A multi-dimensional feature comprehensive prediction method and device based on analytic hierarchy process

Abstract

The application discloses a kind of multi-dimensional feature comprehensive prediction method and device based on analytic hierarchy process, wherein the method includes using poll model prediction result, futures model prediction result and gambling model prediction result, according to hierarchical comprehensive prediction model, obtains the final prediction result of day.The method of the application and the best weight of poll model prediction result, futures model prediction result and gambling model prediction result determined by hierarchical analysis method are used to generate more accurate election prediction results by combining the prediction data of the day.The method not only makes up for the limitations of single prediction technology, but also improves the comprehensive reliability and interpretability of the prediction results, providing a new solution for predicting election results.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and in particular to a multidimensional feature comprehensive prediction method and apparatus based on the analytic hierarchy process. Background Technology

[0002] Public opinion polls are one of the most traditional methods of election prediction. These surveys are typically conducted by professional organizations, mainstream media, or university research teams. They collect data based on sampling survey theory, refine the data with expert opinions, analyze voters' political leanings through opinion polls, and provide predictions. The advantage of this method is its ability to quickly reflect the latest changes in public opinion, especially the dynamic impact that may occur as the election approaches, and to provide information about voter demographics. However, the disadvantage of this method is that it is easily affected by various factors such as sample bias, sample size, institutional subjectivity, and changes in voter intentions, often resulting in some degree of bias in the results. Summary of the Invention

[0003] The main technical problems solved by this invention include: how to overcome the problems of single prediction method, large result deviation and insufficient timeliness in the above-mentioned prior art, thereby improving the overall reliability and interpretability of prediction results.

[0004] To address the aforementioned issues, this invention proposes a multi-dimensional feature-based comprehensive prediction method based on the Analytic Hierarchy Process (AHP). This method determines the optimal weights for polling data, futures market data, and betting data through AHP, and then combines these weights with the day's prediction data to generate more accurate election predictions. This approach not only overcomes the limitations of single prediction techniques but also improves the overall reliability and interpretability of the prediction results, providing a novel solution for predicting election outcomes.

[0005] In a first aspect, embodiments of this application provide a multidimensional feature comprehensive prediction method based on the analytic hierarchy process (AHP), the method comprising:

[0006] Based on the poll data and poll model of the day, the poll model prediction results are obtained, and the poll model prediction results indicate the predicted vote share.

[0007] Based on the futures price series data and the futures model, the futures model prediction results are obtained, and the futures model prediction results indicate the vote rate prediction results;

[0008] Based on the daily odds sequence of betting data and the betting model, the betting model prediction results are obtained, and the betting model prediction results indicate the vote rate prediction results;

[0009] Using the prediction results from the public opinion poll model, the futures model, and the betting model, the final prediction result for the day is obtained based on the hierarchical comprehensive prediction model.

[0010] In a specific embodiment of the present invention, the above-mentioned prediction results using the polling model, futures model, and betting model are used to obtain the final prediction result for the day based on a hierarchical comprehensive prediction model, specifically including:

[0011] Based on the time points, the prediction results of the polling model, the futures model, and the betting model are matched according to the time series.

[0012] Based on the importance assessment values ​​between the predictions of the public opinion polling model, the futures model, and the betting model, a judgment matrix is ​​established.

[0013] The weight vector is obtained by using the eigenvector corresponding to the largest eigenvalue of the judgment matrix;

[0014] Based on the weight vector, a hierarchical comprehensive prediction model is obtained;

[0015] The results of the polling model, the futures model, and the betting model are input into the hierarchical integrated prediction model to obtain the final prediction result for the day.

[0016] In a specific embodiment of the present invention, obtaining the poll model prediction result based on the poll data and the poll model of the day specifically includes:

[0017] By using data from previous social polls, we obtained the overall scores and institutional effect ratings of the polling organizations included in each constituency.

[0018] By using the overall scores and institutional effect ratings of various polling organizations, the polling model is trained to obtain the optimal strategy and parameter configuration;

[0019] Based on the poll data of the day and the poll model with the optimal strategy and parameter configuration, the poll model prediction results are obtained.

[0020] In a specific embodiment of the present invention, obtaining the futures model prediction result based on the daily futures price sequence data and the futures model specifically includes:

[0021] Obtain the futures price sequence data for the current day, which includes futures price sequence data based on timestamps;

[0022] After cleaning and stabilizing the daily futures price series data, we obtain daily futures price series data that meets the stationarity requirement.

[0023] The parameters of the futures model are determined using the autocorrelation function and the partial autocorrelation function. The futures model is an ARIMA model.

[0024] Input the futures price series data of the day into the futures model to obtain the futures price prediction trend, where the futures price prediction trend indicates the predicted value of the win rate;

[0025] The prediction results of the futures model are obtained by using the predicted win rate and the inverse cumulative distribution function of the standard normal distribution.

[0026] In a specific embodiment of the present invention, the above-mentioned method of obtaining a betting model prediction result based on the daily odds sequence of betting data and the betting model, wherein the betting model prediction result indicates the vote rate prediction result, specifically includes:

[0027] Obtain historical odds sequences from betting data and perform data cleaning on these sequences to ensure their continuity and integrity.

[0028] By using the historical odds sequence of each candidate, the daily odds data of each candidate is converted into the vote percentage, resulting in the vote percentage sequence;

[0029] The vote rate sequence is stabilized, and the parameters of the betting model are determined using the autocorrelation function and partial autocorrelation function.

[0030] Obtain the odds sequence for the day, convert the training sequence for the day into the vote rate sequence for the day, and then input the vote rate sequence into the betting model to obtain the prediction result of the betting model.

[0031] In a specific embodiment of the present invention, the above-mentioned method of obtaining the overall score and institutional effect rating of each polling organization included in each constituency through historical social poll data specifically includes:

[0032] Collect and organize the last polling data results of various organizations in the selected electoral districts, and convert the polling data results into vote percentages to obtain the vote percentages of the polling organizations in previous years;

[0033] The vote percentages of polling organizations in previous years were normalized by year to obtain the vote percentages of polling organizations in yearly units.

[0034] Collect the actual results of previous elections in various regions to obtain the vote share of candidates in each election region on an annual basis;

[0035] The error of each polling organization in the electoral district over the years is calculated based on the vote share of polling organizations on an annual basis and the vote share of candidates in the electoral district on an annual basis.

[0036] After normalizing the errors of various polling organizations in the electoral region over the years, they are ranked. Based on the preset ranking level distribution ratio, the institutional effect of the polling organizations included in the electoral region in each year is rated to determine their level.

[0037] The overall scores of each polling organization were calculated based on the year weight and the number of surveys, and the overall scores were normalized.

[0038] The normalized overall scores of all polling organizations in the electoral district are ranked, and the ranking is determined according to the preset distribution ratio of the ranking levels. This yields the comprehensive institutional effect rating of all polling organizations in the electoral district.

[0039] In a specific embodiment of the present invention, the above-mentioned use of the overall scores and institutional effect ratings of various polling organizations to train the polling model and obtain the optimal strategy and parameter configuration specifically includes:

[0040] By combining institutions with different ratings, each combination of institutions can be considered a strategy, and all combinations of institutions together constitute the strategy space.

[0041] For the polling data included under each strategy, time series data is generated by sorting the data according to the release time of the polling agencies' forecasts for each year.

[0042] For time series forecasts for each year, a weighted average is calculated using the exponential moving average method to obtain the polling model forecasts for all candidates.

[0043] For the undecided group, the polling model predictions for all candidates were normalized by linear weighting and exponential adjustment, and normalization was performed using two strategies: herd mentality and spiral of silence.

[0044] For each type of institutional strategy and normalization strategy in each year, there are polling model prediction results. In order to minimize the error of the prediction result corresponding to the candidate with the highest prediction result and the corresponding actual vote result, an objective function is constructed, and the objective function optimizes the parameters of the polling model.

[0045] The gradient descent method is used to solve the minimization problem, and the result of minimizing the prediction error is output. Under each institutional strategy and normalization strategy, the corresponding optimal prediction result is calculated to form a set of prediction results for all strategies.

[0046] By comparing the errors between the predicted results of each strategy and the actual vote results, the institutional strategy, normalization strategy, and corresponding weight parameters that minimize the error are selected as the optimal strategy and parameter configuration for the polling model.

[0047] Secondly, embodiments of this application provide a multi-dimensional feature comprehensive prediction device based on the analytic hierarchy process (AHP), the device comprising:

[0048] The first processing module is used to obtain the poll model prediction results based on the poll data and poll model of the day. The poll model prediction results indicate the predicted vote share.

[0049] The second processing module is used to obtain the futures model prediction results based on the futures price sequence data and the futures model. The futures model prediction results indicate the vote rate prediction results.

[0050] The third processing module is used to obtain the betting model prediction result based on the daily odds sequence of betting data and the betting model. The betting model prediction result indicates the vote rate prediction result.

[0051] The fourth processing module is used to obtain the final prediction result for the day by utilizing the prediction results of the public opinion poll model, the futures model, and the betting model, based on the hierarchical comprehensive prediction model.

[0052] In a specific embodiment of the present invention, the fourth processing module includes:

[0053] Based on the time points, the prediction results of the polling model, the futures model, and the betting model are matched according to the time series.

[0054] Based on the importance assessment values ​​between the predictions of the public opinion polling model, the futures model, and the betting model, a judgment matrix is ​​established.

[0055] The weight vector is obtained by using the eigenvector corresponding to the largest eigenvalue of the judgment matrix;

[0056] Based on the weight vector, a hierarchical comprehensive prediction model is obtained;

[0057] The results of the polling model, the futures model, and the betting model are input into the hierarchical integrated prediction model to obtain the final prediction result for the day.

[0058] In a specific embodiment of the present invention, the first processing module includes:

[0059] By using data from previous social polls, we obtained the overall scores and institutional effect ratings of the polling organizations included in each constituency.

[0060] By using the overall scores and institutional effect ratings of various polling organizations, the polling model is trained to obtain the optimal strategy and parameter configuration;

[0061] Based on the poll data of the day and the poll model with the optimal strategy and parameter configuration, the poll model prediction results are obtained.

[0062] In a specific embodiment of the present invention, the second processing module includes:

[0063] Obtain the futures price sequence data for the current day, which includes futures price sequence data based on timestamps;

[0064] After cleaning and stabilizing the daily futures price series data, we obtain daily futures price series data that meets the stationarity requirement.

[0065] The parameters of the futures model are determined using the autocorrelation function and the partial autocorrelation function. The futures model is an ARIMA model.

[0066] Input the futures price series data of the day into the futures model to obtain the futures price prediction trend, where the futures price prediction trend indicates the predicted value of the win rate;

[0067] The prediction results of the futures model are obtained by using the predicted win rate and the inverse cumulative distribution function of the standard normal distribution.

[0068] In a specific embodiment of the present invention, the third processing module includes:

[0069] Obtain historical odds sequences from betting data and perform data cleaning on these sequences to ensure their continuity and integrity.

[0070] By using the historical odds sequence of each candidate, the daily odds data of each candidate is converted into the vote percentage, resulting in the vote percentage sequence;

[0071] The vote rate sequence is stabilized, and the parameters of the betting model are determined using the autocorrelation function and partial autocorrelation function.

[0072] Obtain the odds sequence for the day, convert the training sequence for the day into the vote rate sequence for the day, and then input the vote rate sequence into the betting model to obtain the prediction result of the betting model.

[0073] In a specific embodiment of the present invention, the first processing module includes:

[0074] Collect and organize the last polling data results of various organizations in the selected electoral districts, and convert the polling data results into vote percentages to obtain the vote percentages of the polling organizations in previous years;

[0075] The vote percentages of polling organizations in previous years were normalized by year to obtain the vote percentages of polling organizations in yearly units.

[0076] Collect the actual results of previous elections in various regions to obtain the vote share of candidates in each election region on an annual basis;

[0077] The error of each polling organization in the electoral district over the years is calculated based on the vote share of polling organizations on an annual basis and the vote share of candidates in the electoral district on an annual basis.

[0078] After normalizing the errors of various polling organizations in the electoral region over the years, they are ranked. Based on the preset ranking level distribution ratio, the institutional effect of the polling organizations included in the electoral region in each year is rated to determine their level.

[0079] The overall scores of each polling organization were calculated based on the year weight and the number of surveys, and the overall scores were normalized.

[0080] The normalized overall scores of all polling organizations in the electoral district are ranked, and the ranking is determined according to the preset distribution ratio of the ranking levels. This yields the comprehensive institutional effect rating of all polling organizations in the electoral district.

[0081] In a specific embodiment of the present invention, the first processing module includes:

[0082] By combining institutions with different ratings, each combination of institutions can be considered a strategy, and all combinations of institutions together constitute the strategy space.

[0083] For the polling data included under each strategy, time series data is generated by sorting the data according to the release time of the polling agencies' forecasts for each year.

[0084] For time series forecasts for each year, a weighted average is calculated using the exponential moving average method to obtain the polling model forecasts for all candidates.

[0085] For the undecided group, the polling model predictions for all candidates were normalized using two strategies: conformity and the spiral of silence effect.

[0086] For each type of institutional strategy and normalization strategy in each year, there are polling model prediction results. In order to minimize the error of the prediction result corresponding to the candidate with the highest prediction result and the corresponding actual vote result, an objective function is constructed, and the objective function optimizes the parameters of the polling model.

[0087] The gradient descent method is used to solve the minimization problem, and the result of minimizing the prediction error is output. Under each institutional strategy and normalization strategy, the corresponding optimal prediction result is calculated to form a set of prediction results for all strategies.

[0088] By comparing the errors between the predicted results of each strategy and the actual vote results, the institutional strategy, normalization strategy, and corresponding weight parameters that minimize the error are selected as the optimal strategy and parameter configuration for the polling model.

[0089] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the multidimensional feature comprehensive prediction method based on the analytic hierarchy process.

[0090] Fourthly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the multidimensional feature comprehensive prediction method based on the analytic hierarchy process as described above.

[0091] Compared with existing technologies, it has the following outstanding advantages:

[0092] This invention's method incorporates institutional effect ratings to screen poll data and combines dynamic information from futures and odds markets, integrating the advantages of three prediction methods. This effectively reduces bias caused by the limitations of a single data source and significantly improves the accuracy of election prediction results. By combining information from different dimensions such as polls and market trading behavior, the prediction results are more comprehensive, clearly reflecting the multiple influencing factors of voter intentions and possessing strong interpretability. Employing exponential moving averages and time-series prediction models, it reflects both long-term trends and captures dynamic changes closer to the election, allowing for timely adjustments to the prediction results to cope with rapidly changing election environments. Integrating multi-source data avoids over-reliance on single methods and resources, reduces dependence on large-scale opinion polls, and minimizes the impact of sampling bias and data noise, further improving resource utilization efficiency and the stability of the prediction model. Furthermore, this method is applicable to predictions in single electoral districts as well as nationwide election result forecasts, possessing broad applicability and promotional value. This invention provides an efficient and reliable solution for result prediction in complex election environments. Attached Figure Description

[0093] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0094] Figure 1 This is a flowchart illustrating the multidimensional feature comprehensive prediction method based on the analytic hierarchy process in an embodiment of the present invention.

[0095] Figure 2 This is a flowchart illustrating the multidimensional feature comprehensive prediction method based on the analytic hierarchy process in an embodiment of the present invention.

[0096] Figure 3 This is a schematic diagram of the multidimensional feature comprehensive prediction device based on the analytic hierarchy process in an embodiment of the present invention;

[0097] Figure 4 This is a schematic diagram of the computer hardware of the present invention. Detailed Implementation

[0098] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0099] It should also be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0100] It should also be understood that, in various embodiments of the present invention, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0101] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0102] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0103] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0104] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0105] To make the above-mentioned features and effects of the present invention clearer and easier to understand, specific embodiments are described below in conjunction with the accompanying drawings. This specification discloses one or more embodiments incorporating the features of the present invention. The disclosed embodiments are merely illustrative. The scope of protection of the present invention is not limited to the disclosed embodiments, but is defined by the appended claims.

[0106] The following are system embodiments corresponding to the above method embodiments. This embodiment can be implemented in conjunction with the above embodiments. The relevant technical details mentioned in the above embodiments are still valid in this embodiment, and will not be repeated here to reduce repetition. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the above embodiments.

[0107] In election forecasting, economic indicator-based methods can explain the long-term impact of the economy on voter behavior. These methods use economic data (such as unemployment rate, GDP growth rate, and inflation rate) to predict election results. They comprehensively analyze national macroeconomic indicators and construct predictive models to estimate the vote share in the general election. Their theoretical basis is the influence of economic conditions on voter behavior. The advantage of this method is that it does not rely directly on voter intention data, avoiding sampling bias found in opinion polls. However, its disadvantages include relatively simplistic model assumptions, neglecting the influence of socio-cultural factors or candidate-specific factors on voter choices; the models are based on long-term historical data, making it difficult to reflect short-term changes before the election; and the predictive accuracy is limited when candidates are evenly matched.

[0108] In election prediction, social media and big data analytics methods analyze voter discussions and sentiments on social media platforms to extract public opinion trends and predict election results. The advantage is that the data is updated in real time, capturing rapidly changing voter sentiment. The disadvantages are that the platform's user base may not be representative, and the algorithm needs to process large amounts of noisy data, potentially leading to unstable prediction results.

[0109] To address the above issues, this application proposes to improve the accuracy of election prediction by combining multi-source data and integrating the prediction results of three different methods. It innovates upon existing prediction techniques.

[0110] An improved social opinion polling data prediction model is developed: Historical polling data is graded and filtered using institutional effects, selecting only high-rated results as input. An exponential moving average method is used to calculate the current vote share of more authoritative polls in each constituency, reducing errors caused by sample bias or institutional bias in traditional polls. Furthermore, to address the undecided factor in polling data, social mentality modeling is conducted by incorporating herd mentality and the spiral of silence effect, reducing the influence of undecided groups on the prediction results.

[0111] A futures market data forecasting model is introduced: a time-series forecasting model is used to analyze the changing trend of the win rate in the futures market and convert it into the vote percentage. This method, based on market trading behavior, can effectively reflect investors' long-term expectations of the election results.

[0112] Applying political betting data prediction models: Similarly, time-series prediction models are used to process odds win rate data and convert it into vote rate, thereby capturing the public's immediate opinion on the election results.

[0113] Finally, the optimal weights for the three methods were determined using hierarchical analysis, and these weights were combined with the predicted data for the day to generate more accurate election predictions. This method not only overcomes the limitations of single prediction techniques but also improves the overall reliability and interpretability of the prediction results, providing a completely new solution for predicting election outcomes.

[0114] This invention aims to overcome the problems of single prediction methods, large result bias, and insufficient timeliness in the existing technologies. It proposes a multi-dimensional feature comprehensive prediction modeling method based on the analytic hierarchy process for social polls, futures markets, and political betting data.

[0115] The following key technical points emerged during the development of this invention:

[0116] 1. This paper proposes a social polling data modeling method based on multi-dimensional time-series analysis and social mentality modeling. This method employs a hierarchical modeling structure, covering both inter-election and intra-election levels. It also incorporates time-series analysis techniques, such as autoregressive models, moving averages, and sliding windows, to more accurately capture the changing trends of polling data and improve the stability of the data analysis results. Furthermore, addressing the undecided factors in polling data, this method combines conformity psychology and the spiral of silence effect to model social mentality, comprehensively and accurately reflecting public opinion and reducing the influence of undecided groups on prediction results. The technical benefits of this method are that it optimizes the analytical framework of polling data, improves the stability and accuracy of prediction results, thereby enhancing the overall credibility of social polls and providing more reliable data support for policy making, election prediction, and social sentiment research.

[0117] 2. This paper proposes a futures market data modeling method based on time-series predictive analysis. This method transforms the futures election win rate into the vote rate and uses time-series analysis techniques to model futures market trading data, identifying the dynamic relationship between futures market behavior and election result expectations. The technical advantage of this method is that it fully utilizes market behavior data, enhances the sensitivity of the predictive model to economic and investment expectations, and improves the accuracy and practicality of predictions.

[0118] 3. This paper proposes a political betting data modeling method based on time-series predictive analysis. This method transforms political betting data into vote percentages and utilizes time-series analysis techniques to mine the betting tendencies of the public in the betting market, identifying their correlation with election results, thereby constructing a more real-time and sensitive predictive model. The technical effect of this method is that by combining the betting tendencies of the public, it enhances the predictive model's ability to dynamically capture election results, thus improving the real-time performance and reliability of election predictions.

[0119] 4. A multi-dimensional feature-based comprehensive prediction modeling method based on the analytic hierarchy process (AHP) is proposed. This method optimizes the weight allocation of multiple prediction models through hierarchical analysis, enabling them to work synergistically across different dimensions to generate the final prediction result. The technical advantage of this method is that it fully leverages the strengths of each prediction model, effectively compensates for the limitations of a single model, improves the overall accuracy and reliability of the prediction results, makes election predictions more robust, and provides stronger data support for election analysis and decision-making.

[0120] The methods of this application will be described in detail below with reference to specific embodiments:

[0121] like Figure 1 As shown in the figure, this application provides a multi-dimensional feature comprehensive prediction method based on the analytic hierarchy process, including the following steps:

[0122] 110. Based on the poll data and poll model of the day, the poll model prediction results are obtained, and the poll model prediction results indicate the predicted vote share.

[0123] 120. Based on the futures price series data and the futures model, the futures model prediction results are obtained, and the futures model prediction results indicate the vote rate prediction results.

[0124] 130. Based on the daily odds sequence of betting data and the betting model, the betting model prediction results are obtained, and the betting model prediction results indicate the vote rate prediction results.

[0125] 140. Using the forecast results from the public opinion poll model, the forecast results from the futures model, and the forecast results from the betting model, the final forecast result for the day is obtained according to the hierarchical comprehensive forecast model.

[0126] In some embodiments, step 140 specifically includes:

[0127] 141. Based on the time points, match the prediction results of the public opinion model, the futures model, and the betting model according to the time series.

[0128] 142. Based on the importance assessment values ​​between the prediction results of the public opinion poll model, the futures model, and the betting model, establish a judgment matrix.

[0129] 143. Obtain the weight vector based on the eigenvector corresponding to the largest eigenvalue of the judgment matrix.

[0130] 144. Based on the weight vector, a hierarchical comprehensive prediction model is obtained.

[0131] 145. Input the forecast results from the public opinion poll model, the futures model, and the betting model into the hierarchical comprehensive forecast model to obtain the final forecast result for the day.

[0132] In some embodiments, step 110 specifically includes:

[0133] 111. By using data from previous social polls, obtain the overall scores and institutional effect ratings of the polling organizations included in each constituency.

[0134] 112. Using the overall scores and institutional effect ratings of various polling organizations, train the polling model to obtain the optimal strategy and parameter configuration.

[0135] 113. Based on the poll data of the day and the poll model with the optimal strategy and parameter configuration, the prediction result of the poll model is obtained.

[0136] In some embodiments, step 120 specifically includes:

[0137] Obtain the futures price series data for the current day, which includes futures price series data with timestamps as the sequence.

[0138] After cleaning and stabilizing the daily futures price series data, we obtain daily futures price series data that meets the stationarity requirement.

[0139] The parameters of the futures model are determined using the autocorrelation function and the partial autocorrelation function. The futures model is an ARIMA model.

[0140] Input the futures price series data of the day into the futures model to obtain the futures price prediction trend, where the futures price prediction trend indicates the predicted value of the win rate.

[0141] The prediction results of the futures model are obtained by using the predicted win rate and the inverse cumulative distribution function of the standard normal distribution.

[0142] In some embodiments, step 130 specifically includes:

[0143] Obtain historical odds sequences from betting data and perform data cleaning on these sequences to ensure their continuity and integrity.

[0144] By using the historical odds sequence of each candidate, the daily odds data of each candidate is converted into the vote percentage, resulting in the vote percentage sequence.

[0145] The vote rate sequence is stabilized, and the parameters of the betting model are determined using the autocorrelation function and partial autocorrelation function.

[0146] Obtain the odds sequence for the day, convert the training sequence for the day into the vote rate sequence for the day, and then input the vote rate sequence into the betting model to obtain the prediction result of the betting model.

[0147] In some embodiments, step 111 specifically includes:

[0148] Collect and organize the last polling data results from various organizations in the selected electoral districts, and convert the polling data results into vote percentages to obtain the vote percentages of each polling organization.

[0149] The vote percentages of polling organizations in previous years were normalized by year to obtain the vote percentages of polling organizations in yearly units.

[0150] Collect the actual results of previous elections in the region to obtain the vote share of candidates in the election region on an annual basis.

[0151] The error of each polling organization in the electoral district over the years is calculated based on the vote share of polling organizations on an annual basis and the vote share of candidates in the electoral district on an annual basis.

[0152] After normalizing the errors of various polling organizations in the electoral region over the years, they are ranked. Based on the preset distribution ratio of the ranking levels, the institutional effects of the polling organizations included in the electoral region in each year are rated to determine their level.

[0153] The overall scores of each polling organization were calculated based on the year weight and the number of surveys, and then normalized.

[0154] The normalized overall scores of all polling organizations in the electoral district are ranked, and the ranking is determined according to the preset distribution ratio of the ranking levels. This yields the comprehensive institutional effect rating of all polling organizations in the electoral district.

[0155] In some embodiments, step 112 specifically includes:

[0156] By combining institutions with different ratings, each combination of institutions can be considered a strategy, and all combinations together constitute the strategy space.

[0157] For the polling data included under each strategy, time series data is generated by sorting the data according to the release time of the polling agency's forecast data for each year.

[0158] For time series forecasts for each year, a weighted average is calculated using the exponential moving average method to obtain the polling model forecasts for all candidates.

[0159] For the undecided group, the polling model predictions for all candidates were normalized using two strategies: conformity and the spiral of silence effect.

[0160] For each type of institutional strategy and normalization strategy in each year, there are polling model prediction results. In order to minimize the error of the prediction result corresponding to the candidate with the highest prediction result and the corresponding actual vote result, an objective function is constructed, and the objective function optimizes the parameters of the polling model.

[0161] The gradient descent method is used to solve the minimization problem, and the result of minimizing the prediction error is output. Under each institutional strategy and normalized strategy, the corresponding optimal prediction result is calculated to form a set of prediction results for all strategies.

[0162] By comparing the errors between the predicted results of each strategy and the actual vote results, the institutional strategy, normalization strategy, and corresponding weight parameters that minimize the error are selected as the optimal strategy and parameter configuration for the polling model.

[0163] Compared with the prior art, the present invention has the following beneficial effects:

[0164] 1. Improve prediction accuracy: By introducing institutional effect rating to screen poll data and combining dynamic information from the futures and odds markets, the advantages of the three prediction methods are combined, effectively reducing the bias caused by the limitations of a single data source and significantly improving the accuracy of election prediction results.

[0165] 2. Enhance the explanatory power of the prediction model: By combining information from different dimensions such as public opinion polls and market transaction behavior, the prediction results are more comprehensive and can more clearly reflect the multiple influencing factors of voters' intentions, thus possessing a strong ability to explain the results.

[0166] 3. Improve timeliness: By adopting the exponential moving average method and time series forecasting model, it can reflect long-term trends and capture dynamic changes close to the election, so that the forecast results can be adjusted in a timely manner to cope with the rapidly changing election environment.

[0167] 4. Optimize resource utilization: Integrate multi-source data to avoid over-reliance on single methods and resources, reduce dependence on large-scale opinion polls, and reduce the impact of sampling bias and data noise, thereby further improving resource utilization efficiency and the stability of prediction models.

[0168] 5. Strong applicability: The method of this invention is applicable to both prediction of election results in a single electoral district and prediction of election results nationwide, and has a wide range of applicable scenarios and promotional value.

[0169] In summary, this invention outperforms existing technologies in terms of accuracy, timeliness, explanatory power, and resource optimization in election prediction, providing an efficient and reliable solution for predicting results in complex election environments.

[0170] The method of this application will now be described in detail with reference to another specific embodiment. For example... Figure 2 As shown in the figure, this application provides a multi-dimensional feature comprehensive prediction method based on the analytic hierarchy process, including the following steps:

[0171] Step 1: Calculation of the social polling model

[0172] 1) Rating the effectiveness of polling organizations by using data from previous social polls.

[0173] ① Collect and organize the results of the last public opinion poll conducted by various organizations in the selected electoral districts in previous terms;

[0174] ② Poll data is usually calculated based on the number of votes a candidate receives. Therefore, when analyzing this data, it needs to be converted into a vote percentage. The formula for this is:

[0175]

[0176] in This represents the predicted vote share of the candidate by the nth polling organization in year t. This represents the number of votes predicted by the nth polling organization in year t. This represents the total number of votes predicted by the nth polling organization in year t, where N is the number of polling organizations in that electoral district in that year, and T is the total number of years used for rating the institutional effect within the selected time range.

[0177] ③ To analyze the overall effect of polling organizations over the years, we normalized the vote share of the nth polling organization in year t within the selected time range. The calculation formula is as follows:

[0178]

[0179] Where N is the number of polling organizations in that election district in that year, and T is the total number of years used for rating the institutional effect within the selected time range.

[0180] ④ For the actual results in this election region in year t, convert them into vote percentages. The calculation formula is as follows:

[0181]

[0182] in Let represent the actual number of poll votes received by the candidate in year t. This represents the actual total number of votes in that constituency in year t.

[0183] ⑤ For the nth polling agency in the t-th year of the electoral district, calculate the polling agency error of the candidate. The calculation formula is as follows:

[0184]

[0185] Where N represents the number of institutions in that electoral district in that year, and T represents the total number for that year.

[0186] ⑥ The polling organization for this candidate is incorrect. (n,t) To normalize to the [0,1] interval, the normalization formula is as follows:

[0187]

[0188] ⑦ In year t, the N polling organizations included are normalized to the organizational error value calculated in ⑥ for that year. Sort the polling organizations and, based on the preset sorting level distribution ratio, rate the institutional effect of the N polling organizations in the electoral region in year t, and determine their respective levels.

[0189] ⑧ Based on ⑦, calculate the overall scores of N polling organizations within the selected time range according to the year weight and the number of surveys. Taking polling organization n1 as an example, its overall score is... The calculation method is as follows:

[0190]

[0191] Where α1 is the weighting coefficient for the number of polls, and N total It is the total number of polls conducted by organization n1, E weighted This is the overall error value, where T is the total number of years, and w t This is the weight of the institution in year t. It is the normalized error of institution n1 in year t, and its formula is:

[0192]

[0193] 9. Repeat step 8 for each polling organization to calculate the overall score for each organization.

[0194] ⑩ Normalize the overall scores of each polling organization to the [0,1] interval. The normalization formula is as follows:

[0195]

[0196] Where N represents the number of polling organizations in that constituency. It is the overall score of the nth polling organization. It is the normalized score of the nth polling organization.

[0197] The normalized overall scores of all polling organizations in the electoral district are ranked, and the ranking is determined according to the preset distribution ratio of the ranking levels. This yields the comprehensive institutional effect rating of all polling organizations in the electoral district.

[0198] Repeat for each electoral district The process yields the overall scores and institutional effect ratings for each polling organization within each constituency.

[0199] 2) Learning the optimal strategy for the polling model using historical polling data: Historical polling data is processed to calculate the overall institutional effect score and rating of all institutions, serving as the basis for model optimization. Subsequently, a strategy for selecting institutions is formulated, and the institutional data included in the selected strategy is extracted. The polling data for each electoral region is then sorted according to the poll release date to construct a time series. An exponential moving average is then used to apply a weighted average to the time series data, generating the polling model output for each electoral region in that year. During this process, the optimal strategy S and the parameters of the exponential moving average are optimized using gradient descent to minimize the model's prediction error on historical data. Furthermore, to address the influence of undecided groups, conformity and the spiral of silence effect are used for normalization adjustments, further improving the reliability and applicability of the polling model.

[0200] ① Define the institutional strategy space S∈{S1,S2,…,S} M} where M is the total number of strategies. This invention combines institutions with different ratings, with each combination of institutions serving as a strategy, and all combinations of institutions together constitute the strategy space S.

[0201] ② For the polling data included under each strategy, time series data is generated by sorting the data according to the release time of the polling agency's predictions in year t. Then, the prediction data is weighted and processed in conjunction with the overall score of that polling agency. The calculation process is as follows:

[0202]

[0203] Where P S (l,t,α1) represents the predicted value of the time series after weight adjustment in year t; This represents the normalized vote share of the l-th polling organization in the time series corresponding to this strategy in year t, which is calculated from step 1) ③. The overall score of the l-th polling organization in the time series is calculated by step 1) ⑩. α1 is the weighting coefficient of the number of polls, and L is the length of the polling data time series under this strategy.

[0204] ③ For the predicted value P of the time series in year t S (l,t,α1), this invention uses the exponential moving average method to perform a weighted average to obtain the polling model prediction results, and the formula is as follows:

[0205] Y predict (α1,α2,t)=EMA(α1,α2,t) (10)

[0206] EMA(α1,α2,t) is an exponential moving average weighting function, which is calculated recursively. This effectively assigns higher weights to recent data, ensuring that the prediction results are sensitive to the latest information.

[0207] ④ Repeat steps ①-③ for each candidate to obtain the following set of polling model predictions for all candidates:

[0208] Y i (α1,α2,t),i=1,2,...,I ​​(11)

[0209] Where Y i (α1,α2,t) represents the poll model prediction result for the i-th candidate; I represents the total number of candidates.

[0210] ⑤ For the undecided group, two strategies can be used for normalization: conformity and the spiral of silence effect. Conformity can be represented by linear weighted normalization, with the following formula:

[0211] Y norm,i (α1,α2,t)=Y i (α1,α2,t)·(1+U),i=1,2,...,I ​​(12)

[0212] Where Y norm,i (α1, α2, t) represents the normalized result of the i-th candidate; U is the proportion of the undecided group, satisfying 0 ≤ U ≤ 1. This normalization method ensures that undecided voters are distributed according to the distribution of declared voters, which conforms to the herd mentality.

[0213] The spiral of silence effect can be represented using exponentially adjusted normalization, as shown in the following formula:

[0214]

[0215] Where k is the spiral reinforcement coefficient, usually taken as k=1. This normalization method reflects the public opinion effect, that is, the leader is more likely to attract the support of undecided voters, making the prediction more in line with the laws of social psychology.

[0216] ⑥ For each type of institutional strategy and normalization strategy in year t, there is a polling model prediction result Y. norm,i (α1, α2, t), select the prediction result corresponding to the candidate i1 with the highest prediction result. and the corresponding actual election results To ensure that the weight parameters α1 and α2 are optimized within the selected time range to minimize the error between the poll model's predicted results and the actual vote count, the objective function of the optimization problem is defined as follows:

[0217]

[0218] ⑦ Use the gradient descent method to solve the minimization problem and output the solution that minimizes the prediction error.

[0219] ⑧ Under each institutional strategy and normalization strategy, calculate the corresponding optimal prediction result to form a set of prediction results for all strategies. By comparing the errors between the prediction results of each strategy and the actual vote results, select the institutional strategy S that minimizes the error. m Normalization strategy and corresponding weight parameters This serves as the optimal strategy and parameter configuration for the polling model.

[0220] 3) Output of the polling model results for this election

[0221] ① Based on the optimal strategy and parameter configuration obtained in step 2), select the corresponding polling organization and generate time series data by sorting the release time of the polling organization's prediction data. Then, combine the overall effect score of the polling organization to weight the prediction data. The specific calculation process is as follows:

[0222]

[0223] Where L now For the optimal strategy S m The length of the time series formed by the poll data; t now This indicates that the selected time frame only includes this election. The normalized vote share of the l-th polling organization in the current polling time series is calculated by the method in step 1) ③. The overall score of the l-th polling organization in the time series is calculated using the method in step 1) ⑩. The weighting coefficient for the number of polls with the smallest overall error in the historical poll data; This represents the time-series forecast value after weighting the current poll data.

[0224] ② The time series forecast values ​​are weighted and averaged using the exponential moving average method to obtain the daily forecast result of the polling model. The calculation formula is as follows:

[0225]

[0226] in The weights of the smoothing factors that minimize the overall error of the poll data from previous years. It is an exponential moving average weighted function.

[0227] ③ Based on the normalization strategy determined in step 2), the prediction results are normalized and output according to the formula described in step 2) ⑤.

[0228] ④ Update the poll data input daily, repeat steps ①-③ to output and record the poll model results for the day, and generate a visualization curve of the poll model prediction results.

[0229] Step Two: Calculation of the Futures Market Model

[0230] ① Data Acquisition: Obtain futures price series data P containing timestamp t. t This includes, but is not limited to, core indicators such as opening price, closing price, highest price, and lowest price.

[0231] ② Data cleaning: Preprocess the raw data and fill in missing values ​​by averaging the corresponding values ​​of the preceding and following dates to ensure the continuity and integrity of the time series data.

[0232] ③ Data stationarization: The ADF test is used to determine the stationarity of the time series. For time series that are detected as non-stationary, they are transformed into stationary series through differencing. The differencing formula is as follows:

[0233] ΔP t =P t -P t-1 (17)

[0234] Where, ΔP t If a sequence is not stationary after a first difference, it can be subjected to second or multiple differences until the sequence satisfies stationarity.

[0235] ④ Model Parameter Confirmation: Using the ARIMA time series forecasting model, based on the stationary time series, the parameters of the ARIMA model (autoregressive order p, moving average order q, differencing order d) are determined using the autocorrelation function (ACF) and partial autocorrelation function (PACF). The formula is as follows:

[0236]

[0237] Where P t Let t be the t-th value in the time series, and T be the total length of the time series. ρ is the mean of the time series. k The autocorrelation coefficient φ is the result of fitting an autoregressive model to the value lagged by k, representing the direct impact of subsequent k on the current value. i,k The autoregressive coefficient can be directly calculated from time series data and represents the total correlation between series values.

[0238] In the parameters of the ARIMA model, the autoregressive order p is the maximum lag value that makes the significant correlation between PACF disappear, the moving average order is the maximum lag value that makes the significant correlation between ACF disappear, and the difference order is determined by the data stabilization process.

[0239] ⑤ Model Prediction: After determining the model parameters, the ARIMA model is used to model and predict futures price data in order to obtain future futures price trends.

[0240] ⑥ Data Transformation: Given that futures data can essentially be considered a form of expression of the win rate, the predicted win rate needs to be converted into the corresponding vote percentage. Since the vote percentages of candidates follow a normal distribution, their mean can be used as an estimate of the vote percentage, thus establishing the relationship between the win rate and the vote percentage as follows:

[0241]

[0242] The winning rate is W A The vote share was D A The inverse transformation yields the following conversion formulas for the win rate and the vote rate:

[0243] D A =0.5-σ·Φ -1 (1-W A ) (twenty one)

[0244] Where Φ -1 (x) is the inverse function of the cumulative distribution function of the standard normal distribution. Generally, σ = 3 is taken, at which point the vote rate D... A The calculation depends only on the winning percentage W. A This is the inverse cumulative distribution function of the standard normal distribution. This method can quickly and accurately convert the winning probability into the vote rate for modeling or simulation analysis.

[0245] ⑦ Output the futures model results daily, forming a visualization curve of the futures model prediction results.

[0246] Step 3: Calculation of the Political Betting Model

[0247] ① Data Collection: Obtain the historical odds sequence X = {x1, x2, ..., x} of political betting data. T}, where x t Let T be the odds for day t, and T be the total number of days in the data.

[0248] ② Data cleaning: Preprocess the raw data and fill in missing values ​​by averaging the corresponding values ​​of the preceding and following dates to ensure the continuity and integrity of the time series data.

[0249] ③ Odds Conversion: First, convert the odds data into win rates. Suppose there are two candidates, A and B, and the bookmaker offers odds of 0 for each candidate. A and O B Then the normalized win rates W for candidates A and B are respectively A and W Bfor:

[0250]

[0251] Then convert it into the percentage of votes, the formula is:

[0252] D A =0.5-σ·Φ -1 (1-W A ), D B =0.5-σ·Φ -1 (1-W B ) (twenty three)

[0253] ④ Data stationarization: Perform stationarity tests on the time series, using unit root tests (such as the ADF test) to determine whether the series is stationary. If the series is non-stationary, use the differencing method to transform it into a stationary series;

[0254] ⑤ Parameter selection: Based on the stationary time series, the parameters of the ARIMA model (autoregression order p, moving average order q, and differencing order d) are determined using the autocorrelation function (ACF) and partial autocorrelation function (PACF). The differencing order d is determined by the data stationary process, the autoregression order p is determined by the PACF cutoff point, and the moving average order q is determined by the ACF cutoff point.

[0255] ⑥ Model Prediction: Predict the odds for the next day using the ARIMA model;

[0256] ⑦ Output the political betting model results daily, forming a visualization curve of the political betting model prediction results.

[0257] Step 4: Hierarchical Integrated Prediction Modeling

[0258] ① Time Series Matching: The obtained results from social polls, futures markets, and political betting models are matched one by one according to their time series. This ensures that the results of the three models accurately correspond within the same time period at each point in time, so that subsequent fusion calculations can be performed.

[0259] ② Confirmation of the hierarchical integrated prediction model: The hierarchical analysis method was selected as the hierarchical integrated prediction model to fuse multi-source data. The basic form of this model is as follows:

[0260] y = w1x1 + w2x2 + w3x3 (24)

[0261] Where w1, w2, and w3 are the weights of the results from the social poll, futures market, and political betting models, respectively, and x1, x2, and x3 are the predicted values ​​of the three types of models, respectively.

[0262] ③ Judgment Matrix Construction: Based on the assessment of the relative importance of different data sources, a comparison matrix for hierarchical analysis is constructed. Judgment matrix A is established by comparing each element at each level pairwise.

[0263]

[0264] Where a ij Representation Model M i Compared to model M j The relative importance of factors can be assessed, for example, by determining the relative importance of public opinion polls to a forecasting target versus the relative importance of the futures market. This is typically determined by expert scoring or data-driven methods, and a common evaluation scale is an integer value from 1 to 9, where 1 represents equal importance and 9 represents extremely important.

[0265] ④ Hierarchical weight calculation: The weights are calculated using the analytic hierarchy process (AHP), with the eigenvector method being a commonly used approach. The weight vector is given by the normalized eigenvector corresponding to the largest eigenvalue of matrix A.

[0266] Aw=λ max w (26)

[0267] Where is the largest eigenvalue of the judgment matrix; w = (w1, w2, w3) T The weights for each model satisfy the normalization constraint and can be calculated using the normalized geometric mean method.

[0268] ⑤ Model Validation: After determining the weights, model validation is performed to ensure the high accuracy of the prediction results obtained by the hierarchical analysis method. Backtesting can be conducted using historical data, or the model can be compared with actual results to adjust the weight calculation method or evaluation scale, ensuring the model's reliability.

[0269] ⑥ Model Output: After determining the weights, obtain the predicted values ​​of the three models for the day, and substitute the data into the hierarchical comprehensive prediction model formula to calculate the output for the day. This output result is the predicted value of current election-related indicators (such as vote share) after integrating multi-source data such as social polls, futures markets, and political betting.

[0270] ⑦ Daily Data Update: Update data daily, calculate the predicted values ​​of the three models for social opinion polls, futures market, and political betting, and repeat steps ①-⑥ to generate the output results of the hierarchical comprehensive prediction model for the day.

[0271] Next, the method of this application will be described in detail with reference to another specific embodiment. For example... Figure 3 As shown, a multidimensional feature synthesis prediction device based on the analytic hierarchy process (AHP) is provided. The device includes:

[0272] The first processing module is used to obtain the poll model prediction results based on the poll data and poll model of the day. The poll model prediction results indicate the predicted vote share.

[0273] The second processing module is used to obtain the futures model prediction results based on the futures price sequence data and the futures model. The futures model prediction results indicate the vote rate prediction results.

[0274] The third processing module is used to obtain the betting model prediction result based on the daily odds sequence of betting data and the betting model. The betting model prediction result indicates the vote rate prediction result.

[0275] The fourth processing module is used to obtain the final prediction result for the day by utilizing the prediction results of the public opinion poll model, the futures model, and the betting model, based on the hierarchical comprehensive prediction model.

[0276] In a specific embodiment of the present invention, the fourth processing module includes:

[0277] Based on the time points, the prediction results of the polling model, the futures model, and the betting model are matched according to the time series.

[0278] Based on the importance assessment values ​​between the predictions of the public opinion polling model, the futures model, and the betting model, a judgment matrix is ​​established.

[0279] The weight vector is obtained by using the eigenvector corresponding to the largest eigenvalue of the judgment matrix;

[0280] Based on the weight vector, a hierarchical comprehensive prediction model is obtained;

[0281] The results of the polling model, the futures model, and the betting model are input into the hierarchical integrated prediction model to obtain the final prediction result for the day.

[0282] In a specific embodiment of the present invention, the first processing module includes:

[0283] By using data from previous social polls, we obtained the overall scores and institutional effect ratings of the polling organizations included in each constituency.

[0284] By using the overall scores and institutional effect ratings of various polling organizations, the polling model is trained to obtain the optimal strategy and parameter configuration;

[0285] Based on the poll data of the day and the poll model with the optimal strategy and parameter configuration, the poll model prediction results are obtained.

[0286] In a specific embodiment of the present invention, the second processing module includes:

[0287] Obtain the futures price sequence data for the current day, which includes futures price sequence data with timestamps as the sequence;

[0288] After cleaning and stabilizing the daily futures price series data, the daily futures price series data that satisfies stationarity is obtained.

[0289] The parameters of the futures model, which is an ARIMA model, are determined using the autocorrelation function and the partial autocorrelation function.

[0290] Input the futures price series data of the day into the futures model to obtain the futures price prediction trend, where the futures price prediction trend indicates the predicted value of the win rate;

[0291] The prediction results of the futures model are obtained by using the predicted win rate and the inverse cumulative distribution function of the standard normal distribution.

[0292] In a specific embodiment of the present invention, the third processing module includes:

[0293] Obtain historical odds sequences from betting data and perform data cleaning on these sequences to ensure their continuity and integrity.

[0294] By using the historical odds sequence of each candidate, the daily odds data of each candidate is converted into the vote percentage, resulting in the vote percentage sequence;

[0295] The vote rate sequence is stabilized, and the parameters of the betting model are determined using the autocorrelation function and partial autocorrelation function.

[0296] Obtain the odds sequence for the day, convert the training sequence for the day into the vote rate sequence for the day, and then input the vote rate sequence into the betting model to obtain the prediction result of the betting model.

[0297] In a specific embodiment of the present invention, the first processing module includes:

[0298] Collect and organize the last polling data results of various organizations in the selected electoral districts, and convert the polling data results into vote percentages to obtain the vote percentages of the polling organizations in previous years;

[0299] The vote percentages of polling organizations in previous years were normalized by year to obtain the vote percentages of polling organizations in yearly units.

[0300] Collect the actual results of previous elections in various regions to obtain the vote share of candidates in each election region on an annual basis;

[0301] The error of each polling organization in the electoral district over the years is calculated based on the vote share of polling organizations on an annual basis and the vote share of candidates in the electoral district on an annual basis.

[0302] After normalizing the errors of various polling organizations in the electoral region over the years, they are ranked. Based on the preset ranking level distribution ratio, the institutional effect of the polling organizations included in the electoral region in each year is rated to determine their level.

[0303] The overall scores of each polling organization were calculated based on the year weight and the number of surveys, and the overall scores were normalized.

[0304] The normalized overall scores of all polling organizations in the electoral district are ranked, and the ranking is determined according to the preset distribution ratio of the ranking levels. This yields the comprehensive institutional effect rating of all polling organizations in the electoral district.

[0305] In a specific embodiment of the present invention, the first processing module includes:

[0306] By combining institutions with different ratings, each combination of institutions can be considered a strategy, and all combinations of institutions together constitute the strategy space.

[0307] For the polling data included under each strategy, time series data is generated by sorting the data according to the release time of the polling agencies' forecasts for each year.

[0308] For time series forecasts for each year, a weighted average is calculated using the exponential moving average method to obtain the polling model forecasts for all candidates.

[0309] For the undecided group, the poll model predictions for all candidates were normalized using two strategies: conformity and the spiral of silence effect.

[0310] For each type of institutional strategy and normalization strategy in each year, there are polling model prediction results. In order to minimize the error of the prediction result corresponding to the candidate with the highest prediction result and the corresponding actual vote result, an objective function is constructed, and the objective function optimizes the parameters of the polling model.

[0311] The gradient descent method is used to solve the minimization problem, and the result of minimizing the prediction error is output. Under each institutional strategy and normalization strategy, the corresponding optimal prediction result is calculated to form a set of prediction results for all strategies.

[0312] By comparing the errors between the predicted results of each strategy and the actual vote results, the institutional strategy, normalization strategy, and corresponding weight parameters that minimize the error are selected as the optimal strategy and parameter configuration for the polling model.

[0313] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the multidimensional feature comprehensive prediction method based on the analytic hierarchy process.

[0314] This application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the multidimensional feature comprehensive prediction method based on the analytic hierarchy process as described above.

[0315] In addition, combined Figure 1 The multidimensional feature synthesis prediction method based on the analytic hierarchy process described in this application can be implemented by electronic devices, such as computer devices. Figure 4 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of this application.

[0316] In some embodiments, the computer device may further include a communication interface 83 and a bus 80. For example, Figure 4 As shown, the processor 81, memory 82, and communication interface 83 are connected through bus 80 and complete communication with each other.

[0317] Specifically, the processor 81 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0318] The memory 82 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 81.

[0319] The processor 81 reads and executes the computer program instructions stored in the memory 82 to implement any of the multidimensional feature comprehensive prediction methods based on the analytic hierarchy process in the above embodiments.

[0320] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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.

[0321] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A multidimensional feature comprehensive prediction method based on the analytic hierarchy process (AHP), characterized in that, The method is used in the method including: Based on the poll data and poll model of the day, the poll model prediction results are obtained, and the poll model prediction results indicate the predicted vote share; Based on the futures price series data and the futures model, the futures model prediction results are obtained, and the futures model prediction results indicate the vote rate prediction results; Based on the daily odds sequence of betting data and the betting model, the betting model prediction result is obtained, and the betting model prediction result indicates the vote rate prediction result; Using the forecast results from the polling model, the futures model, and the betting model, the final forecast result for the day is obtained based on the hierarchical comprehensive forecast model.

2. The multidimensional feature comprehensive prediction method based on the analytic hierarchy process according to claim 1, characterized in that, Using the prediction results from the polling model, the futures model, and the betting model, and based on the hierarchical comprehensive prediction model, the final prediction result for the day is obtained, specifically including: Based on the time points, the prediction results of the public opinion model, the prediction results of the futures model, and the prediction results of the betting model are matched according to the time series; Based on the importance assessment values ​​between the polling model prediction results, the futures model prediction results, and the betting model prediction results, a judgment matrix is ​​established. The weight vector is obtained based on the eigenvector corresponding to the largest eigenvalue of the judgment matrix; Based on the weight vector, a hierarchical comprehensive prediction model is obtained; The prediction results from the polling model, the futures model, and the betting model are input into the hierarchical comprehensive prediction model to obtain the final prediction result for the day.

3. The multidimensional feature comprehensive prediction method based on the analytic hierarchy process according to claim 1, characterized in that, The process of obtaining the poll model prediction results based on the poll data and the poll model of the day specifically includes: By using data from previous social polls, we obtained the overall scores and institutional effect ratings of the polling organizations included in each constituency. By using the overall scores and institutional effect ratings of various polling organizations, the polling model is trained to obtain the optimal strategy and parameter configuration; Based on the poll data of the day and the poll model with the optimal strategy and parameter configuration, the prediction result of the poll model is obtained.

4. The multidimensional feature comprehensive prediction method based on the analytic hierarchy process according to claim 1, characterized in that, The process of obtaining futures model prediction results based on the day's futures price sequence data and the futures model specifically includes: Obtain the futures price sequence data for the current day, which includes futures price sequence data with timestamps as the sequence; After cleaning and stabilizing the daily futures price series data, the daily futures price series data that satisfies stationarity is obtained. The parameters of the futures model, which is an ARIMA model, are determined using the autocorrelation function and the partial autocorrelation function. The futures price sequence data for the day is input into the futures model to obtain the futures price prediction trend, wherein the futures price prediction trend indicates the predicted value of the win rate; The prediction results of the futures model are obtained by using the predicted win rate and the inverse cumulative distribution function of the standard normal distribution.

5. The multidimensional feature comprehensive prediction method based on the analytic hierarchy process according to claim 1, characterized in that, The process involves obtaining a betting model prediction result based on the daily odds sequence of betting data and the betting model. This prediction result indicates the predicted vote percentage, and specifically includes: Obtain the historical odds sequence of the betting data, and perform data cleaning on the historical odds sequence to ensure the continuity and integrity of the historical odds sequence; By using the historical odds sequence of each candidate, the daily odds data of each candidate is converted into the vote percentage, resulting in the vote percentage sequence; The vote rate sequence is stabilized, and the parameters of the betting model are determined using the autocorrelation function and the partial autocorrelation function. Obtain the odds sequence for the day, convert the training sequence for the day into the vote rate sequence for the day, and input the vote rate sequence into the betting model to obtain the prediction result of the betting model.

6. The multidimensional feature comprehensive prediction method based on the analytic hierarchy process according to claim 3, characterized in that, The above-mentioned data from previous public opinion polls were used to obtain the overall scores and institutional effect ratings of the various polling organizations included in each constituency, specifically including: Collect and organize the last polling data results of each organization in the selected electoral district, and convert the polling data results into vote percentages to obtain the vote percentages of each polling organization. The vote percentages of the aforementioned polling organizations in previous years were normalized by year to obtain the vote percentages of the polling organizations in yearly units. Collect the actual results of previous elections in various regions to obtain the vote share of candidates in each election region on an annual basis; The error of each polling organization in the electoral district over the years is calculated based on the vote share of polling organizations on an annual basis and the vote share of candidates in the electoral district on an annual basis. After normalizing the errors of various polling organizations in the electoral region over the years, they are ranked. Based on the preset ranking level distribution ratio, the institutional effect of the polling organizations included in the electoral region in each year is rated to determine their level. The overall scores of each polling organization were calculated based on the year weight and the number of surveys, and the overall scores were normalized. The normalized overall scores of all polling organizations in the electoral district are ranked, and the ranking is determined according to the preset distribution ratio of the ranking levels. This yields the comprehensive institutional effect rating of all polling organizations in the electoral district.

7. The multidimensional feature comprehensive prediction method based on the analytic hierarchy process according to claim 3, characterized in that, The method of using the overall scores and institutional effect ratings of various polling organizations to train the polling model and obtain the optimal strategy and parameter configuration specifically includes: By combining institutions with different ratings, each combination of institutions can be considered a strategy, and all combinations of institutions together constitute the strategy space. For the polling data included under each strategy, time series data is generated by sorting the data according to the release time of the polling agencies' forecasts for each year. For time series forecasts for each year, a weighted average is calculated using the exponential moving average method to obtain the polling model forecasts for all candidates. For the undecided group, the poll model prediction results of all candidates are subjected to linear weight normalization and exponential adjustment normalization, and the normalization is performed using two strategies: herd mentality and spiral of silence effect. For each type of institutional strategy and normalization strategy in each year, there are polling model prediction results. In order to minimize the error of the prediction result corresponding to the candidate with the highest prediction result and the corresponding actual vote result, an objective function is constructed. The objective function optimizes the parameters of the polling model. The gradient descent method is used to solve the minimization problem, and the result of minimizing the prediction error is output. Under each institutional strategy and normalization strategy, the corresponding optimal prediction result is calculated to form a set of prediction results for all strategies. By comparing the errors between the predicted results of each strategy and the actual vote results, the institutional strategy, normalization strategy, and corresponding weight parameters that minimize the error are selected as the optimal strategy and parameter configuration for the polling model.

8. A multidimensional feature synthesis prediction device based on the analytic hierarchy process, characterized in that, The device includes: The first processing module is used to obtain the poll model prediction result based on the poll data and poll model of the day, wherein the poll model prediction result indicates the predicted vote share; The second processing module is used to obtain the futures model prediction result based on the futures price sequence data and the futures model, wherein the futures model prediction result indicates the vote rate prediction result; The third processing module is used to obtain the betting model prediction result based on the daily odds sequence of betting data and the betting model, wherein the betting model prediction result indicates the vote rate prediction result; The fourth processing module is used to obtain the final prediction result for the day by utilizing the prediction results of the public opinion model, the futures model, and the betting model, based on the hierarchical comprehensive prediction model.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the multidimensional feature synthesis prediction method based on the analytic hierarchy process as described in any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the multidimensional feature comprehensive prediction method based on the analytic hierarchy process as described in any one of claims 1 to 7.

Images