Data processing method and device, storage medium and electronic equipment

Abstract

The invention discloses a data processing method and device, a storage medium and electronic equipment, and relates to the technical field of data processing.The method comprises the steps that firstly, a to-be-fused original 5G message data set is obtained based on a 5G message network; converting the original 5G message data set into an initial feature fusion data set in a target format; and performing 5G information fusion on the initial feature fusion data set based on a pre-trained convolutional neural network model to obtain target 5G feature fusion data. Compared with the prior art, the method has the advantages that the original 5G message data set is converted into the initial feature fusion data set in the target format, the initial feature fusion data set is subjected to 5G information fusion based on the pre-trained convolutional neural network model, and the capability of automatically extracting and fusing the multi-modal features of the model is utilized; the method effectively meets the fusion requirements of complex and changeable media features and semantic information, improves the adaptability and flexibility of the message fusion process, and enhances the integrity of feature expression.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a data processing method and device, a storage medium and an electronic device. BACKGROUND

[0002] With the rapid development of the fifth generation wireless systems (5G), the traditional text message has been difficult to meet the needs of users for multimedia content interaction and service integration. The 5G message, also known as the Rich Communication Suite (RCS) message, as a new generation of message service, supports text, pictures, audio, video, location, card, menu and other forms of rich media, realizes the multimedia of the message and the interaction of the service, and the user can complete information acquisition, business handling and payment and other operations in the message window.

[0003] At present, in order to realize the fusion processing of different types of messages, the message content from different terminals and platforms is mainly converted and packaged through standardized protocols and message middleware, and the text, image, audio and video data are unified into a predefined message structure to realize the message interconnection across devices and systems.

[0004] However, due to the differences in message coding, media format and supported functions of different manufacturer terminals, operating systems and application platforms, it is difficult to fully process complex and variable media features and semantic information by relying on fixed rules and protocol mapping, resulting in poor adaptability of message fusion and insufficient feature expression. SUMMARY

[0005] Therefore, the present application provides a data processing method and device, a storage medium and an electronic device, which mainly aims to improve the technical problem that the current fixed rule and protocol mapping method is difficult to fully process complex and variable media features and semantic information, and the adaptability of message fusion is poor and the feature expression is insufficient.

[0006] In a first aspect, the present application provides a data processing method, comprising:

[0007] Based on the 5G message network, obtaining an original 5G message data set to be fused;

[0008] Converting the original 5G message data set into an initial feature fusion data set of a target format;

[0009] Based on a pre-trained convolutional neural network model, performing 5G information fusion on the initial feature fusion data set to obtain a target 5G feature fusion data.

[0010] In a second aspect, the present application provides a data processing device, comprising:

[0011] an acquisition module configured to acquire, based on a 5G message network, an original 5G message dataset to be fused;

[0012] a conversion module configured to convert the original 5G message dataset into an initial feature fusion dataset in a target format;

[0013] a processing module configured to perform 5G information fusion on the initial feature fusion dataset based on a pre-trained convolutional neural network model to obtain target 5G feature fusion data.

[0014] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the method of the first aspect.

[0015] In a fourth aspect, the present application provides an electronic device comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, the processor implementing the method of the first aspect when executing the computer program.

[0016] In a fifth aspect, the present application provides a computer program product having a computer program stored thereon, the computer program being executed by a processor to implement the method of the first aspect.

[0017] According to the above technical solution, the data processing method, device, storage medium and electronic device provided by the present application first acquire, based on a 5G message network, an original 5G message dataset to be fused; then convert the original 5G message dataset into an initial feature fusion dataset in a target format; and then perform 5G information fusion on the initial feature fusion dataset based on a pre-trained convolutional neural network model to obtain target 5G feature fusion data. Compared with the prior art, the present application converts the original 5G message dataset into an initial feature fusion dataset in a target format to realize unified representation of heterogeneous messages, and performs 5G information fusion on the initial feature fusion dataset based on a pre-trained convolutional neural network model to utilize the ability of the model to automatically extract and fuse multi-modal features, effectively cope with the fusion requirements of complex and variable media features and semantic information such as text, image, audio and video, and improve the adaptability and flexibility of the message fusion process and enhance the completeness of feature expression.

[0018] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the following detailed embodiments of the present application can be implemented in accordance with the content of the description, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following detailed embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0019] The accompanying drawings, which are incorporated herein and constitute part of the specification, illustrate embodiments consistent with the application and, together with the description, further serve to explain the principles of the application.

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings required by the embodiments or prior art description will be briefly introduced as follows. Obviously, for those skilled in the field, other drawings can also be obtained based on these drawings without creative labor.

[0021] Figure 1 A flowchart of a data processing method provided by an embodiment of the present application is shown;

[0022] Figure 2 A flowchart of another data processing method provided by an embodiment of the present application is shown;

[0023] Figure 3 A flowchart of an example provided by an embodiment of the present application is shown;

[0024] Figure 4 A structural diagram of a data processing device provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0025] The embodiments of the present application will be described in more detail below with reference to the accompanying drawings. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0026] In order to improve the technical problems that the current fixed rule and protocol mapping mode is difficult to fully process complex and variable media features and semantic information, the adaptability of message fusion is poor, and the feature expression is insufficient. The present embodiment provides a data processing method, which can be applied to a 5G message fusion platform, a communication service node or a message processing gateway, etc. server, such as shown in the figure, the method comprises: Figure 1

[0027] Step 101, based on the 5G message network, obtaining the original 5G message data set to be fused.

[0028] For example, the original 5G message data set to be fused is collected from the 5G message network, wherein the original 5G message data can come from different message service network elements, user equipment, servers, etc.

[0029] ​In some examples, the collected original 5G message dataset can be cleaned and formatted to identify and remove invalid, noisy or abnormal data, ensuring the integrity and reliability of the input data. On this basis, according to the input requirements of the subsequent processing module, the cleaned data is normalized, standardized or encoded, so that the feature values of different sources and different dimensions are unified to the same numerical range or distribution interval, improving the consistency and comparability of the data. This preprocessing process helps to enhance the accuracy of feature mapping, providing a high-quality data basis for subsequent format conversion based on feature relationship mapping table and multi-modal feature fusion, while reducing the adverse effects of data bias or magnitude difference on model training and fusion effect.

[0030] Step 102, transforming the original 5G message dataset into an initial feature fusion dataset of the target format.

[0031] For example, the acquired original 5G message dataset can be cleaned and formatted to remove invalid, noisy or abnormal data, and normalized, standardized or encoded converted for multi-modal content such as text, image, audio, video, etc. Based on the pre-constructed original 5G message data feature relationship mapping table, the features of messages of different formats are mapped to unify them into a standard message dataset with standard field structure and semantic consistency, and then an initial feature fusion dataset of the target format containing multi-dimensional semantic information and unified structure is generated.

[0032] Step 103, performing 5G information fusion on the initial feature fusion dataset based on a pre-trained convolutional neural network model to obtain target 5G feature fusion data.

[0033] For example, the convolutional neural network has the ability to automatically learn the feature representation of the input data without relying on manually designed rules or prior feature engineering. Through the mixed embedding layer, the multi-modal 5G message is uniformly represented as a vector, and the network can convert different types of messages such as text, image, audio, video, etc. into a computable numerical form. In the forward propagation process, each parallel convolution path extracts local features of the corresponding modality, and as the network level deepens, the features evolve from low-level pixels or word frequencies to high-level semantic expressions. The fully connected layer further integrates these high-order features, and the feature maps output by different convolution paths are spliced or weighted in the feature fusion layer, so that the network can automatically capture cross-modal correlations and improve the expression ability of the overall features. This structure shows good adaptability when processing 5G messages with diverse types, complex semantics and implicit features, and has efficient, accurate and stable processing performance.

[0034] Compared with the prior art, by applying the technical scheme of the embodiment, first, based on the 5G message network, the original 5G message data set to be fused is obtained; then the original 5G message data set is converted into an initial feature fusion data set of a target format; and then the initial feature fusion data set is subjected to 5G information fusion based on a pre-trained convolutional neural network model to obtain target 5G feature fusion data. By converting the original 5G message data set into an initial feature fusion data set of a target format, unified representation of heterogeneous messages is realized, and the initial feature fusion data set is subjected to 5G information fusion based on a pre-trained convolutional neural network model, the ability of the model to automatically extract and fuse multi-modal features is utilized, the fusion requirements of complex and variable media features and semantic information such as text, image, audio, and video are effectively met, the adaptability and flexibility of the message fusion process are improved, and the completeness of feature expression is enhanced.

[0035] In order to further illustrate the specific implementation process of the method of the embodiment, the embodiment provides a method as shown in Figure 2 The method comprises the following steps:

[0036] Step 201, based on a 5G message network, an original 5G message data set to be fused is obtained.

[0037] For example, the original 5G message data set to be fused can be collected from multiple message service network elements, user terminal devices, and application servers. The original 5G message data set contains various rich media message types such as text, image, audio, video, location information, and interactive cards, and its source covers different manufacturer terminals, operating systems, and communication nodes. The data format is heterogeneous, and the message content carries basic communication metadata such as time stamp, sender identifier, receiver identifier, and transmission protocol parameters.

[0038] Step 202, the original 5G message data set is converted into an original feature message vector.

[0039] In some examples, key features can be extracted from each original 5G message, which can include text content, media type, message length, sending time, etc. These features are converted into numerical form to construct an original feature message vector.

[0040] Step 203, based on a pre-constructed original 5G message data feature relationship mapping table, the original feature message vector of different formats is mapped into a standard message data set of a target format.

[0041] For example, an original 5G message data feature relationship mapping table can be pre-constructed, which defines the mapping relationship between original 5G message features of different formats. The mapping table can help convert original 5G messages of different formats into a standard format data set.

[0042] Exemplarily, for the format of each original 5G message in the original 5G message dataset to be fused, the feature mapping of each original 5G message in the format of each original 5G message is performed based on the pre-constructed original 5G message data feature relationship mapping table, so as to convert the original feature message vector into the standard format 5G message dataset.

[0043] In some examples, for each original 5G message in the original 5G message dataset to be fused, the pre-constructed original 5G message data feature relationship mapping table is called for feature mapping operation according to the specific message format thereof. The mapping table defines the correspondence between message fields of different sources and types and standard features, and can convert heterogeneous message structures into unified semantic expressions. Through the mapping process, each feature in the original feature message vector is reorganized and converted to meet the preset standard format requirements.

[0044] For example, a standard format 5G message dataset can be defined, which can include all possible message types, fields, structures, etc., to ensure that all messages can be uniformly represented. The finally generated standard format 5G message dataset is consistent in structure, field definition and feature dimension, and all messages are normalized to the same representation, thereby eliminating the format fragmentation problem caused by differences in device manufacturers, operating systems or communication protocols, and providing a unified and standardized data input basis for subsequent enhancement matrix construction, multi-source feature fusion and deep model processing.

[0045] Step 204, based on the corresponding enhancement matrix of each 5G enhancement message in the standard message dataset, performing feature fusion on the standard message dataset to obtain an initial feature fusion dataset.

[0046] Exemplarily, various enhancement message types can be identified in the standard format 5G message dataset. These enhancement message types can include rich text, pictures, videos, audio, location information, etc. For each type of enhancement message, a corresponding enhancement matrix is constructed, and the enhancement matrix describes the features of the message in a numerical form, such as the pixel distribution of an image, the semantic vector of text, or the time-frequency features of audio and video, etc.

[0047] For example, for a picture message, the enhancement matrix can include pixel values, color distribution, texture features, etc. of the picture; for a text message, the enhancement matrix can include word frequency, TF-IDF value, text length, etc.

[0048] In some examples, first, the enhancement matrix corresponding to each 5G enhancement message in the standard format 5G message dataset is determined, and each enhancement matrix and the data fusion feature corresponding to each enhancement matrix are fused to obtain an initial 5G feature fusion dataset.

[0049] Optionally, step 204 can specifically include: identifying an enhanced message type in the standard message data set; constructing an enhanced matrix corresponding to the enhanced message type based on the enhanced message type, the enhanced matrix being used to represent the features of the messages of the enhanced message type; determining data fusion features matched with each enhanced matrix according to the original message content and / or external context information of the standard message data set; and fusing the enhanced matrix and the data fusion features corresponding to the enhanced matrix to generate an initial feature fusion data set.

[0050] In some examples, based on the enhanced matrix, the data fusion features matched with each enhanced matrix are determined, which are not only derived from the original message content in the standard message data set, but also extended to external context information such as user behavior data and network state information, to supplement the static features of the messages and enhance the dynamic semantic expression capability thereof; the obtained data fusion features are compatible with the enhanced matrix in terms of dimension and semantics, ensuring that the two can be jointly processed in a unified space; and then the enhanced matrix and the data fusion features corresponding thereto are fused by means of weighted average, feature splicing or linear transformation, so as to deeply fuse the media features of the messages and the context perception information, and finally generate an initial feature fusion data set containing multi-dimensional and multi-level information.

[0051] For example, through the above steps, the enhanced matrices corresponding to the 5G enhanced messages in the standard format 5G message data set can be determined, and these enhanced matrices are fused with the corresponding data fusion features to obtain an initial 5G feature fusion data set. The fused data set is more rich and comprehensive, which is helpful for subsequent data analysis and machine learning application.

[0052] For example, the data fusion features corresponding to each enhanced matrix are extracted by the initial convolutional neural network:

[0053] output s (i, j) = tanh(con s (i, j))

[0054] wherein, output s is the convolutional layer output, (i, j) represents the matrix position coordinates corresponding to the enhanced matrix data, s represents the number of convolutional layers, tanh represents the activation function, and con s represents the convolutional layer operation.

[0055] Optionally, the method of the embodiment can further include: determining the similarity measure between the feature vectors in the initial feature fusion data set based on the feature distribution of the enhanced matrix; dividing the feature vectors in the initial feature fusion data set into a plurality of message populations based on the similarity measure; and performing differentiated information fusion processing on each message population.

[0056] For example, clustering analysis can be performed on all feature vectors by the metric, and messages with similar media features and context behavior patterns are grouped into the same group, so as to divide the initial feature fusion dataset into multiple message populations, such as a marketing message population mainly with text-image interaction, a media notification population mainly with audio-video transmission, or a local service request population characterized by geographic location and service action; different information fusion strategies are configured for different populations according to their feature preferences and fusion needs

[0057] For example, the data fusion features corresponding to each enhancement matrix extracted by the initial convolutional neural network are taken as a plurality of groups of input vectors, and the input vectors are divided into a plurality of different message populations, to obtain 5G feature fusion data for representing different message populations:

[0058]

[0059] wherein M ij represents the initial 5G feature fusion dataset of different message populations, c represents the number of data fusion features corresponding to the enhancement matrix as input vectors, μ i represents the feature coordinates of the enhancement matrix, c j represents the standard value for dividing the input vectors into message populations, c k represents the standard value of the kth message population, and ε represents the division coefficient of the message population.

[0060] Step 205, based on the pre-trained convolutional neural network model, 5G information fusion is performed on the initial feature fusion dataset to obtain target 5G feature fusion data.

[0061] For example, based on the initial 5G feature fusion dataset and the pre-trained convolutional neural network model (such as a newly designed convolutional neural network 5G-FusionNet), 5G information fusion can be performed to obtain target 5G feature fusion data.

[0062] In some examples, first, an initial 5G feature fusion dataset is obtained, which should contain the features of various types of 5G messages after preliminary fusion. The dataset is divided into a training set, a validation set and a test set to facilitate subsequent model training and evaluation. A pre-trained convolutional neural network model (such as 5G-FusionNet) is loaded. The model can be trained on a large amount of 5G message data and already has the ability to extract and fuse 5G message features. Necessary preprocessing is performed on the initial 5G feature fusion dataset, such as data standardization, normalization, etc., to ensure that the data meets the requirements of the model input. The preprocessed data is input into the loaded convolutional neural network model. The model will process the input data according to its learned feature extraction and fusion capabilities and output the fused 5G features. The optimized model is used to process the test set to obtain the target 5G feature fusion data. These data will contain 5G message features fused by the convolutional neural network model and can be directly used for subsequent tasks (such as classification, prediction, etc.).

[0063] Optionally, step 205 can specifically include: inputting the initial feature fusion dataset into a pre-trained convolutional neural network model, performing uniform embedding representation on different types of message features through a hybrid embedding layer in the convolutional neural network model to generate multi-modal embedding vectors; using a plurality of parallel convolution paths in a fusion convolution layer to respectively extract local features of subsets of message types in the multi-modal embedding vectors, and output feature maps of each type of message; performing fusion processing on the feature maps from different convolution paths in a feature fusion layer to generate a fusion feature vector; and performing nonlinear transformation and dimension mapping on the fusion feature vector through a fully connected layer to output the target 5G feature fusion data.

[0064] The pre-trained convolutional neural network model includes an input layer, a hybrid embedding layer, a fusion convolution layer, a feature fusion layer and a fully connected layer connected in sequence. The input layer (Input Layer) can be used to receive original 5G message data as input, and can also input combinations of text, image, audio, video and other multimedia 5G data as input, and then perform necessary preprocessing according to the data type, such as text cleaning, image normalization, audio feature extraction, etc.

[0065] Exemplarily, a Hybrid Embedding Layer is used to convert different types of 5G message data into a unified embedding representation. For example, for text messages, word embeddings (e.g., Word2Vec, GloVe) are used to convert text into vectors; for image messages, convolutional layers are used to extract features; for audio and video messages, appropriate feature extraction methods are used, and embedding representations of various types of messages are output. A Fusion Convolutional Layer is used to fuse message features of different embedding representations through convolution operations. For example, by designing multiple parallel convolution paths, each path is for a type of message. Each convolution path includes convolution kernels, activation functions, and pooling layers (e.g., max pooling) to output a fused feature map. A Feature Fusion Layer is used to fuse feature maps of different convolution paths and fuse the feature maps together using methods such as concatenation or weighted averaging to output a fused feature vector.

[0066] It should be noted that the number of layers of the Hybrid Embedding Layer can be adaptively adjusted according to changes in model error or the amount of 5G fusion message data, specifically:

[0067]

[0068] wherein, represents the number of layers that need to be added to the Hybrid Embedding Layer, b and θ are respectively a first preset adjustment factor and a second preset adjustment factor, ΔE represents a change rate of the model error or the amount of 5G fusion message data.

[0069] In some examples, a Fully Connected Layer is used to further process and classify the fused features. The fused feature vector is input to the Fully Connected Layer, and an appropriate activation function (e.g., L ωa ) is used for output, and target 5G feature fusion data is output.

[0070] Activation function which can be represented by the following formula:

[0071]

[0072] wherein, represents the activation function under the current convolutional neural network parameters, represents the function expression of the Fully Connected Layer under the current convolutional neural network parameters, and γ represents a preset hyperparameter.

[0073] In some examples, the mixed embedding layer can process multiple types of 5G messages, convert the messages into a unified vector representation through different types of embedding methods; the fusion convolution layer adopts a parallel convolution path design, which can extract features for different types of messages and realize the fusion of features through convolution operations. The feature fusion layer fuses the features of different convolution paths together through methods such as splicing or weighted averaging to obtain a fused feature vector. The fully connected layer further processes and classifies the fused features to output target 5G feature fusion data.

[0074] In some examples, the convolutional neural network in this embodiment uniformly embeds multi-modal 5G messages such as text, images, audio, and video through a mixed embedding layer, converts different types of message content into computable numerical representations, and adopts multiple parallel convolution paths in the fusion convolution layer to extract local features of each type of message, preserving the uniqueness and spatial or temporal structure information of each modality; then in the feature fusion layer, the feature maps output by different paths are spliced or weighted integrated to realize deep fusion of cross-modal features and form high-dimensional and semantically rich fusion feature vectors; finally, through the fully connected layer, the fused features are nonlinearly transformed and dimensionally mapped to output structure-unified target 5G feature fusion data, which not only contains multi-modal features of the original message, but also fuses context awareness and high-level semantic information, and can support subsequent message classification, service recommendation, or interaction decision-making application tasks.

[0075] Optionally, the above fusion processing of the feature maps from different convolution paths in the feature fusion layer to generate a fusion feature vector can specifically include: obtaining the feature maps generated by the multiple parallel convolution paths output by the fusion convolution layer; splicing the feature maps according to the channel dimension or the spatial dimension to form a high-dimensional spliced feature map; and sequentially performing normalization operation and nonlinear activation processing on the high-dimensional spliced feature map to output the fusion feature vector.

[0076] For example, the feature maps can be spliced according to the channel dimension or the spatial dimension to form a high-dimensional spliced feature map, which enables information of different modalities to be represented in a unified space, enhancing the learning ability of cross-modal features; then the spliced high-dimensional feature map is sequentially subjected to normalization operation and nonlinear activation processing, the normalization operation helps to eliminate the dimensional difference between different features, ensuring the stability of subsequent processing, and the nonlinear activation processing introduces the nonlinear expression ability of the model, improving the expressiveness of the feature vector.

[0077] In some examples, as Figure 3As shown, based on the convolutional neural network for 5G message fusion, first, the original 5G message data set to be fused in the 5G message network can be obtained, which is converted into an original feature message vector, and for the format of each original 5G message, a feature mapping is performed based on a pre-constructed mapping table to convert the original feature message vector into a 5G message data set in a standard format; the enhancement matrix corresponding to each enhanced message in the standard format data set is determined, and these matrices are fused with the respective data fusion features to generate an initial 5G feature fusion data set; finally, based on the states in the initial fusion data set and the pre-trained convolutional neural network model, 5G information fusion is performed on each fusion state to obtain the target 5G feature fusion data.

[0078] For example, the convolutional neural network converts multiple types of 5G messages such as text, images, audio, and video into a unified embedding representation through a hybrid embedding layer, achieving unified modeling of heterogeneous data. The fusion convolutional layer sets multiple parallel paths to perform convolution operations on the embedding representations of each type of message, extract local features, and preserve modality specificity. The feature maps output by different paths are spliced or weighted integrated in the feature fusion layer to achieve deep fusion of cross-modal features. The fully connected layer performs nonlinear transformation on the fused high-dimensional feature vector to output target 5G feature fusion data with consistent structure, which contains multi-modal semantic information and can be used for downstream tasks such as classification, recommendation, or interactive decision-making.

[0079] Compared with the prior art, the embodiment first performs feature mapping of each original 5G message in the format of each original 5G message based on a pre-constructed original 5G message data feature relationship mapping table, to convert the original feature message vector into a 5G message data set in a standard format, then fuse each enhancement matrix and the respective data fusion features corresponding to each enhancement matrix to obtain an initial 5G feature fusion data set, and finally perform 5G information fusion on each fusion state based on the fusion states in the initial 5G feature fusion data set and the pre-trained convolutional neural network model to obtain target 5G feature fusion data, which can effectively process different types of 5G messages, realize feature fusion, and ultimately obtain target 5G feature fusion data through the processing of the fully connected layer, providing support for subsequent applications.

[0080] Further, as Figure 1 and Figure 2 The embodiment provides a data processing apparatus for the specific implementation of the method as shown in Figure 4 The apparatus includes an acquisition module 31, a conversion module 32, and a processing module 33.

[0081] The acquisition module 31 is configured to acquire an original 5G message data set to be fused based on a 5G message network;

[0082] The conversion module 32 is configured to convert the original 5G message data set into an initial feature fusion data set in a target format;

[0083] The processing module 33 is configured to perform 5G information fusion on the initial feature fusion data set based on a pre-trained convolutional neural network model to obtain target 5G feature fusion data.

[0084] In some examples of the embodiment, the conversion module 32 is specifically configured to convert the original 5G message data set into an original feature message vector; map the original feature message vector in different formats into a standard message data set in a target format based on a pre-constructed original 5G message data feature relationship mapping table; and perform feature fusion on the standard message data set based on an enhancement matrix corresponding to each 5G enhanced message in the standard message data set to obtain the initial feature fusion data set.

[0085] In some examples of the embodiment, the conversion module 32 is specifically further configured to identify an enhanced message type in the standard message data set; construct an enhancement matrix corresponding to the enhanced message type based on the enhanced message type, the enhancement matrix being used to represent the features of the messages of the enhanced message type; determine data fusion features matched with each enhancement matrix according to the original message content and / or external context information of the standard message data set; and perform fusion processing on the enhancement matrix and the data fusion features corresponding to the enhancement matrix to generate the initial feature fusion data set.

[0086] In some examples of the embodiment, the conversion module 32 is specifically further configured to determine a similarity measure between the feature vectors in the initial feature fusion data set based on the feature distribution of the enhancement matrix; divide the feature vectors in the initial feature fusion data set into a plurality of message populations based on the similarity measure; and perform differentiated information fusion processing on each message population.

[0087] In some examples of the embodiment, the processing module 33 is specifically configured to input the initial feature fusion data set into the pre-trained convolutional neural network model, perform uniform embedding representation on message features of different types through a hybrid embedding layer in the convolutional neural network model to generate a multi-modal embedding vector, perform local feature extraction on a subset of message types in the multi-modal embedding vector through a plurality of parallel convolution paths in a fusion convolution layer to output feature maps of each type of message, perform fusion processing on the feature maps from different convolution paths in a feature fusion layer to generate a fusion feature vector, and perform non-linear transformation and dimension mapping on the fusion feature vector through a fully connected layer to output the target 5G feature fusion data.

[0088] In some examples of the embodiment, the processing module 33 is further configured to acquire feature maps generated by the plurality of parallel convolution paths of the fusion convolution layer output; concatenate the feature maps in the channel dimension or the spatial dimension to form a high-dimensional concatenated feature map; and sequentially perform normalization operation and nonlinear activation processing on the high-dimensional concatenated feature map to output the fusion feature vector.

[0089] It should be noted that the other corresponding descriptions of the functions of the various functional units involved in the data processing device provided in the embodiment can be referred to the corresponding descriptions in Figure 1 and Figure 2 , which will not be repeated here.

[0090] Based on the above methods as shown in Figure 1 and Figure 2 , correspondingly, the embodiment also provides a computer readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above methods as shown in Figure 1 and Figure 2 .

[0091] Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a U disk, a mobile hard disk, etc.), and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of various implementation scenarios of the present application.

[0092] Based on the above methods as shown in Figure 1 and Figure 2 , and the virtual device embodiment as shown in Figure 4 , in order to achieve the above purpose, the embodiments of the present application also provide an electronic device, such as a personal computer, a server, a notebook computer, a smart robot, etc. The device includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to implement the above methods as shown in Figure 1 and Figure 2 .

[0093] Optionally, the above-mentioned physical device can also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, etc. The user interface can include a display screen (Display), an input unit such as a keyboard (Keyboard), etc. The optional user interface can also include a USB interface, a card reader interface, etc. The network interface can optionally include a standard wired interface, a wireless interface (such as a WI-FI interface), etc.

[0094] Those skilled in the art can understand that the above-mentioned entity device structure provided by the embodiment does not constitute a limitation on the entity device, and can include more or fewer components, or combine certain components, or different component arrangements.

[0095] The storage medium can also include an operating system, a network communication module. The operating system is a program for managing hardware and software resources of the above-mentioned entity device, supporting the running of information processing programs and other software and / or programs. The network communication module is used to realize the communication between the components in the storage medium and the communication with other hardware and software in the information processing entity device.

[0096] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary general hardware platform, or by hardware. By applying the scheme of the embodiment, compared with the prior art, the embodiment can efficiently process large-scale message data through automatic extraction and fusion of multi-modal features of 5G messages, significantly improve the data processing speed and analysis accuracy, and reduce the dependence on artificial rules and intervention, thereby providing stable and reliable data support for enterprises. The method endows the system with unified understanding ability of heterogeneous messages such as text, image, audio and video, forms deep feature expression, supports accurate modeling of user behavior, interaction mode and context environment, and thus digs out more valuable data insights. Enterprises can optimize service processes, improve product design, and develop intelligent new business forms based on these insights. At the operation level, the automatic fusion mechanism reduces the long-term investment in manpower and computing resources, and improves the maintainability and scalability of the system. The strong generalization ability of the model enables it to adapt to different industry scenarios and dynamically changing communication environments, effectively dealing with compatibility, abnormal message processing and other risks. On the user side, the system can realize personalized content pushing and interaction guidance combined with historical behavior and real-time context, enhancing the intuitiveness and responsiveness of services. As a key technology in the 5G message ecosystem, the method of the embodiment can provide bottom support for the digital service upgrade of financial, e-commerce, public service and other fields, help to realize the new interaction paradigm of message as a service, and promote the deep integration of communication capabilities and industry applications and industrial transformation.

[0097] It has to be noted that, in the present document, relational terms are intended only to convey a possible relationship between elements or

[0098] The above description is merely that of the specific embodiments of the application and as such is not to be taken in a limiting sense. Various modifications and co nti n uations will be evident to those skilled in the art that do not depart from the spirit and scope of the application as defined by the appended claims. The specific embodiments presented, therefore, are not to be considered in a limiting sense, but are presented for purposes of illustration only in conformance with the above-stated description. It is not the intention to limit the application to the described embodiments but rather the intention is to cover all modifications and alternatives falling within the spirit and scope of the application as defined by the appended claims.

Claims

1. A data processing method, characterized by, The method comprises the following steps: Based on the 5G message network, the original 5G message data set to be fused is obtained; The original 5G message data set is converted into an initial feature fusion data set of a target format; Based on the pre-trained convolutional neural network model, the initial feature fusion data set is subjected to 5G information fusion to obtain target 5G feature fusion data.

2. The method of claim 1, wherein, The original 5G message data set is converted into an initial feature fusion data set of a target format, which comprises the following steps: The original 5G message data set is converted into an original feature message vector; Based on the pre-constructed original 5G message data feature relationship mapping table, the original feature message vector of different formats is mapped into a standard message data set of a target format; Based on the 5G enhanced message corresponding to the enhanced matrix in the standard message data set, the standard message data set is subjected to feature fusion to obtain the initial feature fusion data set.

3. The method of claim 2, wherein, The initial feature fusion data set is obtained by fusing the enhanced matrix and the data fusion feature corresponding to the enhanced matrix, which comprises the following steps: Identify the enhanced message type in the standard message data set; Based on the enhanced message type, the enhanced matrix corresponding to the enhanced message type is constructed, and the enhanced matrix is used to represent the features of the messages of the enhanced message type; According to the original message content and / or external context information of the standard message data set, determine the data fusion features matched with each enhanced matrix; The enhanced matrix and the data fusion features corresponding to the enhanced matrix are fused to generate the initial feature fusion data set.

4. The method of claim 3, wherein, After the initial feature fusion data set is generated by fusing the enhanced matrix and the data fusion features corresponding to the enhanced matrix, the method further comprises the following steps: Based on the feature distribution of the enhanced matrix, determine the similarity measure between the feature vectors in the initial feature fusion data set; Based on the similarity measure, the feature vectors in the initial feature fusion data set are divided into multiple message populations; Each message population is subjected to differentiated information fusion processing.

5. The method of claim 1, wherein, The initial feature fusion data set is input into the pre-trained convolutional neural network model, and the mixed embedding layer in the convolutional neural network model is used to uniformly embed different types of message features to generate a multi-modal embedding vector; A plurality of parallel convolution paths in the fusion convolution layer are used to respectively extract local features of a subset of the corresponding message type in the multi-modal embedding vector, and output feature maps of each type of message; In the feature fusion layer, the feature maps from different convolution paths are fused to generate a fusion feature vector; The fusion feature vector is subjected to nonlinear transformation and dimension mapping through the full connection layer to output the target 5G feature fusion data. The initial feature fusion data set is obtained by fusing the enhanced matrix and the data fusion features corresponding to the enhanced matrix, which comprises the following steps:

6. The method of claim 5, wherein, ​ acquire feature maps generated by a plurality of parallel convolution paths of the fusion convolution layer output; concatenate the feature maps according to a channel dimension or a spatial dimension to form a high-dimensional concatenated feature map; perform a normalization operation and a nonlinear activation process on the high-dimensional concatenated feature map in sequence to output the fusion feature vector.

7. A data processing apparatus, characterized by, comprise: an acquisition module configured to acquire an original 5G message dataset to be fused based on a 5G message network; a conversion module configured to convert the original 5G message dataset into an initial feature fusion dataset in a target format; a processing module configured to perform 5G information fusion on the initial feature fusion dataset based on a pre-trained convolutional neural network model to obtain a target 5G feature fusion data.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by a processor to implement the method in any one of claims 1-6.

9. An electronic device comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, The processor executes the computer program to implement the method in any one of claims 1-6.

10. A computer program product having stored thereon a computer program, characterized in that, The computer program product is executed by a processor to implement the method in any one of claims 1-6.

Images