A method and device for recommending a message delivery group

Abstract

The application discloses a message putting group recommendation method and device, constructs a heterogeneous graph attention network, extracts the fusion feature information of the terminal user according to the heterogeneous graph attention network, obtains the message content of the message to be put, extracts the features of the message content, obtains the message feature information, inputs the fusion feature information and the message feature information into the putting group recommendation model, obtains the to-be-put object of the message to be put after the putting group recommendation model processing, and recommends and puts the message to be put according to the to-be-put object. The application extracts the fusion feature information and the message feature information of the terminal user by constructing the heterogeneous graph attention network, obtains the to-be-put object of the message to be put after the putting group recommendation model processing, improves the accuracy of 5G industry message putting, and maximally reduces the harassment to the terminal user.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and more specifically to a method and apparatus for recommending message delivery groups. Background Technology

[0002] 5G messaging service, based on the user terminal's native SMS entry point, provides users with the ability to send and receive media content such as text, images, audio, video, location, and contacts, including point-to-point messaging, group messaging, group chat messaging, and inter-application messaging. Compared to the limited functionality of traditional SMS, 5G messaging not only broadens the scope of information sending and receiving, supporting users to use multimedia content such as text, audio, video, cards, and location, but also extends the depth of the interactive experience. Users can complete services such as service search, discovery, interaction, and payment within the message window, creating a one-stop information window.

[0003] 5G messaging provides enhanced messaging services between individuals and applications for industry customers, realizing Messaging as a Platform (MaaP). It also introduces a new messaging interaction mode: chatbots. Users can intuitively and conveniently enjoy various 5G application services such as bill payment, ticketing, hotel booking, logistics tracking, restaurant reservations, and food delivery through the 5G messaging window. Chatbots are a dialogue-based service provided by industry customers to end users. This service is typically based on artificial intelligence software, simulating human intelligent dialogue to provide users with specific service functions.

[0004] The 5G messaging system includes a 5G Messaging Center (5GMC), a MaaP system (including a MaaP platform management module and the MaaP platform), and group chat servers, among other equipment. The 5G Messaging Center is the core network element of the 5G messaging system. It has access and routing modules and functions, is deployed as part of the overall Virtual Network Architecture (VNF), and also possesses the processing capabilities and external interfaces of a short message center. This network element will uniformly provide processing, sending, storage, and forwarding functions for short messages and basic multimedia messages. The MaaP system is the core network element of the industry 5G messaging system. This network element will provide industry users with access to 5G commercial messaging services and message uplink and downlink capabilities, and provide users with functions such as industry chatbot search, detail query, and message uplink and downlink. The group chat server provides group chat functionality for 5G messaging, including group chat message sending and receiving, and group information management.

[0005] The 5G messaging application open platform can help industry customers achieve A2P communication across multiple scenarios on demand. Enterprises can quickly deploy messaging applications through the platform without complex code development, helping them easily create their own 5G messaging applications. However, existing 5G messaging target audience recommendation schemes mainly rely on big data tags of end users, that is, targeting users with tags related to message content. This scheme is somewhat intrusive to users and its targeting accuracy is still insufficient. Summary of the Invention

[0006] In view of the above problems, the present invention is proposed to provide a method and apparatus for recommending message delivery groups that overcomes or at least partially solves the above problems.

[0007] According to one aspect of the present invention, a method for recommending message delivery groups is provided, comprising:

[0008] Construct a heterogeneous graph attention network and extract fusion feature information of end users based on the heterogeneous graph attention network;

[0009] Obtain the message content of the message to be delivered, and extract features from the message content to obtain message feature information;

[0010] The fused feature information and the message feature information are input into the target audience recommendation model. After processing by the target audience recommendation model, the target audience for the message to be delivered is obtained.

[0011] The message to be delivered is recommended and delivered based on the target audience.

[0012] According to another aspect of the present invention, a recommendation device for message delivery groups is provided, comprising:

[0013] The network building module is used to build heterogeneous graph attention networks;

[0014] The fusion feature information extraction module is used to extract the fusion feature information of the end user based on the heterogeneous graph attention network;

[0015] The message feature extraction module is used to obtain the message content of the message to be delivered, extract features from the message content, and obtain message feature information.

[0016] The processing module is used to input the fused feature information and the message feature information into the delivery group recommendation model, and obtain the delivery target of the message to be delivered after processing by the delivery group recommendation model;

[0017] The recommendation module is used to recommend and deliver the message to be delivered based on the target audience.

[0018] According to another aspect of the present invention, a computing device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus;

[0019] The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the above-described message delivery group recommendation method.

[0020] According to another aspect of the present invention, a computer storage medium is provided, the storage medium storing at least one executable instruction that causes a processor to perform an operation corresponding to the above-described recommendation method for a message delivery group.

[0021] According to a method and apparatus for recommending message delivery groups based on the present invention, a heterogeneous graph attention network is constructed, and fusion feature information of terminal users is extracted based on the heterogeneous graph attention network; the message content of the message to be delivered is obtained, and feature extraction is performed on the message content to obtain message feature information; the fusion feature information and message feature information are input into a delivery group recommendation model, and the delivery group recommendation model processes the information to obtain the delivery targets of the message to be delivered; and the message to be delivered is recommended and delivered based on the delivery targets. This invention improves the accuracy of message delivery in the 5G industry by constructing a heterogeneous graph attention network, extracting fusion feature information and message feature information of terminal users, and processing the information through a delivery group recommendation model to obtain the delivery targets of the message to be delivered, thereby minimizing harassment to terminal users.

[0022] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0023] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0024] Figure 1 A flowchart of a message delivery group recommendation method provided by an embodiment of the present invention is shown;

[0025] Figure 2 A schematic diagram of a heterogeneous graph network provided in an embodiment of the present invention is shown;

[0026] Figure 3This illustration shows a schematic diagram of the implementation process of a message delivery group recommendation method provided by an embodiment of the present invention;

[0027] Figure 4 This diagram illustrates the structure of a recommendation device for message delivery groups provided in an embodiment of the present invention.

[0028] Figure 5 A schematic diagram of the structure of a computing device provided in an embodiment of the present invention is shown. Detailed Implementation

[0029] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0030] Graph Networks (GN) are sets of functions organized in a graph structure within a topological space for relational reasoning. A GN consists of interconnected GN blocks, also known as nodes in neural network implementations. Connections between nodes are called "edges," representing dependencies between them. The properties of nodes and edges in a GN are the same as in a graph structure. Graph Attention Networks (GAT) introduce an attention mechanism into GNs, assigning different weights to different neighboring nodes during the aggregation of features from the central node. This results in differentiated attention to neighboring nodes, focusing on more correlated slice nodes while ignoring less correlated ones. Traditional Graph Convolutional Networks (GCNs) and other GNs use static, non-adaptive propagation rules, failing to capture which neighboring node contributes more to the central node's classification. Furthermore, in real-world data, not all edges represent the same level of correlation.

[0031] In this embodiment, the graph refers to a relationship graph composed of three types of heterogeneous data in the user network of the messaging terminal, which can be represented as G = (V, E). The nodes in the graph represent the three types of heterogeneous data, E is the set of edges, and the edges represent four relationships among the three types of heterogeneous data. The node feature set is represented by h, and the feature hi of each vertex is a high-dimensional vector. Therefore, the scheme in this embodiment can be regarded as a node classification problem, and the final output is whether the terminal user node is the target of the chatbot message in this industry.

[0032] Figure 1 A flowchart illustrating an embodiment of a message delivery group recommendation method according to the present invention is shown, as follows: Figure 1 As shown, the method includes the following steps:

[0033] Step S110: Construct a heterogeneous graph attention network and extract the fusion feature information of the end user based on the heterogeneous graph attention network.

[0034] Figure 2 This is a schematic diagram of a heterogeneous graph network provided in an embodiment of the present invention, such as... Figure 2 As shown, a heterogeneous graph network composed of three types of heterogeneous nodes is constructed. The node types in the graph include three categories: big data tag nodes, whose characteristics include big data tag name and tag type; end-user nodes, whose characteristics include end-user status information such as account opening location, online time, current registration location, and average revenue per user (ARPU); and industry chatbot nodes, whose characteristics include industry status information such as industry affiliation, average daily message volume, and number of followed users. The relationships between the three types of nodes are of two types: first, the relationship between big data tags and end-users is that the end-user belongs to one or more big data tags; second, the relationship between end-users and industry chatbots is that the end-user follows one or more industry chatbots.

[0035] In this embodiment, the calculation of GAT is divided into two steps: calculating the attention coefficient and weighted summation. The GAT model is implemented by stacking graph attention layers. The input of each graph attention layer is the set of alarm text features generated by the node. The input is the set of node features, and the formula is as follows (1):

[0036]

[0037] The output is a new set of alarm text node features as shown in equation (2):

[0038]

[0039] To calculate the weight of each neighbor node, an F×F' shared weight matrix W is applied to each node to calculate the attention coefficient, which can represent the importance of node j relative to node i. The attention coefficient is calculated as follows (3):

[0040]

[0041] Where a and W are parameters.

[0042] To make the attention coefficient easier to calculate and compare, the logistic regression model softmax is used to regularize all neighboring nodes j of node i, as shown in the following formula (4):

[0043]

[0044] The feature representation of node i after passing through the graph attention layer is obtained by processing by equations (1)-(4), which is the new feature output by GAT for each vertex i. This new feature integrates neighborhood (including neighboring nodes) information, where σ is the activation function.

[0045]

[0046] Where, N i Let represent the set of neighboring nodes of node i, and let coefficient 'a' be the coefficient used for weighted summation during each convolution.

[0047] In an optional approach, step S110 further includes: collecting big data tag feature information, industry chatbot status information, and terminal user status information from the 5G messaging open platform according to a preset time granularity; obtaining big data tag feature data, industry chatbot feature data, and terminal user feature data respectively based on the big data tag feature information, industry chatbot status information, and terminal user status information; and constructing a heterogeneous graph attention network with the terminal user as the central node based on the big data tag feature data, industry chatbot feature data, and terminal user feature data.

[0048] In this embodiment, big data tag feature information with time granularity of T, industry chatbot status information, and terminal user status information are collected from the 5G messaging open platform. Feature data of three types of heterogeneous nodes are extracted, and a heterogeneous graph attention network composed of the three types of heterogeneous nodes is constructed with the terminal user as the central node based on the relationship between the three types of nodes.

[0049] In an optional approach, step S110 further includes: inputting big data label feature data and end-user feature data into the neighbor node-level graph attention layer, and outputting a first end-user feature fused with the big data label feature data; inputting industry chatbot feature data and end-user feature data into the neighbor node-level graph attention layer, and outputting a second end-user feature fused with the industry chatbot feature data; and inputting the first end-user feature and the second end-user feature into the central node-level graph attention layer for aggregation processing to obtain the fused feature information of the end-user.

[0050] Figure 3 This is a schematic diagram illustrating the implementation process of an embodiment of the present invention, such as... Figure 3 As shown, the fused feature information of the end user is first extracted by the end user feature extractor. The big data label feature data and the end user feature data are then input into the neighbor node-level graph attention layer. The first step is to learn the weights of the importance of the first type of neighbor node "big data label" of end user i. Specifically, this is achieved by inputting the state feature h of end user i. ci The attribute characteristics h of the big data tags of its neighboring nodes bi After being transformed into feature vectors through word embedding layers, the vectors are input into the neighbor node-level graph attention layer, and the output is the first terminal user feature h, which is fused with big data label feature data. ci 1 The second step involves learning the weights of the importance of the second type of neighboring nodes "industry chatbots" of end user i, by inputting the state features h of end user i. ci Features of industry chatbots and their neighboring nodes h bi After being transformed into feature vectors through word embedding layers, the vectors are input into the neighbor node-level graph attention layer, and the output is the second end-user feature h, which is fused with industry chatbot feature data. ci 2 ;The first terminal user characteristics h ci 1 Second terminal user characteristics h ci 2 The input center node-level graph attention layer performs aggregation processing, and the output is the fused feature information h of the terminal user that simultaneously fuses the features of two types of neighbor nodes. ci ′.

[0051] It should be noted that, as Figure 3As shown, the word embedding layer converts each word into a vector. The input data dimension is z, and the output is set to convert the words into 64-dimensional spatial vectors. The input sequence length is F, so the shape of the output data of the word embedding layer is (None, F, 64). The function of the word embedding layer is to perform vector mapping on the input words, converting the index of each word into a 64-dimensional fixed-shape vector. The neighbor node-level graph attention layer has 256 convolutional kernels and the activation function is set to "relu". The center node-level graph attention layer has 128 convolutional kernels and the activation function is set to "relu".

[0052] Step S120: Obtain the message content of the message to be delivered, extract features from the message content, and obtain message feature information.

[0053] In an optional approach, step S120 further includes: obtaining a message delivery request; and extracting the message content of the message to be delivered based on the message delivery request.

[0054] In an optional approach, step S120 further includes: performing word segmentation on the message content, mapping each word to a vector through a word embedding layer to obtain a mapping vector for each word; inputting the mapping vector of each word into a convolutional layer for text feature extraction to obtain multiple text feature information; processing the multiple text feature information through a max pooling layer to extract the text feature information with the largest feature value as the message feature information.

[0055] In this step, the industry customer initiates a message delivery request to the 5G Messaging Open Platform. This request carries the message content to be delivered. The system obtains the message content, extracts its features, and obtains message feature information. For example... Figure 3As shown, the message content can be feature extracted using a message feature extractor. Specifically, the message content is segmented, and each word is converted into a mapping vector using the word embedding function of the first layer, i.e., the word embedding layer. The input data dimension is j, and the output is set to convert the words into 64-dimensional spatial vectors. The input sequence length is G, so the shape of the output data of the word embedding layer is (None, G, 64). Further, text features are extracted through the second and fourth convolutional layers (Conv1D) to obtain multiple text feature information. The number of convolutional kernels is 48 (i.e., the output dimension), the spatial window length of the convolutional kernels is set to 2 (i.e., the convolutional kernel reads 2 words consecutively each time), and the activation function is set to "ReLU". The multiple text feature information is then processed through the third and fifth max pooling layers (MaxPooling1D) to extract the text feature information with the largest feature value as the message feature information. The pooling window size is set to 2, and the max pooling layer retains the maximum value among the feature values ​​extracted by the convolutional kernels, discarding all other feature values. Furthermore, the input can be "flattened" using a sixth-layer flattening layer, transforming the three-dimensional input into a two-dimensional one. The flattening layer is used for the transition from the convolutional layer to the subsequent fully connected layer.

[0056] Step S130: Input the fused feature information and message feature information into the target audience recommendation model. After processing by the target audience recommendation model, the target audience for the message to be delivered is obtained.

[0057] In one alternative approach, the training steps of the group recommendation model include: extracting fusion feature information of end-user samples based on a heterogeneous graph attention network, and collecting historical delivery messages and the corresponding delivery object labels; extracting features from the historical delivery messages to obtain historical message feature information; and training the model using the fusion feature information of end-user samples, the historical message feature information, and the corresponding delivery object labels to obtain the group recommendation model.

[0058] Specifically, the heterogeneous graph attention network is used to extract the fusion feature information of end-user samples and collect historical delivery messages and the corresponding delivery object labels. Features are extracted from the historical delivery messages to obtain historical message feature information, which is then encoded into an integer sequence. At the same time, the recommended delivery objects are manually labeled according to the message content of each message to be delivered.

[0059] Specifically, the heterogeneous graph attention network includes a large data tag feature dataset, an industry chatbot feature dataset, an end-user feature dataset, and a set of messages to be delivered. The encoding sequence length of each node feature is defined as F (F is also the longest length in the dataset), with each data entry padded to the length of F, and its dictionary size is z. The encoding sequence length of the message to be delivered is defined as G (G is the longest length in the dataset), with each data entry padded to the length of G, and its dictionary size is j. The tag matrix Y represents manually labeled whether the end-user is the target audience for the industry chatbot message, and has an N*1 shape. A total dataset is constructed using the fused feature information of end-user samples, historical message feature information, and the corresponding target audience tags of historical delivered messages. This total dataset is divided into training data and test data, with 80% of the dataset used as training data and the remaining 20% ​​as test data. The training set is used to train the target audience recommendation model, and the test set is used to evaluate and validate the model.

[0060] In an optional approach, step S130 further includes: concatenating and processing the fused feature information and message feature information through a merging layer to obtain a context vector of a preset length; processing the context vector through a fully connected layer to obtain a prediction result of the target audience for the message to be delivered; and obtaining the target audience for the message to be delivered based on the prediction result of the target audience.

[0061] Specifically, the fused feature vector of the end user output by the end user feature extractor and the message feature vector output by the message feature extractor are concatenated by a concatenation layer along the column dimension to form a fixed-length context vector. This context vector is then input into a fully connected layer (Dense) to obtain the predicted target audience for the message. The fully connected layer has only one neuron, uses a sigmoid activation function, and outputs whether the end user is a target audience for the chatbot message in that industry. i (where 1 represents yes and 0 represents no), the target audience for the message to be delivered is obtained by organizing the prediction results of the target audience of multiple terminal users.

[0062] It should be noted that, in order to improve the accuracy of the prediction results during the training of the group recommendation model, for each terminal user, it is also necessary to calculate the error between the prediction result and the actual result for that terminal user i. The training objective is to minimize this error, and the objective function can be the binary logarithmic loss function (binary_crossentropy) as shown in equation (6):

[0063]

[0064] The training epochs can be set to 1000 (epochs = 1000). The gradient descent optimization algorithm uses the Adam optimizer to improve the learning speed of traditional gradient descent (optimizer = 'adam'). The convolutional neural network can find the optimal weight values ​​that minimize the objective function through gradient descent. The convolutional neural network will learn the weight values ​​autonomously through training. The training set is used to train the model so that the objective function is as small as possible. After each training round, the test set is used to evaluate and verify the group recommendation model. After the group recommendation model converges, the weights of the group recommendation model are derived.

[0065] Step S140: Recommend and deliver the message to the target audience based on the target audience.

[0066] In this step, the target audience recommendation model ultimately outputs whether the end user is a candidate for receiving chatbot messages in the industry, and returns the result to the 5G Messaging Open Platform. The 5G Messaging Open Platform then feeds back the target audience to the industry customer and recommends messages to be delivered according to the target audience.

[0067] The method in this embodiment constructs a heterogeneous graph attention network (HGIFN) centered on the end user, consisting of end users, industry chatbots, and big data tags. This HGIFN is combined with a convolutional neural network. The HGIFN comprises neighbor node-level graph attention and center node-level graph attention. With the end user as the center node, neighbor node-level attention aggregation is performed with two adjacent types of neighbor nodes. The method learns the different importance of the neighbor nodes (i.e., industry chatbots and big data tags) in predicting whether the user is a target for industry message delivery. Different weights are assigned to the relationships between nodes based on the difference in importance. Then, the first end user feature, which incorporates big data tag feature information, and the second end user feature, which incorporates industry chatbot state information, are aggregated. The output fused feature information of the end user and the message feature information of the message to be delivered are merged through a merging layer and input into a classifier composed of fully connected layers. Finally, the method outputs whether the end user is a target for the industry chatbot message, thereby improving the accuracy of 5G industry message delivery and minimizing harassment to end users.

[0068] Figure 4 A schematic diagram of an embodiment of a message delivery group recommendation device according to the present invention is shown. Figure 4 As shown, the device includes: a network construction module 410, a fusion feature information extraction module 420, a message feature extraction module 430, a processing module 440, and a recommendation module 450.

[0069] Network building module 410 is used to build heterogeneous graph attention networks.

[0070] In an alternative approach, the network construction module 410 is further configured to: collect big data tag feature information, industry chatbot status information, and terminal user status information from the 5G messaging open platform according to a preset time granularity; obtain big data tag feature data, industry chatbot feature data, and terminal user feature data respectively based on the big data tag feature information, industry chatbot status information, and terminal user status information; and construct a heterogeneous graph attention network with the terminal user as the central node based on the big data tag feature data, industry chatbot feature data, and terminal user feature data.

[0071] The fusion feature information extraction module 420 is used to extract the fusion feature information of the end user based on the heterogeneous graph attention network.

[0072] In an optional manner, the feature extraction module 420 is further configured to: input big data label feature data and end-user feature data into the neighbor node-level graph attention layer, and output the first end-user feature after fusing the big data label feature data; input industry chatbot feature data and end-user feature data into the neighbor node-level graph attention layer, and output the second end-user feature after fusing the industry chatbot feature data; input the first end-user feature and the second end-user feature into the central node-level graph attention layer for aggregation processing to obtain the fused feature information of the end-user.

[0073] The message feature extraction module 430 is used to obtain the message content of the message to be delivered, extract features from the message content, and obtain message feature information.

[0074] In an alternative embodiment, the message feature extraction module 430 is further configured to: obtain a message delivery request; and extract the message content of the message to be delivered based on the message delivery request.

[0075] In an optional manner, the message feature extraction module 430 is further configured to: perform word segmentation on the message content, map each word segmentation to a vector through a word embedding layer to obtain a mapping vector for each word; input the mapping vector of each word into a convolutional layer for text feature extraction to obtain multiple text feature information; process the multiple text feature information through a max pooling layer to extract the text feature information with the largest feature value as the message feature information.

[0076] The processing module 440 is used to input the fused feature information and message feature information into the target audience recommendation model, and after processing by the target audience recommendation model, the target audience for the message to be delivered is obtained.

[0077] In an optional manner, the message feature extraction module 430 is further configured to: combine and process the fused feature information and message feature information through a merging layer to obtain a context vector of a preset length; process the context vector through a fully connected layer to obtain the target audience prediction result of the message to be delivered; and obtain the target audience of the message to be delivered based on the target audience prediction result.

[0078] In one alternative approach, the device further includes a recommendation model building module, which is used to extract fusion feature information of terminal user samples based on a heterogeneous graph attention network, and collect historical delivery messages and the corresponding delivery object labels; extract features from the historical delivery messages to obtain historical message feature information; and train the model using the fusion feature information of terminal user samples, the historical message feature information, and the corresponding delivery object labels of the historical delivery messages to obtain a delivery group recommendation model.

[0079] The recommendation module 450 is used to recommend and deliver messages to the target audience based on their preferences.

[0080] The device in this embodiment constructs a heterogeneous graph attention network and extracts fusion feature information of terminal users based on the heterogeneous graph attention network; it obtains the message content of the message to be delivered, extracts features from the message content to obtain message feature information; it inputs the fusion feature information and message feature information into a delivery group recommendation model, and after processing by the delivery group recommendation model, it obtains the target audience for the message to be delivered; and it recommends and delivers the message to the target audience based on the target audience. This device, by constructing a heterogeneous graph attention network, extracting fusion feature information and message feature information of terminal users, and processing them through a delivery group recommendation model to obtain the target audience for the message to be delivered, improves the accuracy of 5G industry message delivery and minimizes harassment to terminal users.

[0081] This invention provides a non-volatile computer storage medium storing at least one executable instruction that can execute a recommendation method for message delivery groups in any of the above method embodiments.

[0082] Executable instructions can specifically be used to cause the processor to perform the following operations:

[0083] Construct a heterogeneous graph attention network and extract fusion feature information of end users based on the heterogeneous graph attention network;

[0084] Obtain the message content of the message to be delivered, extract features from the message content, and obtain message feature information;

[0085] The fusion feature information and message feature information are input into the target audience recommendation model. After processing by the target audience recommendation model, the target audience for the message to be delivered is obtained.

[0086] Recommend and deliver messages to the target audience based on their preferences.

[0087] Figure 5 The diagram shows a structural schematic of an embodiment of the computing device of the present invention. The specific embodiments of the present invention do not limit the specific implementation of the computing device.

[0088] like Figure 5 As shown, the computing device may include:

[0089] Processor, Communications Interface, Memory, and Communications Bus.

[0090] The processor, communication interface, and memory communicate with each other via a communication bus. The communication interface is used to communicate with other network elements, such as clients or other servers. The processor executes programs, specifically the steps described in the recommended method embodiment for message delivery to groups.

[0091] Specifically, the program may include program code, which includes computer operation instructions.

[0092] The processor may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The server may include one or more processors of the same type, such as one or more CPUs; or it may include processors of different types, such as one or more CPUs and one or more ASICs.

[0093] Memory is used to store programs. Memory may include high-speed RAM, and may also include non-volatile memory, such as at least one disk drive.

[0094] Specifically, the program can be used to cause the processor to perform the following operations:

[0095] Construct a heterogeneous graph attention network and extract fusion feature information of end users based on the heterogeneous graph attention network;

[0096] Obtain the message content of the message to be delivered, extract features from the message content, and obtain message feature information;

[0097] The fusion feature information and message feature information are input into the target audience recommendation model. After processing by the target audience recommendation model, the target audience for the message to be delivered is obtained.

[0098] Recommend and deliver messages to the target audience based on their preferences.

[0099] The algorithms or displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, the embodiments of the present invention are not directed to any particular programming language. It should be understood that the content of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention.

[0100] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0101] Similarly, it should be understood that, in order to simplify the invention and aid in understanding one or more of the various inventive aspects, features of the embodiments of the invention are sometimes grouped together in a single embodiment, figure, or description thereof in the above description of exemplary embodiments of the invention. However, this disclosure should not be construed as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as reflected in the following claims, inventive aspects lie in fewer than all features of a single foregoing disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into this detailed description, wherein each claim itself is a separate embodiment of the invention.

[0102] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or device so disclosed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0103] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments can be used in any combination.

[0104] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components according to the embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0105] It should be noted that the above embodiments are illustrative of the invention and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be construed as limiting the order of execution.

Claims

1. A method for recommending a message delivery group, characterized by, The method comprises the steps of: constructing a heterogeneous graph attention network to extract fusion feature information of terminal users according to the heterogeneous graph attention network; obtaining message content of a to-be-launched message, performing feature extraction on the message content to obtain message feature information; inputting the fusion feature information and the message feature information into a launch group recommendation model, and obtaining a to-be-launched object of the to-be-launched message after processing by the launch group recommendation model; recommending and launching the to-be-launched message according to the to-be-launched object; wherein the step of constructing the heterogeneous graph attention network further comprises: collecting big data label feature information, industry chat robot state information and terminal user state information from a 5G message open platform according to a preset time granularity; obtaining big data label feature data, industry chat robot feature data and terminal user feature data according to the big data label feature information, the industry chat robot state information and the terminal user state information respectively; constructing a heterogeneous graph attention network with a terminal user as a center node according to the big data label feature data, the industry chat robot feature data and the terminal user feature data; the step of extracting fusion feature information of terminal users according to the heterogeneous graph attention network further comprises: inputting the big data label feature data and the terminal user feature data into a neighbor node level graph attention layer to output first terminal user features fused with the big data label feature data; inputting the industry chat robot feature data and the terminal user feature data into a neighbor node level graph attention layer to output second terminal user features fused with the industry chat robot feature data; inputting the first terminal user features and the second terminal user features into a center node level graph attention layer for aggregation processing to obtain fusion feature information of the terminal user.

2. The method of claim 1, wherein, The training steps of the launch group recommendation model comprise: extracting fusion feature information of terminal user samples according to the heterogeneous graph attention network, and collecting historical launch messages and launch object labels corresponding to the historical launch messages; performing feature extraction on the historical launch messages to obtain historical message feature information; performing model training by using the fusion feature information of the terminal user samples, the historical message feature information and the launch object labels corresponding to the historical launch messages to obtain a launch group recommendation model.

3. The method of claim 1, wherein, The step of obtaining message content of a to-be-launched message further comprises: obtaining a message delivery request; extracting message content of a to-be-launched message according to the message delivery request.

4. The method according to any one of claims 1 to 3, characterized in that, The step of performing feature extraction on the message content to obtain message feature information further comprises: performing word segmentation processing on the message content, mapping each word segmentation to a vector through a word embedding layer to obtain a mapping vector of each word; inputting the mapping vector of each word into a convolution layer to perform text feature extraction, and obtaining a plurality of text feature information; processing the plurality of text feature information through a maximum value pooling layer to extract text feature information with the maximum feature value as the message feature information.

5. The method of claim 1, wherein, The step of inputting the fusion feature information and the message feature information into the launch group recommendation model, and obtaining a to-be-launched object of the to-be-launched message after processing by the launch group recommendation model further comprises: The fusion feature information and the message feature information are spliced and merged by a merging layer to obtain a context vector with a preset length; A prediction result of a delivery group of the to-be-delivered message is obtained by processing the context vector through a full connection layer; A to-be-delivered object of the to-be-delivered message is obtained according to the prediction result of the delivery group.

6. A message delivery group recommendation apparatus characterized by comprising: Comprise: A network construction module is configured to construct a heterogeneous graph attention network; A fusion feature information extraction module is configured to extract fusion feature information of a terminal user according to the heterogeneous graph attention network; A message feature extraction module is configured to obtain message content of a to-be-delivered message, and extract features of the message content to obtain message feature information; A processing module is configured to input the fusion feature information and the message feature information into a delivery group recommendation model, and obtain a to-be-delivered object of the to-be-delivered message after processing by the delivery group recommendation model; A recommendation module is configured to recommend and deliver the to-be-delivered message according to the to-be-delivered object; The network construction module is further configured to: Collect big data label feature information, industry chat robot state information, and terminal user state information from a 5G message open platform according to a preset time granularity; Obtain big data label feature data, industry chat robot feature data, and terminal user feature data according to the big data label feature information, the industry chat robot state information, and the terminal user state information, respectively; Construct a heterogeneous graph attention network with a terminal user as a center node according to the big data label feature data, the industry chat robot feature data, and the terminal user feature data; The fusion feature information extraction module is further configured to: Input the big data label feature data and the terminal user feature data into a neighbor node level graph attention layer to output first terminal user features fused with the big data label feature data; Input the industry chat robot feature data and the terminal user feature data into a neighbor node level graph attention layer to output second terminal user features fused with the industry chat robot feature data; Input the first terminal user features and the second terminal user features into a center node level graph attention layer for aggregation processing to obtain fusion feature information of the terminal user.

7. A computing device, comprising: Comprise: A processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface complete communication with each other through the communication bus; The memory is configured to store at least one executable instruction, and the executable instruction causes the processor to perform operations corresponding to a message delivery group recommendation method according to any one of claims 1-5.

8. A computer storage medium, characterized in that The storage medium stores at least one executable instruction, and the executable instruction causes the processor to perform operations corresponding to a message delivery group recommendation method according to any one of claims 1-5.

Images