Business demand prediction method and device, equipment, storage medium and program product

Abstract

The invention discloses a business demand prediction method and device, equipment, a storage medium and a program product, which are used for rapidly and accurately predicting business demands and enabling a demand prediction process to have interpretability. The business demand prediction method comprises the following steps: constructing a first feature relation graph of a business object based on an object feature set of the business object; performing graph attention operation on the first feature relation graph to obtain an attention coefficient between nodes and node features of the nodes in the first feature relation graph; based on the attention coefficient between the nodes, reconstructing the first feature relation graph to obtain a second feature relation graph of the business object; and predicting business demand information of the business object based on the node features of the nodes and the second feature relation graph.

Description

Technical Field

[0001] The present application relates to the field of deep learning technology, and in particular to a business demand prediction method, device, equipment, storage medium and program product. Background Art

[0002] In all walks of life, understanding business needs is an important basis for formulating service recommendation strategies, product design, and service provision. With the popularization of the Internet and the popularity of mobile devices, the data generated by business objects on the Internet is growing exponentially.

[0003] Therefore, how to quickly and accurately predict business needs from these massive amounts of data still requires further solutions. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide a business demand forecasting method, apparatus, device, storage medium and program product for quickly and accurately forecasting business demand and making the demand forecasting process explainable.

[0005] In order to achieve the above objectives, the present application embodiment adopts the following technical solutions:

[0006] In a first aspect, an embodiment of the present application provides a method for predicting business demand, including:

[0007] Based on the object feature set of the business object, construct a first feature relationship graph of the business object;

[0008] Performing a graph attention operation on the first feature relationship graph to obtain attention coefficients between nodes and node features of nodes in the first feature relationship graph;

[0009] Reconstructing the first feature relationship graph based on the attention coefficients between the nodes to obtain a second feature relationship graph of the business object;

[0010] Based on the node features of the nodes and the second feature relationship graph, business demand information of the business object is predicted.

[0011] In a second aspect, an embodiment of the present application provides a business demand prediction device, including:

[0012] A construction unit, configured to construct a first feature relationship graph of the business object based on an object feature set of the business object;

[0013] A computing unit, configured to perform a graph attention operation on the first feature relationship graph to obtain attention coefficients between nodes and node features of nodes in the first feature relationship graph;

[0014] A reconstruction unit, configured to reconstruct the first feature relationship graph based on the attention coefficients between the nodes to obtain a second feature relationship graph of the business object;

[0015] A prediction unit is used to predict the business demand information of the business object based on the node feature of the node and the second feature relationship graph.

[0016] In a third aspect, an embodiment of the present application provides an electronic device, including:

[0017] processor;

[0018] a memory for storing instructions executable by the processor;

[0019] The processor is configured to execute the instructions to implement the business demand forecasting method provided in the first aspect.

[0020] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium. When instructions in the storage medium are executed by a processor of an electronic device, the electronic device can execute the business demand prediction method provided in the first aspect.

[0021] In the fifth aspect, an embodiment of the present application provides a computer program product, characterized in that the computer program product includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to enable a computer to execute part or all of the steps in the business demand forecasting method provided in the first aspect.

[0022] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects:

[0023] Firstly, based on the object feature set of the business object, the first feature relationship graph of the business object is constructed, which can intuitively reflect the complex relationship between different object features; by performing graph attention operation on the first feature relationship graph, this complex relationship can be effectively captured and learned, and feature extraction can be performed quickly and accurately to obtain the attention coefficients between nodes and the node features of the nodes in the first feature relationship graph. The attention coefficients can accurately reflect the degree of correlation between object features, and the node features of the nodes contain the hidden information behind the object features, which have stronger representation capabilities than the object features; further, the first feature relationship graph is reconstructed based on the attention coefficients between nodes, and the obtained second feature relationship graph can more intuitively and accurately reflect the complex relationship between different object features compared to the first feature relationship graph; business demand prediction based on the second feature relationship graph and node features can effectively combine data of different modalities, utilize the correlation between different object features and the hidden information behind the object features, and more comprehensively understand and analyze the object feature set, thereby improving the accuracy and efficiency of business demand prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0025] Figure 1 A flowchart of a method for predicting business demand provided by an embodiment of the present application;

[0026] Figure 2A A first characteristic relationship diagram provided for an embodiment of the present application;

[0027] Figure 2B Another first characteristic relationship diagram provided for an embodiment of the present application;

[0028] Figure 3 A first characteristic relationship diagram and a reconstructed second characteristic relationship diagram provided in an embodiment of the present application;

[0029] Figure 4 Another embodiment of the present application provides a flowchart of a method for predicting business demand;

[0030] Figure 5 A schematic diagram of the structure of a business demand prediction device provided for one embodiment of the present application;

[0031] Figure 6 A schematic diagram of the structure of an electronic device provided for one embodiment of the present application. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0033] The terms "first", "second", etc. in this specification and claims are used to distinguish similar objects, and are not used to describe a particular order or precedence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein. In addition, "and / or" in this specification and claims means at least one of the connected objects, and the character " / " generally means that the objects associated with each other are in an "or" relationship.

[0034] Some concept descriptions:

[0035] Graph Attention Networks (GAT): is a graph neural network model used to process graph structured data. It introduces an attention mechanism between nodes to effectively capture the relationships and features between nodes.

[0036] Multimodal Large Language Models (MLLM): A large neural network model that can process multiple types of data (such as text, images, audio, etc.) at the same time. It aims to effectively integrate and utilize data of different modalities, so as to more comprehensively understand and process complex multimodal data.

[0037] Multi-layer Perceptron (MLP): is a common artificial neural network result, consisting of one or more hidden layers and an output layer.

[0038] In order to quickly and accurately predict business needs from massive data, an embodiment of the present application proposes a business demand prediction method. First, based on the object feature set of the business object, a first feature relationship graph of the business object is constructed, which can intuitively reflect the complex relationship between different object features; by performing graph attention operation on the first feature relationship graph, this complex relationship can be effectively captured and learned, and feature extraction is performed quickly and accurately to obtain the attention coefficient between nodes and the node features of the nodes in the first feature relationship graph. The attention coefficient can accurately reflect the degree of association between object features, and the node features of the nodes contain hidden information behind the object features, and have stronger representation capabilities than the object features; further, the first feature relationship graph is reconstructed based on the attention coefficient between nodes, and the obtained second feature relationship graph can more intuitively and accurately reflect the complex relationship between different object features compared to the first feature relationship graph; user demand prediction is performed based on the second feature relationship graph and the node features, which can effectively combine data of different modalities, and use the association relationship between different object features and the hidden information behind the object features to more comprehensively understand and analyze the user feature set, thereby improving the accuracy and efficiency of user demand prediction.

[0039] It should be understood that the business demand prediction method proposed in the embodiment of the present application can be executed by an electronic device. As an example, it can be executed by software in an electronic device. The so-called electronic device here can include terminal devices, such as smart phones, tablet computers, laptops, desktop computers, intelligent voice interaction devices, smart home appliances, smart watches, vehicle terminals, aircraft, etc.; or, the electronic device can also include a server, such as an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0040] In one implementation, the implementation environment applicable to a method for predicting business demand provided in an embodiment of the present application includes a client and a server. The client may be an electronic device with a business access function, such as but not limited to a smart phone, a tablet computer, a laptop computer, a desktop computer, an intelligent voice interaction device, a smart home appliance, a smart watch, a vehicle terminal, an aircraft, etc. Alternatively, the client may also be software running in an electronic device.

[0041] A server is an electronic device with computing capabilities. It can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0042] The client and the server exchange data through the network. The client periodically collects the object feature set of the business object and sends it to the server for storage. The server constructs a first feature relationship graph of the business object, predicts the business demand information of the business object based on the first feature relationship graph, and performs business processing based on the business demand information, and returns the business processing results to the client to support the client's business access function. For example, in a service push scenario, the server predicts the business object's demand for a certain service based on the first feature relationship graph. If the demand is greater than a preset threshold, the service message related to the service is pushed to the client. For another example, in a resource allocation scenario, the server predicts the number of resources required by the business object based on the first feature relationship graph, and then issues a corresponding number of resources to the client for the business object to use. For another example, in a friend recommendation scenario, the server predicts other social objects of interest to the business object based on the first feature relationship graph, and then pushes other social objects to the client so that the business object can add these social objects as friends.

[0043] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0044] Please refer to Figure 1 , is a flow chart of a method for predicting business demand provided by an embodiment of the present application, the method comprising the following steps:

[0045] S102: construct a first feature relationship graph of the business object based on the object feature set of the business object.

[0046] The business object here refers to any object that needs to be predicted. For example, in a service push scenario, the service provider often needs to push corresponding service messages to the user based on the user's service needs. At this time, the business object is the user to be predicted. For another example, in a communication resource allocation scenario, the resource server often needs to allocate communication resources to the client based on the client's resource needs. At this time, the business object is the client to which the resources are allocated. For another example, in a friend recommendation scenario, the social platform often needs to recommend new friends to the user based on the user's historical social records. At this time, the business object is the user to be recommended as a friend.

[0047] If demand forecasting is to be performed for multiple business objects, steps S102 to S108 may be performed for each business object to obtain business demand information of the business object.

[0048] The object feature set contains multiple object features of the business object, which are used to describe the characteristics of the business object. For example, in the service push scenario, the object feature set is used to describe the user's portrait, which specifically includes but is limited to user features such as the user's age, gender, marital status, close friends, income, browsing history, shopping software click history, psychological characteristics, etc. It is worth noting that all user features contained in the object feature set are obtained after user authorization and anonymization. For another example, in the resource allocation scenario, the object feature set is used to describe the client's resource usage, which specifically includes but is not limited to the client's historical resource application volume, historical resource utilization rate, current business concurrency, etc.

[0049] The first feature relationship graph describes the association relationship between different object features in the object feature set in the form of a graph. In the above S102, the first feature relationship graph can be constructed in various ways, which can be selected according to actual needs.

[0050] In one embodiment, the first feature relationship graph includes three levels of nodes, the first level nodes represent business objects, each second level node represents a category, and the category is obtained by classifying the object feature set. Each third level node represents an object feature, and the object feature belongs to the category represented by the second level node adjacent to the third level node.

[0051] In this case, the above S102 includes the following steps: classifying the object feature set to obtain multiple subsets; for each subset, creating a first-level node corresponding to the business object and a second-level node corresponding to the subset, and creating a connecting edge between the first-level node and the second-level node; for each subset, creating a third-level node corresponding to the object feature in the subset, and creating a connecting edge between the third-level node and the second-level node.

[0052] For example, in the service push scenario, considering that too large or too small a number of categories may affect the adequacy of the expression of object features, in order to improve the adequacy of the expression of user features, the object feature set of object i is divided into 6 categories, namely basic information, social relations, economic status, behavioral characteristics, psychological characteristics, and hobbies. Thus, the object feature set is divided into 6 subsets, each of which corresponds to a category. Among them, the subset corresponding to the basic information includes but is not limited to object features such as age, common address, contact information, gender, and education, the subset corresponding to the social relationship includes object features such as marital status, close friends, and family members, the subset corresponding to the economic status includes object features such as monthly income and monthly consumption, the subset corresponding to the behavioral characteristics includes user features such as browsing history and shopping software clicked, the subset corresponding to the psychological characteristics includes object features such as indecision and comparison, and the subset corresponding to hobbies includes object features such as liking food and liking sports.

[0053] Then, create the first-level node corresponding to object i, and for each subset, create the second-level node corresponding to the subset and the third-level node corresponding to each object feature in the subset. Figure 2A As shown, object i is the first-level node, nodes 1 to 6 are second-level nodes, and the remaining nodes are third-level nodes. Among them, nodes 1_1 to 1_M correspond to the M object features contained in the subset "basic information", nodes 2_1 to 2_N correspond to the N object features contained in the subset "social relations", nodes 3_1 to 3_O correspond to the O object features contained in the subset "economic status", nodes 4_1 to 4_P correspond to the P object features contained in the subset "behavioral characteristics", nodes 5_1 to 5_Q correspond to the Q object features contained in the subset "psychological characteristics", and nodes 6_1 to 6_R correspond to the R object features contained in the subset "interests and hobbies".

[0054] Next, for each subset, a connection edge is created between the first-level node and the second-level node corresponding to the subset, indicating that the subset belongs to object i; in addition, a connection edge is created between the second-level node corresponding to the subset and the third-level node corresponding to the object feature contained in the subset, indicating that the object feature corresponding to the third-level node belongs to the subset corresponding to the second-level node. Thus, we get Figure 2A The first characteristic relationship diagram is shown on the right.

[0055] In the above implementation, after classifying the object feature set, corresponding nodes are created for the user, each subset, and each object feature, and corresponding connection edges are created between the nodes with inclusion relationships. The first feature relationship graph formed can intuitively and accurately reflect the attribution relationship between the business object and the object feature set and the category to which the object feature belongs.

[0056] In another embodiment, the above S102 includes the following steps: for each object feature in the object feature set, based on the level to which the object feature belongs, creating a corresponding node for the object feature; based on the hierarchical relationship between the object features in the object feature set, creating a first connecting edge between the nodes corresponding to different object features; based on the association relationship between the nodes corresponding to the object features of the same level, creating a second connecting edge between the nodes corresponding to the object features of the same level.

[0057] For example, Figure 2B As shown in the friend recommendation scenario, the object feature set includes the object identifier of business object i, the historical social platforms of n historical friends of business object i (friend 1 to friend n), and the social features on the historical social platforms. In this case, the basic information of business object i is used as the first-level information to create the first-level node, that is, Figure 2B Business object i in; take the user ID of each historical friend as the second-level feature and create a second-level node, that is, Figure 2B Friends 1 to friends n in the second level nodes, and create the first connection edge between these second level nodes and the first level nodes (such as Figure 2B The historical social platforms used by each friend are used as the third-level features to create third-level nodes, that is, Figure 2B In the social platform 11 to social platform 1m, social platform 21 to social platform 2k and social platform n1 to social platform nq, a first connection edge is created between the third-level node and the second-level node of each friend (such as Figure 2B The social features on each social platform are used as the fourth-level features to create the fourth-level nodes, that is, Figure 2B In the social features 111 to 1m1, social features 211 to 2k1, and social features n11 to nq1, a first connection edge is created between the fourth-level node and the third-level node of each friend (such as Figure 2B ) as shown by the solid line in .

[0058] Furthermore, since friend 1 and friend 2 are friends with each other, a second connection edge is created between the node corresponding to friend 1 and the node corresponding to friend 2 (e.g. Figure 2B As shown by the dotted line in ). Since the social platform 1m and the social platform 21 are the same social platform, a second connection edge is created between the node corresponding to the social platform 1m and the node corresponding to the social platform 21 (as shown in FIG. Figure 2B ).

[0059] The first feature relationship graph created by the above implementation can intuitively and accurately reflect the hierarchical relationship between object features in the object feature set and the association relationship between object features at the same level.

[0060] The above only shows some implementation methods of the above S102. Of course, it should be understood that the above S102 can also be implemented in other ways, and the present embodiment of the application does not limit this.

[0061] S104, performing a graph attention operation on the first feature relationship graph to obtain attention coefficients between nodes and node features of the nodes in the first feature relationship graph.

[0062] Graph attention operation refers to the introduction of attention mechanism in the process of feature extraction of graph structure data. By capturing and learning the complex relationship between different nodes, dynamically adjusting the weights between nodes, analyzing the attention coefficients between different nodes, and then realizing weighted aggregation of node features of different nodes based on the attention coefficient, localized feature extraction is realized. In this way, the obtained attention coefficient can accurately reflect the degree of correlation between node features. The extracted node features contain the hidden information behind the object features, and have stronger representation ability than object features. It also gives the feature extraction process a certain degree of interpretability, that is, it reflects how the degree of correlation between different nodes affects the node feature extraction results.

[0063] In the above S104, the attention coefficient between nodes includes the attention coefficient between each node and its adjacent nodes in the first feature relationship graph. The attention coefficient can be determined in various ways, which can be selected according to actual needs. In one embodiment, for each node, the method for determining the attention coefficient between the node and its adjacent nodes includes: step A1, feature extraction of the first feature relationship graph, the first candidate node feature of the node in the first relationship graph and the second candidate node feature of the adjacent node of the node; step A2, based on the first candidate node feature and the second candidate node feature, a graph attention operation is performed to obtain the correlation coefficient between the node and the adjacent node; step A3, the correlation coefficient is normalized to obtain the attention coefficient between the node and the adjacent node.

[0064] In step A1, as an example, based on the first feature relationship graph, an initial feature matrix X and an initial adjacency matrix A are generated; then, the initial feature matrix X and the initial adjacency matrix A are input into the pre-trained GAT network for feature extraction to obtain a candidate feature matrix for each node in the first feature relationship graph.

[0065] Among them, the initial feature matrix X includes the feature vector of each node in the first feature relationship graph, and the feature vector is obtained by mapping the object feature corresponding to the node to the specified dimensional space. The initial adjacency matrix A represents the connection edges between each node and other nodes in the first feature relationship graph. For each node, the candidate node feature of the node is called the first candidate node feature, and the candidate node feature of the adjacent node of the node is called the second candidate node feature. In practical applications, the dimensions of the candidate node features can be set according to experience to avoid problems such as too large dimensions increasing the amount of calculation and too small dimensions causing the loss of important information. For example, if the dimension of the object feature is large, the dimension of the candidate node feature can be set to a smaller value to avoid too large dimensions and increase the amount of calculation; if the dimension of the object feature is small, the dimension of the candidate node feature can be set to a larger value to avoid too small dimensions and cause the loss of important information in the object feature.

[0066] In the above step A2, the correlation coefficient between the node and the adjacent node is used to represent the degree of association between the node and the adjacent node. The graph attention operation on the first candidate node feature and the second candidate node feature can be implemented in various ways. As an example, the first candidate node feature is linearly mapped to obtain a first mapping feature; the second candidate node feature is linearly mapped to obtain a second mapping feature; the first mapping feature and the second mapping feature are spliced ​​to obtain a spliced ​​feature; the spliced ​​feature is single-valued mapped to obtain the correlation coefficient between the node and the adjacent node.

[0067] It is worth noting that, for each node, if the node has multiple adjacent nodes, the correlation coefficient between the node and each of its adjacent nodes can be determined by the above method. Specifically, for each node, the correlation coefficient between the node and each of its adjacent nodes can be determined by the following formula (1).

[0068] e ij =a([Wh i ||Wh j ]),j∈N i (1)

[0069] Among them, e ij represents the correlation coefficient between node i and adjacent node j; h i represents the first candidate node feature, that is, the candidate node feature of node i; h j represents the second candidate node feature, that is, the candidate node feature of the adjacent node j; N i represents the number of neighboring nodes of node i; W is a shared parameter used to increase the dimension of candidate node features; || represents concatenation; a(·) represents a single-valued mapping function used to map the concatenated high-dimensional features to a real number.

[0070] In the above step A3, the attention coefficient between a node and its adjacent nodes indicates the degree of influence of the adjacent nodes on the node. If the attention coefficient is larger, the degree of influence of the adjacent nodes on the node is greater; conversely, the degree of influence of the adjacent nodes on the node is smaller. As an example, for each node, the attention coefficient between the node and each of its adjacent nodes can be determined by the following formula (2).

[0071]

[0072] Among them, a ij represents the attention coefficient between node i and adjacent node j; LeakyReLU(·) represents the activation function; exp represents the exponential operation; e ij represents the correlation coefficient between node i and its adjacent node j; e ik represents the correlation coefficient between node i and the kth adjacent node; N i Represents the number of adjacent nodes of node i.

[0073] In practical applications, the above steps A2 and A3 can also be implemented through the GAT model.

[0074] In the above S104, in one embodiment, for each node, the node feature of the node is determined in a manner including: determining the node feature of the node based on the second candidate node feature and the attention coefficient between the node and the adjacent nodes.

[0075] Specifically, as an example, based on the attention coefficient between the node and the adjacent nodes, the first candidate node feature and the second candidate node feature are weighted summed to obtain the node feature of the node.

[0076] As another example, based on the attention coefficient between the node and the adjacent nodes, the features of the second candidate node are fused to obtain the fused features; and the fused features are operated based on the activation function to obtain the node features of the node.

[0077] For example, the node characteristics of a node can be determined by the following formula (3).

[0078]

[0079] Among them, h′ i represents the node feature of node i; σ(·) represents the activation function, a ij represents the attention coefficient between node i and adjacent node j; h j represents the second candidate node feature, that is, the candidate node feature of the adjacent node j; W is a shared parameter used to increase the dimension of the second candidate node feature; N i Represents the number of adjacent nodes of node i.

[0080] In the above implementation, for each node, based on the attention coefficient between the node and its adjacent nodes, the candidate node features of all adjacent nodes of the node are fused and activated. The obtained node features contain neighborhood information, that is, the hidden information behind the object features, and have stronger representation capabilities than object features.

[0081] The embodiment of the present application shows some implementation methods of the above S104. Of course, it should be understood that the above S104 can also be implemented in other ways.

[0082] S106, reconstructing the first feature relationship graph based on the attention coefficients between the nodes to obtain a second feature relationship graph of the business object.

[0083] Since the attention coefficient between nodes reflects the degree of influence of different nodes on each other, the first feature relationship graph is reconstructed based on the attention coefficient between nodes. Compared with the first feature relationship, the obtained second feature relationship graph can accurately and intuitively reflect the degree of influence of different nodes on each other, that is, the degree of influence of different object features on business objects, without changing the graph structure, that is, without changing the connection relationship between nodes. On the one hand, it can more intuitively and accurately reflect the complex relationship between different object features, and on the other hand, it helps to increase the interpretability of the business demand forecasting process, that is, it explains how different object features affect the business demand forecasting results of business objects.

[0084] In the above S106, the first characteristic relationship graph may be reconstructed in various ways, which may be selected according to actual needs.

[0085] In one embodiment, in order to make the degree of influence of different nodes on each other more intuitively displayed in the second feature relationship diagram, the above S106 includes the following steps: Step B1, based on the attention coefficient between the node and the adjacent nodes, determine the edge weight of the connecting edge between the node and the adjacent nodes; Step B2, based on the edge weight, updates the attributes of the connecting edge to obtain the second feature relationship diagram of the business object.

[0086] In the above step B1, the edge weight of the connecting edge represents the influence degree of the connected adjacent node on the node. The edge weight of the connecting edge is proportional to the attention coefficient between the connected nodes. The edge weight of the connecting edge can be determined by various methods, which can be selected according to actual needs, and the embodiment of the present application does not limit this.

[0087] As an example, for each node, the attention coefficient between the node and its adjacent nodes can be used as the edge weight of the connecting edge between the node and its adjacent nodes.

[0088] As another example, in order to better reflect the degree of influence of object features corresponding to different nodes on business objects as a whole through the edge weights of each connecting edge, when the first feature relationship graph includes multi-level nodes, the level of the node is higher than the level of the node's adjacent nodes, then the edge weight of the connecting edge between the node and its adjacent nodes is determined in the following manner: determine the sum of the attention coefficients between all nodes in the first feature relationship graph; determine the target connecting edge from the adjacent node to the first-level node from the first feature relationship graph; determine the product of the attention coefficients between the nodes connected by the target connecting edge; and determine the edge weight of the connecting edge between the node and its adjacent nodes based on the ratio of the product to the sum of the attention coefficients.

[0089] For example, Figure 3 Taking the first feature relationship graph shown as an example, the first feature relationship graph includes three-level nodes, indicating the attention coefficient between the first-level node "user i" and the connected second-level node 1, indicating the attention coefficient between the second-level node 1 and its adjacent third-level node 1_1. Based on this, the edge weight of the connecting edge between the second-level node 1 and the third-level node 1_1 can be determined by the following formula (4).

[0090]

[0091] in, represents the edge weight of the edge connecting the second-level node 1 and its adjacent third-level node 1_1; represents the sum of the attention coefficients between the second-level node 1 and all its adjacent third-level nodes, represents the sum of the attention coefficients between the second-level node 2 and all its adjacent third-level nodes, Represents the sum of the attention coefficients between the second-level node 6 and all its adjacent third-level nodes.

[0092] In the above step B2, the attributes of the connecting edge may include, but are not limited to: thickness, color, line type, transparency, etc. Figure 3 As shown in the first characteristic relationship graph, the edge weights of the connecting edges between each node and the adjacent nodes are in the following order from large to small: the edge weight of the connecting edge between node "user i" and node 1> the edge weight of the connecting edge between node "user i" and nodes 2 to 5> the edge weight of the connecting edge between node "user i" and node 6> the edge weight of the connecting edge between node 1 and node 1_1=the edge weight of the connecting edge between node 4 and node 4_1>the edge weight of the connecting edges between other nodes. Based on this, based on the edge weight of each connecting edge, the thickness of each connecting edge is updated, and the following can be obtained: Figure 3 The second characteristic relationship diagram is shown on the right.

[0093] The embodiment of the present application shows some implementation methods of the above S106. Of course, it should be understood that the above S106 can also be implemented in other ways, and the embodiment of the present application does not limit this.

[0094] S108, predicting business demand information of the business object based on the node feature and the second feature relationship graph of the node.

[0095] The business demand information is used to describe the business object's demand for a certain business service. The business services here may include, but are not limited to: communication resources, products, service messages, business data, etc.

[0096] The business demand information of the business object varies according to the prediction task. For example, in a service push scenario, if the prediction task is to predict the business object's (i.e., user's) demand for a product, the business demand information of the business object includes the business object's demand for the product; if the prediction task is to predict whether the business object currently needs a certain service, the business demand information of the business object is the probability that the user needs the service; if the prediction task is to predict whether the business object currently needs a certain service and determine the key factors affecting the business demand, the business demand information of the business object includes the probability that the business object needs the service, and the key factors affecting the probability. For another example, in a communication resource allocation scenario, if the prediction task is the amount of resources required by the business object (i.e., client), the business demand information of the business object includes the amount of communication resources required by the business object.

[0097] In one implementation, the node feature and the second feature relationship graph of the node may be input into a pre-trained demand prediction model for prediction to obtain business demand information of the business object.

[0098] In another embodiment, in order to perform demand forecasting more accurately, prompt words are generated based on forecasting task information; and business demand information of the business object is forecasted based on the second feature relationship graph, node features of the nodes, and the prompt words.

[0099] As an example, the description text of the prediction task information, the description text of the second feature relationship graph, and the description text of the node features can be spliced ​​together as prompt words, and the prompt words, the second feature relationship graph, and the node features of the nodes can be input into a pre-trained demand prediction model. The prompt words can be used to guide the model to analyze the second feature relationship graph and the node features to accurately predict the business demand information of the business object.

[0100] For example, Figure 3 Taking the second characteristic relationship diagram shown in FIG. 1 as an example, if the prediction task is to predict whether the business object currently needs a certain service and determine the key factors affecting the business demand, the following prompt words are generated:

[0101] "Input 1 is the second feature relationship diagram of business object i, where the thickness of the connecting edge indicates the degree of influence of the feature corresponding to the node on business object i; Input 2 is the node feature of each node. Based on the above two inputs, please predict the business needs of the business object and perform key object feature analysis to determine the key factors affecting the business needs."

[0102] In the embodiment of the present application, the demand forecasting model can adopt various language models in the field, which can be set according to actual needs, and the embodiment of the present application does not limit this. In a more preferred solution, considering that the second feature relationship graph and the node features belong to data of different modes, in order to effectively combine data of different modes to more accurately forecast demand, the demand forecasting model can adopt the MLLM model.

[0103] In order to facilitate the understanding of the business demand prediction method provided in the embodiment of the present application, Figure 4 Taking as an example, the implementation process of the business demand prediction method provided in the embodiment of the present application is described.

[0104] Please refer to Figure 4 , is a flow chart of a business demand forecasting method provided by another embodiment of the present application, which is implemented with the help of the GAT model and the MLLM model.

[0105] like Figure 4 As shown, first, the object feature set of business object i is divided into multiple subsets, each subset corresponds to a category, and corresponding nodes are created for business object i, each subset, and the object features in each subset, respectively; further, connecting edges are created between the nodes corresponding to business object i and the nodes corresponding to each subset, and connecting edges are created between the nodes corresponding to each subset and the nodes corresponding to each object feature in the subset, so as to obtain the first feature relationship graph of business object i.

[0106] Then, the first feature relationship graph is input into the GAT model for graph attention operation to obtain the attention coefficients between nodes in the first feature relationship graph and the node features of each node.

[0107] Next, based on the attention coefficients between the nodes, the first feature relationship graph is reconstructed to obtain the second feature relationship graph of the business object i.

[0108] Finally, the prompt words of the prediction task, the node features of each node, and the second feature relationship diagram are input into the MLLM model for demand prediction to obtain the demand information of business object i.

[0109] The business demand forecasting method provided in the above embodiment is based on deep learning technology to forecast business demand, and has stronger real-time and high efficiency. In particular, the MLLM model is adopted, which provides the possibility of using multimodal data to quickly process and analyze mechanical energy in a short time, and can update the forecast results in time to provide timely decision support for the business party. Therefore, this business demand forecasting method based on deep learning technology has broader application prospects and potential in the face of an increasingly complex and changing business environment.

[0110] Secondly, the MLLM model can integrate multimodal information from various data sources, understand and analyze data more comprehensively, and thus improve the accuracy of demand forecasting. Given its ability to handle diverse inputs, the MLLM model performs well in generalizing new data and has broad application prospects and potential.

[0111] In addition, compared with traditional feature engineering that relies on domain knowledge and professional experience, the GAT model can quickly and accurately capture the complex relationship between different object features from the first feature relationship graph, analyze the impact of different object features on model performance based on the graph attention mechanism, and dynamically adjust the degree of influence between nodes to achieve localized feature extraction without manual design, which is convenient for better optimizing the performance of the entire system in an end-to-end manner and improving the interpretability of subsequent demand forecasting processes. Therefore, the graph attention operation scheme based on the GAT model can improve the accuracy and efficiency of attention operations and feature extraction.

[0112] One or more embodiments of the present application provide a business demand prediction method. First, based on the object feature set of the business object, a first feature relationship graph of the business object is constructed, which can intuitively reflect the complex relationship between different object features. By performing a graph attention operation on the first feature relationship graph, this complex relationship can be effectively captured and learned, and feature extraction is performed quickly and accurately to obtain the attention coefficient between nodes and the node features of the nodes in the first feature relationship graph. The attention coefficient can accurately reflect the degree of association between object features, and the node features of the nodes contain hidden information behind the object features, and have stronger representation capabilities than the object features. Further, the first feature relationship graph is reconstructed based on the attention coefficient between nodes, and the obtained second feature relationship graph can more intuitively and accurately reflect the complex relationship between different object features compared to the first feature relationship graph. Business demand prediction is performed based on the second feature relationship graph and the node features, which can effectively combine data of different modalities, and can use the association relationship between different object features and the hidden information behind the object features to more comprehensively understand and analyze the object features, which can not only improve the accuracy and efficiency of business demand prediction, but also reflect the influence of different object features on the demand prediction results, so that the demand prediction process has interpretability.

[0113] The business demand prediction method provided in the embodiment of the present application can be applied to business demand prediction in various business scenarios.

[0114] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0115] Based on the same inventive concept, the present application embodiment also provides a business demand prediction device. Figure 5 , is a structural diagram of a business demand prediction device provided in an embodiment of the present application, the device includes: a construction unit 510, a calculation unit 520, a reconstruction unit 530 and a prediction unit 540.

[0116] The construction unit 510 is used to construct a first feature relationship graph of the business object based on the object feature set of the business object.

[0117] The operation unit 520 is used to perform a graph attention operation on the first feature relationship graph to obtain the attention coefficients between nodes and the node features of the nodes in the first feature relationship graph.

[0118] The reconstruction unit 530 is used to reconstruct the first feature relationship graph based on the attention coefficients between the nodes to obtain a second feature relationship graph of the business object.

[0119] The prediction unit 540 is used to predict the business demand information of the business object based on the node feature of the node and the second feature relationship graph.

[0120] In another embodiment, the building block is used to:

[0121] Classifying the object feature set to obtain multiple subsets;

[0122] Creating a first-level node corresponding to the business object and a second-level node corresponding to the subset, and creating a connecting edge between the first-level node and the second-level node;

[0123] A third-level node corresponding to the object feature in the subset is created, and a connecting edge is created between the third-level node and the second-level node.

[0124] In another embodiment, performing a graph attention operation on the first feature relationship graph to obtain attention coefficients between nodes and node features of nodes in the first feature relationship graph includes:

[0125] Performing feature extraction on the first feature relationship graph to obtain a first candidate node feature of a node in the first feature relationship graph and a second candidate node feature of an adjacent node of the node;

[0126] Performing a graph attention operation based on the first candidate node feature and the second candidate node feature to obtain a correlation coefficient between the node and the adjacent node;

[0127] Normalizing the correlation coefficient to obtain an attention coefficient between the node and the adjacent node;

[0128] Based on the second candidate node feature and the attention coefficient between the node and the adjacent node, the node feature of the node is determined.

[0129] In another embodiment, when the operation unit performs a graph attention operation based on the first candidate node feature and the second candidate node feature to obtain a correlation coefficient between the node and the adjacent node, the operation unit performs the following steps:

[0130] Performing linear mapping on the first candidate node feature to obtain a first mapping feature;

[0131] Performing linear mapping on the second candidate node feature to obtain a second mapping feature;

[0132] splicing the first mapping feature and the second mapping feature to obtain a spliced ​​feature;

[0133] The concatenated features are subjected to single-value mapping to obtain the correlation coefficient between the node and the adjacent nodes.

[0134] In another embodiment, the computing unit performs the following steps when determining the node feature of the node based on the second candidate node feature and the attention coefficient between the node and the adjacent node:

[0135] Based on the attention coefficient between the node and the adjacent node, the second candidate node features are fused to obtain fused features;

[0136] The fused features are operated based on an activation function to obtain node features of the nodes.

[0137] In another embodiment, the attention coefficient between the nodes includes an attention coefficient between the node and an adjacent node of the node;

[0138] The reconstruction unit is used for:

[0139] Determine an edge weight of an edge connecting the node and the adjacent node based on an attention coefficient between the node and the adjacent node;

[0140] Based on the edge weight, the attribute of the connecting edge is updated to obtain a second characteristic relationship graph of the business object.

[0141] In another embodiment, the first feature relationship graph includes multiple levels of nodes, the level of the node is higher than the level of the adjacent node of the node; the edge weight of the connecting edge between the node and the adjacent node is determined by:

[0142] Determine the sum of attention coefficients between all nodes in the first feature relationship graph;

[0143] Determine, from the first feature relationship graph, a target connection edge from the adjacent node to the first-level node;

[0144] Determine the product of attention coefficients between nodes connected by the target connection edge;

[0145] Based on the ratio of the product to the sum of the attention coefficients, determine the edge weight of the edge connecting the node node and the adjacent node

[0146] In another embodiment, the prediction unit is used to:

[0147] Generate prompt words based on the predicted task information;

[0148] Based on the second feature relationship graph, the node features of the nodes and the prompt words, business demand information of the business object is predicted.

[0149] Obviously, the business demand prediction device 500 provided in the embodiment of the present application can be used as the above Figure 1 The execution subject of the business demand prediction method shown in FIG. Figure 1 The functions realized are as follows. Since the principles are the same, they will not be described in detail.

[0150] Figure 6 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. Figure 6At the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and a memory. The memory may include a memory, such as a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. Of course, the electronic device may also include hardware required for other services.

[0151] The processor, network interface and memory can be interconnected through an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0152] The memory is used to store the program. Specifically, the program may include a program code, and the program code includes a computer operation instruction. The memory may include a memory and a non-volatile memory, and provides instructions and data to the processor.

[0153] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a user demand prediction device at the logical level. The processor executes the program stored in the memory and is specifically used to perform the following operations:

[0154] Based on the object feature set of the business object, construct a first feature relationship graph of the business object;

[0155] Performing a graph attention operation on the first feature relationship graph to obtain attention coefficients between nodes and node features of nodes in the first feature relationship graph;

[0156] Reconstructing the first feature relationship graph based on the attention coefficients between the nodes to obtain a second feature relationship graph of the business object;

[0157] Based on the node features of the nodes and the second feature relationship graph, business demand information of the business object is predicted.

[0158] The above application Figure 1The method performed by the user demand prediction device disclosed in the illustrated embodiment can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in software form. The above processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware decoding processor for execution, or a combination of hardware and software modules in the decoding processor for execution. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0159] The electronic device may also perform Figure 1 The method is used to realize the user demand prediction device. Figure 1 , Figure 4 The functions of the illustrated embodiment will not be described in detail in the embodiments of the present application.

[0160] Of course, in addition to software implementation methods, the electronic device of the present application does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc. That is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0161] The embodiment of the present application also provides a computer-readable storage medium, which stores one or more programs, wherein the one or more programs include instructions, which, when executed by a portable electronic device including a plurality of application programs, enable the portable electronic device to execute Figure 1 The method of the embodiment shown is specifically used to perform the following operations:

[0162] Based on the object feature set of the business object, construct a first feature relationship graph of the business object;

[0163] Performing a graph attention operation on the first feature relationship graph to obtain attention coefficients between nodes and node features of nodes in the first feature relationship graph;

[0164] Reconstructing the first feature relationship graph based on the attention coefficients between the nodes to obtain a second feature relationship graph of the business object;

[0165] Based on the node features of the nodes and the second feature relationship graph, business demand information of the business object is predicted.

[0166] An embodiment of the present application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute some or all of the steps in the user demand prediction method provided in the embodiment of the present application.

[0167] In short, the above description is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0168] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0169] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0170] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0171] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

Claims

1. A method for predicting business demand, characterized in that: include: Based on the object feature set of the business object, construct a first feature relationship graph of the business object; Performing a graph attention operation on the first feature relationship graph to obtain attention coefficients between nodes and node features of nodes in the first feature relationship graph; Reconstructing the first feature relationship graph based on the attention coefficients between the nodes to obtain a second feature relationship graph of the business object; Based on the node features of the nodes and the second feature relationship graph, business demand information of the business object is predicted.

2. The method according to claim 1, characterized in that The performing graph attention operation on the first feature relationship graph to obtain attention coefficients between nodes and node features of nodes in the first feature relationship graph includes: Performing feature extraction on the first feature relationship graph to obtain a first candidate node feature of a node in the first feature relationship graph and a second candidate node feature of an adjacent node of the node; Performing a graph attention operation based on the first candidate node feature and the second candidate node feature to obtain a correlation coefficient between the node and the adjacent node; Normalizing the correlation coefficient to obtain an attention coefficient between the node and the adjacent node; Based on the second candidate node feature and the attention coefficient between the node and the adjacent node, the node feature of the node is determined.

3. The method according to claim 2, characterized in that The performing graph attention operation based on the first candidate node feature and the second candidate node feature to obtain a correlation coefficient between the node and the adjacent node includes: Performing linear mapping on the first candidate node feature to obtain a first mapping feature; Performing linear mapping on the second candidate node feature to obtain a second mapping feature; splicing the first mapping feature and the second mapping feature to obtain a spliced ​​feature; The concatenated features are subjected to single-value mapping to obtain the correlation coefficient between the node and the adjacent nodes.

4. The method according to claim 2, characterized in that: The determining the node feature of the node based on the second candidate node feature and the attention coefficient between the node and the adjacent node includes: Based on the attention coefficient between the node and the adjacent node, the second candidate node feature is fused to obtain a fused feature; The fused features are operated based on an activation function to obtain node features of the node.

5. The method according to claim 1, characterized in that The attention coefficient between the nodes includes the attention coefficient between the node and the adjacent node of the node; The reconstructing the first feature relationship graph based on the attention coefficients between the nodes to obtain a second feature relationship graph of the business object includes: Determine an edge weight of an edge connecting the node and the adjacent node based on an attention coefficient between the node and the adjacent node; Based on the edge weight, the attribute of the connecting edge is updated to obtain a second characteristic relationship graph of the business object.

6. The method according to claim 5, characterized in that The first feature relationship graph includes multiple levels of nodes, and the level of the node is higher than the level of the adjacent node of the node; the edge weight of the connecting edge between the node and the adjacent node is determined by: Determine the sum of attention coefficients between all nodes in the first feature relationship graph; Determine, from the first feature relationship graph, a target connection edge from the adjacent node to the first-level node; Determine the product of attention coefficients between nodes connected by the target connection edge; Based on the ratio of the product to the sum of the attention coefficients, an edge weight of the edge connecting the node and the adjacent node is determined.

7. The method according to any one of claims 1 to 6, characterized in that The predicting the business demand information of the business object based on the node feature of the node and the second feature relationship graph includes: Generate prompt words based on the predicted task information; Based on the second feature relationship graph, the node features of the nodes and the prompt words, business demand information of the business object is predicted.

8. A business demand forecasting device, characterized in that: include: A construction unit, configured to construct a first feature relationship graph of the business object based on an object feature set of the business object; A computing unit, configured to perform a graph attention operation on the first feature relationship graph to obtain attention coefficients between nodes and node features of nodes in the first feature relationship graph; A reconstruction unit, configured to reconstruct the first feature relationship graph based on the attention coefficients between the nodes to obtain a second feature relationship graph of the business object; A prediction unit is used to predict the business demand information of the business object based on the node feature of the node and the second feature relationship graph.

9. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the business demand forecasting method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the business demand forecasting method as described in any one of claims 1 to 7.

11. A computer program product, characterized in that The computer program product comprises a non-transitory computer-readable storage medium storing a computer program, wherein the computer program is operable to cause a computer to execute part or all of the steps in the business demand forecasting method according to any one of claims 1 to 7.

Images