Event-based asset value prediction method and device, computer equipment and medium

Abstract

The invention relates to the technical field of artificial intelligence, and discloses an event-based asset value prediction method and device, computer equipment and a medium, and the method comprises the steps: obtaining a historical value sequence of a specified asset and a historical event emotion value sequence, and generating an emotion enhancement value sequence based on the historical event emotion value sequence, and extracting a plurality of preset index values based on the historical value sequence, and inputting the plurality of preset index values into a preset dual-channel hybrid neural network for analysis to obtain a predicted asset value. The beneficial effects of the invention are that the method achieves the comprehensive analysis of the event emotion and the historical price through the two-channel mixed neural model, improves the prediction capability, and improves the accuracy of asset value prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of event-based asset value prediction, and in particular to an event-based asset value prediction method, device, computer equipment and medium. Background Art

[0002] Traditional asset price prediction models mostly rely on historical market data, economic indicators and fundamental analysis, which makes it difficult to fully reflect the immediate changes in the market and the impact of unexpected events.

[0003] In recent years, the impact of emergencies on market prices has gradually become more prominent, especially in the digital age, where information spreads rapidly and widely, and the public's emotions and reactions can instantly affect asset prices. For example, emergencies such as natural disasters or negative news often trigger public panic reactions, leading to large fluctuations in market prices. Traditional price prediction models often fail to capture the subtle changes in market sentiment when dealing with these emergencies, causing the prediction results to deviate from the actual situation and increase the risk of credit decisions.

[0004] In addition, the financial market itself has a good emotional response mechanism. Market sentiment can not only affect investors' decisions, but also cause drastic price fluctuations in a short period of time. However, quantifying the impact of market sentiment is still a complex task. Existing methods usually focus on the analysis of technical indicators or fundamental data, lacking in-depth understanding and quantification of market sentiment. Summary of the invention

[0005] Based on this, it is necessary to propose an event-based asset value prediction method, device, computer equipment and medium for the existing event-based asset value prediction problem.

[0006] An event-based asset value prediction method, the method comprising: Get the historical value sequence of the specified asset and the historical event sentiment value sequence; generating an emotion enhancement value sequence based on the historical event emotion value sequence, and extracting a plurality of preset indicator values ​​based on the historical value sequence; Inputting a plurality of the preset indicator values ​​into one channel of a preset dual-channel hybrid neural network, and inputting the emotion enhancement value sequence into another channel of the preset dual-channel hybrid neural network, obtains the predicted value and confidence interval of the designated asset.

[0007] Furthermore, the preset dual-channel hybrid neural network includes an LSTM network, a CNN network, a gated fusion layer and an output layer, wherein the output end of the LSTM network and the output end of the CNN network are respectively connected to the gated fusion layer, and the gated fusion layer is connected to the output layer; The step of inputting the multiple preset index values into one channel of a preset dual-channel hybrid neural network, and inputting the emotion-enhanced value sequence into the other channel of the preset dual-channel hybrid neural network to obtain the predicted value and confidence interval of the specified asset includes: Inputting the multiple preset index values into the LSTM network of the preset dual-channel hybrid neural network to obtain a first output value sequence and a first confidence sequence, and inputting the emotion-enhanced value sequence into the CNN network of the preset dual-channel hybrid neural network to obtain a second output value sequence and a second confidence sequence; Dynamically weighting the first output value and the second output value through a gating mechanism to obtain a comprehensive output value, and weighting the first confidence sequence and the second confidence sequence to obtain a comprehensive confidence sequence; Inputting the comprehensive output value and the comprehensive confidence sequence into the output layer to generate the predicted value and confidence interval of the specified asset.

[0008] Further, the step of obtaining the historical value sequence and historical event sentiment value sequence of the specified asset includes: Obtaining the transaction information of the specified asset within a specified historical time period, and the event data of the specified asset within the specified historical time period; Performing data analysis on the transaction information to obtain the historical value sequence, and obtaining the sentiment scores of the event data through a preset sentiment model, thereby obtaining the historical event sentiment value sequence.

[0009] Further, the step of obtaining the transaction information of the specified asset within the specified historical time period includes: Obtaining the transaction data of the specified asset on each platform within the specified historical time period; Performing data cleaning on each piece of the transaction data to obtain complete transaction data; Performing standardization processing on the complete transaction data to unify the time granularity, thereby obtaining the transaction information.

[0010] Further, the step of generating an emotion-enhanced value sequence based on the historical event sentiment value sequence includes: Analyzing the causal relationship between each sentiment score and value fluctuation in the historical event sentiment value sequence based on the Granger causality test method to calculate the emotion factor weights; Generating the emotion-enhanced value sequence according to the emotion factor weights.

[0011] Further, after the step of inputting the multiple preset index values into one channel of a preset dual-channel hybrid neural network and inputting the emotion-enhanced value sequence into the other channel of the preset dual-channel hybrid neural network to obtain the predicted value and confidence interval of the specified asset, the method further includes: Obtain the current value of the specified asset, and based on the deviation value of the predicted value relative to the current value; Determine whether the deviation value exceeds a preset deviation value; If it exceeds the preset deviation value, determine whether the confidence interval is greater than a preset confidence interval; If the confidence interval is greater than the preset confidence interval, issue a warning to a specified person.

[0012] Further, before the step of inputting the multiple preset index values into one channel of a preset dual-channel hybrid neural network and inputting the emotion-enhanced value sequence into the other channel of the preset dual-channel hybrid neural network to obtain the predicted value and confidence interval of the specified asset, the method further includes: Retrieve multiple sets of sample data from a preset database, and divide the sample data into a training data set and a validation data set according to a preset ratio; one set of the sample data consists of a training emotion-enhanced value sequence of a specified asset, multiple training preset index values, and corresponding labels, where the labels are the predicted value and the confidence interval; Input the multiple training preset index values in the training data set into one channel of an initial dual-channel hybrid neural network, and input the emotion-enhanced value sequence into the other channel of the initial dual-channel hybrid neural network for training, so as to obtain a temporary initial dual-channel hybrid neural network; Use the validation data set to validate the temporary initial dual-channel hybrid neural network to obtain a validation result, and determine whether the validation result is passed; If the validation result is passed, record the temporary initial dual-channel hybrid neural network as the preset dual-channel hybrid neural network.

[0013] An event-based asset value prediction device, the device includes: An acquisition module, configured to acquire a historical value sequence and a historical event sentiment value sequence of a specified asset; A generation module, configured to generate an emotion-enhanced value sequence based on the historical event sentiment value sequence, and extract multiple preset index values based on the historical value sequence; An input module for inputting multiple of the preset metric values into one channel of a preset dual-channel hybrid neural network, and inputting the emotion-enhanced value sequence into another channel of the preset dual-channel hybrid neural network to obtain the predicted value and confidence interval of the specified asset.

[0014] A computer device includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor performs the following steps: Obtain the historical value sequence and historical event sentiment value sequence of a specified asset; Generate an emotion-enhanced value sequence based on the historical event sentiment value sequence, and extract multiple preset metric values based on the historical value sequence; Input multiple of the preset metric values into one channel of a preset dual-channel hybrid neural network, and input the emotion-enhanced value sequence into another channel of the preset dual-channel hybrid neural network to obtain the predicted value and confidence interval of the specified asset. A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the processor performs the following steps: Obtain the historical value sequence and historical event sentiment value sequence of a specified asset; Generate an emotion-enhanced value sequence based on the historical event sentiment value sequence, and extract multiple preset metric values based on the historical value sequence; Input multiple of the preset metric values into one channel of a preset dual-channel hybrid neural network, and input the emotion-enhanced value sequence into another channel of the preset dual-channel hybrid neural network to obtain the predicted value and confidence interval of the specified asset.

[0015] Advantages of the present invention: By obtaining the historical value sequence and historical event sentiment value sequence of a specified asset, generating an emotion-enhanced value sequence based on the historical event sentiment value sequence, and extracting multiple preset metric values based on the historical value sequence, and inputting them into a preset dual-channel hybrid neural network for analysis, the comprehensive analysis of event sentiment and historical price is realized through the dual-channel hybrid neural model, the prediction ability is improved, and the accuracy of asset value prediction is further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Wherein: Figure 1 is an application environment diagram of an event-based asset value prediction method in an embodiment; Figure 2 is a flowchart of an event-based asset value prediction method in an embodiment; Figure 3 is a structural block diagram of an event-based asset value prediction device in an embodiment; Figure 4 is a structural block diagram of a computer device in an embodiment. Detailed implementation manners

[0018] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0019] Figure 1 is an application environment diagram of event-based asset value prediction in an embodiment. Refer to Figure 1 , this event-based asset value prediction method is applied to an event-based asset value prediction system. The event-based asset value prediction system includes a terminal 110 and a server 120. The terminal 110 and the server 120 are connected through a network. The terminal 110 may specifically be a desktop terminal or a mobile terminal, and the mobile terminal may specifically be at least one of a mobile phone, a tablet computer, a notebook computer, etc. The server 120 may be implemented by an independent server or a server cluster composed of multiple servers. The terminal 110 is used to collect event information and value information of a specified asset, and the server 120 is used to implement asset value prediction.

[0020] Such as Figure 2 shown, in an embodiment, an event-based asset value prediction method, device, computer device, and medium are provided. This method can be applied to both the terminal and the server. This embodiment takes the application to the server as an example for illustration. The event-based asset value prediction method specifically includes the following steps: S1: Obtain the historical value sequence and the historical event sentiment value sequence of the specified asset; S2: Generate an emotion-enhanced value sequence based on the historical event sentiment value sequence, and extract multiple preset index values based on the historical value sequence; S3: Input the multiple preset index values into one channel of a preset dual-channel hybrid neural network, and input the emotion-enhanced value sequence into the other channel of the preset dual-channel hybrid neural network to obtain the predicted value and confidence interval of the specified asset.

[0021] As described in step S1 above, obtain the historical value sequence and historical event sentiment value sequence of the specified asset. Among them, the historical value sequence usually refers to the price change data of the asset over a past period of time, which can be obtained from public markets, exchanges or relevant financial databases. The historical price can reflect the cyclical trends and volatility of the market. The historical event sentiment value sequence is the result of sentiment analysis obtained by analyzing event data related to the asset. The event data can come from multiple channels such as news reports, social media, and forum discussions. Sentiment analysis processes text information through natural language processing (NLP) techniques to identify the positive, negative or neutral tendencies of public sentiment. This sentiment value sequence will provide important market psychological factors for subsequent analysis, enabling value prediction to not only be limited to historical price changes, but also consider the impact of market sentiment on asset prices.

[0022] As described in step S2 above, generate an emotion-enhanced value sequence based on the historical event sentiment value sequence, and extract multiple preset index values based on the historical value sequence. Among them, generating the emotion-enhanced value sequence is by a method of combining event sentiment values with the historical value sequence. This process usually involves a weighted model, where the positive and negative impacts of event sentiment values will have corresponding impacts on the historical value of the asset. For example, in the case of increasing the positive emotion impact of an event on the historical price, the emotion-enhanced value sequence may show a higher predicted value, and vice versa. Extracting multiple preset index values from the historical value sequence is to provide multi-dimensional information for the subsequent prediction model. These preset indicators often include volatility, mean reversion, price trend, moving average, trading volume, etc. By analyzing these indicators, the market performance of the asset can be more comprehensively reflected, providing the necessary background information for the model. The process of extracting these indicators usually involves mathematical calculations and statistical analysis to ensure that the characteristics of historical data can be fully utilized.

[0023] As described in step S3 above, multiple preset index values are input into one channel of a preset dual-channel mixed neural network, and the emotion-enhanced value sequence is input into the other channel of the preset dual-channel mixed neural network to obtain the predicted value and confidence interval of the specified asset. A dual-channel mixed neural network is used for asset value prediction. Specifically, this model consists of two independent channels. One channel processes multiple preset index values, and the other channel processes the emotion-enhanced value sequence. With this structure, the model can simultaneously focus on historical value indicators and event sentiment information, and fully integrate the advantages of both to improve the accuracy of prediction. In actual operation, first, the extracted multiple preset index values are input into one channel so that the model can learn the features related to historical price changes. The deep neural network structure of this channel can extract complex patterns and non-linear relationships in the data through multiple hidden nodes. The emotion-enhanced value sequence is input into the other channel to enable the model to capture the direct impact of event sentiment on market fluctuations. Such a design allows the model to learn at different levels, enabling it to have strong expressive ability. After training, the dual-channel mixed neural network will integrate the information from the two channels and finally output the predicted value and its confidence interval of the specified asset. Among them, the predicted value can be the predicted price at a future time point or the predicted price within a future period. Similarly, the confidence interval can be the confidence interval at a future time point or the confidence interval within a future period. The confidence interval provides the uncertainty of the prediction to support credit decisions. The successful implementation of this process can make full use of multiple information sources, thereby enhancing the prediction ability of the model and making the asset value prediction both accurate and flexible.

[0024] In one embodiment, the preset dual-channel mixed neural network includes an LSTM network, a CNN network, a gated fusion layer, and an output layer. Among them, the output ends of the LSTM network and the CNN network are respectively connected to the gated fusion layer, and the gated fusion layer is connected to the output layer; The step S3 of inputting multiple preset index values into one channel of a preset dual-channel mixed neural network and inputting the emotion-enhanced value sequence into the other channel of the preset dual-channel mixed neural network to obtain the predicted value and confidence interval of the specified asset includes: S301: Input multiple preset index values into the LSTM network of the preset dual-channel mixed neural network to obtain a first output value sequence and a first confidence sequence, and input the emotion-enhanced value sequence into the CNN network of the preset dual-channel mixed neural network to obtain a second output value sequence and a second confidence sequence; S302: Dynamically weight the first output value and the second output value through a gating mechanism to obtain a comprehensive output value, and weight the first confidence sequence and the second confidence sequence to obtain a comprehensive confidence sequence; S303: Input the comprehensive output value and the comprehensive confidence sequence into the output layer to generate the predicted value and the confidence interval of the specified asset.

[0025] As described in the above steps S301 - S302, multiple preset metric values are input into the Long Short - Term Memory network (LSTM) of a preset dual - channel hybrid neural network, thereby obtaining a first output value sequence and a first confidence sequence. LSTM is a powerful Recurrent Neural Network (RNN) that can effectively process sequence data, especially performing well in capturing time dependencies and long - term context information. In this channel, the LSTM network generates an output value sequence representing the historical data of the asset by gradually reading and updating each preset metric. At the same time, the emotion - enhanced value sequence is input into the Convolutional Neural Network (CNN) of the network to generate a second output value sequence and a second confidence sequence. CNN is good at processing images and local features, and can extract local patterns and features of the emotion sequence through convolutional layers, thereby capturing the impact of event information on asset prices. CNN obtains features through the sliding window technique and refines the information into an output value sequence for subsequent analysis. Finally, through this step, the LSTM network and the CNN network respectively output sequence values and confidence information related to historical value metrics and emotion - enhanced information, which constitute the input of the subsequent layer and provide basic data for further fusion and weighting. A gating mechanism is used to dynamically weight the first output value and the second output value obtained in the previous step to generate a comprehensive output value. In addition, the first confidence sequence and the second confidence sequence also need to be weighted to form a comprehensive confidence sequence. The gating mechanism is an intelligent weighting strategy designed to dynamically adjust the weights of different outputs according to the prevalence of input information. In this case, first, a weight factor between the first output value and the second output value is calculated based on specific conditions or context information to ensure that information sources with a greater impact on the prediction are given priority under specific circumstances. For example, if market sentiment fluctuates greatly, the weight of the emotion - enhanced value may be increased to ensure that the impact of emotion information on the prediction is fully considered. After dynamic weighting, the resulting comprehensive output value will represent a new prediction result that combines historical value metrics and event information. Similarly, using the same gating mechanism to weight the confidence sequence will help obtain a comprehensive confidence sequence, which can reflect the reliability of the prediction result. By effectively integrating the two output information, this step ensures that the prediction ability of the final model is more powerful and flexible, and can automatically adjust the focus under different market conditions. The comprehensive output value and the comprehensive confidence sequence are input into a preset output layer to generate the predicted value of the specified asset and its confidence interval. The output layer usually consists of a fully - connected layer that takes the previously weighted and fused comprehensive output value as input and adds a non - linear transformation to enhance the expressive power of the model. In this layer, the comprehensive output value is processed to generate the final predicted value, which is an estimate of the future price of the specified asset and aims to provide a reference for credit decisions, investment decisions, or market analysis. At the same time, the comprehensive confidence sequence will also be processed through the output layer to calculate the confidence interval.Confidence intervals provide a quantification of the uncertainty of predicted values. This interval reveals the range within which the actual price may fall, provides the reliability of the prediction results, and reduces potential risks in the decision-making process. It achieves the goal of event-based asset value prediction and provides a powerful tool for financial market participants to manage risks and opportunities.

[0026] In one embodiment, step S1 of obtaining the historical value sequence and the historical event sentiment value sequence of the specified asset includes: S101: Obtain the transaction information of the specified asset within the specified historical time period, and the event data of the specified asset within the specified historical time period; S102: Perform data analysis on the transaction information to obtain the historical value sequence, and obtain the sentiment scores of the event data through a preset sentiment model, so as to obtain the historical event sentiment value sequence.

[0027] As described in steps S101 - S102 above, the acquisition of the historical event sentiment value sequence is realized. Obtaining transaction information usually involves extracting real-time or historical transaction data from relevant financial markets, exchanges, or databases. These data include important indicators such as price, trading volume, opening price, closing price, highest price, and lowest price. These transaction data provide key insights into how the asset performs in the market, helping to analyze its price trends, volatility characteristics, and market reactions. On the other hand, the acquisition of event data covers information from multiple channels such as social media, news reports, and forum comments. Through web crawler technology or API interfaces, the system can automatically capture historical event data related to the specified asset, including the discussion heat, sentiment tendency of the public towards the asset, and the occurrence of unexpected events. These event data can effectively reflect the fluctuations of market sentiment and have an immediate impact on asset prices. Perform data analysis on the transaction information to extract the historical value sequence. Usually, this process includes calculating price changes day by day or week by week and organizing them into a sequence, for example, by calculating the closing price, highest price, and lowest price of each day to construct a time series. The generated historical value sequence can intuitively reflect the historical dynamics of the asset price, showing trends and volatility characteristics. Secondly, use a preset sentiment model to calculate the sentiment scores of the event data. The sentiment model usually adopts natural language processing (NLP) technology, which can be a rule-based model, or a machine learning or deep learning model. In a specific embodiment, the sentiment model is a BERT-based model. The event data will be processed to identify the sentiment tendency (such as positive, negative, or neutral) of each event message and convert it into a specific sentiment score. Finally, the obtained historical event sentiment value sequence will be able to be analyzed in parallel with the historical value sequence, providing a more comprehensive perspective.

[0028] In one embodiment, step S101 of obtaining the transaction information of the specified asset within the specified historical time period includes: S1011: Obtain the transaction data of the specified asset on each platform within the specified historical time period; S1012: Clean the transaction data to obtain complete transaction data; S1013: Standardize the complete transaction data to unify the time granularity, thereby obtaining the transaction information.

[0029] As described in steps S1011 - S1013 above, the original transaction data obtained from multiple trading platforms is cleaned. Errors or inconsistencies in the data are processed and corrected to ensure the quality and accuracy of the data. The original transaction data often contains various problems, such as missing values, duplicate records, inconsistent formats, outliers, etc. If these problems are not solved, they may have a negative impact on subsequent data analysis and model construction. The process of data cleaning usually includes the following steps. First, the system needs to identify and remove duplicate data items to ensure that each transaction record is unique. Next, missing values are processed. Common methods include filling in missing values (such as using the mean, median, or mode) or directly deleting data rows with too many missing values. During the data conversion process, the format consistency of the information should also be ensured, for example, unifying the date and time format so that time series analysis can be carried out smoothly during subsequent analysis. After completing the data cleaning, the system will generate a clean and complete transaction data set to ensure that the data used in subsequent analysis has high reliability and accuracy. The standardized processing of the already cleaned complete transaction data aims to unify the time granularity so that subsequent data analysis can proceed smoothly. Since asset transactions may have different time intervals on different platforms (such as minute, hourly, or daily transaction data, etc.), and the analysis model usually needs to be processed according to a unified time granularity. The standardized processing usually includes the following steps. First, a suitable time granularity needs to be selected, for example, in units of days, hours, or other suitable time periods. Next, all transaction data needs to be unified in format. For example, the transaction records are aggregated according to the selected time granularity, which may involve summing or averaging the price and trading volume within a certain time window. To ensure coverage, interpolation techniques or forward filling methods can be used to process time points without records to ensure that there is transaction information in each time period.

[0030] In one embodiment, step S2 of generating an emotion-enhanced value sequence based on the historical event emotion value sequence includes: S201: Analyze the causal relationship between each emotion score and value fluctuation in the historical event emotion value sequence based on the Granger causality test method to calculate the emotion factor weight; S202: Generate the emotion enhancement value sequence according to the emotion factor weights.

[0031] As described in the above steps S201 - S202, the Granger causality test method is used to analyze the causal relationship between each emotion score in the historical event emotion value sequence and the value fluctuation. The Granger causality test method is a method used to determine whether one time series can help predict another time series. Specifically, first, it is necessary to decompose the historical event emotion value sequence to identify the time series of different emotion scores (such as positive emotion, negative emotion, and neutral emotion). These emotion scores are usually generated after performing sentiment analysis on event data and can reflect the public's emotional attitude towards a specific asset. For example, a positive emotion score may be related to a price increase, while a negative emotion score may correspond to a price decrease. Subsequently, through the Granger causality test method, the system will perform statistical analysis on the historical emotion scores and the corresponding historical value fluctuations to confirm which emotion scores can significantly affect the change in asset prices. By calculating the relevant F - statistic and P - value, the significance of the causal relationship can be judged, and finally, the weight of each emotion score on the asset value fluctuation can be calculated, which is called the emotion factor weight. Based on the emotion factor weights, an emotion enhancement value sequence is generated. This process involves combining the event emotion scores with the weights to better capture the impact of market sentiment on asset value, thereby forming more sensitive prediction data. Specifically, each historical emotion score is multiplied by its corresponding emotion factor weight. Such a calculation can quantify the influence of different emotion scores on the same scale, enabling the effective comparison and aggregation of the impacts of different emotions on the asset. For example, if the positive emotion score is very high during a certain period and its corresponding weight is also large, it indicates that the public's psychological expectation for the asset during this period is very optimistic, and this factor will have a significant impact on the emotion enhancement value of the asset. After completing the weighted calculation of all historical event emotion scores, the system will generate a new time series, namely the emotion enhancement value sequence. This sequence represents the expected value change of the asset after considering the influence of public sentiment. Therefore, the emotion enhancement value sequence can introduce the real - time market sentiment into the traditional value prediction model, ensuring more accurate and flexible predictions. Finally, the generated emotion enhancement value sequence is used as an input for subsequent asset value prediction. This process not only enhances the dynamic response ability of the prediction model but also improves its adaptability to sudden market sentiment events, providing more accurate information support for financial decision - making.

[0032] In one embodiment, after step S3 of inputting the multiple preset index values into one channel of a preset dual-channel hybrid neural network and inputting the emotion-enhanced value sequence into the other channel of the preset dual-channel hybrid neural network to obtain the predicted value and confidence interval of the specified asset, the following steps are further included: S401: Obtain the current value of the specified asset and determine the deviation value of the predicted value relative to the current value; S402: Determine whether the deviation value exceeds a preset deviation value; S403: If the deviation value exceeds the preset deviation value, determine whether the confidence interval is greater than a preset confidence interval; S404: If the confidence interval is greater than the preset confidence interval, issue a warning to a specified person.

[0033] As described in the above steps S401 - S404, obtain the current market value of the specified asset. This information usually comes from the latest transaction data or real - time market quotes. The current value is the real - time price of the asset, providing a reference point for subsequent analysis to be centered around market dynamics. Calculate the deviation value between the predicted value and the current value. Specifically, the deviation value refers to the difference between the predicted value and the actual market price. In this way, the effectiveness of the prediction model relative to the market reality can be evaluated. For example, if the predicted value is significantly higher than the current value, it may indicate that the model is too optimistic; if the predicted value is significantly lower than the current value, it may reflect a pessimistic attitude of the model towards the market situation. Judge whether the obtained deviation value exceeds the preset deviation value. The preset deviation value is an important threshold, usually used to evaluate the rationality of the prediction result, ensuring that the result output by the prediction model is within a reasonable range and avoiding the negative impact of possible misleading information on the decision - making process. The setting of the preset deviation value is usually based on historical market data analysis, economic indicators, or the test results of the model, aiming to define the range of "normal" and "abnormal". When the deviation value exceeds this set threshold, it indicates that the prediction result deviates greatly from the actual market situation, which may reflect problems in the model or sudden changes in the market. Through this judgment, the reliability problem of the prediction result can be identified in a timely manner and an important basis for subsequent analysis or warning can be provided. Judge if, when the deviation value exceeds the preset deviation value, the relevant confidence interval is greater than the set preset confidence interval, where the preset confidence interval is a pre - set interval that can be determined according to the distribution of historical prediction errors. The confidence interval is a quantification of the uncertainty of the prediction result, reflecting the range within which the predicted value may fall at a given confidence level. If the confidence interval is large, it indicates a high uncertainty of the prediction result, which may lead to a decrease in the trust of market participants in this prediction. On the contrary, if the confidence interval is small, it means that the model has a stronger grasp of the predicted value and the market sentiment is more definite. During the judgment process, the system will compare the dynamically measured confidence interval with the preset value to decide whether further action is needed. If the confidence interval is indeed greater than the preset confidence interval, it means that the current prediction signal has a high uncertainty, and then the user is prompted to give a warning. The realization of this warning system can be through email, SMS, mobile application notifications, or system - built - in message push and other methods to quickly convey the warning information to relevant decision - makers. Through this warning mechanism, enterprises can respond accordingly at the first time when market anomalies occur, preventing potential losses. This not only enhances the enterprise's ability to control market dynamics but also improves the reaction speed and decision - making flexibility, ultimately providing a guarantee for the overall operation safety and asset management of the enterprise.

[0034] In one embodiment, before the step S3 of inputting the multiple preset index values into one channel of a preset dual-channel hybrid neural network and inputting the emotion-enhanced value sequence into another channel of the preset dual-channel hybrid neural network to obtain the predicted value and confidence interval of the specified asset, the following steps are further included: S211: Retrieve multiple sets of sample data from a preset database, and divide the sample data into a training data set and a validation data set according to a preset ratio; one set of the sample data consists of a training emotion-enhanced value sequence of a specified asset, multiple training preset index values, and corresponding labels, and the labels are the predicted value and the confidence interval; S212: Input the multiple training preset index values in the training data set into one channel of the initial dual-channel hybrid neural network, and input the emotion-enhanced value sequence into another channel of the initial dual-channel hybrid neural network for training, so as to obtain a temporary initial hybrid neural network; S213: Use the validation data set to validate the temporary initial hybrid neural network to obtain a validation result, and determine whether the validation result is passed; S214: If the validation result is passed, record the temporary initial hybrid neural network as the preset dual-channel hybrid neural network.

[0035] As described in the above steps S211 - S214, multiple sets of sample data are retrieved from a preset database for establishing and training a dual - channel hybrid neural network. The sample data will include a training sentiment - enhanced value sequence of a specified asset and multiple preset metric values. The sentiment - enhanced value sequence reflects the impact of the corresponding event sentiment score on the asset price, while the multiple preset metric values are quantitative expressions of the historical fluctuations of the asset price. These data can provide input features for the network, helping to form an accurate prediction. Next, the sample data is divided into a training data set and a validation data set according to a preset ratio. Common ratios are 70% for training and 30% for validation, or 80% / 20%, etc. Through such a division, it is ensured that the network can learn rich features during the training phase, and at the same time, the validation set is used to test the effectiveness of the model and avoid the problem of overfitting. The training sentiment - enhanced value sequence and multiple training preset metric values in the training data set are respectively input into the initial dual - channel hybrid neural network for training. The dual - channel hybrid neural network is designed to make full use of two types of information (sentiment and metrics) to achieve more accurate asset value prediction. Specifically, multiple of the said training preset metric values are fed into one channel of the network, which usually mainly consists of a structure suitable for processing sequence data such as LSTM, aiming to capture the dynamic behavior and periodic changes in time - series data. At the same time, other training preset metric values are input into another channel, which is mainly responsible for analyzing data related to event impacts. Through a convolutional neural network (CNN) or a recurrent neural network (such as LSTM), a sliding window model can be used to identify patterns in the sentiment sequence and extract key features. With this structure, the network can simultaneously consider historical value fluctuations and the current sentiment state to form a comprehensive prediction model. After multiple iterative trainings, the network parameters will be adjusted to minimize the prediction loss function, forming a temporary initial hybrid neural network. This step is crucial in the whole process, ensuring that the model can learn complex non - linear relationships and thus prepare for subsequent validation and application. The validation data set is divided from the sample data and is different from the training data set. Its main role is to verify the generalization ability of the model. This means that even if the model performs well on the training data, it does not necessarily mean that it can make reliable predictions on new real - world data. By inputting the validation data set into the current model, the system will calculate its prediction results and compare them with the actual labels. The validation results are usually measured by a series of evaluation metrics such as mean squared error (MSE), mean absolute error (MAE), R - squared value, etc. If the performance metrics of the model meet the preset requirements, it can be judged that the validation result is "validation passed". Otherwise, further tuning and improvement are required. If during the validation process, the model shows good performance and passes the preset evaluation criteria, it is considered an effective model suitable for subsequent applications.

[0036] Refer to Figure 3, the present invention also provides an event-based asset value prediction device, the device comprising: An acquisition module 902, configured to acquire a historical value sequence and a historical event sentiment value sequence of a specified asset; A generation module 904, configured to generate an emotion-enhanced value sequence based on the historical event sentiment value sequence, and extract a plurality of preset index values based on the historical value sequence; An input module 906, configured to input the plurality of preset index values into one channel of a preset dual-channel hybrid neural network, and input the emotion-enhanced value sequence into another channel of the preset dual-channel hybrid neural network, to obtain a predicted value and a confidence interval of the specified asset.

[0037] In one embodiment, the preset dual-channel hybrid neural network includes an LSTM network, a CNN network, a gating fusion layer, and an output layer, wherein the output ends of the LSTM network and the CNN network are respectively connected to the gating fusion layer, and the gating fusion layer is connected to the output layer; the input module 906 includes: A preset index value input sub-module, configured to input the plurality of preset index values into the LSTM network of the preset dual-channel hybrid neural network, to obtain a first output value sequence and a first confidence sequence, and input the emotion-enhanced value sequence into the CNN network of the preset dual-channel hybrid neural network, to obtain a second output value sequence and a second confidence sequence; A dynamic weighting sub-module, configured to dynamically weight the first output value and the second output value through a gating mechanism, to obtain a comprehensive output value, and weight the first confidence sequence and the second confidence sequence, to obtain a comprehensive confidence sequence; A comprehensive confidence sequence input sub-module, configured to input the comprehensive output value and the comprehensive confidence sequence into the output layer, to generate a predicted value and a confidence interval of the specified asset.

[0038] In one embodiment, the acquisition module 902 includes: An event data acquisition sub-module, configured to acquire transaction information of the specified asset within a specified historical time period, and event data of the specified asset within the specified historical time period; A data analysis sub-module, configured to perform data analysis on the transaction information, to obtain the historical value sequence, and obtain a sentiment score of the event data through a preset sentiment model, so as to obtain a historical event sentiment value sequence.

[0039] In one embodiment, the event data acquisition sub-module includes: An acquisition unit, configured to acquire transaction data of the specified asset on each platform within a specified historical time period; A cleaning unit for cleaning each piece of the transaction data to obtain complete transaction data; A processing unit for performing standardization processing on the complete transaction data to unify the time granularity, thereby obtaining the transaction information.

[0040] In one embodiment, the generation module 904 includes: An emotional factor weight calculation sub-module for analyzing the causal relationship between each emotional score and value fluctuation in the historical event sentiment value sequence based on the Granger causality test method to calculate the emotional factor weight; An emotional enhanced value sequence generation sub-module for generating the emotional enhanced value sequence according to the emotional factor weight.

[0041] In one embodiment, the event-based asset value prediction device further includes: A current value acquisition module for acquiring the current value of the specified asset and according to the deviation value of the predicted value relative to the current value; A deviation value judgment module for judging whether the deviation value exceeds a preset deviation value; A confidence interval judgment module for judging whether the confidence interval is greater than a preset confidence interval if the preset deviation value is exceeded; An early warning sending module for sending an early warning to a specified person if the confidence interval is greater than the preset confidence interval.

[0042] In one embodiment, the event-based asset value prediction device further includes: A sample data retrieval module for retrieving multiple groups of sample data from a preset database and dividing the sample data into a training data set and a verification data set according to a preset ratio; one group of the sample data is composed of a training emotional enhanced value sequence of a specified asset, multiple training preset index values, and corresponding labels, and the labels are predicted values and confidence intervals; A training emotional enhanced value sequence input module for inputting the multiple training preset index values in the training data set into one channel of a dual-channel hybrid neural initial network, and inputting the emotional enhanced value sequence into another channel of the dual-channel hybrid neural initial network for training, thereby obtaining a temporary hybrid neural initial network; A verification module for verifying the temporary hybrid neural initial network by using the verification data set to obtain a verification result and judging whether the verification result is verification passed; A marking module for marking the temporary hybrid neural initial network as a preset dual-channel hybrid neural network if the verification result is verification passed.

[0043] Figure 4 The internal structure diagram of a computer device in an embodiment is shown. The computer device may specifically be a terminal or a server. As Figure 4 shown, the computer device includes a processor, a memory, and a network interface connected through a system bus. Among them, the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program. When the computer program is executed by the processor, the processor can implement the method for predicting asset value based on events. The internal memory may also store a computer program. When the computer program is executed by the processor, the processor can execute the method for predicting asset value based on events. Those skilled in the art can understand that Figure 4 the structure shown in

[0044] In one embodiment, a computer device is proposed, including a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor performs the following steps: Obtain the historical value sequence and historical event sentiment value sequence of a specified asset; Generate an emotion-enhanced value sequence based on the historical event sentiment value sequence, and extract multiple preset index values based on the historical value sequence; Input the multiple preset index values into one channel of a preset dual-channel hybrid neural network, and input the emotion-enhanced value sequence into the other channel of the preset dual-channel hybrid neural network to obtain the predicted value and confidence interval of the specified asset.

[0045] By setting up a dual-channel hybrid neural network to analyze event data and historical value sequences, the comprehensive analysis of event sentiment and historical prices is realized through the dual-channel hybrid neural model, improving the prediction ability and further improving the accuracy of asset value prediction.

[0046] In one embodiment, a computer-readable storage medium is proposed, storing a computer program. When the computer program is executed by the processor, the processor performs the following steps: Obtain the historical value sequence and historical event sentiment value sequence of a specified asset; Generate an emotion-enhanced value sequence based on the historical event sentiment value sequence, and extract multiple preset index values based on the historical value sequence; Input multiple of the preset index values into one channel of a preset dual-channel hybrid neural network, and input the emotion-enhanced value sequence into the other channel of the preset dual-channel hybrid neural network to obtain the predicted value and confidence interval of the specified asset.

[0047] By setting up a dual-channel hybrid neural network to analyze event data and historical value sequences, a comprehensive analysis of event sentiment and historical prices can be achieved through the dual-channel hybrid neural model, improving the prediction ability and thus enhancing the accuracy of asset value prediction.

[0048] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it may include the processes of the above method embodiments. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application may include non-volatile and / or volatile memories. Non-volatile memories may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0049] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, 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, it should be considered as the scope recorded in this specification.

[0050] The above-described embodiments merely represent several implementation manners of the present application. Their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the patent of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. An event-based asset value prediction method, characterized in that The method includes: Obtaining a historical value sequence and a historical event sentiment value sequence of a specified asset; Generating an emotion-enhanced value sequence based on the historical event sentiment value sequence, and extracting multiple preset index values based on the historical value sequence; Inputting the multiple preset index values into one channel of a preset dual-channel hybrid neural network, and inputting the emotion-enhanced value sequence into another channel of the preset dual-channel hybrid neural network to obtain the predicted value and confidence interval of the specified asset.

2. The event-based asset value prediction method according to claim 1, wherein The preset dual-channel hybrid neural network includes an LSTM network, a CNN network, a gating fusion layer, and an output layer. Among them, the output ends of the LSTM network and the CNN network are respectively connected to the gating fusion layer, and the gating fusion layer is connected to the output layer; The step of inputting the multiple preset index values into one channel of a preset dual-channel hybrid neural network, and inputting the emotion-enhanced value sequence into another channel of the preset dual-channel hybrid neural network to obtain the predicted value and confidence interval of the specified asset includes: Inputting the multiple preset index values into the LSTM network of the preset dual-channel hybrid neural network to obtain a first output value sequence and a first confidence sequence, and inputting the emotion-enhanced value sequence into the CNN network of the preset dual-channel hybrid neural network to obtain a second output value sequence and a second confidence sequence; Dynamically weighting the first output value and the second output value through a gating mechanism to obtain a comprehensive output value, and weighting the first confidence sequence and the second confidence sequence to obtain a comprehensive confidence sequence; Inputting the comprehensive output value and the comprehensive confidence sequence into the output layer to generate the predicted value and confidence interval of the specified asset.

3. The event-based asset value prediction method according to claim 1, wherein The step of obtaining a historical value sequence and a historical event sentiment value sequence of a specified asset includes: Obtaining the transaction information of the specified asset within a specified historical period, and the event data of the specified asset within the specified historical period; Performing data analysis on the transaction information to obtain the historical value sequence, and obtaining the sentiment scores of the event data through a preset sentiment model, so as to obtain the historical event sentiment value sequence.

4. The event-based asset value prediction method according to claim 1, wherein The step of obtaining the transaction information of the specified asset within a specified historical period includes: Obtaining the transaction data of the specified asset on each platform within a specified historical period; Performing data cleaning on each piece of transaction data to obtain complete transaction data; Performing standardization processing on the complete transaction data to unify the time granularity, so as to obtain the transaction information.

5. The event-based asset value prediction method according to claim 1, wherein The step of generating an emotion-enhanced value sequence based on the historical event sentiment value sequence includes: Analyzing the causal relationship between each sentiment score and value fluctuation in the historical event sentiment value sequence based on the Granger causality test method to calculate the emotion factor weight; Generating the emotion-enhanced value sequence according to the emotion factor weight.

6. The event-based asset value prediction method according to claim 1, wherein After the step of inputting the multiple preset index values into one channel of a preset dual-channel hybrid neural network and inputting the emotion-enhanced value sequence into the other channel of the preset dual-channel hybrid neural network to obtain the predicted value and confidence interval of the specified asset, the method further includes: Obtaining the current value of the specified asset and based on the deviation value of the predicted value relative to the current value; Judging whether the deviation value exceeds a preset deviation value; If the deviation value exceeds the preset deviation value, judging whether the confidence interval is greater than a preset confidence interval; If the confidence interval is greater than the preset confidence interval, sending a warning to a specified person.

7. The event-based asset value prediction method according to claim 1, characterized in that, Before the step of inputting the multiple preset index values into one channel of a preset dual-channel hybrid neural network and inputting the emotion-enhanced value sequence into the other channel of the preset dual-channel hybrid neural network to obtain the predicted value and confidence interval of the specified asset, the method further includes: Retrieving multiple groups of sample data from a preset database and dividing the sample data into a training data set and a validation data set according to a preset ratio; one group of the sample data consists of a training emotion-enhanced value sequence of a specified asset, multiple training preset index values, and corresponding labels, and the labels are the predicted value and the confidence interval; Inputting the multiple training preset index values in the training data set into one channel of an initial dual-channel hybrid neural network and inputting the emotion-enhanced value sequence into the other channel of the initial dual-channel hybrid neural network for training, so as to obtain a temporary initial hybrid neural network; Validating the temporary initial hybrid neural network by using the validation data set to obtain a validation result and judging whether the validation result is a pass; If the validation result is a pass, recording the temporary initial hybrid neural network as the preset dual-channel hybrid neural network.

8. An event-based asset value prediction device, characterized in that, The device includes: An acquisition module, configured to acquire a historical value sequence and a historical event sentiment value sequence of a specified asset; A generation module, configured to generate an emotion-enhanced value sequence based on the historical event sentiment value sequence and extract multiple preset index values based on the historical value sequence; An input module, configured to input the multiple preset index values into one channel of a preset dual-channel hybrid neural network and input the emotion-enhanced value sequence into the other channel of the preset dual-channel hybrid neural network to obtain the predicted value and confidence interval of the specified asset.

9. A computer-readable storage medium, characterized in that, A computer program is stored, and when the computer program is executed by a processor, the processor is caused to execute the steps of the event-based asset value prediction method according to any one of claims 1 to 7.

10. A computer device, characterized in that, The device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor is caused to execute the steps of the event-based asset value prediction method according to any one of claims 1 to 7.

Images