A modeling and aggregation method and system for multi-feature subjective probability type prediction

Abstract

The application provides a kind of multi-feature subjective probability type prediction modeling and aggregation method and system, comprising: obtaining expert probability type judgment result, and establishing the linear prediction model of expert probability type judgment result based on lens model;Generate prediction result through the linear prediction model, determine the error of prediction result;According to the prediction result containing prediction error, extract potential information source, and decompose problem variable characteristics and expert behavior characteristics;Estimate the weight given by expert to different information source based on the decomposed expert behavior characteristics, cluster expert weight, form multiple cluster groups;From each cluster group, select the expert with the closest estimated weight distance from the cluster center as the representative of each cluster group, and aggregate the probability type judgment through multiple heuristic methods, the application solves the problem that existing multi-element expert probability type judgment prediction is difficult to accurately aggregate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of decision analysis, in particular to a modeling and aggregation method and system for multi-feature subjective probability type prediction. BACKGROUND

[0002] When data is sparse, decision makers need to consult experts for predictions on relevant variables, such as gross domestic product, sales of new products, losses caused by disasters, and the number of times a new movie is played on streaming media. In order to predict a continuous variable, experts are often required to provide multiple probability predictions, which collectively reflect the location, width and shape of the distribution of uncertain variables, and are widely used in meteorology, probability risk assessment, finance, and operational management.

[0003] In practical applications, decision makers not only care about the predictability of expert judgments, but also care about their interpretability, i.e., different sources of information and ways of processing information. Policy makers need to understand the common or different information, motivations and background factors on which expert predictions are based, which is also crucial for decision transparency. The basic goal of aggregating expert predictions is to pool information sources, reconcile different views, and quantify uncertainty, rather than eliminating inconsistencies or uncertainty. Information extraction is particularly important when the predicted results can only be obtained in the distant future, or when peer prediction cannot verify accurate results.

[0004] Existing research on aggregating probability predictions to extract collective wisdom includes simple averaging, weighted averaging based on aggregation performance, prediction correlation analysis based on "shared information problem", and weighted aggregation based on shared information and private information. The above research either calibrates predictions based on historical data or requires special experimental settings (such as coordination mechanisms, competitive game mechanisms, additional estimates of other expert predictions, etc.). The analysis of expert prediction correlation focuses on the covariance matrix of prediction errors, and does not consider the information sources and information processing methods used by each expert prediction when aggregating predictions. SUMMARY

[0005] The present application provides a modeling and aggregation method and system for multi-feature subjective probability type prediction to solve the problem of inaccurate aggregation of existing multi-expert probability type judgment prediction.

[0006] The present application provides a modeling and aggregation method for multi-feature subjective probability type prediction, comprising:

[0007] Obtaining expert probability type judgment results, establishing a linear prediction model of expert probability type judgment results based on a mirror model;

[0008] Generating a prediction result through the linear prediction model, and determining the error of the prediction result;

[0009] According to the prediction result containing the prediction error, potential information sources are extracted, and problem variable characteristics and expert behavior characteristics are decomposed;

[0010] Based on the decomposed expert behavior characteristics, weights given by experts to different information sources are determined, the expert weights are clustered, and multiple cluster groups are formed;

[0011] From each cluster group, an expert with the estimated weight closest to the cluster center is selected as the representative of each cluster group, and a probability type judgment is aggregated through multiple heuristic methods.

[0012] According to the modeling and aggregation method of the multi-feature subjective probability type prediction provided by the application, the expert probability judgment result is obtained, a linear prediction model of the expert probability type judgment result is established based on a lens model, and the linear prediction model comprises:

[0013] The expert probability type judgment includes a probability value of the prediction error and a probability value not containing the prediction error;

[0014] Based on the lens model and the prediction error model, the probability value of the prediction error and the probability value not containing the prediction error are analyzed, and an analysis result is generated;

[0015] According to the weight distribution of the probability judgment not containing the prediction error and the prediction error in the analysis result, a linear prediction model is established.

[0016] According to the modeling and aggregation method of the multi-feature subjective probability type prediction provided by the application, the prediction result is generated through the linear prediction model, and the error of the prediction result is determined, which comprises:

[0017] The sampling error and the judgment error are determined based on the prediction result generated by the linear prediction model;

[0018] The sampling error depends on the random sampling of information clues and how the expert linearly combines the sampled information;

[0019] The judgment error depends on the cognitive error of the expert on the probability level;

[0020] The error of the probability prediction result is determined by combining the sampling error and the judgment error.

[0021] According to the modeling and aggregation method of the multi-feature subjective probability type prediction provided by the application, according to the prediction result containing the prediction error, potential information sources are extracted, and problem variable characteristics and expert behavior characteristics are decomposed, which comprises:

[0022] A probability type prediction matrix is established based on the prediction result containing the prediction error;

[0023] decompose the probabilistic prediction matrix into a problem variable characteristic matrix and an expert behavior characteristic matrix;

[0024] solving the problem variable characteristic matrix and the expert behavior characteristic matrix by maximum likelihood method to generate problem variable characteristics and expert behavior characteristics.

[0025] According to the multi-feature subjective probabilistic prediction modeling and aggregation method provided by the application, the weights given by experts to different information sources are determined based on the decomposed expert behavior characteristics, the expert weights are clustered to form a plurality of clustering groups, and the method comprises the following steps:

[0026] Based on the probabilistic prediction model, the problem variable characteristics and the expert behavior characteristics are solved, and the weights given by experts to the observable information clues and the unobservable information clues are estimated.

[0027] Based on the weight values given by the experts, similar weight values are divided into the same group to complete clustering and form a plurality of clustering groups.

[0028] According to the multi-feature subjective probabilistic prediction modeling and aggregation method provided by the application, the expert whose estimated weight is closest to the clustering center is selected as the representative of each clustering group, and the aggregation of probabilistic judgments is performed by a plurality of heuristic methods, and the method comprises the following steps:

[0029] The clustering center is determined, the weight values given by the experts of each clustering group are compared with the weight values of the clustering center, and a comparison result is generated.

[0030] According to the comparison result, the expert whose estimated weight is closest to the clustering center is selected as the representative of each clustering group.

[0031] The aggregation of probabilistic judgments is performed by a plurality of heuristic methods based on the representative of each clustering group.

[0032] According to the multi-feature subjective probabilistic prediction modeling and aggregation method provided by the application, the method further comprises the following steps:

[0033] Data visualization is performed, and the correlation between each information clue and the problem variable, the weight values given by each expert to different information clues, and the weighted weights of different aggregation methods are displayed in the form of charts.

[0034] The application further provides a multi-feature subjective probabilistic prediction modeling and aggregation system, and the system comprises:

[0035] A linear prediction model establishing module is configured to acquire the expert probabilistic judgment result, and establish a linear prediction model of the expert probabilistic judgment result based on a lens model.

[0036] a prediction error module configured to generate a prediction result by using the linear prediction model, and determine an error of the prediction result;

[0037] an extraction potential information source module configured to extract a potential information source according to the prediction result containing the prediction error, and decompose problem variable features and expert behavior features;

[0038] a clustering grouping module configured to determine weights given by experts to different information sources based on the decomposed expert behavior features, cluster the expert weights, and form a plurality of clustering groups;

[0039] a prediction aggregation module configured to select an expert with an estimated weight closest to a clustering center in each clustering group as a representative of the clustering group, and aggregate a plurality of heuristic methods for probabilistic judgment.

[0040] The application further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the modeling and aggregation method of the multi-feature subjective probabilistic prediction according to any one of the above.

[0041] The application further provides a non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program is executable on a processor to implement the modeling and aggregation method of the multi-feature subjective probabilistic prediction according to any one of the above.

[0042] The application provides a modeling and aggregation method and system of multi-feature subjective probabilistic prediction, which proposes a linear prediction model considering sampling error and judgment error based on a lens model in the field of psychology, analyzes and models multi-element expert evaluation, and designs an estimation algorithm capable of decomposing expert behavior features and problem variable features, so as to represent information sources of each predicted variable and weights given by experts to different information sources. The expert weights obtained based on the analyzed and modeled expert evaluation are clustered, and one expert is selected from each clustering group for prediction aggregation, so that a small number of experts represent the wisdom of the group, and thus the aggregation of expert probabilistic prediction is accurately performed without historical data calibration. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0044] Figure 1 is one of the flowcharts of the modeling and aggregation method of multi-feature subjective probabilistic prediction provided by the present application.

[0045] Figure 2 Figure 2 is a flowchart of a modeling and aggregation method of a multi-feature subjective probability type prediction provided by the present application;

[0046] Figure 3 Figure 3 is a flowchart of a modeling and aggregation method of a multi-feature subjective probability type prediction provided by the present application;

[0047] Figure 4 Figure 4 is a flowchart of a modeling and aggregation method of a multi-feature subjective probability type prediction provided by the present application;

[0048] Figure 5 Figure 5 is a flowchart of a modeling and aggregation method of a multi-feature subjective probability type prediction provided by the present application;

[0049] Figure 6 Figure 6 is a flowchart of a modeling and aggregation method of a multi-feature subjective probability type prediction provided by the present application;

[0050] Figure 7 Figure 7 is a schematic diagram of a module connection of a modeling and aggregation system of a multi-feature subjective probability type prediction provided by the present application;

[0051] Figure 8 Figure 8 is a structural schematic diagram of an electronic device provided by the present application.

[0052] Reference signs:

[0053] 110: linear prediction model establishment module; 120: prediction error module; 130: extraction potential information source module; 140: clustering grouping module; 150: prediction aggregation module;

[0054] 810: processor; 820: communication interface; 830: memory; 840: communication bus. DETAILED DESCRIPTION

[0055] In order to make the objectives, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below with reference to the drawings in the present application. Obviously, the described embodiments are some embodiments of the present application, but not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0056] The present application will be described below with reference to the drawings. Figures 1-6 A modeling and aggregation method of a multi-feature subjective probability type prediction provided by the present application is described, which comprises:

[0057] S100, acquire expert probability type judgment result, and establish a linear prediction model of the expert probability type judgment result based on a lens model;

[0058] S200, generate a prediction result through the linear prediction model, and determine an error of the prediction result;

[0059] S300, extract potential information sources according to the prediction result containing the prediction error, and decompose problem variable characteristics and expert behavior characteristics;

[0060] S400, determine weights given by experts to different information sources based on the decomposed expert behavior characteristics, cluster the expert weights, and form a plurality of clustering groups;

[0061] S500, select an expert with an estimated weight closest to a clustering center in each clustering group as a representative of each clustering group, and aggregate the probability type judgments through a plurality of heuristic methods.

[0062] In the application, the analysis of the expert prediction correlation focuses on the covariance matrix of the prediction error, and the information sources and information processing methods used by each expert are not considered when the prediction aggregation is performed. Under the premise that the prediction result is calibrated without historical prediction data, a linear prediction model is proposed to capture the information sharing structure and possible prediction error in the probability type judgment. It is assumed that each expert uses a "lens model" to linearly combine a plurality of information clues, and a Gaussian process model is used to explain the correlation between the sampling and judgment errors between the quantile judgments at different probability levels. In the research on the mining of the information sharing structure, some (or all) actual information clues used by each expert are not observed by the decision maker. The application uses the matrix decomposition method to extract "problem variable characteristics" (the potential information clues behind the probability judgment of each predicted problem variable) and "expert behavior characteristics" (how much weight each expert gives to these information clues) from the given probability type judgment, and theoretically proves that clustering the estimated weights of the information clues and the actual weights of the information clues has the same result, that is, the weights of the potential information clues can be used to define the correlation between the expert predictions. Based on this, the experts are clustered according to the weight similarity of the estimated expert prediction related information, and the sub-population is obtained by selecting representatives from each clustering group, so as to realize the prediction aggregation of the analytical modeling based on the multi-element expert evaluation.

[0063] Acquire expert probability type judgment result, and establish a linear prediction model of the expert probability type judgment based on a lens model, which comprises:

[0064] S101, the expert probability type judgment comprises a probability value of a prediction error and a probability value not containing the prediction error;

[0065] S102, based on the lens model and the prediction error model, analyzing the probability value of the prediction error and the probability value not containing the prediction error, and generating an analysis result;

[0066] S103, establishing a linear prediction model according to the weight distribution of the probability judgment without the prediction error in the analysis result and the prediction error.

[0067] In the present application, a linear prediction model of expert probability judgment is established based on the "lens model" in the field of psychology. The probability judgment of the expert is composed of the sum of the probability value not containing the prediction error and the prediction error. The probability value not containing the prediction error is represented as a linear combination of multiple information cues (containing observable information cues and unobservable potential information cues), and each information cue is subject to a certain probability distribution. The weight given to each information cue is greater than or equal to 0, and the sum is 1. In addition, all experts can see the same set of information cues for different variables, and the same expert gives the same set of weights to the information cues for different variables.

[0068] The prediction result is generated by the linear prediction model, and the error of the prediction result is determined, including:

[0069] S201, determining a sampling error and a judgment error based on the prediction result generated by the linear prediction model;

[0070] S202, the sampling error depends on the random sampling of information cues and how the expert linearly combines the sampled information;

[0071] S203, the judgment error depends on the cognitive error of the expert on the probability level;

[0072] S204, determining the error of the probability prediction result by combining the sampling error and the judgment error.

[0073] In the present application, the prediction error of the probability judgment given by the expert can be decomposed into a sampling error and a judgment error. The sampling error depends on two factors, one is the random sampling of information cues, which is subject to a Gaussian process; the other determining factor is how the expert linearly combines the sampled information, i.e. the weight coefficient of the expert. The judgment error mainly comes from the cognitive error of the expert on the probability level, which is also a Gaussian process, and the covariance function of the judgment error has:

[0074] Heteroscedasticity, although the expected value of the judgment error is 0, but its variance is different at different probability levels, which is a non-stationary Gaussian process;

[0075] Symmetry, the variance of the judgment error has symmetry at the probability level of 0.5;

[0076] Independence of variables, the error in judgment is independent between different problem variables.

[0077] Thus, the judgment error can be modeled as the slope related to the probability value not containing the prediction error.

[0078] After confirming the prediction error of the probability type judgment given by the expert, it is helpful to extract the potential information source on the basis of considering the error and improve the accuracy of the prediction aggregation.

[0079] According to the prediction result containing the prediction error, the potential information source is extracted, and the problem variable characteristics and the expert behavior characteristics are decomposed, including:

[0080] S301, establishing a probability type prediction matrix based on the prediction result containing the prediction error;

[0081] S302, decomposing the probability type prediction matrix into a problem variable characteristic matrix and an expert behavior characteristic matrix;

[0082] S303, iteratively solving the problem variable characteristic matrix and the expert behavior characteristic matrix by the maximum likelihood method, to generate the problem variable characteristics and the expert behavior characteristics.

[0083] In the present application, the probability type prediction matrix Y KI is decomposed into a problem variable characteristic matrix φ QI and an expert behavior characteristic matrix β KQ , that is, represented as: Where Y KI is a K*I function matrix, each column being a probability type prediction function of K experts on problem i; β is a K*Q matrix, each row being the weight of Q information clues of expert k; φ is a Q*I function matrix, each column being the quantile function of Q information clues on problem i.

[0084] The unknown parameter θ={β,φ} can be estimated by using the maximum likelihood method. For the above maximum likelihood estimation, the following multi-step iterative algorithm is used to solve:

[0085]

[0086]

[0087] Based on the decomposed expert behavior characteristics, the weights given by the experts to different information sources are determined, the expert weights are clustered, and a plurality of clustering groups are formed, including:

[0088] S401, solving the problem variable characteristics and the expert behavior characteristics based on the probability type prediction matrix, and estimating the weights given by the experts to the observable information clues and the unobservable information clues;

[0089] S402、based on the weight value given by the expert, the similar weight value is divided into the same group, clustering is completed, and multiple clustering groups are formed.

[0090] In the present application, based on the multi-element expert evaluation analysis modeling, the weight given by the expert to different information sources (including observable information clues and unobservable potential information clues) is estimated.

[0091] The expert weight is clustered, and in the present application, the clustering method includes k-means clustering, Gaussian mixture clustering, spectral clustering, kernel clustering, density-based clustering, hierarchical clustering and the like.

[0092] From each clustering group, the expert whose estimated weight is closest to the clustering center is selected as the representative of each clustering group, and the aggregation of probabilistic judgment is carried out by various heuristic methods, including:

[0093] S501, determine the clustering center, compare the weight value given by the expert of each clustering group with the weight value of the clustering center, and generate a comparison result;

[0094] S502, according to the comparison result, the expert whose estimated weight is closest to the clustering center is selected as the representative of each clustering group;

[0095] S503, the aggregation of probabilistic judgment is carried out by various heuristic methods based on the representative of each clustering group.

[0096] In the present application, from each clustering group, the expert whose estimated weight is closest to the clustering center is selected as the representative of each group, and the aggregation of probabilistic judgment is carried out by three heuristic methods, which are:

[0097] Ignoring the simple average of the size of each group and the intra-group variance;

[0098] Only consider the size of the group, and take the size of the group as the weight;

[0099] The size of the group and the intra-group variance are considered at the same time, and the weighted average based on the Tweedie formula contraction is used, the greater the intra-group variance, the greater the weight adjustment.

[0100] The modeling and aggregation method of multi-feature subjective probabilistic prediction also includes:

[0101] Data visualization is carried out, the correlation between each information clue and the problem variable, the weight given by each expert on different information clues and the weighted weight of selecting different aggregation methods are displayed in the form of charts.

[0102] In the present application, based on modeling and aggregation of multi-feature subjective probability type prediction, the information source and the expert weight are estimated, the estimated information clues for different problem variables are represented in the form of broken line chart in a graph, the correlation of each information clue and the problem variable can be intuitively presented, the estimation of the weight of the information clue given by different experts can be represented by different colors of the bar chart, the length of the bar chart represents the weight, the weight given by each expert on different information clues can be clearly seen, and it can be compared to reveal which experts give more weight on some information clues with higher correlation.

[0103] By the modeling and aggregation method of multi-feature subjective probability type prediction provided by the present application, a linear prediction model considering sampling error and judgment error is proposed based on the lens model in the field of psychology, the multi-expert evaluation is analytically modeled, and an estimation algorithm capable of decomposing the characteristics of expert behavior and problem variables is designed, which respectively represents the information source of each predicted variable and the weight of different information sources given by experts. The expert weights obtained based on the analytical modeling of expert evaluation are clustered, and one expert is selected from each cluster group for prediction aggregation, so as to accurately perform the aggregation of expert probability type prediction under the premise of no historical data calibration.

[0104] Reference Figure 7 The present application also discloses a modeling and aggregation system of multi-feature subjective probability type prediction, which comprises:

[0105] The linear prediction model establishing module 110 is used for acquiring the expert probability type judgment result, and establishing a linear prediction model of the expert probability type judgment result based on the lens model;

[0106] The prediction error module 120 is used for generating a prediction result through the linear prediction model, and determining the error of the prediction result;

[0107] The potential information source extraction module 130 is used for extracting potential information sources according to the prediction result containing the prediction error, and decomposing the problem variable characteristics and the expert behavior characteristics;

[0108] The clustering grouping module 140 is used for determining the weight of different information sources given by experts based on the decomposed expert behavior characteristics, clustering the expert weights, and forming a plurality of cluster groups;

[0109] The prediction aggregation module 150 is configured to select an expert with an estimated weight closest to the cluster center as a representative of each cluster group from each cluster group, and to perform a probabilistic judgment aggregation through a plurality of heuristic methods.

[0110] The linear prediction model establishment module is configured to analyze the probability value of the prediction error and the probability value of not containing the prediction error based on the lens model and the prediction error model, and to generate an analysis result.

[0111] The linear prediction model establishment module is configured to analyze the probability value of the prediction error and the probability value of not containing the prediction error based on the lens model and the prediction error model, and to generate an analysis result.

[0112] The linear prediction model establishment module is configured to analyze the probability value of the prediction error and the probability value of not containing the prediction error based on the lens model and the prediction error model, and to generate an analysis result.

[0113] The prediction error module is configured to determine a sampling error and a judgment error based on a prediction result generated by the linear prediction model.

[0114] The sampling error depends on the random sampling of the information cues and how the experts linearly combine the sampled information.

[0115] The judgment error depends on the cognitive error of the experts on the probability level.

[0116] The prediction error module is configured to determine an error of the probabilistic prediction result by combining the sampling error and the judgment error.

[0117] The latent information source extraction module is configured to establish a probabilistic prediction matrix based on the prediction result containing the prediction error.

[0118] The latent information source extraction module is configured to decompose the probabilistic prediction matrix into a problem variable feature matrix and an expert behavior feature matrix.

[0119] The latent information source extraction module is configured to iteratively solve the problem variable feature matrix and the expert behavior feature matrix through a maximum likelihood method, and to generate problem variable features and expert behavior features.

[0120] The clustering grouping module is configured to solve the problem variable features and the expert behavior features based on the probabilistic matrix, and to estimate the weights given by the experts to the observable information cues and the unobservable information cues.

[0121] The clustering grouping module is configured to divide similar weight values into the same group based on the weight values given by the experts, to complete clustering, and to form a plurality of cluster groups.

[0122] The prediction aggregation module is configured to determine a cluster center, to compare the weight values given by the experts of each cluster group with the weight values of the cluster center, and to generate a comparison result.

[0123] The prediction aggregation module is configured to select an expert with an estimated weight closest to the cluster center as a representative of each cluster group according to the comparison result.

[0124] Aggregation of the probabilistic judgments by various heuristic methods based on the representatives of each cluster group.

[0125] The modeling and aggregation system for multi-feature subjective probabilistic prediction provided by the application, by the lens model in the field of psychology, proposes a linear prediction model considering sampling error and judgment error, analyzes and models the multi-element expert evaluation, and designs an estimation algorithm that can decompose the expert behavior characteristics and problem variable characteristics, respectively representing the information source of each predicted variable and the weight of different information sources given by the expert. Based on the expert weight obtained by the analytical modeling of the expert evaluation, a cluster is selected from each cluster group to predict and aggregate, a small number of expert representatives represent the wisdom of the group, so as to accurately aggregate the expert probabilistic prediction without historical data calibration.

[0126] Figure 8 An example of an entity structure diagram of an electronic device is shown as Figure 8 The electronic device can include a processor 810, a communications interface 820, a memory 830, and a communications bus 840, wherein the processor 810, the communications interface 820, and the memory 830 communicate with each other through the communications bus 840. The processor 810 can call the logical instructions in the memory 830 to execute a modeling and aggregation method for multi-feature subjective probabilistic prediction, which includes: obtaining expert probabilistic judgment results, establishing a linear prediction model of the expert probabilistic judgment results based on a lens model;

[0127] Generating a prediction result through the linear prediction model, determining the error of the prediction result;

[0128] Extracting potential information sources from the prediction result containing the prediction error, and decomposing the problem variable characteristics and the expert behavior characteristics;

[0129] Determining the weight given by the expert to different information sources based on the decomposed expert behavior characteristics, clustering the expert weights to form multiple cluster groups;

[0130] From each cluster group, select the expert whose estimated weight is closest to the cluster center as the representative of each cluster group, and aggregate the probabilistic judgments by various heuristic methods.

[0131] Further, the logic instructions in the memory 830 described above can be implemented in the form of software functional units and sold or used as standalone products, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application or parts of the present application that essentially contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0132] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to execute a multi-feature subjective probability type prediction modeling and aggregation method provided by the above-mentioned methods, the method comprising: obtaining expert probability type judgment results, establishing a linear prediction model of the expert probability type judgment results based on a lens model;

[0133] generating a prediction result through the linear prediction model, and determining an error of the prediction result;

[0134] extracting potential information sources from the prediction result containing the prediction error, and decomposing problem variable features and expert behavior features;

[0135] determining weights given by experts to different information sources based on the decomposed expert behavior features, clustering the expert weights to form a plurality of clustering groups;

[0136] from each clustering group, selecting an expert whose estimated weight is closest to the clustering center as a representative of each clustering group, and performing aggregation of probability type judgments through a plurality of heuristic methods.

[0137] In another aspect, the present application also provides a non-transitory computer readable storage medium, which stores a computer program, the computer program is executed by a processor to implement a multi-feature subjective probability type prediction modeling and aggregation method provided by the above-mentioned methods, the method comprising: obtaining expert probability type judgment results, establishing a linear prediction model of the expert probability type judgment results based on a lens model;

[0138] generating a prediction result through the linear prediction model, and determining an error of the prediction result;

[0139] According to the prediction result containing the prediction error, a potential information source is extracted, and problem variable characteristics and expert behavior characteristics are decomposed;

[0140] Based on the decomposed expert behavior characteristics, weights given by experts to different information sources are determined, the expert weights are clustered, and multiple cluster groups are formed;

[0141] From each cluster group, an expert with the estimated weight closest to the cluster center is selected as the representative of each cluster group, and a probability type judgment is aggregated through multiple heuristic methods.

[0142] The device embodiments described above are only illustrative, wherein the units illustrated as separate components can or can not be physically separated, and the components illustrated as units can or can not be physical units, i.e., they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0143] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course, they can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0144] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A modeling and aggregation method for multi-feature subjective probabilistic prediction, characterized in that, The method comprises the following steps: obtaining expert probabilistic judgment results, and establishing a linear prediction model of the expert probabilistic judgment results based on a lens model; generating a prediction result through the linear prediction model, and determining an error of the prediction result; extracting potential information sources from the prediction result containing the prediction error, and decomposing problem variable characteristics and expert behavior characteristics; determining weights given by experts to different information sources based on the decomposed expert behavior characteristics, clustering the expert weights, and forming a plurality of clustering groups; from each clustering group, selecting an expert whose estimated weight is closest to the clustering center as a representative of each clustering group, and performing aggregation of probabilistic judgments through a plurality of heuristic methods.

2. The method of modeling and aggregating of multi-feature subjective probabilistic forecasts according to claim 1, characterized in that, The method of obtaining expert probabilistic judgment results and establishing a linear prediction model of the expert probabilistic judgment results based on a lens model comprises the following steps: The expert probabilistic judgment comprises a probability value of the prediction error and a probability value not containing the prediction error; analyzing the probability value of the prediction error and the probability value not containing the prediction error based on the lens model and a prediction error model, and generating an analysis result; establishing a linear prediction model according to the weight distribution of the probabilistic judgment not containing the prediction error and the prediction error in the analysis result.

3. The method of modeling and aggregating multi-feature subjective probabilistic forecasts according to claim 1, wherein, The method of generating a prediction result through the linear prediction model and determining an error of the prediction result comprises the following steps: determining a sampling error and a judgment error based on the prediction result generated by the linear prediction model; wherein the sampling error depends on random sampling of information clues and how the experts linearly combine the sampled information; the judgment error depends on the cognitive error of the experts on the probability level; determining the error of the probability prediction result by combining the sampling error and the judgment error.

4. The method of modeling and aggregating multi-feature subjective probabilistic forecasts according to claim 1, wherein, The method of extracting potential information sources from the prediction result containing the prediction error and decomposing problem variable characteristics and expert behavior characteristics comprises the following steps: establishing a probabilistic prediction matrix based on the prediction result containing the prediction error; decomposing the probabilistic prediction matrix into a problem variable characteristic matrix and an expert behavior characteristic matrix; iteratively solving the problem variable characteristic matrix and the expert behavior characteristic matrix by a maximum likelihood method to generate the problem variable characteristics and the expert behavior characteristics.

5. The method of claim 1, wherein, The method of determining weights given by experts to different information sources based on the decomposed expert behavior characteristics, clustering the expert weights, and forming a plurality of clustering groups comprises the following steps: solving the problem variable characteristics and the expert behavior characteristics based on the probabilistic prediction matrix, and estimating the weights given by the experts to the observable information clues and the unobservable information clues; based on the weight values given by the experts, dividing similar weight values into the same group, completing clustering, and forming a plurality of clustering groups.

6. The method of modeling and aggregating multi-feature subjective probabilistic forecasts according to claim 1, wherein, The method of selecting an expert whose estimated weight is closest to the clustering center as a representative of each clustering group from each clustering group, and performing aggregation of probabilistic judgments through a plurality of heuristic methods comprises the following steps: determining a clustering center, comparing the weight values given by the experts of each clustering group with the weight values of the clustering center, and generating a comparison result; selecting an expert whose estimated weight is closest to the clustering center as a representative of each clustering group according to the comparison result; performing aggregation of probabilistic judgments based on the representative of each clustering group through a plurality of heuristic methods.

7. The method of modeling and aggregating of multi-feature subjective probabilistic forecasts according to claim 1, characterized in that, Also included are: Data visualization is carried out, and the correlation between each information clue and the problem variable, the weight given by each expert on different information clues, and the weighted weight of selecting different aggregation methods are displayed in the form of charts.

8. A system for modeling and aggregating multi-feature subjective probabilistic forecasts, the system comprising: The system comprises: A linear prediction model establishment module is configured to obtain expert probability type judgment results, and establish a linear prediction model of the expert probability type judgment results based on a lens model; A prediction error module is configured to generate a prediction result through the linear prediction model, and determine an error of the prediction result; An extraction potential information source module is configured to extract a potential information source according to the prediction result containing the prediction error, and decompose problem variable characteristics and expert behavior characteristics; A clustering grouping module is configured to determine the weight given by experts to different information sources based on the decomposed expert behavior characteristics, cluster the expert weights, and form a plurality of clustering groups; A prediction aggregation module is configured to select an expert whose estimated weight distance is closest to the clustering center as a representative of each clustering group from each clustering group, and aggregate the probability type judgments through a plurality of heuristic methods.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the modeling and aggregation method of the multi-feature subjective probability type prediction according to any one of claims 1 to 7 when executing the program.

10. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program implements the modeling and aggregation method of the multi-feature subjective probability type prediction according to any one of claims 1 to 7 when executed by the processor.

Images