A document duplication intelligent evaluation system and method based on big data model

Abstract

The present invention relates to the technical field of data processing, and discloses a document duplicate checking intelligent evaluation system and method based on a big data model, the method comprising: collecting document texts of an academic platform, extracting document features of each document text, establishing a document database, extracting document keywords in the document database, and training a random forest model, when the matching accuracy of the trained random forest model reaches a matching accuracy threshold, outputting a document domain model, determining the duplicate checking domain of a target duplicate checking document based on the document domain model, determining the corresponding duplicate checking document domain in the document database according to the duplicate checking domain, constructing a duplicate checking diagram based on the result of document processing, determining the suspected document corresponding to the target duplicate checking document according to the duplicate checking diagram, and determining the duplicate checking evaluation level of the target duplicate checking document according to the duplicate checking similarity. The present invention avoids the error of artificially determining the duplicate checking domain and ensures the accuracy of the duplicate checking evaluation by utilizing a big data model.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a system and method for intelligent evaluation of document duplication checking based on a big data model. Background Art

[0002] With the development of information technology and the Internet, the number of academic research documents is increasing, and document duplicate checking has become an important task in the field of academic research and publishing. Especially in the fields of scientific research and high-tech industries, how to effectively identify misconduct such as plagiarism, plagiarism and duplicate publication in documents has become an important part of maintaining academic integrity and promoting scientific research progress. At present, document duplicate checking usually relies on manual experience judgment, which makes it difficult to effectively identify the true semantics and disciplinary background of the document, ignores the hierarchical structure and semantic characteristics of the document, and leads to deviations in the positioning of the duplicate checking field when there are similar fields, thus affecting the accuracy of the duplicate checking evaluation.

[0003] Therefore, it is necessary to design an intelligent evaluation system and method for document duplication checking based on big data models to solve the problems existing in current technologies. Summary of the invention

[0004] In view of this, the present invention proposes an intelligent evaluation system and method for document duplication checking based on a big data model, aiming to solve the problem that document duplication checking relies on manual experience judgment, is difficult to effectively identify the true semantics and subject background of the document, ignores the hierarchical structure and semantic characteristics of the document, and leads to deviations in the positioning of the duplication checking field when similar fields exist, thereby affecting the accuracy of the duplication checking evaluation.

[0005] In one aspect, the present invention proposes a method for intelligent evaluation of document duplication based on a big data model, comprising: Collect document texts from the academic platform, extract document features of each document text, distribute and store all document texts according to the document features, and establish a document database; Extracting keywords from the document database and training a random forest model; when the matching accuracy of the trained random forest model reaches a matching accuracy threshold, outputting a document domain model, preprocessing the document to be checked for duplicates, determining a target document to be checked for duplicates, determining a check for duplicates domain for the target document to be checked for duplicates based on the document domain model, and determining a corresponding check for duplicates document domain in the document database based on the check for duplicates domain; Obtain all duplicate check document texts in the duplicate check document field, perform document processing on all duplicate check document texts, construct a duplicate check diagram based on the document processing results, determine suspected documents corresponding to the target duplicate check document according to the duplicate check diagram, compare the suspected documents with the target duplicate check document, and determine the duplicate check similarity; The plagiarism check evaluation level of the target plagiarism check document is determined according to the plagiarism check similarity.

[0006] Furthermore, when collecting document texts from the academic platform, extracting document features of each document text, distributing and storing all document texts according to the document features, and establishing a document database, the following steps are included: Count the same literature texts on the academic platform, retain the largest literature texts among the same literature texts, and remove the remaining literature texts; Constructing a duplicate check text set by combining the retained document text and different document texts, analyzing the duplicate check text set based on natural language processing, and removing stop words from the duplicate check text set; A dependency analysis model is used to analyze the relationship between words in each document text in the deleted duplicate check text set, determine noun phrases, use noun phrases that conform to the syntactic structure as document keywords, and use the document keywords of each document text as the document features; The document texts containing a preset number of identical document features are counted and an analog document library is established; the document texts not containing a preset number of identical document features are counted and an approximate document library is established; and the document texts that cannot be counted are established and stored in a difference document library; The document database includes the analogy document library, the approximate document library and the difference document library.

[0007] Furthermore, when extracting the keywords of the documents in the document database and training the random forest model, when the matching accuracy of the trained random forest model reaches the matching accuracy threshold, the document domain model is output, including: Obtain all document features of the analogy document library, record them as analogy features, obtain all document features of the approximate document library, record them as similarity features, obtain all document features of the difference document library, record them as difference features, and construct a feature data set with the analogy features, the similarity features and the difference features; The feature data set is sampled according to a preset ratio to obtain a training set and a test set, a grid search is used to find model establishment parameters of the random forest model, and a random forest model is constructed; Fitting the random forest model using the training set, substituting the test set into the fitted random forest model and determining the matching accuracy, and presetting a matching accuracy threshold; When the matching accuracy is greater than or equal to the matching accuracy threshold, the random forest model is determined as a document domain model; When the matching accuracy is less than the matching accuracy threshold, the grid search is continued to be used to re-search the model establishment parameters of the random forest model, the random forest model is re-established, the training set is fitted with the re-determined random forest model, the test set is substituted into the re-fitted random forest model and the matching accuracy is re-determined, until the determined matching accuracy is greater than or equal to the matching accuracy threshold, and the document domain model is output.

[0008] Furthermore, when the document to be checked for duplicates is preprocessed to determine the target document to be checked for duplicates, and the checking field of the target document to be checked for duplicates is determined based on the document field model, it includes: Comparing the document format of the document to be checked for duplicate content with the standard document format, if the document format is different from the standard document format, converting the document format to the standard document format, and if the document format is the same as the standard document format, not performing format conversion; Preprocessing the documents to be checked for duplicates after the comparison results, wherein the preprocessing includes removing stop words, unifying the case of punctuation marks, deleting the authors of the documents, and deleting the time of the documents, and determining the target document to be checked for duplicates based on the preprocessing results; The target duplicate check document is parsed using the dependency analysis model to determine the target document features of the target duplicate check document, and the target document features are substituted into the document domain model to determine the duplicate check domain of the target duplicate check document.

[0009] Furthermore, when all the duplicate check document texts are processed and a duplicate check diagram is constructed based on the document processing results, it includes: The document processing includes deleting the author of the document and deleting the time of the document, and obtaining similar documents based on the document processing results; Establish a duplicate checking node for each similar document and obtain the page number of each similar document; If there is a similar document citing another similar document, a sub-reference node of the duplicate checking node is generated for the cited similar document, and a sub-feature node of the duplicate checking node is generated for each document feature of each similar document; A duplicate checking diagram is constructed according to the duplicate checking node, the sub-reference node and the sub-feature node.

[0010] Furthermore, when determining the suspected document corresponding to the target duplicate check document according to the duplicate check diagram, it includes: The suspected value of each duplicate checking node is determined according to the duplicate checking diagram, and the suspected value is obtained by the following formula: ; in, Indicates The suspected value of the duplicate check node, Indicates The number of child reference nodes of a duplicate check node, Indicates The number of sub-feature nodes of a duplicate check node, Indicates The number of pages of similar documents corresponding to each duplicate check node; The suspected document corresponding to the target duplicate check document is determined according to the suspected value of each duplicate check node.

[0011] Furthermore, when determining the suspected document corresponding to the target duplicate check document according to the suspected value of each duplicate check node, it includes: If the largest suspected value is unique, the similar document represented by the duplicate checking node corresponding to the largest suspected value is used as the suspected document corresponding to the target duplicate checking document; If the largest suspected value is not unique, all similar documents represented by the duplicate check node corresponding to the largest suspected value are counted, and the similar document with the least number of pages is used as the suspected document corresponding to the target duplicate check document.

[0012] Furthermore, when comparing the suspected document with the target duplicate check document to determine the duplicate check similarity, it includes: The similarity of the duplicate check is obtained by the following formula: ; in, Indicates the similarity of duplicate checking. Indicates the page number of the suspected document, Indicates the number of pages of the target document to be checked for duplicate content. Indicates the maximum number of pages in the suspected document and the target duplicate check document. Indicates the document characteristics of the suspected document, Indicates the target document characteristics of the target duplicate check document, Indicates the number of words in the suspected document. Indicates the number of words in the target document. , , and represents the weight coefficient, and , , and The value range is (0, 1].

[0013] Furthermore, when determining the plagiarism evaluation level of the target plagiarism check document according to the plagiarism check similarity, it includes: Presetting a first preset duplication check similarity and a second preset duplication check similarity, wherein the first preset duplication check similarity is greater than the second preset duplication check similarity; When the plagiarism similarity is greater than or equal to the first preset plagiarism similarity, the plagiarism evaluation level of the target plagiarism document is determined as a first-level plagiarism; When the plagiarism check similarity is less than the first preset plagiarism check similarity and greater than or equal to the second preset plagiarism check similarity, the plagiarism check evaluation level of the target plagiarism check document is determined as a second-level plagiarism check; When the similarity of the duplicate check is less than the second preset similarity of the duplicate check, the evaluation level of the duplicate check of the target duplicate check document is determined as the third level of duplicate check; The duplication degrees of the first-level duplication check, the second-level duplication check and the third-level duplication check decrease in sequence.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: by collecting the document texts of the academic platform and extracting the document features, the document texts of various fields can be fully stored, so that the duplicate check evaluation can be carried out within a unified framework, thereby ensuring the accuracy of the duplicate check evaluation, by training the random forest model with the document keywords in the document database, using the big data model to learn the relationship therein, thereby outputting the document field model, so that the document duplicate check can intelligently identify the duplicate check field of the target duplicate check document, avoiding the deviation in the duplicate check field caused by human subjective judgment, and improving the accuracy of the duplicate check evaluation, constructing a duplicate check diagram that can intuitively display the correlation between the processing results of each document, accurately deriving the suspected document corresponding to the target duplicate check document, making the duplicate check evaluation have good pertinence, thereby achieving in-depth analysis. The duplicate check evaluation level of the target duplicate check document is dynamically determined according to the duplicate check similarity, thereby improving the reliability of the duplicate check evaluation.

[0015] On the other hand, the present application also provides a document duplicate checking intelligent evaluation system based on a big data model, which is used to apply the above-mentioned document duplicate checking intelligent evaluation method based on a big data model, including: The document collection module is configured to collect document texts from the academic platform, extract document features of each document text, distribute and store all document texts according to the document features, and establish a document database; The duplicate checking domain module is configured to extract keywords of documents in the document database and train a random forest model. When the matching accuracy of the trained random forest model reaches a matching accuracy threshold, the document domain model is output, the document to be checked for duplicates is preprocessed, the target duplicate checking document is determined, the duplicate checking domain of the target duplicate checking document is determined based on the document domain model, and the corresponding duplicate checking document domain in the document database is determined according to the duplicate checking domain; The duplicate checking analysis module is configured to obtain all duplicate checking document texts in the duplicate checking document field, perform document processing on all duplicate checking document texts, construct a duplicate checking diagram based on the document processing results, determine suspected documents corresponding to the target duplicate checking document according to the duplicate checking diagram, compare the suspected documents with the target duplicate checking document, and determine the similarity of the duplicate checking; The plagiarism checking evaluation module is configured to determine the plagiarism checking evaluation level of the target plagiarism checking document according to the plagiarism checking similarity.

[0016] It can be understood that the above-mentioned intelligent evaluation system and method for document duplication checking based on a big data model have the same beneficial effects and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings: Figure 1 A flowchart of a method for intelligent evaluation of document duplication based on a big data model provided by an embodiment of the present invention; Figure 2 A functional block diagram of a document duplication checking intelligent evaluation system based on a big data model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features described in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0019] In some embodiments of the present application, see Figure 1 As shown, a method for intelligent evaluation of document duplication based on a big data model includes: S100: Collect document texts from the academic platform, extract document features of each document text, distribute and store all document texts according to the document features, and establish a document database.

[0020] S200: Extract keywords from the document database and train a random forest model. When the matching accuracy of the trained random forest model reaches the matching accuracy threshold, the document domain model is output, the document to be checked for duplicates is preprocessed, the target document to be checked for duplicates is determined, the target document to be checked for duplicates is determined based on the document domain model, and the corresponding document domain to be checked for duplicates in the document database is determined based on the check for duplicates domain.

[0021] S300: Acquire all duplicate check document texts in the duplicate check document field, perform document processing on all duplicate check document texts, construct a duplicate check diagram based on the result of the document processing, determine the suspected document corresponding to the target duplicate check document according to the duplicate check diagram, compare the suspected document with the target duplicate check document, and determine the duplicate check similarity.

[0022] S400: Determine the evaluation level of the target document to be checked for plagiarism according to the similarity of the plagiarism check.

[0023] Specifically, we use tools such as web crawlers to collect document texts from academic platforms, and extract the document features and distribute them in the established document database, which effectively avoids missing documents and ensures the diversity of documents in the document database. It can comprehensively consider the results of all relevant fields and research directions, thereby improving the coverage and accuracy of document duplicate checking, and thus improving the accuracy of the evaluation. By extracting keywords from the document database and training with the random forest model, it can automatically learn and identify the research fields and content characteristics of different documents. The random forest model is an integrated learning model that can automatically learn the features and patterns between data from a large amount of data, thereby approaching the global optimal solution. By training the random forest model, when the model matching accuracy reaches the set matching accuracy threshold, it indicates that the model has good stability and learning ability at this time, and then the trained random forest model is output, that is, the document domain model. When using the document domain model for prediction, the document to be checked for duplicates is first preprocessed to ensure that the target duplicate check document meets the requirements of the duplicate check. The duplicate check domain of the target duplicate check document is determined based on the document domain model, reducing the error of human intervention. For example, although high-voltage circuit breakers and low-voltage circuit breakers belong to the same category of circuit breakers, their actual application fields and scenarios are completely different. Using the document domain model to identify different categories of fields from a large amount of data can accurately locate the duplicate check field to which the target duplicate check document belongs, so as to find the relevant documents in the corresponding duplicate check document field in the document database, thereby improving the pertinence and accuracy of document duplicate check. According to the determined field of duplicate-checked documents, all duplicate-checked document texts are found in the document database to construct a duplicate-checking diagram, which avoids the risk of missing suspected documents and intuitively presents the relationship between the target duplicate-checked document and the duplicate-checked document text, so as to find suspected documents that are relatively close to the target duplicate-checked document. By comparing the target duplicate-checked document with the suspected document, the duplicate-checking similarity is obtained, and the duplicate-checking assessment level of the target duplicate-checked document is determined, which avoids the influence of subjective factors on human duplicate-checking judgment, resulting in inaccurate duplicate-checking assessment, and improves the credibility of the duplicate-checking assessment.

[0024] It can be understood that by using the document domain model to determine the target document to be checked for plagiarism, the plagiarism checking process becomes targeted, avoiding the accumulation of errors caused by manual plagiarism checking ignoring subtle differences between plagiarism checking fields or relying on manual judgment, thereby improving the accuracy of plagiarism checking assessment.

[0025] In some embodiments of the present application, when collecting literature texts from an academic platform, extracting the literature features of each literature text, and performing distributed storage on all literature texts according to the literature features to establish a literature database, it includes: counting the same literature texts on the academic platform, retaining the literature text with the largest literature size among the same literature texts, and deleting the remaining literature texts. Constructing a duplicate-check text set from the retained literature texts and different literature texts, analyzing the duplicate-check text set based on natural language processing, deleting the stop words in the duplicate-check text set, using a dependency analysis model to parse the relationships between the words in each literature text in the duplicate-check text set after deletion, determining noun phrases, taking the noun phrases that conform to the syntactic structure as literature keywords, and taking the literature keywords of each literature text as literature features. Counting the literature texts with a preset number of the same literature features and establishing an analogous literature library, counting the literature texts without a preset number of the same literature features and establishing an approximate literature library, and storing the literature texts that cannot be counted by establishing a differential literature library. The literature database includes an analogous literature library, an approximate literature library, and a differential literature library.

[0026] Specifically, there may be the same literature texts on the academic platform. For example, the same literature text is recorded on both CNKI and X-MOL, and duplicate storage may occur during the storage process. Therefore, the same literature texts on the academic platform are counted. When the literature texts are uploaded to the academic platform, they will be affected by the mechanisms of the academic platform itself. For example, the encoding and file compression of the academic platform. For the same literature texts, only the literature text with the largest literature size is retained, so as to be closer to the original literature at the time of initial upload, ensuring that the literature texts included in the literature database are complete. If the literature sizes of multiple literature texts are the same, any one of them can be randomly selected. Constructing a duplicate-check text set from the retained literature texts and the literature texts without the same literature texts, and performing natural language processing to delete the stop words (such as meaningless words like "of", "is", "in", etc.). Stop words appear frequently in the literature texts of the duplicate-check text set but do not contribute to the theme content of the literature texts. If they are not deleted, stop words will affect the literature features and the effect of subsequent duplicate-check similarity calculation. Thus, the accuracy of text analysis is improved. Using a dependency analysis model to parse the relationships between the words in each literature text in the duplicate-check text set after deletion. The dependency analysis model is pre-trained, and the determination of literature keywords by the dependency analysis model is mature and relatively lengthy, so no more description will be given here.

[0027] It can be understood that the preset number is preferably 3. Document texts with 3 identical document features are stored in the analogy document library, document texts without 3 identical document features are stored in the approximate document library, and document texts that cannot be counted are stored in the difference document library. For example, if there are two document texts with the same document features, and their document features are urban transportation, signal system and control center, then both document texts with the same document features are stored in the analogy document library, indicating that the fields involved in the two document texts are relatively consistent, thereby effectively distributing and storing the document texts, providing a clear framework for subsequent duplicate checking.

[0028] In some embodiments of the present application, when extracting document keywords from a document database and training a random forest model, when the matching accuracy of the trained random forest model reaches a matching accuracy threshold, the document domain model is output, including: Obtain all document features of the analogy document library, record them as analogy features, obtain all document features of the approximate document library, record them as similar features, obtain all document features of the difference document library, record them as difference features, and construct a feature data set with analogy features, similar features and difference features; The feature data set is sampled according to a preset ratio to obtain a training set and a test set, and a grid search is used to find the model establishment parameters of the random forest model to build a random forest model; Use the training set to fit the random forest model, substitute the test set into the fitted random forest model and determine the matching accuracy, and pre-set the matching accuracy threshold; When the matching accuracy is greater than or equal to the matching accuracy threshold, the random forest model is determined as the literature domain model; When the matching accuracy is less than the matching accuracy threshold, grid search is continued to be used to re-find the model establishment parameters of the random forest model, the random forest model is re-established, the training set is fitted with the re-determined random forest model, the test set is substituted into the re-fitted random forest model and the matching accuracy is re-determined, until the determined matching accuracy is greater than or equal to the matching accuracy threshold, and the document domain model is output.

[0029] Specifically, the feature data set covers the fields where the features of each document are located. Usually 50%-70% of the data is used as the training set, and the rest is used as the test set. Ensure that both the training set and the test set contain various document features to improve the generalization ability of the model. Grid search exhaustively searches the model parameter establishment parameters in the parameter space of the model, such as the maximum depth of the tree, and uses the training set to fit the random forest model. By integrating multiple decision trees and averaging the prediction results of multiple trees, the risk of overfitting is reduced and the accuracy and stability of the model are improved. Substitute the test set into the trained random forest model and calculate the matching accuracy of the model. The matching accuracy reflects the performance of the model on unknown data and measures the performance of the model. The matching accuracy threshold is preferably 0.8. If the matching accuracy is greater than or equal to the matching accuracy threshold, it means that the trained model tends to be stable and can stably approach the global optimal solution. The trained model is then determined as the output document domain model. If the matching accuracy is less than the matching accuracy threshold, it means that the model training has failed, and it is necessary to re-find the model establishment parameters of the random forest model and continue to substitute the training set and test set until it is greater than or equal to the matching accuracy threshold. Otherwise, the cycle will continue, ensuring the accuracy of the document domain model in determining the subsequent duplicate checking domain.

[0030] In some embodiments of the present application, when preprocessing the document to be checked for duplicates, determining the target document to be checked for duplicates, and determining the field of the target document to be checked for duplicates based on the document field model, it includes: comparing the document format of the document to be checked for duplicates with the standard document format, if the document format and the standard document format are different, converting the document format to the standard document format, if the document format and the standard document format are the same, no format conversion is performed, preprocessing the document to be checked for duplicates after the comparison result, the preprocessing includes removing stop words, unifying the case of punctuation marks, deleting the document author and deleting the document time, determining the target document to be checked for duplicates based on the preprocessing result, parsing the target document to be checked for duplicates using a dependency analysis model, determining the target document features of the target document to be checked for duplicates, and substituting the target document features into the document field model to determine the field of the target document to be checked for duplicates.

[0031] Specifically, comparing the document to be checked for duplicate content with the standard document format and converting the format when necessary can ensure the uniformity of the document format and avoid errors in the duplicate checking process caused by inconsistent formats. The standard document format is determined by GB / T 7714-2015. The efficiency and accuracy of duplicate checking evaluation are improved by unifying the case of punctuation marks, removing stop words, deleting the document author, and deleting irrelevant information such as the document time. The dependency analysis model is continued to be used to extract the target document features of the target duplicate checking document and substitute them into the document domain model. It can accurately determine the duplicate checking domain for the target duplicate checking document, reduce erroneous judgments or omissions due to human experience, and improve the reliability of duplicate checking evaluation.

[0032] In some embodiments of the present application, when document processing is performed on all duplicate-checked document texts and a duplicate-checking diagram is constructed based on the document processing results, the following steps are included: Document processing includes deleting the author of the document and the time of deleting the document, and obtaining similar documents based on the document processing results; Establish a duplicate checking node for each similar document and obtain the page number of each similar document; If there is a similar document citing another similar document, a sub-reference node of the duplicate checking node is generated for the cited similar document, and a sub-feature node of the duplicate checking node is generated for each document feature of each similar document; A duplicate checking diagram is constructed based on the duplicate checking nodes, sub-reference nodes and sub-feature nodes.

[0033] In some embodiments of the present application, when determining the suspected document corresponding to the target duplicate check document according to the duplicate check diagram, it includes: Determine the suspected value of each duplicate check node according to the duplicate check diagram. The suspected value is obtained by the following formula: ; in, Indicates The suspected value of the duplicate check node, Indicates The number of child reference nodes of a duplicate check node, Indicates The number of sub-feature nodes of a duplicate check node, Indicates The number of pages of similar documents corresponding to each duplicate check node; Determine the suspected document corresponding to the target duplicate check document according to the suspected value of each duplicate check node.

[0034] Specifically, after determining the target duplicate check field of the duplicate check document, the corresponding duplicate check document field can be found in the document database according to the duplicate check field, and the duplicate check field and the duplicate check document field are one-to-one corresponding. Document processing is performed on all duplicate check document texts in the duplicate check document field, ensuring that the analysis of duplicate check focuses on the substantive content of similar documents, rather than irrelevant information such as the author of the document and the time of the document, thereby improving the reliability of the duplicate check evaluation. By establishing a duplicate check node and its sub-reference nodes and sub-feature nodes to determine the duplicate check diagram, the citations and relevance between similar documents can be intuitively displayed, and each similar document can be arranged in a hierarchical manner to facilitate detailed comparison and analysis, and the dimensions between similar documents can be deeply analyzed to identify subtle differences. The suspected value of the duplicate check node corresponding to each similar document is calculated, which helps to find the suspected document closest to the target duplicate check document, thereby improving the accuracy of the duplicate check evaluation.

[0035] In some embodiments of the present application, when determining the suspected document corresponding to the target duplicate check document according to the suspected value of each duplicate check node, it includes: If the largest suspected value is unique, the similar document represented by the duplicate checking node corresponding to the largest suspected value is used as the suspected document corresponding to the target duplicate checking document; If the largest suspected value is not unique, all similar documents represented by the duplicate checking node corresponding to the largest suspected value are counted, and the similar document with the least number of pages is used as the suspected document corresponding to the target duplicate checking document.

[0036] Specifically, when the largest suspected value is unique, it means that there is only one similar document represented by a duplicate checking node that is highly similar to the target duplicate checking document. In this case, this similar document is directly used as the suspected document of the target duplicate checking document. When the largest suspected value is not unique, it means that there are multiple similar documents represented by duplicate checking nodes that are highly similar to the target duplicate checking document. The similar document with the least number of pages is selected as the suspected document of the target duplicate checking document. This not only improves the efficiency of the duplicate checking evaluation, but also avoids the analysis of other lengthy information, such as: description of non-document characteristics, explanation of human experience, etc., thereby ensuring the reliability of the duplicate checking evaluation.

[0037] In some embodiments of the present application, when comparing the suspected document with the target duplicate check document to determine the duplicate check similarity, it includes: The similarity of duplicate checking is obtained by the following formula: ; in, Indicates the similarity of duplicate checking. Indicates the page number of the suspected document, Indicates the number of pages of the target document to be checked for duplicate content. Indicates the maximum number of pages in the suspected document and the target duplicate check document. Indicates the document characteristics of the suspected document, Indicates the target document characteristics of the target duplicate check document, Indicates the number of words in the suspected document. Indicates the number of words in the target document to be checked for duplicate content. , , and represents the weight coefficient, and , , and The value range is (0, 1].

[0038] In some embodiments of the present application, when determining the plagiarism evaluation level of the target plagiarism check document according to the plagiarism check similarity, it includes: Preset a first preset duplication check similarity and a second preset duplication check similarity, the first preset duplication check similarity being greater than the second preset duplication check similarity; When the plagiarism similarity is greater than or equal to the first preset plagiarism similarity, the plagiarism evaluation level of the target plagiarism document is determined as a first-level plagiarism; When the plagiarism similarity is less than the first preset plagiarism similarity and greater than or equal to the second preset plagiarism similarity, the plagiarism evaluation level of the target plagiarism document is determined as the second level plagiarism; When the similarity of the duplicate check is less than the second preset similarity of the duplicate check, the evaluation level of the duplicate check of the target duplicate check document is determined as the third level of duplicate check; The degree of duplication in the first-level, second-level and third-level plagiarism checks decreases in sequence.

[0039] Specifically, by comprehensively considering multiple factors (number of document pages, number of document words, and document characteristics), the error in duplicate evaluation caused by judgment of a single dimension is avoided. The number of words in the target duplicate document and the number of words in the suspected document are obtained by deleting the author of the document, deleting the document time, and removing stop words, avoiding irrelevant information from affecting the judgment of duplicate checking. The weight coefficient allows for flexible adjustment of different factors according to actual conditions, avoiding the situation where suspected documents cannot be handled under a unified standard, and helps to improve the adaptability and flexibility of duplicate evaluation. Setting different levels of duplicate evaluation according to the similarity of duplicate checking can accurately reflect the degree of repetition between the suspected document and the target duplicate document. The pre-set first preset duplicate checking similarity and the second preset duplicate checking similarity can be numerically adjusted according to the actual situation of duplicate checking to ensure accurate evaluation of each target duplicate checking document, thereby improving the flexibility and accuracy of duplicate checking evaluation.

[0040] In summary, the beneficial effects of the present invention are: by collecting the document texts of the academic platform and extracting the document features, the document texts of various fields can be fully stored, and the duplicate check evaluation is carried out within a unified framework, thereby ensuring the accuracy of the duplicate check evaluation, by training the random forest model with the document keywords in the document database, using the big data model to learn the relationship therein, thereby outputting the document field model, so that the document duplicate check can intelligently identify the duplicate check field of the target duplicate check document, avoiding the deviation in the duplicate check field caused by human subjective judgment, and improving the accuracy of the duplicate check evaluation, constructing a duplicate check diagram that can intuitively show the correlation between the processing results of each document, accurately deriving the suspected document corresponding to the target duplicate check document, so that the duplicate check evaluation has good pertinence, thereby achieving in-depth analysis. The duplicate check evaluation level of the target duplicate check document is dynamically determined according to the duplicate check similarity, and the reliability of the duplicate check evaluation is improved.

[0041] In another preferred embodiment based on the above embodiment, refer to Figure 2As shown, this embodiment provides a document duplicate checking intelligent evaluation system based on a big data model, which is used to apply the above-mentioned document duplicate checking intelligent evaluation method based on a big data model, including: The document collection module is configured to collect document texts from the academic platform, extract document features of each document text, distribute and store all document texts according to the document features, and establish a document database; The duplicate checking domain module is configured to extract keywords from the document database and train a random forest model. When the matching accuracy of the trained random forest model reaches a matching accuracy threshold, the document domain model is output, the document to be checked for duplicates is preprocessed, the target duplicate checking document is determined, the duplicate checking domain of the target duplicate checking document is determined based on the document domain model, and the corresponding duplicate checking document domain in the document database is determined according to the duplicate checking domain; The duplicate checking analysis module is configured to obtain all duplicate checking document texts in the duplicate checking document field, and perform document processing on all duplicate checking document texts, construct a duplicate checking diagram based on the document processing results, determine suspected documents corresponding to the target duplicate checking document according to the duplicate checking diagram, compare the suspected documents with the target duplicate checking document, and determine the similarity of the duplicate checking; The duplicate checking evaluation module is configured to determine the duplicate checking evaluation level of the target duplicate checking document according to the duplicate checking similarity.

[0042] Specifically, the document collection module collects document texts from academic platforms and extracts document features to establish a document database, ensuring the accuracy of subsequent duplicate checking and evaluation, while providing comprehensive data support for the analysis of big data models. The duplicate checking field module trains the random forest model through the document keywords in the document database. During the training process, the big data model (random forest model) captures the relationships contained in the data, thereby deeply learning the document keywords, thereby providing an accurate prediction of the duplicate checking field for the target duplicate checking document, so as to find the corresponding duplicate checking document field in the document database, effectively narrowing the duplicate checking scope and avoiding errors in the duplicate checking document field caused by human experience judgment. The duplicate checking analysis module is responsible for processing and analyzing all duplicate checking document texts in the duplicate checking document field, thereby constructing a duplicate checking diagram. The duplicate checking diagram can intuitively display the relationship between the duplicate checking document texts after document processing, which is convenient for locating suspected documents, thereby ensuring the accuracy of the duplicate checking evaluation. The duplicate checking evaluation module determines the duplicate checking evaluation level of the target duplicate checking document based on the calculated duplicate checking similarity. The system classifies and subdivides the results of the duplicate checking evaluation, clarifies the degree of duplication between the target duplicate checking document and the suspected document, thus laying the foundation for subsequent improvements or customization of duplicate checking reports.

[0043] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0044] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0045] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0046] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for intelligent evaluation of document duplication based on a big data model, characterized in that: include: Collect document texts from the academic platform, extract document features of each document text, distribute and store all document texts according to the document features, and establish a document database; Extracting keywords from the document database and training a random forest model; when the matching accuracy of the trained random forest model reaches a matching accuracy threshold, outputting a document domain model, preprocessing the document to be checked for duplicates, determining a target document to be checked for duplicates, determining a check for duplicates domain for the target document to be checked for duplicates based on the document domain model, and determining a corresponding check for duplicates document domain in the document database based on the check for duplicates domain; Obtain all duplicate check document texts in the duplicate check document field, perform document processing on all duplicate check document texts, construct a duplicate check diagram based on the document processing results, determine suspected documents corresponding to the target duplicate check document according to the duplicate check diagram, compare the suspected documents with the target duplicate check document, and determine the duplicate check similarity; The plagiarism check evaluation level of the target plagiarism check document is determined according to the plagiarism check similarity.

2. The method for intelligent evaluation of document duplication based on a big data model according to claim 1 is characterized in that: When collecting document texts from the academic platform, extracting document features of each document text, distributing and storing all document texts according to the document features, and establishing a document database, the following steps are included: Count the same literature texts on the academic platform, retain the largest literature texts among the same literature texts, and remove the remaining literature texts; Constructing a duplicate check text set by combining the retained document text and different document texts, analyzing the duplicate check text set based on natural language processing, and removing stop words from the duplicate check text set; A dependency analysis model is used to analyze the relationship between words in each document text in the deleted duplicate check text set, determine noun phrases, use noun phrases that conform to the syntactic structure as document keywords, and use the document keywords of each document text as the document features; The document texts containing a preset number of identical document features are counted and an analog document library is established; the document texts not containing a preset number of identical document features are counted and an approximate document library is established; and the document texts that cannot be counted are established and stored in a difference document library; The document database includes the analogy document library, the approximate document library and the difference document library.

3. The intelligent evaluation method for document duplication checking based on a big data model according to claim 2 is characterized in that: The document keywords are extracted from the document database, and the random forest model is trained. When the matching accuracy of the trained random forest model reaches the matching accuracy threshold, the document domain model is output, including: Obtain all document features of the analogy document library, record them as analogy features, obtain all document features of the approximate document library, record them as similarity features, obtain all document features of the difference document library, record them as difference features, and construct a feature data set with the analogy features, the similarity features and the difference features; The feature data set is sampled according to a preset ratio to obtain a training set and a test set, a grid search is used to find model establishment parameters of the random forest model, and a random forest model is constructed; Fitting the random forest model using the training set, substituting the test set into the fitted random forest model and determining the matching accuracy, and presetting a matching accuracy threshold; When the matching accuracy is greater than or equal to the matching accuracy threshold, the random forest model is determined as a document domain model; When the matching accuracy is less than the matching accuracy threshold, the grid search is continued to be used to re-search the model establishment parameters of the random forest model, the random forest model is re-established, the training set is fitted with the re-determined random forest model, the test set is substituted into the re-fitted random forest model and the matching accuracy is re-determined, until the determined matching accuracy is greater than or equal to the matching accuracy threshold, and the document domain model is output.

4. The method for intelligent evaluation of document duplication based on a big data model according to claim 3 is characterized in that: When preprocessing the document to be checked for duplicates, determining the target document to be checked for duplicates, and determining the check field of the target document to be checked for duplicates based on the document field model, it includes: Comparing the document format of the document to be checked for duplicate content with the standard document format, if the document format is different from the standard document format, converting the document format to the standard document format, and if the document format is the same as the standard document format, not performing format conversion; Preprocessing the documents to be checked for duplicates after the comparison results, wherein the preprocessing includes removing stop words, unifying the case of punctuation marks, deleting the authors of the documents, and deleting the time of the documents, and determining the target document to be checked for duplicates based on the preprocessing results; The target duplicate check document is parsed using the dependency analysis model to determine the target document features of the target duplicate check document, and the target document features are substituted into the document domain model to determine the duplicate check domain of the target duplicate check document.

5. The method for intelligent evaluation of document duplication based on big data model according to claim 4 is characterized in that: When all the duplicate documents are processed and a duplicate checking diagram is constructed based on the results of the document processing, it includes: The document processing includes deleting the document author and deleting the document time, and obtaining similar documents based on the document processing results; Establish a duplicate checking node for each similar document and obtain the page number of each similar document; If there is a similar document citing another similar document, a sub-reference node of the duplicate checking node is generated for the cited similar document, and a sub-feature node of the duplicate checking node is generated for each document feature of each similar document; A duplicate checking diagram is constructed according to the duplicate checking node, the sub-reference node and the sub-feature node.

6. The intelligent evaluation method for checking duplicate documents based on a big data model according to claim 5 is characterized in that: When determining the suspected document corresponding to the target duplicate check document according to the duplicate check diagram, it includes: The suspected value of each duplicate checking node is determined according to the duplicate checking diagram, and the suspected value is obtained by the following formula: ； in, Indicates The suspected value of the duplicate check node, Indicates The number of child reference nodes of a duplicate check node, Indicates The number of sub-feature nodes of a duplicate check node, Indicates The number of pages of similar documents corresponding to each duplicate check node; The suspected document corresponding to the target duplicate check document is determined according to the suspected value of each duplicate check node.

7. The intelligent evaluation method for document duplication checking based on a big data model according to claim 6 is characterized in that: When determining the suspected document corresponding to the target duplicate check document according to the suspected value of each duplicate check node, it includes: If the largest suspected value is unique, the similar document represented by the duplicate checking node corresponding to the largest suspected value is used as the suspected document corresponding to the target duplicate checking document; If the largest suspected value is not unique, all similar documents represented by the duplicate check node corresponding to the largest suspected value are counted, and the similar document with the least number of pages is used as the suspected document corresponding to the target duplicate check document.

8. The intelligent evaluation method for document duplication checking based on a big data model according to claim 7 is characterized in that: When comparing the suspected document with the target duplicate check document to determine the duplicate check similarity, it includes: The similarity of the duplicate check is obtained by the following formula: ； in, Indicates the similarity of duplicate checking. Indicates the page number of the suspected document, Indicates the number of pages of the target document to be checked for duplicate content. Indicates the maximum number of pages in the suspected document and the target duplicate check document. Indicates the document characteristics of the suspected document, Indicates the target document characteristics of the target duplicate check document, Indicates the number of words in the suspected document. Indicates the number of words in the target document. , , and represents the weight coefficient, and , , and The value range is (0, 1].

9. The method for intelligent evaluation of document duplication based on a big data model according to claim 8, characterized in that: When determining the plagiarism check evaluation level of the target plagiarism check document according to the plagiarism check similarity, it includes: Presetting a first preset duplication check similarity and a second preset duplication check similarity, wherein the first preset duplication check similarity is greater than the second preset duplication check similarity; When the plagiarism similarity is greater than or equal to the first preset plagiarism similarity, the plagiarism evaluation level of the target plagiarism document is determined as a first-level plagiarism; When the plagiarism check similarity is less than the first preset plagiarism check similarity and greater than or equal to the second preset plagiarism check similarity, the plagiarism check evaluation level of the target plagiarism check document is determined as a second-level plagiarism check; When the similarity of the duplicate check is less than the second preset similarity of the duplicate check, the evaluation level of the duplicate check of the target duplicate check document is determined as the third level of duplicate check; The duplication degrees of the first-level duplication check, the second-level duplication check and the third-level duplication check decrease in sequence.

10. A document duplicate checking intelligent evaluation system based on a big data model, used for applying the document duplicate checking intelligent evaluation method based on a big data model as described in any one of claims 1 to 9, characterized in that: include: The document collection module is configured to collect document texts from the academic platform, extract document features of each document text, distribute and store all document texts according to the document features, and establish a document database; The duplicate checking domain module is configured to extract the keywords of the documents in the document database and train the random forest model. When the matching accuracy of the trained random forest model reaches the matching accuracy threshold, the document domain model is output, the document to be checked for duplicates is preprocessed, the target duplicate checking document is determined, the duplicate checking domain of the target duplicate checking document is determined based on the document domain model, and the corresponding duplicate checking document domain in the document database is determined according to the duplicate checking domain; The duplicate checking analysis module is configured to obtain all duplicate checking document texts in the duplicate checking document field, perform document processing on all duplicate checking document texts, construct a duplicate checking diagram based on the document processing results, determine suspected documents corresponding to the target duplicate checking document according to the duplicate checking diagram, compare the suspected documents with the target duplicate checking document, and determine the similarity of the duplicate checking; The plagiarism checking evaluation module is configured to determine the plagiarism checking evaluation level of the target plagiarism checking document according to the plagiarism checking similarity.

Images