Project establishment repeatability detection method and device based on ocean engineering scientific research project

Abstract

The invention relates to the technical field of natural language processing, in particular to a project establishment repeatability detection method and device based on ocean engineering scientific research projects. According to the method, word segmentation processing is carried out on historical scientific research projects by constructing an ocean engineering terminology dictionary, and meanwhile, independent processing is carried out on numerical parameters, so that the problems of low terminology recognition accuracy, misrecognition and the like are solved; when the text similarity is detected, semantic similarity compensation is introduced, invisible correlation is effectively recognized, and the matching limitation is broken through; in addition, the similarity of item attributes is considered, repeatability detection is comprehensively carried out based on text similarity and attribute similarity, and the defects of plain text detection are effectively overcome.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, specifically to a method and apparatus for detecting the reproducibility of marine engineering research projects. Background Technology

[0002] With the vigorous development of academic research and technological development, multidisciplinary integration is becoming increasingly common, leading to overlapping and duplicate project approvals, which is particularly evident in research institutions. The reasons for this include inconsistent cross-regional and cross-organizational management requirements, insufficient scope of plagiarism detection databases, and technical barriers to duplication identification.

[0003] To avoid duplicate research projects and repetitive research outputs, a large number of researchers are needed for evaluation. This not only depends on the breadth and depth of researchers' expertise but is also tedious and inefficient. For high-tech enterprises and science and technology reform demonstration enterprises, duplicate project approvals can also lead to significant losses. Summary of the Invention

[0004] In view of this, the present invention provides a method and apparatus for detecting the repetitiveness of marine engineering research projects, in order to solve the problem of how to achieve automatic detection of duplicate project initiation.

[0005] In a first aspect, the present invention provides a method for detecting the duplication of marine engineering research projects. The method includes: segmenting historical research projects based on a pre-built marine engineering terminology dictionary, and constructing a matching library based on the text fields and attribute fields corresponding to each project. Terms containing numerical parameters are preferentially segmented using technical solutions, numerical values, and unit binding. The method also includes: determining the first similarity between the text fields of the project to be detected and the text fields corresponding to each project in the matching library using a semantic similarity-weighted compensation BM25 algorithm; determining the second similarity between the attribute fields of the project to be detected and the attribute fields corresponding to each project in the matching library based on a preset similarity calculation rule; and determining whether the project to be detected is a duplicate project based on the first and second similarities.

[0006] In this invention, a dictionary of marine engineering terminology is constructed to segment historical research projects, and numerical parameters are processed separately to solve problems such as low accuracy in terminology recognition and misidentification. When detecting text similarity, semantic similarity compensation is introduced to effectively identify implicit associations and overcome matching limitations. In addition, the similarity of project attributes is considered, and repetition detection is performed based on a combination of text similarity and attribute similarity to effectively compensate for the shortcomings of pure text detection.

[0007] In one optional implementation, historical research projects are segmented based on a pre-built marine engineering terminology dictionary, and a matching library is constructed using the text fields and attribute fields corresponding to each project. Terms containing numerical parameters are prioritized for segmentation using a method that binds technical solutions, numerical values, and units. This includes: acquiring marine engineering terminology and classifying it according to project type, technical field, and technical point; performing polysemous word annotation on the classified marine engineering terminology to construct a marine engineering terminology dictionary; segmenting terms containing numerical parameters in historical research projects using a preset regular expression to obtain terms bound by technical solutions, numerical values, and units; using the marine engineering terminology dictionary to perform forward and backward matching on the text fields in historical research projects, retaining the longest term matching result; the text fields include title, background, target, and content; and extracting attribute information from historical research projects as tags, which are then combined with the bound terms and the longest term matching result and stored in a structured manner to construct the matching library.

[0008] This invention utilizes a pre-built marine engineering terminology dictionary to accurately segment historical research projects, prioritizing the segmentation of terms containing numerical parameters. It also combines forward and reverse matching to retain the longest term result and extracts attribute information for structured storage to construct a matching library. This improves the accuracy of text segmentation for marine engineering research projects, solves the segmentation challenges of polysemous words and terms containing numerical parameters, and standardizes the storage of project text and attributes. It provides a more accurate and structured matching foundation for subsequent research project duplication detection, thereby improving the efficiency and accuracy of research project management and analysis.

[0009] In one optional implementation, the BM25 algorithm with semantic similarity weighted compensation is used to determine the first similarity between the text field of the item to be detected and the text field corresponding to each item in the matching library. This includes: using the BM25 algorithm to determine the similarity between the text field of the item to be detected and the text field corresponding to each item in the matching library, where the text field includes title, background, target, and content; performing a weighted calculation based on the similarity corresponding to the title, background, target, and content to obtain the BM25 similarity; using a semantic similarity algorithm to determine the semantic similarity between the text field in the item to be detected and its synonyms and the synonyms of the text field corresponding to each item in the matching library; and determining the first similarity between the text field of the item to be detected and the text field corresponding to each item in the matching library based on the BM25 similarity and the semantic similarity.

[0010] This invention employs the BM25 algorithm with semantic similarity weighted compensation. On one hand, by combining the BM25 algorithm with multiple text fields such as title and background, and through weighted calculation, it can initially measure text similarity from the perspective of word frequency. On the other hand, it uses semantic similarity algorithms to mine semantic associations such as synonyms, compensating for the shortcomings of BM25 which relies solely on literal matching. The combination of these two methods to determine the first similarity score can more accurately determine the similarity between the text of the item to be detected and the text of items in the matching library, improving the accuracy and comprehensiveness of text matching in scenarios such as duplicate detection of marine engineering scientific research projects, and reducing misjudgments and omissions caused by literal differences or missing semantic associations.

[0011] In one optional implementation, determining the second similarity between the attribute fields of the project to be detected and the attribute fields corresponding to each project in the matching library based on a preset similarity calculation rule includes: extracting the core attributes, secondary attributes, and auxiliary attributes of the project to be detected, wherein the core attributes include the project field and professional direction, the secondary attributes include the implementation part and the person in charge, and the auxiliary attributes include the funding range and the start and end time; determining the attribute similarity between each attribute of the project to be detected and the attribute fields corresponding to each project in the matching library based on the preset similarity calculation rule; and determining the second similarity based on the attribute similarity and a preset weight.

[0012] This invention extracts core, secondary, and auxiliary attributes of the project under test, calculates attribute similarity based on preset rules, and determines a second similarity score by combining weights. This allows for a more detailed and focused measurement of project attribute matching. Core attributes focus on the project's domain and professional direction, ensuring accurate matching of the project's core positioning; secondary and auxiliary attributes supplement dimensions such as implementation details, responsible persons, and funding, comprehensively covering project characteristics. This hierarchical weighted approach not only captures key project attributes to ensure accurate matching of the core, but also considers secondary and auxiliary attributes to improve the comprehensiveness of the matching, effectively improving the accuracy and rationality of attribute matching in scenarios such as repeatability testing of marine engineering research projects.

[0013] In one optional implementation, determining whether a project to be detected is a duplicate project based on the first similarity and the second similarity includes: determining a comprehensive similarity based on the weighted sum of the first similarity and the second similarity; filtering historical research projects in the matching library based on the relationship between the comprehensive similarity and a preset threshold; and reviewing the filtered historical research projects to determine whether the project is a duplicate project.

[0014] In this invention, a comprehensive similarity score is achieved by weighted summing of the first similarity (textual similarity) and the second similarity (attribute similarity). This score is based on textual semantic matching and corrects for biases by using project attribute matching, ensuring comprehensiveness in the judgment. Furthermore, a preset threshold is used to filter historical projects, reducing invalid comparisons and improving efficiency. Finally, a review process is applied to handle suspected cases near the threshold, preventing algorithmic misjudgments. Overall, this invention achieves greater accuracy in repeatability detection of marine engineering research projects, reducing misjudgments and omissions caused by single-dimensional judgments or lack of review, while also considering detection efficiency and adapting to the characteristics of marine engineering projects and the needs of research management.

[0015] In one alternative implementation, the historical research projects are those that have undergone data anonymization and permission classification. Before building the matching database, the historical research projects are further processed by removing stop words, punctuation marks, and text regularization.

[0016] In this invention, data anonymization and access control are first applied to historical research projects to effectively protect sensitive information, ensure data security, and enable differentiated access. Before constructing the matching database, stop words, punctuation marks, and text regularization are performed to filter meaningless words and interference symbols, standardize text format, and reduce redundant information interference. This allows subsequent work such as duplicate detection of research projects based on the matching database to conduct text analysis and matching more accurately and efficiently, while ensuring data security, thereby improving the quality and efficiency of research project management and analysis.

[0017] In one optional implementation, after segmenting historical research projects based on a pre-built marine engineering terminology dictionary and constructing a matching library with the text fields and attribute fields corresponding to each project, the method further includes: when a new non-repeating project is added to the matching library, using a preset model to extract out-of-vocabulary words of the new project and updating the marine engineering terminology dictionary; and performing anomaly detection processing on the matching library according to a preset cycle.

[0018] In this invention, after constructing the matching library, when adding new non-repeating items, out-of-vocabulary words are extracted using a preset model to update the terminology dictionary. This allows the dictionary to promptly incorporate emerging terms in the field of marine engineering, maintaining its timeliness and professionalism, and improving the accuracy of subsequent word segmentation and other operations. By performing anomaly detection processing on the matching library according to a preset cycle, abnormal data problems within the library can be identified and resolved in a timely manner, ensuring the quality of the matching library data. This provides a more reliable data foundation for work such as duplication detection of scientific research projects based on the matching library, thereby improving the overall accuracy and stability of marine engineering scientific research project management and analysis.

[0019] Secondly, this invention provides a device for detecting the duplication of marine engineering research projects. The device includes: a matching library construction module, used to segment historical research projects based on a pre-built marine engineering terminology dictionary, and construct a matching library with text fields and attribute fields corresponding to each project, wherein terms containing numerical parameters are preferentially segmented using technical solutions, numerical values, and unit binding; a first similarity determination module, used to determine the first similarity between the text fields of the project to be detected and the text fields corresponding to each project in the matching library using the BM25 algorithm with semantic similarity weighted compensation; a second similarity determination module, used to determine the second similarity between the attribute fields of the project to be detected and the attribute fields corresponding to each project in the matching library based on preset similarity calculation rules; and a detection module, used to determine whether the project to be detected is a duplicate project based on the first and second similarities.

[0020] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the above-described method for repeatability testing of marine engineering research projects based on the first aspect or any corresponding embodiment thereof.

[0021] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the method for repeatability testing of marine engineering research projects based on the first aspect or any corresponding embodiment described above.

[0022] Fifthly, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the method for repeatability testing of marine engineering research projects based on the first aspect or any corresponding embodiment described above. Attached Figure Description

[0023] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating a method for repeatability testing of marine engineering research projects according to an embodiment of the present invention.

[0025] Figure 2 This is a flowchart illustrating another method for repeatability testing of marine engineering research projects according to an embodiment of the present invention.

[0026] Figure 3 This is a structural block diagram of a repeatability testing device for marine engineering research projects according to an embodiment of the present invention.

[0027] Figure 4 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0028] As mentioned in the background section, duplicate project detection is a pressing issue that needs to be addressed. Automatically determining the duplication of project proposals based on key information and textual semantics is a feasible solution.

[0029] However, marine engineering, as a strategic emerging industry, involves multiple fields such as marine resource development, marine environmental protection, and marine energy utilization, and has the following distinct characteristics:

[0030] 1) Multidisciplinary nature. Individual projects often involve knowledge from multiple disciplines such as oceanography, structural engineering, materials science, and fluid mechanics. For example, research on offshore wind power foundation structures needs to consider marine environmental loads, structural dynamics, and material fatigue characteristics simultaneously.

[0031] 2) Regional specificity. The implementation of the project is highly dependent on the specific marine environmental conditions. For example, the technical solutions for the "Drilling Platform Design in the Deep Water Area of ​​the South China Sea" and the "Drilling Platform Design in the Shallow Water Area of ​​the Bohai Sea" are completely different due to differences in environmental parameters such as water depth, waves, and ocean currents.

[0032] 3) Parameter sensitivity. Small changes in key parameters can lead to fundamental adjustments in the technical approach. A difference of 2 to 3 meters in water depth may require two completely different solutions: a semi-submersible platform and a tension leg platform.

[0033] 4) Dynamic environmental factors. The marine environment is highly dynamic, and factors such as seasonal changes and climate change have a significant impact on the implementation of the project. The "Winter Ice Zone Ship Navigation Study" and the "Summer Ice Zone Ship Navigation Study" require different technical solutions.

[0034] 5) Data Sensitivity. Marine engineering data often involves sensitive information such as maritime rights and information security, and therefore has extremely high requirements for data security and compliance.

[0035] Currently, duplication detection can be achieved through methods such as keyword matching and similarity calculation. Keyword matching, however, only detects duplication through simple keyword matching, such as calculating the overlap rate of keywords in the title and abstract. It fails to consider the semantic specificity of marine engineering terminology, resulting in low accuracy in recognizing terms like "pile scour" and "vortex-induced vibration," and cannot distinguish key differences such as "single-pile foundation scour protection" and "multi-pile foundation scour mechanism." Similarly, methods like similarity calculation do not consider the parameter sensitivity unique to marine engineering, treating "water depth 30m" and "water depth 35m" as highly similar. In actual engineering, this 5m difference can lead to a foundation type changing from a single pile to a jacket structure, and is particularly ineffective when processing short texts (average 300-500 words) of marine engineering research project abstracts.

[0036] In view of this, this embodiment constructs a marine engineering terminology dictionary to segment historical research projects and processes numerical parameters separately to solve problems such as low accuracy in terminology recognition and misidentification. When detecting text similarity, semantic similarity compensation is introduced to effectively identify implicit associations and overcome matching limitations. In addition, considering the similarity of project attributes, duplicate detection is performed based on a combination of text similarity and attribute similarity to effectively compensate for the shortcomings of pure text detection.

[0037] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] According to an embodiment of the present invention, a method for repeatability testing of marine engineering research projects is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0039] This embodiment provides a method for repeatability testing of marine engineering research projects, which can be used in electronic devices such as computers, mobile phones, and tablets. Figure 1 This is a flowchart of a method for repeatability testing of marine engineering research projects according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps:

[0040] Step S101: Segment historical research projects based on a pre-built marine engineering terminology dictionary, and build a matching library with the text fields and attribute fields corresponding to each project. Among them, terms containing numerical parameters are segmented first by binding technical solutions, numerical values ​​and units.

[0041] Specifically, the marine engineering terminology dictionary is a pre-built dictionary containing basic terms related to marine engineering. As a custom industry dictionary, it is used for text segmentation processing of historical research projects. This avoids the problem of reduced accuracy in recognizing marine engineering terminology caused by using general dictionaries to extract keywords. It should be noted that these historical research projects can be projects that have already been approved and processed, serving as the basis for duplication detection. Furthermore, these historical research projects typically include multiple projects; each project is stored separately when building the matching database for subsequent detection and matching. Each project is also appended with metadata tags (such as year, field, person in charge, etc.) as attribute fields.

[0042] In addition, before segmenting each historical research project, terms containing numerical parameters are processed first; that is, the numerical values ​​and their corresponding technical solutions and units are extracted and bound first. This ensures accurate detection of terms containing parameters. For example, terms like "single pile water depth 30 meters" and "jacket foundation water depth 35 meters" should be extracted and bound first.

[0043] Step S102: The BM25 algorithm with semantic similarity weighted compensation is used to determine the first similarity between the text field of the item to be detected and the text field corresponding to each item in the matching library. The BM25 (Best Matching 25) algorithm is an information retrieval and text similarity calculation algorithm. Its core objective is to quantify the importance of words in the text and calculate the similarity between the query text and the text in the document set, thereby achieving accurate text matching and retrieval. Specifically, in this embodiment, the BM25 algorithm is used to calculate the similarity between the text field of the item to be detected (query text) and the text field corresponding to each item in the matching library (text in the document set).

[0044] Furthermore, the BM25 algorithm primarily relies on word literal overlap to calculate similarity, which may lead to missed detections when terms are different but semantically similar. Therefore, this embodiment introduces semantic similarity as compensation based on the BM25 algorithm. That is, the BM25 algorithm and semantic similarity are used together to calculate similarity, and the results are weighted to obtain the first similarity score.

[0045] Step S103: Determine the second similarity between the attribute field of the item to be detected and the attribute field corresponding to each item in the matching library based on a preset similarity calculation rule. Specifically, for the attribute field, this embodiment uses a preset similarity calculation rule to calculate the similarity. The attribute field may include different types of attributes, such as text or numerical values. Therefore, for different types of attribute fields, this embodiment uses different similarity calculation rules to calculate the similarity of the attribute fields to obtain the second similarity.

[0046] Step S104: Determine whether the item to be detected is a duplicate project based on the first similarity and the second similarity. Specifically, after determining the first similarity and the second similarity, the two similarities are combined to determine whether there are duplicate projects in the matching library.

[0047] This embodiment provides a method for repeatability testing of marine engineering research projects, the process of which includes the following steps:

[0048] Step S201: Based on the pre-built marine engineering terminology dictionary, the historical scientific research projects are segmented into words, and a matching library is built with the text fields and attribute fields corresponding to each project. Among them, terms containing numerical parameters are segmented first by binding technical solutions, numerical values ​​and units.

[0049] Specifically, step S201 includes:

[0050] Step S2011 involves acquiring marine engineering terminology and categorizing it according to project type, technical field, and technical point. Specifically, marine engineering terminology can be obtained from relevant standards or databases. For example, basic terms can be extracted from the national standard "Marine Engineering Terminology" (GB / T15918-2020), the terminology database of the International Institute of Marine Technology, and core marine engineering journals (such as "China Offshore Platform" and "Ocean Engineering"). Simultaneously, the extracted terms are categorized into three levels: project type (wind power, oil and gas, submarine cables, marine observation, etc.), technical field (foundation structure, dynamic response, environmental load, construction and installation, etc.), and technical point (e.g., "pile foundation scour" is further divided into "local scour," "overall scour," and "protective measures").

[0051] Step S2012 involves performing polysemous word annotation on the categorized marine engineering terms to construct a marine engineering terminology dictionary. Specifically, polysemous terms among the acquired basic terms require semantic annotation to achieve semantic disambiguation. For example, the term "scour": when the context includes keywords such as "pillar foundation," "seabed," and "local," it is determined to have a specific meaning in marine engineering; when the context includes "river" and "riverbank," it is determined to have a general meaning. It should be noted that the constructed dictionary can be dynamically loaded in subsequent use, that is, it supports the real-time expansion of new terms, improving the ability to identify out-of-vocabulary words.

[0052] Step S2013 involves segmenting terms containing numerical parameters in historical research projects using a preset regular expression to obtain terms bound to technical solutions, numerical values, and units. Specifically, before segmenting historical research projects using a marine engineering terminology dictionary, this embodiment prioritizes processing terms containing numerical parameters. During processing, this embodiment uses designed regular expressions for segmentation. For example, a regular expression containing the format of technical terms + numerical values ​​+ units is designed to match the text in historical research projects, thereby achieving the matching and extraction of terms containing numerical parameters.

[0053] Step S2014: Use a marine engineering terminology dictionary to perform forward and reverse matching on the text fields in historical research projects, retaining the longest term matching result. The text fields include title, background, target, and content.

[0054] Specifically, when performing word segmentation on the text field, this embodiment performs semantic segmentation according to the title, background, target, and content. That is, for each historical scientific research project, the word segmentation results are saved separately according to the title, background, target, and content. In particular, when using a marine engineering terminology dictionary for word segmentation, this embodiment uses the Jieba word segmentation algorithm, that is, loading the marine engineering terminology dictionary into the Jieba word segmentation engine for term matching.

[0055] During matching, first perform forward scanning and matching (scanning from left to right), and then perform reverse scanning and matching (scanning from right to left). For example, taking the text "Research on Submarine Pipeline Buckling Analysis Technology" as an example, the longest term in the custom dictionary is preferentially matched from left to right; during forward scanning, first attempt to match "Submarine Pipeline Buckling Analysis", if the match is successful, it is marked as a word segmentation unit, and then continue to scan the subsequent "Technology Research"; if the match is unsuccessful, shorten by 1 character (such as "Submarine Pipeline Buckling") and re-match until a term in the dictionary is matched or split into single characters. During reverse matching (scanning from right to left), scan from right to left according to the "maximum matching length", first attempt to match "Buckling Analysis Technology Research", and then shorten it to "Pipeline Buckling Analysis Technology", "Submarine Pipeline Buckling Analysis" (if it exists in the dictionary), and mark it as a word segmentation unit; finally, compare the results of forward matching and reverse matching. If the length of the term matched in a certain direction is longer and there is a clear record in the custom dictionary, then preferentially retain this result.

[0056] Step S2015, extract the attribute information in the historical scientific research projects as tags, combine the bound terms and the longest term matching results, and store them in a structured manner to construct a matching library.

[0057] Among them, before constructing the matching library, for the text in the historical scientific research projects, perform processing such as stop word removal, punctuation removal, and text regularization to improve the efficiency of subsequent matching. Specifically, when performing stop word removal, construct a multi-level stop word library, including general stop words (such as "of", "is"), domain-independent words (such as "research", "analysis"), and redundant words for scientific research projects (such as "project", "topic"). Reduce noise interference through regular expression matching and词性过滤(remove function words, auxiliary words, etc.). When performing punctuation removal, use the Unicode punctuation mapping table to cover full-width and half-width symbols in Chinese and English, and retain necessary mathematical symbols (such as "≥", "%") to avoid loss of technical index information. When performing text regularization, uniformly convert English words to lowercase (such as converting "BIM" to "bim"), standardize the number format (such as converting "1 million" to "1000000"), and at the same time normalize dates, units (such as "MPa", "kN") to ensure semantic consistency in subsequent calculations.

[0058] Specifically, when constructing the matching library, store the data after the above processing into the Elasticsearch distributed database by field (title, background, objective, content), and establish an inverted index to support fast retrieval. Attach metadata tags (such as year, field, person in charge) to each project to form a multi-dimensional data cube.

[0059] In step S2016, when a new non-repeating item is added to the matching library, an out-of-vocabulary word for the new item is extracted using a preset model, and the marine engineering terminology dictionary is updated. Specifically, the constructed matching library can be connected to the API interface of the scientific research management system to achieve automatic entry of new items into the library. When storing a new item in the matching library, the processing described in steps S2013, S2014, and S2015 above is also required. Furthermore, new technologies may be used in the new project; therefore, for the new project, the terminology recognition module can be triggered, i.e., an out-of-vocabulary word (such as "hydrogen-driven marine ranch") is extracted using a preset model such as BiLSTM-CRF (Bidirectional Long Short-Term Memory Network-Conditional Random Field) and the dictionary is updated.

[0060] Based on the addition of new projects, the matching repository can be continuously updated. Accordingly, a Git-like version tree is introduced into the matching repository. Each time the matching repository is updated, such as when a new project is added, a version node is automatically generated. This version node includes information such as a timestamp, version snapshot, and version description. Simultaneously, a tree structure is built for the generated version nodes in chronological order, thus forming the version tree. Based on this version tree, it is possible to trace back to a specific time point. For example, by using a target time point (such as all "offshore wind power infrastructure" projects in Q1 2023), the corresponding version node can be quickly located using the timestamp index of the version tree.

[0061] Step S2017: Perform anomaly detection processing on the matching database according to a preset cycle. Specifically, for the generated matching database, it is necessary to periodically detect outliers, such as null values ​​or garbled versions. At the same time, low-quality data can be marked by a manual review interface to ensure the reliability of the data in the database.

[0062] Step S202: The BM25 algorithm with semantic similarity weighted compensation is used to determine the first similarity between the text field of the item to be detected and the text field corresponding to each item in the matching library.

[0063] Specifically, step S202 includes:

[0064] Step S2021: Use the BM25 algorithm to determine the similarity between the text field of the item to be detected and the text field corresponding to each item in the matching library. The text field includes title, background, target and content.

[0065] Specifically, the BM25 algorithm limits the impact of keyword frequency on scoring in the calculation of term frequency (TF), adding "word saturation" k and "field length reduction" b. The formula for calculating the similarity between the text field Q of the item to be detected and a certain text d in each item D in the matching database is:

[0066]

[0067] In the formula, W i For the word q in text Q i The weights, R(q) i ·d) is the word q i The similarity between and text d.

[0068] W i Usually, its IDF value is taken, that is:

[0069]

[0070] In the formula, N is the total number of documents, n(q) i ) represents the number of texts containing the word. To ensure that the denominator is not zero, the smoothing coefficient is set to 0.5.

[0071]

[0072] In the formula, f i For q i The frequency in d, qf i For q i In Q, k1 represents the parameter controlling the word frequency saturation rate, where parameter K is expressed by the following formula:

[0073]

[0074] In the formula, dl is the length of d, avgdl is the average length of all text D, and b represents the field length specification.

[0075] In summary, the BM25 algorithm can be expressed by the following formula:

[0076]

[0077] In addition, before using the BM25 algorithm for similarity calculation, the parameters of the BM25 algorithm are optimized. Specifically, considering the text characteristics of marine engineering, the adaptability of professional knowledge, and processing experience, the optimal parameters are determined through a grid search method. Among them, k1 = 1.2 (controlling the word frequency saturation rate to avoid excessive influence from high-frequency words); b = 0.75 (adjusting the document length normalization coefficient to ensure a moderate penalty for long texts); avgdl is dynamically calculated through a historical topic database to adapt the weight allocation to texts of different lengths.

[0078] It should be noted that for the text fields in the project to be detected, the same method as the term extraction in historical scientific research projects can be used for term extraction. Specifically, refer to the processing in steps S2013, S2014 and S2015 above to obtain the terms in the project to be detected and the corresponding tags of the project to be detected.

[0079] Step S2022 involves a weighted calculation based on the similarity of the title, background, target, and content to obtain the BM25 similarity score. Specifically, when using the BM25 algorithm for similarity calculation, the similarity of the title, background, target, and content of the item to be detected is calculated separately. Then, the calculation results of each part are weighted and summed to obtain the BM25 similarity score. In this embodiment, the title, background, target, and content are assigned weight coefficients of 0.4, 0.2, 0.2, and 0.2 respectively to highlight the criticality of the title. For example, the matching weight of "prefabricated building" in the title is significantly higher than that of the descriptive text in the background.

[0080] Step S2023: A semantic similarity algorithm is used to determine the semantic similarity between the text fields and synonyms of the text fields in the project to be detected and the synonyms of the text fields corresponding to each project in the matching library. Specifically, based on the similarity calculated using the BM25 algorithm, this embodiment expands the text in the project to be detected and the text in the matching library with synonyms, and then uses the expanded synonyms to calculate semantic similarity, thereby alleviating the problem of missed detections caused by word mismatch. The semantic similarity can be calculated using the Word2Vec model, or other methods can be used; this embodiment does not specifically limit this approach.

[0081] Step S2024: Determine the first similarity between the text field of the item to be detected and the text field corresponding to each item in the matching library based on the similarity and semantic similarity calculated by the BM25 algorithm. Specifically, after determining the similarity calculated by the two similarity calculation methods, the two similarities can be weighted and calculated to obtain the first similarity.

[0082] For example, the query text Q = "deep-water floating platform mooring system design" matches the text D = "FPSO mooring system optimization scheme" in the matching database.

[0083] The similarity score obtained by the BM25 algorithm is 0.65 (due to the difference in terminology between "mooring system" and "anchoring system").

[0084] After the synonym expansion, Q' includes "mooring system / anchoring system", and D' includes "anchoring system / mooring device";

[0085] Based on extended synonyms, semantic similarity is calculated using the following formula: Sim_{\text{sem}}=0.88 (Word2Vec identifies the semantic association between "floating platform" and "FPSO");

[0086] The final score is obtained by weighting the two similarity scores: 0.7×0.65+0.3×0.88=0.719.

[0087] Step S203: Determine the second similarity between the attribute field of the item to be detected and the attribute field corresponding to each item in the matching library based on the preset similarity calculation rules.

[0088] Specifically, step S203 includes:

[0089] Step S2031: Extract the core attributes, secondary attributes, and auxiliary attributes of the project to be detected. The core attributes include the project field and professional direction; the secondary attributes include the implementation part and the person in charge; and the auxiliary attributes include the funding range and start and end time. It should be noted that the attribute fields for each project in the matching database also include core attributes, secondary attributes, and auxiliary attributes. Specifically, these attributes are all structured data with fixed dictionary values.

[0090] Step S2032: Determine the attribute similarity between each attribute of the project to be detected and the corresponding attribute field of each project in the matching library based on a preset similarity calculation rule. Specifically, for attribute fields, there are different attribute types. For example, project field and professional direction may be multi-valued attributes, while implementing department and person in charge may be unique attributes, and funding range and start and end time are numerical attributes. For multi-valued attributes, an improved Jaccard coefficient can be used, i.e., the formula... (α = 0.1 is used to smooth out empty set anomalies) Calculate similarity; for unique attributes, Boolean matching can be used to calculate similarity, such as 1 for the same and 0 for different. For numerical attributes, since these values ​​are all range values, the similarity can be calculated by calculating the range overlap rate.

[0091] Step S2033: Determine the second similarity based on attribute similarity and preset weights. Specifically, for the core attributes of project field and professional direction, the weights are 0.3 respectively; for the secondary attributes of implementing department and person in charge, the weights are 0.15 respectively; and for the auxiliary attributes of funding range and start and end time, the weights are 0.1 respectively. The second similarity is obtained by multiplying the similarity of each attribute by its corresponding weight and then summing them.

[0092] Step S204: Determine whether the item to be detected is a duplicate project based on the first similarity and the second similarity.

[0093] Specifically, step S204 includes:

[0094] Step S2041: Determine the comprehensive similarity based on the weighted sum of the first similarity and the second similarity; wherein, in this embodiment, the weight of the first similarity is 0.7 and the weight of the second similarity is 0.3, and the formula for calculating the comprehensive similarity is S. 综合 =0.7S 文本 +0.3S 属性The weights for two similarity scores are assigned using ROC curves to ensure that text dominates but attribute corrections are not misjudged.

[0095] Step S2042: Based on the relationship between the comprehensive similarity and the preset threshold, historical research projects in the matching library are filtered. Specifically, the preset threshold can be adjusted according to the actual situation, such as setting it to 0.75, or adjusting it to 0.85 depending on the scenario, etc. During the filtering process, projects in the matching library whose comprehensive similarity to the project to be detected exceeds the preset threshold can be filtered out.

[0096] Step S2043 involves reviewing the selected historical research projects to determine if any projects are duplicated. Specifically, when there are many selected projects, they can be sorted in descending order of comprehensive similarity, and the top-ranked projects (e.g., the top ten projects) can be output. Projects with a long time span, such as those exceeding 5 years, can also be filtered to avoid interference from outdated projects.

[0097] When historical research projects with a similarity score exceeding a preset threshold are selected, duplicate projects can be identified, requiring manual review to determine the final result. Additionally, projects with a similarity score near a preset threshold (e.g., between 0.7 and 0.8) can also trigger manual review, generating a table highlighting similar sections and comparing differences to aid decision-making.

[0098] It should be noted that before performing the above processing on the projects to be tested and historical research projects in the matching library, security and compliance processing must be carried out first, namely, data anonymization (such as replacing the person in charge's name with the employee ID), hierarchical access control (such as data isolation within departments), and audit log tracking (operation traceability) to meet the enterprise's data security requirements.

[0099] As a specific application embodiment of the present invention, such as Figure 2 As shown, the method for repeatability testing of marine engineering research projects is implemented through the following process:

[0100] 1. Construction of a dictionary of marine engineering terminology.

[0101] The constructed dictionary is the foundation of duplicate detection and includes the following key technical steps:

[0102] 1.1 Establish a terminology database. Specifically, basic terms are extracted from the national standard "Terminology for Marine Engineering" (GB / T 15918-2020), the terminology database of the International Institute of Marine Technology, and core marine engineering journals (such as "China Offshore Platform" and "Ocean Engineering"), and classified in three levels according to engineering category, technical field, and technical point.

[0103] 1.2, Semantic annotation of polysemous words. Specifically for polysemous terms, semantic disambiguation rules are established. For example, for "scouring", when the context contains keywords such as "pile foundation", "seabed", "local", etc., it is determined as the specific meaning in ocean engineering; when the context contains "river", "riverbank", it is determined as the general meaning.

[0104] 2. Domain-specific word segmentation algorithm.

[0105] 2.1, Text word segmentation. Specifically based on Jieba word segmentation, combined with a custom industry dictionary, semantic segmentation is performed on the title, background, objectives, and content of scientific research projects. For professional terms (such as "BIM collaborative design", "geotechnical engineering investigation"), word segmentation is optimized. By comparing the forward and reverse matching results, the longest term matching result is preferentially retained to avoid incorrect segmentation of "subsea / pipeline / buckling / analysis" and ensure that professional vocabulary is not incorrectly split. For terms containing numerical parameters (such as "water depth 50m", "wave height period 8s"), a regular expression priority segmentation pattern is designed to ensure that the parameter and unit are not split. At the same time, through the dynamic loading of the user dictionary function, real-time expansion of new terms is supported, and the recognition ability of out-of-vocabulary words is improved.

[0106] 2.2, Stop word removal. A multi-level stop word library is constructed, including general stop words (such as "of", "is"), domain-irrelevant words (such as "research", "analysis"), and redundant words for scientific research projects (such as "project", "topic"). Through regular expression matching and词性过滤(如去除虚词、助词), noise interference is reduced.

[0107] 2.3, Punctuation removal. The Unicode punctuation symbol mapping table is used to cover full-width and half-width symbols in Chinese and English, and necessary mathematical symbols (such as "≥", "%") are retained to avoid loss of technical index information.

[0108] 2.4, Text regularization. English words are uniformly converted to lowercase (such as "BIM" is converted to "bim"), digital formats are standardized (such as "1 million" is converted to "1000000"), and at the same time, dates and units (such as "MPa", "kN") are normalized to ensure semantic consistency in subsequent calculations.

[0109] 3. Establish a matching library. The construction and maintenance of the matching library directly affect the coverage and timeliness of the detection system.

[0110] 3.1, Multi-dimensional and structured storage. The preprocessed historical project data is stored in the Elasticsearch distributed database by field (title, background, objectives, content), and an inverted index is established to support fast retrieval. Each project is attached with metadata tags (such as year, field, person in charge) to form a multi-dimensional data cube.

[0111] It should be noted that the part "词性过滤(如去除虚词、助词)" in the original text seems to be an incomplete or incorrect expression. I translated it as "词性过滤(如去除虚词、助词)" as it is, but it might need to be corrected in the original context.3.2 Dynamic Update Mechanism. The system connects to the scientific research management system API to obtain data on newly approved projects in real time. This triggers the terminology recognition module (e.g., BiLSTM-CRF model) to extract out-of-vocabulary terms (such as "hydrogen-driven marine ranching") and update the terminology database. A Git-like version tree is used to store historical versions of the matching database, supporting backtracking to project status at specific time points (e.g., all "offshore wind power infrastructure" projects in Q1 2023).

[0112] 3.3 Data Cleaning. Regularly perform outlier checks (such as null values ​​and garbled text) on the matching database, and combine this with manual review of the interface to mark low-quality data, ensuring the reliability of the data in the database.

[0113] 4. Calculate text similarity.

[0114] 4.1 The BM25 algorithm is used to calculate text similarity. The parameters of the BM25 algorithm are determined using a grid search method, taking into account the characteristics of marine engineering texts, adaptability to professional knowledge, and processing experience.

[0115] 4.2 Multi-field weighted strategy. Weight coefficients of 0.4, 0.2, 0.2, and 0.2 are assigned to the title, background, target, and content respectively, highlighting the key importance of the title. For example, the matching weight of "prefabricated building" in the title is significantly higher than that of the descriptive text in the background.

[0116] 4.3 Semantic Enhancement. Based on BM25, synonym expansion is incorporated (e.g., "mooring system" and "anchoring system" are considered equivalent terms). The Word2Vec model is used to calculate semantic similarity, and the BM25 score is weighted to compensate for any missed detections caused by word mismatch.

[0117] Case Description: The query text Q = "Deepwater Floating Platform Mooring System Design" matches the text D = "FPSO Mooring System Optimization Scheme" in the matching database.

[0118] BM25 score, 0.65 (due to the difference in terminology between "mooring system" and "anchoring system");

[0119] After the synonym expansion, Q' includes "mooring system / anchoring system", and D' includes "anchoring system / mooring device";

[0120] Semantic similarity, Sim_{\text{sem}} = 0.88 (Word2Vec identifies the semantic association between "floating platform" and "FPSO");

[0121] Final score: 0.7×0.65+0.3×0.88=0.719.

[0122] 5. Calculate item attribute similarity. Item attributes are structured data with fixed dictionary values. By calculating the similarity of item attributes, a quick determination can be achieved.

[0123] 5.1 Attribute Classification and Weight Allocation: Attributes are divided into three categories: ① Core attributes (project field, professional direction, weights of 0.3 each); ② Secondary attributes (implementing department, person in charge, weights of 0.15 each); ③ Auxiliary attributes (funding range, start and end time, total weight of 0.1).

[0124] The similarity calculation rules include the following methods:

[0125] For unique attributes (such as person in charge), Boolean matching is used (1 for the same, 0 for different).

[0126] For multi-valued attributes (such as those involving majors), the improved Jaccard coefficient is used to calculate similarity.

[0127] 6. Calculate the overall similarity.

[0128] 6.1 A weighted average is used to determine the overall similarity. The weight of text similarity is 0.7, and the weight of attribute similarity is 0.3. This weighting is verified by ROC curve to ensure that text is dominant but attribute corrections may lead to misjudgment.

[0129] 6.2 Dynamic Threshold Setting. Based on historical data distribution, a baseline threshold of 0.75 is set, allowing users to adjust it according to specific scenarios (e.g., a strict mode threshold of 0.85). For suspected cases near the threshold (e.g., the 0.7-0.8 range), a manual review interface is triggered, outputting a table of highlighted similar paragraphs and differences to aid decision-making.

[0130] 6.3 Result Sorting and Filtering. For projects exceeding the threshold, the Top 10 results are output in descending order of overall similarity, and outdated topics with a time span exceeding 5 years (configurable) are filtered to avoid interference from outdated projects.

[0131] 7. System Implementation and Optimization.

[0132] This embodiment employs a microservice architecture. The system is divided into three microservice modules: preprocessing, computation, and storage. Service registration and API gateway are implemented based on Spring Cloud, supporting horizontal scaling and high-concurrency requests. A continuous learning mechanism is also employed, creating a feedback loop: user-corrected data (such as false positives / false negatives) is automatically stored in the training set, and BM25 parameters and weight allocation are periodically fine-tuned to achieve model self-optimization. Furthermore, data anonymization (e.g., replacing the responsible person's name with an employee ID), hierarchical access control (e.g., data isolation within departments), and audit log tracking (operation traceability) meet enterprise data security requirements.

[0133] The present invention has the following effects:

[0134] 1. The detection accuracy is significantly improved, the false positive rate is reduced, and the false negative rate is decreased.

[0135] By optimizing the domain terminology database (e.g., synonymous expansion of "mooring system" and "anchoring system") and parameter sensitivity matching (e.g., triggering scheme changes due to a 5m difference in water depth), the problem of misidentification of professional terms by the traditional BM25 is solved. The Word2Vec semantic compensation mechanism is introduced to effectively identify implicit associations such as "floating platform" and "FPSO," overcoming the limitations of keyword matching.

[0136] 2. Breakthrough in detection efficiency.

[0137] Improved response speed. Based on a microservice architecture and parallel reduction algorithm, a single detection takes less than 1.5 seconds and supports real-time comparison of tens of thousands of projects.

[0138] Resource consumption is reduced. Through incremental update mechanisms and dynamic weight adjustments, the overall computational overhead is reduced, making it suitable for large-scale scientific research management scenarios.

[0139] 3. Enhanced domain adaptability.

[0140] Specialized rules for marine engineering. Pioneering hierarchical matching of sea area types and normalization of environmental parameters, these rules resolve misjudgments caused by regional specificity (South China Sea vs. Bohai Sea) and dynamic environmental factors (winter / summer ice zones).

[0141] Security and compliance assurance. Data anonymization (responsible person → employee ID) and hierarchical access control meet the compliance requirements for sensitive marine engineering data (such as military sea area projects).

[0142] 4. The system has outstanding scalability.

[0143] Modular design. Supports independent algorithm upgrades (such as replacing the BERT model) and can be migrated to scenarios such as patent novelty searches and technical solution comparisons.

[0144] Self-optimization mechanism. Closed-loop training using user error correction data enables dynamic tuning of BM25 parameters (k1=1.2, b=0.75) and weight allocation.

[0145] This embodiment also provides a repeatability testing device for marine engineering research projects. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0146] This embodiment provides a device for detecting the repeatability of marine engineering research projects, such as... Figure 3 As shown, it includes:

[0147] The matching library construction module 31 is used to segment historical scientific research projects based on a pre-built marine engineering terminology dictionary, and to build a matching library with the text fields and attribute fields corresponding to each project. Among them, terms containing numerical parameters are segmented first by binding technical solutions, numerical values ​​and units.

[0148] The first similarity determination module 32 is used to determine the first similarity between the text field of the item to be detected and the text field corresponding to each item in the matching library using the BM25 algorithm with semantic similarity weighted compensation.

[0149] The second similarity determination module 33 is used to determine the second similarity between the attribute field of the item to be detected and the attribute field corresponding to each item in the matching library based on a preset similarity calculation rule;

[0150] The detection module 34 is used to determine whether the item to be detected is a duplicate project based on the first similarity and the second similarity.

[0151] Further functional descriptions of the above modules are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0152] This invention also provides a computer device having the above-described features. Figure 3 The device shown is a repeatability testing device for marine engineering research projects.

[0153] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 4 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 4 Take a processor 10 as an example.

[0154] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GPA), or any combination thereof.

[0155] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0156] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device as shown by a landing page for an app. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, which can be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0157] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0158] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0159] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0160] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0161] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for detecting the repetition of a project based on a marine engineering scientific research project, characterized in that, The method comprises: Based on the pre-constructed marine engineering professional term dictionary, the historical scientific research projects are segmented, and a matching library is constructed with the text field and the attribute field corresponding to each project, wherein the terms containing numerical parameters are preferentially segmented in the form of binding of technical solutions, numerical values and units; The first similarity between the text field of the to-be-detected project and the text field corresponding to each project in the matching library is determined by using the BM25 algorithm with semantic similarity weighted compensation. The second similarity between the attribute field of the to-be-detected project and the attribute field corresponding to each project in the matching library is determined based on a preset similarity calculation rule. Whether the to-be-detected project is repeated is determined based on the first similarity and the second similarity.

2. The method of claim 1, wherein, Based on the pre-constructed marine engineering professional term dictionary, the historical scientific research projects are segmented, and a matching library is constructed with the text field and the attribute field corresponding to each project, wherein the terms containing numerical parameters are preferentially segmented in the form of binding of technical solutions, numerical values and units, comprising: Marine engineering professional terms are obtained and classified according to engineering types, technical fields and technical points; Polysemous word sense annotation is performed on the classified marine engineering professional terms to construct a marine engineering professional term dictionary; The terms containing numerical parameters in the historical scientific research projects are segmented according to a preset regular expression to obtain terms bound by technical solutions, numerical values and units; The text field of the historical scientific research projects is positively matched and inversely matched by using the marine engineering professional term dictionary, and the longest term matching result is retained, the text field including title, background, target and content; The attribute information in the historical scientific research projects is extracted as labels, combined with the bound terms and the longest term matching result, and stored in a structured manner to construct a matching library.

3. The method of claim 1, wherein, The first similarity between the text field of the to-be-detected project and the text field corresponding to each project in the matching library is determined by using the BM25 algorithm with semantic similarity weighted compensation, comprising: The similarity between the text field of the to-be-detected project and the text field corresponding to each project in the matching library is determined by using the BM25 algorithm, the text field including title, background, target and content; The similarity of the BM25 algorithm is obtained by weighted calculation based on the similarity of the title, background, target and content; The semantic similarity between the text field of the to-be-detected project and the synonym of the text field and the synonym of the text field corresponding to each project in the matching library is determined by using the semantic similarity algorithm. The first similarity between the text field of the to-be-detected project and the text field corresponding to each project in the matching library is determined based on the similarity of the BM25 algorithm and the semantic similarity.

4. The method of claim 1, wherein, The second similarity between the attribute field of the to-be-detected project and the attribute field corresponding to each project in the matching library is determined based on a preset similarity calculation rule, comprising: The core attribute, the secondary attribute and the auxiliary attribute of the to-be-detected project are extracted, the core attribute including the project field and the professional direction, the secondary attribute including the implementation part and the person in charge, and the auxiliary attribute including the fund interval and the start and end time; The attribute similarity between each attribute of the to-be-detected project and the attribute field corresponding to each project in the matching library is determined based on a preset similarity calculation rule. Determine the second similarity based on attribute similarity and preset weight.

5. The method of claim 1, wherein, Determine whether the to-be-detected project is a repeated project based on the first similarity and the second similarity, including: Determine a comprehensive similarity based on a weighted sum of the first similarity and the second similarity; Screen historical scientific research projects in the matching library based on a relationship between the comprehensive similarity and a preset threshold; Determine whether the to-be-detected project is a repeated project based on the first similarity and the second similarity.

6. The method of claim 1, wherein, The historical scientific research projects are historical scientific research projects after data desensitization and permission grading, and further include stop word removal, punctuation removal, and text normalization processing on the historical scientific research projects before constructing the matching library.

7. The method of claim 1, wherein, After the historical scientific research projects are segmented based on the pre-constructed marine engineering professional term dictionary and the matching library is constructed with the text field and the attribute field corresponding to each project, the method further includes: When a non-repeated project is added in the matching library, a preset model is used to extract the out-of-vocabulary word of the added project, and the marine engineering professional term dictionary is updated; Abnormality detection processing is performed on the matching library according to a preset period.

8. A device for detecting the repeatability of a marine engineering research project based on the establishment of a project, characterized in that, The device includes: A matching library construction module is configured to segment the historical scientific research projects based on a pre-constructed marine engineering professional term dictionary, and construct a matching library with a text field and an attribute field corresponding to each project, wherein a term containing a numerical parameter is preferentially segmented in a technical solution, numerical value, and unit binding manner; A first similarity determination module is configured to determine a first similarity between a text field of a to-be-detected project and a text field corresponding to each project in the matching library by using a BM25 algorithm with semantic similarity weighted compensation; A second similarity determination module is configured to determine a second similarity between an attribute field of the to-be-detected project and an attribute field corresponding to each project in the matching library based on a preset similarity calculation rule; A detection module is configured to determine whether the to-be-detected project is a repeated project based on the first similarity and the second similarity.

9. A computer device, comprising: It includes: A memory and a processor are communicatively connected, and the memory stores computer instructions, and the processor executes the computer instructions to perform the method for detecting the repeatability of the project based on the marine engineering scientific research project according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions, and the computer instructions are used to make the computer execute the method for detecting the repeatability of the project based on the marine engineering scientific research project according to any one of claims 1 to 7.

Images