A method and system for allocating review tasks

Abstract

The present application provides a method and system for allocating review tasks. Among them, a target review task is received, and the research topic of the target review task is determined; reviewers matching the research topic are screened out; the review ability and response speed of the reviewers for the target review task are predicted, and the corresponding review suitability index is calculated according to the prediction results; reviewers whose review suitability index meets the predetermined conditions are selected to form a review group; an integrated learning strategy is used to fuse the review opinions of multiple reviewers corresponding to the review group to form a review conclusion for the target review task. The technical solution provided by the present application can calculate the corresponding review suitability index according to the predicted review ability and response speed of the reviewers for the target review task, so as to screen out reviewers suitable for the target review task, thereby improving the allocation efficiency of the review task.

Description

Technical Field

[0001] Embodiments of the present application relate to the field of computer technology, and in particular, to a method and system for allocating review tasks. Background Art

[0002] In the field of modern academic publishing, the review process is a key link to ensure research quality and academic integrity. With the increasing number of research results, the number of review tasks is also increasing continuously, which puts higher requirements on the efficiency and quality of the review process. The traditional manual method of allocating review tasks is not only time-consuming and laborious, but also prone to uneven review quality.

[0003] The currently commonly used manual allocation method mainly relies on the experience judgment of the editorial department. Usually, an editor manually selects a suitable reviewer according to the theme of the manuscript and the professional field of the reviewer.

[0004] However, although this method is intuitive and easy to understand, it is inefficient when dealing with a large number of manuscripts, and it is difficult to ensure that the most suitable candidate can be found every time, resulting in a low efficiency of review task allocation. Summary of the Invention

[0005] Embodiments of the present application provide a method and system for allocating review tasks to solve the problem of low efficiency in allocating review tasks in the prior art.

[0006] In a first aspect, embodiments of the present application provide a method for allocating review tasks, including:

[0007] Receiving a target review task and determining the research theme of the target review task;

[0008] In a pre-constructed reviewer database, screening out reviewers who match the research theme;

[0009] Based on the historical review records and current available time of the reviewers, predicting the review ability and response speed of the reviewers for the target review task through data analysis, and calculating the corresponding review suitability index according to the prediction results;

[0010] Selecting reviewers whose review suitability index meets the predetermined conditions to form a review group, and planning suitable review time periods for each reviewer in the review group through an intelligent time management algorithm, so that each reviewer can provide review opinions for the target review task at the corresponding review time;

[0011] Using an integrated learning strategy to fuse the review opinions of multiple reviewers corresponding to the review group to form a review conclusion for the target review task.

[0012] Optionally, the historical review records at least include: review time data and quality assessment data;

[0013] Predicting the review ability and response speed of the reviewer for the target review task through data analysis based on the historical review records and current available time of the reviewer, including:

[0014] Predicting the review response speed of the reviewer for the target review task according to the review time data in the historical review records of the reviewer and the current available time;

[0015] Predicting the review ability of the reviewer for the target review task based on the quality assessment data in the historical review records of the reviewer;

[0016] Taking the review response speed and the review ability as the prediction results;

[0017] Among them, through the following calculation formula, predicting the review response speed of the reviewer for the target review task:

[0018] ;

[0019] Among them, is the review time data in the historical review records of the reviewer; is the current available time of the reviewer; represents the expected value of the actual completion time and the current available time under the condition of the posterior distribution for the actual completion time ; is the posterior distribution calculated according to the historical review time and the current available time; is the expected completion time of the reviewer for the target review task, that is, the predicted value of the review response speed; the numerator is the combined influence of the review time and the current available time; the denominator is the result of integrating all possible review times multiplied by their corresponding current available times ;

[0020] Among them, through the following calculation formula, predicting the review ability of the reviewer for the target review task:

[0021] ;

[0022] Among them, is the feature vector of the reviewer; is the prediction function of the random forest model; are the parameters of the random forest model; is the review ability score for historical review tasks; is the number of samples in the training dataset; is the review ability score of the reviewer for the target review task; denotes finding a set of optimal parameters , such that the actual review ability scores of all training samples and the model prediction values have the minimum sum of squared differences.

[0023] Optionally, predicting the review response speed of the reviewer for the target review task based on the review time data in the reviewer's historical review records and the currently available time includes:

[0024] Construct a Bayesian statistical model based on the review time data in the reviewer's historical review records to describe the prior distribution of the review time;

[0025] Use the Bayesian update rule to update the prior distribution according to the latest review time data of the reviewer obtained to obtain the posterior distribution;

[0026] Based on the posterior distribution and combined with the currently available time of the reviewer, estimate the expected completion time of the reviewer for the target review task as the predicted value of the reviewer's review response speed.

[0027] Optionally, predicting the review ability of the reviewer for the target review task based on the quality assessment data in the reviewer's historical review records includes:

[0028] Collect the quality assessment data in the reviewer's historical review records, including the acceptance rate of the review comments, the detail level of the review reports, and the usefulness scores of the review comments;

[0029] Construct a training dataset according to the quality assessment data, where each piece of training data in the training dataset includes the feature vector of the reviewer and the review ability score for the historical review task, and the feature vector includes the reviewer's professional title, institution, professional field, previous review times, and average review time;

[0030] Train a random forest model, using the feature vectors in the training dataset as inputs and the corresponding review ability scores of the historical review tasks as labels to learn the mapping relationship between the review ability and the feature vectors;

[0031] Output the review ability score of the reviewer for the target review task using the trained random forest model, and use it as the result of predicting the review ability of the reviewer for the target review task.

[0032] Optionally, in the process of predicting the review ability and response speed of the reviewer for the target review task through data analysis, it further includes:

[0033] Use a deep neural network to model the behavior pattern of the obtained reviewer, and combine the obtained review preferences, historical behavior trajectories, and external influencing factors of the reviewer to refine the prediction of the review ability and the response speed, so as to update the prediction result;

[0034] Introduce a reinforcement learning algorithm, and dynamically adjust the parameters of the prediction model according to the completion of the historical review tasks by the reviewer at different time points, so as to improve the prediction accuracy of the prediction result. Among them, the adjusted parameters of the prediction model at least include the prior probability distribution parameters in the Bayesian statistical model, the hyperparameters in the random forest algorithm, the reward function, the discount factor in the reinforcement learning algorithm, and the weight and bias term parameters in the deep neural network.

[0035] Optionally, calculating the corresponding review suitability index according to the prediction result includes:

[0036] According to the current available time of the reviewer, use the time series analysis algorithm to predict the available time periods of the reviewer in a future period of time;

[0037] Calculate the review suitability index of the reviewer according to the prediction result and the available time period;

[0038] Among them, through the following calculation formula, calculate the review suitability index of the reviewer:

[0039] ;

[0040] Among them, represents the review suitability index; represents the number of reviewers; is the comprehensive weight factor of the reviewer, a score comprehensively determined based on the feature vector of the reviewer and the review ability score for historical review tasks; represents the expected completion time of the reviewer for the target review task, is a very small positive number used to prevent division by zero errors; is related to is the exponential factor related to , a non - linear coefficient determined based on the statistical distribution characteristics of the past response times of reviewers to emphasize the importance of the review response time; represents the predicted review ability score of the reviewer for the target review task; is related to the exponential factor, using the square root , used to adjust the influence degree of the review ability score; represents the evaluation value of the available time period of the reviewer for the target review task, which is a quantitative evaluation of the future available time period predicted based on the time - series analysis algorithm; is related to the exponential factor, using , used to adjust the importance of the available time period; represents the number of time conflicts of the reviewer with other review tasks in the future period, which is a negative indicator; is related to the exponential factor, using , used to adjust the influence degree of the number of conflicts;

[0041] Optionally, screening out reviewers matching the research topic in the pre - constructed reviewer database includes:

[0042] Extracting keywords and phrases in the target review task to determine the core elements of the research topic;

[0043] Converting the core elements into a structured data format and matching the core elements in the structured data format with the professional field information of the reviewers stored in the reviewer database to identify reviewers matching the research topic;

[0044] Based on the professional field matching degree, past review experience and review quality evaluation information of the reviewers, screening out the optimal list of reviewers;

[0045] Selecting eligible reviewers from the optimal list of reviewers.

[0046] Optionally, selecting reviewers whose review suitability index reaches a predetermined condition to form a review group and planning appropriate review time periods for each reviewer in the review group through an intelligent time - management algorithm includes:

[0047] Setting a predetermined threshold according to the review suitability index of the reviewers and screening out reviewers higher than the predetermined threshold;

[0048] Including the screened - out reviewers in the list of candidate review groups;

[0049] Analyze the current workload and current available time of each reviewer in the candidate reviewer group list using an intelligent time management algorithm;

[0050] Based on the current workload and current available time of each reviewer, combined with the available time periods of each reviewer in a future period of time, allocate the most suitable review time period for each reviewer through an optimization algorithm to ensure that each reviewer can focus on the review task of the target review task within the specified time period, so as to improve the review efficiency and quality.

[0051] Optionally, using an integrated learning strategy to fuse the review opinions of multiple reviewers corresponding to the review group to form a review conclusion for the target review task, including:

[0052] Collect the review opinions provided by each reviewer in the review group and convert the review opinions into standardized review scores;

[0053] Adopt an integrated learning framework and input the standardized review scores into an integrated model. The integrated model is composed of multiple basic models. Among them, the basic models at least include support vector machines, logistic regression, and neural networks, and each basic model independently evaluates the review scores;

[0054] Based on the evaluation results of multiple basic models, determine the final review conclusion through a weighted voting or averaging strategy. Among them, the weighted voting strategy assigns different weights according to the historical performance of each basic model to ensure the objectivity and fairness of the final review conclusion;

[0055] Among them, through the following calculation formula, form a review conclusion for the target review task:

[0056] ;

[0057] Among them, is the review score converted from the final review conclusion, is the standardized review score converted from the review opinions provided by each reviewer in the review group; is the prediction function of the th basic model; is the number of basic models; is the weight of the th basic model; is the adjustment factor; is the historical loss value of the th basic model;

[0058] In a second aspect, an embodiment of the present application provides a review task allocation system, including:

[0059] A determination module, configured to receive a target review task and determine the research topic of the target review task;

[0060] A screening module, configured to screen out reviewers matching the research topic in a pre-constructed reviewer database;

[0061] A prediction module, configured to predict the review ability and response speed of the reviewer based on the reviewer's historical review records and current available time, and calculate a corresponding review suitability index according to the prediction result;

[0062] A processing module, configured to select reviewers whose review suitability index reaches a predetermined condition to form a review group, and plan a suitable review time period for each reviewer in the review group through an intelligent time management algorithm, so that each reviewer provides review opinions on the target review task at the corresponding review time;

[0063] A generation module, configured to use an integrated learning strategy to fuse the review opinions of multiple reviewers corresponding to the review group to form a review conclusion for the target review task.

[0064] In an embodiment of the present application, a target review task is received, and the research topic of the target review task is determined; reviewers matching the research topic are screened out; the review ability and response speed of the reviewers for the target review task are predicted, and a corresponding review suitability index is calculated according to the prediction result; reviewers whose review suitability index reaches a predetermined condition are selected to form a review group; an integrated learning strategy is used to fuse the review opinions of multiple reviewers corresponding to the review group to form a review conclusion for the target review task. The technical solution provided by the present application can calculate a corresponding review suitability index according to the predicted review ability and response speed of the reviewers for the target review task, so as to screen out reviewers suitable for the target review task, thereby improving the allocation efficiency of the review task.

[0065] These aspects or other aspects of the present application will be more clearly understood in the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0067] Figure 1 Flowchart of a method for allocating review tasks provided by an embodiment of the present application;

[0068] Figure 2 Schematic structural diagram of a review task allocation system provided by an embodiment of the present application;

[0069] Figure 3 Schematic structural diagram of a computing device provided by an embodiment of the present application. Detailed implementation manners

[0070] In order to enable those skilled in the art of the present technology to better understand the solution of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application.

[0071] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0072] Figure 1 A flowchart of a method for allocating review tasks provided by an embodiment of the present application is as Figure 1 shown, and the method includes:

[0073] 101. Receive a target review task and determine the research topic of the target review task;

[0074] In this step, an editor or a system receives an article to be reviewed and clarifies the research field or specific topic to which the article belongs.

[0075] In an embodiment of the present application, assume that an editor receives a manuscript submitted to the Journal of Computer Vision and Pattern Recognition via email. The editor reads the abstract and keywords and confirms that the research belongs to the topic of "Application of Deep Learning in Image Recognition".

[0076] 102. Screen out reviewers in a pre-constructed reviewer database who match the research topic;

[0077] In this step, the established expert database is used to find suitable reviewers according to the research field.

[0078] In an embodiment of the present application, the system automatically screens out a list of reviewers marked with relevant expertise tags such as "deep learning" and "image processing" from a database containing information of thousands of scholars.

[0079] Optionally, the step of "screening out reviewers who match the research topic from the pre - constructed reviewer database" in step 102 includes: extracting keywords and phrases in the target review task to determine the core elements of the research topic; converting the core elements into a structured data format, and matching the core elements in the structured data format with the professional field information of the reviewers stored in the reviewer database to identify reviewers who match the research topic; screening out an optimal list of reviewers based on the professional field matching degree, past review experience, and review quality evaluation information of the reviewers; and selecting eligible reviewers from the optimal list of reviewers.

[0080] Keyword: A word that can summarize the main content or research focus of an article.

[0081] Phrase: A linguistic unit consisting of more than two words that expresses a specific meaning.

[0082] Core element: The most crucial part of the research topic refined by analyzing the text content.

[0083] Structured data format: Converting unstructured information into a form that is easy for computers to process.

[0084] Professional field information: Describing the specific disciplinary scope in which a reviewer specializes during their career.

[0085] Matching degree: An indicator measuring the correlation between a reviewer's professional background and the topic of the paper to be reviewed.

[0086] Review experience: Referring to how many times a reviewer has participated in similar types of review work in the past.

[0087] Review quality evaluation: A score for the quality of the feedback provided by a reviewer based on historical records.

[0088] First, use natural language processing technology to automatically identify the main words in the target review task document that can represent the research direction. Determine the core elements and convert them into structured data: further refine these words to form a concise research summary and encode it in a machine - readable form. Second, compare the data obtained above with the profiles of all reviewers stored in the database to find those who cover the knowledge required for the current research in their professional fields. Then consider the historical performance of each candidate (such as the number of reviews, the favorable review rate received, etc.) and select the most suitable candidates for this review work. Finally, make a final decision from the list generated in the previous step based on actual situations such as time arrangements and other factors.

[0089] In an embodiment of the present application, assume that there is a paper on "the application of artificial intelligence in medical image diagnosis" that needs to be reviewed. First, the system will automatically capture frequently mentioned terms in the text, such as "deep learning", "image recognition", "tumor detection", etc., as keywords. Then, after organizing this information, a short description is formed: "Explore how to improve the accuracy of medical image analysis using advanced AI technology". Next, this description is encoded and compared with the existing reviewer database, and several experts specializing in the development of health information technology are found to be very suitable. Finally, based on their past work experiences and evaluations given by their peers, three candidates with the best performance are selected. Considering that two of them are busy with other projects, the third expert is invited to conduct this review.

[0090] Adopting this automated and refined method can effectively reduce the subjective bias brought by manual intervention. At the same time, it can also locate the real reviewers with the corresponding knowledge background faster, so as to ensure that each manuscript can receive an objective, fair and professional evaluation. In addition, this method helps to promote interdisciplinary communication, enabling researchers in different fields to learn from each other and jointly promote scientific progress.

[0091] 103. Based on the historical review records and current available time of the reviewer, predict the review ability and response speed of the reviewer for the target review task through data analysis, and calculate the corresponding review suitability index according to the prediction results;

[0092] In this step, analyze the past performance of each potential reviewer (such as review quality, on-time rate) and their current time arrangement to evaluate their ability to complete new tasks.

[0093] In an embodiment of the present application, each candidate reviewer is scored, for example, based on factors such as the number of reviews they have done in the past, quality feedback, average response time, and recent work burden. Finally, a value between 0 and 10 is generated for each person as the suitability index.

[0094] Optionally, in step 103, "the historical review records at least include: review time data, quality assessment data; based on the historical review records and current available time of the reviewer, predict the review ability and response speed of the reviewer for the target review task", including: predicting the review response speed of the reviewer for the target review task according to the review time data in the historical review records of the reviewer and the current available time; predicting the review ability of the reviewer for the target review task based on the quality assessment data in the historical review records of the reviewer; taking the review response speed and the review ability as the prediction results.

[0095] Review time data: Records the time required for reviewers to complete previous review tasks.

[0096] Quality assessment data: Includes quality feedback information on review comments from other experts or authors.

[0097] Current available time: Refers to the amount of time that a reviewer can use for new tasks in a future period.

[0098] Review response speed: The expected speed at which a reviewer starts reviewing after receiving a new manuscript.

[0099] Review ability: The ability of a reviewer to provide high-quality feedback estimated based on past performance.

[0100] Review suitability index: A value obtained by combining the response speed and review ability, used to measure whether the reviewer is suitable for the current task.

[0101] First, use the data on the time taken by the reviewer to complete similar tasks before (such as the average review cycle), and combine it with their current schedule (i.e., available time) to estimate how long it will take him / her to start reviewing a new manuscript. Second, analyze the quality scores of the reviewer's previous reviews, such as the proportion of positive reviews, specific scores, etc., and estimate the level of review comments that he / she can provide based on this. Finally, combine the above two factors and use a specific algorithm to generate a score representing the overall suitability of the reviewer.

[0102] In the embodiments of this application, assume there is a reviewer A. His historical review records show that in the past five years, he has completed a total of 50 reviews of articles in the field of machine learning, with an average of 7 days per article. In addition, these reviews received an 85% favorable review rate. Now there is a new paper on deep neural network optimization technology that needs to be reviewed. By checking A's schedule, it is found that he has two weeks of free time in the next month. So, the fitness of A for this task can be estimated through the following formula:

[0103] Review response speed = (Average number of days of past reviews) / (Number of available weeks currently * 7)

[0104] Assume that A can start reviewing immediately within the next two weeks, then:

[0105] Review response speed = 7 / (2 * 7) ≈ 0.5 (indicating that it can start approximately within half a week after receiving the task);

[0106] The review ability score directly uses the favorable review rate: 85%;

[0107] Review suitability index = Review response speed * weight + Review ability * weight;

[0108] If the response speed weight is set to 0.3 and the ability weight is set to 0.7, then:

[0109] Review suitability index = 0.5 * 0.3 + 85 * 0.7 = 0.15 + 59.5 = 59.65;

[0110] In this way, the editor can not only more scientifically select suitable reviewers, but also better plan the timeline of the entire review process. This not only helps to speed up the publication of research results, but also ensures that each manuscript receives high-quality professional review. In addition, this method can also promote fair competition, enabling excellent reviewers to have more opportunities, thereby motivating the entire academic community to improve the review standards.

[0111] This application takes into account that in the field of academic publishing, the efficiency and quality of the review process are crucial. The following problems exist in the prior art: First, editors usually rely on subjective judgment to estimate the response speed of reviewers, lacking objective data support; Second, the available time of reviewers is often not fully considered, resulting in the assigned tasks may exceed the time range of reviewers, thus affecting the speed and quality of the review; Finally, traditional prediction methods fail to fully utilize the information in historical data and cannot provide accurate prediction results. Therefore, the embodiments of the present invention propose a method based on Bayesian statistics to predict the review response speed of reviewers for target review tasks, and conduct a comprehensive evaluation in combination with the current available time to solve the above existing problems and improve the efficiency and accuracy of the review process.

[0112] The specific optional solution is as follows:

[0113] Optionally, "predicting the review response speed of the reviewer for the target review task according to the review time data in the reviewer's historical review records and the current available time" in step 103 includes:

[0114] Predict the review response speed of the reviewer for the target review task through the following calculation formula:

[0115] ;

[0116] Wherein, is the review time data in the reviewer's historical review records; is the current available time of the reviewer; represents the expected value of the actual completion time and the current available time under the condition of the given historical review time data according to the posterior distribution ; is the posterior distribution calculated based on historical review times and current available times; is the expected completion time of the reviewer for the target review task, i.e., the predicted value of the review response speed; numerator is the combined influence of the review time and the current available time; denominator is for all possible review times multiplied by their corresponding current available times and then the result of integration;

[0117] Review time data : Records of the time spent by the reviewer in completing similar review tasks in the past.

[0118] Current available time : The amount of time the reviewer can use for new review tasks.

[0119] Expected value : Under the given conditions of historical review time and current available time, the expectation of the actual completion time according to the posterior distribution.

[0120] Posterior distribution : The probability distribution after combining prior knowledge (such as the general performance of the reviewer) with the observed data (such as historical review times).

[0121] Review response speed : The predicted expected completion time of the reviewer for a specific task.

[0122] First, obtain the time data of the reviewer for completing similar types of manuscripts in the past. Second, understand the actual time the reviewer can use for review in a future period. Further, use Bayes' formula to combine the historical review time data with the current available time to obtain a posterior distribution describing the likelihood of the reviewer completing a new task. Then use the provided formula to calculate the expected completion time of the reviewer for the new task. Finally, combine the predicted response speed and other factors (such as review ability) to obtain a comprehensive evaluation index.

[0123] In an embodiment of the present application, assume that a reviewer B has completed the review work of 100 papers in the biomedical field in the past five years, with an average review time of 10 days each time. Now there is a new biomedical research article that needs to be reviewed. Checking B's schedule reveals that he has 20 days available for review in the next month.

[0124] Historical review time data days, current available time days.

[0125] Calculate the posterior distribution: , since this is a simplified example, it can be assumed that the result of the integral part is a constant, such that is simplified to a proportional relationship.

[0126] Calculate the review response speed according to the formula: days, which means that after considering the current available time, it is expected that Reviewer B will complete the review task in about 10 days.

[0127] By adopting this method, the review response speed of the reviewer can be predicted more accurately, thus helping the editor to reasonably arrange the review tasks and avoid delays caused by insufficient time of the reviewer. At the same time, by combining historical data with the actual situation, the reliability of the prediction is improved, which helps to enhance the efficiency of the entire review process. In addition, this method can also be used as a basis to further develop more complex models to more comprehensively evaluate the overall performance of the reviewer.

[0128] Optionally, "predicting the review response speed of the reviewer for the target review task according to the review time data in the reviewer's historical review records and the current available time" in step 103 includes: constructing a Bayesian statistical model according to the review time data in the reviewer's historical review records to describe the prior distribution of the review time; using the Bayesian update rule to update the prior distribution according to the latest review time data of the reviewer obtained to obtain the posterior distribution; based on the posterior distribution, combining the current available time of the reviewer, estimating the expected completion time of the reviewer for the target review task as the predicted value of the review response speed of the reviewer.

[0129] Prior distribution: The probability distribution of the possible values of a parameter based on existing knowledge before obtaining new observational data.

[0130] Posterior distribution: The probability distribution that combines prior information and new observational data, reflecting the updated state of knowledge.

[0131] Bayesian update rule: A mathematical formula used to calculate the posterior distribution from the prior distribution and new data.

[0132] Expected completion time: The time required for the reviewer to complete a specific review task predicted by the model.

[0133] Review response speed: That is, the expected completion time, representing the speed from the start of the review by the reviewer to the submission of the review comments.

[0134] First, use the time data in the historical review records of the reviewers to establish an initial probability distribution model describing the review time. Second, when the latest review time data is obtained, use Bayes' theorem to update the prior distribution to form a posterior distribution reflecting the latest situation. Finally, combine the current available time of the reviewers and use the posterior distribution to estimate the expected time for them to complete new tasks.

[0135] Suppose there is a normal distribution as the prior distribution, where the mean and the variance are obtained from the historical data. When new review time data appears, the following Bayes update rule can be used:

[0136] The updated mean ;

[0137] The updated variance ;

[0138] where is the variance of the new data, assuming it is known or can be estimated from the data.

[0139] In the embodiment of the present application, assume that a reviewer C has completed the review work of 50 papers in the field of machine learning in the past, with an average time-consuming of 8 days and a standard deviation of 2 days. This means that the prior distribution can be represented by a normal distribution, with the mean , and the variance .

[0140] Recently, C has completed 3 more review works, which took 7 days, 9 days, and 6 days respectively. These new data can be added to the model: the average value of the new data , assuming the standard deviation of the new data (for simplicity), now apply the Bayes update rule:

[0141] The updated mean ;

[0142] The updated variance ;

[0143] If reviewer C has 20 days available in the next month, it can be expected that he will complete the new review task in about 7.5 days, which is reasonable considering his current available time.

[0144] By introducing and continuously updating the Bayesian statistical model, the estimation of the reviewers' response speed can be dynamically adjusted, making the prediction closer to the actual situation. This not only helps the editor make a more reasonable time arrangement but also ensures that the reviewers can complete the review work in their best state, thus improving the efficiency and quality of the entire review process. In addition, this data-driven method can better handle the changes in the reviewers' workload and promote the smooth progress of academic exchanges.

[0145] Optionally, the "calculating the corresponding review suitability index according to the prediction result" in step 103 includes: according to the current available time of the reviewer, using the time series analysis algorithm to predict the available time periods of the reviewer in a future period; calculating the review suitability index of the reviewer according to the prediction result and the available time periods.

[0146] Time series analysis: A statistical method used to analyze data points arranged in chronological order to discover trends, periodic or seasonal patterns.

[0147] Available time period: The specific time interval during which a reviewer can be used for review work in a future period obtained according to the prediction result.

[0148] Review suitability index: An index calculated by combining the review response speed and the future available time period, used to measure the suitability of a reviewer to undertake the current review task.

[0149] First, use the time series analysis algorithm to predict the available time period: By performing time series analysis on the historical work schedule data of the reviewer, identify their daily work patterns and potential time rules, so as to predict the future available time period. Secondly, combine the predicted available time period with the time length required for the review task to evaluate whether this time period is sufficient to complete the review task, and calculate an index reflecting the review suitability accordingly.

[0150] In the embodiment of the present application, assume there is a reviewer D, and his historical records show that he has been completely free for two days on average every week in the past two years and can be used for review work. By using ARIMA (Autoregressive Integrated Moving Average Model) to perform time series analysis on D's work schedule data, the daily available time distribution of the next month for him can be predicted.

[0151] Assume that the predicted values of the available hours per day for reviewer D in the next month have been obtained through the ARIMA model. For example:

[0152] The first week: 4 hours on Monday, 6 hours on Tuesday, 3 hours on Wednesday...

[0153] The second week: 2 hours on Monday, 5 hours on Tuesday, 7 hours on Wednesday...

[0154] This data can help determine which days are most suitable for review.

[0155] If the target review task is expected to take 10 hours to complete, then the above prediction results can be used to evaluate which days are most suitable as review periods. For example:

[0156] If Monday and Tuesday of the first week are selected (4 + 6 = 10 hours), the review suitability index is relatively high because it exactly meets the task requirements and has strong continuity. If Tuesday and Wednesday of the second week are selected (5 + 7 = 12 hours), although the total duration exceeds the required time, it also indicates that reviewer D has sufficient buffer space during this period to handle possible delays or other emergencies.

[0157] A simple review suitability index formula can be defined as follows: Suitability index = (Predicted available time / Required time) × Consecutive days coefficient. The consecutive days coefficient can be set according to actual needs. For example, a higher weight can be given to two consecutive days.

[0158] Assuming the coefficient for two consecutive days is 1.2, the suitability index for the first week is: Suitability index 1 = (10 / 10) × 1.2 = 1.2, and the suitability index for the second week is: Suitability index 2 = (12 / 10) × 1.2 = 1.44. From this example, it can be seen that Tuesday and Wednesday of the second week are more suitable for arranging this review task.

[0159] By introducing time series analysis to predict the future available time periods of reviewers and calculating the review suitability index accordingly, editors can arrange review tasks more precisely. This method not only takes into account the current available time of reviewers but also foresees future time arrangements, helping to avoid task delays caused by time conflicts. In addition, it can help editors optimize resource allocation, ensure that each reviewer can complete the review work in the best state, thereby improving the efficiency and quality of the entire review process. Ultimately, this data-driven method promotes the smooth progress of academic exchanges and speeds up the release of research results.

[0160] 104. Select reviewers whose review suitability index meets the predetermined conditions to form a review team, and use an intelligent time management algorithm to plan suitable review periods for each reviewer in the review team, so that each reviewer can provide review opinions on the target review task during the corresponding review time;

[0161] In this step, appropriate reviewers are selected to form a team based on the suitability index obtained in the previous step, and specific review time windows are assigned to them considering their personal schedules.

[0162] In the embodiment of the present application, 5 reviewers with scores exceeding 7 points are selected to form a review group. The available time periods of them are checked using time optimization software, and then the manuscript is sent out while specifying the expected feedback date.

[0163] Optionally, "selecting reviewers whose review suitability index meets the predetermined conditions to form a review group, and planning appropriate review time periods for each reviewer in the review group through an intelligent time management algorithm" in step 104 includes: setting a predetermined threshold according to the review suitability index of the reviewers, and screening out reviewers higher than the predetermined threshold; including the screened reviewers in the list of candidate review groups; using the intelligent time management algorithm to analyze the current workload and current available time of each reviewer in the list of candidate review groups; based on the current workload and current available time of each reviewer, combined with the available time periods of each reviewer in a future period of time, allocating the most suitable review time period for each reviewer through an optimization algorithm to ensure that each reviewer can focus on the review task of the target review task during the specified period, so as to improve the review efficiency and quality.

[0164] Predetermined threshold: A minimum standard value set to screen out eligible reviewers.

[0165] List of candidate review groups: A list of potential reviewers formed after preliminary screening.

[0166] Intelligent time management algorithm: An automated algorithm used to analyze and plan the time allocation of reviewers to ensure the best execution of review tasks.

[0167] Current workload: The quantity and complexity of all work tasks that a reviewer is currently dealing with.

[0168] Current available time: The amount of time that a reviewer currently has available for new tasks.

[0169] Optimization algorithm: A mathematical method used to find the optimal solution or an approximate optimal solution, which is used here to determine the most suitable review time period for each reviewer.

[0170] First, set a threshold based on the review suitability index and select reviewers whose scores are higher than this threshold as candidates. Second, include the selected reviewers in a temporary list for further evaluation. Further, use an intelligent time management algorithm to analyze the work status and time resources of each reviewer in the candidate list in detail. Finally, based on information such as workload, available time, and available time slots, use an optimization algorithm to arrange the best review time window for each reviewer.

[0171] In the embodiment of this application, assume that there are four reviewers E, F, G, and H, and their review suitability indices are 1.5, 1.8, 2.0, and 1.7 respectively. The set predetermined threshold is 1.6.

[0172] E(1.5) < the predetermined threshold (1.6), so E is not selected.

[0173] F(1.8) > the predetermined threshold (1.6), so F is selected.

[0174] G(2.0) > the predetermined threshold (1.6), so G is selected.

[0175] H(1.7) > the predetermined threshold (1.6), so H is selected.

[0176] The final candidate review panel list includes three reviewers F, G, and H.

[0177] Using the intelligent time management algorithm, the following information can be obtained:

[0178] F: The current workload is low, and there are an average of 5 hours of available time per day in the next week.

[0179] G: The current workload is high, but there is more free time on weekends.

[0180] H: The current workload is moderate, and there are 3 fixed hours of available time every night.

[0181] It is necessary to allocate review time for a paper that is expected to take 20 hours to complete. Assume that a linear programming-based optimization algorithm is used to allocate time:

[0182] For F, he can be arranged to invest 5 hours per day this week, for a total of 25 hours.

[0183] For G, he can be arranged to invest 10 hours each on the two weekend days, for a total of 20 hours.

[0184] For H, he can be arranged to invest 3 hours every night in the next week, for a total of 21 hours.

[0185] To ensure the quality and efficiency of the review process, reviewers who can commit more continuous time may be given priority. For example, G can be selected to complete the review concentratedly on weekends, or F can be allowed to complete it dispersedly during normal times. If three people need to participate together, it can be reasonably allocated according to their respective optimal time periods.

[0186] By introducing a review suitability index and an intelligent time management algorithm, this solution can form a review panel more scientifically and systematically and reasonably arrange the review time periods of each member. This not only helps to ensure that the review tasks can be completed with high quality within the specified time but also avoids review quality problems caused by overloading. In addition, this method improves the transparency and fairness of the entire review process, promotes the effective allocation of review resources, thus accelerating the publication speed of academic achievements. At the same time, it also reduces the workload of editors, enabling them to focus on other key links and further enhancing the overall operation efficiency of the publishing institution.

[0187] 105. Use an ensemble learning strategy to integrate the review opinions of multiple reviewers corresponding to the review panel to form a review conclusion for the target review task.

[0188] In this step, all reviewers' comments are comprehensively collected, and machine learning methods are used for summary analysis to give a comprehensive evaluation and suggestions.

[0189] In the embodiments of this application, natural language processing technology is applied to analyze the positive and negative feedback in each review report; a weighted voting mechanism or other aggregation models are used to decide whether to accept the manuscript for review after modification or directly reject it. The final report includes not only the overall decision but also provides specific improvement directions for the author's reference.

[0190] Optionally, the "using an ensemble learning strategy to integrate the review opinions of multiple reviewers corresponding to the review panel to form a review conclusion for the target review task" in step 105 includes: collecting the review opinions provided by each reviewer in the review panel and converting the review opinions into standardized review scores; adopting an ensemble learning framework and inputting the standardized review scores into an ensemble model, where the ensemble model is composed of multiple base models, and the base models at least include support vector machines, logistic regression, and neural networks, and each base model independently evaluates the review scores; based on the evaluation results of multiple base models, the final review conclusion is determined through a weighted voting or averaging strategy, where the weighted voting strategy assigns different weights according to the historical performance of each base model to ensure the objectivity and fairness of the final review conclusion.

[0191] Ensemble learning: A machine learning method that improves the overall prediction performance by combining the prediction results of multiple models (base models).

[0192] Standardized review scores: Convert the text opinions provided by reviewers into numerical scores in a unified format for easy machine processing.

[0193] Support Vector Machine (SVM): A supervised learning model used for classification and regression analysis.

[0194] Logistic Regression: A generalized linear model used for binary or multi-class classification problems.

[0195] Neural Network: A computational model that simulates the structure of the human brain and is suitable for complex pattern recognition and non-linear relationship modeling.

[0196] Weighted voting: Assign different weights according to the historical performance of each model, and then aggregate these weighted results to obtain the final decision.

[0197] Average strategy: Directly take the average of all model evaluation results as the final conclusion.

[0198] First, obtain the review opinions from each reviewer and convert them into a scoring form under a unified standard. Second, design an ensemble learning system that includes multiple types of basic models (such as SVM, Logistic Regression, Neural Network, etc.). Then, let each basic model evaluate the standardized review scores respectively. Finally, use the weighted voting or average strategy to integrate the evaluation results of each basic model to obtain the final review conclusion.

[0199] In the embodiments of this application, assume there is a review panel consisting of three reviewers A, B, and C. They each provided a review opinion on a certain paper. Convert these opinions into standard scores from 0 to 10: The score given by A is 8.5, the score given by B is 7.0, and the score given by C is 9.0. Next, input these scores into an ensemble learning framework that includes three basic models: SVM, Logistic Regression, and Neural Network. Assume the previous performances of these three models are as follows:

[0200] The historical accuracy rate of SVM is 80%, and the weight is set to 0.3;

[0201] The historical accuracy rate of Logistic Regression is 75%, and the weight is set to 0.25;

[0202] The historical accuracy rate of Neural Network is 85%, and the weight is set to 0.45;

[0203] The evaluation result of SVM for the score is 8.0;

[0204] The evaluation result of Logistic Regression for the score is 7.5;

[0205] The evaluation result of the neural network for the score is 8.8;

[0206] To obtain the final review conclusion, a weighted voting strategy can be used to synthesize these three evaluation results: Final score = (8.0 × 0.3) + (7.5 × 0.25) + (8.8 × 0.45) Final score = (8.0 × 0.3) + (7.5 × 0.25) + (8.8 × 0.45) Final score = 2.4 + 1.875 + 3.96 Final score = 2.4 + 1.875 + 3.96 Final score = 8.235 Final score = 8.235.

[0207] If a simple average strategy is adopted, it can be directly calculated: Final score = (8.0 + 7.5 + 8.8) / 3 Final score = (8.0 + 7.5 + 8.8) / 3 Final score = 8.1 Final score = 8.1.

[0208] In this example, whether it is the weighted voting or the average strategy, relatively high final scores are obtained, indicating that the quality of this paper is good. However, the weighted voting strategy takes into account the historical performance of each model, so its result is more valuable for reference.

[0209] By using an ensemble learning strategy to fuse the opinions of multiple reviewers, this method can reduce the subjective bias brought by a single reviewer and improve the consistency and reliability of the review conclusion. At the same time, since multiple types of machine learning models are integrated, different types of data features can be better captured, thus enhancing the robustness and generalization ability of the entire system. In addition, the weighted voting strategy makes those models with better historical performance have a greater influence, further ensuring the objectivity and fairness of the final review conclusion. This not only helps to improve the review efficiency, but also enhances the transparency of academic review and promotes the high-quality publication of scientific research results.

[0210] This application takes into account that in the field of academic publishing, the review process is a key link to ensure the quality of research. However, there are the following problems in the existing technologies: First, the opinions of a single reviewer may have subjective biases, affecting the objectivity of the final conclusion; Second, the opinions among different reviewers may be inconsistent, making it difficult for editors to make a comprehensive judgment; Finally, with the development of data science, how to use machine learning technologies to improve the consistency and accuracy of review conclusions has become an urgent problem to be solved. Therefore, the embodiments of the present invention propose a method based on an ensemble learning strategy to fuse the opinions of multiple reviewers and form a comprehensive review conclusion for the target review task through a specific weighted formula to solve the above existing problems and improve the objectivity and fairness of the review process.

[0211] Optionally, "using an ensemble learning strategy to fuse the review opinions of multiple reviewers corresponding to the review panel to form a review conclusion for the target review task" in step 105 includes:

[0212] Form a review conclusion for the target review task through the following calculation formula:

[0213] ;

[0214] where is the review score converted from the final review conclusion, is the standardized review score converted from the review opinion provided by each reviewer in the review panel; is the prediction function of the th base model; is the number of base models; is the weight of the th base model; is the adjustment factor; is the historical loss value of the th base model; is the normalization term to ensure that the sum of the weights of all base models is 1.

[0215] First, obtain the review opinions from each reviewer and convert them into a scoring form under a unified standard. Second, select multiple types of machine learning models as base models, such as support vector machines, logistic regression, neural networks, etc. Further, calculate the weights of each base model using the given formula, and the weights depend on the historical performance of the model (i.e., the historical loss value). Finally, multiply the prediction results of each base model by the corresponding weights and then sum them to obtain the final review score.

[0216] Suppose there is a review panel consisting of three reviewers A, B, and C. They each provided a review opinion on a certain paper. Convert these opinions into a standard score from 0 to 10: The score given by A is 8.5, the score given by B is 7.0, and the score given by C is 9.0.

[0217] Next, input these scores into an ensemble learning framework that includes three base models: SVM, logistic regression, and neural network. Suppose the previous performances of these three models are as follows: The historical loss value of SVM is 0.3, the historical loss value of logistic regression is 0.4, and the historical loss value of neural network is 0.2. Set the adjustment factor λ = 1λ = 1.

[0218] Calculate the weights of each base model:

[0219] ;

[0220] Suppose the base model evaluated the standardized review scores as follows: The evaluation result of the SVM for the scores is 8.0, the evaluation result of the logistic regression for the scores is 7.5, and the evaluation result of the neural network for the scores is 8.8.

[0221] Calculate the final review score:

[0222] ;

[0223] By using an ensemble learning strategy and combining a specific weighting formula to fuse the opinions of multiple reviewers, this method can reduce the subjective bias brought by a single reviewer, improve the consistency and reliability of the review conclusion. At the same time, since multiple types of machine learning models are integrated and the weight assignment takes into account the historical performance of each model, it can better capture different types of data features, thus enhancing the robustness and generalization ability of the entire system. In addition, this weighting method makes the models with better historical performance have a greater influence, further ensuring the objectivity and fairness of the final review conclusion. This not only helps improve the review efficiency, but also enhances the transparency of academic review and promotes the high-quality publication of research results.

[0224] This application takes into account that in the field of academic publishing, the review process is a key link to ensure research quality. However, there are the following problems in the existing technologies: First, editors usually evaluate the capabilities of reviewers based on limited information (such as personal experience and simple statistics), and this method lacks systematicness and objectivity; Second, traditional evaluation methods fail to fully utilize the rich information in historical data, resulting in inaccurate evaluation of reviewers' capabilities; Finally, with the development of machine learning technologies, how to use advanced algorithms to improve the accuracy of predicting review capabilities has become an urgent problem to be solved. Therefore, the embodiments of the present invention propose a method based on a random forest model to predict the review capabilities of reviewers for a target review task, so as to solve the above existing problems and improve the efficiency and quality of the review process.

[0225] Optionally, predicting the review capabilities of the reviewer for the target review task based on the quality evaluation data in the reviewer's historical review records includes:

[0226] Predict the review capabilities of the reviewer for the target review task through the following calculation formula:

[0227] ;

[0228] where is the feature vector of the reviewer; is the prediction function of the random forest model; is a parameter of the random forest model; is the review ability score for historical review tasks; is the number of samples in the training dataset; is the review ability score of the reviewer for the target review task; represents finding a set of optimal parameters , such that the actual review ability scores of all training samples and the model predicted values have the minimum sum of squared differences.

[0229] First, relevant features are extracted from the historical review records of reviewers to form a feature vector . Then, the random forest algorithm is used as the prediction model, which can handle a large number of features and provide stable prediction results. Secondly, the dataset in the historical review records is used to train the model by minimizing the sum of squared differences between the actual scores and the predicted scores to find the optimal parameters . Finally, the relevant features of the target review task are input into the trained model to obtain the predicted review ability score of the reviewer for this task .

[0230] Suppose there is a dataset of historical review records of 100 reviewers, and each record includes the following features: professional background (numerically represented), number of past reviews, average score, review completion time, article type preference. These features form a feature vector . In addition, each record also includes an actual review ability score . 100 samples are drawn from the historical data, each sample contains 5 features, and the corresponding actual review ability scores. For example:

[0231] Sample 1: [1, 20, 9.0, 7, 8.5] -> 8.8

[0232] Sample 2: [2, 15, 8.5, 5, 7.0] -> 7.5 ...

[0233] Sample 100: [3, 30, 9.5, 6, 9.0] -> 9.2

[0234] Use the random forest algorithm to train the model. Suppose after training, the model finds the optimal parameters .

[0235] Now it is necessary to predict the review ability of a new reviewer A for a certain paper. The feature vector of reviewer A is: professional background: 2, number of past reviews: 18, average score: 8.7, review completion time: 6, article type preference: 8.0.

[0236] Input these features into the trained random forest model: Suppose the predicted score output by the model is 8.6.

[0237] By using the random forest model to predict the reviewing ability of reviewers, this method can more systematically utilize the information in historical data, improving the accuracy and reliability of predictions. At the same time, the random forest model has strong robustness and generalization ability, and can handle high-dimensional and complex data structures. In addition, this data-driven method reduces the influence of subjective judgment, enhancing the transparency and fairness of the review process. This not only helps editors make more reasonable reviewer selections, but also promotes the high-quality publication of research results and enhances the trust in the entire academic community.

[0238] Optionally, predicting the reviewing ability of the reviewer for the target reviewing task based on the quality assessment data in the reviewer's historical review records includes: collecting the quality assessment data in the reviewer's historical review records, including the acceptance rate of the review comments, the detail level of the review reports, and the usefulness score of the review comments; constructing a training data set according to the quality assessment data, where each piece of training data in the training data set contains the feature vector of the reviewer and the reviewing ability score for the historical reviewing task, and the feature vector includes the reviewer's title, institution, professional field, previous review times, and average review time; training a random forest model, using the feature vectors in the training data set as input and the reviewing ability scores for the corresponding historical reviewing tasks as labels to learn the mapping relationship between the reviewing ability and the feature vectors; using the trained random forest model to output the reviewing ability score of the reviewer for the target reviewing task and taking it as the result of predicting the reviewing ability of the reviewer for the target reviewing task.

[0239] Quality assessment data: Refers to the data set for evaluating the past work performance of reviewers.

[0240] Acceptance rate: The proportion of review comments accepted and revised by the author.

[0241] Detail level: The measurement standard for the content detail of review reports.

[0242] Usefulness score: The score given for the value of review comments based on the feedback from peers or other readers.

[0243] Feature vector: A set of numerically represented attribute collections used to describe the characteristics of an object (here, the reviewer).

[0244] Random forest model: An ensemble learning method that uses multiple decision trees to vote to determine the final classification result.

[0245] First, extract the quality assessment data of each reviewer from the historical records, such as the opinion adoption rate. Secondly, organize the information collected above into a form that includes the reviewer's characteristics and their corresponding ability scores. Further, use the random forest algorithm, with the feature vector as the input variable and the historical ability scores as the output variable, to train a model that can reflect the relationship between the two. Finally, use the trained model to predict the ability scores of potential reviewers for the newly submitted target review tasks.

[0246] In the embodiment of the present application, assume that a certain journal needs to find suitable reviewers for a paper in the field of machine learning. First, the system will automatically screen out all registered reviewers with relevant background knowledge and collect their historical review records; then, calculate the eigenvalue of each candidate based on these records, such as the average time required for each review, the number of articles published in the professional field, etc.; then, use the previously established random forest model to predict the scores of all candidates; finally, recommend the top-scoring experts as the best candidates for the editor to refer to and select.

[0247] By implementing this solution, the objectivity and fairness in the review process can be significantly improved, the bias caused by human factors can be reduced, and at the same time, the review cycle can be effectively shortened, and the dissemination speed of scientific research results can be accelerated. In addition, it can also help young scholars obtain growth opportunities faster and promote the healthy development of the academic community.

[0248] Optionally, in the process of predicting the review ability and response speed of the reviewer for the target review task through data analysis, it further includes: using a deep neural network to model the behavior pattern of the obtained reviewer, and combining the obtained review preferences, historical behavior trajectories, and external influencing factors of the reviewer to refine the prediction of the review ability and the response speed to update the prediction result; introducing a reinforcement learning algorithm to dynamically adjust the parameters of the prediction model according to the completion of the historical review tasks by the reviewer at different time points to improve the prediction accuracy of the prediction result, where the adjusted parameters of the prediction model at least include the prior probability distribution parameters in the Bayesian statistical model, the hyperparameters in the random forest algorithm, the reward function, discount factor in the reinforcement learning algorithm, and the weight and bias term parameters in the deep neural network.

[0249] Deep neural network: An artificial intelligence algorithm that simulates the way the human brain processes information and can be used to identify complex patterns in data.

[0250] Review preference: Refers to the selection tendency shown by the reviewer when accepting or rejecting a review invitation.

[0251] Historical behavior trajectory: including but not limited to behavior data accumulated over a long period, such as review time, frequency, field preference, etc.

[0252] External influencing factors: factors such as holidays and conference arrangements, which may affect the available time of reviewers.

[0253] Bayesian statistical model: a statistical method that updates the posterior probability using the prior probability distribution.

[0254] Hyperparameter: a parameter that needs to be manually set in a machine learning model, such as the number of trees in a random forest.

[0255] Reward function: a function that defines the goal in reinforcement learning and is used to guide the learning direction of the algorithm.

[0256] Discount factor: a weight that measures the importance of future rewards and is used to calculate the cumulative reward in reinforcement learning.

[0257] First, use a deep neural network to learn the behavior patterns of reviewers, with the input including multi-dimensional information such as personal preferences and historical behavior trajectories. Second, combine the above model with data considering external factors to obtain more accurate prediction results regarding review ability and response speed. Further, use reinforcement learning techniques to continuously optimize the model parameters based on actual feedback to ensure that the model evolves self-adaptively over time. Specifically, it is to regularly or on-demand update the key parameter values such as the prior probability in the Bayesian model, the hyperparameters of the random forest, and the weights of the deep learning network.

[0258] In the embodiment of this application, assume that a certain journal is looking for a suitable reviewer to review a latest research article in the field of artificial intelligence. First, the system will collect the historical behavior data of all potential reviewers in this field, such as which topics of articles they have reviewed in the past and the average time required to complete a review. Then, use a trained deep neural network model to analyze this set of data to obtain the estimated review quality score of each candidate for this specific article and the estimated speed of their possible response. At the same time, considering that there will be a large international AI conference soon, which may make some experts very busy, the priority of those known participants is appropriately reduced in the final recommended list. Finally, as more actual review result data accumulates, the system will also automatically tune various algorithm parameters used internally, thereby continuously improving the accuracy of future task prediction.

[0259] By introducing advanced deep learning and reinforcement learning technologies, this solution can not only reflect the real working status of reviewers to a greater extent but also flexibly adapt to the changing needs in different situations, providing a more scientific and reasonable decision-making support tool for editors. This helps to further improve the transparency and fairness in the academic communication process and also promotes the rapid dissemination of high-quality research results.

[0260] In this application, considering that in the academic publishing process, selecting appropriate reviewers requires not only considering their professional capabilities and historical performance but also taking into account the reviewers' response speed and available time slots. Existing methods often focus only on one aspect, such as determining reviewers solely based on review capabilities or response time, which may result in the selected reviewers not being suitable for the current task requirements in actual operation. For example, they may delay the review due to being busy. Therefore, this application proposes a method that comprehensively considers multiple factors to calculate the review suitability index, aiming to improve the overall efficiency and quality of the review process by more comprehensively evaluating the matching degree of reviewers.

[0261] Optionally, calculating the review suitability index of the reviewer based on the prediction result and the available time slot includes:

[0262] Calculate the review suitability index of the reviewer through the following calculation formula:

[0263] ;

[0264] Wherein, represents the review suitability index; represents the number of reviewers; is the comprehensive weight factor of the reviewer, a score comprehensively determined based on the feature vector of the reviewer and the review ability score for historical review tasks; represents the expected completion time of the reviewer for the target review task, is a very small positive number used to prevent division by zero errors; is related to the exponential factor, using the exponential function to emphasize the importance of the review response time, a non-linear coefficient determined based on the statistical distribution characteristics of the reviewers' past response times; represents the predicted review ability score of the reviewer for the target review task; is related to the exponential factor, using the square root to adjust the influence degree of the review ability score; Represents the evaluation value of the available time period for the target review task by the predicted reviewer, which is a quantitative evaluation of the future available time period predicted based on the time series analysis algorithm; Is related to The exponential factor, using , is used to adjust the importance of the available time period; Represents the number of time conflicts with other review tasks that the reviewer has in the future, which is a negative indicator; Is related to The exponential factor, using , is used to adjust the influence degree of the number of conflicts;

[0265] Comprehensive weight factor : A weighted value determined based on the reviewer's feature vector and the scores of their historical review tasks.

[0266] Predicted review response time : The time required for the reviewer to complete the target review task predicted by the model.

[0267] Exponential factor : A coefficient used to adjust the influence degree of each factor on the final suitability index.

[0268] Review ability score : A quantitative indicator reflecting the professional level of the reviewer for a specific task.

[0269] Available time period evaluation value : The amount of available time for the reviewer in the future obtained through methods such as time series analysis.

[0270] Number of time conflicts : The number of other review tasks that the reviewer has already accepted, considered as a negative factor.

[0271] First, collect all relevant information of all potential reviewers, including but not limited to personal background, historical review records, etc., and clean and format them. Secondly, set the corresponding weight factors , exponential factors and a very small positive number for each reviewer. Then substitute the above parameters into the provided calculation formula to calculate the review suitability index of each candidate respectively. Finally, rank the candidate list from high to low according to the obtained suitability index, and select the most suitable candidate for invitation.

[0272] In the embodiments of the present application, it is assumed that a journal needs to find suitable reviewers for an article on machine learning. The system first identifies 5 experts with relevant professional knowledge backgrounds as alternative candidates. For one of the experts (designated as the 1st), the known data is as follows:

[0273] Comprehensive weight factor , the predicted review response time \(T_1 = 7\) days, exponential factor Review ability score \(C_1 = 4.6\) (out of 5), exponential factor , the available time period evaluation value \(A_1 = 10\) hours / week, exponential factor , the number of time conflicts \(K_1 = 2\), exponential factor , extremely small positive number .

[0274] Use the given formula to calculate the review suitability index of this expert :

[0275] ;

[0276] After substituting the specific values, we get:

[0277] ;

[0278] Repeat this process to calculate the suitability indices of the other four experts and make a final selection based on this.

[0279] By adopting such a multi-dimensional evaluation system, it can more accurately reflect the adaptation situation of each reviewer under specific conditions, thereby helping the editor quickly find the most suitable reviewer. This not only improves the fairness and transparency in the review process but also ensures the quality and timeliness of paper review, and further promotes the effective dissemination and development of scientific research results.

[0280] Figure 2 Figure [0000738] shows a schematic structural diagram of a review task allocation system provided by an embodiment of the present application. The device includes: Figure 2 As shown, the device includes:

[0281] A determination module 21, configured to receive a target review task and determine the research topic of the target review task;

[0282] A screening module 22, configured to screen out reviewers matching the research topic in a pre-constructed reviewer database;

[0283] A prediction module 23, configured to predict the review ability and response speed of the reviewer based on the historical review records and current available time of the reviewer, and calculate the corresponding review suitability index according to the prediction results;

[0284] A processing module 24, configured to select reviewers whose review suitability indices meet a predetermined condition to form a review group, and plan a suitable review time period for each reviewer in the review group through an intelligent time management algorithm, so that each reviewer provides a review opinion on the target review task during the corresponding review time;

[0285] A generating module 25, configured to use an integrated learning strategy to fuse the review opinions of multiple reviewers corresponding to the review group to form a review conclusion for the target review task.

[0286] Figure 2 The described review task allocation system can execute Figure 1 The described review task allocation method in the illustrated embodiment, and its implementation principle and technical effects will not be elaborated further. For the review task allocation system in the above embodiment, the specific manners in which each module and unit perform operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0287] In a possible design, Figure 2 The review task allocation system in the illustrated embodiment can be implemented as a computing device, such as Figure 3 shown, and this computing device can include a storage component 31 and a processing component 32;

[0288] The storage component 31 stores one or more computer instructions, and among them, the one or more computer instructions are called and executed by the processing component 32.

[0289] The processing component 32 is configured to: receive a target review task, and determine the research topic of the target review task; screen out reviewers matching the research topic; predict the review ability and response speed of the reviewers for the target review task, and calculate the corresponding review suitability index according to the prediction results; select reviewers whose review suitability indices meet a predetermined condition to form a review group; use an integrated learning strategy to fuse the review opinions of multiple reviewers corresponding to the review group to form a review conclusion for the target review task.

[0290] An embodiment of the present application also provides a computer storage medium storing a computer program, and when the computer program is executed by a computer, it can implement the Figure 1 review task allocation method in the illustrated embodiment.

[0291] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated here.

[0292] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative labor.

[0293] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0294] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present application.

Claims

1. A method for assigning manuscript review tasks, characterized in that: include: Receive target review tasks and determine the research topics of the target review tasks; From the pre-built reviewer database, select reviewers that match the research topic; Based on the reviewer's historical review records and current available time, predict the reviewer's reviewing ability and response speed for the target review task through data analysis, and calculate the corresponding review suitability index according to the prediction result; Selecting reviewers whose review suitability index meets the predetermined conditions to form a review group, and planning a suitable review time period for each reviewer in the review group through an intelligent time management algorithm, so that each reviewer can provide review opinions for the target review task at the corresponding review time; Using an integrated learning strategy to integrate the review opinions of multiple reviewers corresponding to the review group to form a review conclusion for the target review task; The historical review records include at least: review time data and quality assessment data; The step of predicting the reviewing ability and response speed of the reviewer for the target reviewing task through data analysis based on the reviewer's historical reviewing records and current available time includes: Predicting the review response speed of the reviewer for the target review task based on the review time data in the reviewer's historical review records and the current available time; Predicting the reviewing ability of the reviewer for the target reviewing task based on the quality assessment data in the reviewer's historical review records; Taking the review response speed and the review ability as prediction results; The following calculation formula is used to predict the review response speed of the reviewer for the target review task: ； in, The review time data in the reviewer's historical review records; is the current available time of the reviewer in question; Represents the data at a given historical review time and current available time Under the condition of Actual completion time Expected value; It is the posterior distribution calculated based on the historical review time and the current available time; is the expected completion time of the reviewer for the target review task, that is, the predicted value of the review response speed; is the combined effect of review time and current available time; the denominator It is the time for all possible review Multiply it by its corresponding current available time The result of post-integration; The following calculation formula is used to predict the reviewer's review ability for the target review task: ； in, is the feature vector of the reviewer; is the prediction function of the random forest model; are the parameters of the random forest model; It is the review ability score for historical review tasks; is the number of samples in the training dataset; is the reviewer's review ability score for the target review task; Indicates finding a set of optimal parameters , so that the actual review ability score of all training samples With the model prediction value The sum of squared differences between them is minimal.

2. The method according to claim 1, characterized in that The predicting the review response speed of the reviewer for the target review task according to the review time data in the reviewer's historical review records and the current available time includes: Based on the review time data in the historical review records of the reviewers, a Bayesian statistical model is constructed to describe the prior distribution of the review time; Using the Bayesian update rule, the prior distribution is updated according to the latest review time data of the reviewer to obtain the posterior distribution; Based on the posterior distribution and in combination with the current available time of the reviewer, the expected completion time of the reviewer for the target review task is estimated as a predicted value of the reviewer's review response speed.

3. The method according to claim 1, characterized in that: The predicting the reviewing ability of the reviewer for the target reviewing task based on the quality assessment data in the reviewer's historical review records includes: Collecting quality assessment data from the reviewers' historical review records, including the adoption rate of the review opinions, the level of detail of the review reports, and the usefulness scores of the review opinions; Constructing a training data set based on the quality assessment data, wherein each piece of training data in the training data set contains a feature vector of the reviewer and a reviewing ability score for a historical review task, wherein the feature vector includes the reviewer's title, institution, professional field, number of past reviews, and average review time; Training a random forest model, using the feature vectors in the training data set as input and the reviewing ability scores of the corresponding historical review tasks as labels, to learn the mapping relationship between the reviewing ability and the feature vectors; The trained random forest model is used to output the reviewer's review ability score for the target review task, and used as a result of predicting the reviewer's review ability for the target review task.

4. The method according to claim 1, characterized in that: In the process of predicting the reviewer's reviewing ability and response speed for the target reviewing task through data analysis, it also includes: Using a deep neural network to model the behavior patterns of the reviewers, combining the review preferences, historical behavior trajectories, and external influencing factors of the reviewers, to refine the prediction of the reviewing ability and the response speed, so as to update the prediction results; A reinforcement learning algorithm is introduced to dynamically adjust the parameters of the prediction model according to the completion status of the historical review tasks by the reviewers at different time points to improve the prediction accuracy of the prediction results, wherein the adjusted parameters of the prediction model include at least the prior probability distribution parameters in the Bayesian statistical model, the hyperparameters in the random forest algorithm, the reward function and discount factor in the reinforcement learning algorithm, and the weight and bias parameters in the deep neural network.

5. The method according to claim 4, characterized in that The calculation of the corresponding review suitability index according to the prediction results includes: Based on the current available time of the reviewer, a time series analysis algorithm is used to predict the available time period of the reviewer in the future; Calculating the review suitability index of the reviewer according to the prediction result and the available time period; The review suitability index of the reviewer is calculated by the following formula: ； in, represents the review suitability index; represents the number of reviewers; is the reviewer's comprehensive weight factor, which is a score determined based on the reviewer's feature vector and the reviewer ability score for historical review tasks; It indicates the expected completion time of the reviewer for the target review task. is a very small positive number used to prevent division by zero errors; is with The relevant exponential factor, using the exponential function , to emphasize the importance of reviewer response time, a nonlinear coefficient determined based on the statistical distribution characteristics of reviewers' past response times; It represents the predicted reviewer's review ability score for the target review task; is with The relevant exponential factor is expressed in square roots. , used to adjust the impact of reviewer ability scores; It represents the estimated value of the available time period of the reviewer for the target review task, which is a quantitative evaluation of the future available time period predicted by the time series analysis algorithm; is with The relevant exponential factor is , used to adjust the importance of the available time periods; It indicates the number of time conflicts that the reviewer will have with other review tasks in the future, which is a negative indicator; is with The relevant exponential factor is , used to adjust the impact of the number of conflicts.

6. The method according to claim 1, characterized in that The reviewers matching the research topic are screened out from the pre-built reviewer database, including: Extract keywords and phrases from the target review task to identify the core elements of the research topic; Converting the core elements into a structured data format, and matching the core elements of the structured data format with the professional field information of the reviewers stored in the reviewer database to identify reviewers matching the research topic; Filter out the best list of reviewers based on their professional field matching, past review experience, and review quality evaluation information; Select qualified reviewers from the optimal reviewer list.

7. The method according to claim 1, characterized in that The step of selecting reviewers whose review suitability index meets predetermined conditions to form a review group, and planning a suitable review time period for each reviewer in the review group through an intelligent time management algorithm includes: According to the review suitability index of the reviewers, a predetermined threshold is set, and reviewers with a suitability index higher than the predetermined threshold are screened out; Include the selected reviewers in the candidate review team list; Analyzing the current workload and current available time of each reviewer in the candidate review panel list using an intelligent time management algorithm; Based on each reviewer's current workload and current available time, combined with each reviewer's available time period in the future, an optimization algorithm is used to assign the most suitable review time period to each reviewer, to ensure that each reviewer can focus on the review task of the target review task within the specified time period, so as to improve the review efficiency and quality.

8. The method according to claim 7, characterized in that The use of an integrated learning strategy to integrate the review opinions of multiple reviewers corresponding to the review group to form a review conclusion for the target review task includes: Collecting the review opinions provided by each reviewer in the review team and converting the review opinions into standardized review scores; Using an integrated learning framework, the standardized review score is input into an integrated model, wherein the integrated model is composed of multiple basic models, wherein the basic models at least include support vector machines, logistic regression, and neural networks, and each basic model independently evaluates the review score; Based on the evaluation results of the multiple basic models, the final review conclusion is determined by weighted voting or averaging strategy, wherein the weighted voting strategy assigns different weights according to the historical performance of each basic model to ensure the objectivity and fairness of the final review conclusion; The review conclusion for the target review task is formed by the following calculation formula: ； in, It is the review score converted from the final review conclusion. It is the standardized review score converted from the review opinions provided by each reviewer in the review team; It is The prediction function of the base model; is the number of base models; It is The weights of the base models; is the regulating factor; It is The historical loss value of the base model; is a normalization term that ensures that the sum of the weights of all base models is 1.

9. A system for distributing manuscript review tasks, applied to the method for distributing manuscript review tasks according to any one of claims 1 to 8, characterized in that: include: A determination module, used to receive a target review task and determine a research topic of the target review task; A screening module is used to screen out reviewers matching the research topic from a pre-built reviewer database; A prediction module, used to predict the reviewing ability and response speed of the reviewer through data analysis based on the reviewer's historical review records and current available time, and calculate the corresponding review suitability index according to the prediction results; A processing module is used to select reviewers whose review suitability index meets the predetermined conditions to form a review group, and plan a suitable review time period for each reviewer in the review group through an intelligent time management algorithm, so that each reviewer can provide review opinions for the target review task at the corresponding review time; A generation module is used to use an integrated learning strategy to integrate the review opinions of multiple reviewers corresponding to the review group to form a review conclusion for the target review task.

Images