Academic ability assessment method based on sequence modeling

Abstract

The invention discloses an academic ability assessment method based on sequence modeling, which belongs to the technical field of academic assessment of scholars and comprises the steps of extracting potential factors, creating an academic ability assessment model based on Transform and outputting an academic ability assessment result of the scholars. According to the method, the academic ability of the scholar is evaluated by considering the decision factor of academic career success of the scholar and modeling the academic career of the scholar based on Transform, so that the deviation caused by single-dimension academic ability evaluation of the scholar is avoided, the growth trajectory of the scholar is represented based on the time sequence, the dynamic attribute of the academic career of the scholar is mined, and the academic ability of the scholar is evaluated. The neglect of dynamic attributes in a scholar growth track caused by only paying attention to static attributes of scholars is avoided, a rewarding and evaluation historical record is used as priori knowledge for evaluating the scholars, the defect that an existing evaluation method cannot utilize the existing priori knowledge accumulated by the scholars for comprehensive evaluation is overcome, scholar quantitative indexes which are easier to understand are quantitatively output, and the evaluation efficiency of the scholars is improved. And the academic ability of the scholar can be evaluated more accurately and reasonably.

Description

Technical Field

[0001] The invention relates to an academic ability evaluation method, in particular to an academic ability evaluation method based on sequence modeling, and belongs to the technical field of academic evaluation of scholars. Background Art

[0002] Scholars' academic evaluation ability can not only help scholars develop themselves better, but also guide employers to scientifically evaluate and manage talents. In recent years, many researchers have explored the impact of various factors on academic careers, such as co-author networks and citations. However, scholars' academic careers are affected by various confounding factors. It is very challenging to model scholars' academic careers with quantitative and systematic methods. There has been no unified definition and standard for quantifying scholars' academic ability. For this reason, researchers have proposed some efficient evaluation methods. Among them, there are two main widely recognized evaluation methods based on papers. One is the evaluation method based on the number of citations of papers, such as h-index, i10-index and the impact factor of the journal in which the paper is published. The other is the number and speed of scholars' publication of papers. In addition, some technical studies focus on the attributes of scholars such as their collaborator network, work unit, gender, etc., to evaluate the impact of scholars' academic ability and career development. However, such evaluation methods have the following shortcomings: (1) They only evaluate scholars' academic level from a single perspective, such as the number of publications, citations, and cooperation ability, ignoring the interactions between the above factors and failing to scientifically consider the impact of the above factors on academic ability; (2) Existing evaluation methods often only focus on the static attributes of scholars and ignore the dynamic attributes of scholars in their growth trajectory; (3) Existing evaluation methods fail to utilize the existing prior knowledge accumulated by scholars, such as historical records of professional title evaluation and academic honor selection. Based on the above shortcomings, there is currently a lack of methods for systematically evaluating scholars' academic level based on their academic careers. Therefore, the present invention proposes an academic ability evaluation method based on sequence modeling. The method can take into account the determinants of the success of scholars' academic careers, evaluate scholars' academic abilities based on Transformer modeling of scholars' academic careers, avoid the deviation caused by a single dimension in evaluating scholars' academic abilities, characterize scholars' growth trajectories based on time series, and explore the dynamic attributes of scholars' careers, avoid ignoring the dynamic attributes in scholars' growth trajectories by focusing only on scholars' static attributes, and use historical records of awards and honors as prior knowledge for evaluating scholars, thus solving the problem that existing evaluation methods cannot use the existing prior knowledge accumulated by scholars for comprehensive evaluation, and quantitatively output scholars' quantitative indicators that are easier to understand, so as to make a more accurate and reasonable evaluation of scholars' academic abilities. Summary of the invention

[0003] The main purpose of the present invention is to solve the problem that the existing academic ability assessment of scholars only assesses the academic level of scholars from a single perspective, such as the number of papers published, the number of citations, the ability to cooperate, etc., ignores the interaction between the above factors, fails to scientifically consider the impact of the above factors on academic ability, only focuses on the static attributes of scholars, ignores the dynamic attributes of scholars in their growth trajectory, and fails to utilize the existing prior knowledge accumulated by scholars, such as the shortcomings of historical records such as title evaluation and academic honor selection, and provides an academic ability assessment method based on sequence modeling.

[0004] The purpose of the present invention can be achieved by adopting the following technical solutions:

[0005] An academic ability assessment method based on sequence modeling includes the following steps:

[0006] S1. Extraction of latent factors: Use the collection of papers of scholars and their authors to perform text classification based on BERT to obtain the research field of scholars, use the total number of papers published by scholars and the growth rate of papers published to obtain the academic productivity of scholars, obtain the academic influence of scholars based on h-index, i10-index and citation volume, analyze the collaborator network of scholars based on node2vec algorithm to obtain the academic distance of scholars, use graph convolutional neural network to analyze the collaborator network of scholars to obtain the academic circle of scholars, and then combine the gender of scholars and work units to form the time series vector of eight dimensions of academic age, academic productivity, academic influence, academic distance, academic circle, research field, work unit and gender of scholars. This group of time series vectors is;

[0007] S2. Create an academic ability assessment model based on Transformer: For a recipient of an academic honor, obtain their academic age, and assess the academic ability of the scholar based on the time difference between the recipients of the academic honor. Given a scholar, compare the similarity between him and the scholars who have already received the academic honor, and assess the academic ability of the scholar based on the similarity of the academic honor recipients. Based on these two scholar assessment standards, create a scholar assessment model based on Transformer, take the scholar's career data as input, use the results obtained from the regression task to predict the number of years it takes for the scholar to obtain a certain academic honor, and use the results obtained from the classification task to assess the similarity between the scholar and the scholars who have already received the academic honor.

[0008] S3. Output the academic ability assessment results of scholars.

[0009] As a further method of the present invention, the academic age in S1 is related to the time of the scholar's scientific research career, which is used to reflect the scholar's growth potential and ability to maintain creativity. The academic age is the time from the publication of the scholar's first public academic paper to the evaluation year. The expression of academic age is:

[0010] t i =ey i -sy i ;

[0011] Where: i is the scholar number, t i is the academic age of scholar i, ey i is the year in which scholar i was evaluated, sy i The year when scholar i published his first public academic paper.

[0012] As a further method of the present invention, the academic productivity in S1 is the cumulative output of academic achievements each year in the academic career of the scholar. For a given scholar i and his academic age t i , academic productivity is related to the annual number of papers published, the average annual number of papers published, the cumulative number of papers published, and the growth rate of papers published. When the evaluation year is t, the annual number of papers published is the number of papers published by scholar i in the tth year, and the cumulative number of papers published is the number of papers published by scholar i from the first publication to the tth year. i The cumulative number of papers published in a year. The formula for the cumulative number of papers published is:

[0013]

[0014] Where: N i (t i ) is the cumulative number of papers published by scholars, n i (t) is the number of papers published by scholar i in year t;

[0015] The growth rate of publications and the cumulative number of publications N of scholar i i (t i ) and the assessment year is subject to Among them, α i is the growth rate of scholar i’s publications.

[0016] As a further method of the present invention, the quantitative indicators of academic influence in S1 are the total citations, annual average citations, citation growth rate, h-index and i10-index of scholar i in year t, where the total citations are c i (t), the average annual citation volume is The citation growth rate is

[0017] As a further method of the present invention, the academic distance in S1 is the influence of the scholar's social ability on the development of his academic career. The academic distance is the ratio of the sum of the vector cosines represented by the low-dimensional vectors of scholar i and cooperating scholar j to the number of cooperating scholars. The formula of academic distance is:

[0018]

[0019] In the formula: sdi is the academic distance, (x i ,x j ) is the low-dimensional vector representation of scholar i and collaborating scholar j calculated in the scholar collaborator network based on the node2vec algorithm, and N is the number of high-impact scholars in the collaborator network. High-impact scholars are scholars whose total citation volume is higher than a certain set threshold and scholars with senior academic titles such as academician and fellow.

[0020] As a further method of the present invention, the academic circle in S1 uses a graph convolutional neural network to extract the characteristics of the scholar's cooperation network structure, and establishes a collaborator network subgraph for each scholar i at his academic age t, wherein each node of the collaborator network subgraph is a scholar, each edge of the collaborator network subgraph is a cooperative relationship between scholars, and the weight of the edge is the number of collaborations between scholars. A two-layer graph convolutional neural network is used to learn the low-dimensional vector representation of scholar i in the collaborator network subgraph, and the low-dimensional vector representation is cn i (t), is the academic circle attribute of scholar i in the evaluation year t.

[0021] As a further method of the present invention, the research field in S1 takes into account the differences in the research fields of scholars. For a given scholar i, a vector is calculated using his paper output in t years as his research field. This vector is rf i (t), vector rf i (t) is obtained by encoding the text information of scholar i’s paper using a BERT-based text classifier. The vector rf i The formula for (t) is:

[0022]

[0023] Where: M is the total number of papers by scholar i, z is the paper number of scholar i, P z is the text information of the paper numbered z by scholar i, Field_Cls(P z ) is the result of the text classifier classifying the paper numbered z. The text classifier is trained based on the papers and their field classification label data in Microsoft Academic.

[0024] As a further method of the present invention, the employment information of scholar i in S1 is represented by four-dimensional vector embedding, the employment information is employ_emb, the gender of scholar i is a one-dimensional vector, the gender of scholar i is Gender, when Gender = (1), the gender of scholar i is male, when Gender = (0), the gender of scholar i is female.

[0025] As a further method of the present invention, the time series vector of scholar i in S1 is used to characterize the academic characteristics of scholar i from the first publication to the evaluation year. The time series vector of scholar i is expressed as:

[0026]

[0027] Where: X i is the time series vector representation of scholar i, is the eight-dimensional academic ability vector representation of scholar i’s first published paper, An eight-dimensional vector representation of the academic ability of the scholar in the year he or she was evaluated.

[0028] As a further method of the present invention, in S2, the classification task selects cross entropy as the loss function during training, and the regression task adopts MSE as the loss function during training.

[0029] The beneficial technical effects of the present invention are:

[0030] (1) Compared with the academic ability assessment based on a single dimension, the method provided by the present invention integrates various attributes of scholars and avoids the bias caused by the academic ability assessment of scholars based on a single dimension;

[0031] (2) Characterizing the growth trajectory of scholars based on time series can fully explore the dynamics of scholars' academic careers;

[0032] (3) Use the historical records of awards and honors as prior knowledge for evaluating scholars;

[0033] (4) Combining graph neural network technology and text classifiers, it is possible to capture potential factors such as academic age, academic productivity, academic influence, academic distance, academic circle, research field, work unit, and gender. It is able to consider the determinants of a scholar's academic career success and combine Transformer modeling to quantitatively output scholars' academic abilities into easier-to-understand quantitative indicators, thereby making a more accurate and reasonable assessment of scholars' academic abilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic diagram of an evaluation model for the academic ability evaluation method based on sequence modeling according to the present invention. DETAILED DESCRIPTION

[0035] In order to make the technical solution of the present invention more clear and specific to those skilled in the art, the present invention is further described in detail below in conjunction with embodiments and drawings, but the implementation manner of the present invention is not limited thereto.

[0036] like Figure 1 As shown, the academic ability assessment method based on sequence modeling provided in this embodiment includes the following steps:

[0037] S1. Extraction of latent factors: Use the collection of papers of scholars and their authors to perform text classification based on BERT to obtain the research field of scholars, use the total number of papers published by scholars and the growth rate of papers published to obtain the academic productivity of scholars, obtain the academic influence of scholars based on h-index, i10-index and citation volume, analyze the collaborator network of scholars based on node2vec algorithm to obtain the academic distance of scholars, use graph convolutional neural network to analyze the collaborator network of scholars to obtain the academic circle of scholars, and then combine the gender of scholars and work units to form the time series vector of eight dimensions of academic age, academic productivity, academic influence, academic distance, academic circle, research field, work unit and gender of scholars. This group of time series vectors is;

[0038] S2. Create an academic ability assessment model based on Transformer: For a recipient of an academic honor, obtain their academic age, and assess the academic ability of the scholar based on the time difference between the recipients of the academic honor. Given a scholar, compare the similarity between him and the scholars who have already received the academic honor, and assess the academic ability of the scholar based on the similarity of the academic honor recipients. Based on these two scholar assessment standards, create a scholar assessment model based on Transformer, take the scholar's career data as input, use the results obtained from the regression task to predict the number of years it takes for the scholar to obtain a certain academic honor, and use the results obtained from the classification task to assess the similarity between the scholar and the scholars who have already received the academic honor.

[0039] S3. Output the academic ability assessment results of scholars.

[0040] The beneficial technical effects of the academic ability assessment method based on sequence modeling proposed by the present invention are:

[0041] (1) Compared with the academic ability assessment based on a single dimension, the method provided by the present invention integrates various attributes of scholars and avoids the bias caused by the academic ability assessment of scholars based on a single dimension;

[0042] (2) Characterizing the growth trajectory of scholars based on time series can fully explore the dynamics of scholars' academic careers;

[0043] (3) Use the historical records of awards and honors as prior knowledge for evaluating scholars;

[0044] (4) Combined with graph neural network technology and text classifier, the potential factors of academic age, academic productivity, academic influence, academic distance, academic circle, research field, work unit and gender can be captured, and the determinants of the success of scholars' academic careers can be considered. Combined with Transformer modeling, the academic ability of scholars can be quantitatively output to more easily understand the quantitative indicators of scholars, and the academic ability of scholars can be more accurately and reasonably evaluated. The method of evaluating the academic ability of scholars in the present invention can be used for the display of academic websites, and the comprehensive academic ability evaluation of scholars can be displayed on the homepage of scholars; for different research fields, according to the quantitative evaluation of scholars, a ranking list of scholars in the field is generated to realize the expert recommendation function; auxiliary institutions and employing units evaluate the comprehensive ability of academic talents, which can be applied to talent recruitment, promotion and award and excellence evaluation; it can realize the self-evaluation of scholars and help scholars realize the scientific improvement of their own academic ability.

[0045] The academic age in S1 is related to the time they have been engaged in scientific research, and is used to reflect the scholar's growth potential and ability to maintain creativity. The academic age is the time from the publication of the scholar's first public academic paper to the evaluation year. The expression of academic age is:

[0046] t i =ey i -sy i ;

[0047] Where: i is the scholar number, t i is the academic age of scholar i, ey i is the year in which scholar i was evaluated, sy i The year when scholar i published his first public academic paper.

[0048] The academic productivity in S1 is the cumulative output of academic achievements each year in the academic career of the scholar. For a given scholar i and his academic age t i , academic productivity is related to the annual number of papers published, the average annual number of papers published, the cumulative number of papers published, and the growth rate of papers published. When the evaluation year is t, the annual number of papers published is the number of papers published by scholar i in the tth year, and the cumulative number of papers published is the number of papers published by scholar i from the first publication to the tth year. i The cumulative number of papers published in a year. The formula for the cumulative number of papers published is:

[0049]

[0050] Where: N i (t i ) is the cumulative number of publications by scholars, n i (t) is the number of papers published by scholar i in year t;

[0051] The growth rate of publications and the cumulative number of publications N of scholar ii (t i ) and the assessment year is subject to Among them, α i is the growth rate of scholar i’s publications.

[0052] The quantitative indicators of academic influence in S1 are the total citations, annual average citations, citation growth rate, h-index and i10-index of scholar i in year t, where the total citations are c i (t), the average annual citation volume is The citation growth rate is

[0053] The academic distance in S1 is the impact of the scholar's social ability on his academic career development. The academic distance is the ratio of the sum of the vector cosines represented by the low-dimensional vectors of scholar i and cooperating scholar j to the number of cooperating scholars. The formula of academic distance is:

[0054]

[0055] In the formula: sd i is the academic distance, (x i ,x j ) is the low-dimensional vector representation of scholar i and collaborating scholar j calculated in the scholar collaborator network based on the node2vec algorithm, and N is the number of high-impact scholars in the collaborator network. High-impact scholars are scholars whose total citation volume is higher than a certain set threshold and scholars with senior academic titles such as academician and fellow.

[0056] For example, when a scholar is elected as an ACM / IEEE Fellow or academician, he or she needs to be recommended by fellows or academicians in related research fields. Collaboration between scholars with high-impact scholars is conducive to the improvement of their academic level. The academic distance is defined as the average distance between scholar i and high-impact scholars in the collaborator network. The smaller the distance, the more advantages the scholar has in his or her cooperative attributes, which describes the closeness of the cooperation between scholars and high-impact scholars.

[0057] The academic circle in S1 uses a graph convolutional neural network to extract the structural features of the scholar's cooperation network, and establishes a collaborator network subgraph for each scholar i at his academic age t, where each node of the collaborator network subgraph is a scholar, each edge of the collaborator network subgraph is a cooperative relationship between scholars, and the weight of the edge is the number of collaborations between scholars. A two-layer graph convolutional neural network is used to learn the low-dimensional vector representation of scholar i in the collaborator network subgraph, and the low-dimensional vector representation is cn i (t), is the academic circle attribute of scholar i in the evaluation year t.

[0058] Although academic distance describes the degree of cooperation between scholars and high-impact scholars, other rich structural features in the collaborator network have not been reused. Therefore, the present invention uses graph convolutional neural networks to extract the structural features of scholars' cooperation networks and enrich the use of collaborator network structures.

[0059] The research field in S1 takes into account the differences in the research fields of scholars. For a given scholar i, a vector is calculated using his paper output in t years as his research field. This vector is rf i (t), vector rf i (t) is obtained by encoding the text information of scholar i’s paper using a BERT-based text classifier. The vector rf i The formula for (t) is:

[0060]

[0061] Where: M is the total number of papers by scholar i, z is the paper number of scholar i, P z is the text information of the paper numbered z by scholar i, Field_Cls(P z ) is the result of the text classifier classifying the paper numbered z. The text classifier is trained based on the papers and their field classification label data in Microsoft Academic.

[0062] The employment information of scholar i in S1 is represented by four-dimensional vector embedding, where the employment information is employ_emb, and the gender of scholar i is a one-dimensional vector, where the gender of scholar i is Gender. When Gender = (1), the gender of scholar i is male, and when Gender = (0), the gender of scholar i is female.

[0063] The time series vector of scholar i in S1 is used to characterize the academic characteristics of scholar i from the first publication to the evaluation year. The time series vector of scholar i is expressed as:

[0064]

[0065] Where: X i is the time series vector representation of scholar i, is the eight-dimensional academic ability vector representation of scholar i’s first published paper, An eight-dimensional vector representation of the academic ability of the scholar in the year he or she was evaluated.

[0066] In S2, the classification task selects cross entropy as the loss function during training, and the regression task uses MSE as the loss function during training.

[0067] The Transformer-based scholar evaluation model uses the career data of scholars as input to meet the following requirements: (1) In the regression task, it can predict how many years it will take for scholars to obtain a certain academic honor, and effectively evaluate the academic ability of scholars; (2) Through the classification task, it can compare the similarity between scholars and scholars who have already received academic honors, which can facilitate the observation of the level and level of the academic ability of scholars.

[0068] When performing classification tasks, taking the election of scholars as IEEE Fellows as an example, the scholar's eight-dimensional time series vector is input into the evaluation model to output the probability of the scholar being elected as IEEE Fellow, that is, the similarity between the scholar and scholars who have been elected as IEEE Fellows. The higher the similarity, the stronger the individual's academic ability. During training, the training data uses the data of scholars who have been elected as IEEE Fellows as positive examples in the classification task, non-IEEE Fellow scholars as negative examples, and the calendar year as a variable for training; when setting the regression task, the eight-dimensional matrix is ​​input into the evaluation model, and the model outputs how many years it takes for the scholar to be successfully selected as an IEEE Fellow. The smaller the year value, the stronger the scholar's academic ability. During training, the data of scholars who have been elected as IEEE Fellows is used, and the model is trained with MSE as the loss function until the model converges.

[0069] In summary, in this embodiment, the method for evaluating the academic ability of scholars in the present invention can be used for the display of academic websites, and the comprehensive academic ability evaluation is displayed on the homepage of scholars; for different research fields, according to the quantitative evaluation of scholars, a ranking list of scholars in the field is generated to realize the expert recommendation function; auxiliary institutions and employing units evaluate the comprehensive ability of academic talents, which can be applied to talent recruitment, promotion and award and evaluation links; it can realize the self-evaluation of scholars and help scholars realize the scientific improvement of their own academic ability. For example, when scholars are elected as ACM / IEEE Fellows or academicians, they need Fellows or academicians in related research fields to recommend them. The cooperation between scholars and high-impact scholars is conducive to the improvement of their academic level. The academic distance is defined as the average distance between scholar i and high-impact scholars in the collaborator network. The smaller the distance, the more advantages scholars have in their cooperative attributes, which describes the closeness of cooperation between scholars and high-impact scholars. Although the academic distance describes the closeness of cooperation between scholars and high-impact scholars, other rich structural features in the collaborator network have not been reused. Therefore, the present invention uses a graph convolutional neural network to extract the structural features of the scholar's cooperation network and enrich the use of the collaborator network structure. The Transformer-based scholar evaluation model uses the career data of scholars as input to meet the following requirements: (1) In the regression task, it can predict how many years it will take for scholars to obtain a certain academic honor, and effectively evaluate the academic ability of scholars; (2) Through the classification task, it can compare the similarity between scholars and scholars who have already received academic honors, which can facilitate the observation of the level and level of the academic ability of scholars. When performing classification tasks, taking the election of scholars as IEEE Fellows as an example, the scholar's eight-dimensional time series vector is input into the evaluation model to output the probability of the scholar being elected as IEEE Fellow, that is, the similarity between the scholar and scholars who have been elected as IEEE Fellows. The higher the similarity, the stronger the individual's academic ability. During training, the training data uses the data of scholars who have been elected as IEEE Fellows as positive examples in the classification task, non-IEEE Fellow scholars as negative examples, and the calendar year as a variable for training; when setting the regression task, the eight-dimensional matrix is ​​input into the evaluation model, and the model outputs how many years it takes for the scholar to be successfully selected as an IEEE Fellow. The smaller the year value, the stronger the scholar's academic ability. During training, the data of scholars who have been elected as IEEE Fellows is used, and the model is trained with MSE as the loss function until the model converges.

[0070] The above description is only a further embodiment 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 concepts of the present invention within the scope disclosed by the present invention, which belong to the protection scope of the present invention.

Claims

1. An academic ability assessment method based on sequence modeling, characterized in that: The steps include: S1. Extraction of latent factors: Use the collection of papers of scholars and their authors to perform text classification based on BERT to obtain the research field of scholars, use the total number of papers published by scholars and the growth rate of papers published to obtain the academic productivity of scholars, obtain the academic influence of scholars based on h-index, i10-index and citation volume, analyze the collaborator network of scholars based on node2vec algorithm to obtain the academic distance of scholars, use graph convolutional neural network to analyze the collaborator network of scholars to obtain the academic circle of scholars, and then combine the gender of scholars and work units to form the time series vector of eight dimensions of academic age, academic productivity, academic influence, academic distance, academic circle, research field, work unit and gender of scholars. This group of time series vectors is; S2. Create an academic ability assessment model based on Transformer: For a recipient of an academic honor, obtain their academic age, and assess the academic ability of the scholar based on the time difference between the recipients of the academic honor. Given a scholar, compare the similarity between him and the scholars who have already received the academic honor, and assess the academic ability of the scholar based on the similarity of the academic honor recipients. Based on these two scholar assessment standards, create a scholar assessment model based on Transformer, take the scholar's career data as input, use the results obtained from the regression task to predict the number of years it takes for the scholar to obtain a certain academic honor, and use the results obtained from the classification task to assess the similarity between the scholar and the scholars who have already received the academic honor. S3. Output the academic ability assessment results of scholars.

2. The method for evaluating academic ability based on sequence modeling as claimed in claim 1, characterized in that: The academic age in S1 is related to the time they have been engaged in scientific research, and is used to reflect the scholar's growth potential and ability to maintain creativity. The academic age is the time from the publication of the scholar's first public academic paper to the evaluation year. The expression of academic age is: t i =yes i -and i ; Where: i is the scholar number, t i is the academic age of scholar i, ey i is the year in which scholar i was evaluated, sy i The year when scholar i published his first public academic paper.

3. The method for evaluating academic ability based on sequence modeling as claimed in claim 2, characterized in that: The academic productivity in S1 is the cumulative output of academic achievements each year in the academic career of the scholar. For a given scholar i and his academic age t i , academic productivity is related to the annual number of papers published, the average annual number of papers published, the cumulative number of papers published, and the growth rate of papers published. When the evaluation year is t, the annual number of papers published is the number of papers published by scholar i in the tth year, and the cumulative number of papers published is the number of papers published by scholar i from the first publication to the tth year. i The cumulative number of papers published in a year. The formula for the cumulative number of papers published is: Where: N i (t i ) is the cumulative number of papers published by scholars, n i (t) is the number of papers published by scholar i in year t; The growth rate of publications and the cumulative number of publications N of scholar i i (t i ) and the assessment year is subject to Among them, α i is the growth rate of scholar i’s publications.

4. The method for evaluating academic ability based on sequence modeling as claimed in claim 2, characterized in that: The quantitative indicators of academic influence in S1 are the total citations, annual average citations, citation growth rate, h-index and i10-index of scholar i in year t, where the total citations are c i (t), the average annual citation volume is The citation growth rate is 5. The method for evaluating academic ability based on sequence modeling as claimed in claim 2, characterized in that: The academic distance in S1 is the impact of the scholar's social ability on his academic career development. The academic distance is the ratio of the sum of the vector cosines represented by the low-dimensional vectors of scholar i and cooperating scholar j to the number of cooperating scholars. The formula of academic distance is: In the formula: sd i is the academic distance, (x i ,x j ) is the low-dimensional vector representation of scholar i and collaborating scholar j calculated in the scholar collaborator network based on the node2vec algorithm, and N is the number of high-impact scholars in the collaborator network. High-impact scholars are scholars whose total citation volume is higher than a certain set threshold and scholars with senior academic titles such as academician and fellow.

6. The method for evaluating academic ability based on sequence modeling as claimed in claim 2, characterized in that: The academic circle in S1 uses a graph convolutional neural network to extract the structural features of the scholar's cooperation network, and establishes a collaborator network subgraph for each scholar i at his academic age t, where each node of the collaborator network subgraph is a scholar, each edge of the collaborator network subgraph is a cooperative relationship between scholars, and the weight of the edge is the number of collaborations between scholars. A two-layer graph convolutional neural network is used to learn the low-dimensional vector representation of scholar i in the collaborator network subgraph, and the low-dimensional vector representation is cn i (t), is the academic circle attribute of scholar i in the evaluation year t.

7. The method for evaluating academic ability based on sequence modeling as claimed in claim 2, characterized in that: The research field in S1 takes into account the differences in the research fields of scholars. For a given scholar i, a vector is calculated using his paper output in t years as his research field. This vector is rf i (t), vector rf i (t) is obtained by encoding the text information of scholar i’s paper using a BERT-based text classifier. The vector rf i The formula for (t) is: Where: M is the total number of papers by scholar i, z is the paper number of scholar i, P z is the text information of the paper numbered z by scholar i, Field_Cls(P z ) is the result of the text classifier classifying the paper numbered z. The text classifier is trained based on the papers and their field classification label data in Microsoft Academic.

8. The method for evaluating academic ability based on sequence modeling as claimed in claim 2, characterized in that: The employment information of scholar i in S1 is represented by four-dimensional vector embedding, where the employment information is employ_emb, and the gender of scholar i is a one-dimensional vector, where the gender of scholar i is Gender. When Gender = (1), the gender of scholar i is male, and when Gender = (0), the gender of scholar i is female.

9. The method for evaluating academic ability based on sequence modeling as claimed in claim 1, characterized in that: The time series vector of scholar i in S1 is used to characterize the academic characteristics of scholar i from the first publication to the evaluation year. The time series vector of scholar i is expressed as: Where: X i is the time series vector representation of scholar i, is the eight-dimensional academic ability vector representation of scholar i’s first published paper, An eight-dimensional vector representation of the academic ability of the scholar in the year he or she was evaluated.

10. The method for evaluating academic ability based on sequence modeling according to claim 1, characterized in that: In S2, the classification task selects cross entropy as the loss function during training, and the regression task uses MSE as the loss function during training.

Images