Object evaluation method and device, equipment, medium and program product

Abstract

The invention provides an object evaluation method, and relates to the field of artificial intelligence. The method comprises the steps of obtaining Q data sequence features of a to-be-evaluated object, and calculating a complexity index of each data sequence feature in the Q data sequence features; the Q data sequence features are divided into K data sets according to the complexity index of each data sequence feature, each data set comprises at least one data sequence feature in the corresponding complexity index range, and K is an integer larger than or equal to 2; calling pre-trained K machine learning models from a computer storage space to process data in the K data sets in a one-to-one correspondence manner to obtain K initial evaluation results, the contents of the K machine learning models being different from each other; and performing fusion processing based on the K initial evaluation results and the respective first weight coefficients to obtain a final evaluation result of the to-be-evaluated object.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence, and more specifically, to an object evaluation method, apparatus, device, medium, and program product. Background Art

[0002] Object evaluation refers to the process of collecting, analyzing and interpreting data and information about the object to be evaluated, and judging the object according to specific standards to obtain the desired objective evaluation results.

[0003] In the related art, object evaluation mainly relies on manual judgment based on features within a certain range. Manual judgment is greatly affected by subjective factors, the accuracy of judgment is not high, and once the judgment is made, it will not be updated in real time as the object itself changes. Therefore, a method of using machine learning models is provided to evaluate and judge, by uniformly inputting all the features of the object into a single machine learning model to output the predicted evaluation results.

[0004] In the process of realizing the inventive concept of the present disclosure, the inventors found that the features of the objects to be evaluated are diverse, each with different characteristics. A single machine learning model may not be able to fully process all the features, resulting in more computer resources being consumed for data processing and result prediction, reducing computer performance, and failing to obtain accurate prediction results. Summary of the invention

[0005] In view of the above problems, the present disclosure provides an object evaluation method, apparatus, device, medium and program product.

[0006] One aspect of an embodiment of the present disclosure provides an object evaluation method, including: obtaining Q data sequence features of an object to be evaluated, where Q is an integer greater than or equal to 2; calculating a complexity index of each data sequence feature in the Q data sequence features, where the complexity index of each data sequence feature is obtained based on the sample entropy of the feature; dividing the Q data sequence features into K data sets according to the complexity index of each data sequence feature, wherein each data set includes at least one data sequence feature within a corresponding complexity index range, where K is an integer greater than or equal to 2; calling K pre-trained machine learning models from a computer storage space to process the data in the K data sets one by one to obtain K initial evaluation results, where the contents of the K machine learning models are different; and performing fusion processing based on the K initial evaluation results and their respective first weight coefficients to obtain a final evaluation result of the object to be evaluated.

[0007] According to an embodiment of the present disclosure, before performing fusion processing based on the K initial evaluation results and their respective first weight coefficients, it also includes: training the initial K machine learning models to obtain the K machine learning models, wherein during the training process, the K first weight coefficients are updated according to the prediction result errors of the one-to-one corresponding K machine learning models, and the prediction result error characterizes the degree of difference between the prediction results of each machine learning model for the training samples and the sample labels.

[0008] According to an embodiment of the present disclosure, updating K of the first weight coefficients during the training process includes: after each round of training, determining the prediction result errors of each of the K machine learning models; obtaining the sum of the K prediction result errors, and the ratio of the prediction result error of each machine learning model to the sum of the errors; and updating the ratio of each machine learning model to the first weight coefficient of the model.

[0009] According to an embodiment of the present disclosure, calculating the complexity index of each data sequence feature among Q data sequence features includes: performing phase space reconstruction on each data sequence feature to obtain a corresponding reconstructed data sequence feature; determining a first sample entropy value of each data sequence feature, and a second sample entropy value of the reconstructed data sequence feature corresponding to the data sequence feature; obtaining the complexity index of each data sequence feature according to the difference between the first sample entropy value and the second sample entropy value; for each data sequence feature, when its first sample entropy value is less than the second sample entropy value of the corresponding reconstructed data sequence feature, it has a first complexity index and is divided into the second data set; otherwise, it has a second complexity index and is divided into the third data set.

[0010] According to an embodiment of the present disclosure, calculating the complexity index of each data sequence feature among Q data sequence features includes: calculating the third sample entropy value of each data sequence feature; obtaining multiple historical evaluation data of the object to be evaluated; obtaining the sample entropy value of each data sequence feature in the multiple historical evaluation data, and determining the historical sample entropy range; if the third sample entropy value has a third complexity index within the historical sample entropy range, it is divided into the second data set, otherwise it has a fourth complexity index and is divided into the third data set.

[0011] According to an embodiment of the present disclosure, the K machine learning models include an integrated autoregressive integrated moving average model and a long short-term memory neural network model, and calling the pre-trained K machine learning models from the computer storage space to process the data in the K data sets one by one includes: calling the pre-trained integrated autoregressive integrated moving average model to process the data in the second data set, and calling the pre-trained long short-term memory neural network model to process the data in the third data set.

[0012] According to an embodiment of the present disclosure, pre-training the integrated autoregressive integrated moving average model includes: determining L autoregressive integrated moving average models as L base learners, where L is an integer greater than or equal to 2; binary encoding the L base learners and assigning L second weight coefficients respectively; training the L base learners based on a genetic learning algorithm to update the L second weight coefficients to obtain the third weight coefficients of each of the L base learners, and the integrated autoregressive integrated moving average model includes the L base learners and the L third weight coefficients.

[0013] According to an embodiment of the present disclosure, obtaining Q data sequence features of the object to be evaluated includes: obtaining a first data set of the object to be evaluated, the first data set including original data obtained from the object to be evaluated; and processing the first data set based on a set empirical mode decomposition method to obtain the Q data sequence features.

[0014] According to an embodiment of the present disclosure, the object to be evaluated includes an enterprise to be evaluated, the Q data sequence features are obtained from the enterprise's basic data, enterprise operating data, enterprise financial data and the enterprise's transaction data in financial institutions, and the initial evaluation result includes the enterprise size category of the enterprise to be evaluated.

[0015] Another aspect of an embodiment of the present disclosure provides an object evaluation device, including: a feature acquisition module, used to acquire Q data sequence features of an object to be evaluated, where Q is an integer greater than or equal to 2; an index calculation module, used to calculate the complexity index of each data sequence feature in the Q data sequence features, wherein the complexity index of each data sequence feature is obtained based on the sample entropy of the feature; a feature partitioning module, used to partition the Q data sequence features into K data sets according to the complexity index of each data sequence feature, wherein each data set includes at least one data sequence feature within a corresponding complexity index range, and K is an integer greater than or equal to 2; an initial evaluation module, used to call K pre-trained machine learning models from a computer storage space to process the data in the K data sets one by one to obtain K initial evaluation results, and the contents of the K machine learning models are different; a weighted processing module, used to perform fusion processing based on the K initial evaluation results and their respective first weight coefficients to obtain a final evaluation result of the object to be evaluated.

[0016] Another aspect of an embodiment of the present disclosure provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the method as described above.

[0017] Another aspect of the embodiments of the present disclosure further provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to execute the method as described above.

[0018] Another aspect of the embodiments of the present disclosure further provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0019] One or more of the above embodiments have the following beneficial effects: by dividing the data sequence features into multiple data sets and using multiple machine learning models for parallel processing, the computing efficiency of the computer can be improved and the prediction speed of the evaluation results can be accelerated. By using multiple machine learning models to process different data sets and fusing the results, the advantages of each model in processing different complexity indicators can be comprehensively utilized to improve the accuracy and credibility of the evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The above contents and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0021] Figure 1 The application scenario diagram of the object evaluation method according to the embodiment of the present disclosure is schematically shown;

[0022] Figure 2 A flowchart of an object evaluation method according to an embodiment of the present disclosure is schematically shown;

[0023] Figure 3 Schematically shows a flow chart of extracting features according to an embodiment of the present disclosure;

[0024] Figure 4 A flowchart of calculating a complexity index according to an embodiment of the present disclosure is schematically shown;

[0025] Figure 5 A flowchart of training the integrated autoregressive integrated moving average model according to an embodiment of the present disclosure is schematically shown;

[0026] Figure 6 A flowchart of training the integrated autoregressive integrated moving average model according to another embodiment of the present disclosure is schematically shown;

[0027] Figure 7 Schematically shows a flow chart of updating K first weight coefficients according to an embodiment of the present disclosure;

[0028] Figure 8 A schematic diagram of a structure of an object evaluation device according to an embodiment of the present disclosure is shown; and

[0029] Fig. 9A block diagram of an electronic device suitable for implementing an object evaluation method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0030] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0031] In the technical solution of the present invention, the user information (including but not limited to user personal information, user image information, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data comply with the relevant laws, regulations and standards of relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0032] Taking enterprise assessment as an example, existing commercial banks mainly rely on manual judgment to identify enterprise scale, combining the characteristics of customers within a certain range, and the judgment is not updated dynamically as the enterprise continues to develop. Manual judgment of customer enterprise scale is greatly affected by the subjective factors of bank account managers, the accuracy of judgment is not high, and after the judgment is made, it will not be updated in real time as the customer enterprise develops, which is not conducive to the bank's classification management of customers and affects the bank's classification marketing of customers based on the size of the customer enterprise.

[0033] In some embodiments of the present disclosure, an object evaluation method is provided, which can improve the computing efficiency of the computer and accelerate the prediction speed of the evaluation results by dividing the data sequence features into multiple data sets according to complexity and using multiple machine learning models for parallel processing. By using multiple machine learning models to process different data sets and fusing the results, the advantages of each model in processing different complexity indicators can be comprehensively utilized to improve the accuracy and credibility of the evaluation.

[0034] Figure 1 The following schematically shows an application scenario diagram of the object evaluation method according to an embodiment of the present disclosure. It should be noted that: Figure 1What is shown are merely examples to which the embodiments of the present disclosure can be applied, so as to help those skilled in the art understand the technical content of the present disclosure, but it does not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios.

[0035] like Figure 1 As shown, the application scenario 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for a communication link between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0036] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only examples).

[0037] The terminal devices 101 , 102 , and 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.

[0038] The server 105 may be a server that provides various services, such as a background management server (only an example) that provides support for websites browsed by users using the terminal devices 101, 102, and 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0039] Server 105 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud computing, network services, and middleware services.

[0040] In applications such as client applications and web applications (abbreviated as APP in English), the client (i.e., front end) and the server (i.e., back end) can communicate data through network messages, for example, obtain training samples from the front end, receive the user's training instructions on the front end interface, and send the training samples to the server in response to the training instructions to train the initial model stored in the computer storage space. After the training is completed, the machine learning model is deployed to a specific computer storage space, such as a local terminal device or a cloud database on the server side, in response to the deployment instructions sent by the user through the front end. And in response to the evaluation instructions sent by the user through the front end, the data sequence features of the object are obtained and the machine learning model is called for processing.

[0041] It should be noted that the object evaluation method provided in the embodiment of the present invention can be executed by the terminal device 101, 102, 103 or the server 105. Accordingly, the object evaluation device provided in the embodiment of the present invention can generally be set in the terminal device 101, 102, 103 or the server 105.

[0042] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0043] The following will be based on Figure 1 The scenario described is based on the example of the enterprise to be evaluated. Figure 2 to Figure 7 The object evaluation method of the embodiment of the present disclosure is described in detail. It is understandable that the object evaluation method provided by the present disclosure is not limited to enterprise scale evaluation, but can also be used in scenarios such as enterprise or personal credit evaluation, product quality evaluation, mechanical structure life evaluation or enterprise risk evaluation.

[0044] Figure 2 The flowchart of the object evaluation method according to the embodiment of the present disclosure is schematically shown.

[0045] like Figure 2 As shown, this embodiment includes:

[0046] In operation S210, Q data sequence features of the object to be evaluated are obtained, where Q is an integer greater than or equal to 2.

[0047] The objects to be evaluated include enterprises that need to be evaluated. Q data sequence features are obtained from the basic data of the enterprise, the business data of the enterprise, the financial data of the enterprise, and the transaction data of the enterprise in financial institutions. The initial evaluation results include the enterprise scale category of the enterprise to be evaluated. Data sequence features refer to the characteristic data sequence of each dimension of data in the enterprise to be evaluated. For example, the business data of the enterprise can include the annual business data. The financial data of the enterprise can include the annual financial data. Other dimensional data can also be collected according to time and processed in the form of vector sequences.

[0048] Each dimension of data may have multiple data series features. For example, corporate financial data may include data series features corresponding to assets, data series features corresponding to liabilities, data series features corresponding to working capital, and so on.

[0049] In operation S220, a complexity index of each data sequence feature among the Q data sequence features is calculated, and the complexity index of each data sequence feature is obtained based on the sample entropy of the feature.

[0050] The complexity index characterizes the complexity of measuring the characteristics of a data sequence. It can directly use the sample entropy as the index value, or it can be obtained by performing corresponding calculations on the sample entropy. For example, the difference and degree of change between two sample entropies of the same data sequence feature at different time points can be determined. Another example is to reconstruct the same data sequence feature to obtain features of different sequence lengths, thereby calculating the difference and degree of change between the sample entropies. For example, the sample entropy of Q data sequence features can be sorted, and the complexity index of each data sequence feature can be assigned according to the sorting result.

[0051] In operation S230, the Q data sequence features are divided into K data sets according to the complexity index of each data sequence feature, wherein each data set includes at least one data sequence feature within the corresponding complexity index range, and K is an integer greater than or equal to 2.

[0052] Each data set is a subset of data sequence features divided according to complexity indicators, including at least one data sequence feature within different complexity ranges.

[0053] In operation S240, K pre-trained machine learning models are called from the computer storage space to process the data in the K data sets one by one to obtain K initial evaluation results, and the contents of the K machine learning models are different.

[0054] Exemplarily, the K machine learning models include traditional machine learning models, such as support vector machines, logistic regression, random forest, autoregressive integrated moving average model (ARIMA), etc., and may also be deep learning models, such as long short-term memory neural network (LSTM), BERT, etc.

[0055] Computer storage space may include local cache, memory, database, or remotely connected cloud database, cloud storage, etc. Different computing resources may be allocated to each model at runtime, and the allocation method may be determined based on the number of model parameters and the size of input data.

[0056] The contents of the K machine learning models include at least one of the model algorithm category, model architecture, model parameter value, or model parameter quantity. For example, among any two machine learning models, one is an LSTM model and the other is a support vector machine model, which belong to different model algorithm categories, and the corresponding model architectures, model parameter values, or model parameter quantities are also different. For example, among any two machine learning models, both are LSTM models, but at least one of their model architectures, model parameter values, or model parameter quantities is different.

[0057] K machine learning models process the data in K data sets one by one, that is, characterize each machine learning model to process data sequence features with different complexity indicators. The data sequence features with which complexity indicators each machine learning model is used to process can be determined according to the characteristics of its model algorithm. For example, the logistic regression model is suitable for processing low complexity, while the LSTM model is suitable for processing high complexity. It can also be determined in advance based on the actual processing time of each model. For example, each model processes the data sequence features within each complexity range, and for each complexity range, the model with the shortest processing time and / or the most accurate prediction result is determined for processing.

[0058] In operation S250, a fusion process is performed based on the K initial evaluation results and their respective first weight coefficients to obtain a final evaluation result of the object to be evaluated.

[0059] The initial evaluation results refer to the prediction results obtained after each machine learning model processes the corresponding data set.

[0060] According to the embodiments of the present disclosure, by dividing the data sequence features into multiple data sets and using multiple machine learning models for parallel processing, the computing efficiency of the computer can be improved and the prediction speed of the evaluation results can be accelerated. By using multiple machine learning models to process different data sets and fusing the results, the advantages of each model in processing different complexity indicators can be comprehensively utilized to improve the accuracy and credibility of the evaluation.

[0061] Figure 3 The flowchart of extracting features according to an embodiment of the present disclosure is schematically shown.

[0062] like Figure 3 As shown, this embodiment is one of the embodiments of operation S210, including:

[0063] In operation S310 , a first data set of an object to be evaluated is obtained, where the first data set includes original data obtained from the object to be evaluated, for example, the original data is represented in the form of a serialized vector.

[0064] For example, it can be the original data of the enterprise to be evaluated extracted from the database. The original data refers to the initial data of the enterprise to be evaluated. For example, it can be the transfer amount directly obtained from a financial institution.

[0065] In operation S320, the first data set is processed based on the set empirical mode decomposition method to obtain Q data sequence features.

[0066] The ensemble empirical mode decomposition (EEMD) method mainly decomposes a large number of irregular, nonlinear, and non-stationary data sequences into multiple independent intrinsic mode functions, overcoming the problems of data confusion and data aliasing, and can decompose the characteristics of the data sequence into multiple independent intrinsic mode functions (IMFs). For example, feature extraction is achieved based on the following formulas 1 and 2:

[0067] x n1 (t) = x(t) + ε n1 (t) Formula 1

[0068]

[0069] Among them, x(t) represents the original sequence of the first data set at time t, ε n1 (t) represents noise information, c j1，i1 (t) represents the IMF component with j1 noise information added, i1 represents the i1th order IMF component, n1 and m1 represent the number of IMF components and the number of times noise information is added, respectively, and i1, j1, n1, and m1 are all greater than or equal to 1.

[0070] Referring to Formula 1 and Formula 2, the original data of each dimension of the enterprise to be evaluated is collected to obtain the first data set. Referring to Formula 1, for each original sequence x(t), noise information ε is added n1 (t), if the original sequence corresponding to the financial data is added with the newly added financial data, a new sequence x is obtained. n1 (t). For each x n1 (t) Perform EMD decomposition to obtain the respective IMF components c j，i (t) and residual value, repeat m1 times. Finally, perform ensemble average calculation according to Formula 2 to obtain the final IMF of EEMD decomposition, i.e., c i (t). The decomposed c i (t) Applied to the original data sequence of the sample entropy sequence data X = [x(1), x(2), ..., x(N)] below.

[0071] According to an embodiment of the present disclosure, by processing the first data set through the ensemble empirical mode decomposition method, multiple independent intrinsic mode functions can be extracted, thereby more accurately representing the characteristics of the object to be evaluated and improving the accuracy and credibility of subsequent model prediction evaluation.

[0072] Figure 4 The flowchart of calculating the complexity index according to an embodiment of the present disclosure is schematically shown.

[0073] like Figure 4 As shown, this embodiment is one of the embodiments of operation S220, including:

[0074] In operation S410, each data sequence feature is reconstructed in phase space to obtain a corresponding reconstructed data sequence feature.

[0075] Phase space reconstruction is to process the original data sequence features and convert them into a set of points in a multidimensional space. For example, through the delayed embedding method, one-dimensional time series data is converted into a two-dimensional or higher-dimensional data point set. The dimensions before and after reconstruction are different.

[0076] In operation S420, a first sample entropy value of each data sequence feature and a second sample entropy value of a reconstructed data sequence feature corresponding to the data sequence feature are determined.

[0077] In operation S430, the complexity index of each data sequence feature is obtained according to the difference between the first sample entropy value and the second sample entropy value. By comparing the first sample entropy value and the second sample entropy value, the complexity of the data sequence feature is further measured, and an index is provided to describe its complexity.

[0078] There are three parameters for calculating the sample entropy value, which are the data sequence length N, the dimension m2 and the tolerance r. The data sequence features are divided by obtaining the first sample entropy value and the second sample entropy value.

[0079] Assuming that the data sequence feature X = [x(1), ..., x(2), ..., x(N)], the sequence X is reconstructed in phase space to obtain X = [x m2 (1), x m2 (2), …, x m2 (N-m2+1)], where x m2 (i2) = [x(i2), x(i2+1), ..., x(i2+m2-1)], i2 is greater than or equal to 1 and less than or equal to N-m2+1. The entropy value of each sample is calculated as follows:

[0080] d[x m2 (i2), x m2 (j2)]=max k∈[0，m-1] |x(i2+k)-x(j2+k)| Formula 3

[0081]

[0082]

[0083]

[0084] Among them, d[x m2 (i2), x m2 (j2)] represents x m2 (i2), x m2 The absolute value of the maximum difference between the corresponding elements (j2) represents the distance between the two, and j2 is greater than or equal to 1 and less than or equal to N-m2. represents d[x m2 (i2), x m2 (j2)]<r, E(N,m2,r) represents the value of sample entropy, B m2 (r) represents x m2 (i2), x m2 (j2) The probability of matching m points under similarity tolerance r.

[0085] In operation S430, the data sequence features where the second sample entropy value is smaller than the first sample entropy value are merged into a second data set S′=[S 1 , …, S i , …, S N ], and the rest are modeled separately for interval construction, and the third data set S″=[S 1 , …, S i , …, S N ].

[0086] Referring to the method of calculating the second sample entropy value and the first sample entropy value as above, each data sequence feature can be reconstructed in phase space to obtain two reconstructed data sequence features to calculate the second sample entropy value and the first sample entropy, and the values ​​of the two reconstructed data sequence features N or m2 are different.

[0087] In some embodiments, for each data sequence feature, when its first sample entropy value is less than the second sample entropy value of the corresponding reconstructed data sequence feature, it has a first complexity index and is classified into the second data set; otherwise, it has a second complexity index and is classified into the third data set. The second complexity index represents a greater complexity than the first complexity index.

[0088] In other embodiments, only the third sample entropy value of each data sequence feature may be determined to obtain a complexity index, for example, the third sample entropy value of each data sequence feature is calculated, multiple historical evaluation data of the object to be evaluated are obtained, the sample entropy value of each data sequence feature in the historical evaluation data is obtained, the historical sample entropy range is determined, and it is determined whether the first sample entropy value is within the historical sample entropy range. If so, it has the third complexity index and is divided into the second data set, otherwise it has the fourth complexity index and is divided into the third data set. The fourth complexity index represents a greater degree of complexity than the third complexity index.

[0089] In other embodiments, for the data sequence features divided into the third data set, a separate model is also established to perform interval construction so as to adapt it to the processing format of the corresponding machine learning model, such as the different processing formats of input data between the LSTM model and the ARIMA model.

[0090] For example, the K machine learning models include an integrated autoregressive integrated moving average model and a long short-term memory neural network model, and the K pre-trained machine learning models are called from the computer storage space to process the data in the K data sets one by one, including:

[0091] A pre-trained integrated autoregressive integrated moving average model is called to process the data in the second data set, and a pre-trained long short-term memory neural network model is called to process the data in the third data set.

[0092] The process of training K machine learning models is further described below.

[0093] Based on the company's basic information, number of employees, social security, transaction data, working capital, and financial news, combined with machine learning models and data sequence characteristics, the following training steps can be used to obtain K machine learning models to evaluate the size of the company.

[0094] Step 1: Data Collection and Preparation

[0095] Collect data on the company's basic information, number of employees, social security, transaction data, working capital and financial news, and perform data cleaning and pre-processing to ensure the accuracy and completeness of the data.

[0096] Step 2: Feature Engineering and Data Sequence Feature Extraction

[0097] According to different data types and goals, feature engineering and data sequence feature extraction are performed. For example, basic information can extract features such as the number of employees and registered capital. Social security data can extract features such as payment status and number of insured persons. Transaction data can extract features such as transaction amount and transaction frequency. Liquidity can extract features such as cash flow and balance sheet. Financial news can perform sentiment analysis and keyword extraction. Extracting data sequence features can be performed according to the above operations on EEMD. Finally, the training data set is obtained.

[0098] Step 3: Model training and evaluation

[0099] For example, several machine learning models, such as decision trees, random forests, or neural networks, are used to train and evaluate the extracted features. The model can use supervised learning or unsupervised learning methods to classify or cluster data based on labels or targets.

[0100] In some embodiments, the K machine learning models include an integrated autoregressive integrated moving average model to be trained and a long short-term memory neural network model to be trained.

[0101] Figure 5 The flowchart of training an integrated autoregressive integrated moving average model according to an embodiment of the present disclosure is schematically shown. Figure 6 The flowchart of training an integrated autoregressive integrated moving average model according to another embodiment of the present disclosure is schematically shown.

[0102] like Figure 5 As shown, the pre-trained integrated autoregressive integrated moving average model includes:

[0103] In operation S510, L autoregressive integrated moving average models are determined as L base learners, where L is an integer greater than or equal to 2.

[0104] In operation S520, L base learners are binary-coded and assigned L second weight coefficients respectively.

[0105] In operation S530, L base learners are trained based on a genetic learning algorithm to update L second weight coefficients to obtain third weight coefficients of the L base learners respectively, and the integrated autoregressive integrated moving average model includes L base learners and L third weight coefficients.

[0106] This embodiment integrates multiple ARIMA models as base learners, encodes all base learners in binary format, and then uses a genetic algorithm to select the optimal learner combination. Figure 6 Operation S510 to Operation S530 are further described.

[0107] The ARIMA model can predict the changing trend by explaining the relationship between data series. It includes three parameters, which represent the order p of the autoregressive model, the number of differences d, and the order q of the moving average model. The model calculation expression is as follows:

[0108]

[0109] Among them, X t represents the sequence characteristics at time t, X t-p , X t-j3 They represent the sequence values ​​of the first p and first j3 moments at time t, respectively. is the autoregressive coefficient. j3 is the moving average coefficient.

[0110] The establishment of the integrated ARIMA model is mainly divided into the following two steps:

[0111] Perform a stationarity test and a difference operation to determine the number of differences d.

[0112] Set the value range of order p and q: Assuming that the value range of order p is (0, p) and the value range of q is (0, q), a total of (p+1)*(q+1) ARIMA models are constructed for training.

[0113] Assume that the set of ARIMA models constructed is A = [a(1), ..., a(i), ..., a(N)], and use the binary weight coding method to assign an initial weight w to each model. i (0), w i (0) is randomly initialized to 0 or 1, and the optimal ARIMA model set is expressed as:

[0114]

[0115] where w i (t) represents the weight value after t genetic calculations. The objective function of the genetic algorithm is as follows:

[0116]

[0117] In the formula, m3 represents the number of samples that need to be verified. represents the predicted value of the i4th sample under the j4th model, y i4Indicates the true value corresponding to the i4th sample. Finally, the output result Z = [z(1), ..., z(i4), ..., z(N)] of the ARIMA model combination with the best generalization ability is obtained, where i4 and j4 are greater than or equal to 1.

[0118] According to the embodiments of the present disclosure, by integrating multiple autoregressive integrated moving average models, the advantages of different models can be utilized to improve the accuracy of prediction. By using a genetic learning algorithm to train and optimize the model, the overfitting problem of the model can be reduced and the stability of the prediction can be improved. By optimizing the weights of the base learner through a genetic learning algorithm, the contribution of each model can be automatically adjusted, so that the prediction result of the integrated model is more accurate.

[0119] In other embodiments, the LSTM model is trained using a training data set. For example, an error back propagation (BP) algorithm can be used to correct the size of the parameters in the initial model during the training process, so that the reconstruction error loss becomes smaller and smaller, aiming to obtain the optimal LSTM model parameters, such as the weight matrix.

[0120] It can be understood that the samples for training the integrated autoregressive integrated moving average model can be obtained according to the data sequence feature acquisition method in the above-mentioned second data set, and the samples for training the LSTM model can be obtained according to the data sequence feature acquisition method in the above-mentioned third data set.

[0121] Step 4: Scale Assessment and Validation

[0122] By training K machine learning models, we can process the validation set to evaluate the scale of the enterprise and determine whether the prediction results are accurate.

[0123] Therefore, we can comprehensively consider the company's basic information, number of employees, social security, transaction data, working capital, financial news and other data, and use machine learning models and data sequence characteristics to evaluate the size of the company.

[0124] In the training process, before operation S250 performs fusion processing based on the K initial evaluation results and their respective first weight coefficients, the training stage further includes:

[0125] The initial K machine learning models are trained to obtain K machine learning models, wherein during the training process, the K first weight coefficients are updated according to the prediction result errors of the one-to-one corresponding K machine learning models, and the prediction result error characterizes the difference between the prediction results of each machine learning model for the training samples and the sample labels.

[0126] During the training process, according to the prediction result errors of the K corresponding machine learning models, the K first weight coefficients are updated to adjust the contribution of each model to the final result, so that the weight of the model with smaller error is increased and the weight of the model with larger error is reduced. By evaluating the error and updating the weight coefficient, the accuracy and generalization ability of the model can be improved.

[0127] Figure 7 The flowchart of updating K first weight coefficients according to an embodiment of the present disclosure is schematically shown.

[0128] like Figure 7 As shown, this embodiment includes:

[0129] In operation S710, after each round of training, the prediction result errors of the K machine learning models are determined. For example, during the training process, the samples in the training data set are divided into multiple batches, and one batch of samples is input in each round of training.

[0130] In operation S720, the sum of the K prediction result errors and the ratio of the prediction result error of each machine learning model to the sum of the errors are obtained. The purpose of obtaining the sum of the K prediction result errors is to comprehensively evaluate the prediction performance of all models and calculate the ratio of the prediction error of each model to the sum of the total errors to reflect the relative contribution of each model in the prediction.

[0131] In operation S730, the ratio of each machine learning model is updated to the first weight coefficient of the model.

[0132] In order to improve the judgment accuracy of the combined model, the inverse error method is used to perform weighted judgment on the results of the integrated ARIMA model and LSTM model training. The prediction result error ε i The calculation of is as follows:

[0133]

[0134] where y' i5 Represents the predicted value, y i5 Represents the actual value. Fusion function g i5 The calculation is as follows:

[0135]

[0136]

[0137] g i5 =w 1 z i5 +W 2 h i5

[0138] Among them, ε1 , ε 2 They refer to the errors of the integrated ARIMA model and LSTM model, respectively, and w 1 、w 2 They represent the corresponding weight coefficients respectively. i5 is the initial evaluation result of the integrated ARIMA model, h i5 is the initial evaluation result of the LSTM model.

[0139] In operation S720, ε is obtained. 2 , ε 1 , in operation S250, the fusion function g is used i5 The fusion process is implemented to obtain the final prediction result of the object to be evaluated. The role of updating the ratio of each model to the first weight coefficient is to adjust the weights between the models to reflect their relative credibility in the prediction. This can more effectively integrate the prediction results of each model into the final decision.

[0140] In the object evaluation method of this embodiment, the EEMD processing flow is first used to decompose a variety of chaotic original data into multiple independent intrinsic modal functions as the original data set, and then the sample entropy method is used to extract the features of the original data set, the data set is classified, and the sequence features within a certain range selected by the sample entropy are applied to the ARIMA model, and the customer enterprise scale score values ​​under multiple influencing factors are trained using the ARIMA model, and then the household enterprise scale score values ​​of multiple influencing factors are integrated using the genetic algorithm. The remaining data selected by the sample entropy method is separately modeled for interval construction and the LSTM model is applied to the sequence data to train the customer enterprise scale score values ​​under multiple influencing factors. Finally, the ARIMA model and the LSTM model are combined and revised using the inverse error method to output the final score value of the enterprise scale within the error range, so as to realize a method for determining the enterprise scale based on all aspects of the customer's influencing factors, which can be updated in real time and has a higher accuracy rate.

[0141] According to the embodiments of the present disclosure, the sequence data of all features can be covered, and the ARIMA model can better process the data with low complexity of the sequence data. Within a certain sample entropy value range, the genetic algorithm is used to select some ARIMA models with high accuracy, which can obtain higher accuracy than the method of using a single model. The remaining features apply the LSTM model, which can better process the data with high complexity of the sequence data and calculate the score of the customer's enterprise scale. And during training, the optimal solution of the ARIMA model and the LSTM model is combined and trained using the inverse error method, and the final score value of the enterprise scale within the error range is finally output, so as to realize the enterprise scale determination result based on all aspects of the customer's influencing factors. On the one hand, the comprehensiveness of the determination result can be improved, and on the other hand, the accuracy of the determination result can be improved. The evaluation process is more flexible, not based on the prediction of a certain historical point in time, but based on the determination results dynamically obtained in each time series.

[0142] By using multiple machine learning models and fusing their prediction results, the advantages of each model can be combined, the bias and error of a single model can be reduced, and the accuracy of the overall prediction can be improved. Since each model has its own unique feature extraction and learning methods, their responses to input data of different complexity may be different. By fusing the prediction results of multiple models, the instability of a single model can be reduced, making the overall prediction results more stable and reliable.

[0143] Based on the above object evaluation method, the present disclosure also provides an object evaluation device. Figure 8 The device is described in detail.

[0144] Figure 8 The structure block diagram of the object evaluation device according to the embodiment of the present disclosure is schematically shown.

[0145] like Figure 8 As shown, the object evaluation device 800 of this embodiment includes a feature acquisition module 810, an index calculation module 820, a feature division module 830, an initial evaluation module 840, and a weighted processing module.

[0146] The feature acquisition module 810 may perform operation S210 to acquire Q data sequence features of the object to be evaluated, where Q is an integer greater than or equal to 2.

[0147] The index calculation module 820 may perform operation S220 to calculate the complexity index of each of the Q data sequence features, where the complexity index of each data sequence feature is obtained based on the sample entropy of the feature.

[0148] The feature partitioning module 830 may perform operation S230, which is used to partition the Q data sequence features into K data sets according to the complexity index of each data sequence feature, wherein each data set includes at least one data sequence feature within the corresponding complexity index range, and K is an integer greater than or equal to 2.

[0149] The initial evaluation module 840 can execute operation S240, which is used to call K pre-trained machine learning models from the computer storage space to process the data in K data sets one by one, and obtain K initial evaluation results. The contents of the K machine learning models are different.

[0150] The weighted processing module 850 may execute operation S250 to perform fusion processing based on the K initial evaluation results and their respective first weight coefficients to obtain a final evaluation result of the object to be evaluated.

[0151] In some embodiments, the feature acquisition module 810 may perform operations S310 to S320, which will not be described in detail herein.

[0152] In some embodiments, the feature calculation module 820 may perform operations S410 to S430, which will not be described in detail herein.

[0153] In some embodiments, the weighted processing module 850 may perform operations S710 to S730, which will not be described in detail herein.

[0154] In some embodiments, the object evaluation device 800 may further include a model training module, which may train K machine learning models, for example, and may be used to perform operations S510 to S530, which will not be described in detail herein.

[0155] It should be noted that the object evaluation device 800 includes two components for performing the above Figure 2 to Figure 7 The modules of each step of any embodiment described. The implementation methods, technical problems solved, functions realized, and technical effects achieved of each module / unit / subunit, etc. in the device part embodiment are respectively the same or similar to the implementation methods, technical problems solved, functions realized, and technical effects achieved of each corresponding step in the method part embodiment, and will not be repeated here.

[0156] According to an embodiment of the present disclosure, any multiple modules among the feature acquisition module 810, the index calculation module 820, the feature division module 830, the initial evaluation module 840, and the weighted processing module can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module.

[0157] According to an embodiment of the present disclosure, at least one of the feature acquisition module 810, the index calculation module 820, the feature division module 830, the initial evaluation module 840, and the weighted processing module can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or by any one of the three implementation methods of software, hardware, and firmware, or by an appropriate combination of any of them. Alternatively, at least one of the feature acquisition module 810, the index calculation module 820, the feature division module 830, the initial evaluation module 840, and the weighted processing module can be at least partially implemented as a computer program module, and when the computer program module is run, the corresponding function can be performed.

[0158] Fig. 9 A block diagram of an electronic device suitable for implementing an object evaluation method according to an embodiment of the present disclosure is schematically shown.

[0159] like Fig. 9 As shown, the electronic device 900 according to an embodiment of the present disclosure includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage part 908 into a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include an onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0160] In RAM 903, various programs and data required for the operation of electronic device 900 are stored. Processor 901, ROM 902 and RAM 903 are connected to each other via bus 904. Processor 901 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 902 and / or RAM 903. It should be noted that the program can also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in one or more memories.

[0161] According to an embodiment of the present disclosure, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to the bus 904. The electronic device 900 may further include one or more of the following components connected to the I / O interface 905: an input portion 906 including a keyboard, a mouse, etc. An output portion 907 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc. A storage portion 908 including a hard disk, etc. And a communication portion 909 including a network interface card such as a LAN card, a modem, etc. The communication portion 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the I / O interface 905 as needed. A removable medium 911, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 910 as needed, so that the computer program read therefrom is installed into the storage portion 908 as needed.

[0162] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments. It may also exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0163] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus, or a device. For example, according to an embodiment of the present disclosure, a computer-readable storage medium may include the ROM 902 and / or RAM 903 described above and / or one or more memories other than ROM 902 and RAM 903.

[0164] The embodiment of the present disclosure also includes a computer program product, which includes a computer program, and the computer program contains program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the method provided by the embodiment of the present disclosure.

[0165] The above functions defined in the system / device of the embodiment of the present disclosure are performed when the computer program is executed by the processor 901. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0166] In one embodiment, the computer program may be based on a tangible storage medium such as an optical storage device, a magnetic storage device, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of a signal on a network medium, and downloaded and installed through the communication part 909, and / or installed from a removable medium 911. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0167] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 909, and / or installed from the removable medium 911. When the computer program is executed by the processor 901, the above functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the system, device, means, module, unit, etc. described above can be implemented by a computer program module.

[0168] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).

[0169] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0170] It will be appreciated by those skilled in the art that the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways, even if such combinations and / or combinations are not explicitly described in the present disclosure. In particular, the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways without departing from the spirit and teachings of the present disclosure. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0171] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above separately, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. The scope of the present disclosure is defined by the attached claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A method for evaluating an object, comprising: Obtain Q data sequence features of the object to be evaluated, where Q is an integer greater than or equal to 2; Calculate the complexity index of each data sequence feature among the Q data sequence features, where the complexity index of each data sequence feature is obtained based on the sample entropy of the feature; Dividing the Q data sequence features into K data sets according to the complexity index of each data sequence feature, wherein each data set includes at least one data sequence feature within the range of the corresponding complexity index, and K is an integer greater than or equal to 2; Calling K pre-trained machine learning models from a computer storage space to process the data in the K data sets one by one to obtain K initial evaluation results, wherein the contents of the K machine learning models are different; A fusion process is performed based on the K initial evaluation results and their respective first weight coefficients to obtain a final evaluation result of the object to be evaluated.

2. The method according to claim 1, wherein: Before performing fusion processing based on the K initial evaluation results and their respective first weight coefficients, the method further includes: The initial K machine learning models are trained to obtain the K machine learning models, wherein during the training process, the K first weight coefficients are updated according to the prediction result errors of the one-to-one corresponding K machine learning models, and the prediction result error characterizes the degree of difference between the prediction results of each machine learning model for the training samples and the sample labels.

3. The method according to claim 2, wherein: Updating the K first weight coefficients during the training process includes: After each round of training, the prediction result errors of the K machine learning models are determined; Obtain the sum of the K prediction result errors and the ratio of the prediction result error of each machine learning model to the sum of the errors; The ratio of each machine learning model is updated to the first weight coefficient of the model.

4. The method according to claim 1, wherein: The complexity indicators for calculating each of the Q data sequence features include: Performing phase space reconstruction on each of the data sequence features to obtain corresponding reconstructed data sequence features; Determine a first sample entropy value of each data sequence feature and a second sample entropy value of a reconstructed data sequence feature corresponding to the data sequence feature; Obtaining a complexity index of each data sequence feature according to a difference between the first sample entropy value and the second sample entropy value; For each data sequence feature, when its first sample entropy value is less than the second sample entropy value of the corresponding reconstructed data sequence feature, it has the first complexity index and is divided into the second data set; otherwise, it has the second complexity index and is divided into the third data set.

5. The method according to claim 1, wherein: The complexity indicators for calculating each of the Q data sequence features include: Calculating a third sample entropy value of each data sequence feature; Obtain multiple historical evaluation data of the object to be evaluated; Obtaining sample entropy values ​​of the characteristics of each data sequence in the multiple historical evaluation data, and determining the historical sample entropy range; If the third sample entropy value has a third complexity index within the historical sample entropy range, it is classified into the second data set; otherwise, it has a fourth complexity index and is classified into the third data set.

6. The method according to claim 4 or 5, wherein: The K machine learning models include an integrated autoregressive integrated moving average model and a long short-term memory neural network model, and calling the pre-trained K machine learning models from the computer storage space to process the data in the K data sets one by one includes: The pre-trained integrated autoregressive integrated moving average model is called to process the data in the second data set, and the pre-trained long short-term memory neural network model is called to process the data in the third data set.

7. The method according to claim 6, wherein: Pre-training the integrated autoregressive integrated moving average model includes: Determine L autoregressive integrated moving average models as L basis learners, where L is an integer greater than or equal to 2; Binary encoding the L base learners and assigning them L second weight coefficients respectively; The L base learners are trained based on a genetic learning algorithm to update the L second weight coefficients to obtain respective third weight coefficients of the L base learners. The integrated autoregressive integrated moving average model includes the L base learners and the L third weight coefficients.

8. The method according to claim 1, wherein: Obtaining Q data sequence features of the object to be evaluated includes: Acquire a first data set of the object to be evaluated, wherein the first data set includes original data acquired from the object to be evaluated; The first data set is processed based on the set empirical mode decomposition method to obtain the Q data sequence features.

9. The method according to any one of claims 1 to 5, 7 and 8, wherein: The object to be evaluated includes the enterprise to be evaluated, the Q data sequence features are obtained from the enterprise's basic data, enterprise operating data, enterprise financial data and the enterprise's transaction data in financial institutions, and the initial evaluation result includes the enterprise size category of the enterprise to be evaluated.

10. An object evaluation device, comprising: A feature acquisition module is used to obtain Q data sequence features of the object to be evaluated, where Q is an integer greater than or equal to 2; An index calculation module, used to calculate the complexity index of each data sequence feature among the Q data sequence features, wherein the complexity index of each data sequence feature is obtained based on the sample entropy of the feature; A feature partitioning module, configured to partition the Q data sequence features into K data sets according to the complexity index of each data sequence feature, wherein each data set includes at least one data sequence feature within the range of the corresponding complexity index, and K is an integer greater than or equal to 2; An initial evaluation module, used to call K pre-trained machine learning models from a computer storage space to process the data in the K data sets one by one, and obtain K initial evaluation results, wherein the contents of the K machine learning models are different; The weighted processing module is used to perform fusion processing based on the K initial evaluation results and their respective first weight coefficients to obtain a final evaluation result of the object to be evaluated.

11. An electronic device, comprising: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors are enabled to execute the method according to any one of claims 1 to 9.

12. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to execute the method according to any one of claims 1 to 9.

13. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 9 is implemented.

Images