium, which stores machine-executable instructions. When the machine-executable instructions are called and executed by a processor, the machine-executable instructions prompt the processor to implement the above-mentioned data evaluation method and the training method of the quantum graph convolutional neural network model. The specific implementation can be found in the method embodiment, which will not be repeated here.

[0130] The computer program product of the data evaluation method and the training method of the quantum graph convolutional neural network model provided in the embodiments of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the method in the previous method embodiment. The specific implementation can be found in the method embodiment, which will not be repeated here.

[0131] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A data evaluation method, characterized in that: The method comprises: Obtain the user evaluation data set to be processed; Processing the to-be-processed user evaluation data set to obtain a to-be-processed quantum state corresponding to the to-be-processed user evaluation data set; Inputting the quantum state to be processed into a pre-trained quantum graph convolutional neural network model, and outputting the evaluation result of the user evaluation data set to be processed; The step of processing the user evaluation data set to be processed to obtain the quantum state to be processed corresponding to the user evaluation data set to be processed includes: Based on the user evaluation data set to be processed, determining a user evaluation network graph and a first multi-dimensional node feature vector; Based on the user evaluation network graph, construct an initial adjacency matrix and a degree matrix; Standardizing the initial adjacency matrix based on the degree matrix to obtain a standardized adjacency matrix; Based on the standardized adjacency matrix, determining the quantum state to be processed corresponding to the user evaluation data set to be processed; The quantum graph convolutional neural network model includes a multi-layer graph convolution module, a measurement module, and a fully connected layer connected in sequence; The first-layer graph convolution module is used to perform convolution processing on the received standardized adjacency matrix, the first multi-dimensional node feature vector, and the quantum bits corresponding to the quantum state to be processed, and output a first aggregated node feature vector; For each layer of graph convolution modules other than the first layer of graph convolution modules, the graph convolution module of this layer is used to perform convolution processing on the received standardized adjacency matrix, the aggregation node feature vector output by the previous layer of graph convolution modules, and the quantum bits corresponding to the quantum state to be processed, and output a second aggregation node feature vector; The measurement module is used to measure the expected value of each of the quantum bits based on the second aggregation node feature vector output by the last layer of graph convolution module; The fully connected layer is used to perform full connection processing on the expected value of each of the quantum bits, and output the evaluation result of the user evaluation data set to be processed.

2. The method according to claim 1, characterized in that Based on the standardized adjacency matrix, the step of determining the quantum state to be processed corresponding to the user evaluation data set to be processed includes: Based on the standardized adjacency matrix, the first multi-dimensional node feature vector is processed to obtain a second multi-dimensional node feature vector; Performing dimensionality reduction processing on the second multi-dimensional node feature vector to obtain a third multi-dimensional node feature vector; The third multi-dimensional node feature vector is encoded to obtain a quantum state to be processed corresponding to the user evaluation data set to be processed.

3. The method according to claim 2, characterized in that The step of encoding the third multi-dimensional node feature vector to obtain the quantum state to be processed corresponding to the user evaluation data set to be processed includes: For each node feature vector in the third multi-dimensional node feature vector, taking the arc sine of the node feature vector to obtain a quantum gate rotation angle corresponding to the node feature vector; Through a preset rotation quantum gate, multiple quantum gate rotation angles corresponding to the third multi-dimensional node feature vector are encoded and processed to obtain a quantum state to be processed corresponding to the user evaluation data set to be processed.

4. The method according to claim 1, characterized in that: Each layer of the graph convolution module includes multiple unitary modules; each of the unitary modules is composed of a preset rotation quantum gate.

5. A training method for a quantum graph convolutional neural network model, which is applied to the data evaluation method according to any one of claims 1 to 4, characterized in that: The method comprises: Acquire a preset data set; wherein the preset data set includes: a training sample set and a test sample set; each user evaluation data training sample in the training sample set carries a first evaluation label; Inputting the preset data set into an initial quantum graph convolutional neural network model, and outputting a first training result corresponding to the preset data set; Selecting a second training result corresponding to the training sample set from the first training result; Based on a preset loss function and the first evaluation label, the loss value of the prediction result corresponding to the second training result is calculated, and the initial quantum graph convolutional neural network model is trained based on the loss value to obtain a trained quantum graph convolutional neural network model.

6. The method according to claim 5, characterized in that Each user evaluation data test sample in the test sample set carries a second evaluation label, and the method further includes: Inputting the preset data set into the quantum graph convolutional neural network model, and outputting a first test result corresponding to the preset data set; Selecting a second test result corresponding to the test sample set from the first test result; For each user evaluation data test sample in the test sample set, comparing the second evaluation label corresponding to the user evaluation data test sample with the corresponding second test result to obtain a comparison result; The accuracy of the quantum graph convolutional neural network model is determined according to the comparison results corresponding to each user evaluation data test sample.

7. A data evaluation device, characterized in that: The device comprises: A first acquisition module is used to acquire a user evaluation data set to be processed; A processing module, used to process the user evaluation data set to be processed to obtain a quantum state to be processed corresponding to the user evaluation data set to be processed; A first output module, used to input the quantum state to be processed into a pre-trained quantum graph convolutional neural network model, and output the evaluation result of the user evaluation data set to be processed; The processing module is also used for: Based on the user evaluation data set to be processed, determining a user evaluation network graph and a first multi-dimensional node feature vector; Based on the user evaluation network graph, construct an initial adjacency matrix and a degree matrix; Standardizing the initial adjacency matrix based on the degree matrix to obtain a standardized adjacency matrix; Based on the standardized adjacency matrix, determining the quantum state to be processed corresponding to the user evaluation data set to be processed; The quantum graph convolutional neural network model includes a multi-layer graph convolution module, a measurement module, and a fully connected layer connected in sequence; The first-layer graph convolution module is used to perform convolution processing on the received standardized adjacency matrix, the first multi-dimensional node feature vector, and the quantum bits corresponding to the quantum state to be processed, and output a first aggregated node feature vector; For each layer of graph convolution modules other than the first layer of graph convolution modules, the graph convolution module of this layer is used to perform convolution processing on the received standardized adjacency matrix, the aggregation node feature vector output by the previous layer of graph convolution modules, and the quantum bits corresponding to the quantum state to be processed, and output a second aggregation node feature vector; The measurement module is used to measure the expected value of each of the quantum bits based on the second aggregation node feature vector output by the last layer of graph convolution module; The fully connected layer is used to perform full connection processing on the expected value of each of the quantum bits, and output the evaluation result of the user evaluation data set to be processed.

8. A training device for a quantum graph convolutional neural network model, which is arranged in the data evaluation device as claimed in claim 7; characterized in that: The device comprises: The second acquisition module is used to acquire a preset data set; wherein the preset data set includes: a training sample set and a test sample set; each user evaluation data training sample in the training sample set carries a first evaluation label; An input module, used to input the preset data set into the initial quantum graph convolutional neural network model, and output a first training result corresponding to the preset data set; A selection module, configured to select a second training result corresponding to the training sample set from the first training result; A calculation module is used to calculate the loss value of the prediction result corresponding to the second training result based on a preset loss function and the first evaluation label, and train the initial quantum graph convolutional neural network model based on the loss value to obtain a trained quantum graph convolutional neural network model.

9. An electronic device, characterized in that: It includes a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor executes the machine executable instructions to implement the data evaluation method described in any one of claims 1 to 4, or the training method of the quantum graph convolutional neural network model described in any one of claims 5 to 6.

10. A machine-readable storage medium, characterized in that: The machine-readable storage medium stores machine-executable instructions. When the machine-executable instructions are called and executed by the processor, the machine-executable instructions prompt the processor to implement the data evaluation method described in any one of claims 1 to 4, or the training method of the quantum graph convolutional neural network model described in any one of claims 5 to 6.

Images