Information processing method, device and equipment

Abstract

The present invention discloses a method, device and equipment for processing information, the method comprising: obtaining a first type of message; generating an element topology diagram of the first type of message; obtaining a first input matrix and a second input matrix based on the element topology diagram of the first type of message; inputting the first input matrix and the second input matrix into a trained first network model for processing to obtain a first output; the first network model is a graph encoder; inputting the first output into a second network model for processing to obtain a second type of message; the second network model is a sequence encoder. Through the above-mentioned method, the present invention solves the problems of disorder and incoherence that occur easily after the first type of rich media type message is converted into the second type of message, and achieves the beneficial effect of increasing the user experience of the called party.

Description

Technical Field

[0001] The present invention relates to the field of information processing technology, and in particular to an information processing method, device and equipment. Background Art

[0002] 5G messaging, based on the native SMS interface of mobile devices, allows users to send and receive text, images, audio, video, location, contacts, and other media content. This includes point-to-point messaging, group messaging, group chat messaging, and messaging between peers and applications. Compared to the single-function traditional SMS, 5G messaging not only broadens the scope of information sending and receiving, supporting users with multimedia content such as text, audio and video, cards, and location, but also extends the depth of the interactive experience. Users can complete service search, discovery, interaction, payment, and other services within the message window, creating a one-stop service information window.

[0003] In the existing technology, when converting 5G messages to ordinary text messages, all the text contents in the 5G messages are simply merged. For rich media type 5G messages, the coherence after conversion to ordinary text messages is generally poor, the reading experience is not friendly enough, and problems such as disorder and incoherence are prone to occur. Summary of the Invention

[0004] In view of the above problems, embodiments of the present invention are proposed to provide an information processing method, apparatus, and device that overcome the above problems or at least partially solve the above problems.

[0005] According to one aspect of an embodiment of the present invention, a method for processing information is provided, including:

[0006] Get the first type of message;

[0007] generating an element topology diagram of the first type of message;

[0008] Obtaining a first input matrix and a second input matrix according to the element topology graph of the first type of message;

[0009] Inputting the first input matrix and the second input matrix into a trained first network model for processing to obtain a first output; the first network model is a graph encoder;

[0010] The first output is input into a second network model for processing to obtain a second type of message; the second network model is a sequence encoder.

[0011] According to another aspect of an embodiment of the present invention, there is provided an information processing device, including:

[0012] An acquisition module, configured to acquire a first type of message;

[0013] A processing module is used to generate an element topology diagram of the first type of message; obtain a first input matrix and a second input matrix based on the element topology diagram of the first type of message; input the first input matrix and the second input matrix into a trained first network model for processing to obtain a first output; the first network model is a graph encoder; the first output is input into a second network model for processing to obtain a second type of message; the second network model is a sequence encoder.

[0014] According to another aspect of an embodiment of the present invention, there is provided a computing device, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus;

[0015] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute an operation corresponding to the above-mentioned information processing method.

[0016] According to another aspect of the embodiments of the present invention, a computer storage medium is provided, wherein the storage medium stores at least one executable instruction, and the executable instruction enables a processor to execute an operation corresponding to the above-mentioned information processing method.

[0017] According to the solution provided by the above embodiment of the present invention, by obtaining a first type of message; generating an element topology graph of the first type of message; obtaining a first input matrix and a second input matrix based on the element topology graph of the first type of message; inputting the first input matrix and the second input matrix into a trained first network model for processing to obtain a first output; the first network model is a graph encoder; the first output is input into a second network model for processing to obtain and output a second type of message; the second network model is a sequence encoder. This solves the problem of disorder and incoherence that occurs when rich media type 5G messages are converted into ordinary text messages, and achieves the beneficial effect of improving the user experience of the called party.

[0018] The above description is only an overview of the technical solutions of the embodiments of the present invention. In order to more clearly understand the technical means of the embodiments of the present invention, they can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the embodiments of the present invention more obvious and easy to understand, the specific implementation methods of the embodiments of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the embodiments of the present invention. The same reference numerals are used throughout the accompanying drawings to denote the same components. In the accompanying drawings:

[0020] Figure 1 A flow chart showing a method for processing information provided by an embodiment of the present invention is shown;

[0021] Figure 2 A model diagram consisting of a graph encoder and a sequence decoder provided by an embodiment of the present invention is shown;

[0022] Figure 3 A schematic diagram of an LSTM neuron provided by an embodiment of the present invention is shown;

[0023] Figure 4 The following is a flowchart of the online conversion of 5G messages to text messages provided by an embodiment of the present invention;

[0024] Figure 5 A schematic diagram showing the structure of an information processing device provided by an embodiment of the present invention;

[0025] Figure 6 A schematic structural diagram of a computing device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0026] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0027] Figure 1 FIG. 1 is a flow chart showing a method for processing information provided by an embodiment of the present invention. Figure 1 As shown, the method includes the following steps:

[0028] Step 11, obtaining a first type of message;

[0029] Step 12: Generate an element topology diagram of the first type of message;

[0030] Step 13: obtaining a first input matrix and a second input matrix according to the element topology diagram of the first type of message;

[0031] Step 14: input the first input matrix and the second input matrix into a trained first network model for processing to obtain a first output; the first network model is a graph encoder;

[0032] Step 15: Input the first output into a second network model for processing to obtain a second type of message, and output it; the second network model is a sequence encoder.

[0033] The information processing method described in this embodiment is as follows: obtaining a first type of message; generating an element topology diagram of the first type of message; obtaining a first input matrix and a second input matrix based on the element topology diagram of the first type of message; inputting the first input matrix and the second input matrix into a trained first network model for processing to obtain a first output; the first network model is a graph encoder; the first output is input into a second network model for processing to obtain a second type of message, and outputs it; the second network model is a sequence encoder. This solves the problem that the first type of rich media type messages (such as 5G messages) are prone to disorder and incoherence after being converted into the second type of messages (such as text short messages), and achieves the beneficial effect of increasing the user experience of the called party.

[0034] In an optional embodiment of the present invention, step 12 may include:

[0035] Step 121: Decompose the first type of message to obtain at least one element;

[0036] Specifically, the elements include cards, pictures, videos, locations, etc. but are not limited to the above.

[0037] Step 122: Generate an element topology graph of the first type of message based on the logical relationship between the elements, expressed as G = (V, E), where V is the set of nodes corresponding to the elements V = {V1, V2, V3, ..., V N}, E is the set of edges of the element topology graph, and the edges represent nodes V i and node V j The logical relationship between them.

[0038] Specifically, the elements after decomposition of the first type of message are taken as nodes, the text content and element type of each element are taken as node features, and the logical relationship between elements is taken as the connection relationship of the nodes to form the element topology graph of the first type of message: G = (V, E). Where V is the set of nodes corresponding to the elements V = {V1, V2, V3, ..., V N}, E is the set of edges of the element topology graph, and the edges represent nodes V i and node V j The logical relationship between them.

[0039] In another optional embodiment of the present invention, step 13 may include:

[0040] Step 131: Encode the edges in the element topology graph of the first type of message to obtain a first input matrix A, where A is an N*N matrix, where N is the number of nodes in the element topology graph, and the first input matrix A is an adjacency matrix;

[0041] Specifically, the first input matrix A is the logical relationship between the first type of messages, e ij Represents element node V i and element node V j The logical relationship between them includes parallel, primary and secondary, general and specific, etc., but is not limited to the element relationship described above.

[0042] Step 132: Obtain a second input matrix X based on the characteristic sequence of each element in the element topology diagram of the first type of message. The second input matrix X is an N*F matrix, where F is the encoding length of the characteristic sequence of the element, and the second input matrix X is a characteristic matrix.

[0043] Specifically, according to the element topology diagram of the first type of message, the encoding sequence length of each element feature is defined as F, the longest length F in the first type of message set is taken as its encoding sequence length, and the length of each data is filled with F to obtain a second input matrix X with a shape of N*F, and its dictionary size is taken as element_vocab_size (element name size).

[0044] The following combination Figure 2 Explain the specific implementation process of steps 14 and 15 above:

[0045] In another optional embodiment of the present invention, step 14 may include:

[0046] Step 141: Input the first input matrix and the second input matrix into the trained graph encoder for processing to obtain a first output Z = GCN(X, A); wherein GCN is a graph convolutional neural network, A is the first input matrix, and X is the second input matrix.

[0047] In this embodiment, the graph encoder is composed of a GCN (graph convolutional neural network), which extracts the logical relationship between the elements of the first type of message and the text content features of each element node, projects the nodes and edges of the element topology into a low-dimensional vector space, and obtains a first output Z, Z = GCN (X, A), A is the first input matrix, X is the second input matrix, and Z is the first output, that is, the potential space vector output by the graph encoder.

[0048] In another optional embodiment of the present invention, step 141 may include:

[0049] Step 1411: Input the first input matrix and the second input matrix into the first layer of the trained graph convolutional neural network (GCN) for processing to obtain the first layer output of the GCN.

[0050] Specifically, the first layer of the graph convolutional neural network GCN is the input layer, and the first input matrix and the second input matrix are input into the trained input layer for processing to obtain the first layer output of the GCN.

[0051] Step 1412: The output of the first layer of the GCN is input into the second layer of the GCN for processing to obtain the second layer output. The second layer of the GCN is a graph convolution layer.

[0052] Step 1413: The second layer output of the GCN is input into the third layer of the GCN for processing to obtain the third layer output. The third layer output of the GCN is a graph convolution layer.

[0053] In step 1414, the third layer output of the GCN is input into the fourth layer of the GCN for processing to obtain the fourth layer output, where the fourth layer of the GCN is a graph convolution layer; the fourth layer output of the GCN is the latent space vector Z = GCN (X, A) of the nodes and edges of the element topology graph of the first type of message.

[0054] In this embodiment, the number of convolution kernels of the second layer of the GCN is greater than the number of convolution kernels of the third layer of the GCN, and the number of convolution kernels of the third layer of the GCN is greater than the number of convolution kernels of the fourth layer of the GCN.

[0055] Specifically, the second layer of the graph convolutional neural network GCN is a graph convolution layer (GCN), the number of convolution kernels is 256, the activation function is set to "relu", and the first layer output of the GCN is used to extract the topological features of the first type of message elements to obtain the second layer output.

[0056] The third layer of the graph convolutional neural network GCN is a graph convolution layer (GCN), the number of convolution kernels is 128, and the activation function is set to "relu". The topological features of the first type of message elements are extracted from the second layer output of the GCN to obtain the third layer output;

[0057] The fourth layer of the graph convolutional neural network (GCN) is a graph convolutional layer (GCN), with 64 convolution kernels and a "lamda" activation function. The third layer output of the GCN is used to extract the topological features of the first type of message elements to obtain the fourth layer output. The fourth layer output of the GCN is the latent space vector Z = GCN(X, A) of the nodes and edges of the element topology graph of the first type of message.

[0058] In another optional embodiment of the present invention, step 15 may include:

[0059] Step 151: input the first output into a sequence decoder for processing, and output a target type message;

[0060] Specifically, the sequence decoder is composed of an LSTM layer (long short-term memory neural layer), which extracts topological features from the latent space vector representation of the element topological nodes and edges of the first type of message to generate a target-normalized target type message y, y = LSTM (Z), where y is the target type message, LSTM is the long short-term memory layer, and Z is the first output. The LSTM (long short-term memory neural network) is a special type of recurrent neural network that can remember long-term information by controlling the storage time of values ​​in the cache, making it suitable for time series prediction. Each neuron has four inputs and one output, and each neuron has a cell that stores the remembered value.

[0061] Step 152: Compare the target type message with a message corresponding to a preset label matrix to obtain a second type message, where the preset label matrix is ​​a correct message text converted from the pre-labeled first type message to the second type message.

[0062] Specifically, the preset label matrix Y is the correct message text of the first type of message converted from the manually pre-labeled first type of message to the regularized second type of message, and its shape is N*K. The longest length K of the regularized second type of message set is taken as its index sequence length, and its dictionary size is taken as sortedText_vocab_size (classified text name size). Each regularized second type of message can be represented as {y1, y2, y3, ..., y K}.

[0063] In yet another optional embodiment of the present invention, step 151 may include:

[0064] Step 1511: input the first output into the first layer of the sequence decoder for processing to obtain the first layer output of the sequence decoder, where the first layer of the sequence decoder is a long short-term memory layer (LSTM).

[0065] Specifically, the number of neurons in the first layer of the sequence decoder is set to 64, and the activation function is set to "relu"; the first output Z is input to this layer for processing to obtain the first layer output of the sequence decoder.

[0066] Step 1512: The first layer output of the sequence decoder is input into the second layer of the sequence decoder for processing to obtain the second layer output of the pre-column decoder, where the second layer of the sequence decoder is a long short-term memory layer (LSTM).

[0067] Specifically, the number of neurons in the second layer of the sequence decoder is set to 128, and the activation function is set to "relu"; the output of the first layer is input to this layer for processing to obtain the output of the second layer of the sequence decoder.

[0068] Step 1513: Input the second layer output of the sequence decoder to the third layer of the sequence decoder for processing to obtain the third layer output of the sequence decoder. The third layer of the sequence decoder is a long short-term memory layer (LSTM).

[0069] Specifically, the number of neurons in the third layer of the sequence decoder is set to 256, and the activation function is set to "relu"; the output of the second layer is input to this layer for processing to obtain the output of the third layer of the sequence decoder.

[0070] The number of neurons in the first layer of the sequence decoder is smaller than the number of neurons in the second layer of the sequence decoder, and the number of neurons in the second layer of the sequence decoder is smaller than the number of neurons in the third layer of the sequence decoder;

[0071] Step 1514: The third layer output of the sequence decoder is input into the fourth layer of the sequence decoder for processing to obtain the fourth layer output of the sequence decoder. The fourth layer of the sequence decoder is a fully connected layer, and the fourth layer output of the sequence decoder is the target type message.

[0072] Specifically, the fourth layer of the sequence decoder is a fully connected (Dense) layer, containing sortedText_vocab_size (the size of the classified text name) Dense fully connected neurons, with the activation function set to "softmax". The softmax output result is fed into the multi-class cross entropy loss function. The shape of the output data of this layer is None, sortedText_vocab_size (none, the size of the classified text name).

[0073] In this embodiment, each LSTM neuron contains three gates, including a forget gate, an input gate, and an output gate, but is not limited to the above.

[0074] like Figure 3 As shown, first pass Forget useless information, then pass Add new information and then pass Integrate new information with old information and then Output the information about the next timestamp that the current LSTM unit has learned, and finally pass Y t =σ(W′h t ) Output the neuron.

[0075] Long short-term memory neural network has a good effect on learning long time series. Each connection line in the LSTM unit contains a corresponding weight. t represents the input vector, ht represents the hidden state, C t represents the neuron state at time t, Y t represents the output of the neuron, W is the trainable weight matrix, and b is the bias vector.

[0076] In the above embodiment of the present invention, the first network model is trained through the following process:

[0077] Data collection and preprocessing:

[0078] We collected historical caller-originated 5G messages (i.e., the first type of messages mentioned above) and used each element in the 5G message as a node. The text content and element type of each element served as node features, and the logical relationships between elements served as node connections. This formed an element topology graph for the 5G message. We converted each topology graph into the model inputs: the adjacency matrix A and the feature matrix X.

[0079] The element topology of 5G messages can be expressed as G = (V, E), where V is the set of element nodes V = {V1, V2, V3, ..., V N}, E is the set of edges.

[0080] The adjacency matrix A is the logical relationship between the elements of the 5G message, e ij Represents element node V i and element node V j The logical relationships between them can include parallel, primary and secondary, general and specific relationships, and each logical relationship is encoded. The shape is N*N (N is the number of nodes).

[0081] The feature matrix X represents the sequence of text content and element type within each element. Define the encoding sequence length of each element feature as F. Take the longest length F in the 5G message element set as its encoding sequence length. Pad the length of each piece of data to F. Therefore, the feature matrix X can be represented as an N*F feature matrix, with its vocabulary size as element_vocab_size.

[0082] The total dataset is divided into a training set and a test set, with 90% of the total dataset being the training set and 10% being the test set. The training set is used to train the model, and the test set is used to test the model.

[0083] Building the model:

[0084] Build a neural network from graph to sequence structure, which consists of a graph encoder and a sequence decoder. Figure 2 The encoder shown uses a graph convolutional layer to extract the logical relationships between the elements of the 5G message and the text content features of each element node, and projects the nodes and edges of the element topology into a low-dimensional vector space;

[0085] The LSTM layer of the sequence decoder is then used to learn the relationship between the topological vector representation of the 5G message elements after the spatial features are extracted and the correct regularized text message, and finally output the regularized text message sequence.

[0086] Graph Encoder: This consists of a graph convolutional neural network that maps the topology of the input caller-originated 5G message elements into a latent space vector representation of topological nodes and edges, Z = GCN(X, A).

[0087] The first layer is the input layer: it inputs the adjacency matrix A and feature matrix X of the 5G message element topology;

[0088] The second layer is the graph convolution layer (Graph Conv): the number of convolution kernels is 256, and the activation function is set to "relu". The convolution layer is used to extract the topological features of 5G message elements;

[0089] The third layer is the graph convolution layer (Graph Conv): the number of convolution kernels is 128, and the activation function is set to "relu";

[0090] The fourth layer is the graph convolution layer (Graph Conv): the number of convolution kernels is 64, and the activation function is set to "lambda". It outputs the latent space vector representation Z of the topological nodes and edges of the 5G message element initiated by the caller;

[0091] The graph convolution layer is a type of neural network layer. Each neural network layer can be written as a nonlinear function: H () =f(H () ,A), H(0)=X is the input data, H(L)=Z is the output data, L is the number of layers of the neural network, and different f() and parameters also determine different models, for example: W (l) is the parameter matrix of the lth neural network layer, σ() is the nonlinear activation function (such as ReLU), A is the first input matrix, and D is the node diagonal matrix of A.

[0092] Sequence decoder: It is composed of a long short-term memory neural network, responsible for extracting topological features from the latent space vector representation of the topological nodes and edges of the 5G message elements initiated by the caller, and generating the target regularized text SMS sequence, y = LSTM (Z).

[0093] The fifth layer is the long short-term memory (LSTM) layer: the number of neurons is set to 64, the activation function is set to "relu", and the latent space vector of the input topology is represented by Z;

[0094] The sixth layer is the long short-term memory layer (LSTM): the number of neurons is set to 128, and the activation function is set to "relu";

[0095] The seventh layer is the long short-term memory (LSTM) layer: the number of fully connected neurons is 256, and the activation function is set to "relu";

[0096] The sixth fully connected (Dense) layer (output layer) contains sortedText_vocab_size fully connected neurons, with the activation function set to "softmax". The softmax output is fed into a multi-class cross entropy loss function. The output data of this layer has a shape of (None, sortedText_vocab_size).

[0097] After obtaining the second type of message, calculate the error between the predicted target type message and the second type of message, and train the target to minimize the error. The objective function selects the 'categorical_crossentropy' multi-class logarithmic loss function: Where n is the number of samples, m is the number of categories, yi is the target type message, logy i This is the second type of message. The number of training rounds is set to 1000 (epochs = 1000), and the Adam optimizer is selected for the gradient descent optimization algorithm to improve the learning speed of traditional gradient descent (optimizer = 'adam'). Neural networks use gradient descent to find the optimal weight values ​​that minimize the objective function. The neural network autonomously learns these weights through training. The training set is used to minimize the objective function, and the test set is used to evaluate and verify the model after each round of training. Weights are derived after the model converges.

[0098] like Figure 4 As shown, in an embodiment of the present invention, a specific implementation process of a method for processing information is provided. The first type of message is described by taking a 5G message as an example, and the second type of message is described by taking an ordinary text short message as an example, including the following:

[0099] Step 41: The 5G messaging terminal or industry chatbot initiates a 5G messaging service process to the SIP access module of the calling 5GMC through an invite signaling.

[0100] Step 42: The SIP access module sends the 5G message to the message processing module of the calling 5GMC through the invite signaling;

[0101] In step 43, the 5GMC message processing module extracts the XML text in the body-text field of the signaling message and sends it to the 5G message preprocessing module, which constructs a 5G message element topology map based on the XML text. Specifically, each element in the 5G message is used as a node, the text content in each element is used as the node feature, and the logical relationship between the elements is used as the node connection relationship.

[0102] Step 44: Send the adjacency matrix A and feature matrix X representing the 5G message element topology graph to a 5G message-to-text SMS generator composed of a pre-trained graph-to-sequence neural network; where the adjacency matrix A represents the logical relationship between the elements of the 5G message, and the feature matrix X represents the text content and element type of each element;

[0103] Step 45: The 5G message to text SMS generator uses a graph-to-sequence network to learn the logical relationships between the elements in the 5G message, thereby reorganizing the text content in each element and ultimately converting the 5G message in a rich media format into a regular text SMS.

[0104] Step 46: Send the ordinary text message to the SMS processing module of the calling 5GMC;

[0105] Step 47: Finally, the calling 5GMC finds the called MSC through the HLR / HSS and sends the ordinary text message to the called user terminal.

[0106] The above embodiment of the present invention collects historical caller-originated 5G messages, treating each element in each 5G message as a node, the text content and element type in each element as node features, and the logical relationships between elements as node connections, thereby forming an element topology graph of the 5G message. Each topology graph is converted into the model input, namely, an adjacency matrix A and a feature matrix X, where the adjacency matrix A represents the logical relationships between the elements of the 5G message, the feature matrix X represents the text content and element type in each element, and the label matrix Y represents the manually labeled conversion of each caller-originated 5G message into a regularized text message. A graph-to-sequence network model consisting of a graph encoder and a sequence decoder is constructed, where the graph encoder uses a graph convolutional layer to extract the logical relationships between the elements of the caller-originated 5G message and the text content features of each element node, and projects the nodes and edges of the element topology into a low-dimensional vector space. The long short-term memory layer of the sequence decoder then learns the relationship between the 5G message element topology vector representation after the extracted spatial features and the correct regularized text message, ultimately outputting a regularized text message sequence. This will improve the user experience of the called party when converting 5G messages to ordinary text messages.

[0107] Figure 5 FIG. 5 is a schematic diagram showing a structure 50 of an information processing device provided by an embodiment of the present invention. Figure 5 As shown, the device includes:

[0108] An acquisition module 51 is configured to acquire a first type of message;

[0109] The processing module 52 is used to generate an element topology diagram of the first type of message; obtain a first input matrix and a second input matrix based on the element topology diagram of the first type of message; input the first input matrix and the second input matrix into the trained first network model for processing to obtain a first output; the first network model is a graph encoder; the first output is input into the second network model for processing to obtain a second type of message, and output it; the second network model is a sequence encoder.

[0110] Optionally, the processing module 52 is further configured to decompose the first type message to obtain at least one element;

[0111] According to the logical relationship between the elements, the element topology graph of the first type of message is generated, which is expressed as G = (V, E), where V is the set of nodes corresponding to the elements V = {V1, V2, V3, ..., VN}, and E is the set of edges of the element topology graph, where the edges represent the nodes V i and node V j The logical relationship between them.

[0112] Optionally, the processing module 52 is further configured to encode edges in the element topology graph of the first type of message to obtain a first input matrix A, where A is an N*N matrix, where N is the number of nodes in the element topology graph;

[0113] A second input matrix X is obtained according to the characteristic sequence of each element in the element topology diagram of the first type of message. The second input matrix X is an N*F matrix, where F is the encoding length of the characteristic sequence of the element.

[0114] Optionally, the processing module 52 is further used to input the first input matrix and the second input matrix into a trained graph encoder for processing to obtain a first output Z=GCN(X, A); wherein GCN is a graph convolutional neural network, A is the first input matrix, and X is the second input matrix.

[0115] Optionally, the processing module 52 is further configured to input the first input matrix and the second input matrix into a first layer of a trained graph convolutional neural network GCN for processing to obtain a first layer output of the GCN;

[0116] The output of the first layer of the GCN is input into the second layer of the GCN for processing to obtain the second layer output, where the second layer of the GCN is a graph convolution layer;

[0117] The second layer output of the GCN is input into the third layer of the GCN for processing to obtain the third layer output, and the third layer output of the GCN is the graph convolution layer;

[0118] The third layer output of the GCN is input into the fourth layer of the GCN for processing to obtain the fourth layer output, where the fourth layer of the GCN is a graph convolution layer; the fourth layer output of the GCN is the latent space vector Z=GCN(X, A) of the nodes and edges of the element topology graph of the first type of message, the number of convolution kernels of the second layer of the GCN is greater than the number of convolution kernels of the third layer of the GCN, and the number of convolution kernels of the third layer of the GCN is greater than the number of convolution kernels of the fourth layer of the GCN.

[0119] Optionally, the processing module 52 is further configured to input the first output into a sequence decoder for processing and output a target type message;

[0120] The target type message is compared with a message corresponding to a preset label matrix to obtain a second type message, wherein the preset label matrix is ​​a correct message text converted from a pre-labeled first type message to a second type message.

[0121] Optionally, the processing module 52 is further configured to input the first output into the first layer of the sequence decoder for processing to obtain the first layer output of the sequence decoder, where the first layer of the sequence decoder is a long short-term memory layer LSTM;

[0122] The first layer output of the sequence decoder is input into the second layer of the sequence decoder for processing to obtain the second layer output of the sequence decoder, where the second layer of the sequence decoder is a long short-term memory layer LSTM;

[0123] The second layer output of the sequence decoder is input into the third layer of the sequence decoder for processing to obtain the third layer output of the sequence decoder, wherein the third layer of the sequence decoder is a long short-term memory layer (LSTM); the number of neurons in the first layer of the sequence decoder is smaller than the number of neurons in the second layer of the sequence decoder, and the number of neurons in the second layer of the sequence decoder is smaller than the number of neurons in the third layer of the sequence decoder;

[0124] The third layer output of the sequence decoder is input into the fourth layer of the sequence decoder for processing to obtain the fourth layer output of the sequence decoder. The fourth layer of the sequence decoder is a fully connected layer, and the fourth layer output of the sequence decoder is the target type message.

[0125] It should be noted that this embodiment is an apparatus embodiment corresponding to the above method embodiment, and all implementation methods in the above method embodiment are applicable to the embodiment of this apparatus and can achieve the same technical effects.

[0126] An embodiment of the present invention provides a non-volatile computer storage medium, wherein the computer storage medium stores at least one executable instruction, and the computer executable instruction can execute an information processing method in any of the above method embodiments.

[0127] Figure 6 The schematic diagram of the structure of the computing device provided by the embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the computing device.

[0128] like Figure 6 As shown, the computing device may include: a processor, a communication interface, a memory, and a communication bus.

[0129] The processor, communication interface, and memory communicate with each other via a communication bus. The communication interface is used to communicate with other devices, such as clients or other server network elements. The processor is used to execute programs, specifically, the steps described in the aforementioned embodiment of a method for processing information on a computing device.

[0130] Specifically, the program may include program codes including computer operation instructions.

[0131] The processor may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention. The one or more processors included in the computing device may be processors of the same type, such as one or more CPUs, or processors of different types, such as one or more CPUs and one or more ASICs.

[0132] Memory is used to store programs. The memory may include high-speed RAM memory, and may also include non-volatile memory (non-volatile memory), such as at least one disk storage.

[0133] The program can be specifically used to cause the processor to execute an information processing method in any of the above-mentioned method embodiments. The specific implementation of each step in the program can refer to the corresponding description of the corresponding steps and units in the above-mentioned information processing method embodiment, and will not be repeated here. Those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working process of the above-mentioned devices and modules can refer to the corresponding process description in the above-mentioned method embodiment, and will not be repeated here.

[0134] The algorithm or display provided herein is not inherently related to any particular computer, virtual system or other device. Various general-purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing such systems. In addition, the embodiment of the present invention is not directed to any specific programming language. It should be understood that various programming languages ​​can be utilized to implement the content of the embodiment of the present invention described herein, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the embodiment of the present invention.

[0135] In the description provided herein, numerous specific details are described. However, it is understood that embodiments of the present invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques are not shown in detail so as not to obscure the understanding of this description.

[0136] Similarly, it should be understood that in order to streamline the embodiments of the invention and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the invention, various features of the embodiments of the invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed approach should not be interpreted as reflecting an intention that the claimed embodiments of the invention require more features than are expressly recited in each claim. Rather, as reflected in the claims below, inventive aspects lie in less than all of the features of the individual embodiments disclosed above. Accordingly, the claims that follow the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of the invention.

[0137] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition may be divided into multiple submodules or subunits or subcomponents. All features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed herein may be combined in any combination, except that at least some of such features and / or processes or units are mutually exclusive. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0138] Furthermore, those skilled in the art will appreciate that although some embodiments herein include certain features included in other embodiments but not other features, combinations of features from different embodiments are intended to be within the scope of the present invention and to form different embodiments. For example, in the claims below, any of the claimed embodiments may be used in any combination.

[0139] The various component embodiments of the present invention can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all of the functions of some or all of the components according to the embodiments of the present invention. The embodiments of the present invention can also be implemented as a device or apparatus program (e.g., a computer program and a computer program product) for executing part or all of the methods described herein. Such a program implementing an embodiment of the present invention can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0140] It should be noted that the above embodiments illustrate rather than limit the embodiments of the invention, and that a person skilled in the art may devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between brackets should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The embodiments of the invention may be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third etc. does not indicate any order. These words may be interpreted as names. The steps in the above embodiments should not be understood as limiting the order of execution unless otherwise specified.

Claims

1. A method for processing information, characterized in that: include: Obtain a first type of message; the first type of message is a 5G message; generating an element topology diagram of the first type of message; Obtaining a first input matrix and a second input matrix according to the element topology graph of the first type of message; Inputting the first input matrix and the second input matrix into a trained first network model for processing to obtain a first output; the first network model is a graph encoder; Input the first output into a second network model for processing to obtain a second type of message, and output the second type of message; the second network model is a sequence decoder, and the second type of message is a normal text short message; Inputting the first output into the second network model for processing to obtain the second type message further includes: inputting the first output into a sequence decoder for processing to output a target type message; comparing the target type message with a message corresponding to a preset label matrix to obtain the second type message, wherein the preset label matrix is ​​a correct message text converted from the pre-labeled first type message to the second type message; The step of inputting the first output into a sequence decoder for processing and outputting a target type message further includes: inputting the first output into the first layer of the sequence decoder for processing to obtain the first layer output of the sequence decoder, wherein the first layer of the sequence decoder is a long short-term memory layer LSTM; inputting the first layer output of the sequence decoder into the second layer of the sequence decoder for processing to obtain the second layer output of the sequence decoder, wherein the second layer of the sequence decoder is a long short-term memory layer LSTM; inputting the second layer output of the sequence decoder into the third layer of the sequence decoder for processing to obtain the third layer output of the sequence decoder, wherein the third layer of the sequence decoder is a long short-term memory layer LSTM; the number of neurons in the first layer of the sequence decoder is less than the number of neurons in the second layer of the sequence decoder, and the number of neurons in the second layer of the sequence decoder is less than the number of neurons in the third layer of the sequence decoder; inputting the third layer output of the sequence decoder into the fourth layer of the sequence decoder for processing to obtain the fourth layer output of the sequence decoder, wherein the fourth layer of the sequence decoder is a fully connected layer, and the fourth layer output of the sequence decoder is the target type message.

2. The information processing method according to claim 1, characterized in that: Generating an element topology diagram of the first type of message includes: Decomposing the first type of message to obtain at least one element; According to the logical relationship between the elements, the element topology graph of the first type of message is generated, which is expressed as G = (V, E), where V is the set of nodes corresponding to the elements V = {V1, V2, V3, ..., V N }, E is the set of edges of the element topology graph, and the edges represent nodes V i and node V j The logical relationship between them.

3. The information processing method according to claim 1, characterized in that: Obtaining a first input matrix and a second input matrix according to the element topology diagram of the first type of message includes: Encoding the edges in the element topology graph of the first type of message to obtain a first input matrix, where the first input matrix is ​​an N*N matrix, where N is the number of nodes in the element topology graph; A second input matrix is ​​obtained according to the characteristic sequence of each element in the element topology diagram of the first type of message. The second input matrix is ​​an N*F matrix, where F is the encoding length of the characteristic sequence of the element.

4. The information processing method according to claim 1, characterized in that: Inputting the first input matrix and the second input matrix into the trained first network model for processing to obtain a first output includes: The first input matrix and the second input matrix are input into the trained graph encoder for processing to obtain a first output Z=GCN(X,A); wherein GCN is a graph convolutional neural network, A is the first input matrix, and X is the second input matrix.

5. The information processing method according to claim 4, characterized in that: The first input matrix and the second input matrix are input into the trained graph encoder for processing to obtain a first output, including: Input the first input matrix and the second input matrix into the first layer of the trained graph convolutional neural network GCN for processing to obtain the first layer output of the GCN; The output of the first layer of the GCN is input into the second layer of the GCN for processing to obtain the second layer output, where the second layer of the GCN is a graph convolution layer; The second layer output of the GCN is input into the third layer of the GCN for processing to obtain the third layer output, and the third layer output of the GCN is the graph convolution layer; The third layer output of the GCN is input into the fourth layer of the GCN for processing to obtain the fourth layer output, where the fourth layer of the GCN is a graph convolution layer; the fourth layer output of the GCN is the latent space vector Z=GCN(X, A) of the nodes and edges of the element topology graph of the first type of message, the number of convolution kernels of the second layer of the GCN is greater than the number of convolution kernels of the third layer of the GCN, and the number of convolution kernels of the third layer of the GCN is greater than the number of convolution kernels of the fourth layer of the GCN.

6. An information processing device, characterized in that: include: An acquisition module, configured to acquire a first type of message, wherein the first type of message is a 5G message; A processing module, configured to generate an element topology graph of the first type of message; Obtaining a first input matrix and a second input matrix according to the element topology graph of the first type of message; Input the first input matrix and the second input matrix into a trained first network model for processing to obtain a first output; the first network model is a graph encoder; the first output is input into a second network model for processing to obtain a second type of message and output it; the second network model is a sequence decoder, and the second type of message is a normal text short message; The processing module is further configured to: input the first output into a sequence decoder for processing, and output a target type message; Comparing the target type message with a message corresponding to a preset label matrix to obtain a second type message, wherein the preset label matrix is ​​a correct message text converted from the pre-labeled first type message to the second type message; The processing module is further configured to: input the first output into the first layer of the sequence decoder for processing to obtain the first layer output of the sequence decoder, wherein the first layer of the sequence decoder is a long short-term memory layer LSTM; The first layer output of the sequence decoder is input into the second layer of the sequence decoder for processing to obtain the second layer output of the sequence decoder, where the second layer of the sequence decoder is a long short-term memory layer LSTM; The second layer output of the sequence decoder is input into the third layer of the sequence decoder for processing to obtain the third layer output of the sequence decoder, wherein the third layer of the sequence decoder is a long short-term memory layer (LSTM); the number of neurons in the first layer of the sequence decoder is smaller than the number of neurons in the second layer of the sequence decoder, and the number of neurons in the second layer of the sequence decoder is smaller than the number of neurons in the third layer of the sequence decoder; The third layer output of the sequence decoder is input into the fourth layer of the sequence decoder for processing to obtain the fourth layer output of the sequence decoder. The fourth layer of the sequence decoder is a fully connected layer, and the fourth layer output of the sequence decoder is the target type message.

7. A computing device comprising: A processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute an operation corresponding to the information processing method according to any one of claims 1 to 5.

8. A computer storage medium, wherein at least one executable instruction is stored in the storage medium, wherein the executable instruction enables a processor to execute an operation corresponding to the information processing method according to any one of claims 1 to 5.

Images