A method for constructing an automatic selection model for paper submission based on multi-dimensional feature analysis

Abstract

The present invention discloses a method for constructing an automatic selection model for paper submission based on multi-dimensional feature analysis, and relates to the technical field of automatic selection model construction for paper submission. First, data collection and university subject portrait construction are carried out, and data such as comprehensive rankings, category rankings, and first- and second-level subject rankings of universities are obtained from multiple authoritative channels to construct portraits. Then, the universities to be reviewed and submitted for review are represented by vectors, and the relevant features of the universities are converted into vector form through a specific mapping function. Then, the scoring weighting factors of the universities to be reviewed are calculated, and the number of reviews, total review time, and the number of blacklist experts are comprehensively considered. The weight coefficients are determined by scientific methods, and then the cosine similarity is used to calculate the similarity between the vectors of the submitted and received universities. The present invention can improve the efficiency and accuracy of paper submission and selection, comprehensively select suitable universities based on multiple factors, dynamically adapt to data changes, ensure academic fairness, and can be self-optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic selection model construction for paper submission, and in particular to a method for constructing an automatic selection model for paper submission based on multi-dimensional feature analysis. Background Art

[0002] In the current academic research field, paper publication is an important way to display and communicate scientific research results. The paper submission and review process is crucial to ensure academic quality. However, the traditional paper submission and review model has many problems.

[0003] In the traditional paper submission process, editors usually rely on manual experience to select reviewing universities or experts. This manual selection method faces the problem of low efficiency. Editors need to spend a lot of time to understand the strength of relevant disciplines in various universities and the situation of potential review experts, and screen suitable review objects from many universities one by one. This process is lengthy and cumbersome, which seriously affects the timeliness of paper submission and causes the paper to experience a long waiting time from submission to the review stage.

[0004] Moreover, manual selection is difficult to comprehensively consider various relevant factors. For example, the editor may only be familiar with the comprehensive rankings of some universities, but it is difficult to accurately grasp the specific rankings of specific disciplines in universities, the scope of subject coverage and other details. This one-sided consideration often leads to unreasonable selection of universities for review, and there may be a situation where the matching degree between the universities for review and the research direction of the paper is not high. For example, sending a highly specialized science and engineering paper to a university with obvious advantages in liberal arts for review will not only fail to guarantee the quality of the review, but may also delay the timely publication of the author's research results.

[0005] Furthermore, the existing review model lacks dynamism and data support. When editors select universities for review, it is difficult to obtain real-time data on the universities' review activity, review efficiency, and expert credibility. For example, although some universities have strong academic strength, the number of review experts has decreased recently due to various reasons, and the review efficiency has dropped significantly. However, manual selection cannot detect this change in time, which affects the smoothness of the entire paper submission and review process.

[0006] In addition, with the continuous development of academic research and the increase in the number of universities, the number of paper submissions has also been rising year by year. The manual review model is increasingly unable to cope with the massive amount of papers and numerous potential reviewing universities, and it is difficult to meet the growing needs of academic exchanges.

[0007] In order to solve the defects in the above-mentioned traditional paper submission model, there is an urgent need for an automatic paper submission selection model that can perform analysis based on multi-dimensional features and is automated, precise and dynamic. It can comprehensively consider multiple factors such as the comprehensive ranking of universities, subject rankings, review data, etc., quickly and accurately screen out suitable universities for paper submission, improve the efficiency and quality of paper submission, and ensure that academic research results can be reviewed and published in a timely and fair manner. Summary of the invention

[0008] The present invention proposes a method for constructing an automatic selection model for paper submission based on multi-dimensional feature analysis to solve the problems mentioned in the above-mentioned prior art.

[0009] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: a method for constructing an automatic selection model for paper submission based on multi-dimensional feature analysis, comprising the following steps:

[0010] S1. Data collection and university subject portrait construction steps:

[0011] Set college collection , through multiple channels for colleges and universities h1-h n Collect comprehensive rankings , School Category Ranking , first-level discipline ranking , List of Secondary Disciplines Characteristic data and comprehensive ranking data are derived from the comprehensive strength rankings of universities published annually by education evaluation agencies; school category rankings are obtained from the comparison of similar universities based on the classification standards of universities by education authorities; first-level discipline rankings are taken from discipline evaluation reports published by academic organizations or evaluation agencies; and the second-level discipline list is extracted from the discipline construction page of the official website of the university or the discipline information database filed by the education department to construct a discipline portrait. ;

[0012] S2. Steps for vector representation of the receiving university:

[0013] The vector representation of the receiving college is , where f is the mapping function constructed using deep learning technology, R e Indicates the comprehensive ranking of the universities under review, C e Indicates the category ranking of the universities receiving the review, S 1，e S represents the first-level discipline ranking of the reviewing university. 2，eIt represents the list of secondary disciplines of the universities under review. Specifically, a deep neural network is used, and its architecture includes an input layer, multiple hidden layers, and an output layer. The number of nodes in the input layer is consistent with the number of collected university features, and the number of nodes in the output layer is set according to the dimension of the vector space. During the training process, the university data samples with accurate annotations are used, and the weights and biases of the neural network are continuously optimized through the back-propagation algorithm, so that the generated vectors represent the multi-dimensional features of the universities.

[0014] S3. Vector representation steps for universities submitting for review:

[0015] The vector representation of the universities submitting the review is , where R S Indicates the comprehensive ranking of the universities submitted for review, C S Indicates the category ranking of the universities submitted for review, S 1，S Indicates the first-level discipline ranking of the university submitted for review, S 2，S Represents the secondary discipline list of the universities submitting for review. The training and optimization methods of the mapping function f are the same as the vector representation steps of the universities receiving the review.

[0016] S4. Calculation steps of scoring weighting factors:

[0017] Setting the scoring weighting factors for the universities receiving the review , α, β, and γ are weight coefficients. The weight coefficients are determined by rigorously constructing a pairwise comparison judgment matrix, and the relative importance of each indicator is accurately judged based on historical data. Specifically, if the expert determines that the importance of the number of reviews relative to the total review time is 3 based on experience, then the corresponding position in the judgment matrix is ​​accurately assigned a value of 3;

[0018] Normalize each element of the judgment matrix by dividing it by the sum of its column, then average each row of the normalized matrix to obtain the weight, and strictly calculate the consistency ratio CR, where , , CI represents the consistency index, is the maximum eigenvalue of the judgment matrix, n is the order of the judgment matrix. If the consistency test fails to meet the standard, the hierarchical analysis method is used to carefully adjust the judgment matrix until it meets the consistency requirements. The number of reviews is calculated from the historical records of the paper submission system; the total review time is accurately calculated by accurately recording the submission start time and the review result return time of each paper, and the average review time of a single paper is obtained by combining the number of reviews; the number of blacklist experts is calculated from the number of experts marked as unqualified in the expert database;

[0019] S5. Vector similarity calculation steps:

[0020] The cosine similarity is used to calculate the similarity between the submitting university and the receiving university. , normalize the calculation results so that their value range is strictly limited to between 0 and 1, where It represents the similarity between the submitted university vector and the received university vector. The specific normalization operation is as follows: Let the original cosine similarity calculation result be x, and the normalized result be , where min and max are the minimum and maximum values ​​of all similarity calculation results respectively;

[0021] S6. Comprehensive scoring model steps:

[0022] Combine the recommendation index with the weighting factor to form a comprehensive score To prevent outliers from causing excessive interference to the comprehensive score, the weighting factor W e Reasonably set upper and lower limits, based on extensive historical data analysis, set W e The upper limit is 10 and the lower limit is 0. When the calculation result exceeds this range, the boundary value is taken as the final weighting factor;

[0023] S7, threshold setting and sorting steps:

[0024] Set a threshold and select universities with comprehensive scores higher than the threshold ,in represents the set of recommended universities, h i represents a certain university in the university set H, S e，i Indicates college h i The threshold T is adjusted dynamically and flexibly according to the detailed distribution of historical submission data and actual specific needs. The adjustment method is as follows: in-depth analysis of the distribution curve of the comprehensive score in the historical submission data, and determination of the threshold position according to different selection objectives;

[0025] Set the threshold at the high segment of the distribution curve; if it tends to be high coverage, lower the threshold. At the same time, fine-tune the threshold based on the feedback information from actual review tasks, including the actual review quality and matching degree of the recommended universities.

[0026] Furthermore, it also includes feedback optimization steps: collecting feedback information from reviewers, the feedback information includes an evaluation of the degree of match between the paper and the discipline of the receiving university, an evaluation of the review quality, and a comprehensive evaluation of the rationality of the review time; adjusting feature weights and recommendation algorithms based on feedback results. If the feedback evaluation corresponding to a certain feature is lower than the threshold, the weight of the feature in the model is appropriately reduced; if it is found that the recommendation algorithm produces unreasonable recommendations, the algorithm is optimized and improved in a targeted manner.

[0027] Furthermore, it also includes a model evaluation step: using a test data set to evaluate the model, the test data set contains paper samples and university samples, and the distribution of the samples is consistent with the actual review scenario; using accuracy and recall rate indicators to measure model performance, the accuracy is calculated as the exact ratio of the number of accurately recommended universities to the total number of recommended universities, and the recall rate is calculated as the accurate ratio of the number of accurately recommended universities to the actual number of universities that should be recommended; according to the evaluation results, timely discover and improve model defects. If the accuracy is found to be lower than the threshold, adjust the features or optimize the algorithm.

[0028] Furthermore, it also includes data preprocessing steps: comprehensive cleaning, efficient denoising and standardization of the collected feature data; during the cleaning process, duplicate records, erroneous values ​​and missing values ​​in the data are removed; denoising processing uses data smoothing algorithms, specifically the moving average method, to reduce random fluctuations in the data; standardization processing maps data with different characteristics to preset intervals, including using normalization methods to convert data to intervals or using standardization methods to make the data conform to the normal distribution.

[0029] Furthermore, in the steps of data collection and university subject portrait construction, the characteristic data is updated regularly, and the update cycle is determined according to the update frequency of the data source.

[0030] Furthermore, in the vector representation step of the receiving universities and the vector representation step of the submitting universities, regularization technology is introduced to prevent model overfitting and improve the generalization ability of the model.

[0031] Furthermore, in the step of calculating the scoring weighting factors, the weight coefficients α, β, and γ are dynamically adjusted, and the values ​​of each coefficient are adjusted according to the characteristics of different subject areas or paper types.

[0032] Furthermore, in the vector similarity calculation step, an improved cosine similarity algorithm is used to consider the correlation between features and improve the accuracy of similarity calculation.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] First, improve selection efficiency. Through data collection and algorithm processing, we can quickly analyze many universities and automatically select suitable universities for review, reducing the tedious operation of manual one-by-one comparison, and greatly shortening the time from submission to review.

[0035] Secondly, improve the accuracy of selection. Comprehensive university rankings, subject rankings, subject lists and other multi-dimensional characteristics are taken into consideration, while also considering factors such as the review activity of the receiving university, average review time and expert credibility, to ensure that the submitting university and the paper are highly matched in terms of subject and major, and to ensure the quality of the review.

[0036] Furthermore, it has dynamic adaptability. The model can adjust the recommendation of universities for review in a timely manner according to the dynamic changes of relevant university data, such as updates of subject rankings and changes in review status, so that the selection results always remain consistent with the actual situation and avoid unreasonable submission due to information lag.

[0037] In addition, academic fairness is promoted. Screening based on objective data and scientific algorithms reduces the interference of human factors. Whether it is a well-known university or a relatively unknown university with strength in a specific field, everyone has the opportunity to participate in appropriate paper reviews, ensuring a fair and just academic review environment.

[0038] Finally, through feedback optimization and model evaluation mechanisms, we can continuously improve ourselves and continuously enhance model performance to better serve the review of academic papers. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a schematic diagram of a method for constructing an automatic selection model for paper submission based on multi-dimensional feature analysis proposed in the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0042] In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined. In addition, the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, and it can be the internal connection of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0043] Reference Figure 1 :A method for constructing an automatic selection model for paper submission based on multi-dimensional feature analysis, comprising the following steps:

[0044] S1. Data collection and university subject portrait construction steps:

[0045] Set college collection , through multiple channels for colleges and universities h1-h n Collect comprehensive rankings , School Category Ranking , first-level discipline ranking , List of Secondary Disciplines Characteristic data. The comprehensive ranking data comes from the comprehensive strength ranking list of universities published annually by professional and authoritative education evaluation institutions. The list is based on multi-dimensional evaluations of the university's teaching level, scientific research strength, and faculty quality; the school category ranking is based on the classification standards of universities by the education authorities and is obtained from comparisons of similar universities; the first-level discipline ranking is taken from the discipline evaluation report published by authoritative academic organizations or professional evaluation institutions in various disciplines, and its evaluation considers key factors such as the discipline's scientific research achievements, academic influence, and faculty allocation; the second-level discipline list is directly extracted from the discipline construction page of the university's official website or the detailed discipline information database filed by the education department to construct a discipline portrait .

[0046] S2. Steps for vector representation of the receiving university:

[0047] The vector representation of the receiving college is , where f is the mapping function constructed using deep learning technology, R e Indicates the comprehensive ranking of the universities under review, C e Indicates the category ranking of the universities receiving the review, S 1，e S represents the first-level discipline ranking of the reviewing university.2，e It represents the list of secondary disciplines of the universities under review. Specifically, a deep neural network is used, and its architecture includes an input layer, multiple hidden layers, and an output layer. The number of input layer nodes is consistent with the number of collected university features, and the number of output layer nodes is set according to the dimension of the vector space. During the training process, a large number of university data samples with accurate annotations are used, and the weights and biases of the neural network are continuously optimized through the back propagation algorithm, so that the generated vector can accurately represent the multi-dimensional characteristics of universities, greatly improving the accuracy and effectiveness of vector representation.

[0048] S3. Vector representation steps for universities submitting for review:

[0049] The vector representation of the universities submitting the review is , where R S Indicates the comprehensive ranking of the universities submitted for review, C S Indicates the category ranking of the universities submitted for review, S 1，S Indicates the first-level discipline ranking of the university submitted for review, S 2，S Represents the list of secondary disciplines of the universities submitting for review. The training and optimization methods of the mapping function f are the same as the vector representation steps of the receiving universities, ensuring that the vectors of the universities submitting for review can accurately reflect their characteristics and provide a basis for subsequent precise selection.

[0050] S4. Calculation steps of scoring weighting factors:

[0051] Setting the scoring weighting factors for the universities receiving the review , α, β, γ are weight coefficients. The weight coefficients are determined by rigorously constructing a pairwise comparison judgment matrix, and the relative importance of each indicator is accurately judged based on the in-depth opinions of experts in the professional field or a large amount of historical data. For example, if experts determine that the importance of the number of reviews relative to the total review time is 3 based on experience, then the corresponding position in the judgment matrix is ​​accurately assigned a value of 3. Each element of the judgment matrix is ​​normalized by dividing it by the sum of the column, and then the average of each row of the normalized matrix is ​​obtained to obtain the weight. At the same time, the consistency ratio CR is strictly calculated, where , , CI represents the consistency index, is the maximum eigenvalue of the judgment matrix, and n is the order of the judgment matrix, which ensures the high consistency of the judgment matrix. If the consistency test fails to meet the standard, the hierarchical analysis method is used to carefully adjust the judgment matrix until it fully meets the consistency requirements. The number of reviews is accurately counted from the detailed historical records of the paper submission system, which can truly reflect the active participation of the reviewers; the total review time is accurately calculated by accurately recording the submission start time and the review result return time of each paper, and the average review time of a single paper is obtained by combining the number of reviews, which effectively reflects the review speed and efficiency; the number of blacklist experts is accurately counted from the strictly managed expert database. The number of experts marked as unqualified directly affects the reputation of colleges and universities.

[0052] S5. Vector similarity calculation steps:

[0053] The cosine similarity is used to calculate the similarity between the submitting university and the receiving university. , the calculation results are normalized so that their numerical range is strictly limited to between 0 and 1, which is convenient for the subsequent accurate calculation and scientific comparison of the comprehensive score. represents the similarity between the submitted university vector and the received university vector. The specific normalization operation is as follows: Let the original cosine similarity calculation result be x, and the normalized result be , where min and max are the minimum and maximum values ​​of all similarity calculation results, respectively, to ensure the accuracy and effectiveness of normalization.

[0054] S6. Comprehensive scoring model steps:

[0055] Combine the recommendation index with the weighting factor to form a comprehensive score To prevent outliers from causing excessive interference to the comprehensive score, the weighting factor W e Reasonable upper and lower limits are set to ensure the stability and reliability of the scoring results. For example, based on extensive historical data analysis, W e The upper limit is 10 and the lower limit is 0. When the calculation result exceeds this range, the boundary value is taken as the final weighting factor to ensure the scientific nature of the scoring model.

[0056] S7, threshold setting and sorting steps:

[0057] Set a threshold and select universities with comprehensive scores higher than the threshold ,in represents the set of recommended universities, h i represents a certain university in the university set H, S e，i Indicates college h i The threshold T is adjusted dynamically and flexibly according to the detailed distribution of historical submission data and actual specific needs. The adjustment method is as follows: In-depth analysis of the distribution curve of the comprehensive score in the historical submission data, and accurate determination of the threshold position according to different selection goals (such as focusing on high accuracy or high coverage). If high accuracy is pursued, the threshold is set at the high segment of the distribution curve; if high coverage is preferred, the threshold is appropriately lowered. At the same time, combined with the detailed feedback information of the actual submission task, such as the actual review quality and matching degree of the recommended universities, the threshold is fine-tuned to ensure the high quality of the selection results.

[0058] The present invention also includes a feedback optimization step: comprehensively collect detailed feedback information from reviewers, which covers the accurate evaluation of the matching degree between the paper and the subject of the receiving university, the detailed evaluation of the review quality, the comprehensive evaluation of the rationality of the review time, etc. According to the feedback results, the feature weights and recommendation algorithms are scientifically adjusted. If the feedback evaluation corresponding to a certain feature is generally not ideal, the weight of the feature in the model is appropriately reduced; if it is found that the recommendation algorithm produces unreasonable recommendations under specific circumstances, the algorithm is optimized and improved in a targeted manner to continuously improve the model performance and improve the accuracy and effectiveness of the selection.

[0059] The present invention also includes a model evaluation step: regularly use a carefully designed test data set to comprehensively evaluate the model. The test data set contains widely representative paper samples and university samples, and the distribution of the samples is highly similar to the actual complex review scenario. Key indicators such as accuracy and recall rate are used to accurately measure model performance. The accuracy rate is calculated as the exact ratio of the number of accurately recommended universities to the total number of recommended universities, and the recall rate is calculated as the exact ratio of the number of accurately recommended universities to the actual number of universities that should be recommended. According to the evaluation results, model defects are discovered and effectively improved in a timely manner. If the accuracy rate is found to be low, an in-depth analysis of the reasons may be unreasonable feature selection or loopholes in the algorithm, and then the features are adjusted or the algorithm is optimized in a targeted manner to ensure that the model always maintains good performance.

[0060] The present invention also includes a data preprocessing step: comprehensive cleaning, efficient denoising and standardization of the collected feature data. During the cleaning process, duplicate records, obvious error values ​​(such as illogical ranking data) and missing values ​​in the data are strictly removed (using reasonable mean filling or other applicable methods). De-noising uses advanced data smoothing algorithms, such as the moving average method, to effectively reduce random fluctuations in the data. Standardization maps data with different characteristics to specific reasonable intervals, such as using normalization methods to convert data to intervals or using standardization methods to make data conform to normal distribution, significantly improving data quality and laying a solid and reliable foundation for model construction.

[0061] In the present invention, in the steps of data collection and university subject portrait construction, the feature data is updated regularly, and the update cycle is determined according to the update frequency of the data source to ensure the timeliness of the data so that the model can make accurate selections based on the latest information.

[0062] In the present invention, in the step of vector representation of universities for review and the step of vector representation of universities for review, regularization technology is introduced to prevent overfitting of the model, improve the generalization ability of the model, and ensure that the characteristic vectors of universities can be accurately represented under different data conditions.

[0063] In the present invention, in the scoring weighting factor calculation step, the weight coefficients α, β, and γ are dynamically adjusted, and the values ​​of each coefficient are adjusted according to the characteristics of different subject areas or paper types to make the scoring weighting factor more in line with actual needs.

[0064] The method for constructing an automatic selection model for paper submission based on multi-dimensional feature analysis is characterized in that, in the vector similarity calculation step, an improved cosine similarity algorithm is used to consider the correlation between features, so as to further improve the accuracy of similarity calculation, thereby improving the overall performance of the selection model.

[0065] The above are only preferred specific implementation modes of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and inventive concepts of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for constructing an automatic selection model for paper submission based on multi-dimensional feature analysis, characterized in that: The following steps are involved: S1. Data collection and university subject portrait construction steps: Set college collection , through multiple channels for colleges and universities h1-h n Collect comprehensive rankings , School Category Ranking , first-level discipline ranking , List of Secondary Disciplines Feature data; build subject portraits ; S2. Steps for vector representation of the receiving university: The vector representation of the receiving college is , where f is the mapping function constructed using deep learning technology, R e Indicates the comprehensive ranking of the universities under review, C e Indicates the category ranking of the universities receiving the review, S 1，e S represents the first-level discipline ranking of the reviewing university. 2，e It represents the list of secondary disciplines of the universities under review, uses the university data samples with accurate annotations, and continuously optimizes the weights and biases of the neural network through the back propagation algorithm, so that the generated vectors represent the multi-dimensional characteristics of the universities; S3. Vector representation steps for universities submitting for review: The vector representation of the universities submitting the review is , where V S Indicates the comprehensive ranking of the universities submitted for review, C S Indicates the category ranking of the universities submitted for review, S 1，S Indicates the first-level discipline ranking of the university submitted for review, S 2，S Represents the secondary discipline list of the universities submitting for review. The training and optimization methods of the mapping function f are the same as the vector representation steps of the universities receiving the review. S4. Calculation steps of scoring weighting factors: Setting the scoring weighting factors for the universities receiving the review , α, β, γ are weight coefficients, and the weight coefficients are determined by building a comparison judgment matrix; Each element of the judgment matrix is ​​normalized by dividing it by the sum of its column, and then the weight is obtained by averaging each row of the normalized matrix. At the same time, the consistency ratio CR is calculated, where , , CI represents the consistency index, is the maximum eigenvalue of the judgment matrix, and n is the order of the judgment matrix; S5. Vector similarity calculation steps: The cosine similarity is used to calculate the similarity between the submitting university and the receiving university. , the calculation results are normalized, where It represents the similarity between the submitted university vector and the received university vector. The specific normalization operation is as follows: Let the original cosine similarity calculation result be x, and the normalized result be , where min and max are the minimum and maximum values ​​of all similarity calculation results respectively; S6. Comprehensive scoring model steps: Combine the recommendation index with the weighting factor to form a comprehensive score , for the weighting factor W e Reasonably set the upper and lower limits, based on historical data analysis, set W e The upper limit is 10 and the lower limit is 0. When the calculation result exceeds this range, the boundary value is taken as the final weighting factor.

2. The method for constructing an automatic selection model for paper submission based on multi-dimensional feature analysis according to claim 1 is characterized in that: Also includes: Threshold setting and sorting steps: Set the threshold and select universities with comprehensive scores higher than the threshold ,in represents the set of recommended universities, h i represents a certain university in the university set H, S e，i Indicates college h i The comprehensive score of the review data is dynamically adjusted based on the detailed distribution of historical review data and actual specific needs. Feedback optimization steps: Collect feedback information from reviewers, including the evaluation of the degree of match between the paper and the discipline of the receiving university, the evaluation of the review quality, and a comprehensive evaluation of the rationality of the review time; adjust the feature weights and recommendation algorithm based on the feedback results. If the feedback evaluation corresponding to a certain feature is lower than the threshold, appropriately reduce the weight of the feature in the model; if it is found that the recommendation algorithm produces unreasonable recommendations, perform targeted optimization and improvement on the algorithm.

3. The method for constructing an automatic selection model for paper submission based on multi-dimensional feature analysis according to claim 1 is characterized in that: It also includes a model evaluation step: using a test data set to evaluate the model, the test data set contains paper samples and university samples, and the distribution of the samples is consistent with the actual review scenario; using accuracy and recall rate indicators to measure model performance, the accuracy is calculated as the exact ratio of the number of accurately recommended universities to the total number of recommended universities, and the recall rate is calculated as the accurate ratio of the number of accurately recommended universities to the actual number of universities that should be recommended; according to the evaluation results, timely discover and improve model defects. If the accuracy is found to be lower than the threshold, adjust the features or optimize the algorithm.

4. The method for constructing an automatic selection model for paper submission based on multi-dimensional feature analysis according to claim 1 is characterized in that: It also includes data preprocessing steps: comprehensive cleaning, efficient denoising and standardization of the collected feature data; during the cleaning process, duplicate records, erroneous values ​​and missing values ​​in the data are removed; denoising processing uses data smoothing algorithms, specifically the moving average method, to reduce random fluctuations in the data; standardization processing maps data with different characteristics to preset intervals, including using normalization methods to convert data to intervals or using standardization methods to make the data conform to the normal distribution.

5. The method for constructing an automatic selection model for paper submission based on multi-dimensional feature analysis according to claim 1 is characterized in that: During the steps of data collection and university subject profile construction, feature data are updated regularly, and the update cycle is determined based on the update frequency of the data source.

6. The method for constructing an automatic selection model for paper submission based on multi-dimensional feature analysis according to claim 1 is characterized in that: In the steps of vector representation of receiving universities and submitting universities for review, regularization technology is introduced to prevent model overfitting and improve the generalization ability of the model.

7. The method for constructing an automatic selection model for paper submission based on multi-dimensional feature analysis according to claim 1 is characterized in that: In the scoring weighting factor calculation step, the weight coefficients α, β, and γ are dynamically adjusted, and the values ​​of each coefficient are adjusted according to the characteristics of different subject areas or paper types.

8. The method for constructing an automatic selection model for paper submission based on multi-dimensional feature analysis according to claim 1 is characterized in that: In the vector similarity calculation step, an improved cosine similarity algorithm is used to consider the correlation between features and improve the accuracy of similarity calculation.

Images