Enrollment plan admission probability prediction method, system, device and storage medium

Abstract

This invention relates to the field of educational data processing technology, providing a method, system, device, and storage medium for predicting the probability of enrollment in school admission plans. The method includes the following steps: generating a normalized query rank for the year; filtering valid records from the comparison group, counting the number of records less than the target historical boundary and the number of records equal to the target historical boundary, and calculating the joint midpoint rank accordingly; generating a basic probability transition width based on the normalized query rank, adjusting the basic width using a midpoint rank-based hierarchical adjustment factor to obtain the plan probability transition width; inputting the normalized query rank into a bounded, monotonic, and continuous segmented probability kernel to generate the annual admission probability; performing weighted fusion after normalizing the timeliness weights on the valid years, and generating a confidence coefficient based on the annual completeness and within-group sample sufficiency. This invention achieves adaptive rank magnitude, stable handling of missing data and joint boundaries within groups, differentiated responses to different plans within the same group, and unified fusion of missing years, making it suitable for enrollment plan prediction.

Description

Technical Field

[0001] This invention belongs to the field of educational data processing technology, and in particular relates to methods, systems, devices and storage media for predicting the probability of enrollment in school admission plans. Background Technology

[0002] College entrance examination application assistance systems typically compare a candidate's ranking with the lowest admission ranking of previous years, and then output the probability of admission using a linear function, normal function, or fixed threshold. While this method can generate basic results, it still suffers from the following problems when processing real admission data in batches:

[0003] First, ranking is a cumulative quantity that varies with the number of test takers. The same difference in ranking around 20,000th place and around 100,000th place do not have the same meaning; using a fixed transition width will cause inconsistent response speeds across different ranking intervals.

[0004] Second, the competitiveness of different enrollment plans within the same comparison group varies. If only the lowest historical ranking of the target plan is used, it cannot express the plan's relative position within its comparison group. If the rank generation process does not maintain a stable correspondence between the target record identifier and the sorting position, or if all plans are uniformly set to the same rank due to missing values ​​within the group, it will cause the target plan's rank to be distorted and lose the still valid hierarchical information.

[0005] Third, existing piecewise functions are prone to numerical out-of-bounds errors, jumps at connection points, or incorrect directions. The output should not increase as the admission ranking worsens; near the lowest historical admission ranking, the function should smoothly transition from a high probability to zero, rather than abruptly truncate.

[0006] Fourth, the actual data often has missing years. The addition of new majors, changes in major codes, or missing historical data may result in the target plan having only one or two valid years. If a separate set of year weights is set for each missing combination, not only will the rules be redundant, but it will also be difficult to uniformly express the completeness of historical information.

[0007] Fifth, the numerical values ​​generated directly from the historical lowest admission rankings are essentially estimates of the model's admission probability. Only after obtaining the actual application and admission labels and completing time isolation calibration can the values ​​have a statistical interpretation corresponding to the actual admission frequency. Therefore, the system needs to simultaneously save the original estimate, confidence coefficient, and calibration status.

[0008] Therefore, there is a need for an independent forecasting method that is fully defined by input data, computational relationships, parameter constraints, and anomaly handling rules: determining the midpoint order of the target plan within the group on the effective historical records, uniformly normalizing the time-sensitivity weight for any effective annual combination, and providing clear handling rules for continuity, monotonicity, value range, and insufficient data. Summary of the Invention

[0009] The purpose of this invention is to provide a method, system, device, and storage medium for predicting the probability of enrollment in an admissions plan, in order to solve the problems mentioned in the background art.

[0010] This invention is implemented as follows: a method for predicting the probability of enrollment in a student recruitment plan, comprising the following steps:

[0011] Get the candidate's current ranking, the number of valid candidates in the current year, the number of valid candidates in at least one historical year, the historical lowest admission ranking of the target enrollment plan in each historical year, and the comparison group identifier to which the target enrollment plan belongs;

[0012] Based on the current ranking, the number of valid candidates in the current year, and the number of valid candidates in each historical year, generate the annual normalized query ranking within the valid ranking range for each historical year.

[0013] In each historical year, valid records of the lowest historical admission ranking are selected from the comparison group to construct an annual valid set. The number of records in the annual valid set whose lowest historical admission ranking is less than the lowest historical admission ranking of the target enrollment plan, and the number of records whose lowest historical admission ranking is equal to the lowest historical admission ranking of the target enrollment plan are counted. Based on this, the ranking of the target enrollment plan in each historical year is generated.

[0014] Based on the annual normalized query ranking of each historical year, a basic probability transition width with the same ranking unit as the historical minimum admission ranking of the corresponding historical year is generated; based on the parallel midpoint ranking and the number of valid records in the annual valid set, a hierarchical adjustment factor is generated; the basic probability transition width is adjusted using the hierarchical adjustment factor to obtain the planned probability transition width of the target enrollment plan in each historical year.

[0015] Based on the historical lowest admission ranking and the probability transition width of the target enrollment plan in each historical year, determine the probability kernel connection point for each historical year; input the annual normalized query ranking of each historical year into a bounded, monotonic piecewise probability kernel with continuous function values ​​at the probability kernel connection point to generate the annual admission probability of the target enrollment plan in each historical year.

[0016] The basic time-effect weights corresponding to each historical year are limited to the effective years of the historical lowest admission ranking of the target enrollment plan, and the basic time-effect weights are normalized on the effective years to obtain the annual fusion weights of each effective year; the annual fusion weights are used to weight and fuse the admission probabilities of each year to obtain the annual fusion results of the target enrollment plan.

[0017] Based on the sum of the basic timeliness weights corresponding to the valid year, generate the annual completeness factor; based on the number of valid records in the valid set corresponding to the valid year and the preset reference sample size, generate the within-group sample sufficiency factor; based on the annual completeness factor and the within-group sample sufficiency factor, generate the confidence coefficient; output the predicted admission probability results of the target enrollment plan.

[0018] Furthermore, the annual normalized query rank Generate according to the following formula:

[0019] ;

[0020] Where r0 is the current rank; N0 is the number of valid test takers in the current year; N h Let be the number of valid candidates in the h-th historical year; clip indicates that the result is restricted to [1, N]. h ]; round indicates rounding positive numbers up by half, specifically defined as:

[0021] .

[0022] Furthermore, the annual valid set V G,h Construct it in the following way:

[0023] ;

[0024] Where G is the comparison group; L j,h To compare the lowest historical admission ranking of the j-th enrollment plan in the h-th historical year; N h The number of valid candidates in the h-th historical year;

[0025] The parallel midpoint position k i,h Generate according to the following formula:

[0026] ;

[0027] Where, n < This represents the number of historical lowest admission rankings in the annual valid set that are less than the target enrollment plan. The number of records in the valid set for the year that equal the historical lowest admission ranking of the target enrollment plan;

[0028] When the target enrollment plan does not belong to the valid set of years, the corresponding historical year will be marked as invalid for the target enrollment plan, and no parallel midpoint order will be generated;

[0029] This generates the hierarchical adjustment factor q. i,h :

[0030] ;

[0031] Where, m G,h This represents the number of valid records in the valid set for the year.

[0032] Furthermore, the planned probability transition width e i,h Generate according to the following formula:

[0033] ;

[0034] Among them, E h The basic probability transition width is α, which is a preset fixed basic ratio coefficient; the basic probability transition width E h Generate according to the following formula:

[0035] ;

[0036] Where, r h This represents the annual normalized query rank for the h-th historical year.

[0037] After obtaining the planned probability transition width e i,h Then, the following width constraint steps are performed:

[0038] Calculate the upper bound of the width e up :

[0039] ;

[0040] Among them, e max This is the preset upper limit for the planned transition width; It is a preset positive number, and 0 < <1;

[0041] When e min ≤e up At that time, the planned probability transition width is limited to [e] min , eup Within the interval; where e min The preset lower limit of the planned transition width, and 0 <e min ≤e max ;

[0042] When e up <e min When this happens, the corresponding historical year will be marked as invalid and excluded from the valid years.

[0043] Furthermore, the segmented probability kernel p ih Defined as follows:

[0044] Let the target plan's annual historical boundary be The adjusted width is and define:

[0045] ;

[0046] Peak probability and boundary probability satisfy ;

[0047] Scale of Gaussian transition section for:

[0048] ;

[0049] Then the piecewise probability kernel p ih (r h ) is defined as:

[0050] ;

[0051] The piecewise probability kernel p ih (r h )satisfy:

[0052] ;

[0053] .

[0054] Furthermore, the annual fusion weight w i,h and annual integration results Generate according to the following formula:

[0055] ;

[0056] ;

[0057] in, H represents the basic timeliness weight corresponding to the h-th historical year; i Let be the set of valid years for the historical lowest admission ranking of the target enrollment plan; 1(·) is the indicator function; p(i,h) is the annual admission probability of the target enrollment plan in the h-th historical year.

[0058] Furthermore, the annual completeness factor Generate according to the following formula:

[0059] ;

[0060] The within-group sample sufficiency factor Generate according to the following formula:

[0061] ;

[0062] Where, m ref The preset reference sample size is m.ref> 0;

[0063] The confidence coefficient C i Generate according to the following formula:

[0064] ;

[0065] The estimated probability of admission at the planned level, P i Generate according to the following formula:

[0066] ;

[0067] Where p0 is the preset neutral baseline probability value.

[0068] Another object of the present invention is to provide an enrollment plan admission probability prediction system for implementing the above-mentioned enrollment plan admission probability prediction method, comprising:

[0069] The data acquisition module is used to acquire the candidate's current ranking, the number of valid candidates in the current year, the number of valid candidates in at least one historical year, the historical lowest admission ranking of the target enrollment plan in each historical year, and the comparison group identifier to which the target enrollment plan belongs;

[0070] The annual ranking normalization module is used to generate the annual normalized query ranking within the valid ranking range of each historical year based on the current ranking, the number of valid candidates in the current year, and the number of valid candidates in each historical year.

[0071] The effective ranking module is used to select valid records of the lowest historical admission ranking from the comparison group in each historical year, construct the annual effective set, count the number of records in the annual effective set whose lowest historical admission ranking is less than the lowest historical admission ranking of the target enrollment plan, and the number of records whose lowest historical admission ranking is equal to the lowest historical admission ranking of the target enrollment plan, and generate the parallel median ranking of the target enrollment plan in each historical year based on this.

[0072] The transition width adjustment module is used to generate a basic probability transition width with the same ranking unit as the historical lowest admission ranking of the corresponding historical year based on the annual normalized query ranking of each historical year; generate a hierarchical adjustment factor based on the parallel midpoint ranking and the number of valid records in the annual valid set; and adjust the basic probability transition width using the hierarchical adjustment factor to obtain the planned probability transition width of the target enrollment plan in each historical year.

[0073] The annual probability module is used to determine the probability kernel connection point for each historical year based on the historical lowest admission ranking and the probability transition width of the target enrollment plan in each historical year. The annual normalized query ranking of each historical year is input into a bounded, monotonic piecewise probability kernel with continuous function values ​​at the probability kernel connection point to generate the annual admission probability of the target enrollment plan in each historical year.

[0074] The annual fusion module is used to limit the basic time-effect weights corresponding to each historical year to the effective range of the historical lowest admission ranking of the target enrollment plan, and normalize the basic time-effect weights on the effective years to obtain the annual fusion weights for each effective year; the annual fusion weights are then used to weight and fuse the admission probabilities of each year to obtain the annual fusion result of the target enrollment plan.

[0075] The prediction output module generates an annual completeness factor based on the sum of the basic timeliness weights corresponding to the valid year; generates an in-group sample sufficiency factor based on the number of valid records in the valid set corresponding to the valid year and the preset reference sample size; generates a confidence coefficient based on the annual completeness factor and the in-group sample sufficiency factor; and outputs the predicted admission probability results of the target enrollment plan.

[0076] Another object of the present invention is to provide an electronic device including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method.

[0077] Another object of the present invention is to provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0078] The enrollment probability prediction method provided by this invention achieves adaptive ranking, stable handling of missing and tied boundaries within groups, differentiated responses to different plans within the same group, and unified fusion of missing years, making it suitable for enrollment plan prediction. Compared with traditional fixed-window and hierarchical correction-free plan prediction methods, this invention has the following advantages:

[0079] 1. The basic transition width varies with the ranking of the candidates, reducing the scale distortion caused by a fixed ranking window in different ranking ranges;

[0080] 2. The calculation of the midpoint position using only valid annual records can handle missing records and parallel boundaries within a group, and avoid distortion of the correspondence between the target record and the position.

[0081] 3. Different plans within the same comparison group receive different transition widths, allowing the relative competitive positions of the plans to directly participate in the probabilistic response;

[0082] 4. The piecewise probability kernel has continuous function values, monotonic direction, and bounded output at key connection points, which facilitates batch calculation and numerical verification.

[0083] 5. The effective annual weights are uniformly normalized to reduce missing combination branches, and the amount of historical information is explicitly expressed through the confidence coefficient;

[0084] 6. The output retains the yearly results, position, width, weight, and version, and can perform deterministic recalculation on any plan. Attached Figure Description

[0085] Figure 1 This is a flowchart illustrating the method for predicting the probability of enrollment in an admission plan, as provided in an embodiment of the present invention. Detailed Implementation

[0086] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0087] like Figure 1 As shown, in one embodiment of the present invention, a method for predicting the probability of admission to an enrollment plan is provided, comprising the following steps:

[0088] S1. Obtain the candidate's current ranking, the number of valid candidates in the current year, the number of valid candidates in at least one historical year, the historical lowest admission ranking of the target enrollment plan in each historical year, and the comparison group identifier to which the target enrollment plan belongs;

[0089] S2. Based on the current ranking, the number of valid candidates in the current year, and the number of valid candidates in each historical year, generate the normalized query ranking for each year that falls within the valid ranking range.

[0090] S3. In each historical year, select valid records of the lowest historical admission ranking from the comparison group, construct the annual valid set, count the number of records in the annual valid set whose lowest historical admission ranking is less than the lowest historical admission ranking of the target enrollment plan, and the number of records whose lowest historical admission ranking is equal to the lowest historical admission ranking of the target enrollment plan, and generate the joint midpoint ranking of the target enrollment plan in each historical year based on this.

[0091] S4. Based on the annual normalized query ranking of each historical year, generate a basic probability transition width with the same ranking unit as the historical minimum admission ranking of the corresponding historical year; generate a hierarchical adjustment factor based on the parallel midpoint ranking and the number of valid records in the annual valid set; adjust the basic probability transition width using the hierarchical adjustment factor to obtain the planned probability transition width of the target enrollment plan in each historical year.

[0092] S5. Based on the historical lowest admission ranking and the probability transition width of the target enrollment plan in each historical year, determine the probability kernel connection point for each historical year; input the annual normalized query ranking of each historical year into a bounded, monotonic piecewise probability kernel with continuous function values ​​at the probability kernel connection point, and generate the annual admission probability of the target enrollment plan in each historical year.

[0093] S6. Limit the basic time-sensitive weights corresponding to each historical year to the effective years of the lowest historical admission ranking of the target enrollment plan, and normalize the basic time-sensitive weights on the effective years to obtain the annual fusion weights of each effective year; use the annual fusion weights to weight and fuse the admission probabilities of each year to obtain the annual fusion results of the target enrollment plan.

[0094] S7. Generate an annual completeness factor based on the sum of the basic timeliness weights corresponding to the valid year; generate an in-group sample sufficiency factor based on the number of valid records in the valid set corresponding to the valid year and the preset reference sample size; generate a confidence coefficient based on the annual completeness factor and the in-group sample sufficiency factor; output the predicted admission probability results of the target enrollment plan.

[0095] This invention uses a single enrollment plan as the prediction object. Institutions, institutional groups of majors, or sets of majors within the same batch within an institution are only used to establish the historical comparison context of the plan and are not considered as the final prediction object of this invention. In practical application, firstly, the current candidate's ranking is obtained within the same province, subject category, and batch, and a normalized query ranking for the historical year is generated based on the total number of candidates in the current year and historical years. Then, for each target enrollment plan, invalid historical boundaries are filtered within its comparison group, and the midpoint ranking is calculated using the number of records less than the target boundary and the number of records equal to the target boundary. Based on the normalized query ranking for each historical year, a basic transition width is generated within that year's ranking unit, and the midpoint ranking is used to form a hierarchical adjustment factor. Plans that are relatively difficult within the group receive a larger transition width, while plans that are relatively easy within the group receive a smaller transition width. The adjusted width and the historical lowest admission ranking of the target plan together determine three connection points, forming a continuous segmented probability kernel from the high-probability zone, the boundary transition zone, to the zero-value zone. Finally, the annual admission probability is calculated only for years with valid historical boundaries, and the preset time-sensitive weight is restricted to normalization after the valid year. The completeness of the historical year and the sample size of the valid comparison group form a confidence coefficient. In a preferred embodiment, the confidence coefficient is also used to shrink results with insufficient evidence towards a neutral benchmark. This invention achieves the following objectives through the above method:

[0096] 1. Automatically generate a basic probability transition width that is appropriate to the magnitude of the candidate's position;

[0097] 2. Even when there are missing or tied historical positions within a group, the relative order of the target plan can still be calculated correctly;

[0098] 3. Adjust the annual probability transition zone according to the relative ranking of the target plan within the group, so that different plans within the same comparison group can obtain different response scales;

[0099] 4. Using a bounded, monotonic probability kernel with continuous function values, the relationship between the candidate's ranking and the historical lowest admission ranking is converted into the probability of admission in each year.

[0100] 5. Apply uniform timeliness weight normalization to any valid annual combination, and use confidence coefficients to express historical completeness and within-group sample sufficiency;

[0101] 6. Output plan-level results and complete intermediate quantities, enabling deterministic recalculation of the same input and calculation rule version.

[0102] It should be noted that, in the embodiments of the present invention, some terms are defined as follows:

[0103] Current rank: The cumulative rank of the candidate in the specified province, subject category and batch in the current year, denoted as r0. The smaller the value, the better the rank.

[0104] Total number of candidates per year: The number of valid candidates in the current year and historical years under the same statistical caliber, denoted as N0 and N, respectively. h ;

[0105] Annual Normalized Query Rank: The query rank after converting the current rank to the historical annual population scale, denoted as r. h ;

[0106] Historical boundary: The lowest admission ranking of the target enrollment plan i in a historical year h, denoted as L. i,h ;

[0107] Comparison Group: A set of plans for comparison established according to the provincial admission rules and data structure, denoted as G; it can be a set of plans within a college / major group or a set of plans for the same batch within a college;

[0108] Valid set for the year: The set of records in the comparison group whose historical boundaries are finite positive numbers and do not exceed the maximum number of people in that year, denoted as V. G,h ;

[0109] Midpoint rank: The rank of the zero-starting points, denoted as k, is generated based on the number of records less than the target boundary and the number of records equal to the target boundary. i,h ;

[0110] Basic transition width: A positive width generated based on the normalized query rank of a historical year, with the same rank unit as the historical boundary of that year, denoted as E. h ;

[0111] Planned transition width: The ranking width after adjusting the basic transition width using the midpoint order within the target plan group, denoted as e. i,h ;

[0112] Probability of admission by year: The [0,1] value obtained by inputting the annual query ranking into the target plan's annual probability kernel is denoted as p. i,h ;

[0113] Annual fusion result: The original plan-level result obtained after normalizing the time-based weights over the effective year, denoted as

[0114] Confidence coefficient: A [0,1] value formed by the completeness of historical years and the sample size of the effective comparison groups, denoted as C. i .

[0115] In a preferred embodiment of the present invention, the following data is specifically obtained in step S1:

[0116] Current year, province, subject category, batch, and candidate's current ranking ;

[0117] Total number of candidates in the current year and all historical years under the same criteria ;

[0118] The target enrollment plan identifier and its affiliated institution, institution-major group, or intra-institutional comparison set identifier;

[0119] The lowest admission ranking of the target program and other programs within the same comparison group in at least one historical year;

[0120] The optional real application, admission, or enrollment tags are only used for subsequent statistical calibration and evaluation.

[0121] The prediction target is a single enrollment plan. For provinces using university-major group admission, the comparison group is preferably a set of plans within the same university-major group, of the same subject category, and in the same batch; for provinces not using major groups, the comparison group can be a set of plans within the same university of the same subject category and in the same batch. The comparison group only provides the relative ranking and does not change the plan-level output identifier. The ranking and the number of students must be finite positive numbers. If the historical boundary is equal to zero, negative, empty, infinite, or exceeds the upper limit of the corresponding year's number of students, the record will not be included in the valid set for that year. Missing values ​​are not replaced by zero.

[0122] It should be noted that the comparison group can be a college and major group, a group of plans of the same subject category within the college, or other stable plan groups, depending on the provincial admission rules; the group key and the applicable year should be saved in a versioned manner.

[0123] In a preferred embodiment of the present invention, in order to enable the method of the present invention to be executed independently, it is preferable to use population ratio normalization to generate annual normalized query ranking. :

[0124] ;

[0125] Where r0 is the current rank; N0 is the number of valid test takers in the current year; N h Let be the number of valid candidates in the h-th historical year; clip indicates that the result is restricted to [1, N]. h ]; round indicates rounding positive numbers up by half, specifically defined as:

[0126] .

[0127] It should be noted that the above rounding rules and calculation rules are saved together to ensure deterministic recalculation of boundary inputs. When the current year and historical years have already used the same effective population scale, the following settings are applied: ,therefore When the business system has already saved the confirmed historical year query rankings, the rankings can be read directly, but the finiteness and maximum number of people checks are still performed. This step only ensures that the query ranking is within the valid ranking range of the historical year; it does not involve comparison group order or probability calculations. Therefore, this embodiment already includes a complete and replaceable independent input scheme. Furthermore, the annual query ranking can be generated using population ratio normalization, or by searching for the same cumulative percentile or using a confirmed uniform scale ranking; regardless of the method used, the result must be limited to the valid number of people in the historical year.

[0128] In a preferred embodiment of the present invention, for each historical year h, the effective set of years V G,h Construct it in the following way:

[0129] ;

[0130] Where G is the comparison group; L j,h To compare the lowest historical admission ranking of the j-th enrollment plan in the h-th historical year; N h Let m be the number of valid test takers in the h-th historical year. G,h If the target plan is not part of the valid set, then the target plan is unavailable for that year; the absence of other plans does not affect the target plan that is still valid.

[0131] For each effective target plan i, the following statistics are performed:

[0132] ;

[0133] ;

[0134] Then the zero starting point of the target plan is the parallel midpoint position k. i,h Generate according to the following formula:

[0135] ;

[0136] Where, n < This represents the number of historical lowest admission rankings in the annual valid set that are less than the target enrollment plan. The number of records in the valid set for the year that equal the historical lowest admission ranking of the target enrollment plan;

[0137] It should be noted that the midpoint order can be calculated by "less than the number plus half the number of parallel items", or by using a stable rank algorithm that produces the same midpoint result.

[0138] When the target enrollment plan does not belong to the valid set of years, the corresponding historical year will be marked as invalid for the target enrollment plan, and no parallel midpoint order will be generated;

[0139] This generates a hierarchical adjustment factor (i.e., a difficulty direction factor) q. i,h :

[0140] ;

[0141] Where, m G,h This represents the number of valid records in the annual valid set. Programs with lower historical minimum admission rankings are generally more difficult to secure. The closer to 1; the easier the plan. The closer The midpoint order is directly determined by the numerical relationship between valid records, and the same boundary is uniformly processed by the number of parallel records, so it is not affected by the original record order.

[0142] In a preferred embodiment of the present invention, the planned probability transition width e i,h Generate according to the following formula:

[0143] ;

[0144] Among them, E h The base probability transition width; α is a preset fixed base ratio coefficient, preferably 0.75, that is, the planned transition width is formed by three-quarters of the base width and one-quarter of the in-group position adjustment.

[0145] Furthermore, for each historical year, based on the above-mentioned annual normalized query rank, a basic transition width that grows with the rank magnitude is constructed, let:

[0146] ;

[0147] Reduce the normal density relative to the peak to The difference between the right-hand position corresponding to the time and the query rank for that year is defined as the basic transition width:

[0148] ;

[0149] Substituting the default parameters above, the base probability transition width E h It can be simplified to:

[0150] ;

[0151] Where, r h This represents the normalized query rank for the h-th historical year; therefore, the base width is not a fixed number of ranks, but rather varies proportionally with the query rank of the corresponding historical year, maintaining the same rank unit as the historical boundary of that year. Preferably, a positive lower bound can also be set. and the Upper Realm :

[0152] ;

[0153] in, Both are only used to constrain the basic transition width. The transition width with subsequent constraint plans Set and include each calculation rule version separately; omit this step if the basic width limit is not enabled.

[0154] It should be noted that the basic transition width can be determined by the inverse solution formula of the density threshold, or by a function that monotonically increases with the annual normalized query position and has positive upper and lower bounds.

[0155] Furthermore, to ensure the effectiveness of subsequent probability kernel parameters, the planned probability transition width e is obtained. i,h Then, the following width constraint steps are performed:

[0156] Calculate the upper bound of the width e up :

[0157] ;

[0158] Among them, e max This is the preset upper limit for the planned transition width; It is a preset positive number, and 0 < <1;

[0159] When e min ≤e up At that time, the planned probability transition width is limited to [e] min , eup Within the interval; where emin The preset lower limit of the planned transition width, and 0 <e min ≤e max ;

[0160] When e up <e min If this occurs, it indicates that the historical boundary for that year is too small or the parameter configuration is inapplicable. The corresponding historical year will be marked as invalid and the data will be removed from the valid year list. If excluded, the upper and lower bounds of the constraint are not swapped, and the annual probability is not calculated further. The above conditions guarantee... Thus ensuring and It is a positive number.

[0161] In a preferred embodiment of the present invention, the segmented probability kernel p ih Defined as follows:

[0162] Let the target plan's annual historical boundary be The adjusted width is and define:

[0163] ;

[0164] Peak probability and boundary probability satisfy ;

[0165] Scale of Gaussian transition section for:

[0166] ;

[0167] Then the piecewise probability kernel p ih (r h ) is defined as:

[0168] ;

[0169] The piecewise probability kernel p ih (r h )satisfy:

[0170] ;

[0171] .

[0172] Therefore, the piecewise probability kernel is The function values ​​at the three connection points are continuous and monotonically non-increasing with the query position. In implementation, the final value is restricted to [0,1] to handle floating-point errors and abnormal inputs. This protects a combination of structures with continuous, bounded, monotonic function values ​​and finite transition regions; only continuous function values ​​are required, not that the first derivatives at the connection points are equal.

[0173] It should be noted that the high-probability segments and boundary segments of the piecewise probability kernel can be replaced with other bounded monotonic functions; the replacement function should maintain continuous function values ​​at the connection points and return to zero at L+e.

[0174] In a preferred embodiment of the present invention, the set of effective historical years for the target plan is denoted as:

[0175] ;

[0176] If H i If empty, the calculations for weights, minimum sample size, and shrinkage will not be performed; instead, the "Insufficient Data" status will be output directly. The following formula is only used for H. i The case where it is not empty.

[0177] Configure a positive base timeliness weight for each candidate historical year For example, in an embodiment with 2026 as the target year, the following weights are applied to the three historical years of 2023, 2024, and 2025:

[0178] ;

[0179] This invention does not require pre-setting a specific weight for each missing combination, but only uniformly normalizes it over the valid year, thus the annual fusion weight w i,h and annual integration results Generate according to the following formula:

[0180] ;

[0181] ;

[0182] in, H represents the basic timeliness weight corresponding to the h-th historical year; i Let be the set of valid years for the historical lowest admission ranking of the target enrollment plan; 1(·) is the indicator function; p(i,h) is the annual admission probability of the target enrollment plan in the h-th historical year.

[0183] It should be noted that the annual basic timeliness weight can be a fixed configuration, an exponential time decay, or a positive weight selected through rolling verification. Missing years are uniformly excluded through an indicator function and then renormalized.

[0184] In a preferred embodiment of the present invention, the annual completeness factor Generate according to the following formula:

[0185] ;

[0186] Within-group sample sufficiency factor Generate according to the following formula:

[0187] ;

[0188] Where, m ref The preset reference sample size is m. ref> 0;

[0189] Then the confidence coefficient C i Generate according to the following formula:

[0190] ;

[0191] The embodiments of the present invention output at least the original annual fusion results simultaneously. and confidence coefficient C i Preferably, by using a preset neutral baseline probability value p0 (e.g., 0.5) to narrow down the low-confidence results, the planned admission probability estimate P... i Generate according to the following formula:

[0192] ;

[0193] When historical years are complete and the comparison group sample is sufficient The final result is consistent with the year-by-year weighted value; when historical data is insufficient, the output moves closer to the neutral benchmark while retaining the original fusion result for auditing. If low-confidence shrinkage is not enabled, then the following is defined: This ensures that subsequent outputs and optional calibrations always have a definite input. Zero-bit or zero-probability data must not be masqueraded as valid historical data.

[0194] It should be noted that low confidence processing can output the original results and confidence coefficients, and can also further perform shrinkage towards a neutral benchmark or a benchmark from the same batch.

[0195] In a preferred embodiment of the present invention, when there are no real application and admission tags at the candidate level... This represents the "admission probability estimate" or "admission probability score" generated by the historical lowest admission ranking model, and its status is marked as "uncalibrated estimate". This value can be used for comparison of plans within the same calculation rule version, the same province, and the same batch.

[0196] When the business system has real labels, set The monotonically non-decreasing calibration mapping was trained using data strictly earlier than the target prediction year:

[0197] ;

[0198] The calibration function can employ equidistant binning monotonic regression, ordinal-preserving regression, or logistic calibration. After calibration, the output shows a "calibrated estimate" status and records the training year, province, subject category, batch, and calibration version. The calibration module is an optional implementation and does not affect the independent execution of the aforementioned plan-level prediction chain.

[0199] In a preferred embodiment of the present invention, in step S7, each plan outputs at least the following results:

[0200] 1. Plan identifier, current position, and calculation rule version;

[0201] 2. Valid historical year set and normalized query ranking for each year;

[0202] 3. Historical boundaries, effective group size, midpoint rank, and rank factor for each year;

[0203] 4. Basic transition width, annual plan transition width, and probability of admission in different years;

[0204] 5. Annual normalized weights, original annual fusion results, confidence coefficients, and final results;

[0205] 6. Status: "Uncalibrated estimate", "calibrated estimate" or "Insufficient data".

[0206] For each comparison group and each historical year, sorting the effective historical boundaries once allows reuse of the positional order of all plans, with a time complexity of O(n log n). After sorting, the probability calculation for each plan for each year is performed with constant complexity.

[0207] In another embodiment of the present invention, an enrollment plan admission probability prediction system is also provided to implement the above-described enrollment plan admission probability prediction method, specifically including:

[0208] The data acquisition module is used to acquire the candidate's current ranking, the number of valid candidates in the current year, the number of valid candidates in at least one historical year, the historical lowest admission ranking of the target enrollment plan in each historical year, and the comparison group identifier to which the target enrollment plan belongs;

[0209] The annual ranking normalization module is used to generate the annual normalized query ranking within the valid ranking range of each historical year based on the current ranking, the number of valid candidates in the current year, and the number of valid candidates in each historical year.

[0210] The effective ranking module is used to select valid records of the lowest historical admission ranking from the comparison group in each historical year, construct the annual effective set, count the number of records in the annual effective set whose lowest historical admission ranking is less than the lowest historical admission ranking of the target enrollment plan, and the number of records whose lowest historical admission ranking is equal to the lowest historical admission ranking of the target enrollment plan, and generate the parallel median ranking of the target enrollment plan in each historical year based on this.

[0211] The transition width adjustment module is used to generate a basic probability transition width with the same ranking unit as the historical lowest admission ranking of the corresponding historical year based on the annual normalized query ranking of each historical year; generate a hierarchical adjustment factor based on the parallel midpoint ranking and the number of valid records in the annual valid set; and adjust the basic probability transition width using the hierarchical adjustment factor to obtain the planned probability transition width of the target enrollment plan in each historical year.

[0212] The annual probability module is used to determine the probability kernel connection point for each historical year based on the historical lowest admission ranking and the probability transition width of the target enrollment plan in each historical year. The annual normalized query ranking of each historical year is input into a bounded, monotonic piecewise probability kernel with continuous function values ​​at the probability kernel connection point to generate the annual admission probability of the target enrollment plan in each historical year.

[0213] The annual fusion module is used to limit the basic time-effect weights corresponding to each historical year to the effective range of the historical lowest admission ranking of the target enrollment plan, and normalize the basic time-effect weights on the effective years to obtain the annual fusion weights for each effective year; the annual fusion weights are then used to weight and fuse the admission probabilities of each year to obtain the annual fusion result of the target enrollment plan.

[0214] The prediction output module generates an annual completeness factor based on the sum of the basic timeliness weights corresponding to the valid year; generates an in-group sample sufficiency factor based on the number of valid records in the valid set corresponding to the valid year and the preset reference sample size; generates a confidence coefficient based on the annual completeness factor and the in-group sample sufficiency factor; and outputs the predicted admission probability results of the target enrollment plan.

[0215] It should be noted that the modules can be integrated into the same computing node, or they can be separated into an offline batch processing module and an online query module. In the offline stage, the effective sorting and median order are pre-calculated by year and comparison group; in the online stage, the width, year-specific results, and fusion results are calculated after receiving the candidates' rankings.

[0216] In another embodiment of the present invention, an electronic device is also provided, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described above. It should be noted that the electronic device may be a server, a personal computer, a cloud computing node, or a dedicated computing device for processing student admissions data.

[0217] In another embodiment of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the above-described method.

[0218] Example 1: This example uses Jilin's 2026 enrollment plan as an example to illustrate the method of the present invention:

[0219] 1. Data Implementation Basis: The Jilin 2026 enrollment combined table contains 19,531 enrollment records. Based on the historical lowest admission rankings from 2023 to 2025, 10,004 records are complete for all three years, 2,789 records are valid for only two years, 2,480 records are valid for only one year, and an average of 4,258 records are missing for all three years. This demonstrates that the merging of valid rankings within a group and missing years is a common data state in batch calculations.

[0220] The following example uses Hebei University of Engineering's Group 006 from the Jilin 2026 combined admission table. This group has 39 current plans; 37 have valid historical minimum admission rankings for 2025, and 2 plans have empty historical boundaries for the past three years. The candidate's current ranking is taken as 35000. This embodiment uses the path of directly reading the confirmed historical year's query ranking, taking 35000, 35000, and 34731 as the known query rankings for 2023, 2024, and 2025, respectively. If the query rankings are not obtained in advance, they are generated independently according to the population ratio formula.

[0221] 2. Recalculation of basic transition width: The query rank for both 2023 and 2024 is 35000. Substitute this into the above formula:

[0222] ;

[0223] The query rank in 2025 was 34731, corresponding to:

[0224] ;

[0225] All the above values ​​were directly calculated using the above formula. The following values ​​are taken as... The annual base weights are 0.05, 0.10, and 0.85; in this embodiment, the base width limit is not enabled, and the upper and lower bound limits are not triggered for any of the planned widths.

[0226] 3. Complete calculation of the surveying and mapping project: The minimum admission rankings for the "Surveying and Mapping Project" in 2023, 2024, and 2025 are 35294, 36545, and 37905, respectively. The effective group size for all three years is 37.

[0227] In 2023, 28 records were smaller than the target boundary, and 2 records were equal to the target boundary.

[0228] ;

[0229] ;

[0230] ;

[0231] r 2023 =35000 Input the above piecewise probability kernel, and we get:

[0232]

[0233] In 2024, 27 records were smaller than the target boundary, and 5 records were equal to the target boundary.

[0234] ;

[0235] In 2025, the number of records smaller than the target boundary was 28, and the number of records equal to the target boundary was 1.

[0236] ;

[0237] All three years were valid and the comparison group had a sufficient sample size, therefore The planned fusion result is as follows:

[0238] ;

[0239] The estimated probability of admission under this parameter version is 50.932979%. For the same applicants and comparison groups, the recalculated result for "Facility Agriculture Science and Engineering" is 56.264212%, and for "Animal Science" it is 49.146204%. The differences among the three arise from the historical boundaries of each year and their midpoint ranking within the valid comparison groups.

[0240] 4. Missing Year Demonstration: If the above target plan is only valid for 2024 and 2025, the weight of the valid year is obtained by a unified formula:

[0241] ;

[0242] ;

[0243] Annual completeness is . This embodiment takes , the number of samples in the comparison group for both valid years is 37, so ; then take :

[0244] ;

[0245] This calculation does not require a dedicated fusion rule specifically for "missing the year 2023". If the target program has no valid boundaries in all three years, a "insufficient data" status is output, and no pseudo-probability is output.

[0246] 5. Embodiment check result: after recalculating the above data, all three years satisfy 0<e<L, and the annual results all lie within [0, 1]; at substituting the left and right adjacent segments, the function values are uniformly equal to and 0. The absence of two other programs in the group does not prevent "Surveying and Mapping Engineering" from participating in the annual calculation using its own valid boundaries.

[0247] Compared with the prior art: the conventional fixed threshold or fixed rank window solution does not change the transition width according to the magnitude of the annual query rank; whereas the present invention generates a base width consistent with the rank scale of each historical year through E h The conventional solution that only compares the candidate's rank with the historical boundary of the target program does not use the information of the comparison group to which the target program belongs; whereas the present invention calculates the midpoint rank in valid records, and adjusts the probability transition region of the target program using the rank factor. In addition, in the prior art, solutions that do not determine a stable rank based on record values and target record identifiers, or adopt a uniform rank once a missing value is encountered, will cause target record misalignment or information loss; whereas the present invention determines the true rank of the target program through "count of values less than + midpoint of ties", and missing other records will not invalidate the entire comparison group. The prior art solution that sets year weights separately for each missing combination requires maintaining multiple sets of rules; whereas the present invention uses the annual validity indicator for unified normalization, and records the change in information amount caused by missing values with a confidence coefficient.

[0248] It is worth noting that the embodiments of this invention have already provided the input data required for prediction, the annual rank normalization relationship, the basic transition width formula, the intra-group midpoint order rule, the plan width adjustment relationship, the segmented probability kernel, the effective year fusion rule, the confidence coefficient, and the anomaly boundary. Those skilled in the art can obtain a definite plan-level admission probability estimate from the input data based solely on the inputs, formulas, parameter constraints, and anomaly rules disclosed in the embodiments of this invention, without needing to supplement any undisclosed processing steps. The key technology combination of this invention is "annual normalized rank driving basic width - parallel midpoint order on effective records - plan width adjustment - continuous bounded segmented probability kernel - effective year weight normalization - confidence coefficient output"; the specific computing carrier, data storage format, or deployment method used does not change this technology combination or its input-output relationship.

[0249] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods.

[0250] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for predicting the probability of enrollment in an admissions plan, characterized in that, Includes the following steps: Get the candidate's current ranking, the number of valid candidates in the current year, the number of valid candidates in at least one historical year, the historical lowest admission ranking of the target enrollment plan in each historical year, and the comparison group identifier to which the target enrollment plan belongs; Based on the current ranking, the number of valid candidates in the current year, and the number of valid candidates in each historical year, generate the annual normalized query ranking within the valid ranking range for each historical year. In each historical year, valid records of the lowest historical admission ranking are selected from the comparison group to construct an annual valid set. The number of records in the annual valid set whose lowest historical admission ranking is less than the lowest historical admission ranking of the target enrollment plan, and the number of records whose lowest historical admission ranking is equal to the lowest historical admission ranking of the target enrollment plan are counted. Based on this, the ranking of the target enrollment plan in each historical year is generated. Based on the annual normalized query ranking of each historical year, a basic probability transition width with the same ranking unit as the historical minimum admission ranking of the corresponding historical year is generated; based on the parallel midpoint ranking and the number of valid records in the annual valid set, a hierarchical adjustment factor is generated; the basic probability transition width is adjusted using the hierarchical adjustment factor to obtain the planned probability transition width of the target enrollment plan in each historical year. Based on the historical lowest admission ranking and the probability transition width of the target enrollment plan in each historical year, determine the probability kernel connection point for each historical year; input the annual normalized query ranking of each historical year into a bounded, monotonic piecewise probability kernel with continuous function values ​​at the probability kernel connection point to generate the annual admission probability of the target enrollment plan in each historical year. The basic time-effect weights corresponding to each historical year are limited to the effective years of the historical lowest admission ranking of the target enrollment plan, and the basic time-effect weights are normalized on the effective years to obtain the annual fusion weights of each effective year; the annual fusion weights are used to weight and fuse the admission probabilities of each year to obtain the annual fusion results of the target enrollment plan. The annual completeness factor is generated based on the sum of the basic timeliness weights corresponding to the effective years. Based on the number of valid records in the valid set corresponding to the valid year and the preset reference sample size, generate the within-group sample sufficiency factor; generate the confidence coefficient based on the annual completeness factor and the within-group sample sufficiency factor. Output the predicted probability of admission for the target enrollment plan.

2. The method for predicting the probability of admission to an enrollment plan according to claim 1, characterized in that, The annual normalized query rank Generate according to the following formula: ； Where r0 is the current rank; N0 is the number of valid test takers in the current year; N h Let be the number of valid candidates in the h-th historical year; clip indicates that the result is restricted to [1, N]. h ]; round indicates rounding positive numbers up by half, specifically defined as: 。 3. The method for predicting the probability of enrollment in an admission plan according to claim 1, characterized in that, The annual valid set V G,h Construct it in the following way: ； Where G is the comparison group; L j,h To compare the lowest historical admission ranking of the j-th enrollment plan in the h-th historical year; N h The number of valid candidates in the h-th historical year; The parallel midpoint position k i,h Generate according to the following formula: ； Where, n < This represents the number of historical lowest admission rankings in the annual valid set that are less than the target enrollment plan. This represents the number of records in the valid set for the year that match the historical lowest admission ranking of the target enrollment plan. When the target enrollment plan does not belong to the valid set of years, the corresponding historical year will be marked as invalid for the target enrollment plan, and no parallel midpoint order will be generated; This generates the hierarchical adjustment factor q. i,h : ； Where, m G,h This represents the number of valid records in the valid set for the year.

4. The method for predicting the probability of admission to an enrollment plan according to claim 3, characterized in that, The planned probability transition width e i,h Generate according to the following formula: ； Among them, E h The basic probability transition width is α, which is a preset fixed basic ratio coefficient; the basic probability transition width E h Generate according to the following formula: ； Where, r h This is the annual normalized query rank for the h-th historical year; After obtaining the planned probability transition width e i,h Then, the following width constraint steps are performed: Calculate the upper bound of the width e up : ； Among them, e max This is the preset upper limit for the planned transition width; It is a preset positive number, and 0 < <1; When e min ≤e up At that time, the planned probability transition width is limited to [e] min , eup Within the interval; where e min The preset lower limit of the planned transition width, and 0 <e min ≤e max ; When e up <e min When this happens, the corresponding historical year will be marked as invalid and excluded from the valid years.

5. The method for predicting the probability of admission to an enrollment plan according to claim 4, characterized in that, The segmented probability kernel p ih Defined as follows: Let the target plan's annual historical boundary be The adjusted width is and define: ； Peak probability and boundary probability satisfy ; Scale of Gaussian transition section for: ； Then the piecewise probability kernel p ih (r h ) is defined as: ； The piecewise probability kernel p ih (r h )satisfy: ； 。 6. The method for predicting the probability of admission to an enrollment plan according to claim 5, characterized in that, The annual fusion weight w i,h and annual integration results Generate according to the following formula: ； ； in, H represents the basic timeliness weight corresponding to the h-th historical year; i Let be the set of valid years for the historical lowest admission ranking of the target enrollment plan; 1(·) is the indicator function; p(i,h) is the annual admission probability of the target enrollment plan in the h-th historical year.

7. The method for predicting the probability of admission to an enrollment plan according to claim 6, characterized in that, The annual completeness factor Generate according to the following formula: ； The within-group sample sufficiency factor Generate according to the following formula: ； Where, m ref The preset reference sample size is m. ref> 0; The confidence coefficient C i Generate according to the following formula: ； The estimated probability of admission at the planned level, P i Generate according to the following formula: ； Where p0 is the preset neutral baseline probability value.

8. An enrollment plan admission probability prediction system, used to implement the enrollment plan admission probability prediction method according to any one of claims 1-7, characterized in that, include: The data acquisition module is used to acquire the candidate's current ranking, the number of valid candidates in the current year, the number of valid candidates in at least one historical year, the historical lowest admission ranking of the target enrollment plan in each historical year, and the comparison group identifier to which the target enrollment plan belongs; The annual ranking normalization module is used to generate the annual normalized query ranking within the valid ranking range of each historical year based on the current ranking, the number of valid candidates in the current year, and the number of valid candidates in each historical year. The effective ranking module is used to select valid records of the lowest historical admission ranking from the comparison group in each historical year, construct the annual effective set, count the number of records in the annual effective set whose lowest historical admission ranking is less than the lowest historical admission ranking of the target enrollment plan, and the number of records whose lowest historical admission ranking is equal to the lowest historical admission ranking of the target enrollment plan, and generate the parallel median ranking of the target enrollment plan in each historical year based on this. The transition width adjustment module is used to generate a basic probability transition width with the same ranking unit as the historical lowest admission ranking of the corresponding historical year based on the annual normalized query ranking of each historical year; generate a hierarchical adjustment factor based on the parallel midpoint ranking and the number of valid records in the annual valid set; and adjust the basic probability transition width using the hierarchical adjustment factor to obtain the planned probability transition width of the target enrollment plan in each historical year. The annual probability module is used to determine the probability kernel connection point for each historical year based on the historical lowest admission ranking and the probability transition width of the target enrollment plan in each historical year. The annual normalized query ranking of each historical year is input into a bounded, monotonic piecewise probability kernel with continuous function values ​​at the probability kernel connection point to generate the annual admission probability of the target enrollment plan in each historical year. The annual fusion module is used to limit the basic time-effect weights corresponding to each historical year to the effective range of the historical lowest admission ranking of the target enrollment plan, and normalize the basic time-effect weights on the effective years to obtain the annual fusion weights for each effective year; the annual fusion weights are then used to weight and fuse the admission probabilities of each year to obtain the annual fusion result of the target enrollment plan. The prediction output module is used to generate an annual completeness factor based on the sum of the basic timeliness weights corresponding to the valid year; generate an in-group sample sufficiency factor based on the number of valid records in the valid set corresponding to the valid year and the preset reference sample size; and generate a confidence coefficient based on the annual completeness factor and the in-group sample sufficiency factor. Output the predicted probability of admission for the target enrollment plan.

9. An electronic device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-7.

Images