Information processing method and device, computer equipment and storage medium

Abstract

The invention relates to an information processing method and device, computer equipment, a storage medium and a computer program product. The method can be applied to scenes such as cloud technology, artificial intelligence and smart traffic, and comprises the steps of obtaining background information of a target product, and interaction features and information features of the target product in a target time period; generating an interaction feature hypergraph based on the interaction features and the background information, and generating an information feature hypergraph based on the information features and the background information; performing attention processing on each node belonging to the same hyperedge in the interaction feature hypergraph to obtain an updated interaction feature, and performing attention processing on each node belonging to the same hyperedge in the information feature hypergraph to obtain an updated information feature; performing evolution enhancement processing on the updated information features to obtain enhanced information features; and performing information processing on the target product based on the updated interaction feature and the enhanced information feature to obtain trend information of the target product. By adopting the method, the information processing accuracy and efficiency can be improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to an information processing method, apparatus, computer device and storage medium. Background Art

[0002] Product information processing refers to the analysis, processing and application of various information obtained and organized for a specific product to achieve a predetermined purpose. By processing and analyzing product information, decisions can be made more quickly and accurately.

[0003] In existing solutions, product information processing is usually performed by users based on their own processing experience. When there are many products, users are affected by subjective factors and personal abilities, which may lead to low accuracy and efficiency of product information processing. Summary of the invention

[0004] Based on this, it is necessary to provide an information processing method, apparatus, computer device and storage medium that can improve the accuracy and efficiency of information processing in response to the above technical problems.

[0005] In a first aspect, the present application provides an information processing method. The method comprises:

[0006] Obtaining background information of a target product, interaction features and information features of the target product in a target period;

[0007] generating an interaction feature hypergraph based on the interaction feature and the background information, and generating an information feature hypergraph based on the information feature and the background information;

[0008] Performing attention processing on each node belonging to the same hyperedge in the interaction feature hypergraph to obtain updated interaction features, and performing attention processing on each node belonging to the same hyperedge in the information feature hypergraph to obtain updated information features;

[0009] The updated information features are subjected to evolutionary enhancement processing to obtain enhanced information features; the enhanced information features and the updated information features are used to perform information processing on the target product.

[0010] In a second aspect, the present application also provides an information processing device. The device comprises:

[0011] A feature acquisition module, used to acquire background information of a target product, interaction features and information features of the target product in a target period;

[0012] A hypergraph construction module, used to generate an interaction feature hypergraph based on the interaction feature and the background information, and to generate an information feature hypergraph based on the information feature and the background information;

[0013] A feature updating module, configured to perform attention processing on each node belonging to the same hyperedge in the interaction feature hypergraph to obtain updated interaction features, and to perform attention processing on each node belonging to the same hyperedge in the information feature hypergraph to obtain updated information features;

[0014] The feature enhancement module is used to perform evolutionary enhancement processing on the updated information features to obtain enhanced information features; the enhanced information features and the updated information features are used to perform information processing on the target product.

[0015] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0016] Obtaining background information of a target product, interaction features and information features of the target product in a target period;

[0017] generating an interaction feature hypergraph based on the interaction feature and the background information, and generating an information feature hypergraph based on the information feature and the background information;

[0018] Performing attention processing on each node belonging to the same hyperedge in the interaction feature hypergraph to obtain updated interaction features, and performing attention processing on each node belonging to the same hyperedge in the information feature hypergraph to obtain updated information features;

[0019] The updated information features are subjected to evolutionary enhancement processing to obtain enhanced information features; the enhanced information features and the updated information features are used to perform information processing on the target product.

[0020] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0021] Obtaining background information of a target product, interaction features and information features of the target product in a target period;

[0022] generating an interaction feature hypergraph based on the interaction feature and the background information, and generating an information feature hypergraph based on the information feature and the background information;

[0023] Performing attention processing on each node belonging to the same hyperedge in the interaction feature hypergraph to obtain updated interaction features, and performing attention processing on each node belonging to the same hyperedge in the information feature hypergraph to obtain updated information features;

[0024] The updated information features are subjected to evolutionary enhancement processing to obtain enhanced information features; the enhanced information features and the updated information features are used to perform information processing on the target product.

[0025] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0026] Obtaining background information of a target product, interaction features and information features of the target product in a target period;

[0027] generating an interaction feature hypergraph based on the interaction feature and the background information, and generating an information feature hypergraph based on the information feature and the background information;

[0028] Performing attention processing on each node belonging to the same hyperedge in the interaction feature hypergraph to obtain updated interaction features, and performing attention processing on each node belonging to the same hyperedge in the information feature hypergraph to obtain updated information features;

[0029] The updated information features are subjected to evolutionary enhancement processing to obtain enhanced information features; the enhanced information features and the updated information features are used to perform information processing on the target product.

[0030] The above-mentioned information processing method, apparatus, computer equipment, storage medium and computer program product can obtain the background information of the target product, the interaction characteristics and information characteristics of the target product in the target time period, thereby combining the background of the product, historical interaction characteristics and information affecting the interaction, etc., to comprehensively analyze the information of the target product and improve the accuracy of the information processing of the target product; and by constructing a hypergraph, related features can be aggregated together, so that the complex relationship between the target products can be quickly and accurately captured based on the hypergraph to update the features, thereby improving the accuracy and efficiency of the information processing of the target product; in the hypergraph structure, the features represented by some nodes may be more informative than other nodes Through attention processing, the model can pay more attention to these key features, thereby improving the accuracy of information processing of the target product; by performing evolutionary enhancement processing on the updated information features, the enhanced information features can carry more contextual information. The contextual information contains the evolutionary information of the event, which is more conducive to capturing long-term trends and understanding the impact of emergencies on the information of the target product, thereby further improving the accuracy of information processing of the target product; by comprehensively considering the interactive features and the enhanced information features to process the information of the target product, the accuracy of information processing of the target product can be further improved; and no human intervention is required in the entire prediction process, thereby improving the accuracy and efficiency of information processing of the target product. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 An application environment diagram of an information processing method in an embodiment;

[0032] Figure 2 is a flowchart of an information processing method in an embodiment;

[0033] Figure 3 A schematic diagram of a hypergraph in one embodiment;

[0034] Figure 4 A schematic diagram of an interactive feature hypergraph in one embodiment;

[0035] Figure 5 A schematic diagram of an information feature hypergraph in one embodiment;

[0036] Figure 6 is a schematic diagram of a first node in an embodiment;

[0037] Figure 7 is a schematic diagram of a second node in an embodiment;

[0038] Figure 8 A schematic diagram of a matter map in one embodiment;

[0039] Fig. 9 A schematic diagram of an information evolution diagram in one embodiment;

[0040] Fig.10 is a flowchart of an information processing method in another embodiment;

[0041] Fig.11 A schematic diagram of a product selection page in an embodiment;

[0042] Fig.12 A schematic diagram of a prediction result display page in an embodiment;

[0043] Fig.13 A schematic diagram of a trend prediction model processing framework in one embodiment;

[0044] Fig.14 A schematic diagram of a processing framework of a hypergraph attention module in one embodiment;

[0045] Fig.15 A schematic diagram of a processing framework of a matter graph enhancement module in one embodiment;

[0046] Fig.16 A schematic diagram of a fusion and update module processing framework in one embodiment;

[0047] Fig.17 A schematic diagram of a model test result in one embodiment;

[0048] Fig.18 A schematic diagram of a model test result in another embodiment;

[0049] Fig.19 is a structural block diagram of an information processing device in one embodiment;

[0050] Fig. 20 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0052] The information processing method provided in this application involves artificial intelligence machine learning, natural language processing and other technologies, among which:

[0053] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making.

[0054] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, as well as machine learning / deep learning, autonomous driving, smart transportation and other major directions.

[0055] Machine Learning (ML) is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0056] Natural language processing (NLP) is an important direction in the fields of computer science and artificial intelligence. It studies various theories and methods that can achieve effective communication between people and computers using natural language. Natural language processing is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field will involve natural language, that is, the language people use in daily life, so it is closely related to the study of linguistics. Natural language processing technology usually includes text processing, semantic understanding, machine translation, robot question answering, knowledge graph and other technologies.

[0057] The information processing method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other servers. The information processing method can be executed by the terminal 102 or the server 104 alone, or by the terminal 102 and the server 104 in collaboration. In some embodiments, the information processing method is executed by the terminal 102, and the terminal 102 obtains the background information of the target product, the interaction features and information features of the target product in the target time period; generates an interaction feature hypergraph based on the interaction features and the background information, and generates an information feature hypergraph based on the information features and the background information; performs attention processing on each node belonging to the same hyperedge in the interaction feature hypergraph to obtain an updated interaction feature, and performs attention processing on each node belonging to the same hyperedge in the information feature hypergraph to obtain an updated information feature; performs evolution enhancement processing on the updated information feature to obtain an enhanced information feature; the enhanced information feature and the updated information feature are used for information processing of the target product.

[0058] Among them, the terminal 102 can be, but is not limited to, various desktop computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server 104 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The terminal 102 and the server 104 can be directly or indirectly connected via wired or wireless communications, and this application does not limit this.

[0059] In one embodiment, Figure 2 As shown, an information processing method is provided, which is applied to Figure 1 The computer device in the example is used to illustrate, including the following steps:

[0060] S202, obtaining background information of the target product, interaction features and information features of the target product in the target time period.

[0061] The target product refers to the product used for interaction, which can be a financial product, commodity or other item. The financial product can be stocks, bonds, funds, cryptocurrencies, etc., and the commodity can be agricultural products, energy, metals, etc. Other items can be artworks, collectibles, real estate or other items that can be bought and sold. There can be multiple target products, such as stock A, stock B and stock C; or the target product can be model A mobile phone, model B mobile phone and model C mobile phone; or the target product can be book A, clothing B and food C, etc.

[0062] Background information may be various information related to the target product, including but not limited to industry information, regional information, and product audience information involved in the target product. Industry information refers to the type of industry involved in the target product. This information can affect the overall situation of the industry and thus affect the trend of the target product. Regional information refers to the geographical area where the target product is located. Economic, cultural, policy and other factors in different regions will also have an impact on the trend of the target product. Product audience information refers to the characteristics and behaviors of the target consumer group of the target product, which may specifically be information such as age, gender, educational background or consumer labels. This information affects the classification or recommendation of the target product. For example, a product A can be divided into children's clothing according to the age of the audience, men's clothing according to gender, and peripherals of a TV series or movie according to the theme.

[0063] For example, the target product is a certain stock, the stock code of the target product is XYZ, and the industry it belongs to is the technology industry. The stock represents a technology company that mainly engages in software development and Internet services. The company operates globally, but its headquarters is located in Region 1. From the background information, it can be obtained that the industry type of the target product is the technology industry, and the region it belongs to is Region 1; the target product is a mobile phone of model ABC, and the industry it belongs to is the consumer electronics industry. Mobile phone model ABC is a smart phone produced by a well-known mobile phone manufacturer, and the headquarters of the well-known mobile phone manufacturer is in Region 2. From the background information, it can be obtained that the industry type of the target product is the consumer electronics industry, and the region it belongs to is Region 2; the target product is a certain newly listed or to-be-listed product, and the basic description of the product gives the title, description, picture, price, etc. of the product. From these basic descriptions, the product audience information of the target product can be determined.

[0064] The target period refers to a period of time in the past, which can also be called a historical period. It can be a period of time of any length. For example, the target period can be the past few days, weeks, months or years. The target period can include multiple sub-periods. For example, the target period is the past week and the sub-periods are daily.

[0065] The interactive features refer to the features of the interactive activities related to the target product corresponding to each sub-period of the target time period. Specifically, they can be features related to click-through rate, purchase rate, browsing time, comment rate, etc., or they can be features related to transaction quantity, transaction price, transaction amount, etc. For example, if the target time period is the past week and the sub-period can be daily, the interactive features of the target product obtained refer to the interactive features of the target product every day in the past week.

[0066] Information features refer to the features of social activities related to the target product corresponding to each sub-period of the target time period, which can be news, reports, and comments. For example, media reports on the trend of consumers turning to more environmentally friendly and sustainable consumer goods have led businesses to adjust their product lines and add more "green" or "eco-friendly" categories. The target period is the past week, and the sub-period can be daily. The interaction features of the target product obtained refer to the information features of the target product every day in the past week.

[0067] Specifically, the computer device can obtain the background information of the target product, the interaction data and information information of the target product in the target time period from the public data source, and perform feature processing on the interaction data and information information to obtain the interaction features and information features of the target product in the target time period.

[0068] For example, the industry type of the target product can be obtained from public business databases, company official websites, industry reports and other data sources, the interaction data of the target product in the target period can be obtained from the interactive platform, and the news information of the target product in the target period can be obtained from news websites or social media platforms, etc. The interactive platform can be an e-commerce trading platform or a financial trading platform.

[0069] In one embodiment, after obtaining the interaction data and information information of the target product in the target time period, the computer device can input the interaction data and information information into a pre-trained machine learning model, and use the feature extraction network of the machine learning model to extract features from the interaction data and information information respectively to obtain the interaction features and information features of the target product in the target time period.

[0070] S204, generating an interaction feature hypergraph based on the interaction feature and the background information, and generating an information feature hypergraph based on the information feature and the background information.

[0071] Among them, the interactive feature hypergraph is used to represent the interactive features of different target products in the corresponding sub-period, and at the same time, it shows the correlation between the interactive features of different target products. The information feature hypergraph is used to represent the information features of different target products in the corresponding sub-period, and at the same time, it shows the correlation between the information features of different target products.

[0072] It should be noted that hypergraph is an extended concept in graph theory, which is more general and flexible than traditional graph. In traditional graph, an edge connects two nodes, while in hypergraph, an edge can connect multiple nodes, and this connection is called a hyperedge. Figure 3 The figure shows a schematic diagram of a hypergraph, in which node 1 and node 3 are on hyperedge 1 (solid area), and node 2, node 4 and node 5 are on hyperedge 2 (dashed area), that is, node 1 and node 3 are connected by hyperedge 1, and node 2, node 4 and node 5 are connected by hyperedge 2.

[0073] Specifically, for any sub-period in the target time period, the computer device generates an interaction feature hypergraph corresponding to the sub-period based on the interaction features and background information of each target product corresponding to the sub-period; and generates an information feature hypergraph corresponding to the sub-period based on the information features and background information of each target product corresponding to the sub-period.

[0074] For example, multiple target products are stock A, stock B, stock C, stock D, stock E, stock F, and their respective background information is obtained. The target period is the past week, and the current processing sub-period is the first day of this week. Then the interaction features of stock A, stock B, stock C, stock D, stock E, and stock F on the first day are obtained, and the interaction features AJ1, interaction feature BJ1, interaction feature CJ1, interaction feature DJ1, interaction feature EJ1, and interaction feature FJ1 are obtained. Then based on the background information and the interaction features AJ1, interaction feature BJ1, interaction feature CJ1, interaction feature DJ1, interaction feature EJ1, and interaction feature F J1 generates an interactive feature hypergraph corresponding to the first day; obtains the information features of stock A, stock B, stock C, stock D, stock E, and stock F on the first day, and obtains information features AZ1, information features BZ1, information features CZ1, information features DZ1, information features EZ1, and information features FZ1. Then, based on the background information and information features AZ1, information features BZ1, information features CZ1, information features DZ1, information features EZ1, and information features FZ1, an information feature hypergraph corresponding to the first day is generated. The above processing is also performed on the interactive features and information features of different target products on other days of the past week, so that the interactive feature hypergraph and information feature hypergraph corresponding to each day of the past week can be obtained, such as Figure 4 The interactive feature hypergraph of a day in one embodiment is shown. Figure 4 In the middle, hyperedge 1 connects the nodes corresponding to the interaction features AJ1, BJ1, and FJ1, and hyperedge 2 connects the nodes corresponding to the interaction features CJ1, DJ1, and EJ1; Figure 5 The figure shows an information feature hypergraph of a certain day in one embodiment. Figure 5 Hyperedge 1 connects the nodes corresponding to information feature AZ1, information feature BZ1, information feature CZ1 and information feature DZ1, and hyperedge 2 connects the nodes corresponding to information feature EZ1 and information feature FZ1.

[0075] S206, performing attention processing on each node belonging to the same hyperedge in the interaction feature hypergraph to obtain updated interaction features, and performing attention processing on each node belonging to the same hyperedge in the information feature hypergraph to obtain updated information features.

[0076] Among them, the attention mechanism is used to assign weights to different nodes belonging to the same hyperedge, so as to weight the nodes belonging to the same hyperedge and update the features represented by the nodes.

[0077] Specifically, for any hyperedge of the interaction feature hypergraph corresponding to any sub-time period, attention processing is performed on each node on the hyperedge to obtain an attention processing result, and the interaction feature represented by the corresponding node is updated according to the attention processing result to obtain an updated interaction feature; for any hyperedge of the information feature hypergraph corresponding to any sub-time period, attention processing is performed on each node on the hyperedge to obtain an attention processing result, and the information feature represented by the corresponding node is updated according to the attention processing result to obtain an updated information feature.

[0078] For example, the hyperedges corresponding to industry 1 in the information feature hypergraph connect nodes 1, 2, and 3 respectively, where the vector of the information feature of stock A corresponding to node 1 is expressed as [0.992, 0.469, 0.636], the vector of the information feature of stock B corresponding to node 2 is expressed as [0.844, 0.831, 0.707], and the vector of the information feature of stock C corresponding to node 3 is expressed as [0.435, 0.342, 0.067]. Attention processing is performed on nodes 1, 2, and 3 of the hyperedge respectively, and the vector of the updated information feature of stock A corresponding to node 1 is expressed as [2.623, 1.605, 1.272], the vector of the updated information feature of stock B corresponding to node 2 is expressed as [2.429, 1.641, 1.281], and the vector of the updated information feature of stock C corresponding to node 3 is expressed as [1.572, 1.072, 0.785].

[0079] In one embodiment, a computer device may input the interaction features and information features of a target product in a target period outputted by a feature extraction network of a machine learning model into a hypergraph attention processing network of a machine learning model, and generate an interaction feature hypergraph based on the interaction features and background information through the hypergraph attention network, and generate an information feature hypergraph based on the information features and background information, perform attention processing on each node belonging to the same hyperedge in the interaction feature hypergraph to obtain updated interaction features, and perform attention processing on each node belonging to the same hyperedge in the information feature hypergraph to obtain updated information features.

[0080] S208, performing evolutionary enhancement processing on the updated information features to obtain enhanced information features; the enhanced information features and the updated information features are used to perform information processing on the target product.

[0081] Among them, evolutionary enhancement processing refers to the use of external knowledge, relationships or historical information in the model to further process the features, thereby improving the expressiveness of the features and further enhancing the performance and effect of the model. In the embodiment of the present application, evolutionary enhancement processing is used to enrich the updated information features so that it can better capture the evolution and correlation of the information. Information processing can specifically be classification processing, recommendation processing or trend prediction processing.

[0082] Specifically, after obtaining the updated information features, the computer device obtains an external knowledge base containing information about events, relationships, causal relationships, etc., and connects the updated information features with the event nodes in the external knowledge base, and performs evolutionary enhancement processing on the updated information features based on the connected event nodes to obtain enhanced information features, and performs information processing on the target product based on the enhanced information features and the updated information features.

[0083] In one embodiment, after the computer device outputs the updated information features through the hypergraph attention network of the machine learning model, the updated information features are input into the evolution enhancement network, and the updated information features are subjected to evolution enhancement processing through the evolution enhancement network to obtain enhanced information features, and information processing is performed on the target product based on the enhanced information features and the updated information features.

[0084] In one embodiment, after obtaining the updated interaction features and the enhanced information features, the computer device may perform trend prediction on the target product based on the updated interaction features and the enhanced information features to obtain trend information of the target product.

[0085] Among them, trend prediction refers to the prediction of the interaction data of the target product in the future, which can be the prediction of transaction price, transaction volume, etc. Trend information refers to the result of the prediction, which can be rising, falling, or equal status.

[0086] Specifically, after obtaining the updated interactive features and the enhanced information features, the computer device can fuse the updated interactive features and the enhanced information features to obtain fused features, and perform trend prediction on the target product based on the fused features to obtain trend information of the target product. For example, if the trend information is an upward trend, it means that the target product may increase in price or have an increase in sales volume in the future; if the trend information is a downward trend, it means that the price may drop or the sales volume may decrease; if the trend information is a flat trend, it means that the target product may remain basically unchanged in price or sales volume in the future.

[0087] In one embodiment, a computer device may input updated interactive features output by a hypergraph attention network of a machine learning model and enhanced information features output by an evolutionary enhancement network of the machine learning model into a feature fusion network of the machine learning model to obtain fused features, and input the fused features into a trend prediction network, and perform trend prediction on a target product based on the fused features through the trend prediction network to obtain trend information of the target product.

[0088] In one embodiment, the computer device can also classify the target product based on the fusion feature. Specifically, the computer device can input the fusion feature into a pre-trained classification model, and classify and predict the input fusion feature through the classification model to obtain the classification category of the target product. The classification model can be a decision tree, SVM (support vector machine), KNN (K nearest neighbor), deep learning network or other models.

[0089] In one embodiment, the computer device may also perform a recommendation process on the target product based on the fusion feature. Specifically, the computer device may input the fusion feature and the object feature of the object to be recommended into a pre-trained recommendation model, process the input fusion feature and the object feature through the recommendation model, obtain a recommendation score for the target product, and select a product from the target product based on the recommendation score to be finally recommended to the object to be recommended. The recommendation model may be based on matrix decomposition, deep learning (such as a neural network) or other recommendation algorithms.

[0090] The above information processing method can comprehensively analyze the information of the target product by acquiring the background information of the target product, the interaction characteristics and information characteristics of the target product in the target period, thereby combining the background of the product, historical interaction characteristics and information affecting the interaction, and improve the accuracy of the information processing of the target product; and by constructing a hypergraph, the related features can be aggregated together, so that the complex relationship between the target products can be quickly and accurately captured based on the hypergraph to update the characteristics, thereby improving the accuracy and efficiency of the information processing of the target product; in the hypergraph structure, the characteristics represented by some nodes may be more informative than other nodes. Through attention processing, the model can pay more attention to these key characteristics, thereby improving the accuracy of the information processing of the target product; by performing evolutionary enhancement processing on the updated information characteristics, the enhanced information characteristics can carry more context information, and the context information contains the evolutionary information of the event, which is more conducive to capturing long-term trends and understanding the impact of sudden events on the information of the target product, thereby further improving the accuracy of the information processing of the target product; by comprehensively considering the interaction characteristics and the enhanced information characteristics to process the information of the target product, the accuracy of the information processing of the target product can be further improved; and no human intervention is required in the entire prediction process, thereby improving the accuracy and efficiency of the information processing of the target product.

[0091] In one embodiment, the process by which a computer device obtains background information of a target product, interaction features and information features of the target product in a target time period includes the following steps: obtaining background information of the target product; obtaining interaction data and information information of the target product in a target time period; performing time series feature processing on the interaction data to obtain interaction features; and performing text feature processing on the information information to obtain information features.

[0092] Among them, the interaction data in the target period refers to the interaction data corresponding to each sub-period of the target period, and the interaction data in the target period belongs to the time series number; the information information in the target period refers to the information information corresponding to each sub-period, and the information information in the target period belongs to text data.

[0093] Time series feature processing refers to the processing of time series data to extract information about time in order to better capture the time series patterns and trends in the sequence. Text feature processing refers to the process of converting text data into numerical features that can be used by machine learning models.

[0094] Specifically, the computer device obtains the interaction data and information corresponding to each sub-period of the target time period of the target product, and performs time series feature processing on the interaction data corresponding to each sub-period in sequence to obtain the interaction feature corresponding to each sub-period, and performs text feature processing on the information information corresponding to each sub-period to obtain the information feature corresponding to each sub-period.

[0095] In one embodiment, the computer device can input the interaction data corresponding to each sub-period into the GRU network branch of the feature extraction network, perform feature extraction on the interaction data through the GRU (Gated Recurrent Unit) network branch, and obtain the interaction features corresponding to each sub-period; input the information corresponding to each sub-period into the TextCNN (Convolutional Neural Network for Text Classification) network branch of the feature extraction network, perform feature extraction on the information information through the TextCNN network, and obtain the interaction features corresponding to each sub-period.

[0096] Among them, GRU is a recurrent neural network used to capture dependencies and long-term trends in time series, and TextCNN is a convolutional neural network used to capture local features and semantic information in text.

[0097] Specifically, each GRU layer of the GRU network branch processes the input interaction data in turn to obtain the interaction features corresponding to each sub-period; the convolution layer and pooling layer of the TextCNN network processes the input information in turn to obtain the interaction features corresponding to each sub-period.

[0098] In one embodiment, when there are at least two pieces of information corresponding to a sub-period of a target product, feature processing can be performed on each piece of information to obtain at least two information features, and the average feature of the at least two information features can be determined, and the average feature can be determined as the information feature corresponding to the sub-period.

[0099] For example, the information is news, and there are two news about stock A on a certain day, where news 1 is "Company A released its annual financial report, with a substantial increase in profits", and news 2 is "Company A launched a new product". Then, feature processing is performed on news 1 and news 2 respectively to obtain news feature 1 and news feature 2. News feature 3 is obtained by averaging news feature 1 and news feature 2, and news feature 3 is determined as the news feature of stock A on that day.

[0100] In the above embodiment, the computer device can better determine the changes and trends of the target product by combining information of different dimensions such as the background information of the target product, the interaction data and news information in the target time period; time series feature processing and text feature processing can extract meaningful features that can be used for analysis and modeling from the original data. These features can capture the change patterns and trends of the target product in different dimensions, thereby better determining the changes and trends of the target product.

[0101] In one embodiment, the background information includes regional information, and the process of a computer device generating an interactive feature hypergraph based on the interactive features and the background information includes the following steps: taking the interactive features as the first nodes, determining the region corresponding to each first node according to the regional information; dividing each first node into sets according to the region to obtain a corresponding regional node set; connecting the first nodes in each regional node set to obtain an interactive feature hypergraph.

[0102] Specifically, for any sub-time period, the interaction characteristics of different target products in the sub-time period are taken as the first node, and the region corresponding to each target product is determined according to the regional information. The region corresponding to the target product is the region corresponding to the corresponding first node, and the first nodes belonging to the same region are divided into the same set to obtain the regional node set corresponding to each region. For any regional node set, the first nodes in the regional node set are connected to obtain the interaction feature hypergraph corresponding to the sub-time period.

[0103] For example, multiple target products are stock A, stock B, stock C, stock D, stock E, stock F, the target period is the past week, and the current processing sub-period is the first day of this week. Then the interaction features of each stock on the first day are obtained, such as interaction feature AJ1, interaction feature BJ1, interaction feature CJ1, interaction feature DJ1, interaction feature EJ1, and interaction feature FJ1. Each interaction feature is the first node, such as Figure 6 As shown, nodes AJ1, BJ1, CJ1, DJ1, EJ1, and FJ1 are obtained. According to the regional information of each stock, it is determined that stocks C, D, and E all belong to region 1, and stocks A, B, and F belong to region 2. Then, the first nodes are divided into sets according to the region to obtain regional node set 1 and regional node set 2, and nodes CJ1, DJ1, and EJ1 in regional node set 1 are connected, and nodes AJ1, BJ1, and FJ1 in regional node set 2 are connected, as shown in FIG. Figure 4 As shown, the interaction feature hypergraph corresponding to the first day of the week is obtained, in which nodes CJ1, DJ1 and EJ1 are on hyperedge J1, and nodes AJ1, BJ1 and FJ1 are on hyperedge J2. The above method can be used to perform similar processing on the interaction features of other days of the week, thereby obtaining the interaction feature hypergraph of each day of the week.

[0104] In the above embodiment, the computer device can capture the differences in interaction features between different regions by taking the interaction feature as the first node and determining the region corresponding to each first node according to the regional information. Regional information is related to interaction features in many cases. For example, factors such as the economic level and market demand of different regions will affect the interaction behavior. Therefore, associating interaction features with regions helps to extract more meaningful information; each first node is divided into sets according to the region to obtain a corresponding regional node set; the first nodes in each regional node set are connected to obtain an interaction feature hypergraph. This structured expression can better reflect the relationship between interaction features between different regions and within the same region, thereby providing richer information for subsequent prediction work.

[0105] In one embodiment, the background information includes industry information, and the process of a computer device generating an information feature hypergraph based on information features and background information includes the following steps: taking the information features as the second nodes, determining the industry corresponding to each second node according to the industry information; dividing the second nodes into sets according to the industry to obtain corresponding industry node sets; connecting the second nodes in each industry node set to obtain an information feature hypergraph.

[0106] Specifically, for any sub-period, the information features of different target products in the sub-period are used as the second node, and the industry corresponding to each target product is determined based on the industry information. The industry corresponding to the target product is the industry corresponding to the corresponding second node, and the second nodes belonging to the same industry are divided into the same set to obtain the industry node set corresponding to each industry. For any industry node set, the second nodes in the industry node set are connected to obtain the information feature hypergraph corresponding to the sub-period.

[0107] For example, the target products are stock A, stock B, stock C, stock D, stock E, and stock F. The target period is the past week. The sub-period currently being processed is the first day of the week. Then, the information features of each stock on the first day are obtained, such as information feature AZ1, information feature BZ1, information feature CZ1, information feature DZ1, information feature EZ1, and information feature FZ1. Each information feature is used as the second node, such as Figure 7 As shown, nodes AZ1, BZ1, CZ1, DZ1, EZ1, and FZ1 are obtained. According to the industry information of each stock, it is determined that stock A, stock B, stock C, and stock D all belong to industry 1, and stock E and stock F belong to industry 2. Then, the second nodes are divided into sets according to the industry to obtain industry node set 1 and industry node set 2, and nodes AZ1, BZ1, CZ1, and DZ1 in industry node set 1 are connected, and nodes EZ1 and FZ1 in industry node set 2 are connected, as shown in FIG. Figure 5 As shown, the information feature hypergraph corresponding to the first day of the week is obtained, in which nodes AZ1, BZ1, CZ1 and DZ1 are on hyperedge Z1, and nodes EZ1 and FZ1 are on hyperedge Z2. The above method can be used to perform similar processing on the information features of other days of the week, thereby obtaining the information feature hypergraph of each day of the week.

[0108] In the above embodiment, the computer device can capture the differences in information features between different industries by taking information features as second nodes and determining the industries corresponding to each second node according to industry information. Information such as news reports and comments in different industries may have different themes, focuses and emotions. Therefore, associating information features with industries helps to extract more meaningful information; the second nodes are divided into sets according to industries to obtain corresponding industry node sets; the second nodes in each industry node set are connected to obtain an information feature hypergraph. This structured expression can better reflect the relationship between information features between different industries and within the same industry, thereby providing richer information for subsequent prediction work.

[0109] In one embodiment, a computer device performs attention processing on each node belonging to the same hyperedge in an interaction feature hypergraph, and the process of obtaining an updated interaction feature specifically includes the following steps: performing attention processing on a first target node on a first hyperedge in the interaction feature hypergraph, and obtaining a first attention weight of other nodes on the first hyperedge; and updating the interaction feature corresponding to the first target node based on the first attention weight, and obtaining an updated interaction feature corresponding to the first target node.

[0110] The first hyperedge is a hyperedge in the interaction feature graph, the first target node is the target node on the first hyperedge currently being processed, and the other nodes are nodes other than the target node in the nodes on the first hyperedge currently being processed. Figure 4 As shown, if the first hyperedge currently being processed is hyperedge 1, the first target node is node AJ1, and the other nodes are node BJ1 and node FJ1; if the first hyperedge currently being processed is hyperedge 2, the first target node is node CJ1, and the other nodes are node DJ1 and node EJ1.

[0111] Specifically, for any first hyperedge in the interaction feature graph, select target nodes, i.e., first target nodes, from the first hyperedge in turn, and determine the similarity between the first target node and other nodes, determine the first attention weights of other nodes based on the similarity, and fuse the interaction features represented by the other nodes with the interaction features represented by the first target node according to the first attention weights of other nodes to obtain attention fusion features, and use the attention fusion features as the updated interaction features corresponding to the first target node.

[0112] The similarity between the first target node and other nodes refers to the similarity between the interaction features represented by the first target node and the interaction features represented by other nodes.

[0113] It can be understood that by performing the above attention processing on each first node on each first hyperedge in the interaction feature graph, it is possible to update the interaction feature represented by each first node and obtain the updated interaction feature of each first node.

[0114] In the above embodiment, the computer device can calculate the correlation weights between different nodes, that is, the degree of mutual influence between nodes, through attention processing, which helps to accurately capture the correlation between interactive feature nodes so as to better transfer and integrate information. Through the attention weights, the interactive features can be weighted and transferred according to the importance of other nodes, thereby fusing the information of other nodes, thereby capturing important features in a more targeted manner, so as to better determine the changes and trends of the target products.

[0115] In one embodiment, a computer device performs evolutionary enhancement processing on the updated information features, and the process of obtaining the enhanced information features includes the following steps: performing attention processing on the second target node on the second hyperedge in the information feature hypergraph to obtain the second attention weight of other nodes on the second hyperedge; updating the information feature corresponding to the second target node based on the second attention weight to obtain the updated information feature corresponding to the second target node.

[0116] The second hyperedge is a hyperedge in the information feature graph, and the second target node is a node on the second hyperedge currently being processed. Figure 5 As shown, if the second hyperedge currently being processed is hyperedge 1, the second target node is node AZ1, and the other nodes are node BZ1, node CZ1, and node DZ1; if the second hyperedge currently being processed is hyperedge 2, the second target node is node EZ1, and the other node is node FZ1.

[0117] Specifically, for any second hyperedge in the information feature graph, select target nodes, i.e., second target nodes, from the second hyperedge in turn, and determine the similarity between the second target node and other nodes, determine the second attention weights of other nodes based on the similarity, and fuse the information features represented by the other nodes with the information features represented by the second target node according to the second attention weights of other nodes to obtain an attention fusion feature, and use the attention fusion feature as the updated information feature corresponding to the second target node.

[0118] The similarity between the second target node and other nodes refers to the similarity between the information features represented by the second target node and the information features represented by other nodes.

[0119] It can be understood that by performing the above attention processing on each second node on each second hyperedge in the information feature graph, it is possible to update the information features represented by each second node and obtain the updated information features of each second node.

[0120] In the above embodiment, the computer device can calculate the correlation weights between different nodes through attention processing, that is, the degree of mutual influence between nodes, which helps to accurately capture the correlation between information feature nodes so as to better transmit and integrate information. Through the attention weights, information features can be weighted and transmitted according to the importance of other nodes, thereby integrating the information of other nodes, thereby capturing important features in a more targeted manner, so as to better determine the changes and trends of target products.

[0121] In one embodiment, the similarity between the target node and other nodes on a hyperedge satisfies the following relationship:

[0122]

[0123] in, Represents the similarity between the target node s and other nodes k, node s Indicates the features corresponding to the target node s, node k represents the features corresponding to other nodes k, a is the function used for similarity calculation, and δ is the activation function.

[0124] In one embodiment, the weights of other nodes on a hyperedge relative to the target node satisfy the following relationship:

[0125]

[0126] Among them, α s,k represents the weight of other nodes k relative to the target node s, represents the similarity between the target node s and other nodes k, It represents the similarity between the target node s and other nodes h, and H represents that the target node s has a total of H other nodes on this hyperedge.

[0127] In one embodiment, the updated features of the target node on a hyperedge and other nodes satisfy the following relationship:

[0128]

[0129] in, Represents the updated features corresponding to the target node s, node k represents the features corresponding to other nodes k, α a,k Represents the weight of other nodes k relative to the target node s, node s Represents the features corresponding to the target node s.

[0130] It can be understood that the above relationships (1) to (3) can be used to perform attention processing on each node belonging to the same hyperedge in the interaction feature hypergraph to obtain updated interaction features, and can also be used to perform attention processing on each node belonging to the same hyperedge in the information feature hypergraph to obtain updated information features. When used to perform attention processing on each node belonging to the same hyperedge in the interaction feature hypergraph, the features represented by each node are interaction features, and when used to perform attention processing on each node belonging to the same hyperedge in the information feature hypergraph, the features represented by each node are information features.

[0131] In one embodiment, a computer device performs evolutionary enhancement processing on updated information features, and the process of obtaining enhanced information features includes the following steps: obtaining a pre-constructed matter graph; generating an information evolution graph based on the matter graph and updated information features; performing convolution processing on the information evolution graph based on a machine learning model to obtain a convolution result; and determining enhanced information features based on the convolution result.

[0132] Among them, the event graph is a knowledge graph used to represent the causal relationship, regular association and evolution process between events. The event graph in the embodiment of the present application can be constructed manually or automatically extracted from large-scale text data through natural language processing technology. Figure 8 The figure shows a schematic diagram of a causal graph. The causal graph is a directed graph. Each node represents an event. The nodes are connected by directed edges. The direction of the arrow represents the evolution direction of the event. The value on the edge represents the evolution probability. For example, node 0 represents an equity transfer event, and node 15 represents a policy support event. The value on the directed edge from node 0 to node 15 is 0.112, which means that the probability of the equity transfer event evolving into a policy support event is 0.112. The value on the directed edge from node 15 to node 0 is 0.136, which means that the probability of the policy support event evolving into an equity transfer event is 0.136.

[0133] The information evolution graph is a graph structure used to represent the evolutionary relationship and causal relationship of information features over time. The machine learning model can be a graph convolutional network (GCN), which is used to propagate and aggregate the features represented by the nodes in the graph, thereby capturing the correlation and information transfer between nodes.

[0134] Specifically, after obtaining the updated information features, the computer device obtains a pre-constructed causal graph, wherein the nodes in the causal graph represent abstract events, and the edges represent the causal relationship or connection between events. Using the causal graph as a guide, the updated information features are combined with the event relationships in the causal graph to generate an information evolution graph, and the generated information evolution graph is convoluted using a pre-trained machine learning model. Through the convolution processing, the features represented by the nodes can be transmitted to neighboring nodes, thereby achieving evolutionary enhancement of the updated information features represented by each node and obtaining enhanced information features.

[0135] In the above embodiment, the computer device can help the model capture the intrinsic relationship and evolution pattern between different information features by combining external knowledge; by performing convolution processing on the information evolution graph through the machine learning model, important evolution patterns and features can be automatically learned, and the strengthened information features can better capture the correlation and evolution information of information features, thereby better determining the changes and trends of target products.

[0136] In one embodiment, the process of a computer device generating an information evolution graph based on a cause-and-effect graph and updated information features includes the following steps: determining the evolution probability of the updated information features based on the cause-and-effect graph; taking the updated information features as the third node, connecting the third nodes based on the evolution probability of the updated information features to obtain an information evolution graph.

[0137] The evolution probability refers to the probability of evolution between updated information features corresponding to different sub-periods of the same target product, for example, the probability of evolution between updated information features of stock A every day in the past week. The information evolution graph is a directed graph, in which each information feature is a node and each evolution probability is a directed edge.

[0138] Specifically, for any one of the multiple target products, after obtaining the updated information features corresponding to each sub-period within the target time period of the target product, the computer device determines the evolution probability between any two updated information features based on the event graph, and uses the updated information feature as the third node, connects the third node based on the determined evolution probability, and obtains the information evolution graph corresponding to the target product.

[0139] It can be understood that by performing the above processing on the updated information features corresponding to each sub-period of each target product within the target period, an information evolution graph corresponding to each target product can be obtained.

[0140] Taking stock A as an example, the updated information features of stock A in the past week are information feature AZ1, information feature AZ2, information feature AZ3, information feature AZ4, information feature AZ5, information feature AZ6 and information feature AZ7, and the evolution probability is: AZ1->AZ3: 0.18, AZ2->AZ3: 0.26, AZ3->AZ4: 0.17, AZ4->AZ7: 0.22, AZ5->AZ7: 0.15, AZ6->AZ7: 0.35. Each updated information feature is used as the third node, and the corresponding information evolution diagram is generated based on the evolution probability of the updated information feature. Fig. 9 shown.

[0141] In the above embodiment, the computer device can better determine the evolution process between information features by utilizing the rich events and evolutionary relationships contained in the event graph, thereby constructing an accurate information evolution graph, and further enhancing the information features, so as to better determine the changes and trends of the target product in the future.

[0142] In one embodiment, the process of a computer device determining the evolution probability of an updated information feature based on a causal graph includes the following steps: determining a target event corresponding to the updated information feature in the causal graph; obtaining the evolution probability of the target event in the causal graph; and determining the evolution probability of the target event as the evolution probability of the updated information feature.

[0143] The target event refers to an event that matches the updated information feature, and specifically may be an event that is similar to the updated information feature.

[0144] Specifically, for any target product among multiple target products, after obtaining the updated information feature corresponding to each sub-period within the target time period of the target product, the computer device determines, for any updated information feature, the similarity between the events represented by each node in the event graph and the updated information feature, and determines the event corresponding to the maximum similarity as the target event matching the updated information feature. By performing the above processing on each updated information feature of the target product, the target event corresponding to the updated information feature of each sub-period of the target product can be obtained. After obtaining the target event corresponding to the updated information feature of each sub-period, the evolution probability of any two target events in the example graph is determined, and the evolution probability of the two target events in the example graph is determined as the evolution probability between the corresponding two updated information features.

[0145] It can be understood that if two target events are directly connected in the event graph, the evolution probability corresponding to the connecting edge can be determined as the direct evolution probability of the two target events. If the two target events are indirectly connected in the event graph through other nodes, the evolution probability between the two target events is determined based on the indirect evolution probability on the connection path. Specifically, the sub-evolution probabilities corresponding to each path of the connection path can be determined, and the product of each sub-evolution probability can be determined as the evolution probability between the target events corresponding to the head and tail nodes of the corresponding connection path.

[0146] For example, Figure 8 As shown in the figure, if the target event is the litigation event represented by node 2 and the stock price drop event represented by node 9, Figure 8It can be seen that node 2 and node 9 are not directly connected, but node 2 and node 9 are connected through node 11. Therefore, based on the evolution probability 1 from node 11 to node 2 and the evolution probability 2 from node 9 to node 2, the evolution probability of node 9 to node 2 can be determined. Specifically, the product of evolution probability 1 and evolution probability 2 can be determined, and the product can be determined as the evolution probability of node 9 to node 2.

[0147] In the above embodiment, the computer device can better determine the evolution process between information features by utilizing the rich events and evolutionary relationships contained in the event graph, thereby constructing an accurate information evolution graph.

[0148] In one embodiment, the computer device can also perform splicing processing on the updated interaction features and enhanced information features belonging to the same sub-period in the target time period to obtain the spliced ​​features corresponding to each sub-period; perform temporal attention fusion on the spliced ​​features corresponding to each sub-period to obtain fused features; perform information processing on the target product based on the fused features to obtain processed information of the target product.

[0149] Specifically, for any target product, after obtaining the updated interaction features and enhanced information features corresponding to each sub-period within the target time period, for any sub-period, the updated interaction features and enhanced information features of the sub-period can be spliced ​​in the feature dimension to obtain the spliced ​​features corresponding to the sub-period, and the spliced ​​features corresponding to each sub-period are obtained by performing the above processing on the updated interaction features and enhanced information features corresponding to the sub-period of each sub-period, and attention processing is performed on the spliced ​​features corresponding to each sub-period to obtain fused features, and information processing is performed on the target product based on the fused features to obtain processed information of the target product.

[0150] In one embodiment, the information processing may be a trend prediction processing, and the computer device may input the fused features into a trend prediction network, and the trend prediction network may perform trend prediction on the target product based on the fused features to obtain trend information of the target product.

[0151] In one embodiment, the information processing may be classification processing, and the computer device may input the fusion features into a pre-trained classification model, and perform classification prediction on the input fusion features through the classification model to obtain the classification category of the target product. The classification model may be a decision tree, SVM (support vector machine), KNN (K nearest neighbor), deep learning network or other models.

[0152] In one embodiment, the information processing may be recommendation processing, where the computer device may input the fusion features and the object features of the object to be recommended into a pre-trained recommendation model, process the input fusion features and the object features through the recommendation model, obtain the recommendation score of the target product, and select the product that is finally recommended to the object to be recommended from the target products based on the recommendation score. The recommendation model may be based on matrix decomposition, deep learning (such as neural network) or other recommendation algorithms.

[0153] In the above embodiment, the computer device can integrate information from different data sources by splicing the updated interactive features and the enhanced information features, making full use of both interactive and information information, thereby providing a more comprehensive information basis, which helps to more accurately predict the trend of the target product; time is an important factor affecting trends, and by performing time attention fusion on the spliced ​​features, the model can pay more attention to the differences in importance of different sub-periods, better reflect the trend changes in the time series, and thus more accurately predict the future trend of the target product.

[0154] In one embodiment, a computer device performs temporal attention fusion on the spliced ​​features corresponding to each sub-period, and the process of obtaining the fused features includes the following steps: performing temporal attention processing on the spliced ​​features corresponding to the target sub-period to obtain the third attention weights of other sub-periods; and fusing the spliced ​​features corresponding to each sub-period based on the third attention weight to obtain the fused features.

[0155] The target sub-period may be the sub-period closest to the current time. For example, if the target period is the past week, the target sub-period may be the 7th day of the past week.

[0156] Specifically, for any target product among multiple target products, after obtaining the spliced ​​features corresponding to each sub-period within the target time period of the target product, the sub-period closest to the current time is selected as the target sub-period, and the similarity between the spliced ​​features corresponding to the target sub-period and the spliced ​​features of other sub-periods is determined, and the third attention weights of other sub-periods are determined based on the similarity, and the spliced ​​features of other sub-periods are fused with the spliced ​​features of the target sub-period according to the third attention weights of other sub-periods to obtain a fused feature, which is the fused feature of the target product.

[0157] For example, the target period is the past week, and the target sub-period can be the 7th day of the past week. The similarity between the spliced ​​features of each day in the previous 6 days and the spliced ​​features of the target sub-period is calculated respectively to obtain the similarity corresponding to each day in the previous 6 days. Then, based on the similarity corresponding to each day in the previous 6 days, the third attention weight of each day in the previous 6 days is determined. The spliced ​​features of the previous 6 days are weighted according to the attention weights corresponding to each day in the previous 6 days to obtain the weighted result. The weighted result is fused with the spliced ​​fusion feature of the 7th day to obtain the fusion feature, which is the fusion feature of the target product.

[0158] In the above embodiment, the computer device can use time attention processing to enable the model to pay more attention to important sub-periods when calculating fusion features, that is, those periods that have a significant impact on predicting trends. By adjusting the weights, important sub-periods can be highlighted and minor sub-periods can be weakened, so that changes and trends of target products can be better determined.

[0159] In one embodiment, Fig.10 As shown, an information processing method is also provided, which is applied to Figure 1 The computer device in the example is used to illustrate, including the following steps:

[0160] S1002, obtaining background information of the target product, and obtaining interaction data and news information of the target product in the target time period.

[0161] S1004, performing time series feature processing on the interaction data to obtain interaction features, and performing text feature processing on the information information to obtain information features.

[0162] S1006, generating an interaction feature hypergraph based on the interaction feature and the background information, and generating an information feature hypergraph based on the information feature and the background information.

[0163] S1008, performing attention processing on the first target node on the first hyperedge in the interaction feature hypergraph to obtain the first attention weights of other nodes on the first hyperedge; updating the interaction feature corresponding to the first target node based on the first attention weight to obtain the updated interaction feature corresponding to the first target node.

[0164] S1010, perform attention processing on the second target node on the second hyperedge in the information feature hypergraph to obtain the second attention weights of other nodes on the second hyperedge; update the information feature corresponding to the second target node based on the second attention weight to obtain the updated information feature corresponding to the second target node.

[0165] S1012, obtaining a pre-built event graph; generating an information evolution graph based on the event graph and updated information features.

[0166] S1014, performing convolution processing on the information evolution graph based on the machine learning model to obtain a convolution result; and determining enhanced information features based on the convolution result.

[0167] S1016, fusing the updated interactive features and the enhanced information features to obtain fused features.

[0168] S1018, performing information processing on the target product based on the fusion features to obtain processed information of the target product.

[0169] The present application also provides an application scenario, which can specifically be a stock trend prediction scenario. The user can select the product to be predicted on the terminal. The terminal generates a trend prediction request based on the selected product to be predicted, and sends the trend prediction request to the server. The server obtains the background information of the target product associated with the product to be predicted, the interaction features and information features of the target product in the target time period based on the received trend prediction request; wherein the product to be predicted belongs to a set composed of multiple target products, an interaction feature hypergraph is generated based on the interaction features and background information, and an information feature hypergraph is generated based on the information features and background information; attention processing is performed on each node belonging to the same hyperedge in the interaction feature hypergraph to obtain updated interaction features, and attention processing is performed on each node belonging to the same hyperedge in the information feature hypergraph to obtain updated information features; evolutionary enhancement processing is performed on the updated information features to obtain enhanced information features; trend prediction is performed on the target product based on the updated interaction features and enhanced information features to obtain trend information of the target product, and then the server returns the trend information of the product to be predicted in the target product to the terminal, so that the terminal displays the received trend information. Fig.11 As shown, the user can select the product to be predicted for trend prediction on the terminal product selection page. After completing the selection, the trend prediction control is triggered. The terminal generates a trend prediction request in response to the triggering operation of the trend prediction control, and sends the trend prediction request to the server, so that the server performs trend prediction based on the trend prediction request, and displays the prediction information after receiving the trend prediction request, such as Fig.12 shown.

[0170] The present application also provides an application scenario, which can be a stock trend prediction scenario. The application scenario applies the above information processing method. The trend prediction method is specifically implemented by a pre-trained trend prediction model. The trend prediction model processing framework is as follows: Fig.13 As shown in FIG. 1 , the trend prediction model specifically includes (1) feature extraction module, (2) hypergraph attention module, (3) event graph enhancement module, (4) fusion and update module, and (5) prediction module, where:

[0171] (1) Feature Extraction Module (FEM)

[0172] It is used to receive input data and extract features from the input data to obtain the features of the data for subsequent model processing.

[0173] The specific processing process is as follows: the input data includes the historical price information and financial news information of U stocks within the target period (such as W days) every day. For the historical price information, GRU is used to extract features; for the news information, TextCNN is used to extract features.

[0174] The data flow of this module changes as follows: After passing through this module, the data is converted from initial information to vector features. i represents all the news vectors of the i-th stock within W days, L i represents the historical price vector of the i-th stock within W days, L i ,B i The dimensions are (U, W, batch_size, features_dim), batch_size represents the batch size, and features_dim represents the feature dimension.

[0175] (2) HyperGraph Attention Module (HGAM)

[0176] First, we use the industry relationship in the background relationship to divide the news vectors of the same industry into the same hyperedge to build a news hypergraph. Similarly, we use the regional relationship in the background relationship to divide the price information and build a price hypergraph. Fig.14 As shown, the hypergraph attention is then calculated for each type of hypergraph, the importance of each neighbor node in the same hyperedge to the target node is learned, and the features of each node in the same hyperedge are fused based on the importance of each neighbor node to obtain an updated feature vector of the target node. The updated feature vector can specifically be an updated price vector or an updated news vector. Fig.14 middle Represents the updated news vector.

[0177] The data flow of this module changes as follows: After this module, using the background relationship R N We construct W news hypergraphs with U stock news information. The hypergraph nodes are the news features of the i-th stock on a certain day. NUsed to determine the hyperedge. Similarly, W stock price hypergraphs are also constructed. Then the feature vectors of the nodes in the hypergraph are updated. The updated feature vector can be an updated price vector or an updated news vector. The dimension is (U, W, batch_size, features_dim_hg), and features_dim_gh represents the node feature dimension after the hypergraph attention.

[0178] (3) Event Graph Enhancement Module (EGEM)

[0179] The event graph is used as an access model for external knowledge bases. The nodes in the event graph represent abstract events, which represent the summary of a type of events in the news collection. The edges in the graph represent the causal relationship between abstract events, and the numbers on the edges represent the transition probability. Fig.15 As shown in the figure, the financial news nodes of each stock are classified according to the abstract types in the event graph, and then the edges are formed for the nodes according to the causal relationship between these categories, so as to generate the news evolution graph of each stock. Then, GCN is used to capture the node relationship information in the news evolution graph, and the news features of each node in the news evolution graph are enhanced to obtain the enhanced news features. Fig.15 middle Represents the enhanced news vector.

[0180] The data flow of this module changes as follows: After this module, all nodes in W news hypergraphs are used to construct graphs, and a total of U news evolution graphs are constructed. Each news evolution graph has W nodes, and the relationship between the edges in the graph is guided by the knowledge of the event graph. Finally, the node enhancement is completed through gcn, and the news feature dimension is (U, W, batch_size, features_dim_hg).

[0181] (4) Feature Fusion and Updating Module (FFUM)

[0182] First, the financial news features are integrated with the historical price features, and then the temporal attention mechanism is used to further update the information. The features of the day closest to the predicted trading day are selected as the query vector, and the similarity is used to score. After obtaining the weights, the results of each day are weighted and summed to get the final feature representation.

[0183] The data flow of this module changes as follows: Fig.16As shown in the figure, the financial news features enhanced by the event graph and the price features processed by the hypergraph attention are taken, and the two features are fused one by one through iaff (a fusion algorithm) according to the order of each stock every day. The feature dimension after fusion is (U, W, batch_size, features_dim_hg). After that, the fused features are processed by time attention, and the final feature dimension is (U, batch_size, features_dim_hg).

[0184] (5) Stock Movement forecasting Module (SMPM)

[0185] The output of the previous module is input into the fully connected layer with softmax activation function to obtain the probability distribution of three types of price trends (increasing, flat and falling) of each stock on the predicted day t+W.

[0186] The data flow of this module changes as follows: after full connection, the data dimension becomes (U*batch_size,3).

[0187] It should be noted that the operating environment of the above application scenarios is as follows: the model codes are all written in the pytorch framework, the data input dimension is set to 16, the number of experimental rounds is 600, the initial learning rate is set to 1e-3, the batch size is set to 128, W is taken as 5, 10, and 20 in the experiment, and the RTX A6000 GPU is used for hardware acceleration.

[0188] In addition, other methods were selected to compare with the information processing methods used in the above application scenarios on the same data set to test the performance of each method under different window period lengths. The final results are as follows: Fig.17As shown in the figure, EGHAN represents the information processing method provided in the embodiment of the present application, SVM (Support Vector Machine), LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), HATS (Hierarchical Attention Time-Serie), MAN-SF (a prediction method), STHAN (Spatio-Temporal Hierarchical Attention Network), AD-GAT (Adaptive Graph Attention Network), HAGTAN (Temporal-relational hypergraph tri-attention networks, hyper Figure 3 Attention network) represent different trend prediction methods; acc, P, R, F1 are evaluation indicators accuracy, precision, recall and F1 score respectively; W refers to the target period, from Fig.17 It can be seen that the method provided in this application has achieved better results in various indicators compared with other methods.

[0189] In addition, in order to better verify the impact of each functional module of the model, an ablation experiment was also conducted, that is, some modules of the model were removed and then the experiment was conducted. The results are as follows Fig.18 As shown, Fig.18 In the figure, EGHAN-RN and EGHAN-RS respectively represent not using background relations RN and RS to construct hyperedges, but randomly constructing hyperedges. EGHAN-E represents not using the event graph enhancement module. EGHAN-A represents not using iaff for feature fusion, but directly using the splicing method. It can be seen from the results that after removing different modules, the model performance decreases to varying degrees, which proves the effectiveness of these modules.

[0190] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0191] Based on the same inventive concept, the embodiment of the present application also provides an information processing device for implementing the above-mentioned information processing method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in the one or more information processing device embodiments provided below can refer to the limitations on the information processing method above, and will not be repeated here.

[0192] In one embodiment, Fig.19 As shown, an information processing device is provided, including: a feature acquisition module 1902, a hypergraph construction module 1904, a feature update module 1906, a feature enhancement module 1908 and an information processing module 1910, wherein:

[0193] The feature acquisition module 1902 is used to acquire the background information of the target product, the interaction features and information features of the target product in the target time period.

[0194] The hypergraph construction module 1904 is used to generate an interaction feature hypergraph based on the interaction feature and the background information, and to generate an information feature hypergraph based on the information feature and the background information.

[0195] The feature updating module 1906 is used to perform attention processing on each node belonging to the same hyperedge in the interactive feature hypergraph to obtain updated interactive features, and to perform attention processing on each node belonging to the same hyperedge in the information feature hypergraph to obtain updated information features.

[0196] The feature enhancement module 1908 is used to perform evolutionary enhancement processing on the updated information features to obtain enhanced information features; the enhanced information features and the updated information features are used to perform information processing on the target product.

[0197] In the above embodiment, by obtaining the background information of the target product, the interaction features and information features of the target product in the target time period, the information of the target product can be comprehensively analyzed in combination with the background of the product, the historical interaction features and the information affecting the interaction, thereby improving the accuracy of the information processing of the target product; and, by constructing a hypergraph, the related features can be aggregated together, so that the complex relationship between the target products can be quickly and accurately captured based on the hypergraph to update the features, thereby improving the accuracy and efficiency of the information processing of the target product; in the hypergraph structure, the features represented by some nodes may be more informative than other nodes. Through attention processing, the model can pay more attention to these key features, thereby improving the accuracy of the information processing of the target product; by performing evolutionary enhancement processing on the updated information features, the enhanced information features can carry more context information, and the context information contains the evolutionary information of the event, which is more conducive to capturing long-term trends and understanding the impact of sudden events on the information of the target product, thereby further improving the accuracy of the information processing of the target product; by comprehensively considering the interaction features and the enhanced information features to process the information of the target product, the accuracy of the information processing of the target product can be further improved; and no human intervention is required in the entire prediction process, thereby improving the accuracy and efficiency of the information processing of the target product.

[0198] In one embodiment, the feature acquisition module 1902 is used to: acquire background information of the target product; acquire interaction data and information of the target product in a target time period; perform time series feature processing on the interaction data to obtain interaction features; perform text feature processing on the information information to obtain information features.

[0199] In one embodiment, the background information includes regional information; the hypergraph construction module 1904 is used to: take the interactive feature as the first node, and determine the region corresponding to each first node according to the regional information; divide the first nodes into sets according to the region to obtain the corresponding regional node sets; connect the first nodes in each regional node set to obtain the interactive feature hypergraph.

[0200] In one embodiment, the background information includes industry information; the hypergraph construction module 1904 is used to: use information features as second nodes, and determine the industry corresponding to each second node based on the industry information; divide the second nodes into groups according to the industry to obtain corresponding industry node sets; connect the second nodes in each industry node set to obtain an information feature hypergraph.

[0201] In one embodiment, the feature update module 1906 is used to: perform attention processing on the first target node on the first hyperedge in the interaction feature hypergraph to obtain the first attention weight of other nodes on the first hyperedge; update the interaction feature corresponding to the first target node based on the first attention weight to obtain the updated interaction feature corresponding to the first target node.

[0202] In one embodiment, the feature update module 1906 is used to: perform attention processing on the second target node on the second hyperedge in the information feature hypergraph to obtain the second attention weight of other nodes on the second hyperedge; update the information feature corresponding to the second target node based on the second attention weight to obtain the updated information feature corresponding to the second target node.

[0203] In one embodiment, the feature enhancement module 1908 is used to: obtain a pre-constructed cause-and-effect graph; generate an information evolution graph based on the cause-and-effect graph and updated information features; perform convolution processing on the information evolution graph based on a machine learning model to obtain a convolution result; and determine enhanced information features based on the convolution result.

[0204] In one embodiment, the feature enhancement module 1908 is also used to: determine the evolution probability of the updated information feature based on the event graph; use the updated information feature as the third node, connect the third node based on the evolution probability of the updated information feature, and obtain an information evolution graph.

[0205] In one embodiment, the feature enhancement module 1908 is used to: determine the target event corresponding to the updated information feature in the event graph; obtain the evolution probability of the target event in the event graph; and determine the evolution probability of the target event as the evolution probability of the updated information feature.

[0206] In one embodiment, the device also includes an information processing module 1910, which is used to: splice the updated interaction features and enhanced information features belonging to the same sub-period in the target time period to obtain the spliced ​​features corresponding to each sub-period; perform temporal attention fusion on the spliced ​​features corresponding to each sub-period to obtain fused features; perform information processing on the target product based on the fused features to obtain processed information of the target product.

[0207] In one embodiment, the information processing module 1910 is also used to: perform temporal attention processing on the spliced ​​features corresponding to the target sub-period to obtain the third attention weights of other sub-periods; and fuse the spliced ​​features corresponding to each sub-period based on the third attention weight to obtain fused features.

[0208] Each module in the above information processing device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0209] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig. 20 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface, the display unit and the input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, an information processing method is implemented. The display unit of the computer device is used to form a visually visible image, and can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse, etc.

[0210] Those skilled in the art will understand that Fig. 20 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0211] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0212] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0213] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0214] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0215] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0216] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0217] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the present application. It should be noted that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. An information processing method, characterized in that: The method comprises: Obtaining background information of a target product, interaction features and information features of the target product in a target period; generating an interaction feature hypergraph based on the interaction feature and the background information, and generating an information feature hypergraph based on the information feature and the background information; Performing attention processing on each node belonging to the same hyperedge in the interaction feature hypergraph to obtain updated interaction features, and performing attention processing on each node belonging to the same hyperedge in the information feature hypergraph to obtain updated information features; The updated information features are subjected to evolutionary enhancement processing to obtain enhanced information features; the enhanced information features and the updated information features are used to perform information processing on the target product.

2. The method according to claim 1, characterized in that The acquisition of background information of the target product, the interaction features and information features of the target product in the target period include: Obtain background information on the target product; Obtaining interactive data and information about the target product in a target period; Performing time series feature processing on the interaction data to obtain interaction features; Perform text feature processing on the information to obtain information features.

3. The method according to claim 1, characterized in that The background information includes regional information; and generating an interactive feature hypergraph based on the interactive feature and the background information includes: Taking the interaction feature as a first node, determining a region corresponding to each first node according to the region information; Divide the first nodes into sets according to the regions to obtain corresponding regional node sets; The first nodes in each of the regional node sets are connected to obtain an interactive feature hypergraph.

4. The method according to claim 1, characterized in that: The background information includes industry information; and generating an information feature hypergraph based on the information feature and the background information includes: Taking the information feature as the second node, determining the industry corresponding to each of the second nodes according to the industry information; Divide the second nodes into sets according to the industry to obtain a corresponding industry node set; The second nodes in each of the industry node sets are connected to obtain an information feature hypergraph.

5. The method according to claim 1, characterized in that The attention processing is performed on each node belonging to the same hyperedge in the interaction feature hypergraph to obtain updated interaction features, including: Performing attention processing on a first target node on a first hyperedge in the interactive feature hypergraph to obtain first attention weights of other nodes on the first hyperedge; The interaction feature corresponding to the first target node is updated based on the first attention weight to obtain an updated interaction feature corresponding to the first target node.

6. The method according to claim 1, characterized in that The attention processing is performed on each node belonging to the same hyperedge in the information feature hypergraph to obtain updated information features, including: Performing attention processing on a second target node on a second hyperedge in the information feature hypergraph to obtain second attention weights of other nodes on the second hyperedge; The information feature corresponding to the second target node is updated based on the second attention weight to obtain an updated information feature corresponding to the second target node.

7. The method according to claim 1, characterized in that The step of performing evolutionary enhancement processing on the updated information features to obtain enhanced information features includes: Get pre-built event graphs; Generate an information evolution graph based on the event graph and the updated information features; Performing convolution processing on the information evolution graph based on a machine learning model to obtain a convolution result; The enhanced information features are determined based on the convolution results.

8. The method according to claim 7, characterized in that The generating of the information evolution graph based on the event graph and the updated information features includes: Determining the evolution probability of the updated information feature based on the event graph; The updated information feature is used as a third node, and the third nodes are connected based on the evolution probability of the updated information feature to obtain an information evolution graph.

9. The method according to claim 8, characterized in that The step of determining the evolution probability of the updated information feature based on the event graph includes: Determine the target event corresponding to the updated information feature in the event graph; Obtaining the evolution probability of the target event in the event graph; The evolution probability of the target event is determined as the evolution probability of the updated information feature.

10. The method according to any one of claims 1 to 9, characterized in that The method further comprises: Performing splicing processing on the updated interaction features and the enhanced information features belonging to the same sub-period in the target period to obtain spliced ​​features corresponding to each sub-period; Performing temporal attention fusion on the concatenated features corresponding to each of the sub-periods to obtain fused features; Information processing is performed on the target product based on the fusion features to obtain processed information of the target product.

11. The method according to claim 10, characterized in that The step of performing temporal attention fusion on the spliced ​​features corresponding to each of the sub-periods to obtain fused features includes: Performing temporal attention processing on the concatenated features corresponding to the target sub-period to obtain third attention weights of the other sub-periods; The spliced ​​features corresponding to each of the sub-periods are fused based on the third attention weight to obtain a fused feature.

12. An information processing device, characterized in that: The device comprises: A feature acquisition module, used to acquire background information of a target product, interaction features and information features of the target product in a target period; A hypergraph construction module, used to generate an interaction feature hypergraph based on the interaction feature and the background information, and to generate an information feature hypergraph based on the information feature and the background information; A feature updating module, configured to perform attention processing on each node belonging to the same hyperedge in the interaction feature hypergraph to obtain updated interaction features, and to perform attention processing on each node belonging to the same hyperedge in the information feature hypergraph to obtain updated information features; The feature enhancement module is used to perform evolutionary enhancement processing on the updated information features to obtain enhanced information features; the enhanced information features and the updated information features are used to perform information processing on the target product.

13. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 11 are implemented.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 11 are implemented.

15. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 11 are implemented.

Images