Multimedia information fusion pushing method and system based on 5G message

Abstract

The invention relates to the technical field of information pushing, and discloses a multimedia information fusion pushing method and system based on a 5G message, and the method comprises the steps: carrying out the sequence expansion of a user behavior sequence of a target user, and obtaining a target behavior sequence; extracting local behavior characteristics and global behavior characteristics of the target behavior sequence, and performing information fusion on the local behavior characteristics and the global behavior characteristics to obtain user behavior characteristics; calculating a multimedia interest resource set according to the user behavior characteristics, and performing resource filtering on the interest resource set to obtain a multimedia resource set; extracting hierarchical features of each resource in the multimedia resource set, and constructing target features according to the hierarchical features; and compiling a 5G message signal according to the target feature, and pushing multimedia information according to the 5G message signal. According to the invention, the accuracy of multimedia information pushing can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of information push, and in particular, to a multimedia information fusion push method and system based on 5G messages. Background Art

[0002] With the rapid development of the communication industry, the integrated application of big data technology and 5G communication technology in the field of multimedia information is becoming increasingly extensive and in-depth. In the field of multimedia information, compared with traditional text messages, 5G messages can support various media forms such as text, pictures, audio, video, location, contacts, and documents. They can achieve precise push of multimedia information based on multi-dimensional data such as user characteristics and behavior habits, ensuring that users receive personalized and demand-compliant information in different scenarios, while meeting the diverse demands of enterprises for precise marketing and service push, and can expand the coverage of the push audience.

[0003] However, multimedia information from different sources has differences in formats, encodings, resolutions, etc., resulting in a large amount of format conversion and adaptation work required during data fusion, increasing the technical difficulty and complexity, and information loss or quality degradation may occur during the conversion process. At the same time, it is difficult to fully explore the personalized preferences and needs of users during message push, and the pertinence and precision of the push content are insufficient, resulting in users receiving a large amount of multimedia information they are not interested in, reducing the utilization efficiency of information and user satisfaction. Therefore, how to improve the accuracy of multimedia information push has become an urgent problem to be solved. Summary of the Invention

[0004] The present invention provides a multimedia information fusion push method and system based on 5G messages, and its main purpose is to solve the problem of poor accuracy in multimedia information push.

[0005] Obtain the user behavior sequence of the target user, perform sequence expansion on the user behavior sequence to obtain a target behavior sequence; Extract the local behavior features and global behavior features of the target behavior sequence, and perform information fusion on the local behavior features and the global behavior features to obtain user behavior features; Calculate the multimedia interest resource set of the target user according to the user behavior features, and perform resource filtering on the interest resource set to obtain a multimedia resource set; Extract the hierarchical features of each resource in the multimedia resource set, and construct the target features of the multimedia resource set according to the hierarchical features; Compile a 5G message signal according to the target features, and perform multimedia information push according to the 5G message signal.

[0006] Optionally, the sequence augmentation of the user behavior sequence to obtain a target behavior sequence includes: Reverse sort the user behavior sequence to obtain a reverse behavior sequence; Construct a sequence embedding matrix for the reverse behavior sequence, and calculate the behavior preference scores of each user behavior in the user behavior sequence according to the sequence embedding matrix; Select the user behavior with the largest behavior preference score and add it to the user behavior sequence to obtain an updated behavior sequence; Iterate using the updated behavior sequence as the user behavior sequence until the number of iterations is greater than a preset number threshold to obtain a target behavior sequence.

[0007] Optionally, the extraction of the local behavior features and global behavior features of the target behavior sequence includes: Construct a behavior node graph based on the target behavior sequence, and perform self-attention calculation on the behavior node graph to obtain local behavior features; Identify the neighbor nodes of each behavior node in the behavior node graph, and calculate the attention coefficients of the neighbor nodes; Perform feature aggregation on the neighbor nodes according to the attention coefficients to obtain node aggregation features; Update the features of the behavior nodes according to the node aggregation features to obtain global behavior features.

[0008] Optionally, the information fusion of the local behavior features and the global behavior features to obtain user behavior features includes: Perform feature splicing on the local behavior features to obtain local fusion features; Perform feature weighted average on the global behavior features to obtain global fusion features; Perform sum pooling on the local fusion features and the global fusion features to obtain user behavior features.

[0009] Optionally, the calculation of the multimedia interest resource set of the target user according to the user behavior features includes: Calculate the user similarity between the target user and the users in a preset user set according to the user behavior features; Construct a set of similar users according to the user similarity, and obtain the interest resource scores corresponding to the set of similar users; Calculate the user resource score of the target user according to the user similarity and the interest resource scores; Determine the multimedia interest resource set of the target user according to the user resource score.

[0010] Optionally, extracting the hierarchical features of each resource in the multimedia resource set includes: Performing vector transformation and type division on the multimedia resource set to obtain text resource vectors and other resource vectors; Performing multi-layer encoding on the text resource vectors to obtain multi-layer encoding features; Performing mean pooling on the multi-layer encoding features to obtain hierarchical pooling features, and performing feature splicing on the hierarchical pooling features to obtain text hierarchical features; Performing multi-layer convolutional pooling on the other resource vectors to obtain pooling features, and performing fully connected processing on the pooling features to obtain other hierarchical features; Aggregating the text hierarchical features and the other hierarchical features to obtain the hierarchical features of the multimedia resource set.

[0011] Optionally, constructing the target features of the multimedia resource set according to the hierarchical features includes: Constructing a feature vector sequence according to the hierarchical features, and performing attention calculation on the feature vector sequence to obtain attention features; Performing residual connection on the attention features and the target features to obtain connection features; Performing regularization processing on the connection features to obtain regularization features, and performing fully connected feed-forward calculation on the regularization features to obtain feed-forward features; Performing residual connection on the feed-forward features and the regularization features to obtain initial target features; Performing regularization processing on the initial target features to obtain the target features of the multimedia resource set.

[0012] Optionally, formulating a 5G message signal according to the target features: Performing channel coding on the target features to obtain coded information; Performing signal modulation on the coded information to obtain a communication signal of the coded information; Performing wavelet transform on the communication signal to obtain wavelet coefficients; Calculating a coefficient threshold of the communication signal according to the wavelet coefficients, and extracting target coefficients of the wavelet coefficients according to the coefficient threshold; Performing wavelet reconstruction on the target coefficients to obtain a 5G message signal.

[0013] Optionally, performing signal modulation on the coded information to obtain a communication signal of the coded information includes: Determining the transmission distance and channel quality index corresponding to the coded information; When the transmission distance is greater than a preset communication threshold, perform constant envelope data modulation processing on the encoded information to obtain a constant envelope modulation signal; Perform delay compensation processing on the constant envelope modulation signal according to the channel quality index to obtain a communication signal.

[0014] To solve the above problems, the present invention also provides a multimedia information fusion push system based on 5G messages, and the system includes: A sequence expansion module, configured to obtain a user behavior sequence of a target user, and perform sequence expansion on the user behavior sequence to obtain a target behavior sequence; A feature information fusion module, configured to extract local behavior features and global behavior features of the target behavior sequence, and perform information fusion on the local behavior features and the global behavior features to obtain user behavior features; A multimedia resource set calculation module, configured to calculate a multimedia interest resource set of the target user according to the user behavior features, and perform resource filtering on the interest resource set to obtain a multimedia resource set; A target feature construction module, configured to extract hierarchical features of each resource in the multimedia resource set, and construct target features of the multimedia resource set according to the hierarchical features; An information push module, configured to prepare a 5G message signal according to the target features, and perform multimedia information push according to the 5G message signal.

[0015] In the embodiment of the present invention, by performing sequence expansion on the user behavior sequence of the target user, learning the dependency relationship between different user behaviors, and obtaining a more comprehensive and accurate target behavior sequence of user behaviors; extracting local behavior features and global behavior features of the target behavior sequence, and performing information fusion to obtain user behavior features, more rich feature information can be obtained, and thus the behavior motivation and intention of the user can be understood more comprehensively; calculating a multimedia interest resource set according to the user behavior features, and performing resource filtering to obtain a multimedia resource set, removing bad information can ensure the security and accuracy of information push; extracting hierarchical features of each resource, and constructing target features according to the hierarchical features, the feature information in the multimedia resource set can be fused, so that the target features contain the feature information of each resource in the multimedia resource set, providing a basis for subsequent multimedia information push; preparing a 5G message signal according to the target features, and performing multimedia information push according to the 5G message signal, can accurately push the multimedia resource set to the target user, effectively improve the quality of information push, and further improve the accuracy of multimedia information push. Therefore, the multimedia information fusion push method and system based on 5G messages proposed by the present invention can solve the problem of poor accuracy of multimedia information push. Description of the Drawings

[0016] Figure 1 Flow schematic diagram of a multimedia information fusion push method based on 5G messages provided by an embodiment of the present invention; Figure 2 Flow schematic diagram of extracting local behavior features and global behavior features of a target behavior sequence provided by an embodiment of the present invention; Figure 3 Flow schematic diagram of calculating a multimedia interest resource set of a target user provided by an embodiment of the present invention; Figure 4 Functional module diagram of a multimedia information fusion push system based on 5G messages provided by an embodiment of the present invention.

[0017] The realization, functional characteristics and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. Detailed implementation manners

[0018] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0019] The embodiments of the present application provide a multimedia information fusion push method based on 5G messages. The execution subject of the multimedia information fusion push method based on 5G messages includes, but is not limited to, at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiments of the present application. In other words, the multimedia information fusion push method based on 5G messages can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to: a single server, a server cluster, a cloud server, or a cloud server cluster, etc. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms.

[0020] Refer to Figure 1 As shown, it is a flow schematic diagram of a multimedia information fusion push method based on 5G messages provided by an embodiment of the present invention. In this embodiment, the multimedia information fusion push method based on 5G messages includes: S1. Obtain the user behavior sequence of the target user, and perform sequence expansion on the user behavior sequence to obtain a target behavior sequence.

[0021] In the embodiments of the present invention, the user behavior sequence is a set of various types of information generated during the target user's use of the platform's functions and interaction with the platform, sorted according to the generation time. It may include the residence time of the target user when browsing goods, reflecting the degree of attention of the user to a certain commodity; the search keywords of the target user for the commodity, reflecting the demand direction of the target user; the playing frequency of the target user for music, reflecting the music preferences of the target user; and may also include the frequency and time pattern of the target user sharing the location, etc. The user behavior sequence contains rich user preference information. According to the user behavior sequence, the potential preferences and interest points of the user can be analyzed from multiple dimensions. Furthermore, from a vast amount of project resources, those projects that the target user may be interested in can be accurately recommended to the target user.

[0022] Specifically, sequence expansion is to supplement the information in the user behavior sequence to prevent poor accuracy of information push caused by too little behavior information in the user behavior sequence of the target user.

[0023] In the embodiments of the present invention, performing sequence expansion on the user behavior sequence to obtain a target behavior sequence includes: Performing reverse sorting on the user behavior sequence to obtain a reverse behavior sequence; Constructing a sequence embedding matrix of the reverse behavior sequence, and calculating the behavior preference score of each user behavior in the user behavior sequence according to the sequence embedding matrix; Selecting the user behavior with the largest behavior preference score and adding it to the user behavior sequence to obtain an updated behavior sequence; Taking the updated behavior sequence as the user behavior sequence for iteration until the number of iterations is greater than a preset number threshold to obtain a target behavior sequence.

[0024] In the embodiments of the present invention, reversing the front-back order of the user behavior sequence to obtain a reverse behavior sequence. The sequence embedding matrix is a representation method that maps the reverse behavior sequence to a low-dimensional vector space. For the elements in the reverse behavior sequence, a corresponding sequence embedding matrix can be converted through a pre-constructed recurrent neural network (RNN) or Transformer and other architectures using the corresponding initial embedding matrix.

[0025] Furthermore, calculating the behavior preference score of each user behavior is to multiply the sequence embedding matrix by the weight matrix of each user behavior that can be learned using a bidirectional self-attention architecture to obtain the behavior preference score of each user behavior. According to the user phase with the largest behavior preference score, it is added to the user behavior sequence to obtain an updated behavior sequence, and the updated behavior sequence is used for iteration to obtain a target behavior sequence.

[0026] Among them, the bidirectional self-attention architecture includes a multi-head self-attention mechanism (Multi-Head Attention) with residual connections and a feed-forward neural network (Feed-Forward Neural Network). Through the bidirectional self-attention architecture, the dependencies between different user behaviors can be learned, and then the user behavior sequence can be predicted to obtain a more comprehensive and accurate target behavior sequence of user behaviors.

[0027] S2. Extract the local behavior features and global behavior features of the target behavior sequence, and fuse the local behavior features and the global behavior features to obtain user behavior features.

[0028] In the embodiments of the present invention, the local behavior features are the behavior transition feature information included in each individual target behavior in the target behavior sequence, and the global behavior features are the cross-behavior feature information between target behaviors, thereby realizing more fine-grained feature extraction.

[0029] Specifically, referring to Figure 2 as shown, the extraction of the local behavior features and global behavior features of the target behavior sequence includes: S21. Construct a behavior node graph according to the target behavior sequence, and perform self-attention calculation on the behavior node graph to obtain local behavior features; S22. Identify the neighbor nodes of each behavior node in the behavior node graph, and calculate the attention coefficients of the neighbor nodes; S23. Aggregate the features of the neighbor nodes according to the attention coefficients to obtain node aggregation features; S24. Update the features of the behavior nodes according to the node aggregation features to obtain global behavior features.

[0030] In the embodiments of the present invention, the behavior node graph is a node graph constructed with each target behavior in the target behavior sequence as a node. Among them, the edges between the nodes include unidirectional and bidirectional. The unidirectional edge indicates that there is a unidirectional transition between the nodes in the target behavior sequence. For example, the positions of node 1 and node 2 are adjacent in the target behavior sequence. The bidirectional edge indicates that there is a bidirectional transition between the nodes in the target behavior sequence. For example, the positions of node 1 and node 2 are bidirectionally adjacent in the target behavior sequence. For example, the positions of node 1 and node 2 in the target behavior sequence are (...1,2...2,1...).

[0031] Furthermore, the self-attention calculation is to calculate the attention coefficients of each behavior node according to the behavior node graph using the self-attention mechanism, and multiply the attention coefficients with the node features element-wise to obtain local behavior features.

[0032] Specifically, a neighbor node is a user behavior of each behavior node in the target behavior sequence with a time interval within a preset sequence interval. Feature concatenation is performed on the grounding features of each neighbor node, as well as the dot product results of the feature points between each behavior node and its neighbor nodes. The concatenated feature result is multiplied by the dot product feature result and then multiplied by a preset first feature weight, and is activated using the Relu activation function to obtain the attention coefficient of each neighbor node. The attention coefficients of the neighbor nodes are multiplied by the node features of each behavior node respectively and then added together to obtain the final node aggregation feature.

[0033] Specifically, a preset Graph Neural Network (GNN) can be used to update the node features of the behavior nodes and the node aggregation features of the neighbor nodes to obtain the global behavior features that include the nodes themselves and their neighbor nodes.

[0034] In the embodiments of the present invention, information fusion is to fuse the feature information representing the transition feature information of individual target behaviors and the cross-behavior feature information between target behaviors, so as to learn comprehensive user behavior features.

[0035] Specifically, the step of fusing the local behavior features and the global behavior features to obtain user behavior features includes: Performing feature concatenation on the local behavior features to obtain local fusion features; Performing feature weighted average on the global behavior features to obtain global fusion features; Performing sum pooling processing on the local fusion features and the global fusion features to obtain user behavior features.

[0036] In the embodiments of the present invention, feature concatenation is to connect the local behavior features of different behavior nodes in the same dimension to form a higher-dimensional feature representation, which can retain all feature information intact and learn the complex associations between different local behavior features.

[0037] Feature weighted summation is to multiply multiple global behavior features by corresponding weight coefficients respectively and then add them together to obtain a comprehensive feature representation, which can highlight more important features and suppress less important features, and is helpful to improve the accuracy of subsequent multimedia interest resource set calculation.

[0038] Further, the sum pooling processing is to perform pooling processing using the SumPooling function. The local fusion features and the global fusion features are divided into several regions, then the features in each region are summed, and finally the sum results of all regions are concatenated as the user behavior features.

[0039] In the embodiments of the present invention, by fusing the local behavior features and the global behavior features, richer feature information can be obtained, so as to more comprehensively understand the user's behavior motivation and intention, avoid one-sidedness caused by only focusing on local behaviors, and effectively improve the accuracy of subsequent interest resource calculation.

[0040] S3. Calculate the multimedia interest resource set of the target user according to the user behavior features, and filter the resources in the interest resource set to obtain a multimedia resource set.

[0041] In the embodiments of the present invention, the multimedia interest resources are push resources that may be of interest to users and are presented in various forms (such as videos, audios, images, texts, etc.).

[0042] Specifically, refer to Figure 3 As shown, calculating the multimedia interest resource set of the target user according to the user behavior features includes: S31. Calculate the user similarity between the target user and the users in the preset user set according to the user behavior features; S32. Construct a similar user set according to the user similarity, and obtain the interest resource scores corresponding to the similar user set; S33. Calculate the user resource score of the target user according to the user similarity and the interest resource scores; S34. Determine the multimedia interest resource set of the target user according to the user resource score.

[0043] In the embodiments of the present invention, the preset user set can be a set composed of users who have had information interaction with the target user. For example, users such as social network platform communication users, interest club members, and forum netizens. The user similarity between users is calculated using the user behavior features. Among them, the feature similarity between the user behavior features of the users in the user set and the user behavior features of the target user can be calculated to obtain the user similarity.

[0044] Furthermore, select the preset number of users with the largest user similarity to construct a similar user set, and obtain the historical scores of each similar user in the similar user set for multimedia resources to obtain the interest resource scores. Among them, the interest resource score is a quantitative score given by the user to items (such as products, services, applications, courses, etc.) according to their own usage experience, satisfaction, and other relevant factors, usually presented in digital form, such as 1 to 5 stars, 1 to 10 points, etc.

[0045] Specifically, the following formula can be used to calculate the user resource score of the target user: Where represents the target user The user resource score for the resource represents the user similarity between the target user and the users in the set of similar users and the user interest resource score for the resource represents the total number of users in the set of similar users. For the resource the interest resource score represents the total number of users in the set of similar users.

[0046] Specifically, sort the multimedia resources corresponding to the user resource scores from high to low to obtain a multimedia resource sequence, and select a preset number of multimedia resources with the highest scores from the multimedia resource sequence to construct a multimedia interest resource set for information push.

[0047] Specifically, resource filtering is to filter out the bad information in the multimedia interest resource set to ensure the security and accuracy of multimedia information push.

[0048] Specifically, a homophone corpus and a deformed word corpus corresponding to the bad words can be constructed on the basis of the bad word corpus, and the keywords in the multimedia interest resource set are matched according to the corpus, and multimedia interest resources with a preset threshold number of matches are extracted from the multimedia interest resource set to obtain a multimedia resource set.

[0049] Specifically, the multimedia interest resource set can be processed, the text of each resource in the multimedia interest resource set can be extracted, the function words, punctuation marks, etc. can be removed according to the general function word corpus, and the word frequency of the processed phrases can be counted. According to the set quantity, the keywords ranked at the front are extracted in turn to obtain the keywords of each multimedia interest resource in the multimedia interest resource set.

[0050] S4. Extract the hierarchical features of each resource in the multimedia resource set, and construct the target features of the multimedia resource set according to the hierarchical features.

[0051] In the embodiment of the present invention, the hierarchical feature is to classify the feature information of each resource, including low-level and high-level features. The low-level features can intuitively display the information of the resource, and the high-level features can more directly reflect the semantic information and internal meaning of each resource, so as to comprehensively perform feature analysis through the low-level features and high-level features.

[0052] Specifically, the extraction of the hierarchical features of each resource in the multimedia resource set includes: Performing vector conversion and type division on the multimedia resource set to obtain a text resource vector and other resource vectors; Perform multi-layer encoding on the text resource vector to obtain multi-layer encoded features; Perform average pooling on the multi-layer encoded features to obtain hierarchical pooling features, and perform feature concatenation on the hierarchical pooling features to obtain text classification features; Perform multi-layer convolutional pooling on the other resource vector to obtain pooling features, and perform fully connected processing on the pooling features to obtain other classification features; Aggregate the text classification features and the other classification features to obtain the classification features of the multimedia resource set.

[0053] In the embodiments of the present invention, vector conversion can be performed on each resource in the multimedia resource set through a pre-constructed encoding module. For example, different types of vector conversions can be performed on resources of different modalities. For example, the Word2Vec can be used to convert text resources into vectors, the convolutional neural network can be used to perform vector conversion on images and videos, and the audio can be vector-converted through waveform encoding.

[0054] Further, the vectors after vector conversion of the multimedia resource set are divided to obtain a text resource vector corresponding to the text resource and other resource vectors corresponding to other resources.

[0055] Specifically, the text resource vector is encoded using a multi-layer Transformer encoder to obtain multi-layer encoded features for each layer of encoding. Then, average pooling is performed on the encoded features of each layer in the multi-layer encoded features to obtain hierarchical pooling features, and feature concatenation is performed on the hierarchical pooling features to obtain text classification features.

[0056] Specifically, the multi-layer convolutional pooling is the pooling feature of each layer obtained by sequentially performing Dense Block multi-layer convolutional processing, convolutional layer, and pooling layer processing on the other resource vector. Then, each layer of pooling feature is subjected to fully connected processing to obtain other classification features with consistent feature dimensions. Each Dense Block contains multiple convolutional layers, and these convolutional layers are connected together in a specific manner, usually using a 3x3 convolutional kernel. Each layer in the Dense Block will connect its input with the outputs of all previous layers in the channel dimension. Through the Dense Block, the flow and interaction of features between different layers are strengthened, so that each layer can obtain richer feature information, thereby improving the feature content of the other classification features and obtaining more accurate other classification features.

[0057] In the embodiments of the present invention, by classifying the multimedia resource set, feature extraction can be performed on different types of resources, and at the same time, classification features with different feature information can be extracted. Through the classification features, the feature information of the multimedia resources can be comprehensively obtained.

[0058] In the embodiments of the present invention, the target features for constructing the multimedia resource set according to the hierarchical features include: Constructing a feature vector sequence according to the hierarchical features, calculating the attention of the feature vector sequence to obtain attention features; Performing a residual connection between the attention features and the target features to obtain connection features; Performing regularization processing on the connection features to obtain regularized features, and performing a fully connected feedforward calculation on the regularized features to obtain feedforward features; Performing a residual connection between the feedforward features and the regularized features to obtain initial target features; Performing regularization processing on the initial target features to obtain the target features of the multimedia resource set.

[0059] In the embodiments of the present invention, the feature vector sequence is obtained by vector splicing of the hierarchical features of each resource to obtain a feature vector sequence composed of multiple hierarchical features. The feature vector sequence is multiplied by three learnable projection matrices to obtain Q, K, and V vectors. Then, the attention weights of each feature of the feature vector sequence to other features are calculated through Q and K, and the product of the attention weights and V is used as the attention feature.

[0060] Specifically, the feature vector sequence and the attention features are added through a residual connection to obtain connection features. Regularization processing refers to imposing certain constraint conditions on the connection features, enabling the connection features to better perform subsequent feature fusion while meeting specific requirements. For example, regularization processing can be performed through orthogonal constraints of the feature vectors.

[0061] Furthermore, the fully connected feedforward calculation is performed using a pre-constructed fully connected feedforward network (Fully - Connected Feedforward Network). Among them, the fully connected feedforward network includes an input layer, a hidden layer, and an output layer. The regularized features enter the network from the input layer and are passed to the hidden layer neurons after preliminary processing by the input layer neurons. In the hidden layer neurons, the regularized features are multiplied by the corresponding weights, added with biases, and then subjected to a non-linear transformation through an activation function to obtain the output of the neurons. Then, these outputs serve as the input of the next layer and continue to propagate forward until the output layer obtains the final prediction result, that is, the feedforward features.

[0062] Among them, the neurons in the fully connected feedforward network are interconnected in a specific manner, that is, all neurons in the previous layer are connected to all neurons in the next layer, forming a fully connected relationship, and the information is transmitted unidirectionally between the neurons, starting from the input layer and propagating forward layer by layer to the output layer, without feedback connections.

[0063] In an embodiment of the present invention, by constructing a target feature, the feature information in the multimedia resource set can be fused, so that the target feature contains the feature information of each resource in the multimedia resource set, providing a basis for subsequent multimedia information push.

[0064] S5. Compile a 5G message signal according to the target feature, and perform multimedia information push according to the 5G message signal.

[0065] In an embodiment of the present invention, the 5G message signal is a 5G communication signal that can perform data transmission. Through the 5G message signal, multimedia resources can be transmitted, thereby realizing multimedia information push.

[0066] Specifically, the compiling of the 5G message signal according to the target feature: Perform channel coding on the target feature to obtain coded information; Perform signal modulation on the coded information to obtain a communication signal of the coded information; Perform wavelet transform on the communication signal to obtain wavelet coefficients; Calculate a coefficient threshold of the communication signal according to the wavelet coefficients, and extract target coefficients of the wavelet coefficients according to the coefficient threshold; Perform wavelet reconstruction on the target coefficients to obtain a 5G message signal.

[0067] In an embodiment of the present invention, channel coding is to encode the target feature during transmission, thereby improving the reliability and accuracy of transmission. For example, channel coding can be performed by low-density parity-check (LDPC) coding or polarization coding to obtain coded information.

[0068] Furthermore, signal modulation is used to convert the coded information into a form suitable for transmission, improve the anti-interference ability and transmission efficiency of the signal, and encode the information by changing the amplitude, frequency or phase of the coded information to obtain a communication signal suitable for information push. The client can receive and decode through the information compilation method to obtain the multimedia resources included in the target feature.

[0069] Specifically, the performing of signal modulation on the coded information to obtain a communication signal of the coded information includes: Determine the transmission distance and channel quality index corresponding to the coded information; When the transmission distance is greater than a preset communication threshold, perform constant envelope data modulation processing on the coded information to obtain a constant envelope modulation signal; Perform delay compensation processing on the constant envelope modulation signal according to the channel quality index to obtain a communication signal.

[0070] Specifically, the transmission distance is the distance between the terminal where the target user is located and the transmitting base station, and the channel quality indicator is an indicator reflecting the communication channel quality of the terminal where the target user is located. For example, channel quality indicators such as the current channel signal-to-noise ratio and the current channel capacity.

[0071] Specifically, constant envelope data modulation processing refers to a modulation method in which the amplitude of the modulated signal remains constant, and only the phase or frequency changes. Gaussian filtering can be performed on the signal baseband and then modulated to make the spectral characteristics of the signal better.

[0072] Specifically, delay compensation processing is to appropriately delay or advance the signal according to the channel quality indicator, so that the timing of the signal is aligned with the signal under normal delay conditions. For example, when the channel quality is poor, such as in a strong interference environment or a region with severe signal fading, the delay of the received signal may fluctuate greatly. At this time, a larger delay compensation range is required. The signal can be appropriately delayed or advanced in the equalizer or delay adjustment circuit at the receiving end of the target user to align the timing of the signal with the signal under normal delay conditions. Among them, the channel quality can be identified according to the channel quality indicator by using a preset rule.

[0073] Specifically, a request signaling can be sent to obtain the signaling transmission time. According to the preset transmission expected time and the signaling transmission time, the signal transmission delay is calculated, and then delay compensation processing is performed according to the signal transmission delay, so that the communication signal is less interfered by the delay and the time synchronization accuracy of multimedia information push is improved.

[0074] Further, wavelet transform is to perform layer-by-layer wavelet decomposition on the communication signal using wavelet functions to obtain wavelet decomposition coefficients at different levels. Through wavelet decomposition, high-frequency information and low-frequency information in the transformed signal can be obtained. By calculating an appropriate coefficient threshold to screen the wavelet coefficients and setting the wavelet decomposition coefficients smaller than the noise threshold to 0 while keeping other signal values unchanged, the communication signal can be denoised.

[0075] Further, the coefficient threshold of the communication signal is calculated using the following formula: Where, represents the coefficient threshold, represents the signal length of the communication signal, represents the maximum decomposition level of wavelet transform, represents a preset heuristic parameter, represents the median of the absolute values of the wavelet coefficients, represents the wavelet coefficient.

[0076] Specifically, wavelet reconstruction is the inverse process of wavelet transform. In wavelet transform, a communication signal is decomposed into approximation and detail coefficients of different scales. Wavelet reconstruction can reconstruct a signal from the target coefficients for signal recovery, recombine the decomposed signal into the original signal to obtain a 5G message signal, and send the multimedia resource set to the target user through the 5G message signal.

[0077] In the embodiments of the present invention, by compiling the 5G message signal of the multimedia resource set, the signal anti-interference ability and transmission efficiency of the 5G message signal to be sent can be improved. Using the multimedia information push of the 5G message signal, the multimedia resource set is sent to the client corresponding to the target user, so as to accurately push the multimedia resource set to the target user, effectively improve the quality of information push, and further improve the accuracy of multimedia information push.

[0078] As Figure 4 shown, it is a functional module diagram of a multimedia information fusion push system based on 5G messages provided by an embodiment of the present invention.

[0079] The multimedia information fusion push system 400 based on 5G messages according to the present invention can be installed in an electronic device. According to the functions to be realized, the multimedia information fusion push system 400 based on 5G messages may include a sequence expansion module 401, a feature information fusion module 402, a multimedia resource set calculation module 403, a target feature construction module 404, and an information push module 405. The modules in the present invention can also be referred to as units, which refer to a series of computer program segments that can be executed by a processor of an electronic device and can complete fixed functions, and are stored in the memory of the electronic device.

[0080] In this embodiment, the functions of each module / unit are as follows: The sequence expansion module 401 is used to obtain the user behavior sequence of the target user and perform sequence expansion on the user behavior sequence to obtain a target behavior sequence; The feature information fusion module 402 is used to extract the local behavior feature and the global behavior feature of the target behavior sequence, and perform information fusion on the local behavior feature and the global behavior feature to obtain a user behavior feature; The multimedia resource set calculation module 403 is used to calculate the multimedia interest resource set of the target user according to the user behavior feature, and perform resource filtering on the interest resource set to obtain a multimedia resource set; The target feature construction module 404 is used to extract the hierarchical feature of each resource in the multimedia resource set, and construct the target feature of the multimedia resource set according to the hierarchical feature; The information push module 405 is configured to compile a 5G message signal according to the target feature and push multimedia information according to the 5G message signal.

[0081] Specifically, each module in the multimedia information fusion push system 400 based on 5G messages described in the embodiments of the present invention adopts the same technical means as those in the Figures 1 to 3 multimedia information fusion push method based on 5G messages described above and can produce the same technical effects, which will not be elaborated here.

[0082] The present invention also provides an electronic device, which may include a processor, a memory, a communication bus, and a communication interface, and may also include a computer program stored in the memory and executable on the processor, such as a multimedia information fusion push method program based on 5G messages.

[0083] Among them, in some embodiments, the processor may be composed of integrated circuits. For example, it may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits with the same or different functions, including a combination of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips.

[0084] The memory includes at least one type of readable storage medium, which includes flash memory, mobile hard disks, multimedia cards, card-type memories (such as SD or DX memories, etc.), magnetic memories, magnetic disks, optical disks, etc. In some embodiments, the memory may be an internal storage unit of the electronic device, such as the mobile hard disk of the electronic device.

[0085] The communication bus may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. The bus is configured to realize the connection and communication between the memory and at least one processor, etc.

[0086] The communication interface is used for communication between the above-mentioned electronic device and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), and is generally used to establish a communication connection between this electronic device and other electronic devices. The user interface may be a display, an input unit (such as a keyboard), and optionally, the user interface may also be a standard wired interface or a wireless interface.

[0087] Only the electronic device with components is shown in the figure. Those skilled in the art can understand that the structure shown in the figure does not constitute a limitation on the electronic device, and it may include fewer or more components than shown in the figure, or combine certain components, or have different component arrangements.

[0088] Specifically, for the specific implementation method of the above instructions by the processor, reference can be made to the description of the relevant steps in the corresponding embodiments of the accompanying drawings, which will not be elaborated here.

[0089] In several embodiments provided by the present invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0090] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-mentioned exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms.

[0091] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any associated drawing marks in the claims should not be regarded as limiting the claims involved.

[0092] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is a theory, method, technology, and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results.

[0093] In addition, it is obvious that the term "comprising" does not exclude other units or steps, and the singular form does not exclude the plural form. A plurality of units or systems stated in the system claims can also be implemented by one unit or system through software or hardware. Terms such as first, second, etc. are used to denote names and do not denote any particular order.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A multimedia information fusion and push method based on 5G messaging, characterized in that, The method includes: Obtain the user behavior sequence of the target user, perform sequence expansion on the user behavior sequence to obtain a target behavior sequence; Extract the local behavior features and global behavior features of the target behavior sequence, perform information fusion on the local behavior features and the global behavior features to obtain user behavior features; Calculate the multimedia interest resource set of the target user according to the user behavior features, perform resource filtering on the interest resource set to obtain a multimedia resource set; Extract the grading features of each resource in the multimedia resource set, and construct the target features of the multimedia resource set according to the grading features; Compile a 5G message signal according to the target features, and perform multimedia information push according to the 5G message signal.

2. The multimedia information fusion push method based on 5G messages according to claim 1, wherein, The performing sequence expansion on the user behavior sequence to obtain a target behavior sequence includes: Perform reverse sorting on the user behavior sequence to obtain a reverse behavior sequence; Construct a sequence embedding matrix of the reverse behavior sequence, and calculate the behavior preference score of each user behavior in the user behavior sequence according to the sequence embedding matrix; Select the user behavior with the largest behavior preference score and add it to the user behavior sequence to obtain an updated behavior sequence; Iterate the updated behavior sequence as the user behavior sequence until the number of iterations is greater than a preset number threshold to obtain a target behavior sequence.

3. The multimedia information fusion push method based on 5G messages according to claim 1, characterized in that The extracting the local behavior features and global behavior features of the target behavior sequence includes: Construct a behavior node graph according to the target behavior sequence, perform self-attention calculation on the behavior node graph to obtain local behavior features; Identify the neighbor nodes of each behavior node in the behavior node graph, and calculate the attention coefficients of the neighbor nodes; Perform feature aggregation on the neighbor nodes according to the attention coefficients to obtain node aggregation features; Perform feature update on the behavior nodes according to the node aggregation features to obtain global behavior features.

4. The multimedia information fusion push method based on 5G messages according to claim 1, characterized in that, The performing information fusion on the local behavior features and the global behavior features to obtain user behavior features includes: Perform feature splicing on the local behavior features to obtain local fusion features; Perform feature weighted average on the global behavior features to obtain global fusion features; Perform sum pooling processing on the local fusion features and the global fusion features to obtain user behavior features.

5. The multimedia information fusion push method based on 5G messages according to claim 1, wherein The calculating the multimedia interest resource set of the target user according to the user behavior features includes: Calculate the user similarity between the target user and the users in a preset user set according to the user behavior features; Construct a similar user set according to the user similarity, and obtain the interest resource scores corresponding to the similar user set; Calculate the user resource score of the target user according to the user similarity and the interest resource scores; Determine the multimedia interest resource set of the target user according to the user resource score.

6. The multimedia information fusion push method based on 5G messages according to claim 1, wherein The extracting the grading features of each resource in the multimedia resource set includes: Perform vector conversion and type division on the multimedia resource set to obtain text resource vectors and other resource vectors; Perform multi-layer encoding on the text resource vector to obtain multi-layer encoded features; Perform mean pooling on the multi-layer encoded features to obtain hierarchical pooling features, and perform feature concatenation on the hierarchical pooling features to obtain text hierarchical features; Perform multi-layer convolutional pooling on the other resource vectors to obtain pooling features, and perform fully connected processing on the pooling features to obtain other hierarchical features; Aggregate the text hierarchical features and the other hierarchical features to obtain the hierarchical features of the multimedia resource set.

7. The multimedia information fusion push method based on 5G messages according to claim 1, characterized in that Constructing the target features of the multimedia resource set according to the hierarchical features includes: Construct a feature vector sequence according to the hierarchical features, and perform attention calculation on the feature vector sequence to obtain attention features; Perform residual connection on the attention features and the target features to obtain connection features; Perform regularization processing on the connection features to obtain regularization features, and perform fully connected feed-forward calculation on the regularization features to obtain feed-forward features; Perform residual connection on the feed-forward features and the regularization features to obtain initial target features; Perform regularization processing on the initial target features to obtain the target features of the multimedia resource set.

8. The multimedia information fusion push method based on 5G messages according to claim 1, wherein, Compiling a 5G message signal according to the target features: Perform channel encoding on the target features to obtain encoded information; Perform signal modulation on the encoded information to obtain a communication signal of the encoded information; Perform wavelet transform on the communication signal to obtain wavelet coefficients; Calculate the coefficient threshold of the communication signal according to the wavelet coefficients, and extract the target coefficients of the wavelet coefficients according to the coefficient threshold; Perform wavelet reconstruction on the target coefficients to obtain a 5G message signal.

9. The multimedia information fusion push method based on 5G messages according to claim 8, wherein Performing signal modulation on the encoded information to obtain a communication signal of the encoded information includes: Determine the transmission distance and channel quality index corresponding to the encoded information; When the transmission distance is greater than a preset communication threshold, perform constant envelope data modulation processing on the encoded information to obtain a constant envelope modulation signal; Perform delay compensation processing on the constant envelope modulation signal according to the channel quality index to obtain a communication signal.

10. A multimedia information fusion push system based on 5G messaging, characterized in that, The system includes: A sequence expansion module, configured to obtain a user behavior sequence of a target user, and perform sequence expansion on the user behavior sequence to obtain a target behavior sequence; A feature information fusion module, configured to extract local behavior features and global behavior features of the target behavior sequence, and perform information fusion on the local behavior features and the global behavior features to obtain user behavior features; A multimedia resource set calculation module, configured to calculate a multimedia interest resource set of the target user according to the user behavior features, and perform resource filtering on the interest resource set to obtain a multimedia resource set; A target feature construction module, configured to extract hierarchical features of each resource in the multimedia resource set, and construct target features of the multimedia resource set according to the hierarchical features; An information push module, configured to compile a 5G message signal according to the target features, and perform multimedia information push according to the 5G message signal.

Images