An algorithm recommendation method and device, a storage medium and an electronic device

Abstract

The application discloses an algorithm recommendation method and device, a storage medium and an electronic device. The method comprises the following steps: selecting data in a sample database based on a data selection mode in each candidate algorithm, and obtaining a data set corresponding to each candidate algorithm; processing first feature data in the data set corresponding to each candidate algorithm based on a feature processing mode in each candidate algorithm, and obtaining each initial feature set; selecting data in each initial feature set based on a feature selection mode in each candidate algorithm, and obtaining each target feature set; analyzing and processing data in each target feature set based on an analysis mode in each candidate algorithm, and obtaining an analysis result corresponding to each candidate algorithm; and screening each candidate algorithm based on each analysis result, and obtaining a target algorithm. The method can save the time for selecting an algorithm, and improve the speed of selecting an algorithm.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to an algorithm recommendation method and device, a storage medium and an electronic device. BACKGROUND

[0002] In existing intelligent search, intelligent recommendation and other application services, algorithms are needed to determine target recommendation content. Therefore, the selection of algorithms is crucial in various intelligent recommendation and other application services, which guarantees accurate recommendation.

[0003] In the prior art, the selection of algorithms is completed by algorithm engineers. In the process of business application, algorithm engineers evaluate several known algorithms according to experience based on business data structure and process, select several algorithms, then test the algorithms, and compare the test results to determine the final algorithm. However, due to the uneven experience of each algorithm engineer, the paths explored are different, and the solutions obtained are also different. Moreover, the trial process consumes a lot of time, and a small adjustment needs to go through the processes of changing, calculating, waiting and re-verifying, which consumes a lot of time.

[0004] Therefore, there is an urgent need for an algorithm recommendation method to solve the problem of long time consumption in the selection of algorithms in the prior art. SUMMARY

[0005] Therefore, the present application provides an algorithm recommendation method, device, storage medium and electronic device, which mainly aims to solve the problem of long time consumption in the current algorithm selection process.

[0006] To solve the above problems, the present application provides an algorithm recommendation method, comprising:

[0007] Based on the data selection mode in each candidate algorithm, data in the sample database is selected to obtain a data set corresponding to each candidate algorithm;

[0008] Based on the feature processing mode in each candidate algorithm, first feature data in the data set corresponding to each candidate algorithm is processed to obtain an initial feature set corresponding to each candidate algorithm;

[0009] Based on the feature selection mode in each candidate algorithm, data in the initial feature set corresponding to each candidate algorithm is selected to obtain a target feature set corresponding to each candidate algorithm;

[0010] Based on the analysis mode in each candidate algorithm, data in the target feature set corresponding to each candidate algorithm is analyzed and processed to obtain an analysis result corresponding to each candidate algorithm.

[0011] Screening each of the candidate algorithms based on each of the analysis results, to obtain a target algorithm.

[0012] Optionally, the method further comprises: obtaining a plurality of data corresponding to a target project, to obtain the sample database;

[0013] The data in the sample database is selected based on the data selection mode in each of the candidate algorithms, to obtain a data set corresponding to each of the candidate algorithms, specifically comprising:

[0014] Based on the data selection mode in each of the candidate algorithms, first identification information corresponding to the data selection mode is determined;

[0015] The selected data is marked based on each of the first identification information, and the marked data is taken as data in the data set corresponding to each of the candidate algorithms;

[0016] The data selection mode includes any one of the following: leave-out method, leave-one-out method, bootstrap method, and time period screening method.

[0017] Optionally, the initial feature set corresponding to each of the candidate algorithms is obtained by processing first feature data in the data set corresponding to each of the candidate algorithms based on a feature processing mode in each of the candidate algorithms, specifically comprising:

[0018] Based on the feature processing mode in each of the candidate algorithms, second identification information corresponding to each of the feature processing modes is determined;

[0019] The first feature data in the data set corresponding to the candidate algorithm is marked based on the second identification information;

[0020] Each of the first feature data is processed based on the second identification information marked by the first feature data, using a feature processing mode corresponding to the second identification information, and the modified first feature data is taken as data in the initial feature set;

[0021] The feature processing mode includes any one or more of the following: bucket processing, normalization processing, encoding processing, and denoising processing.

[0022] Optionally, the target feature set corresponding to each of the candidate algorithms is obtained by selecting data in the initial feature set corresponding to each of the candidate algorithms based on a feature selection mode in each of the candidate algorithms, specifically comprising:

[0023] Based on the feature selection mode in each of the candidate algorithms, third identification information corresponding to the feature selection mode is determined;

[0024] label the modified first feature data in the initial feature set corresponding to the candidate algorithm based on the third identification information;

[0025] select the modified first feature data based on the third identification information by which the modified first feature data is labeled, using a feature selection mode corresponding to the third identification information, and use the selected modified first feature data as data in the target feature set;

[0026] The feature selection mode includes any one of the following: information gain mode, cross-entropy mode, relative entropy mode, odds ratio mode, and mutual information mode.

[0027] Optionally, the analysis mode in each of the candidate algorithms is used to analyze and process the data in the target feature set corresponding to each of the candidate algorithms to obtain an analysis result corresponding to each of the candidate algorithms, and specifically includes:

[0028] The fourth identification information corresponding to the analysis mode is determined based on the analysis mode in each of the candidate algorithms.

[0029] The selected modified first feature data in the target feature set corresponding to the candidate algorithm is labeled based on the fourth identification information.

[0030] The selected modified first feature data is analyzed and processed based on the fourth identification information by which the selected modified first feature data is labeled, using an analysis mode corresponding to the fourth identification information, to obtain an analysis result corresponding to each of the candidate algorithms.

[0031] The analysis mode includes any one of the following: iterative algorithm, K-means clustering algorithm, web page ranking method, association rule algorithm, classification and regression tree algorithm, Naive Bayes classification algorithm, logistic regression algorithm, support vector machine algorithm, proximity algorithm, decision tree algorithm, and gradient descent tree algorithm.

[0032] Optionally, the candidate algorithms are screened based on the analysis results to obtain a target algorithm, and specifically includes:

[0033] A target evaluation mode is determined.

[0034] The analysis results are evaluated based on the target evaluation mode to obtain an evaluation coefficient corresponding to each of the analysis results.

[0035] The target algorithm is determined from the candidate algorithms based on the evaluation coefficient of each analysis result.

[0036] To solve the above problems, the application provides an algorithm recommendation device, comprising:

[0037] The first selection module is configured to select data in the sample database based on a data selection manner in each candidate algorithm, and obtain a data set corresponding to each candidate algorithm;

[0038] The processing module is configured to process first feature data in the data set corresponding to each candidate algorithm based on a feature processing manner in each candidate algorithm, and obtain an initial feature set corresponding to each candidate algorithm;

[0039] The second selection module is configured to select data in the initial feature set corresponding to each candidate algorithm based on a feature selection manner in each candidate algorithm, and obtain a target feature set corresponding to each candidate algorithm;

[0040] The analysis module is configured to analyze and process data in the target feature set corresponding to each candidate algorithm based on an analysis manner in each candidate algorithm, and obtain an analysis result corresponding to each candidate algorithm;

[0041] The screening module is configured to screen each candidate algorithm based on the analysis result, and obtain a target algorithm.

[0042] Optionally, the first selection module is specifically configured to:

[0043] Based on the data selection manner in each candidate algorithm, first identification information corresponding to the data selection manner is determined;

[0044] Based on each first identification information, the selected data is marked, and the marked data is used as data in the data set corresponding to each candidate algorithm;

[0045] The data selection manner includes any one of the following: leave-out method, leave-one-out method, bootstrap method, and time period screening method.

[0046] To solve the above problems, the application provides a storage medium, comprising: the storage medium stores a computer program, and the computer program is executed by a processor to realize the steps of the algorithm recommendation method in any one of the above.

[0047] To solve the above problems, the application provides an electronic device, comprising at least a memory and a processor, the memory stores a computer program, and the processor realizes the steps of the algorithm recommendation method in any one of the above when executing the computer program on the memory.

[0048] The algorithm recommendation method, device, storage medium and electronic device in the application can automatically select the most suitable target algorithm from a large number of algorithms, and then apply the target algorithm to a target application service such as intelligent recommendation, thereby saving the time for algorithm selection and improving the accuracy of algorithm selection, and solving the problem that due to data with more than ten thousand dimensions, each dimension needs to be calculated by a complex algorithm, a large amount of data needs to be trained, and each subtle adjustment needs to go through the whole process, thereby causing the selection of the algorithm to take a long time.

[0049] The above description is only a summary of the technical solutions of the present application, in order to enable the technical means of the present application to be more clearly understood, and can be implemented according to the content of the specification, and in order to enable the above and other purposes, characteristics and advantages of the present application to be more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0050] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of the preferred embodiments and are not meant to limit the present application. Furthermore, the same reference numerals are used throughout the several drawings to refer to same or like parts. In the drawings:

[0051] Figure 1 A flowchart of an algorithm recommendation method according to an embodiment of the present application;

[0052] Figure 2 A flowchart of an algorithm recommendation method according to another embodiment of the present application;

[0053] Figure 3 A structure block diagram of an algorithm recommendation device according to another embodiment of the present application. DETAILED DESCRIPTION

[0054] The various aspects and features of the present application are described herein with reference to the accompanying drawings.

[0055] It should be understood that various modifications can be made to the embodiments described herein. Therefore, the above description should not be taken as limiting, but merely as exemplification of the embodiments. Those skilled in the art will envision other modifications within the scope and spirit of the application.

[0056] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the present application and, together with the general description of the application given above, and the detailed description of the embodiments given below, serve to explain the principles of the present application.

[0057] These and other characteristics of the present application will become apparent from the following description of the preferred forms given, by way of non-limiting example only, with reference to the attached drawings.

[0058] It should also be understood that, while the present application has been described in terms of certain embodiments, the skilled person will be able to ascertain many other equivalent forms of the present application without departing from the application.

[0059] The above and other aspects, features, and advantages of the present application will become more apparent from the following detailed description, taken in conjunction with the accompanying drawings, which illustrate

[0060] Specific embodiments of the present application are described hereinbelow, with reference to the drawings; however, it should be understood that the disclosed embodiments are merely examples of the present application, which can be practiced in numerous ways. Well-known and / or repetitive functions and structures are not described in detail to avoid obscuring the present application unnecessarily. Therefore, specific structural and functional details disclosed herein are not intended to limit the present application, but merely as a basis for the claims and a representative basis for teaching one skilled in the art to variously employ the present application in virtually any appropriate detailed structure.

[0061] The specification can use phrases such as "in one embodiment", "in another embodiment", "in yet another embodiment", or "in other embodiments", which can refer to one or more of the same or different embodiments under the present application.

[0062] The embodiments of the present application provide an algorithm recommendation method, as shown in the following flowchart, comprising the following steps: Figure 1

[0063] Step S101, based on the data selection manner in each candidate algorithm, selecting data in the sample database to obtain a data set corresponding to each candidate algorithm;

[0064] In the specific implementation process, the sample database can be created in advance based on the target project, that is, a plurality of data corresponding to the target project are obtained to obtain the sample database. The target project in the embodiments of the present application specifically refers to a target application / target platform to which the algorithm is applied, for example, the target project can be shopping software, news software, etc. The candidate algorithm in the present step refers to each push method that can be applied to the shopping software or the news software, each candidate algorithm contains a plurality of types of data processing manners, that is, each candidate algorithm is composed of each type of data processing manner. The data selection manner is one of the types of data processing manners. The data selection manner in the present step can specifically include any one or several of the following: hold-out method, leave-one-out method, bootstrap method, time period screening method, etc. Since there are a plurality of types of data processing manners, a plurality of candidate algorithms can be obtained by combination.

[0065] ​Step S102, based on the feature processing mode in each of the candidate algorithms, processing the first feature data in the data set corresponding to each of the candidate algorithms to obtain an initial feature set corresponding to each of the candidate algorithms;

[0066] In this step, the feature processing mode specifically includes one or more of the following: bucket processing, normalization processing, encoding processing, denoising processing, etc. In the specific implementation process, the data can be stored in columns, and the data columns can be labeled. A single feature and multiple processing modes will generate multiple data, so the marker bit has only one bit of 1. By using this method to process the data, the problem of excessive data and the pressure of data storage can be avoided, and data explosion can be avoided.

[0067] Step S103, based on the feature selection mode in each of the candidate algorithms, selecting data in the initial feature set corresponding to each of the candidate algorithms to obtain a target feature set corresponding to each of the candidate algorithms;

[0068] In the specific implementation process, the feature selection mode includes any one or more of the following: information gain mode, cross-entropy mode, KL divergence mode / relative entropy mode, odds ratio / odds ratio mode, mutual information mode. Since different feature selection modes are used to select data in the same initial feature set, different results will be obtained, so the selection of the feature selection mode as a sub-algorithm in the target algorithm is the key to determining the target algorithm.

[0069] In the specific implementation process, the data can also be labeled by using the labeling method, that is, a feature selection mode marker bit is added, so that multiple selection algorithms can mark the same data, and the combination serial number is additionally fissured. By increasing the marker bit of the feature selection algorithm, the problem of excessive data can be prevented.

[0070] Step S104, based on the analysis mode in each of the candidate algorithms, analyzing and processing the data in the target feature set corresponding to each of the candidate algorithms to obtain an analysis result corresponding to each of the candidate algorithms;

[0071] In this step, the analysis mode specifically includes any one or more of the following: iterative algorithm Adaboost, K-means clustering algorithm K-Means, page ranking method PageRank, association rule algorithm Aproori, classification and regression tree algorithm CART, naive Bayes classification algorithm Nave Bayes, logistic regression algorithm LR, support vector machine algorithm SVM, neighbor algorithm KNN, decision tree algorithm C4.5, gradient descent tree algorithm GDBT.

[0072] In the implementation process of the step, before the analysis and processing of the feature set corresponding to each algorithm, the target computing platform can also be determined, that is, the target computing platform is used to analyze and calculate each target feature data in the corresponding target feature set based on each algorithm, so as to obtain the analysis result, that is, the result of intelligent recommendation. That is, by distributing the calculation amount to each computing platform, the calculation speed can be accelerated.

[0073] Step S105: screening each candidate algorithm based on each analysis result to obtain a target algorithm.

[0074] In the step, after obtaining each analysis result, each analysis result can be evaluated and calculated by a predetermined evaluation method, so as to obtain a corresponding evaluation coefficient, and finally the target algorithm is determined according to the size of the evaluation coefficient.

[0075] The method in the embodiment can automatically select the most suitable target algorithm from a large number of algorithms, and then apply the target algorithm to the target application service such as intelligent recommendation, thereby saving the time of algorithm selection and improving the accuracy of algorithm selection. At the same time, the problem of long time consumption of algorithm selection due to the following reasons is solved: the data has more than ten thousand dimensions, each dimension needs to be calculated by a complex algorithm, a large amount of data needs to be trained, each slight adjustment needs to go through the whole process again, and then the selection of the algorithm is time-consuming.

[0076] Another embodiment of the present application provides an algorithm recommendation method, as shown in Figure 2 The method comprises the following steps:

[0077] Step S201: obtaining a plurality of data corresponding to a target project to obtain a sample database;

[0078] Step S202: determining first identification information corresponding to a data selection mode in each candidate algorithm based on the data selection mode; marking the selected data based on each first identification information, and taking the marked data as data in a data set corresponding to each candidate algorithm;

[0079] The data selection method in this step includes any one of the following: leave-out method, leave-one-out method, bootstrap method, and time period screening method. In the specific implementation process, a first correspondence relationship between each data selection method and each first identification information can be established in advance. When it is necessary to select data in the sample database using the data selection method in the candidate algorithm, the first correspondence relationship can be searched to determine the first identification information corresponding to the data selection method, and then the data in the sample database is labeled using the first identification information to obtain a data set corresponding to the candidate algorithm. Specifically, when labeling based on the first identification information, the identification information of the candidate algorithm can also be added at the same time, that is, a marking bit of the candidate algorithm is added to each data, so that it can be accurately determined which data selection method in the candidate algorithm labels the data. In this step, by labeling the data, the problem of excessive data caused by too many algorithms and too many data sets corresponding to each algorithm can be avoided.

[0080] In step S203, second identification information corresponding to each feature processing method in each candidate algorithm is determined based on the feature processing method. First feature data in the data set corresponding to the candidate algorithm is labeled based on the second identification information. Each first feature data is processed using the feature processing method corresponding to the second identification information based on the second identification information of the labeled first feature data, and the modified first feature data is used as data in the initial feature set.

[0081] In the specific implementation process of this step, the feature processing method includes any one or several of the following: bucket processing, normalization processing, encoding processing, and denoising processing. In the specific implementation process of this step, a second correspondence relationship between each second identification information and each feature processing method can also be established in advance. When it is necessary to process the first feature data in the data set to obtain the initial feature set, the corresponding second identification information can be determined by searching the second correspondence relationship, and then the first feature data in the data set is further labeled using the second identification information. Subsequently, the first feature data can be directly processed through the labeling result, so that the problem of repeated processing and excessive data caused by the same first feature data and the same feature data processing method in multiple candidate algorithms can be avoided, and the data storage pressure is also reduced.

[0082] Step S204, determining third identification information corresponding to the feature selection mode based on the feature selection mode in each of the candidate algorithms; marking the modified first feature data in the initial feature set corresponding to the candidate algorithm based on each of the third identification information; selecting each of the modified first feature data based on the third identification information marked on the modified first feature data, and using the feature selection mode corresponding to the third identification information, and taking the selected modified first feature data as data in each of the target feature sets;

[0083] In the implementation process of the present step, the feature selection mode includes any one of the following: information gain mode, cross entropy mode, KL divergence mode / relative entropy mode, odds ratio mode / ratio mode, mutual information mode. In the implementation process of the present step, a third correspondence relationship between each third identification information and each feature selection mode can also be established, so that when the data in the initial feature set needs to be processed to obtain the target feature set, the corresponding third identification information can be determined by searching the second correspondence relationship, and then the data in the data set is further marked by using the third identification information. Subsequently, the data can be directly processed by using the marking result, so that the problem of repeated processing of data and excessive amount of data caused by the same second feature data and the same feature selection mode involved in multiple candidate algorithms can be avoided.

[0084] Step S205, determining fourth identification information corresponding to each of the analysis modes based on the analysis mode in each of the candidate algorithms; marking the selected modified first feature data in the target feature set corresponding to the candidate algorithm based on the fourth identification information; and analyzing and processing each of the selected modified first feature data based on the fourth identification information marked on the selected modified first feature data, using the analysis mode corresponding to the fourth identification information, to obtain an analysis result corresponding to each of the candidate algorithms.

[0085] In the implementation process of the present step, the analysis mode includes any one of the following: iterative algorithm Adaboost, K-means clustering algorithm K-Means, PageRank, association rule algorithm Aproori, classification and regression tree algorithm CART, Nave Bayes classification algorithm, logistic regression algorithm LR, support vector machine algorithm SVM, KNN algorithm, decision tree algorithm C4.5, and gradient descent tree algorithm GDBT.

[0086] The fourth corresponding relationship between the fourth identification information and the analysis manners can be established, so that when the data in the target feature set needs to be analyzed and processed to obtain an analysis result, the corresponding fourth identification information can be determined by searching the fourth corresponding relationship, and then the data in the data set is further marked by using the fourth identification information, which can avoid the problems of repeated data processing and excessive data quantity.

[0087] In the specific implementation process, before the analysis and processing of the feature set corresponding to each algorithm, the target computing platform can be determined, that is, the target computing platform is used to analyze and calculate each target feature data in the corresponding feature set based on each algorithm, so as to obtain the analysis result, that is, the result of intelligent recommendation. That is, by distributing the calculation amount to each computing platform, the calculation speed can be accelerated. In this step, the target computing platform specifically includes any one of the following: Flink platform, Spark platform, and Hadoop platform.

[0088] In step S206, a target evaluation manner is determined; each analysis result is evaluated based on the target evaluation manner to obtain an evaluation coefficient corresponding to each analysis result; and a target algorithm is determined from each candidate algorithm based on the evaluation coefficient of each analysis result.

[0089] In the specific implementation process, the target evaluation manner includes any one or several of the following: precision, recall, accuracy, F1 score, mean absolute error, mean square error, releasable variance score, and R2 score. That is, the evaluation coefficient can be calculated by using the above evaluation manners, and finally the target algorithm is determined based on the high and low of each evaluation coefficient. In the specific implementation process, each candidate algorithm can be sorted according to the high and low order of the evaluation coefficient, and then at least one target recommended algorithm is determined according to the sorting result, for example, the candidate algorithms in the top three are determined as the target algorithm and are displayed. In the specific implementation process, the sorting result of each candidate algorithm can also be displayed, so that the user can more clearly understand the recommendation level of each candidate algorithm.

[0090] In this embodiment, after the target algorithm is determined, real-time training can be performed, for example, the data of real-time feedback is connected, and the optimal combination of each data processing manner is performed, that is, each data processing manner in the target algorithm is gradually trained, so as to provide real-time algorithm service for the corresponding target project. In this embodiment, the evaluation scores of each candidate algorithm can also be stored as reference data to support subsequent business combination and reduce the number of combinations.

[0091] The intelligent recommendation method in the application can replace the work of algorithm engineers and complete a large amount of exploration and selection work that cannot be completed by algorithm engineers. In the application, candidate algorithms are obtained by intelligently combining various data processing methods, and then corresponding data calculation, data processing and other tasks are intelligently performed according to the added mark information of the data to obtain evaluation coefficients corresponding to the candidate algorithms. Finally, the candidate algorithms are sorted and displayed according to the high-low order of the evaluation coefficients, so that an optimal algorithm is selected and provided to an application programming interface (API) service, without the need for engineers to additionally develop the API and repeatedly process the logic of features. The API can realize the function of data analysis AB test. The method in the embodiment supports real-time business scenarios and real-time training, and can also be intervened by humans, that is, the data analysis method AB test is selected by humans to perform data analysis, so as to more accurately obtain the evaluation results of the candidate algorithms and realize optimal selection. The method in the application can also save a large amount of cost for enterprises and provide data-persuasive support for business development.

[0092] The algorithm recommendation method, device, storage medium and electronic equipment in the application can automatically select the most suitable target algorithm from a large number of algorithms, and then apply the target algorithm to a target application service such as intelligent recommendation, thereby saving the time for algorithm selection and improving the accuracy of algorithm selection. At the same time, the problem that the selection of the algorithm takes a long time due to the fact that the data has more than ten thousand dimensions, each dimension needs to be calculated by a complex algorithm, a large amount of data needs to be trained, and each subtle adjustment needs to go through the entire process again is solved, and at the same time, the subsequent target platform is guaranteed to provide intelligent recommendation based on the target algorithm and historical browsing information of the user.

[0093] Another embodiment of the application provides an algorithm recommendation device, as shown in Figure 3 The device comprises:

[0094] A first selection module 1 is configured to select data in a sample database based on a data selection method in each candidate algorithm to obtain a data set corresponding to each candidate algorithm.

[0095] A processing module 2 is configured to process first feature data in the data set corresponding to each candidate algorithm based on a feature processing method in each candidate algorithm to obtain an initial feature set corresponding to each candidate algorithm.

[0096] A second selection module 3 is configured to select data in the initial feature set corresponding to each candidate algorithm based on a feature selection method in each candidate algorithm to obtain a target feature set corresponding to each candidate algorithm.

[0097] The analysis module 4 is configured to analyze and process data in the target feature set corresponding to each candidate algorithm based on an analysis manner in each candidate algorithm, and obtain an analysis result corresponding to each candidate algorithm.

[0098] The screening module 5 is configured to screen each candidate algorithm based on the analysis result, and obtain a target algorithm.

[0099] In the embodiment, the algorithm recommendation device further includes an acquisition module configured to acquire a plurality of data corresponding to a target project to obtain the sample database; and the first selection module is specifically configured to determine first identification information corresponding to a data selection manner in each candidate algorithm based on the data selection manner, mark the selected data based on each first identification information, and take the marked data as data in a data set corresponding to each candidate algorithm; and the data selection manner includes any one of the following: a leave-out method, a leave-one-out method, a bootstrap method, and a time period screening method.

[0100] In the embodiment, the processing module is specifically configured to determine second identification information corresponding to a feature processing manner in each candidate algorithm based on the feature processing manner, mark first feature data in the data set corresponding to the candidate algorithm based on the second identification information, process each first feature data based on second identification information of the marked first feature data by using a feature processing manner corresponding to the second identification information, and take the modified first feature data as data in the initial feature set; and the feature processing manner includes any one or more of the following: a bucket processing, a normalization processing, an encoding processing, and a noise reduction processing.

[0101] In the embodiment, the second selection module is specifically configured to determine third identification information corresponding to a feature selection manner in each candidate algorithm based on the feature selection manner, mark the modified first feature data in the initial feature set corresponding to the candidate algorithm based on each third identification information, select each modified first feature data based on third identification information of the marked first feature data by using a feature selection manner corresponding to the third identification information, and take the selected modified first feature data as data in each target feature set; and the feature selection manner includes any one of the following: an information gain manner, a cross-entropy manner, a relative entropy manner, a ratio ratio manner, and a mutual information manner.

[0102] In the specific implementation process of the embodiment, the analysis module is specifically configured to: determine fourth identification information corresponding to each analysis manner based on the analysis manners in each candidate algorithm; mark the modified first feature data obtained through selection in the target feature set corresponding to the candidate algorithm based on the fourth identification information; and analyze and process each selected modified first feature data based on the fourth identification information by using the analysis manner corresponding to the fourth identification information, to obtain an analysis result corresponding to each candidate algorithm. The analysis manner includes any one of the following: an iterative algorithm, a K-means clustering algorithm, a webpage ranking method, an association rule algorithm, a classification regression tree algorithm, a Naive Bayes classification algorithm, a logistic regression algorithm, a support vector machine algorithm, a proximity algorithm, a decision tree algorithm, and a gradient descent tree algorithm.

[0103] In the specific implementation process of the embodiment, the screening module is specifically configured to: determine a target evaluation manner; evaluate each analysis result based on the target evaluation manner, to obtain an evaluation coefficient corresponding to each analysis result; and determine a target algorithm from each candidate algorithm based on the evaluation coefficient of each analysis result.

[0104] The algorithm recommendation device in the application can automatically select the most suitable target algorithm from a large number of algorithms, and then apply the target algorithm to a target application service such as intelligent recommendation, thereby saving the time for algorithm selection and improving the accuracy of algorithm selection, and solving the problem that the selection of the algorithm takes a long time due to the fact that the data has more than ten thousand dimensions, each dimension needs to be calculated by a complex algorithm, a large amount of data needs to be trained, and each slight adjustment needs to go through the entire process again.

[0105] Another embodiment of the application provides a storage medium storing a computer program, which, when executed by a processor, implements the following method steps:

[0106] Step one, selecting data in a sample database based on a data selection manner in each candidate algorithm, to obtain a data set corresponding to each candidate algorithm;

[0107] Step two, processing first feature data in the data set corresponding to each candidate algorithm based on a feature processing manner in each candidate algorithm, to obtain an initial feature set corresponding to each candidate algorithm;

[0108] Step three, selecting data in the initial feature set corresponding to each candidate algorithm based on a feature selection manner in each candidate algorithm, to obtain a target feature set corresponding to each candidate algorithm;

[0109] Step four, based on the analysis method in each of the candidate algorithms, data in the target feature set corresponding to each of the candidate algorithms is analyzed and processed to obtain an analysis result corresponding to each of the candidate algorithms;

[0110] Step five, based on each of the analysis results, each of the candidate algorithms is screened to obtain a target algorithm.

[0111] The specific implementation process of the above method steps can be referred to the embodiments of any one of the algorithm recommendation methods described above, which will not be repeated here.

[0112] The storage medium in the present application can automatically select the most suitable target algorithm from a large number of algorithms, and then apply the target algorithm to intelligent recommendation and other target application services, saving the time of algorithm selection and improving the accuracy of algorithm selection, and at the same time solving the problem that due to the data having more than ten thousand dimensions, each dimension needs to be calculated by a complex algorithm, a large amount of data needs to be trained, and each subtle adjustment needs to go through the whole process, thereby making the selection of the algorithm time-consuming.

[0113] Another embodiment of the present application provides an electronic device comprising at least a memory and a processor, the memory storing a computer program, and the processor implementing the following method steps when executing the computer program on the memory:

[0114] Step one, based on the data selection method in each candidate algorithm, data in a sample database is selected to obtain a data set corresponding to each of the candidate algorithms;

[0115] Step two, based on the feature processing method in each of the candidate algorithms, first feature data in the data set corresponding to each of the candidate algorithms is processed to obtain an initial feature set corresponding to each of the candidate algorithms;

[0116] Step three, based on the feature selection method in each of the candidate algorithms, data in the initial feature set corresponding to each of the candidate algorithms is selected to obtain a target feature set corresponding to each of the candidate algorithms;

[0117] Step four, based on the analysis method in each of the candidate algorithms, data in the target feature set corresponding to each of the candidate algorithms is analyzed and processed to obtain an analysis result corresponding to each of the candidate algorithms;

[0118] Step five, based on each of the analysis results, each of the candidate algorithms is screened to obtain a target algorithm.

[0119] The specific implementation process of the above method steps can be referred to the embodiments of any one of the algorithm recommendation methods described above, which will not be repeated here.

[0120] The electronic device in the application can automatically select the most suitable target algorithm from a plurality of algorithms, and then apply the target algorithm to a target application service such as intelligent recommendation, thereby saving the time for algorithm selection and improving the accuracy of algorithm selection, and solving the problem that due to the data having more than ten thousand dimensions, each dimension needs to be calculated by a complex algorithm, a large amount of data needs to be trained, and each subtle adjustment needs to go through the whole process again, thereby making the selection of the algorithm time-consuming.

[0121] The above embodiments are only exemplary embodiments of the application and are not used to limit the application, and the protection scope of the application is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements to the application within the spirit and protection scope of the application, and such modifications or equivalent replacements are also regarded as falling within the protection scope of the application.

Claims

1. An algorithm recommendation method, characterized by, The method comprises the following steps: Obtaining a plurality of data corresponding to a target project to obtain a sample database; the sample database contains a plurality of shopping data, or the sample database contains a plurality of news data; Selecting data in the sample database containing a plurality of shopping data or a plurality of news data based on a data selection mode in each candidate algorithm to obtain a data set corresponding to each candidate algorithm; the candidate algorithm is each push algorithm applied to a shopping software or a news software; Processing first feature data in the data set corresponding to each candidate algorithm based on a feature processing mode in each candidate algorithm to obtain an initial feature set corresponding to each candidate algorithm; Selecting data in the initial feature set corresponding to each candidate algorithm based on a feature selection mode in each candidate algorithm to obtain a target feature set corresponding to each candidate algorithm; Analyzing and processing data in the target feature set corresponding to each candidate algorithm based on an analysis mode in each candidate algorithm to obtain an analysis result corresponding to each candidate algorithm; the analysis result represents a result of intelligent recommendation; Selecting each candidate algorithm based on each analysis result to obtain a target algorithm; The method for selecting data in the sample database containing a plurality of shopping data or a plurality of news data based on a data selection mode in each candidate algorithm to obtain a data set corresponding to each candidate algorithm comprises the following steps: Determining first identification information corresponding to the data selection mode based on the data selection mode in each candidate algorithm; Marking the selected shopping data or news data based on each first identification information, and taking the marked data as data in the data set corresponding to each candidate algorithm; the data is stored in a column, and the data column is marked.

2. The method of claim 1, wherein, The data selection mode comprises any one of the following: leave-out method, leave-one-out method, bootstrap method, and time period screening method.

3. The method of claim 1, wherein, The method for processing first feature data in the data set corresponding to each candidate algorithm based on a feature processing mode in each candidate algorithm to obtain an initial feature set corresponding to each candidate algorithm comprises the following steps: Determining second identification information corresponding to each feature processing mode based on the feature processing mode in each candidate algorithm; Marking first feature data in the data set corresponding to each candidate algorithm based on the second identification information; Processing each first feature data based on the second identification information corresponding to the second identification information based on the second identification information of each first feature data, and taking the modified first feature data as data in the initial feature set; The feature processing mode comprises any one or several of the following: bucket processing, normalization processing, encoding processing, and noise reduction processing.

4. The method of claim 3, wherein, The method for selecting data in the initial feature set corresponding to each candidate algorithm based on a feature selection mode in each candidate algorithm to obtain a target feature set corresponding to each candidate algorithm comprises the following steps: determine third identification information corresponding to the feature selection manner in each of the candidate algorithms based on the feature selection manner; label the modified first feature data in the initial feature set corresponding to the candidate algorithm based on the third identification information; select each of the modified first feature data based on the third identification information by which each of the modified first feature data is labeled, using the feature selection manner corresponding to the third identification information, and take the selected modified first feature data as data in each of the target feature sets; The feature selection manner includes any one of the following: information gain manner, cross entropy manner, relative entropy manner, ratio ratio manner, mutual information manner.

5. The method of claim 1, wherein, The analysis manner in each of the candidate algorithms is used to analyze and process the data in the target feature set corresponding to each of the candidate algorithms to obtain an analysis result corresponding to each of the candidate algorithms, specifically including: determine fourth identification information corresponding to the analysis manner in each of the candidate algorithms based on the analysis manner; label the selected modified first feature data in the target feature set corresponding to the candidate algorithm based on the fourth identification information; analyze and process each of the selected modified first feature data based on the fourth identification information by which each of the selected modified first feature data is labeled, using the analysis manner corresponding to the fourth identification information, to obtain an analysis result corresponding to each of the candidate algorithms; The analysis manner includes any one of the following: iterative algorithm, K-means clustering algorithm, web page ranking method, association rule algorithm, classification regression tree algorithm, naive Bayes classification algorithm, logistic regression algorithm, support vector machine algorithm, proximity algorithm, decision tree algorithm, gradient descent tree algorithm.

6. The method of claim 1, wherein, The analysis result is used to screen each of the candidate algorithms to obtain a target algorithm, specifically including: determine a target evaluation manner; evaluate each of the analysis results based on the target evaluation manner to obtain an evaluation coefficient corresponding to each of the analysis results; determine a target algorithm from each of the candidate algorithms based on the evaluation coefficient of each analysis result.

7. An algorithm recommendation apparatus characterized by comprising: including: an acquisition module configured to acquire a plurality of data corresponding to a target project to obtain a sample database; The sample database contains a plurality of shopping data, or the sample database contains a plurality of news data; A first selection module is configured to select data in a sample database containing a plurality of shopping data or a plurality of news data based on a data selection manner in each candidate algorithm to obtain a data set corresponding to each of the candidate algorithms; the candidate algorithm is each push algorithm applied to a shopping software or a news software; A processing module is configured to process first feature data in a data set corresponding to each of the candidate algorithms based on a feature processing manner in each of the candidate algorithms to obtain an initial feature set corresponding to each of the candidate algorithms; The second selection module is configured to select data in the initial feature set corresponding to each candidate algorithm based on a feature selection manner in each candidate algorithm, and obtain a target feature set corresponding to each candidate algorithm. The analysis module is configured to analyze and process data in the target feature set corresponding to each candidate algorithm based on an analysis manner in each candidate algorithm, and obtain an analysis result corresponding to each candidate algorithm. The analysis result represents a result of intelligent recommendation. The screening module is configured to screen each candidate algorithm based on the analysis result, and obtain a target algorithm. The first selection module is specifically configured to determine first identification information corresponding to a data selection manner in each candidate algorithm based on the data selection manner; mark selected data based on the first identification information, and store the marked data as data in a data set corresponding to each candidate algorithm. The data is stored in a column, and the data column is marked.

8. The algorithm recommendation apparatus of claim 7, wherein, The data selection manner includes any one of the following: a leave-out method, a leave-one-out method, a bootstrap method, and a time period screening method.

9. A storage medium, characterized by The storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the algorithm recommendation method in any one of claims 1-6.

10. An electronic device, comprising: The device at least includes a memory and a processor. The memory stores a computer program, and the processor implements the steps of the algorithm recommendation method in any one of claims 1-6 when executing the computer program stored in the memory.

Images