5G message processing method and system based on artificial intelligence, and storage medium

Abstract

The invention discloses a 5G message processing method and system based on artificial intelligence, and a storage medium, and relates to the technical field of data processing, and the method comprises the steps: analyzing modal type message data to set a time interception window, comprehensively setting a time interception window for the previous message condition and modal type message data condition of a user, and obtaining the modal type message data of the user; the time window can preliminarily intercept and divide continuous and relatively associated messages in time, and a reliable basis is provided for subsequent message unification and association. The method comprises the following steps: constructing a cross-modal alignment model, performing unified alignment on features under all modal types through the cross-modal alignment model, performing association analysis on the features, extracting value information on the basis of association information, improving the reliability and depth of message processing, completing deep mining and analysis of messages, and ensuring optimized development of message services.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular, to a 5G message processing method, system, and storage medium based on artificial intelligence. Background Art

[0002] As an upgraded form of text messages, 5G messages support various modal formats such as text, pictures, audio, and video, and have the characteristics of rich media, no need to install an APP, strong interactivity, high security, and wide coverage. These characteristics enable 5G messages to provide a richer information display and interaction experience, meeting the diverse communication needs of users. At the same time, the application of artificial intelligence technology further improves the processing efficiency and intelligence level of 5G messages, and realizes functions such as automatic classification, intelligent reply, and emotion perception of messages through natural language processing, machine learning and other technologies, bringing a more convenient, efficient, and intelligent information service experience to users.

[0003] In the prior art, current message processing often only stays at the surface information level, such as the simple transmission of content such as text, pictures, audio, and video, lacking in-depth mining of message data. This results in poor reliability and low depth of message processing, and cannot meet the growing demand for message service analysis.

[0004] Therefore, how to improve the reliability and depth of message processing is a technical problem to be solved at present. Summary of the Invention

[0005] The purpose of the present invention is to solve the problems existing in the prior art, and a 5G message processing method based on artificial intelligence is proposed, which includes: Collect message data generated by users on the message platform, distinguish the modal types of the message data, and analyze the modal type message data to set a time intercept window; Intercept the message data according to the time intercept window to generate message data segments, and extract the features of each modal type in the message data segments; Construct a cross-modal alignment model, and uniformly align the features under all modal types through the cross-modal alignment model; Perform correlation analysis on the unified features to obtain an associated message data set, extract value information according to the associated message data set, and visually display the value information.

[0006] In some embodiments of the present application, analyzing the modal type message data to set a time intercept window includes: Establish a message data timeline according to the timestamps of the user message data, and determine the relative activity of the user at different time periods; Set the time truncation window length based on the relative activity of the user at different time periods and the frequency of the actual modal type message data at different time periods.

[0007] In some embodiments of the present application, a message data timeline is established according to the timestamps of the user message data, and the relative activity of the user at different time periods is determined, including, Establish a message data timeline according to the timestamps of the user message data, mark the message interaction parameters at the corresponding positions on the message data timeline, and split the overall message data timeline into multiple message data timeline segments according to the total length of the message data timeline; Set the standard time interval through the mean and standard deviation of the time intervals between the previous message data in each message data timeline segment, and distinguish the two interaction types of continuous message interaction and discontinuous message interaction on each message data timeline segment by virtue of the standard time interval; For the two interaction types of continuous message interaction and discontinuous message interaction, determine the activity of each type according to the message interaction parameters respectively, and determine the relative activity of the user at different time periods; ; Wherein, is the relative activity of the user at time period, , 2 are the respective quantities of the messages of the two interaction types of continuous message interaction and discontinuous message interaction, , are respectively the combination weights of the th continuous message interaction and the th discontinuous message interaction, , are respectively the activities of the th continuous message interaction and the th discontinuous message interaction, , are respectively the conversion coefficients of the message quantity and the average interval time between messages of the continuous message interaction, , are respectively the message quantity and the average interval time between messages of the th continuous message interaction at time period, is the first constant at time period.

[0008] In some embodiments of the present application, set the time truncation window length based on the relative activity of the user at different time periods and the frequency of the actual modal type message data at different time periods, including, Calculate the frequency of occurrence of actual message data of each modality type in different time periods, and comprehensively obtain the frequency of occurrence of message data of all modality types to get the frequency of message data of modality types; Determine the time truncation window length according to the relative activity and the frequency of message data of modality types in the same time period.

[0009] In some embodiments of the present application, a cross-modal alignment model is constructed, including, Collect features of all modality types, determine the semantic content of features under each modality type, and divide the features under the modality type into positive sample pairs and negative sample pairs through the semantic content of the features. Both the positive sample pairs and the negative sample pairs include feature combinations of different modality types; Construct a training set and a validation set according to the positive sample pairs and negative sample pairs respectively, initialize the parameters of the model through the training set, describe the loss situation between the model output and the true label through a contrast loss function, and update the initialized model parameters through the evaluation metrics of the gradient descent algorithm and the initialized model parameters on the validation set to minimize the contrast loss function to implement the cross-modal alignment model; During the process of updating the initialized model parameters, dynamically adjust the learning rate in the gradient descent algorithm through a learning rate scheduler to improve the verification effect.

[0010] In some embodiments of the present application, during the process of updating the initialized model parameters, dynamically adjust the learning rate in the gradient descent algorithm through a learning rate scheduler, including, Analyze the positive sample pairs and negative sample pairs in the validation set to determine the target value of the evaluation metric, obtain the deviation amount according to the actual value and the target value of the evaluation metric, and adjust the learning rate through the deviation amount. The evaluation metrics include loss value and accuracy; ; Wherein, is the adjusted learning rate, is the initial learning rate, , are the influence weights of the loss value and the accuracy respectively, , are the loss value and the accuracy respectively, is the second constant.

[0011] In some embodiments of the present application, perform association analysis on the unified features to obtain an associated message data set, and extract valuable information according to the associated message data set, including, Perform association analysis through the Apriori algorithm to obtain an associated message data set, and extract valuable information through a data processing tool in the associated message data set.

[0012] Correspondingly, the present application also provides a 5G message processing system based on artificial intelligence, including: A first module, configured to collect message data generated by users on a message platform, distinguish the modal types of the message data, and analyze the modal type message data to set a time truncation window; A second module, configured to truncate the message data according to the time truncation window, generate message data segments, and extract features under each modal type in the message data segments; A third module, configured to construct a cross-modal alignment model, and uniformly align the features under all modal types through the cross-modal alignment model; A fourth module, configured to perform correlation analysis on the unified features, obtain an associated message data set, extract value information according to the associated message data set, and visually display the value information.

[0013] The present application also has a computer-readable storage medium, and the computer-readable storage medium stores one or more programs, and when the one or more programs are executed by an electronic device including a plurality of application programs, the electronic device is caused to execute the method described above.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. Analyze the modal type message data to set a time truncation window, and comprehensively set a time truncation window based on the user's previous message situation and the modal type message data situation. This time window can initially truncate and divide the relatively related messages that are continuous in time, providing a reliable basis for subsequent message unification and correlation.

[0015] 2. Construct a cross-modal alignment model, uniformly align the features under all modal types through the cross-modal alignment model, perform correlation analysis on the features, extract value information based on the associated information, improve the reliability and depth of message processing, complete the in-depth mining and analysis of messages, and ensure the optimized development of message services. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic flow chart of a 5G message processing method based on artificial intelligence proposed by the present invention; Figure 2 It is a schematic structural diagram of a 5G message processing system based on artificial intelligence proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments.

[0018] Reference Figure 1 , a 5G message processing method based on artificial intelligence, comprising, Step S101, collect message data generated by users on the message platform, distinguish the message data by modal type, and analyze the modal type message data to set a time capture window.

[0019] In this embodiment, a big data acquisition platform is used to obtain and integrate various types of data resources from the Internet through means such as web crawler technology. These data resources can include multiple modalities such as text, images, audio, and video. The data from different message platforms are integrated to form a unified data set. In the integration process, it is necessary to pay attention to issues such as data format unification, deduplication, and outlier processing. Modality refers to the way data is expressed or perceived, such as text, images, audio, etc. In multimodal data, each modality has its unique characteristics and representation methods. The time interception window is set in combination with the user's message interaction habits and the modal type message data. By analyzing the characteristics of different modal type message data and considering user behavior, we can set a suitable time interception window to capture messages that are relatively continuous in time and may be associated. This will provide a good foundation for subsequent message association analysis.

[0020] It should be noted that this solution mainly solves how to determine the correlation between messages, so as to more accurately extract valuable information and explore the deep meaning embodied in the messages.

[0021] In some embodiments of the present application, the modal type message data is analyzed to set the time interception window, including: Establish a message data timeline based on the timestamps of user message data and determine the relative activity of users in different time periods; The time interception window length is set based on the relative activity of users in different time periods and the frequency of actual modal type message data in different time periods.

[0022] In this embodiment, a message data timeline is established based on the user's previous message status, and the activity of user messages in different time periods is analyzed. The time interception window length is set based on the relative activity and the frequency of actual modal type message data in different time periods.

[0023] In some embodiments of the present application, a message data timeline is established based on the timestamp of the user message data, and the relative activity of the user in different time periods is determined, including: Establish a message data timeline according to the timestamp of the user message data, mark the message interaction parameters at the corresponding positions on the message data timeline, and split the entire message data timeline into multiple message data timeline segments according to the total length of the message data timeline; Set a standard time interval based on the mean and standard deviation of the time intervals between previous message data in each message data time axis segment, and distinguish between two interaction types, continuous message interaction and discontinuous message interaction, on each message data time axis segment by virtue of the standard time interval; For the two interaction types of continuous message interaction and discontinuous message interaction, determine the activity levels of their respective types according to the message interaction parameters, and determine the relative activity levels of the user at different time periods; ; Wherein, is the relative activity level of the user at time period, , 2 are the respective numbers of messages of the two interaction types of continuous message interaction and discontinuous message interaction, , are the combination weights of the th continuous message interaction and the th discontinuous message interaction respectively, , are the activity levels of the th continuous message interaction and the th discontinuous message interaction respectively, , are the conversion coefficients of the number of messages and the average interval time between messages of continuous message interaction respectively, , are the number of messages and the average interval time between messages of the th continuous message interaction at time period respectively, is the first constant at time period.

[0024] In this embodiment, the message interaction parameters include relevant parameters such as the number of times of sending, receiving, replying to messages, liking, commenting, etc., and calculate the mean and standard deviation of the time interval. Select a suitable multiple (such as 1 time, 2 times, etc.), and add and subtract the standard deviation of this multiple from the mean as the time interval threshold (standard time interval). Continuous message interaction is multiple messages whose time between messages is within the standard time interval, and discontinuous message interaction is a single message whose time between messages is outside the standard time interval. Determine the activity levels of their respective types according to the message interaction parameters (such as weighted summation), represents the correction of the number of messages and the average interval time between messages on continuous message interaction to the activity level. Both the number of messages and the average interval time can reflect the active situation, is to balance the magnitude of the correction function.

[0025] In some embodiments of the present application, the time truncation window length is set based on the relative activity of the user at different time periods and the frequency of the actual modal type message data at different time periods, including Calculating the frequency of occurrence of the actual message data of each modal type in different time periods, and obtaining the frequency of the modal type message data by integrating the frequencies of the message data of all modal types; Determining the time truncation window length for the relative activity and the frequency of the modal type message data in the same time period.

[0026] In this embodiment, the time truncation window length is set by combining the relative activity and the frequency, which improves the accuracy of truncation.

[0027] Step S102, truncating the message data according to the time truncation window to generate a message data segment, and extracting the features under each modal type in the message data segment.

[0028] In this embodiment, methods such as window subclassing are used to truncate the message data according to the set time truncation window to generate a message data segment. Ensure that the message data within each data segment is continuous in time. The time truncation window lengths in different time periods may be different, and the generated message data segments are numbered, stored, and managed for subsequent feature extraction and correlation analysis.

[0029] Text modality: The TF-IDF (Term Frequency-Inverse Document Frequency) feature extraction method can be used to convert the text data into a feature vector. In addition, techniques such as word embedding (such as Word2Vec, BERT, etc.) can be used to extract deeper text features.

[0030] Image modality: The HOG (Histogram of Oriented Gradients) feature extraction method can be used to capture the local shape and texture information of the image. In addition, a convolutional neural network (CNN) can be used to extract deeper image features.

[0031] Audio modality: The MFCC (Mel Frequency Cepstral Coefficients) feature extraction method can be used to convert the audio signal into a feature vector. In addition, a recurrent neural network (RNN) or a long short-term memory network (LSTM) can be used to extract deeper audio features.

[0032] Step S103, constructing a cross-modal alignment model, and uniformly aligning the features under all modal types through the cross-modal alignment model.

[0033] In this embodiment, since the message involves multiple modal contents, it is necessary to perform cross-modal alignment on the features of different modalities, establish a unified standard and space to facilitate subsequent comparison and correlation analysis. The cross-modal contrastive learning method, such as the UNIMO framework, is adopted to construct a cross-modal alignment model. This model obtains context-based feature representations through the self-attention mechanism to achieve alignment between different modal features. The features of different modal types are input into the cross-modal alignment model, and through the learning and optimization of the model, the features of different modalities are mapped into a unified semantic space to achieve unified alignment of features.

[0034] In some embodiments of the present application, constructing a cross-modal alignment model includes collecting features under all modal types, determining the semantic content of the features under each modal type, and dividing the features under the modal type into positive sample pairs and negative sample pairs through the semantic content of the features. Both the positive sample pairs and the negative sample pairs include feature combinations under different modal types; constructing a training set and a validation set according to the positive sample pairs and the negative sample pairs respectively, initializing the parameters of the model through the training set, describing the loss situation between the model output and the true label through a contrast loss function, and updating the parameters of the initialized model through the evaluation index of the gradient descent algorithm and the initialized model on the validation set to minimize the contrast loss function to implement the cross-modal alignment model; During the process of updating the parameters of the initialized model, the learning rate in the gradient descent algorithm is dynamically adjusted through a learning rate scheduler to improve the validation effect.

[0035] In this embodiment, positive sample pairs: feature pairs from different modalities but representing the same semantic information. The model should make the distance between these sample pairs in the feature space smaller to strengthen the similarity between them. Negative sample pairs: feature pairs from different modalities and representing different semantic information. The model should make the distance between these sample pairs in the feature space larger to distinguish the differences between them. The contrast loss function is a function that measures the difference between the model prediction value and the actual value. It quantifies the performance of the model by calculating the distance between these sample pairs in the feature space. The goal of the contrast loss function is to find the model parameters that minimize the value of the loss function. This is usually achieved through optimization algorithms such as gradient descent. During the training process, the model will continuously adjust the parameters according to the gradient information of the loss function to minimize the distance between positive sample pairs and maximize the distance between negative sample pairs.

[0036] It can be understood that the contrast loss function and the gradient descent algorithm are relatively common methods in this field, and the specific content will not be elaborated here. Other algorithms that can be completed also belong to the protection scope of the present application.

[0037] In some embodiments of the present application, during the process of updating the parameters of the initialization model, the learning rate in the gradient descent algorithm is dynamically adjusted through a learning rate scheduler, including, Analyzing the positive sample pairs and negative sample pairs in the validation set to determine the target value of the evaluation metric, obtaining the deviation amount based on the actual value and the target value of the evaluation metric, and adjusting the learning rate through the deviation amount. The evaluation metrics include the loss value and the accuracy; ; Among them, is the adjusted learning rate, is the initial learning rate, , are the influence weights of the loss value and the accuracy respectively, , are the loss value and the accuracy respectively, is the second constant.

[0038] In this embodiment, modern deep learning frameworks (such as TensorFlow, PyTorch, etc.) provide a variety of learning rate schedulers (such as ReduceLROnPlateau, etc.). These schedulers can dynamically adjust the learning rate according to the changes in the metrics on the validation set, thereby optimizing the training process. The loss value is an important metric for measuring the difference between the predicted value and the actual value of the model. During the training process, if the loss value on the validation set remains high or no longer decreases, it may mean that the model has fallen into a local optimum or the learning rate is too large, resulting in oscillations. At this time, reducing the learning rate may help the model jump out of the local optimum or stabilize the training process. The accuracy is an important metric for measuring the classification performance of the model. During the training process, if the accuracy on the validation set no longer improves, it may mean that the model has converged or the learning rate is too small, resulting in a slow training speed. At this time, adjusting the learning rate may help the model continue to optimize or accelerate the convergence speed.

[0039] In this embodiment, according to the specific task and the characteristics of the validation data set, reasonable target values or target change degrees are set for the loss value and the accuracy. Here, the target values include parameter values and parameter change rates. represents the correction of the initial learning rate due to the deviation between the loss value and the accuracy, exists to balance the size of the correction function.

[0040] Step S104, perform correlation analysis on the unified features to obtain a set of associated message data, extract value information according to the set of associated message data, and visually display the value information.

[0041] In this embodiment, association analysis algorithms such as the Apriori algorithm and the FP-growth algorithm are used to perform association analysis on the unified features. These algorithms can discover frequent item sets and association rules in the data, revealing the relationships between different modal features. The results of the association analysis are stored and managed for subsequent extraction of valuable information and visual display. Extraction of valuable information: Through data mining techniques, such as machine learning algorithms, analyze the associated message data set to extract valuable information. This information can include trend prediction, anomaly detection, user behavior analysis, etc. Visual display: Use data visualization techniques, such as platforms like Smartbi, to display the extracted valuable information in graphical form. Present the relationships and trends between the data through charts, graphs, etc., to help users better understand and utilize the data.

[0042] In this embodiment, surface information refers to the information directly presented by the message content itself, such as text descriptions, image displays, etc. These information are intuitive and easy to understand, but often lack deep associations and interpretations. Valuable information, on the other hand, is hidden beneath the surface information and needs to be revealed through in-depth mining and analysis using artificial intelligence techniques. This valuable information may include users' consumption habits, interest preferences, behavior patterns, etc., which have extremely high commercial value and social value for enterprises and governments.

[0043] In some embodiments of this application, perform association analysis on the unified features to obtain an associated message data set, and extract valuable information according to the associated message data set, including Perform association analysis through the Apriori algorithm to obtain an associated message data set, and extract valuable information from the associated message data set through a data processing tool.

[0044] Correspondingly, this application also provides an artificial intelligence-based 5G message processing system, as Figure 2 shown, including The first module is used to collect the message data generated by users on the message platform, distinguish the modal types of the message data, and analyze the modal type message data to set a time truncation window; The second module is used to truncate the message data according to the time truncation window to generate message data segments, and extract the features under each modal type in the message data segments; The third module is used to construct a cross-modal alignment model, and perform unified alignment on the features under all modal types through the cross-modal alignment model; The fourth module is used to perform association analysis on the unified features to obtain an associated message data set, extract valuable information according to the associated message data set, and perform visual display on the valuable information.

[0045] The present application also provides a computer-readable storage medium storing one or more programs, which, when executed by an electronic device including a plurality of application programs, cause the electronic device to execute the method described above.

[0046] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By analyzing the modal type message data to set a time truncation window, a time truncation window is comprehensively set based on the user's previous message situation and the modal type message data situation. This time window can preliminarily truncate and divide the relatively correlated messages that are continuous in time, providing a reliable basis for subsequent message unification and association.

[0047] 2. Construct a cross-modal alignment model to uniformly align the features under all modal types through the cross-modal alignment model, conduct correlation analysis on the features, extract value information based on the correlation information, improve the reliability and depth of message processing, complete the in-depth mining and analysis of messages, and ensure the optimized development of message services.

[0048] Through the description of the above embodiments, those skilled in the art can clearly understand that the present invention can be implemented by hardware or by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including several instructions for causing a computer device (such as a personal computer, a server, or a network device, etc.) to execute the methods described in various implementation scenarios of the present invention.

[0049] Those skilled in the art can understand that the drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the drawings are not necessarily essential for implementing the present invention.

[0050] Those skilled in the art can understand that the modules in the system in the implementation scenario can be distributed in the system of the implementation scenario according to the description of the implementation scenario, or can be correspondingly changed and located in one or more systems different from the present implementation scenario. The modules in the above implementation scenario can be combined into one module, or further split into multiple sub-modules.

[0051] The above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A 5G message processing method based on artificial intelligence, characterized in that: include, Collect message data generated by users on the message platform, distinguish the message data by modal type, and analyze the modal type message data to set the time interception window; The message data is intercepted according to the time interception window to generate a message data segment, and the features of each modal type in the message data segment are extracted; Construct a cross-modal alignment model to uniformly align features of all modal types; The unified features are subjected to correlation analysis to obtain a correlated message data set, and valuable information is extracted based on the correlated message data set, and the valuable information is visualized.

2. The 5G message processing method based on artificial intelligence according to claim 1, characterized in that: Analyze modal type message data to set the time interception window, including, Establish a message data timeline based on the timestamps of user message data and determine the relative activity of users in different time periods; The time interception window length is set based on the relative activity of users in different time periods and the frequency of actual modal type message data in different time periods.

3. The 5G message processing method based on artificial intelligence according to claim 2 is characterized in that: Establish a message data timeline based on the timestamp of the user's message data, and determine the relative activity of the user in different time periods. include, Establish a message data timeline according to the timestamp of the user message data, mark the message interaction parameters at the corresponding positions on the message data timeline, and split the entire message data timeline into multiple message data timeline segments according to the total length of the message data timeline; The standard time interval is set by the mean and standard deviation of the time intervals between the previous message data of each message data time axis segment, and the two interaction types of continuous message interaction and non-continuous message interaction on each message data time axis segment are distinguished by the standard time interval; For the two interaction types of continuous message interaction and non-continuous message interaction, the activity of each type is determined according to the message interaction parameters, and the relative activity of users in different time periods is determined; ； in, For users Relative activity during the time period, , 2 are the numbers of two types of interaction messages: continuous message interaction and non-continuous message interaction. , Respectively Continuous message interaction and The combined weight of non-continuous message interactions, , Respectively Continuous message interaction and The activity of non-continuous message interaction, , are the conversion coefficients of the number of messages in continuous message interaction and the average interval time between messages, , Respectively in Time period The number of messages in a continuous message interaction and the average interval time between messages, For The first constant under the time period.

4. The 5G message processing method based on artificial intelligence according to claim 2, characterized in that: The time interception window length is set based on the relative activity of users in different time periods and the frequency of actual modal type message data in different time periods. include, Calculate the frequency of occurrence of each modal type message data in different time periods, and combine the frequency of occurrence of all modal type message data to obtain the frequency of modal type message data; The length of the time capture window is determined based on the relative activity and frequency of modal type message data in the same time period.

5. The 5G message processing method based on artificial intelligence according to claim 1, characterized in that: Construct a cross-modal alignment model, including, Collect features under all modal types, determine the semantic content of features under each modal type, and divide the features under the modal type into positive sample pairs and negative sample pairs according to the semantic content of the features. Both positive sample pairs and negative sample pairs include feature combinations under different modal types. Construct training sets and validation sets based on positive sample pairs and negative sample pairs, respectively. Initialize the model parameters through the training set. Use the contrast loss function to describe the loss between the model output and the true label. Update the parameters of the initialized model through the gradient descent algorithm and the evaluation index of the parameters of the initialized model on the validation set to minimize the contrast loss function and realize the cross-modal alignment model. In the process of updating the parameters of the initialization model, the learning rate in the gradient descent algorithm is dynamically adjusted through the learning rate scheduler to improve the verification effect.

6. The 5G message processing method based on artificial intelligence according to claim 5, characterized in that: In the process of updating the parameters of the initialization model, the learning rate in the gradient descent algorithm is dynamically adjusted through the learning rate scheduler. include, Analyze the positive and negative sample pairs in the validation set to determine the target value of the evaluation index. According to the actual value and target value of the evaluation index, the deviation is obtained. The learning rate is adjusted by the deviation. The evaluation index includes loss value and accuracy. ； in, is the adjusted learning rate, is the initial learning rate, , are the influence weights of loss value and accuracy respectively, , are the loss value and accuracy respectively, is the second constant.

7. The 5G message processing method based on artificial intelligence according to claim 1, characterized in that: Perform correlation analysis on the unified features to obtain a correlated message data set, and extract valuable information based on the correlated message data set, including: The Apriori algorithm is used to perform association analysis to obtain an associated message data set, and valuable information is extracted from the associated message data set using data processing tools.

8. A 5G message processing system based on artificial intelligence, characterized in that: include, The first module is used to collect message data generated by users on the message platform, distinguish the message data by modal type, and analyze the modal type message data to set the time interception window; The second module is used to intercept the message data according to the time interception window, generate message data segments, and extract features under each modal type in the message data segments; The third module is used to build a cross-modal alignment model, which uniformly aligns the features of all modal types. The fourth module is used to perform correlation analysis on the unified features to obtain a correlated message data set, extract value information based on the correlated message data set, and visualize the value information.

9. A computer-readable storage medium storing one or more programs, which, when executed by an electronic device including a plurality of application programs, causes the electronic device to execute any one of the methods of claims 1-7.

Images