Method and medium for constructing a prediction model of user enrollment year based on big data analysis

Abstract

A method and medium for constructing a user's application year prediction model based on big data analysis includes: S1): data collection; S2): processing unstructured text data; S3): generating a user year information unit set; S4): data cleaning, cleaning out user year information; S5): generating user year information inference details; S6): calculating user year information unit weight score; S7): generating information unit time weight score; S8): introducing the most recent active priority strategy to weaken the influence of the user's historical year information score; S9): calculating the total score of each user's application year to realize the construction of the user's application year prediction model; S10): verifying the user's application year prediction model. The present invention calculates the user's application prediction year in a scientific and reasonable way, improves the accuracy of the user's application year prediction, and facilitates targeted services according to different user situations.

Description

Technical Field

[0001] The present invention belongs to the field of data processing technology, and in particular relates to a method and medium for constructing a user's college entrance examination application year prediction model based on big data analysis. Background Art

[0002] As the idea of ​​lifelong learning has become more popular, many people are constantly improving themselves. This has led to the birth of platforms that provide learning resources for users. How can the platform accurately understand the year in which users apply for further studies? How can the business end provide better services to users based on their different stages? Taking the postgraduate entrance examination as an example, the number of applicants has continued to grow in recent years, but it fell for the first time in 2024, indicating that the postgraduate entrance examination market has entered a relatively stable stage.

[0003] The current postgraduate entrance examination industry is restricted by market demand, resource input and business model. Existing prediction methods are mostly based on simple statistics or questionnaires. After the statistics or questionnaires, the data obtained needs to be sorted, which requires a large workload and lacks systematicness and accuracy. Summary of the invention

[0004] Purpose of the invention: In order to overcome the above shortcomings, the purpose of the present invention is to provide a method for constructing and applying a user's entrance examination application year prediction model based on big data analysis. By predicting the user's application year based on big data analysis, an accurate prediction of the user's application year is achieved, which effectively improves the efficiency and accuracy of institutions or platforms in course development, refined operations and personalized delivery.

[0005] Technical solution: In order to achieve the above-mentioned purpose, the present invention provides a method for constructing a user's college entrance examination application year prediction model based on big data analysis, comprising the following steps:

[0006] S1): Data collection, i.e., collecting all behavior data of users on the terminal platform containing year information, wherein the behavior data includes structured text data and unstructured text data;

[0007] S2): Processing unstructured text data, that is, based on the idea of ​​sentence segmentation, each piece of collected unstructured text data is segmented based on punctuation marks to ensure that continuous unstructured text data is decomposed into smaller, easier-to-process units, namely sentences or short sentences;

[0008] S3): firstly process the unstructured text data processed in S2) into structured data by regularization and single-line-to-multiple-line conversion method according to the structured text data structure in S1), then integrate the formed structured data with the structured text data collected in S1), and record a piece of year information data of a user as a user year information unit to generate a user year information unit set;

[0009] S4): data cleaning, i.e., first setting the key dimensions for cleaning, performing data normalization on all user year information units in the user year information unit set, extracting all year expressions containing year information in the user year information units, and cleaning out the year information;

[0010] S5): Based on the year expression obtained by data regularization in S4), the user year information is inferred according to the specific scenario under the key dimension of cleaning, and the user year information inference details are generated; if the year information cannot be cleaned out through the key dimension of cleaning, the backup year is used to infer the application year;

[0011] S6): Based on the inference details of the user year information obtained in S5), determine the information source and key dimension weighting according to the reliability of the information source data, and calculate the user year information unit weight score;

[0012] S7): Based on the idea that the farther the time is, the lower the credibility of the information is, the user year information inference details in S5) are calculated based on the number of days from the information occurrence time, and the time weight score of the user year information unit is generated using the sigmoid function formula;

[0013] S8): Introduce the most recently active priority strategy, taking the X-month-XX-day of the user's most recent active year as the dividing node. If the user has new active information for the year of application after X-month-XX-day of the last year of activity (i.e. after the re-examination line of that year is released), it is considered that the user's historical information of the past years will not be scored. At this time, the time weight of the past years information is assigned to 0, that is, only the year information of the user who did not take the preliminary examination is retained, so as to weaken the impact of the user's historical year information on the score;

[0014] S9): Combined with the results in S8), taking the user as a unit, the inferred application year is grouped according to the inferred details of the user's year information, and the total score of the user's application year is calculated based on the product and sum of the primary information source weight, secondary information source weight, key dimension weight and time weight of each piece of information, and the year with the highest score is regarded as the user's predicted application year, thereby realizing the construction of the user's application year prediction model;

[0015] S10): Verify the user application year prediction model, use the minimum sampling method to determine the minimum sample number, and perform sampling evaluation on the user application year prediction model constructed in S9) to ensure the accuracy of the user application year prediction model;

[0016] S11): Apply the user application year prediction model to the CDP population attribute analysis and user group-related special analysis of educational institutions to improve operational efficiency.

[0017] The present invention discloses a user's entrance examination application year prediction model based on big data analysis, which analyzes and organizes the data, performs data regularization and cleaning on the user's year information unit to obtain the year information, and then infers the user's year information inference details based on the year. The product of the information source, key dimension and time weight is used as the information unit score, and the total score of each user's application year is calculated. The year with the highest score is the application prediction year. The user's application prediction year is calculated in a scientific and reasonable manner, which effectively improves the accuracy of the user's application year prediction, makes it convenient for staff to provide targeted services according to the different situations of each user, and also facilitates institutions and platforms to develop courses and carry out precise operations.

[0018] The user's application year prediction model for further studies based on big data analysis of the present invention comprises five key dimensions set in S4), specifically: application year time, application year text, grade at the time, on-the-job application and multiple applications, and a specific inference logic is set in each specific dimension scenario to more accurately extract the year information prompted by the information unit;

[0019] If the year information cannot be inferred through the five key dimensions, the fallback year will be used to infer the application year.

[0020] The method for constructing and applying a user enrollment year prediction model based on big data analysis described in the present invention, wherein the sigmoid function formula is used in S7) to calculate the time weight and generate the time weight score of the user year information unit, and the specific process is as follows:

[0021] The initial time weight formula is as follows:

[0022]

[0023] Among them, x i =The number of days since the information occurred

[0024] For each piece of information in the information unit, the upper value can be obtained through the time weight formula. The formula normalizes the time weight from 0 to 1 to obtain the time weight Time i .

[0025] The method for constructing and applying a user's application year prediction model based on big data analysis of the present invention, wherein S9) calculates the total score of the user for each application year according to the product and accumulation of the primary information source weight, the secondary information source weight, the dimension weight and the time weight of each piece of information, and regards the year with the highest score as the user's predicted application year, and the specific process is as follows:

[0026] First, calculate the total score of information in the year of application:

[0027]

[0028] Among them, Source1 i Indicates the weight of the first-level information source, Source2 i Represents the weight of the secondary information source, Dim i Represents the weight of the key dimension of information extraction, Time i represents the time weight; n j Indicates the cumulative number of messages in postgraduate entrance examination year j; Source1 i 、Source2 i 、Dim i Match the information source weight, information source type weight, and dimension type weight set in the user's year information unit. If a new information source appears, set the default value to 0.5. If the business end believes that this value is unreasonable with the user's actual situation, manually adjust it to a reasonable weight.

[0029] The year with the largest final score among the total scores of information under the application year is the predicted application year, so:

[0030] Predicted application year = max (cumulative score of application year j) corresponding to the year.

[0031] The method for constructing and applying a user enrollment year prediction model based on big data analysis described in the present invention, the formula for determining the minimum number of samples using the minimum sampling method is as follows:

[0032]

[0033] Where n is the sample size, N is the population size, Z is the Z value of the standard normal distribution (the Z value corresponding to the 95% confidence level is 1.96), p is the proportion of a certain feature in the population (assuming the proportion is 0.5), and E is the allowable error range (0.05).

[0034] Through calculation, it was found that as the overall size N increased, the sample size n gradually stabilized and remained at around 400. Finally, 400 samples were selected for accuracy evaluation.

[0035] In the method for constructing and applying a user's application year prediction model based on big data analysis of the present invention, in the process of inferring the user's year information in S5), when the key dimension is the grade at that time, the inference process of the application year is as follows:

[0036] Extract user information. If the user's information is filled in before / after July, add or subtract the difference in years to the year that meets the application conditions according to the extracted grade information;

[0037] If the user's information is filled in before December 25, and the user's extracted information is in-service / graduated / working / master's degree / graduate student, the predicted application year = the information filled in year + 1;

[0038] If the user fills in the information after December 25, and the user extracts information as being employed / graduated / working / master / graduate student, then the predicted application year = the year the information is filled in + 2.

[0039] In the method for constructing and applying a user's application year prediction model based on big data analysis of the present invention, in the process of inferring the user's year information in S5), when the key dimension is the text of the application year, the inference process of the application year is as follows:

[0040] Extract the information filled in by the user. If the words "take this year's exam / preliminary exam / plan / continue" are extracted, the predicted application year = the year filled in + 1;

[0041] If the words "Next year's exam / preliminary exam / planning / continue" are extracted, the predicted year of application = the year of information filled in + 2;

[0042] If the words "exam next year / preliminary exam / planning / continue" are extracted, the predicted application year = the year filled in the information + 3;

[0043] If the test is taken at the end of the year, the predicted year of application = the year of information filled in + 1;

[0044] If the preparation for the preliminary examination starts in 20xx, then the predicted year of application = 20xx.

[0045] In the process of inferring the user's year information in S5), when the key dimension is the number of the application year, the inference process of the application year is as follows:

[0046] If the words "20xx exam / session / year postgraduate entrance examination / year graduation / graduation" are extracted, the predicted application year = 20xx;

[0047] If the words "20xx level / xx level" are extracted, the predicted application year = 20xx+4;

[0048] If the words "xx exam / session / year exam / re-examination / follow-up / graduation / year graduation" are extracted, the predicted application year = 20xx;

[0049] If the words "passed the exam in xx year" are extracted, the predicted year of application = the year of information collection + 1;

[0050] If the extracted information does not meet the above four conditions and the word 20xx is extracted, the predicted application year = 20xx.

[0051] The method for constructing and applying a user's further education application year prediction model based on big data analysis described in the present invention, in the process of inferring the user's year information in S5), when the key dimension is applying while working or applying multiple times, the predicted application year = the year the information is filled in + 1.

[0052] The method for constructing and applying a user's application year prediction model based on big data analysis of the present invention, in the process of inferring the user's year information in S5), if the five key dimensions cannot be inferred, a fallback year is also included, and if the registration time is before October, the predicted application year = the information filled year + 1;

[0053] If the registration time is after October, the predicted application year = the year filled in the information + 2.

[0054] A computer-readable storage medium described in the present invention stores a computer program thereon, and when the program is executed by a processor, the steps in the method for constructing a user's entrance examination application year prediction model based on big data analysis are implemented.

[0055] It can be seen from the above technical solution that the present invention has the following beneficial effects:

[0056] 1. The user's entrance examination application year prediction model based on big data analysis described in the present invention analyzes and organizes the data, and performs data regularization and cleaning on the user's year information unit to obtain the year information, and then infers the user's year information inference details based on the year, and uses the product of information source, key dimension, and time weight as the information unit score to calculate the total score of each user's application year. The year with the highest score is the application prediction year. The user's application prediction year is calculated in a scientific and reasonable manner, which effectively improves the accuracy of the user's application year prediction, facilitates the staff to provide targeted services according to the different situations of each user, and also facilitates institutions and platforms to develop courses and carry out precise operations.

[0057] 2. During the data processing process, the present invention performs text segmentation on unstructured text data based on punctuation marks through a regular extraction method of sentence segmentation, ensuring that continuous unstructured text data is decomposed into smaller, easier-to-process units, facilitating subsequent user information extraction and improving the accuracy of information extraction.

[0058] 3. The present invention sets key dimensions for cleaning, performs data regularization on all user year information units in a user year information unit set, extracts the year expression contained in the information unit, and cleans the year information; thus improving the accuracy of year information data extraction.

[0059] 4. The present invention introduces a recently active priority strategy, which only retains the year information of users who have not taken the preliminary examination, effectively weakening the influence of the user's historical year information score and further improving the accuracy of model construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a structural schematic diagram of the method for constructing and applying a user's college entrance examination application year prediction model based on big data analysis according to the present invention;

[0061] Figure 2 For the present invention Figure 2 A represents a schematic diagram of ordinary regular extraction. Figure 2 B is a schematic diagram of sentence segmentation regularization processing;

[0062] Figure 3 A flow chart for generating a user year information unit set for data integration in the present invention;

[0063] Figure 4 A schematic diagram for inferring the year expression contained in the extracted information unit in the present invention;

[0064] Figure 5 A schematic diagram of the process of calculating the weight score of the user year information unit in the present invention;

[0065] Figure 6 This is a screenshot of the key weight details of the source of the postgraduate entrance examination year prediction model in the present invention;

[0066] Figure 7 It is a schematic diagram of the key dimensions in the present invention;

[0067] Figure 8 This is a table showing the data source in the present invention. DETAILED DESCRIPTION

[0068] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments.

[0069] Example

[0070] like Figure 1 A method for constructing a user's college entrance examination application year prediction model based on big data analysis is shown, comprising the following steps:

[0071] S1): Data collection, i.e., collecting all behavior data of users on the terminal platform containing year information, wherein the behavior data includes structured text data and unstructured text data;

[0072] S2): Processing unstructured text data through regular extraction, that is, based on the idea of ​​sentence segmentation, each piece of collected unstructured text data is segmented based on punctuation marks to ensure that continuous unstructured text data is decomposed into smaller, easier-to-process units, namely sentences or short sentences;

[0073] S3): firstly process the unstructured text data processed in S2) into structured data by regularization and single-line-to-multiple-line conversion method according to the structured text data structure in S1), then integrate the formed structured data with the structured text data collected in S1), and record a piece of year information data of a user as a user year information unit to generate a user year information unit set;

[0074] S4): data cleaning, i.e., first setting the key dimensions for cleaning, performing data normalization on all user year information units in the user year information unit set, extracting all year expressions containing year information in the user year information units, and cleaning out the year information;

[0075] S5): Based on the year expression obtained by data regularization in S4), the user year information is inferred according to the specific scenario under the key dimension of cleaning, and the user year information inference details are generated; if the year information cannot be cleaned out through the key dimension of cleaning, the backup year is used to infer the application year;

[0076] S6): Based on the inference details of the user year information obtained in S5), determine the information source and key dimension weighting according to the reliability of the information source data, and calculate the user year information unit weight score;

[0077] S7): Based on the idea that the farther the time is, the lower the credibility of the information is, the user year information inference details in S5) are calculated based on the number of days from the information occurrence time, and the time weight score of the user year information unit is generated using the sigmoid function formula;

[0078] S8): Introduce the most recently active priority strategy, taking the X-month-XX-day of the user's most recent active year as the dividing node. If the user has new active information for the year of application after X-month-XX-day of the last year of activity (i.e. after the re-examination line of that year is released), it is considered that the user's historical information of the past years will not be scored. At this time, the time weight of the past years information is assigned to 0, that is, only the year information of the user who did not take the preliminary examination is retained, so as to weaken the impact of the user's historical year information on the score;

[0079] S9): Combined with the results in S8), taking the user as a unit, the inferred application year is grouped according to the inferred details of the user's year information, and the total score of the user's application year is calculated based on the product and sum of the primary information source weight, secondary information source weight, key dimension weight and time weight of each piece of information, and the year with the highest score is regarded as the user's predicted application year, thereby realizing the construction of the user's application year prediction model;

[0080] S10): The user application year prediction model is verified, and the minimum sample number is determined using the minimum sampling method. A sampling evaluation is performed on the user application year prediction model constructed in S9) to ensure the accuracy of the user application year prediction model.

[0081] In S2) of this embodiment, the unstructured text data is processed by regular extraction, that is, according to the idea of ​​sentence segmentation, each piece of collected unstructured text data is segmented based on punctuation marks to ensure that the continuous unstructured text data is decomposed into smaller, easy-to-process units, that is, sentences or short sentences; Figure 2 As shown in the figure, it takes the postgraduate entrance examination as an example to compare the ordinary regularization and sentence segmentation idea regularization. It can be seen that the sentence segmentation idea is more accurate in extraction.

[0082] It should be noted that in S5), according to the specific scenarios under the key dimensions, a key dimension actually represents some special scenarios. Figure 4 Year results can be extrapolated.

[0083] In the method for constructing and applying a user's application year prediction model based on big data analysis described in this embodiment, the key dimensions set in S4) are derived from the user year information unit, which includes five key dimensions, specifically: application year time, application year text, grade at the time, application while working, and multiple applications. It should be noted that the user year information unit only includes five key dimensions, and each user year information specifies one key dimension.

[0084] In S6) described in this embodiment, the information source and key dimension weightings are determined according to the reliability of the information source data. It should be noted that all collected information has a source, and different systems collect information in different ways. The information source has been obtained in the data collection stage, so the tracing is completed; it should be noted that the information source level is divided according to the level of the location of the information collection source, such as the first-level source APP; some information collection pages of the second-level source APP. This method is the preferred method in this case. It can be imagined that it can also choose other level division principles according to the actual application scenario.

[0085] Since the reliability of information sources is different, each source is initially artificially weighted according to the actual situation on the business side, and later adjustments are supported. The importance of year information obtained from different key dimensions is different. For example, the year 2026 extracted from the 2026 postgraduate entrance examination is 2026, and the year 2025 extracted from the sophomore is 2025. The former 2026 is the year data directly stated. Compared with it, the credibility of the year 2025 extracted from the sophomore is weaker. Therefore, key dimensions must also be weighted.

[0086] The calculation of the user year information unit weight score = Source1 i *Source2 i *Dim i

[0087] Among them, Source1 i Indicates the weight of the first-level information source, Source2 i Represents the weight of the secondary information source, Dim i Represents the weight of the key dimension of information extraction.

[0088] The method for constructing and applying a prediction model for the year of application for further studies based on big data analysis in this embodiment is as follows: S9) calculates the total score of each information unit on each application year of the user according to the product and accumulation of the primary information source weight, the secondary information source weight, the dimension weight and the time weight of each piece of information, and regards the year with the highest score as the predicted application year of the user. The specific process is as follows:

[0089] First, calculate the total score of information in the year of application:

[0090]

[0091] Among them, Source1 i Indicates the weight of the first-level information source, Source2 i Represents the weight of the secondary information source, Dim i Represents the weight of the key dimension of information extraction, Time i represents the time weight; n j represents the cumulative number of messages in the postgraduate entrance examination year j; if 10 messages are collected in 2025, then n j =10;Source1 i 、Source2 i 、Dim i Match the information source weight, information source type weight, and dimension type weight set in the user's year information unit. If a new information source appears, set the default value to 0.5. If the business end believes that this value is unreasonable with the user's actual situation, manually adjust it to a reasonable weight.

[0092] The year with the largest final score among the total scores of information under the application year is the predicted application year, so:

[0093] Predicted application year = max (cumulative score of application year j) corresponding to the year.

[0094] It should be noted that the default value of the weights set for information units within the system is uniformly set to 0.5.

[0095] like Figure 8 The weight sources shown here are mainly divided into 11 sources: APP personal information, H5 forms, order purchases, consultant follow-up, clue records, school confirmation forms, school reservation forms, user registration forms, follow-up tags, school recommendation mini-programs, and student files in the academic system. Among them, the user's school information can be obtained by analyzing the information in the school confirmation form and school reservation form.

[0096] The method for constructing and applying a user enrollment year prediction model based on big data analysis in this embodiment uses a sigmoid function formula to calculate time weight in S7) to generate a time weight score for the user year information unit. The specific process is as follows:

[0097] The time weight part refers to the commonly used function sigmoid function formula According to the business data, adjust the parameters and preliminarily set the time weight formula as follows:

[0098]

[0099] Among them, x i =The number of days since the information occurred

[0100] For each piece of information in the information unit, the upper value can be obtained through the time weight formula. The formula normalizes the time weight from 0 to 1 to obtain the time weight Time i It should be noted that x i =The number of days from the time the information occurred. If the time is less than one day, it will be calculated as one day.

[0101] The method for constructing and applying a user enrollment year prediction model based on big data analysis described in this embodiment, the formula for determining the minimum number of samples using the minimum sampling method is as follows: S10):

[0102]

[0103] Among them, n is the sample size, N is the size of the population, Z is the Z value of the standard normal distribution (the Z value corresponding to the 95% confidence level is 1.96), p is the proportion of a certain feature in the population (assuming the proportion is 0.5), and E is the allowable error range of ±0.05. It should be noted that the error range can be adjusted according to actual conditions.

[0104] Through calculation, it was found that as the overall size N increased, the sample size n gradually stabilized and remained at around 400. Finally, 400 samples were selected for accuracy evaluation.

[0105] Example 2

[0106] The method for constructing and applying a prediction model for the year of application for further studies based on big data analysis described in this embodiment is the same as that in Example 1. When the year information of the user is in different key dimensions, the inference process of predicting the year of application is as follows: In the method for constructing and applying a prediction model for the year of application for further studies based on big data analysis described in this embodiment, in the process of inferring the year information of the user, when the key dimension is the grade at that time, the inference process of the year of application is as follows: extract the user information, if the user's information is filled in before / after July, add or subtract the difference in years from the year that meets the application conditions according to the extracted grade information;

[0107] If the user's information is filled in before December 25, and the user's extracted information is in-service / graduated / working / master's degree / graduate student, the predicted application year = the information filled in year + 1;

[0108] If the user fills in the information after December 25, and the user extracts information as being employed / graduated / working / master / graduate student, then the predicted application year = the year the information is filled in + 2.

[0109] Since the first test of the postgraduate entrance examination is around December 25th every year, this time node is selected. Also, since users who do not participate in the first test of the current year are generally forced to participate in the first test of the next year, it is assumed here that users before December 25th are taking the postgraduate entrance examination of the current year, and those after December 25th are taking the postgraduate entrance examination of the next year.

[0110] In the method for constructing and applying a user's application year prediction model based on big data analysis described in this embodiment, in the process of inferring the user's year information in S5), when the key dimension is the text of the application year, the inference process of the application year is as follows:

[0111] Extract the information filled in by the user. If the words "take this year's exam / preliminary exam / plan / continue" are extracted, the predicted application year = the year filled in + 1;

[0112] If the words "Next year's exam / preliminary exam / planning / continue" are extracted, the predicted year of application = the year of information filled in + 2;

[0113] If the words "exam next year / preliminary exam / planning / continue" are extracted, the predicted application year = the year filled in the information + 3;

[0114] If the test is taken at the end of the year, the predicted year of application = the year of information filled in + 1;

[0115] If the preparation for the preliminary examination starts in 20xx, then the predicted year of application = 20xx.

[0116] In the process of inferring the user's year information in S5), when the key dimension is the number of the application year, the inference process of the application year is as follows:

[0117] If the words "20xx exam / session / year postgraduate entrance examination / year graduation / graduation" are extracted, the predicted application year = 20xx;

[0118] If the words "20xx level / xx level" are extracted, the predicted application year = 20xx+4;

[0119] If the words "xx exam / session / year exam / re-examination / follow-up / graduation / year graduation" are extracted, the predicted application year = 20xx;

[0120] If the words "passed the exam in xx year" are extracted, the predicted year of application = the year of information collection + 1;

[0121] If the extracted information does not meet the above four conditions and the word 20xx is extracted, the predicted application year = 20xx.

[0122] In the method for constructing and applying a prediction model for the year of application for further studies based on big data analysis described in this embodiment, in the process of inferring the year information of the user in S5), when the key dimension is applying for the examination while working or applying for the examination multiple times, the predicted year of application = the year of information filling + 1. The reason for this operation is affected by the particularity of the postgraduate entrance examination education industry. Generally speaking, the user groups of postgraduate entrance examination users who are working or have taken the postgraduate entrance examination multiple times belong to the group of people who can take the postgraduate entrance examination at any time. When the user clearly indicates that he is working or has graduated, rather than being a postgraduate student or a graduate student, the user has the qualifications and ability to take the postgraduate entrance examination in the year of information collection + 1, so the design marks the user with this year label to avoid information omission.

[0123] The method for constructing and applying a user's entrance examination application year prediction model based on big data analysis described in this embodiment, in the process of inferring the user's year information in S5), if the five key dimensions cannot be inferred, a backup year is also included. If the registration time is before October, the predicted application year = the year filled in the information + 1; if the registration time is after October, the predicted application year = the year filled in the information + 2.

[0124] It should be noted that the user's year information is cleaned from 11 information sources through data cleaning, and the key dimensions to which the cleaned user's year information belongs are viewed. The predicted postgraduate entrance examination year is inferred based on the inference process of predicting the application year under the above-mentioned key dimensions;

[0125] According to the reliability of information source data, determine the information source and key dimension weighting, and calculate the user year information unit weight score;

[0126] Then, the sigmoid function formula is used to calculate the time weight and generate the time weight score of the user's year information unit;

[0127] Finally, the total score of the user in each application year is calculated based on the product and sum of the primary information source weight, secondary information source weight, dimension weight and time weight of each piece of information, and the year with the highest score is regarded as the user's predicted application year.

[0128] Example 3

[0129] A computer-readable storage medium described in the present invention stores a computer program thereon, and when the program is executed by a processor, the steps in the method for constructing a user's entrance examination application year prediction model based on big data analysis are implemented.

[0130] Example 4

[0131] The present embodiment describes an application of a user further education application year prediction model based on big data analysis, and the model constructed by the user further education application year prediction model construction method based on big data analysis described in the present invention is applied to CDP population attribute analysis and user group-related special analysis of educational institutions to improve operational efficiency.

[0132] The above description is only a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements without departing from the principle of the present invention, and these improvements should also be regarded as within the protection scope of the present invention.

Claims

1. A method for constructing a user's college entrance examination application year prediction model based on big data analysis, characterized in that: include: Follow these steps: S1): Data collection, i.e., collecting all behavior data of users on the terminal platform containing year information, wherein the behavior data includes structured text data and unstructured text data; S2): Processing unstructured text data, that is, based on the idea of ​​sentence segmentation, each piece of collected unstructured text data is segmented based on punctuation marks to ensure that continuous unstructured text data is decomposed into smaller, easier-to-process units, namely, sentences or short sentences; S3): first, the unstructured text data processed in S2) is processed according to the structured text data structure in S1) by regularization and single-line conversion to form structured text data, then the formed structured text data and the structured text data collected in S1) are integrated, and a piece of year information data of a user is recorded as a user year information unit, and a user year information unit set is generated; S4): data cleaning, i.e., first setting the key dimensions for cleaning, performing data normalization on all user year information units in the user year information unit set, extracting all year expressions containing year information in the user year information units, and cleaning out the year information; S5): Based on the year expression obtained by data regularization in S4), the user year information is inferred according to the specific scenario under the key dimension of cleaning, and the user year information inference details are generated; if the year information cannot be cleaned out through the key dimension of cleaning, the backup year is used to infer the application year; S6): Based on the inference details of the user year information obtained in S5), determine the information source and key dimension weighting according to the reliability of the information source data, and calculate the user year information unit weight score; S7): Based on the idea that the farther the time is, the lower the credibility of the information is, the user year information inference details in S5) are calculated based on the number of days from the information occurrence time, and the time weight score of the user year information unit is generated using the sigmoid function formula; S8): Introduce the most recently active priority strategy, taking the X-month-XX-day of the user's most recent active year as the dividing node. If the user has new active information for the year of application after X-month-XX-day of the last active year, it is considered that the user's historical information for the past years will not be scored. At this time, the time weight of the past years information is assigned to 0, that is, only the year information of the user who did not take the preliminary examination is retained, so as to weaken the impact of the user's historical year information score; S9): Combined with the results in S8), taking the user as a unit, grouping the inferred application years in the user year information inference details, and calculating the total score of each user's application year, i.e., the user year information unit score, based on the product and accumulation of the primary information source weight, secondary information source weight, key dimension weight and time weight of each piece of information, and taking the year with the highest score as the user's predicted application year, thereby realizing the construction of the user's application year prediction model; S10): Verify the user application year prediction model, use the minimum sampling method to determine the minimum sample number, and perform sampling evaluation on the user application year prediction model constructed in S9) to ensure the accuracy of the user application year prediction model.

2. The method for constructing a user's college entrance examination application year prediction model based on big data analysis according to claim 1, characterized in that: The key dimensions set in S4) include five key dimensions, namely: application year time, application year text, grade at that time, application while working and multiple applications, and specific inference logic is set in each specific dimension scenario to facilitate more accurate extraction of the year information prompted by the information unit.

3. The method for constructing a user's college entrance examination application year prediction model based on big data analysis according to claim 2 is characterized by: In S7), the sigmoid function formula is used to calculate the time weight to generate the time weight score of the user year information unit. The specific process is as follows: The initial time weight formula is as follows: Among them, x i =The number of days since the information occurred For each piece of information in the information unit, the upper value can be obtained through the time weight formula. The formula normalizes the time weight from 0 to 1 to obtain the time weight Time i .

4. The method for constructing a user's college entrance examination application year prediction model based on big data analysis according to claim 1, characterized in that: S9) calculates the total score of the user in each application year according to the product and accumulation of the primary information source weight, secondary information source weight, dimension weight and time weight of each piece of information, and regards the year with the highest score as the user's predicted application year. The specific process is as follows: First, calculate the total score of information in the year of application: Among them, Source1 i Indicates the weight of the first-level information source, Source2 i Represents the weight of the secondary information source, Dim i Represents the weight of the key dimension of information extraction, Time i represents the time weight; n j Indicates the cumulative number of messages in postgraduate entrance examination year j; The year with the largest final score among the total scores of information under the application year is the predicted application year, so: Predicted application year = max (cumulative score of application year j) corresponding to the year.

5. The method for constructing a user's college entrance examination application year prediction model based on big data analysis according to claim 2 is characterized by: The formula for determining the minimum number of samples using the minimum sampling method in S10) is as follows: Where n is the sample size, N is the population size, Z is the Z value of the standard normal distribution, p is the proportion of a certain feature in the population, and E is the allowable error range. Through calculation, it was found that as the overall size N increased, the sample size n gradually stabilized and remained at around 400. Finally, 400 samples were selected for accuracy evaluation.

6. The method for constructing a user's college entrance examination application year prediction model based on big data analysis according to claim 2 is characterized by: In the process of inferring the user's year information in S5), when the key dimension is the grade at that time, the inference process of the application year is as follows: Extract user information. If the user's information is filled in before / after July, add or subtract the difference in years from the year that meets the application conditions according to the extracted user's grade information; If the user's information is filled in before December 25, and the user's extracted information is in-service / graduated / working / master's degree / graduate student, the predicted application year = the information filled in year + 1; If the information is filled in after December 25, and the user extracts information as being employed / graduated / working / master / graduate student, then the predicted application year = the year the information is filled in + 2.

7. The method for constructing a user's college entrance examination application year prediction model based on big data analysis according to claim 2 is characterized by: In the process of inferring the user year information in S5), when the key dimension is the text of the application year, the inference process of the application year is as follows: Extract the information filled in by the user. If the words "take this year's exam / preliminary exam / plan / continue" are extracted, Then the predicted application year = the year filled in + 1; If the words "Next year's exam / preliminary exam / planning / continue" are extracted, the predicted year of application = the year of information filled in + 2; If the words "exam next year / preliminary exam / planning / continue" are extracted, the predicted application year = the year filled in the information + 3; If the test is taken at the end of the year, the predicted year of application = the year of information filled in + 1; If the preparation for the preliminary examination starts in 20xx, then the predicted year of application = 20xx.

8. The method for constructing a user's college entrance examination application year prediction model based on big data analysis according to claim 2 is characterized by: In the process of inferring the user year information in S5), when the key dimension is the number of the application year, the inference process of the application year is as follows: If the words "20xx exam / session / year postgraduate entrance examination / year graduation / graduation" are extracted, the predicted application year = 20xx; If the words "20xx level / xx level" are extracted, the predicted application year = 20xx+4; If the words "xx exam / session / year exam / re-examination / follow-up / graduation / year graduation" are extracted, the predicted application year = 20xx; If the words "passed the exam in xx year" are extracted, the predicted year of application = the year of information collection + 1; If the extracted information does not meet the above four conditions and the word 20xx is extracted, the predicted application year = 20xx.

9. The method for constructing a user's college entrance examination application year prediction model based on big data analysis according to claim 2, characterized in that: In the process of inferring the user's year information in S5), when the key dimension is applying for the exam while working or applying for the exam multiple times, the predicted application year = the year the information is filled in + 1.

10. The method for constructing a user's college entrance examination application year prediction model based on big data analysis according to claim 2, characterized in that: In the process of inferring the user year information in S5), if the year information cannot be cleaned out through the key dimensions of cleaning, the application year is inferred using the backup year. The process of inferring the application year using the backup year is as follows: If the registration time is before October, the predicted registration year = the year filled in + 1; If the registration time is after October, the predicted application year = the year filled in the information + 2.

Images