Data analysis method and system for additive screening

Abstract

The invention relates to the technical field of machine learning, in particular to a data analysis method and system for additive screening, and the method comprises the steps: dividing a plurality of historical data into a backup historical data set and an exercise historical data set; selecting a plurality of historical data from all the historical data belonging to the training historical data set, training and generating an analysis model by using the selected historical data, judging whether the analysis model meets a preset first requirement or not, continuing the next step if the analysis model meets the preset first requirement, and repeating the step if the analysis model does not meet the preset first requirement; and determining a plurality of historical data which are not selected in the exercise historical data set, performing exercise by using the determined plurality of historical data to generate an explanation model, inputting the additive information into the analysis model to obtain effect information, and inputting the additive information into the explanation model to obtain explanation information. According to the invention, the accuracy of the estimated effect information can be ensured.

Description

Technical Field

[0001] This application relates to the technical field of machine learning, and particularly to a data analysis method and system for additive screening. Background Art

[0002] Semen cryopreservation is a commonly used assisted reproductive technology for livestock at present. By screening high- and low-freezing-resistant semen of pigs and analyzing the changes in proteins and lipids therein, differential proteins and differential lipids can be used as additives for new cryoprotectants. Studying the effects of adding different additives to frozen semen on the freezing effect of pig sperm using machine learning can save a large amount of experimental time and is of great significance.

[0003] Similar prior arts include a Chinese patent application with the publication number CN116796159A, which provides a dyeing effect prediction method, a training method and device for a dyeing effect prediction model, and particularly relates to artificial intelligence technology, neural network model technology, and model training technology. The method includes: using a wavelet basis function to decompose the Raman spectrum data of a to-be-tested silk spindle into multiple sub-signal data; determining the characteristics of the to-be-tested silk spindle according to at least part of the target sub-signal data among the multiple sub-signal data; and using a dyeing effect prediction model to predict a first dyeing label according to the characteristics of the to-be-tested silk spindle. Among them, the first dyeing label is used to characterize the dyeing effect level. Another similar prior art is a Chinese patent application with the publication number CN119416066A, which discloses a sewage treatment effect prediction method and system based on deep learning. The method includes data collection, data preprocessing, construction of a treatment effect prediction model, hyperparameter optimization, and sewage treatment effect prediction. The original sewage treatment data is obtained through data collection; a data preprocessing method including data cleaning, data annotation, data encoding, data normalization, and dataset segmentation is adopted; and an improved spatio-temporal attention graph convolutional model is used for sewage treatment effect prediction. However, the accuracy of the effect prediction methods in the above two patent applications needs to be improved. Summary of the Invention

[0004] This application divides a number of historical data into a backup historical data group and a practice historical data group respectively, uses the practice historical data group to practice and generate an analysis model and an explanation model, uses the backup historical data group to perform verification processing on the analysis model and the explanation model, uses the analysis model to obtain the effect information corresponding to the additive information, and uses the explanation model to obtain the explanation information related to the effect information. This application aims to improve the accuracy of the estimated effect information.

[0005] This application provides a data analysis method for additive screening, including the following steps: The preparation module obtains a number of historical data composed of historical additive information and historical effect information, and the preparation module divides the number of historical data into a backup historical data group and a practice historical data group respectively; The practice module selects a number of historical data from all the historical data belonging to the practice historical data group, uses the selected number of historical data to practice generating an analysis model, and the practice module determines whether the analysis model meets a preset first requirement. If it meets, continue to the next step. If it does not meet, repeat this step; The practice module determines a number of historical data that have not been selected in the practice historical data group, uses the determined number of historical data to practice generating an explanation model, and the analysis module inputs the additive information into the analysis model to obtain effect information, and inputs the additive information into the explanation model to obtain explanation information.

[0006] As a preferred technical solution of the present application, the practice module determines whether the analysis model meets a preset first requirement, including the following steps: The practice module sequentially inputs the historical additive information in all the historical data in the backup historical data group into the analysis model to respectively obtain the output effect information; For each output effect information, the practice module calculates the difference between the output effect information and the corresponding historical effect information, and the practice module calculates the average value of all the differences to obtain a difference mean value, and determines whether the difference mean value is less than a preset first threshold.

[0007] As a preferred technical solution of the present application, the practice module uses the determined number of historical data to practice generating an explanation model, including the following steps: The practice module sequentially inputs the historical additive information in the determined number of historical data into the analysis model to respectively obtain the output effect information; For each output effect information, the practice module calculates the difference between the output effect information and the corresponding historical effect information, determines whether the difference is less than the first threshold. If it is, replaces the historical effect information in the corresponding historical data with historical explanation information with high credibility. If it is not, replaces the historical effect information in the corresponding historical data with historical explanation information with low credibility, and the practice module uses the determined number of historical data to practice generating an explanation model.

[0008] As a preferred technical solution of the present application, after the practice module uses the determined number of historical data to practice generating an explanation model, the following steps are further included: For each piece of historical data in the backup historical data group, the practice module inputs the historical additive information in the historical data into the description model to obtain the output description information, inputs the historical additive information in the historical data into the analysis model to obtain the output effect information, and calculates the difference between the output effect information and the corresponding historical effect information; Regarding each piece of historical data in the backup historical data group, the practice module determines whether the description information corresponding to the historical data matches the difference corresponding to the historical data. If so, the count value is incremented by one; if not, the count value remains unchanged; The practice module calculates the ratio of the count value to the total number of all historical data in the backup historical data group, determines whether the ratio is greater than a preset second threshold. If not, the preset second threshold is changed, and the determined several pieces of historical data are used again to practice generating the description model.

[0009] As a preferred technical solution of the present application, the practice module uses the selected several pieces of historical data to practice generating an analysis model, including the following steps: The practice module obtains several pieces of practice data, the practice data includes historical additive information, and the practice module uses the selected several pieces of historical data to perform practice processing on several pre-stored backup models; The practice module divides the several backup models into several categories, selects one category as the core category from the several categories, and the practice module selects several pieces of practice data from all the practice data and performs supplementary processing on the selected several pieces of practice data; The practice module uses the selected several pieces of historical data and the selected several pieces of practice data to perform practice processing on all the backup models in the core category, deletes all the backup models in other categories, determines whether the stop condition is met. If so, all steps are ended; if not, it jumps to the previous step.

[0010] As a preferred technical solution of the present application, the practice module divides the several backup models into several categories, including the following steps: For each backup model, the practice module inputs the historical additive information in each piece of practice data that has not been selected in all the practice data into the backup model to respectively obtain the output effect information, and generates representative data using all the output effect information; Based on the representative data corresponding to each backup model, the practice module divides all the backup models into several categories respectively.

[0011] As a preferred technical solution of the present application, the practice module selects one category as the core category from the several categories, including the following steps: For each category, the practice module calculates the average of the output accuracies of all the fallback models in the category to obtain the average output accuracy; The practice module selects, from all the categories, the category with the maximum average output accuracy as the core category.

[0012] This application also provides a data analysis system for additive screening, including the following modules: A preparation module, configured to obtain a plurality of historical data composed of historical additive information and historical effect information, and divide the plurality of historical data into a fallback historical data group and a practice historical data group respectively; A practice module, configured to select a plurality of historical data from all the historical data belonging to the practice historical data group, use the selected plurality of historical data to practice generating an analysis model, and determine whether the analysis model meets a preset first requirement. If not, repeat this process. If so, determine a plurality of historical data in the practice historical data group that have not been selected, and use the determined plurality of historical data to practice generating an explanation model; An analysis module, configured to input additive information into the analysis model to obtain effect information, and input additive information into the explanation model to obtain explanation information.

[0013] Compared with the prior art, the beneficial effects of this application are at least as follows: In the technical solution provided by this application, first, the preparation module obtains a plurality of historical data composed of historical additive information and historical effect information, and the preparation module divides the plurality of historical data into a fallback historical data group and a practice historical data group respectively; second, the practice module selects a plurality of historical data from all the historical data belonging to the practice historical data group, uses the selected plurality of historical data to practice generating an analysis model, and the practice module determines whether the analysis model meets a preset first requirement. If it meets, continue to the next step. If not, repeat this step; finally, the practice module determines a plurality of historical data in the practice historical data group that have not been selected, uses the determined plurality of historical data to practice generating an explanation model, and the analysis module inputs additive information into the analysis model to obtain effect information, and inputs additive information into the explanation model to obtain explanation information. Through this application, not only can highly accurate effect information corresponding to the additive information be obtained, but also explanation information corresponding to the effect information can be obtained, making it more convenient to use the effect information. Description of the Drawings

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0015] Figure 1 It is a flowchart of the data analysis method for additive screening in the embodiments of the present application; Figure 2 It is a flowchart of the method for generating an explanatory model in the embodiments of the present application; Figure 3 It is a flowchart of the method for generating an analysis model in the embodiments of the present application; Figure 4 It is a schematic diagram of the data analysis system for additive screening in the embodiments of the present application. Detailed implementation manners

[0016] The embodiments of the present application provide a data analysis method and system for additive screening. The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and above-mentioned drawings of the present application are used to distinguish similar objects and do not have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0017] For ease of understanding, the following describes the specific process of the embodiments of the present application. Please refer to Figure 1 , a data analysis method for additive screening in the embodiments of the present application includes the following main steps: Step 1: The preparation module obtains a number of historical data composed of historical additive information and historical effect information, and the preparation module divides the number of historical data into a backup historical data group and a practice historical data group respectively; Step 2: The practice module selects a number of historical data from all the historical data belonging to the practice historical data group, uses the selected number of historical data to practice generating an analysis model, and the practice module determines whether the analysis model meets a preset first requirement. If it meets, continue to the next step. If it does not meet, repeat this step; Step 3: The practice module determines several historical data that have not been selected in the practice historical data group, uses the determined several historical data to practice and generate an explanation model, and the analysis module inputs the additive information into the analysis model to obtain effect information, and inputs the additive information into the explanation model to obtain explanation information.

[0018] Specifically, in order to use machine learning to study the effects of adding different additives to frozen semen on the freezing effect of pig sperm, thereby saving a large amount of experimental time, Steps 1 to 3 are proposed. In Step 1, the preparation module obtains several historical data, which are obtained based on actual experimental data and consist of historical additive information and historical effect information. For the sake of understanding, for example, the historical additive information includes the additive name, additive concentration, and addition amount, and the historical effect information includes the vitality value of the sperm after thawing. The preparation module divides the several historical data into a backup historical data group and a practice historical data group respectively. This embodiment does not specifically limit the division method. Usually, the number of historical data in the practice historical data group is more than the number of historical data in the backup historical data group. In Step 2, the practice module selects several historical data from all the historical data belonging to the practice historical data group. The first several historical data can be randomly selected. The selected several historical data are used to practice and generate an analysis model. Subsequently, the practice module determines whether the analysis model meets a preset first requirement. If it meets, Step 3 is continued. If it does not meet, Step 2 is repeated. It should be noted that when repeating Step 2, the several historical data selected by the practice module from all the historical data belonging to the practice historical data group include all the historical data selected last time, as well as multiple historical data selected this time. This time, any historical data that has not been selected can be selected. In Step 3, the practice module determines several historical data that have not been selected in the practice historical data group, uses the determined several historical data to practice and generate an explanation model. Then, the analysis module inputs the additive information into the analysis model to obtain the predicted effect information, and inputs the additive information into the explanation model to obtain the predicted explanation information. The analysis model and the explanation model are machine learning models in the prior art, and the analysis model and the explanation model will be described in detail below.

[0019] Through the above method, when using machine learning to study the effects of adding different additives to frozen semen on the freezing effect of pig sperm, not only can the predicted effect information with high accuracy be obtained based on the additive information, but also the explanation information about the effect information can be obtained.

[0020] Furthermore, the practice module determines whether the analysis model meets a preset first requirement, including the following steps: Step 1: The training module sequentially inputs the historical additive information in all historical data in the backup historical data group into the analysis model to respectively obtain the output effect information. Step 2: For each piece of output effect information, the training module calculates the difference between the output effect information and the corresponding historical effect information, and the training module calculates the average value of all differences to obtain the difference mean value, and determines whether the difference mean value is less than a preset first threshold.

[0021] Specifically, it is introduced how the training module determines whether the analysis model meets the preset first requirement. In Step 1, the training module sequentially inputs the historical additive information in all historical data in the backup historical data group into the analysis model, aiming to respectively obtain the output effect information. As can be seen from the above, the output effect information can be the estimated vitality value of thawed sperm. In Step 2, for each piece of output effect information, the training module calculates the difference between the output effect information and the corresponding historical effect information. It should be noted that the difference always takes a value greater than 0, and the corresponding historical effect information refers to the historical effect information in the historical data corresponding to the output effect information. Subsequently, the training module calculates the average value of all differences to obtain the difference mean value, and determines whether the difference mean value is less than a preset first threshold. The first threshold is set according to the actual application scenario.

[0022] Furthermore, the training module uses a determined number of historical data to train and generate an explanation model, including the following steps: Step 1: The training module sequentially inputs the historical additive information in the determined number of historical data into the analysis model to respectively obtain the output effect information. Step 2: Regarding each piece of output effect information, the training module calculates the difference between the output effect information and the corresponding historical effect information, determines whether the difference is less than the first threshold. In the case of yes, the historical effect information in the corresponding historical data is replaced with historical explanation information with high credibility, and in the case of no, the historical effect information in the corresponding historical data is replaced with historical explanation information with low credibility, and the training module uses the determined number of historical data to train and generate an explanation model.

[0023] Specifically, please refer to Figure 2, introduce the process of the practice module using a certain number of historical data to practice generating an explanatory model. In step 1, the practice module sequentially inputs the historical additive information in the certain number of historical data into the analysis model to obtain the output effect information respectively. As can be seen from the above, the output effect information can be the predicted vitality value of the thawed sperm. In step 2, for each output effect information, the practice module calculates the difference between the output effect information and the corresponding historical effect information. It should be noted that the difference always takes a value greater than 0. The corresponding historical effect information refers to the historical effect information in the historical data corresponding to the output effect information. Determine whether the difference is less than the first threshold. If it is less, it means that the output effect information is relatively accurate, and the historical effect information in the corresponding historical data is replaced with historical explanatory information with high credibility. If it is greater than or equal to, it means that the output effect information is not accurate enough, and the historical effect information in the corresponding historical data is replaced with historical explanatory information with low credibility. It should be noted that the corresponding historical data refers to the historical data corresponding to the output effect information. Then the practice module uses the certain number of historical data to practice generating an explanatory model. The explanatory model can be an SVM model. This embodiment does not specifically limit the type of the explanatory model.

[0024] Further, after the practice module uses the certain number of historical data to practice generating an explanatory model, the following steps are also included: Step 1: For each historical data in the backup historical data group, the practice module inputs the historical additive information in the historical data into the explanatory model to obtain the output explanatory information, inputs the historical additive information in the historical data into the analysis model to obtain the output effect information, and calculates the difference between the output effect information and the corresponding historical effect information; Step 2: Regarding each historical data in the backup historical data group, the practice module determines whether the explanatory information corresponding to the historical data matches the difference corresponding to the historical data. If so, the count value is incremented by one. If not, the count value remains unchanged; Step 3: The practice module calculates the ratio of the count value to the total number of all historical data in the backup historical data group, determines whether the ratio is greater than the preset second threshold. If not, the preset second threshold is changed, and the certain number of historical data is used again to practice generating an explanatory model.

[0025] Specifically, after introducing the process of using a certain number of historical data in the practice module to practice generating an explanatory model, and before inputting the additive information into the analysis model and the explanatory model in the analysis module, the following additional processing is also carried out by the practice module. In step 1, for each historical data in the backup historical data group, the practice module inputs the historical additive information in the historical data into the explanatory model to obtain the output explanatory information. As known from the above, the content of the output explanatory information can be either high credibility or low credibility. Then, the historical additive information in the historical data is input into the analysis model to obtain the output effectiveness information. As known from the above, the output effectiveness information can be, for example, the estimated motility value of the thawed sperm. Calculate the difference between the output effectiveness information and the corresponding historical effectiveness information. It should be noted that the difference always takes a value greater than 0. The corresponding historical effectiveness information refers to the historical effectiveness information in the historical data corresponding to the output effectiveness information. In step 2, for each historical data in the backup historical data group, the practice module determines whether the explanatory information corresponding to the historical data matches the difference corresponding to the historical data. Matching means that when the content of the explanatory information corresponding to the historical data is of high credibility, the difference corresponding to the historical data is less than the first threshold, and when the content of the explanatory information corresponding to the historical data is of low credibility, the difference corresponding to the historical data is greater than or equal to the first threshold. If the two match, increment the count value by one. It should be noted that the initial value of the count value is 0. If they do not match, keep the count value unchanged. In step 3, the practice module calculates the ratio of the count value to the total number of all historical data in the backup historical data group, and determines whether the ratio is greater than the preset second threshold. The initial value of the second threshold is set according to the actual application scenario. If it is less than or equal to, change the preset second threshold. The specific method can be to decrease the preset second threshold, and then use the certain number of historical data again to practice generating the explanatory model. It should be noted that the certain number of historical data actually refers to the certain number of historical data in which the historical effectiveness information has been replaced with historical explanatory information.

[0026] Furthermore, the practice module uses a selected number of historical data to practice generating an analysis model, including the following steps: Step 1: The practice module obtains a number of practice data, where the practice data includes historical additive information, and the practice module uses the selected number of historical data to perform practice processing on a number of pre-stored backup models; Step 2: The practice module divides the number of backup models into several categories, selects one category as the core category from the several categories, and the practice module selects a number of practice data from all the practice data and performs supplementary processing on the selected number of practice data; Step 3: The practice module uses a selected number of historical data and a selected number of practice data to perform practice processing on all candidate models in the core category, deletes all candidate models in other categories, determines whether the stop condition is met, and if so, ends all steps; if not, jumps to the previous step.

[0027] Specifically, please refer to Figure 3 for the process of the practice module using a selected number of historical data to practice and generate an analysis model. Considering that obtaining a large amount of historical data requires a large amount of time cost, and preparing a large amount of historical data is required to obtain an analysis model with high output accuracy, to solve this technical problem, in Step 1, the practice module obtains a selected number of practice data. It should be noted that the practice data only includes historical additive information and does not include historical effect information. The practice module uses a selected number of historical data to perform practice processing on a number of pre-stored candidate models. The candidate models can be neural network models, and the specific type of the candidate models is not limited in this embodiment. In Step 2, the practice module divides a number of candidate models into a number of categories, selects one category as the core category from the number of categories, and the practice module selects a number of practice data from all the practice data and performs supplementary processing on the selected number of practice data. The selected number of practice data refers to a number of practice data that may improve the output accuracy of the candidate models in the core category. The supplementary processing refers to setting corresponding historical effect information for the historical additive information in the practice data, which can be specifically set manually. In Step 3, the practice module uses a selected number of historical data and the selected number of practice data that have been supplemented to perform practice processing on all candidate models in the core category, and at the same time deletes all candidate models in other categories. Subsequently, the practice module determines whether the stop condition is met. The stop condition can be that the number of candidate models in the core category is 1. If so, the candidate model in the core category is used as the analysis model and all steps are ended. If not, it jumps to Step 2. It should be noted that when the practice module selects a number of practice data from all the practice data again, if the selected practice data has been supplemented, there is no need to supplement it again.

[0028] Further, the practice module divides a number of candidate models into a number of categories, including the following steps: Step 1: For each candidate model, the practice module inputs the historical additive information in each unselected practice data in all the practice data into the candidate model to obtain the output effect information respectively, and generates representative data using all the output effect information; Step 2: Based on the representative data corresponding to each candidate model, the practice module divides all the candidate models into a number of categories respectively.

[0029] Specifically, the process of dividing a number of backup models into several categories for the practice module will be introduced. In the first step, for each backup model, the practice module inputs the historical additive information in each piece of practice data that has not been selected from all the practice data into the backup model, with the aim of obtaining the output effect information respectively. As can be seen from the above, the output effect information can be the predicted motility value of the thawed sperm. Subsequently, representative data is generated using all the output effect information, and the representative data can be represented in the form of a vector. In the second step, based on the representative data corresponding to each backup model, the practice module divides all the backup models into several categories respectively. For the sake of easy understanding, for example, the existing clustering algorithm in the prior art can be used to divide all the backup models into several categories respectively.

[0030] Furthermore, the practice module selects one category from several categories as the core category, including the following steps: Step 1: For each category, the practice module calculates the average value of the output accuracy rates of all the backup models in the category to obtain the average output accuracy rate; Step 2: The practice module selects the category with the maximum corresponding average output accuracy rate among all the categories as the core category.

[0031] Specifically, the process of the practice module selecting one category from several categories as the core category will be introduced. In the first step, for each category, the practice module calculates the average value of the output accuracy rates of all the backup models in the category, with the aim of obtaining the average output accuracy rate. The process of calculating the output accuracy rate of each backup model in the category includes: setting the test data set of the backup model, calculating the average value of the output deviations of the backup model on the test data set, that is, the average output deviation. The output deviation can take a value greater than 0, and the reciprocal of the average output deviation is used as the output accuracy rate, which can be calculated through the existing technology and will not be elaborated here. In the second step, the practice module selects the category with the maximum corresponding average output accuracy rate among all the categories as the core category.

[0032] According to another aspect of the embodiments of the present application, as shown in Figure 4 the present application also provides a data analysis system for additive screening, including a preparation module, a practice module, and an analysis module, to implement a data analysis method for additive screening described above.

[0033] Among them, the functions of each module are as follows: The preparation module is used to obtain a number of historical data composed of historical additive information and historical effect information, and divide the number of historical data into a backup historical data group and a practice historical data group respectively; An exercise module, which is used to select several historical data from all the historical data belonging to the exercise historical data group, use the selected several historical data to exercise and generate an analysis model, and determine whether the analysis model meets a preset first requirement. If it does not meet the requirement, repeat this process. If it meets the requirement, determine several historical data in the exercise historical data group that have not been selected, and use the determined several historical data to exercise and generate an explanation model; An analysis module, which is used to input additive information into the analysis model to obtain effect information, and input additive information into the explanation model to obtain explanation information.

[0034] In summary, in the technical solution provided in this application, first, a preparation module obtains several historical data composed of historical additive information and historical effect information, and the preparation module divides the several historical data into a backup historical data group and an exercise historical data group respectively; second, the exercise module selects several historical data from all the historical data belonging to the exercise historical data group, uses the selected several historical data to exercise and generate an analysis model, and the exercise module determines whether the analysis model meets a preset first requirement. If it meets the requirement, proceed to the next step. If it does not meet the requirement, repeat this step; finally, the exercise module determines several historical data in the exercise historical data group that have not been selected, uses the determined several historical data to exercise and generate an explanation model, and the analysis module inputs additive information into the analysis model to obtain effect information, and inputs additive information into the explanation model to obtain explanation information. Through this application, not only can highly accurate effect information corresponding to the additive information be obtained, but also explanation information corresponding to the effect information can be obtained, and the effect information can be used more conveniently.

[0035] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described system, system, and unit can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0036] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0037] As described above, the above embodiments are only used to illustrate the technical solutions of this application, rather than to limit it; although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of various embodiments of this application.

Claims

1. A data analysis method for additive screening, characterized in that, The method includes the following steps: The preparation module obtains a number of historical data composed of historical additive information and historical effect information, and the preparation module divides the number of historical data into a backup historical data group and a practice historical data group respectively; The practice module selects a number of historical data from all the historical data belonging to the practice historical data group, uses the selected number of historical data to practice generating an analysis model, and the practice module determines whether the analysis model meets a preset first requirement. If it meets, continue to the next step; if it does not meet, repeat this step; The practice module determines a number of historical data that have not been selected in the practice historical data group, uses the determined number of historical data to practice generating an explanation model, and the analysis module inputs the additive information into the analysis model to obtain effect information, and inputs the additive information into the explanation model to obtain explanation information.

2. The method according to claim 1, characterized in that The practice module determines whether the analysis model meets a preset first requirement, including the following steps: The practice module sequentially inputs the historical additive information in all the historical data in the backup historical data group into the analysis model to respectively obtain the output effect information; For each output effect information, the practice module calculates the difference between the output effect information and the corresponding historical effect information, and the practice module calculates the average value of all the differences to obtain a difference mean value, and determines whether the difference mean value is less than a preset first threshold.

3. The method according to claim 2, wherein The practice module uses the determined number of historical data to practice generating an explanation model, including the following steps: The practice module sequentially inputs the historical additive information in the determined number of historical data into the analysis model to respectively obtain the output effect information; Regarding each output effect information, the practice module calculates the difference between the output effect information and the corresponding historical effect information, determines whether the difference is less than the first threshold. If it is, replaces the historical effect information in the corresponding historical data with historical explanation information with high credibility; if it is not, replaces the historical effect information in the corresponding historical data with historical explanation information with low credibility, and the practice module uses the determined number of historical data to practice generating an explanation model.

4. The method according to claim 3, characterized in that, After the practice module uses the determined number of historical data to practice generating an explanation model, the following steps are further included: For each historical data in the backup historical data group, the practice module inputs the historical additive information in the historical data into the explanation model to obtain the output explanation information, inputs the historical additive information in the historical data into the analysis model to obtain the output effect information, and calculates the difference between the output effect information and the corresponding historical effect information; For each historical data in the backup historical data group, the practice module determines whether the explanation information corresponding to the historical data matches the difference corresponding to the historical data. If it does, increments the count value by one; if it does not, keeps the count value unchanged; The practice module calculates the ratio of the count value to the total number of all the historical data in the backup historical data group, determines whether the ratio is greater than a preset second threshold. If it is not, changes the preset second threshold, and uses the determined number of historical data to practice generating an explanation model again.

5. The method according to claim 1, wherein The practice module uses a selected number of historical data to practice generating an analysis model, including the following steps: The practice module obtains a number of practice data, the practice data includes historical additive information, and the practice module uses a selected number of historical data to perform practice processing on a number of pre-stored backup models; The practice module divides a number of backup models into a number of categories, selects one category from the number of categories as the core category, and the practice module selects a number of practice data from all the practice data and performs supplementary processing on the selected number of practice data; The practice module uses a selected number of historical data and a selected number of practice data to perform practice processing on all the backup models in the core category, deletes all the backup models in other categories, determines whether the stop condition is met, and if it is met, ends all the steps, and if it is not met, jumps to the previous step.

6. The method according to claim 5, wherein The practice module divides a number of backup models into a number of categories, including the following steps: For each backup model, the practice module inputs the historical additive information in each unselected practice data in all the practice data into the backup model to respectively obtain the output effect information, and uses all the output effect information to generate representative data; Based on the representative data corresponding to each backup model, the practice module divides all the backup models into a number of categories respectively.

7. The method according to claim 6, wherein The practice module selects one category from the number of categories as the core category, including the following steps: For each category, the practice module calculates the average value of the output accuracy rates of all the backup models in the category to obtain the average output accuracy rate; The practice module selects the category with the largest corresponding average output accuracy rate among all the categories as the core category.

8. A data analysis system for additive screening, which is used to implement the method described in any one of claims 1 to 7, characterized in that, Including the following modules: A preparation module for obtaining a number of historical data composed of historical additive information and historical effect information, and respectively dividing the number of historical data into a backup historical data group and a practice historical data group; A practice module for selecting a number of historical data from all the historical data belonging to the practice historical data group, using the selected number of historical data to practice generating an analysis model, and determining whether the analysis model meets a preset first requirement. If it does not meet, repeat this process. If it meets, determine a number of historical data in the practice historical data group that have not been selected, and use the determined number of historical data to practice generating an explanation model; An analysis module for inputting additive information into the analysis model to obtain effect information, and inputting additive information into the explanation model to obtain explanation information.

Images