Journal relevance analysis method and system based on multi-source data and AI model

Abstract

The application provides a journal correlation analysis method and system based on multi-source data and an AI model, and belongs to the technical field of artificial intelligence, and specifically comprises: a correlation unit identification module responsible for determining the correlation unit of the journal paper of a target unit based on the journal paper data of scientific researchers; a correlation unit updating module for determining the comparison and analysis target for correlation unit updating in the journal paper based on an analysis processing strategy and the number of overlaps between the journal paper and the scientific researchers of the target unit; and a correlation analysis module for determining an AI model-based journal correlation analysis method based on the update processing result of eliminated scientific researchers and the scientific paper data of the scientific researchers of the target unit except the eliminated scientific researchers when performing analysis and processing of the correlation unit, and performing correlation analysis and processing of the journal paper based on the journal correlation analysis method and multi-source data, so as to ensure the comprehensiveness and reliability of the abnormal behavior recognition processing.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence technology, and in particular relates to a method and system for journal association analysis based on multi-source data and AI models. Background Technology

[0002] Power companies (including power grid companies, power generation groups, power research institutes, design institutes, etc.) are the core entities for technological innovation in the power industry. Their researchers publish a large number of papers in domestic and foreign academic journals every year, covering multiple technical fields such as power systems and their automation, high voltage technology, power electronics, new energy, energy storage, smart grids, and power markets.

[0003] In existing technical solutions, power company staff often update journal articles in the database through users' periodic self-reporting, ensuring that the users' journal articles match the actual situation. However, the following technical problems exist: When identifying and analyzing abnormal behaviors such as duplicate submissions in journal articles, there are often discrepancies between the journal article title and the actual institution name. Therefore, how to update and identify the associated institution name, and how to determine the journal's association analysis method based on the updated institution name identification results, so as to ensure the reliability and effectiveness of abnormal behavior identification and processing, has become an urgent technical problem to be solved.

[0004] Specifically, this application provides a method and system for journal association analysis based on multi-source data and AI models. Summary of the Invention

[0005] To achieve the objectives of this invention, the following technical solution is adopted: Specifically, this application provides a journal association analysis system based on multi-source data and AI models, which includes: Related unit identification module, related unit update module, and related analysis module; The associated unit identification module is responsible for determining the associated units of journal articles of the target unit based on the researchers' journal article data. The associated unit update module is responsible for determining the analysis and processing strategy of associated units of the target unit in the journal articles based on the journal article data of each staff member in the target unit. Using the analysis and processing strategy and the number of overlaps between the journal articles and the researchers of the target unit, the module determines the comparative analysis target in the journal articles for associated unit update. The correlation analysis module will exclude researchers who do not need to be referenced when analyzing related units as eliminated researchers. Based on the updated processing results of the eliminated researchers and the research paper data of researchers in the target units other than the eliminated researchers, the module will determine the journal correlation analysis method based on the AI ​​model, and use the journal correlation analysis method and multi-source data to perform correlation analysis processing of journal papers.

[0006] Furthermore, the journal article data of the researchers is determined based on the researchers' journal article entry information in the research management system.

[0007] Furthermore, the associated institution of the target institution's journal article is the institution name that the target institution may be labeled in the journal article.

[0008] Secondly, this application provides a journal association analysis method based on multi-source data and an AI model, applied to the aforementioned journal association analysis system based on multi-source data and an AI model, specifically including: S1 determines the verification matching researchers among the researchers based on the journal article data of each staff member in the target unit, and determines the analysis and processing strategy of the related units of the target unit in the journal articles based on the verification matching researchers among the staff members and the journal article data of the researchers. S2 uses the analysis and processing strategy and the number of overlaps between journal articles and researchers in the target unit to determine the comparative analysis targets in the journal articles for updating related units. Based on the analysis and processing strategy of related units of the target unit in the journal articles and the degree of identification and matching of related units in the comparative analysis targets, the elimination management strategy of the identified targets among the researchers is determined. S3 uses the elimination management strategy to identify researchers among the researchers who do not need to be referenced when calculating the number of overlapping researchers in the analysis and processing of the target unit's related units in journal articles, and identifies them as eliminated researchers. Based on the updated processing results of the eliminated researchers and the research paper data of researchers in the target unit excluding eliminated researchers, an AI model-based journal association analysis method is determined.

[0009] Furthermore, the method for determining the journal association analysis method based on the AI ​​model is as follows: The number of researchers to be eliminated is determined based on the updated processing results of the elimination process. The remaining researchers are those in the target unit who are excluded from the elimination of researchers. The number of remaining researchers is then determined. Based on the number of eliminated researchers, the number of remaining researchers, and the number of research papers published by each remaining researcher, the journal association analysis method based on the AI ​​model is determined.

[0010] The beneficial effects of this invention are as follows: Based on the verification and matching data of researchers and their journal articles, the matching degree of the target unit's related units is determined by updating the related units of the target unit based on the researchers' research articles. Based on the matching degree of the updated related units of the target unit, the analysis and processing strategy of the related units of the target unit in the journal articles is determined, which reduces unnecessary analysis and processing of related units while ensuring the reliability of the updated related units.

[0011] Based on the updated processing results of eliminated researchers, the latency of the current journal correlation analysis is determined. Combined with the number of remaining researchers and the number of research papers by each remaining researcher, the reliability of the updated identification processing of related units is determined. Based on the latency of the current journal correlation analysis and the reliability of the updated identification processing of related units, the AI ​​model-based journal correlation analysis method is determined. This reduces the difficulty of correlation analysis while ensuring the reliability and real-time performance of the identification and processing of abnormal behaviors in journal papers.

[0012] Other features and advantages will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0013] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0014] The above and other features and advantages of the present invention will become more apparent from a detailed description of exemplary embodiments thereof with reference to the accompanying drawings.

[0015] Figure 1 This is a framework diagram of a journal association analysis system based on multi-source data and AI models; Figure 2 This is a flowchart of a journal association analysis method based on multi-source data and AI models; Figure 3 This is a flowchart illustrating the method for determining the analysis and processing strategy for related units of the target unit in a journal article. Detailed Implementation

[0016] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, they are provided so that the invention will be thorough and complete, and the concept of the exemplary embodiments will be fully conveyed to those skilled in the art. The same reference numerals in the drawings denote the same or similar structures, and therefore their detailed description will be omitted.

[0017] The terms “a,” “one,” “the,” and “the” are used to indicate the existence of one or more elements / components / etc.; the terms “including” and “having” are used to indicate an open-ended meaning of inclusion and that other elements / components / etc. may exist in addition to the listed elements / components / etc. Example

[0018] To solve the above problems, according to one aspect of the present invention, such as Figure 1 As shown, a journal association analysis system based on multi-source data and AI models is provided, specifically including: Related unit identification module, related unit update module, and related analysis module; The associated unit identification module is responsible for determining the associated units of journal articles of the target unit based on the researchers' journal article data. The associated unit update module is responsible for determining the analysis and processing strategy of associated units of the target unit in the journal articles based on the journal article data of each staff member in the target unit. Using the analysis and processing strategy and the number of overlaps between the journal articles and the researchers of the target unit, the module determines the comparative analysis target in the journal articles for associated unit update. The correlation analysis module will exclude researchers who do not need to be referenced when analyzing related units as eliminated researchers. Based on the updated processing results of the eliminated researchers and the research paper data of researchers in the target units other than the eliminated researchers, the module will determine the journal correlation analysis method based on the AI ​​model, and use the journal correlation analysis method and multi-source data to perform correlation analysis processing of journal papers.

[0019] Furthermore, the journal article data of the researchers is determined based on the researchers' journal article entry information in the research management system.

[0020] Specifically, the associated institution of the target institution's journal article is the institution name that may be marked in the journal article.

[0021] It should be noted that the journal association analysis method and multi-source data are used to perform association analysis on journal articles, specifically including: Based on the journal association analysis method, when it is determined that the association analysis of the research papers of the researchers of the target institution needs to be carried out, the research papers of multiple journals in the history of the journal paper author are used as a basis. The AI ​​model is used to determine the duplication of the journal paper with the research papers of multiple journals in the history of the journal paper author. Based on the identification results of the duplication, it is determined whether there is any abnormal behavior.

[0022] The journal association analysis AI model is constructed using a multi-granularity text similarity fusion architecture. The model consists of three parts: a feature extraction layer, a similarity calculation layer, and an anomaly classification layer. The feature extraction layer performs multi-granularity feature extraction on the journal article to be analyzed, as well as research papers from multiple journals in the author's historical records: semantic vector representations are extracted from the article title, abstract, and keywords using BERT or RoBERTa pre-trained language models; word frequency features and local word order features are extracted from the main text of the article using TF-IDF and N-gram models; layout features are extracted from the structured content of the article, such as figures, formulas, and data tables, using a layout parsing model; and metadata such as the research project number and experimental data number are added as supplementary features. The similarity calculation layer performs multi-dimensional similarity calculations on the paper to be analyzed and the author's previous papers one by one. These include cosine similarity based on semantic vectors (Sim_sem=cos(V_q, V_h)), overlap based on word frequency features (Sim_tf=|F_q∩F_h| / |F_q∪F_h| (Jaccard coefficient)), and layout similarity based on structural features. The multi-dimensional similarities are then weighted and fused into a comprehensive similarity (Sim_total=λ_1·Sim_sem+λ_2·Sim_tf+λ_3·Sim_layout). The anomaly classification layer receives the comprehensive similarity vector and outputs a classification probability distribution of "normal / duplicate submission / data fabrication" through a fully connected layer and a Softmax activation function.

[0023] During the model training phase, a large number of labeled papers were used for supervised training on the samples. The labeling content included whether there were abnormal behaviors such as duplicate submissions and data fabrication. The cross-entropy loss function was adopted, and the anomaly detection accuracy and recall on the validation set were used as evaluation metrics for hyperparameter tuning. At the same time, a similarity threshold adaptive learning mechanism was introduced. Through contrastive learning, positive sample pairs (papers with reasonable extensions by the same author) and negative sample pairs (paper pairs with duplicate submissions or data fabrication) were trained, enabling the model to distinguish between "reasonable continuation of research" and "abnormal duplication behavior". The model input consists of the journal paper to be analyzed and research papers from multiple journals in the author's history. The output is the comprehensive similarity value of the paper pair and the abnormal behavior classification result (normal / duplicate submission / data fabrication) and the corresponding confidence score. When the confidence score is greater than a preset threshold, abnormal behavior is identified.

[0024] The model deployment phase works in conjunction with the journal association analysis method defined in S3: For researchers using real-time analysis (e.g., those who have been eliminated from research), the model immediately performs similarity comparisons and anomaly identification on newly submitted papers; for researchers using pre-set periodic analysis, the model performs batch comparison analyses at fixed intervals (e.g., 7 days). The model also utilizes information on the overlap between journal papers and the target institution's affiliated institutions and researchers within those institutions, combined with unique research project numbers, the same experimental project, and experimental data for cross-validation. The identification results are fed back to the training dataset to achieve continuous model optimization, thereby reducing the difficulty of association analysis while ensuring the reliability and real-time performance of identifying and handling anomalies such as duplicate submissions and data fabrication.

[0025] Specifically, the abnormal behaviors include duplicate submissions and data fabrication. Example

[0026] Secondly, such as Figure 2 As shown, this application provides a journal association analysis method based on multi-source data and an AI model, applied to the aforementioned journal association analysis system based on multi-source data and an AI model, specifically including: S1 determines the verification matching researchers among the researchers based on the journal article data of each staff member in the target unit, and determines the analysis and processing strategy of the related units of the target unit in the journal articles based on the verification matching researchers among the staff members and the journal article data of the researchers. Specifically, the researchers mentioned in the verification matching section are those whose number of journal articles exceeds a preset threshold. Their published journal articles can be used to identify and process journal articles with non-standard unit name labeling of the target unit.

[0027] Specifically, such as Figure 3 As shown, the method for determining the analysis and processing strategy of the related units of the target unit in the journal article is as follows: In this embodiment, based on the verification and matching of researchers and the journal article data of each researcher, the matching degree of the target unit's related units update processing is determined based on the researchers' research articles. Based on the matching degree of the target unit's related units update processing, the analysis and processing strategy of the target unit's related units in the journal articles is determined. This reduces unnecessary related analysis and processing while ensuring the reliability of the related unit update processing.

[0028] Verification and matching of researchers refers to researchers whose number of journal articles exceeds a preset threshold. This allows for the identification and processing of journal articles with non-standard institution name labeling. Associated institutions refer to the institution names that might be listed in the journal articles.

[0029] By verifying and matching researchers and their journal articles to determine the analytical processing strategy for identifying related units, on the one hand, by identifying researchers with a sufficient number of journal articles as reliable sources of verification information, a more lenient analytical processing strategy for related units can be adopted when the data quality is high; on the other hand, by conducting a graded evaluation of the degree of matching, a comprehensive evaluation of data quality can be achieved, avoiding the use of an overly lenient approach to perform indiscriminate correlation analysis on journal articles, and reducing the computational resources and time costs consumed by unnecessary correlation analysis.

[0030] S11 determines the number of verified matching researchers among the researchers based on the verified matching researchers among the researchers. The statistical verification of the number of researchers provides basic data for the subsequent calculation of the proportion of verification numbers. By statistically verifying the number of researchers, the scale of high-quality researchers in the target unit that can be used to identify related units can be accurately quantified. This provides an objective basis for judging the sufficiency of the verification data, and thus lays the foundation for determining the analysis and processing strategy of related units based on the sufficiency of the verification data.

[0031] The target institution has a total of 10 researchers, P1 to P10. Among them, P1 (8 papers), P2 (7 papers), P3 (6 papers), and P4 (5 papers) are 4 researchers who have published more than 5 journal articles than the preset threshold.

[0032] Specifically, the proportion of verified matching researchers among the researchers is taken as the verification quantity proportion of the researchers. It is determined whether the verification quantity proportion of the researchers is greater than a preset quantity proportion threshold. If so, the analysis and processing strategy for the related units of the target unit in the journal article is determined to be the association analysis strategy. That is, as long as the number of researchers in the journal article that overlap with the target unit reaches or exceeds the first personnel quantity threshold, the association analysis processing of the journal article is performed. If not, proceed to step S12.

[0033] The percentage of verification data is calculated as: (Number of researchers matched for verification) ÷ (Total number of researchers). This percentage determines the sufficiency of the verification data. When the percentage exceeds the threshold, the verification data is sufficient and of high quality, allowing for the adoption of an association analysis strategy (association analysis is performed when the number of overlapping data exceeds the threshold). When the percentage is less than or equal to the threshold, further judgment is applied in step S12. Under sufficient verification data, a more lenient association analysis strategy is employed, expanding the coverage of associated unit identification and promptly identifying more discrepancies in associated unit name labeling. When verification data is insufficient, subsequent refined judgment is applied, thereby reducing the difficulty of data analysis and processing.

[0034] S12 determines the number of journal articles published by each researcher based on their journal article data.

[0035] By statistically analyzing the number of journal articles published by each researcher, and quantifying the output of journal articles by each researcher, we can objectively reflect the size of the sample of articles that each researcher can use for verification with related units. This allows us to determine the identification and processing strategy for related units based on the size of the sample of articles that each researcher can use for verification with related units, ensuring that a relatively lenient association analysis strategy can be adopted while maintaining high verification reliability.

[0036] It is understood that in the above steps, the number of journal articles by the researchers is used as a basis to determine the verification matching weight value of the researchers, and to determine whether the average value of the verification matching weight value of each researcher is greater than the preset matching weight threshold. If so, proceed to step S13; otherwise, determine that the analysis and processing strategy for the related units of the target unit in the journal articles is a restricted analysis strategy. That is, as long as the number of researchers in the journal articles that overlap with the target unit reaches the second personnel number threshold (greater than the first personnel number threshold), the association analysis and processing of the journal articles will be performed.

[0037] The verification matching weight value is a preset weight value corresponding to the number of journal articles published by the researcher. For example, the preset weight mapping is as follows: 8 or more articles → 0.8, 7 articles → 0.7, 6 articles → 0.6, 5 articles → 0.5, 4 articles → 0.4, 3 articles → 0.3, 2 articles → 0.2, 1 article → 0.1.

[0038] The overall richness of scientific research paper data is determined by verifying the mean of the matching weight values. When the mean is greater than the threshold, the data is rich, and the process proceeds to S13 for fine-grained judgment. When the mean is less than or equal to the threshold, a restrictive analysis strategy is adopted (the number of overlapping data must reach a higher threshold for the number of second-level personnel before association analysis is performed). When the overall journal paper data of researchers is not rich, a stricter restrictive analysis strategy is adopted. When the verification sample is insufficient, a stricter update and identification strategy for related units is adopted, thereby reducing the difficulty of data analysis and processing, reducing unnecessary association analysis processing times, and reducing the computational resource consumption of analysis and processing.

[0039] S13 determines the analysis and processing strategy for the related units of the target unit in the journal articles based on the number of verification matching researchers among the researchers and the number of journal articles of each researcher.

[0040] By combining two dimensions of evaluation, we consider both the richness of researchers' publications and the coverage of researchers matched with the validation data. This avoids the strategy selection bias caused by a single dimension. When the validation data quality is high, we adopt a lenient association analysis strategy to expand the coverage of related units. When the validation data quality is insufficient, we adopt a restrictive analysis strategy to reduce the difficulty of data processing.

[0041] Specifically, in the above steps, the verification matching value is determined by the average of the verification matching weight values ​​of each researcher and the proportion of researchers among the researchers who are verified matching. It is then determined whether the verification matching value is greater than a preset matching threshold. If so, the analysis and processing strategy for the related units of the target unit in the journal article is determined to be an association analysis strategy. As long as the number of researchers in the journal article who overlap with the target unit reaches or exceeds the first personnel number threshold, the association analysis and processing of the journal article is performed. If not, the analysis and processing strategy for the related units of the target unit in the journal article is determined to be a restriction analysis strategy. As long as the number of researchers in the journal article who overlap with the target unit reaches or exceeds the second personnel number threshold (greater than the first personnel number threshold), the association analysis and processing of the journal article is performed.

[0042] The verification match value is calculated as the average verification match weight value multiplied by the percentage of verifications. This value comprehensively assesses the quality of the verification data. Its significance lies in employing a lenient association analysis strategy when the verification match value exceeds a threshold, and a stricter strategy (with overlap exceeding a second threshold) when the verification match value is less than or equal to the threshold. This quantitative calculation of the verification match value automates the determination of the analysis and processing strategy, avoiding the subjectivity and inconsistency of manual judgment. Furthermore, the verification match value integrates both paper richness and verification coverage, making strategy selection more precise. It maximizes the identification coverage while ensuring the reliability of associated unit identification, thereby improving the efficiency and accuracy of associated unit update processing.

[0043] S2 uses the analysis and processing strategy and the number of overlaps between journal articles and researchers in the target unit to determine the comparative analysis targets in the journal articles for updating related units. Based on the analysis and processing strategy of related units of the target unit in the journal articles and the degree of identification and matching of related units in the comparative analysis targets, the elimination management strategy of the identified targets among the researchers is determined. Furthermore, the target of the comparative analysis used for updating the associated units in the journal articles is the number of researchers overlapping with the target unit, which is a journal article that meets the quantity threshold corresponding to the analysis and processing strategy of the associated units of the target unit in the journal article.

[0044] Specifically, the degree of identification and matching of related units in the comparison analysis target is determined based on whether the unit of the comparison analysis target belongs to related units. Specifically, based on the unique scientific research project number, whether the unit of the same experimental project, experimental data and paper is determined to be related to the unit of the comparison analysis target, on the one hand, it realizes the identification of abnormal behavior of duplicate submissions, and on the other hand, it realizes the updating of related units of the target unit, thereby identifying and analyzing abnormal behavior of duplicate submissions.

[0045] It should be noted that the method for determining the elimination management strategy for the identified targets among the researchers is as follows: In this embodiment, the efficiency of identifying related units and the frequency of comparison analysis are determined based on the degree of identification and matching of related units in the comparison analysis targets and the number of comparison analysis targets. Furthermore, the efficiency and reliability of the identification analysis are determined by combining the analysis and processing strategies of related units of target units in journal articles. Based on the efficiency of the identification analysis and the frequency of comparison analysis, the elimination management strategy for the identification targets among researchers is determined. This approach minimizes the difficulty of comparison analysis while ensuring the reliability of the updated identification of related units.

[0046] The comparative analysis target refers to journal articles whose number of researchers overlapping with the target institution meets the threshold corresponding to the analysis and processing strategy of related institutions of the target institution in journal articles. The degree of matching is determined based on whether the institution of the comparative analysis target is a related institution. This is determined by unique research project numbers, the same experimental project and experimental data, and the articles themselves. This serves two purposes: firstly, to identify abnormal behavior of duplicate submissions, and secondly, to update the related institutions of the target institution.

[0047] By comparing and analyzing the degree and number of related units in the target, the elimination management strategy is determined. On the one hand, by selectively eliminating researchers with insufficient verification data and low contribution, the number of researchers that need to be compared in the process of identifying related units is reduced while minimizing the difficulty of comparison and analysis. On the other hand, by setting elimination thresholds in combination with the different types of analysis and processing strategies, the reliability of the updated identification and processing of related units is ensured, and incomplete identification of related units due to blind elimination is avoided.

[0048] A journal article has 3 overlapping researchers with the target institution (reaching the threshold of 3 for the limited analysis strategy). This article is the target of the comparison analysis. The unique research project number, experimental project and experimental data, and article description are used to determine whether the institution is an affiliated institution of the target institution. For example, if the research project number is the same, the same experimental project and experimental data are used, and the article description contains descriptions of the institution's research projects, it can be identified as an affiliated institution.

[0049] S21 determines the comparison analysis targets of the related units belonging to the target unit in the comparison analysis targets based on the degree of identification and matching of the related units in the comparison analysis targets, and takes the comparison analysis targets of the related units belonging to the target unit in the comparison analysis targets as the association analysis targets.

[0050] Association analysis target = comparison analysis target that belongs to the related units of the target unit in the comparison analysis target.

[0051] By screening out the target of the comparative analysis, the association analysis targets are identified, and papers whose institutions truly belong to the target institution are identified, providing a basis for subsequent elimination judgment. The institution affiliation of the comparative analysis targets is determined by objective criteria such as unique scientific research project numbers, the same experimental project, and experimental data. This can not only accurately identify abnormal behavior of duplicate submissions, but also promptly discover and update the names of the target institution's associated institutions, keeping the associated institution database up-to-date.

[0052] In the past 30 days, there have been 10 comparative analysis target papers. Among them, 3 papers were identified by their research project numbers as belonging to affiliated institutions of the target institution. These 3 papers are the targets of the association analysis.

[0053] It is understood that step S21 includes the following: determining whether the number of associated analysis targets within the most recent preset time period is greater than the preset threshold for the number of associated targets. If so, the elimination management strategy for the identified targets among the researchers is determined to be no need for updating, thereby ensuring the reliability of the identification of associated units. If not, proceed to step S22.

[0054] The system determines whether the number of related analysis targets within the most recent preset time period is greater than the preset threshold for the number of related targets. When the number of related analysis targets is large, the identification of related units remains effective and no update processing is required. When the number is small, the system proceeds to step S22 for further judgment. When the number of related analysis targets is large recently, it indicates that the identification process of related units is running normally and can continuously discover new related unit information. In this case, maintaining the existing set of researchers can ensure the reliability of the identification of related units and avoid the technical problem of unreliable updates of related units caused by unnecessary elimination processing.

[0055] S22 determines the number of comparison analysis targets based on the aforementioned comparison analysis targets.

[0056] By statistically comparing and analyzing the number of targets, we can accurately assess the current workload of identifying and processing related units, providing data to determine whether researchers need to be eliminated, and avoiding unnecessary elimination when the comparison and analysis frequency is low.

[0057] In the above steps, the number of comparison and analysis targets in the most recent preset time period is obtained, and it is determined whether the number of comparison and analysis targets in the most recent preset time period is greater than the preset threshold for the number of comparison and analysis targets. If yes, proceed to step S23; otherwise, determine that the elimination management strategy for the identified targets among the researchers is that no update processing is required, thereby ensuring the reliability of the identification of related units.

[0058] The frequency of comparison analysis is determined by comparing the number of targets. The significance is that when the number is greater than the threshold, the comparison analysis is frequent, and the process proceeds to step S23 for elimination. When the number is less than or equal to the threshold, the comparison analysis is infrequent and no update is required. Maintaining the stability of the research personnel set when the comparison analysis is infrequent ensures the reliability and continuity of the identification of related units, while avoiding the management complexity and waste of computing resources caused by frequent updates to the research personnel set.

[0059] S23 determines the elimination management strategy for the identified targets among the researchers by using the number of correlation analysis targets within the most recent preset time period, the number of comparison analysis targets, and the analysis and processing strategy of the related units of the target units in the journal articles.

[0060] Based on frequent comparative analysis, S23 makes the final elimination judgment through the identification matching coefficient of related units and analysis and processing strategies. By jointly evaluating the quantitative relationship between the correlation analysis target and the comparison analysis target and the type of analysis and processing strategy, the elimination management strategy is determined by considering both the matching degree of the identification of related units and the strictness of the analysis and processing strategy, thus achieving precise optimization of the research personnel group.

[0061] It is understood that step S23 includes steps S231 and S232, specifically including: S231 determines the identification matching coefficient of the associated unit based on the number of related analysis targets and the number of comparison analysis targets within the most recent preset time period, and judges whether the identification matching coefficient of the associated unit is greater than the preset matching coefficient threshold. If not, proceed to step S232. If yes, determine that the elimination management strategy for the identified targets among the researchers is no longer required to update, thereby ensuring the reliability of the identification of the associated unit.

[0062] The identification matching coefficient of the associated unit = number of associated analysis targets ÷ number of comparison analysis targets. The identification matching coefficient of the associated unit determines the degree of matching of the associated unit identification. Its significance is that when the identification matching coefficient > the threshold, the degree of matching of the associated unit identification is high and no update processing is required; when the identification matching coefficient ≤ the threshold, the degree of matching is low and the process proceeds to S232 for elimination judgment. When the identification matching coefficient of the associated unit is low, it means that the proportion of the current comparison analysis targets that truly belong to the target unit is not high. At this time, the process needs to proceed to step S232 for further evaluation and analysis processing.

[0063] S232 determines whether the analysis and processing strategy for the related units of the target unit in the journal article belongs to the association analysis strategy. If yes, the elimination management strategy for the research personnel identification target is determined as follows: if the verification matching weight value of the research personnel is less than the preset weight threshold, the research personnel are determined to be personnel who do not need to be referenced in the analysis and processing of the journal article. That is, no processing is needed when considering the number of overlapping research personnel in the analysis and processing of the related units of the target unit in the journal article. If no, the elimination management strategy for the research personnel identification target is determined as follows: if the verification matching weight value of the research personnel is less than the set weight threshold (greater than the preset weight threshold), the research personnel are determined to be personnel who do not need to be referenced in the analysis and processing of the journal article. That is, no processing is needed when considering the number of overlapping research personnel in the analysis and processing of the related units of the target unit in the journal article.

[0064] The severity of elimination criteria is determined by analyzing the type of analysis strategy. When the analysis strategy is a correlation analysis strategy (relaxed), there are more targets for comparison analysis, meaning that a stricter preset weight threshold is used to eliminate more low-validation personnel. When the analysis strategy is a restriction analysis strategy (strict), a more lenient weight threshold is used to eliminate fewer personnel to ensure the reliability of identification. Under the correlation analysis strategy, a lower elimination threshold eliminates more researchers with insufficient validation data, reducing the computational load of comparison analysis and improving the efficiency of related unit identification. Under the restriction analysis strategy, a higher elimination threshold eliminates only researchers with severely insufficient validation data, ensuring the reliability of related unit identification and avoiding omissions in related unit identification due to excessive elimination.

[0065] S3 uses the elimination management strategy to identify researchers among the researchers who do not need to be referenced when calculating the number of overlapping researchers in the analysis and processing of the target unit's related units in journal articles, and identifies them as eliminated researchers. Based on the updated processing results of the eliminated researchers and the research paper data of researchers in the target unit excluding eliminated researchers, an AI model-based journal association analysis method is determined.

[0066] Furthermore, the method for determining the journal association analysis method based on the AI ​​model is as follows: In this application, the delay in the current journal correlation analysis is determined based on the update processing results of eliminated researchers. Combined with the number of remaining researchers and the number of research papers by each remaining researcher, the reliability of the update identification processing of related units is determined. Based on the delay in the current journal correlation analysis and the reliability of the update identification processing of related units, the AI ​​model-based journal correlation analysis method is determined. This reduces the difficulty of correlation analysis while ensuring the reliability and real-time nature of the identification and processing of abnormal behaviors in journal papers.

[0067] "Researchers removed" refers to those who do not need to be considered when calculating the number of overlapping researchers in the analysis of related institutions of the target institution in journal articles. "Delay level" refers to the degree of delay in journal relevance analysis based on the analysis strategy for related institutions of the target institution in journal articles; the more researchers removed, the higher the delay level. Abnormal behavior includes duplicate submissions and data fabrication.

[0068] The journal correlation analysis method is determined by updating the results of the elimination of researchers and considering the situation of the remaining researchers. Its advantages are twofold: firstly, it dynamically assesses the latency of correlation analysis based on the number of eliminated researchers, employing real-time full analysis to ensure the real-time nature of anomaly identification when the latency is high; secondly, it assesses the reliability of the correlation unit update and identification process based on the number of remaining researchers and the number of papers, using pre-set cycle batch analysis to reduce the difficulty of analysis and computational consumption when the reliability is high. Thus, while reducing the difficulty of correlation analysis, it simultaneously ensures the reliability and real-time nature of identifying and handling anomalies such as duplicate submissions and data fabrication in journal articles.

[0069] S31 determines the number of researchers to be eliminated based on the updated processing results of the eliminated researchers.

[0070] By statistically analyzing the number of researchers eliminated, we can provide basic data for assessing the delay in subsequent correlation analysis. By statistically analyzing the number of researchers eliminated, we can accurately quantify the scale of personnel that do not need to be referenced in the correlation analysis, providing an objective basis for assessing the delay in the analysis, and thus rationally selecting the processing method of real-time analysis or periodic analysis.

[0071] Understandably, in step S31, it is determined whether the number of researchers to be eliminated exceeds the preset threshold for the number of researchers to be eliminated. If so, the delay in journal correlation analysis based on the analysis and processing strategy of the target unit's related units in the journal article is relatively high. Therefore, the journal correlation analysis method based on the AI ​​model is determined to perform real-time correlation analysis on the research papers of all researchers in the target unit to determine whether there are abnormal situations such as duplicate submissions. That is, if the journal article belongs to the related unit or the target unit, and the author of the journal article is a researcher in the target unit, correlation analysis is performed. If not, the process proceeds to step S32.

[0072] The latency of correlation analysis is determined by the number of researchers eliminated. When the number exceeds a threshold, the latency is high. Method 1 is adopted: real-time correlation analysis is performed on the research papers of all researchers in the target institution. When the number of eliminated researchers is large, continuing to use periodic batch analysis will lead to a high latency in journal correlation analysis, which may miss abnormal behaviors such as duplicate submissions. By adopting full real-time analysis, the latest submitted papers can be immediately correlated, abnormal behaviors can be detected in a timely manner, and the real-time identification and handling of abnormal behaviors can be guaranteed.

[0073] For example, if the threshold of eliminating 6 researchers is greater than 5, then Method 1 is adopted: real-time correlation analysis is performed on the research papers of all researchers (P1 to P10) in the target unit. As long as the journal article belongs to the related unit or the target unit and the author is a researcher in the target unit, the correlation analysis is performed.

[0074] S32 defines the remaining researchers as those in the target unit, excluding those who were eliminated, and determines the number of remaining researchers.

[0075] Determining the number of remaining researchers provides fundamental data for assessing the reliability of subsequent correlation analysis. By statistically analyzing the number of remaining researchers, we can accurately assess the scale of researchers still participating in correlation analysis within the target institution, providing an objective basis for judging the reliability of the analysis and thus enabling the rational selection of full-cycle analysis or differentiated analysis methods.

[0076] Specifically, in step S32, the number of remaining researchers is used to determine whether the number of remaining researchers is greater than a preset value. If so, the reliability of the correlation analysis is relatively high. The AI ​​model-based journal correlation analysis method is determined to perform real-time correlation analysis on the research papers of all researchers in the target unit according to a preset cycle to determine whether there are any abnormalities such as duplicate submissions. That is, if the journal paper belongs to the related unit or the target unit, and the author of the journal paper is a researcher in the target unit, correlation analysis is performed. If not, the process proceeds to step S33.

[0077] The reliability of the correlation analysis is judged by the number of remaining researchers. When the number of remaining researchers is greater than the threshold, the reliability is relatively high. Method 2 is adopted: real-time correlation analysis is performed on the research papers of all researchers in the target unit according to a preset period of 7 days. When the number of remaining researchers is sufficient, the data foundation for the updated identification and processing of the related units is sufficient and the reliability is relatively high. By adopting batch analysis with a preset period, the reliability of the correlation analysis results can be guaranteed, and the computational power consumption caused by real-time analysis of each paper can be reduced, thus achieving a balance between analysis efficiency and reliability.

[0078] S33 determines the journal association analysis method based on the number of eliminated researchers, the number of remaining researchers, and the number of research papers published by each remaining researcher.

[0079] When the number of remaining researchers is insufficient, S33 makes the final determination by using the updated matching coefficient of the associated institutions and the matching coefficient of the association analysis. By jointly evaluating the sum of the verification matching weight values ​​of the remaining researchers and the adequacy of the coverage of associated institutions, the journal association analysis method can be accurately selected. When the reliability is high, a full analysis with a preset cycle is used to reduce the processing difficulty. When the reliability is insufficient, a differentiated analysis (real-time analysis of eliminated researchers and periodic analysis of non-eliminated researchers) is used to balance real-time performance and efficiency.

[0080] Specifically, step S33 includes: S331 determines the update matching coefficient of the associated unit based on the number of remaining researchers and the number of research papers of each remaining researcher. It then determines whether the update matching coefficient of the associated unit is greater than a preset matching coefficient threshold. If so, the AI ​​model-based journal association analysis method is determined to perform real-time association analysis processing on the research papers of all researchers in the target unit according to a preset cycle to determine whether there are any abnormal situations such as duplicate submissions. That is, if the journal paper belongs to the associated unit or the target unit, and the author of the journal paper is a researcher in the target unit, then association analysis processing is performed. If not, then proceed to step S332.

[0081] The update matching coefficient of the associated unit is equal to the sum of the verification matching weights of the remaining researchers. The reliability of the updated identification of associated units is judged by the update matching coefficient. When the update matching coefficient is greater than the threshold, the reliability is relatively high, and Method 2 (preset cycle full analysis) is adopted. When the update matching coefficient is less than or equal to the threshold, it is transferred to S332 for further judgment. When the sum of the verification matching weights of the remaining researchers is high, it indicates that the remaining researchers have abundant paper data and the data foundation for the updated identification of associated units is reliable. The preset cycle full analysis can ensure the reliability of identification and reduce the processing difficulty and computing power consumption of paper-by-paper real-time analysis. When the update matching coefficient is insufficient, it is transferred to S332 to avoid blindly adopting full analysis when the data foundation is insufficient.

[0082] Specifically, the update matching coefficient of the associated unit is determined based on the sum of the verification matching weight values ​​of each remaining researcher.

[0083] S332 determines the matching coefficient of the correlation analysis based on the number of associated units and the number of eliminated researchers, and judges whether the matching coefficient of the correlation analysis is greater than a preset coefficient threshold. If so, the AI ​​model-based journal correlation analysis method is determined to perform real-time correlation analysis processing on the research papers of all researchers of the target unit according to a preset cycle to determine whether there are abnormal situations such as duplicate submissions. That is, if the journal paper belongs to an associated unit or the target unit, and the author of the journal paper is a researcher of the target unit, correlation analysis processing is performed; otherwise, the AI ​​model-based journal correlation analysis method is determined to be... The correlation analysis method involves performing real-time correlation analysis on the research papers of researchers eliminated from the target institution to determine whether there are any anomalies such as duplicate submissions. Specifically, if a journal article belongs to an affiliated institution or the target institution, and the author of the journal article is a researcher from the target institution, correlation analysis is performed. If the researcher does not belong to the eliminated institution, real-time correlation analysis is performed according to a preset cycle to determine whether there are any anomalies such as duplicate submissions.

[0084] The matching coefficient for correlation analysis is calculated as K1 multiplied by the number of associated units, followed by K2 multiplied by the number of researchers eliminated. K1 and K2 are weighting coefficients, which are determined based on the analytic hierarchy process (AHP).

[0085] The sufficiency of associated unit coverage is determined by the matching coefficient of the correlation analysis. When the matching coefficient is greater than the threshold, the associated unit coverage is sufficient, and Method 2 (preset cycle full analysis) is used. When the matching coefficient is less than or equal to the threshold, Method 3 is used: real-time correlation analysis is performed on the research papers of eliminated researchers, while non-eliminated researchers are analyzed according to a preset cycle. When the matching coefficient of the correlation analysis is high, it indicates that the associated unit coverage is sufficient and the overall identification reliability is high, so full analysis with a preset cycle is used to reduce the processing difficulty. When the matching coefficient is low, it indicates that the associated unit coverage is insufficient. Real-time analysis is used to capture abnormal behavior in a timely manner for eliminated researchers (researchers with high analysis and processing delays), while periodic analysis is used for non-eliminated researchers to balance computing power consumption. This ensures the real-time identification and processing of abnormal behavior while reducing the overall analysis and processing difficulty.

[0086] Specifically, the more related units there are and the fewer researchers are eliminated, the higher the matching coefficient of the correlation analysis will be.

[0087] The positive correlation factors of the matching coefficient in correlation analysis have been identified, providing a directional reference for setting the coefficient threshold. By identifying the positive correlation factors of the matching coefficient, we can guide the reasonable setting of the matching coefficient threshold based on the number of associated units and the number of researchers eliminated during the implementation process, making the selection of analysis methods interpretable and adjustable.

[0088] The target institution has a total of 10 researchers, P1 to P10, with the following number of journal articles: P1=8, P2=7, P3=6, P4=5, P5=4, P6=3, P7=3, P8=2, P9=2, P10=1.

[0089] S11, verify that the matched researchers are P1, P2, P3 and P4, a total of 4 people. The verification ratio = 4 ÷ 10 = 0.40, which is not greater than the threshold of 0.50. Proceed to S12.

[0090] S12 — Verify that the mean of the matching weight value is 0.41 and is not greater than the threshold of 0.45. Determine the adoption of the restricted analysis strategy: when the number of journal articles and the number of researchers in the target unit that overlap reaches more than 3 of the second personnel number threshold, the association analysis process will be carried out.

[0091] Following S1, the analysis strategy is a restricted analysis strategy. The comparison analysis targets N articles (journal articles with more than a threshold of overlap with researchers from the target institution) within the last 30 days.

[0092] S21 – Based on unique research project numbers, the same experimental project and experimental data, and paper data, if 3 out of N comparative analysis targets belong to affiliated units of the target unit, then the number of association analysis targets is 3. If the number of association analysis targets in the last 30 days is greater than 2, then the elimination management strategy for the identified targets is determined to be no need for updating.

[0093] Following S2, 6 researchers from P5 to P10 were eliminated, leaving 4 researchers from P1 to P4.

[0094] S31—The threshold of eliminating 6 researchers is greater than 5. Method 1 is adopted: real-time correlation analysis is performed on the research papers of all researchers (P1 to P10) in the target unit.

[0095] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0096] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0097] The above description is merely one or more embodiments of this specification and is not intended to limit this specification. Various modifications and variations can be made to the one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of this specification.

Claims

1. A journal association analysis system based on multi-source data and AI models, characterized in that, Specifically, it includes: Related unit identification module, related unit update module, and related analysis module; The associated unit identification module is responsible for determining the associated units of journal articles of the target unit based on the researchers' journal article data. The associated unit update module is responsible for determining the analysis and processing strategy of associated units of the target unit in the journal articles based on the journal article data of each staff member in the target unit. Using the analysis and processing strategy and the number of overlaps between the journal articles and the researchers of the target unit, the module determines the comparative analysis target in the journal articles for associated unit update. The correlation analysis module will exclude researchers who do not need to be referenced when analyzing related units as eliminated researchers. Based on the updated processing results of the eliminated researchers and the research paper data of researchers in the target units other than the eliminated researchers, the module will determine the journal correlation analysis method based on the AI ​​model, and use the journal correlation analysis method and multi-source data to perform correlation analysis processing of journal papers.

2. The journal association analysis system based on multi-source data and AI models as described in claim 1, characterized in that, The journal article data of the researchers is determined based on the journal article data entered by the researchers in the research management system.

3. The journal association analysis system based on multi-source data and AI models as described in claim 1, characterized in that, The associated institution of the target institution's journal article is the institution name that may be marked in the journal article.

4. The journal association analysis system based on multi-source data and AI models as described in claim 1, characterized in that, The journal association analysis method and multi-source data are used to perform association analysis on journal articles, specifically including: Based on the journal association analysis method, when it is determined that the association analysis of the research papers of the researchers of the target institution needs to be carried out, the research papers of multiple journals in the history of the journal paper author are used as a basis. The AI ​​model is used to determine the duplication of the journal paper with the research papers of multiple journals in the history of the journal paper author. Based on the identification results of the duplication, it is determined whether there is any abnormal behavior.

5. The journal association analysis system based on multi-source data and AI models as described in claim 4, characterized in that, The abnormal behaviors include duplicate submissions and data fabrication.

6. A journal association analysis method based on multi-source data and AI models, applied to the journal association analysis system based on multi-source data and AI models as described in any one of claims 1-5, characterized in that, Specifically, it includes: Based on the journal article data of each staff member in the target unit, the verification matching researchers among the researchers are identified. Based on the verification matching researchers among the staff members and the journal article data of the researchers, the analysis and processing strategy for the related units of the target unit in the journal articles is determined. Using the aforementioned analysis and processing strategy and the number of overlaps between journal articles and researchers in the target institution, the comparative analysis targets in the journal articles used for updating related institutions are determined. Based on the analysis and processing strategy of related institutions of the target institution in the journal articles and the degree of identification and matching of related institutions in the comparative analysis targets, the elimination management strategy of the identified targets among the researchers is determined. Using the elimination management strategy, researchers who do not require reference when calculating the number of overlapping researchers in the analysis and processing of the target institution's related institutions in journal articles are identified and are designated as eliminated researchers. Based on the updated processing results of the eliminated researchers and the research paper data of researchers in the target institution excluding the eliminated researchers, an AI model-based journal association analysis method is determined.

7. The journal association analysis method based on multi-source data and AI models as described in claim 6, characterized in that, The researchers included in the verification matching are those whose number of journal articles exceeds a preset threshold.

8. The journal association analysis method based on multi-source data and AI model as described in claim 6, characterized in that, The method for determining the analysis and processing strategy for the related units of the target unit in the journal article is as follows: Based on the verification-matched researchers among the researchers, determine the number of verification-matched researchers among the researchers; The number of journal articles published by each researcher is determined based on their journal article data. Based on the number of researchers who validated and matched the research personnel and the number of journal articles published by each researcher, an analysis and processing strategy for the related units of the target unit in the journal articles was determined.

9. The journal association analysis method based on multi-source data and AI model as described in claim 6, characterized in that, The method for determining the journal association analysis method based on the AI ​​model is as follows: S31 determines the number of researchers to be eliminated based on the updated processing results of the eliminated researchers; S32 defines the remaining researchers as the researchers in the target unit excluding those who were eliminated, and determines the number of remaining researchers. S33 determines the journal association analysis method based on the number of eliminated researchers, the number of remaining researchers, and the number of research papers published by each remaining researcher.

10. The journal association analysis method based on multi-source data and AI model as described in claim 9, characterized in that, Based on the number of eliminated researchers, the number of remaining researchers, and the number of research papers published by each remaining researcher, the AI-based journal association analysis method is determined, specifically including: S331 uses the number of remaining researchers and the number of research papers of each remaining researcher to determine the update matching coefficient of the associated unit, and determines whether the update matching coefficient of the associated unit is greater than the preset matching coefficient threshold. If so, the journal association analysis method based on the AI ​​model is determined to perform real-time association analysis processing on the research papers of all researchers of the target unit according to the preset cycle to determine whether there are abnormal situations such as duplicate submissions. That is, if the journal paper belongs to the associated unit or the target unit, and the author of the journal paper is a researcher of the target unit, the association analysis processing is performed. If not, proceed to step S332. S332 determines the matching coefficient of the correlation analysis based on the number of associated units and the number of eliminated researchers, and judges whether the matching coefficient of the correlation analysis is greater than a preset coefficient threshold. If so, the AI ​​model-based journal correlation analysis method is determined to perform real-time correlation analysis processing on the research papers of all researchers of the target unit according to a preset cycle to determine whether there are abnormal situations such as duplicate submissions. That is, if the journal paper belongs to an associated unit or the target unit, and the author of the journal paper is a researcher of the target unit, correlation analysis processing is performed; otherwise, the AI ​​model-based journal correlation analysis method is determined to be... The correlation analysis method involves performing real-time correlation analysis on the research papers of researchers eliminated from the target institution to determine whether there are any anomalies such as duplicate submissions. Specifically, if a journal article belongs to an affiliated institution or the target institution, and the author of the journal article is a researcher from the target institution, correlation analysis is performed. If the researcher does not belong to the eliminated institution, real-time correlation analysis is performed according to a preset cycle to determine whether there are any anomalies such as duplicate submissions.

Images