A marine cargo attribute recognition method and system based on multi-algorithm weighted fusion

Abstract

The application discloses a kind of sea freight attribute identification method and system based on multi-algorithm weighted fusion.There is included obtaining including booking and / or bill of lading data to be identified freight transport document data and carries out standardization preprocessing, obtains including cargo control party number, freight data and cargo name standard text standardization freight transport document data;Call preset cargo category keyword library and preset cargo attribute keyword library to carry out attribute identification to cargo name standard text, determine initial cargo attribute identification result;Based on cargo control party number, freight data and cargo name standard text, search in pre-constructed historical freight transport document database, and adopt pre-constructed multi-algorithm weighted fusion method to carry out similarity matching calculation, obtain associated historical freight transport document data;Based on initial cargo attribute identification result and associated historical freight transport document data determine target cargo attribute identification result.Improve the level of logistics freight transport document compliance risk control and cargo attribute automatic identification ability.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for identifying maritime cargo attributes based on multi-algorithm weighted fusion. Background Technology

[0002] In logistics and freight forwarding document processing, to prevent violations and concealment, and to mitigate the risks of maritime transport safety accidents and customs compliance penalties, it is necessary to regularly review and verify monthly new freight document data against historical freight document data. Currently, the industry generally uses a verification method combining manual verification with conventional text matching, which has many technical shortcomings and business pain points, severely restricting the accuracy and efficiency of cargo attribute identification. Specifically, the traditional manual line-by-line verification method is labor-intensive and inefficient when dealing with large volumes of freight document data, and is prone to omissions and human error. Meanwhile, conventional text matching relies solely on basic fuzzy similarity algorithms, easily generating numerous invalid warnings or matching omissions, making it difficult to accurately identify concealment activities where product names are similar but cargo properties are inconsistent, thus failing to meet the efficient, accurate, and traceable business requirements for compliant cargo attribute identification in freight documents. Summary of the Invention

[0003] This invention provides a method and system for identifying maritime cargo attributes based on multi-algorithm weighted fusion, in order to solve the problems of abnormal cargo attribute identification and underreporting.

[0004] According to one aspect of the present invention, a method for identifying maritime cargo attributes based on multi-algorithm weighted fusion is provided, comprising: Obtain the freight document data to be identified, perform standardized preprocessing on the freight document data to be identified, and obtain standardized freight document data. The freight document data to be identified includes booking data and / or bill of lading data. The standardized freight document data includes at least the cargo control party number, cargo flow data and standard text of cargo name. Retrieve the preset cargo category keyword library and the preset cargo attribute keyword library, identify cargo attributes in the standard text of cargo name, and determine the initial cargo attribute identification result corresponding to the freight document data to be identified. Based on the cargo control party number, cargo flow data, and standard text of cargo name, a search is performed in a pre-built historical freight document database, and a pre-built multi-algorithm weighted fusion method is used to calculate similarity matching to obtain the associated historical freight document data corresponding to the freight document data to be identified. Based on the initial cargo attribute identification results and the cargo attribute tagging information corresponding to the associated historical freight document data, the target cargo attribute identification result of the freight document data to be identified is determined.

[0005] Optionally, the freight document data to be identified is subjected to standardized preprocessing to obtain standardized freight document data, including: cleaning each field information in the freight document data to be identified to obtain cleaned field information; for the goods name field in the cleaned field information, calling a pre-built noise regular expression pattern library, and removing redundant descriptive text in the goods name field through regular expression matching to obtain the standard text of the goods name corresponding to the goods name field; replacing the content of the corresponding field in the cleaned field information with the standard text of the goods name corresponding to the goods name field to obtain standardized freight document data.

[0006] Optionally, a preset cargo category keyword library and a preset cargo attribute keyword library are retrieved to perform cargo attribute identification on the standard text of the cargo name, and to determine the initial cargo attribute identification result corresponding to the freight document data to be identified. This includes: performing word segmentation on the standard text of the cargo name to obtain multiple word segmentation units; performing keyword matching on the preset cargo category keyword library based on the multiple word segmentation units to obtain cargo type matching results; performing matching on the preset cargo attribute keyword library based on the multiple word segmentation units to obtain cargo attribute matching results; and determining the cargo attribute identification result corresponding to the freight document data to be identified based on the cargo type matching result and the cargo attribute matching result.

[0007] Optionally, based on the cargo controller number, cargo flow data, and standard text of cargo name, a search is performed in a pre-built historical freight document database, and a pre-built multi-algorithm weighted fusion method is used to calculate similarity matching to obtain the associated historical freight document data corresponding to the freight document data to be identified. This includes: searching the pre-built historical freight document database based on the cargo controller number and cargo flow data to obtain freight document data retrieval results; if the freight document data retrieval results contain at least one historical freight document record, the historical cargo name text is extracted from each historical freight document record, and the similarity between the historical cargo name text and the standard text of cargo name is calculated using the pre-built multi-algorithm weighted fusion method to obtain the similarity data corresponding to each historical freight document record. The pre-built multi-algorithm weighted fusion method includes multiple preset fuzzy matching algorithms; the historical freight document records corresponding to the similarity data that meet the preset similarity threshold in the similarity data corresponding to each historical freight document record are determined as the associated historical freight document data corresponding to the freight document data to be identified.

[0008] Optionally, a pre-constructed multi-algorithm weighted fusion method is used to calculate the similarity between historical cargo name text and standard cargo name text, obtaining similarity data corresponding to each historical freight document record. This includes: calculating multiple basic similarity scores between historical cargo name text and standard cargo name text based on multiple preset fuzzy matching algorithms, including word segmentation set matching algorithm, word segmentation sorting and editing matching algorithm, and local sliding matching algorithm; determining the number of intersections and unions of keyword sets between historical cargo name text and standard cargo name text, and determining the keyword matching similarity score based on the number of intersections and unions of keyword sets; and performing weighted fusion based on multiple basic similarity scores and keyword matching similarity scores to obtain the similarity between historical cargo name text and standard cargo name text.

[0009] Optionally, the method also includes: if the similarity data corresponding to each historical freight document record is lower than the preset similarity threshold, it is determined that there is no associated historical freight document data corresponding to the freight document data to be identified; if there are multiple historical freight document records whose similarity data meets the preset similarity threshold, the similarity reward and punishment mechanism is invoked to adjust the corresponding similarity data.

[0010] Optionally, if it is determined that there is no associated historical freight document data corresponding to the freight document data to be identified, the method further includes: determining the target freight attribute identification result of the freight document data to be identified based on the initial freight attribute identification result.

[0011] Optionally, if multiple historical freight document records have similarity data that meets a preset similarity threshold, a similarity reward and punishment mechanism is invoked to adjust the corresponding similarity data. This includes: extracting dangerous goods marking information corresponding to each of the multiple historical freight document records; if the dangerous goods marking information of each historical freight document record is consistent, the similarity data remains unchanged; if the dangerous goods marking information of at least one historical freight document record is inconsistent with the other records, the similarity data is reduced according to a preset step size.

[0012] Optionally, the target cargo attribute identification result of the freight document data to be identified is determined based on the initial cargo attribute identification result and the cargo attribute marking information corresponding to the associated historical freight document data, including: if the initial cargo attribute identification result is a first attribute, then the first attribute is determined as the target cargo attribute identification result of the freight document data to be identified; if the initial cargo attribute identification result is a second attribute, and the cargo attribute marking information of the associated historical freight document data is a first attribute identifier, then the target cargo attribute identification result of the freight document data to be identified is determined to be a suspected concealed document; if the initial cargo attribute identification result is a second attribute, and the cargo attribute marking information of the associated historical freight document data is a second attribute identifier, then the second attribute is determined as the target cargo attribute identification result of the freight document data to be identified.

[0013] According to another aspect of the present invention, a maritime cargo attribute identification system based on multi-algorithm weighted fusion is provided, comprising: The standardized freight document data determination module is used to acquire freight document data to be identified, perform standardized preprocessing on the freight document data to be identified, and obtain standardized freight document data. The freight document data to be identified includes booking data and / or bill of lading data. The standardized freight document data includes at least the cargo control party number, cargo flow data, and standard text of cargo name. The initial cargo attribute identification result determination module is used to retrieve the preset cargo category keyword library and the preset cargo attribute keyword library, perform cargo attribute identification on the standard text of cargo name, and determine the initial cargo attribute identification result corresponding to the freight document data to be identified. The module for determining associated historical freight document data is used to search the pre-built historical freight document database based on the cargo controller number, cargo flow data and standard text of cargo name, and to perform similarity matching calculation using a pre-built multi-algorithm weighted fusion method to obtain the associated historical freight document data corresponding to the freight document data to be identified. The target cargo attribute identification result determination module is used to determine the target cargo attribute identification result of the freight document data to be identified based on the initial cargo attribute identification result and the cargo attribute tag information corresponding to the associated historical freight document data.

[0014] The technical solution of this invention obtains standardized freight document data by acquiring the freight document data to be identified and performing standardized preprocessing on the data. The freight document data to be identified includes booking data and / or bill of lading data. The standardized freight document data includes at least the cargo control party number, cargo flow data, and standard text of the cargo name. This achieves unified data standards, eliminates the problems of messy, incomplete, and inconsistent formats in the original data, and lays a solid data foundation for subsequent accurate identification. The invention also retrieves a preset cargo category keyword library and a preset cargo attribute keyword library to identify the cargo attributes of the standard text of the cargo name, determining the initial cargo attribute identification result corresponding to the freight document data to be identified. This achieves discrimination based on the standardized keyword library, effectively improving the efficiency of cargo name identification, avoiding subjective bias in manual judgment, and achieving high initial judgment of cargo attributes. The system is efficient, standardized, and regulated. Based on the cargo control party number, cargo flow data, and standard text of cargo name, it searches a pre-built historical freight document database and uses a pre-built multi-algorithm weighted fusion method to calculate similarity matching, obtaining the associated historical freight document data corresponding to the freight document data to be identified. This significantly improves the accuracy of historical data matching and provides reliable historical data support for result verification. Based on the initial cargo attribute identification results and the cargo attribute marking information corresponding to the associated historical freight document data, the system determines the target cargo attribute identification result of the freight document data to be identified. It realizes a dual verification mechanism of intelligent preliminary judgment and historical data verification, which makes up for the limitations of a single identification method, effectively reduces the probability of missed or false judgments in cargo attribute identification, and comprehensively improves the accuracy, reliability, and rigor of freight document cargo attribute identification results. This solution achieves standardized, intelligent, and dual-verification of cargo attribute identification in freight documents. It ensures data accuracy from the source through data preprocessing, enables rapid and accurate identification of cargo names based on a dedicated keyword database, and completes result verification through historical data similarity matching. This effectively solves the problems of low efficiency, strong subjectivity, high rates of missed and false judgments, and concealment of dangerous goods in traditional cargo attribute identification methods. It significantly improves the accuracy, comprehensiveness, and efficiency of cargo attribute identification in freight documents. Furthermore, based on a multi-dimensional data linkage verification mechanism, it strengthens the reliability and credibility of the identification results, providing standardized and traceable data support for freight document security risk control.

[0015] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a method for identifying maritime cargo attributes based on multi-algorithm weighted fusion provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart of a method for identifying maritime cargo attributes based on multi-algorithm weighted fusion, provided in Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of the structure of a maritime cargo attribute identification system based on multi-algorithm weighted fusion provided in Embodiment 3 of the present invention; Figure 4 This is a schematic diagram of a visual UI interaction interface for a cargo nature comparison tool applicable to an embodiment of the present invention; Figure 5 This is a schematic diagram of the visual UI interaction interface of a cargo nature comparison tool applicable to embodiments of the present invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0019] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] Example 1 Figure 1This is a flowchart of a maritime cargo attribute identification method based on multi-algorithm weighted fusion provided in Embodiment 1 of the present invention. This embodiment is applicable to the intelligent identification of maritime cargo attributes in freight document data. This method can be executed by a maritime cargo attribute identification system, which can be implemented in hardware and / or software and can be configured in electronic devices such as computers and servers. Figure 1 As shown, the method includes: S110. Obtain the freight document data to be identified, perform standardized preprocessing on the freight document data to be identified, and obtain standardized freight document data. The freight document data to be identified includes booking data and / or bill of lading data. The standardized freight document data includes at least the cargo control party number, cargo flow data, and standard text of cargo name.

[0021] Specifically, the freight document data to be identified can be understood as the original data object for cargo attribute identification. This can be obtained by capturing complete original freight document form data submitted by users and collected by the system in the ocean freight document business scenario. The freight document data to be identified includes booking data and / or bill of lading data. For example, it can be imported through a visual interactive interface. The appropriate data acquisition method can be adopted according to actual needs, without limitation here. Standardized freight document data can be understood as structured and standardized data representing the freight document data to be identified, providing core data basis for subsequent cargo attribute identification and data matching. It can be generated by standardizing the original freight document data to be identified. Standardization preprocessing includes, but is not limited to, removing irrelevant special characters, eliminating invalid and redundant information, completing key fields, unifying data expression standards, normalizing null values, and unifying format conversion. The cargo control party number can be understood as a unique code used to identify the cargo ownership and control responsibility entity in the freight document business. It can accurately locate the responsible enterprise or institution for the cargo. It can be obtained by extracting a unique identifier code from the cargo owner, freight forwarder, or cargo control entity field of the original freight document data and then standardizing it. Cargo flow data can be specifically understood as the core business data recording the entire process of a single freight shipment, covering transportation routes, batches, cargo quantities, carrier information, loading and unloading nodes, and other flow-related content. This can be achieved by sorting through the transportation flow-related fields in the original freight document data, integrating, deduplicating, and standardizing them to form structured cargo flow information. Standard cargo name text can be specifically understood as a unified and standardized cargo name text, distinct from colloquial, abbreviation, and common names. Based on an industry standard name database, irregular cargo names in the original freight documents can be replaced with synonyms, corrected, and standardized to eliminate ambiguity, ultimately resulting in a standardized and unified standard name text that can be used for keyword comparison.

[0022] Specifically, raw freight document data to be identified is collected through preset data collection methods. Then, the raw freight document data, which is from various sources, has inconsistent formats, and is not standardized, is preprocessed in a standardized manner to extract and generate structured and standardized freight document data, which includes at least the cargo control party number, cargo flow data, and standard text of cargo name.

[0023] In this embodiment, preprocessing effectively removes redundant, erroneous, missing, and non-standard information from the original freight document data to be identified, unifies the format and description standards of freight document data across the entire domain, and solves the problems of inconsistent definitions, inconsistent fields, diverse descriptions of goods, and low data availability in the original freight document data. This not only ensures the integrity, standardization, and uniformity of core business fields, but also provides an accurate, compliant, and unified data foundation for subsequent business processes. It avoids business identification deviations caused by non-standard original data from the source, and significantly improves the overall accuracy and stability of subsequent data processing and risk identification.

[0024] Optionally, the freight document data to be identified is subjected to standardized preprocessing to obtain standardized freight document data, including: cleaning each field information in the freight document data to be identified to obtain cleaned field information; for the goods name field in the cleaned field information, calling a pre-built noise regular expression pattern library, and removing redundant descriptive text in the goods name field through regular expression matching to obtain the standard text of the goods name corresponding to the goods name field; replacing the content of the corresponding field in the cleaned field information with the standard text of the goods name corresponding to the goods name field to obtain standardized freight document data.

[0025] The pre-built noise regularity pattern library can be understood as a set of regularity rules adapted to the business scenario of ocean freight documents, specifically used to identify and filter invalid and redundant content in the cargo name field. It can accurately match noise content in the cargo name text that has no actual business meaning and interferes with standardized recognition, and is the core tool for realizing intelligent and automated standardization of cargo names. Optionally, the method for constructing the noise regular expression pattern library is as follows: Obtain non-standard cargo name text from historical freight document data within a preset historical time period; comprehensively analyze and statistically summarize various types of noise content, such as frequently occurring redundant modifiers, invalid suffixes, remarks, special symbols, and irrelevant supplementary descriptions, in the cargo name data of freight documents; classify and extract various noise features based on the industry standard for cargo name specifications in maritime freight documents; write corresponding regular expression matching expressions to form individual noise identification rules; then verify, deduplicate, and optimize all regular expression rules, eliminating conflicting or invalid rules; finally, integrate all effective and accurate noise regular expression rules to construct a comprehensive noise regular expression pattern library that is adapted to freight document business scenarios. This library can stably and accurately automatically identify and clean various types of noise text in cargo names, ensuring the standardization and accuracy of cargo name preprocessing.

[0026] Specifically, a systematic data cleaning process is performed on all fields of the freight document data to be identified. Invalid data is removed, abnormal fields are corrected, missing information is filled in, and character format is converted to obtain clean and reliable field information. On this basis, for the core cargo name field in the cleaned fields, a pre-built noise regular expression pattern library is called. Through precise regular expression matching rules, redundant modifiers, invalid suffixes, redundant remarks, irrelevant symbols, and other non-business redundant descriptive text in the cargo name are automatically identified and removed, generating a unified, standardized, and interference-free standard cargo name text. Finally, the processed standard cargo name text is used to replace the original cargo name field content in the cleaned field information, ultimately generating standardized freight document data with complete fields, unified format, and standardized content.

[0027] For example, the cargo name field of the freight document data to be identified and historical freight documents is uniformly cleaned. The cleaning operation includes: normalizing null values ​​and converting them all to lowercase; batch removing redundant descriptions (such as "100CTNS", "500KG", "100×50×30", "MODEL-X2024", etc.) using a noise regular expression pattern library; removing irrelevant special characters, keeping only letters, numbers, spaces and hyphens, merging extra spaces, and outputting a clean cargo name text containing only the core description of the cargo.

[0028] In this embodiment, the original freight document data is purified from the source by field cleaning, solving the problems of messy original data and abnormal data interfering with business identification. At the same time, the dedicated regular expression noise pattern library is used to intelligently clean up redundant content in the product name, effectively improving the efficiency and consistency of the standardization of product name processing. It completely eliminates the problems of non-standard descriptions, complex and redundant content, and inconsistent formats of product names in freight documents, and avoids problems such as subsequent keyword matching errors and deviations in cargo attribute identification caused by differences in product name descriptions. It provides accurate, unified, and high-quality standardized data support for core business links such as cargo attribute identification and similarity matching of subsequent freight document data, and greatly improves the accuracy and stability of subsequent attribute identification.

[0029] S120. Retrieve the preset cargo category keyword library and the preset cargo attribute keyword library, identify cargo attributes in the standard text of cargo name, and determine the initial cargo attribute identification result corresponding to the freight document data to be identified.

[0030] The pre-defined cargo category keyword library can be understood as a set of standard terms for cargo types pre-classified and constructed, covering core characteristic words of cargo categories such as chemicals, metals, electronics, textiles, food, and furniture hardware. It is mainly used to identify cargo category attributes and can be constructed by identifying transportation specifications and maritime cargo category catalogs. The pre-defined cargo attribute keyword library can be understood as a set of professional keywords specifically used to identify various dangerous goods, pre-constructed based on cargo attribute transportation specifications and maritime dangerous goods category catalogs. It includes, but is not limited to, UN dangerous goods numbers, maritime dangerous goods standard symbols, and high-risk characteristic words such as flammable, corrosive, and toxic. At the same time, it is configured with brand and model filtering regular expressions to avoid brand differences interfering with category judgment. It is specifically used to identify the dangerous goods risk attributes of cargo. It can be constructed by comparing with national and industry dangerous goods lists, combining the characteristic words of dangerous goods names that frequently appear in the context of freight documents, and sorting, verifying, and eliminating ambiguous words. The initial cargo attribute identification result can be understood as the standardized judgment conclusion of the cargo attribute output for a single freight document data to be identified. It can be achieved by calling two preset keyword libraries, namely the preset cargo category keyword library and the preset cargo attribute keyword library, to perform bidirectional matching and verification on the preprocessed standard text of cargo name. Based on the matched keyword library type, it can automatically identify whether the cargo is dangerous goods, and then generate the corresponding standardized cargo attribute identification result.

[0031] Specifically, by retrieving the pre-configured preset cargo category keyword library and preset cargo attribute keyword library, and using the pre-processed standardized cargo name standard text as the identification object, the cargo attributes are intelligently compared and identified based on the dual-database linkage matching mechanism, accurately identifying whether the cargo in the current freight document belongs to the category of dangerous goods, and outputting the initial cargo attribute identification result corresponding to the freight document data to be identified.

[0032] For example, the preprocessed product name text is segmented and matched to extract the set of core industry goods keywords (such as "lithiumbattery" and "steelpipe"); at the same time, the text is traversed to match the goods attribute keywords and the regular rules of the four-digit UN Dangerous Goods Number (such as "UN3480") to determine whether the initial goods attribute identification results of the corresponding booking and bill of lading data belong to dangerous goods attributes.

[0033] In this embodiment, automatic matching and identification are performed based on a preset cargo category keyword library and a preset cargo attribute keyword library. This achieves the use of standardized keyword rules to avoid the problems of differences in human experience and subjective misjudgment. The identification standard is unified, reusable, and iterative, which effectively improves the distinguishability and accuracy of cargo attribute identification. It can quickly complete the automated identification of large batches of freight document data, significantly improve the efficiency of cargo attribute identification, and at the same time ensure the consistency, standardization, and reliability of the identification results, providing an accurate preliminary judgment basis for subsequent cargo attribute risk review and results.

[0034] Optionally, a preset cargo category keyword library and a preset cargo attribute keyword library are retrieved to perform cargo attribute identification on the standard text of the cargo name, and to determine the initial cargo attribute identification result corresponding to the freight document data to be identified. This includes: performing word segmentation on the standard text of the cargo name to obtain multiple word segmentation units; performing keyword matching on the preset cargo category keyword library based on the multiple word segmentation units to obtain cargo type matching results; performing matching on the preset cargo attribute keyword library based on the multiple word segmentation units to obtain cargo attribute matching results; and determining the initial cargo attribute identification result corresponding to the freight document data to be identified based on the cargo type matching result and the cargo attribute matching result.

[0035] Specifically, the cargo type matching result can be understood as data representing the matching status between the word segmentation units of the standard cargo name text and the cargo type feature keywords. This is achieved by comparing each segmented unit with a preset cargo category keyword database, generating corresponding matching results (matched or not matched) based on whether a match is found within the database. The cargo attribute matching result can be understood as data representing the matching status between the word segmentation units of the standard cargo name text and the cargo attribute feature keywords. This is achieved by simultaneously traversing and matching the same batch of word segmentation units against a preset cargo attribute keyword database, outputting corresponding dangerous goods matching results (matched or not matched) based on whether the word segmentation unit matches a cargo attribute feature keyword. These two types of matching results are independent and mutually comparative, providing bidirectional data for subsequent comprehensive identification of whether cargo belongs to dangerous goods.

[0036] Specifically, intelligent identification of cargo attributes in freight documents is achieved by retrieving preset cargo category keyword libraries and preset cargo attribute keyword libraries. First, the standardized cargo name text is processed with precise word segmentation to break it down into multiple independent, smallest-granularity word units, avoiding the inaccuracy problem caused by whole-sentence matching. Then, each word unit is bidirectionally searched and matched with the preset cargo category keyword library and the preset cargo attribute keyword library to obtain the corresponding cargo type matching results and cargo attribute matching results. Finally, the two types of matching results are combined for comprehensive comparison, cross-validation and logical judgment to accurately determine the initial cargo attribute identification result corresponding to the current freight document data to be identified.

[0037] In this embodiment, fine-grained word segmentation is used to accurately extract the core semantics of the product name. Simultaneously, a dual-database parallel matching and verification logic is employed, using a pre-set product category keyword database and a pre-set product attribute keyword database, forming a two-way recognition constraint. This effectively solves problems such as ambiguity, missed feature recognition, and inability to distinguish between vague product names that easily occur with single keyword matching. It significantly improves the recognition accuracy in scenarios involving complex product names, mixed product names, and similar product names. Furthermore, the entire recognition process is performed based on a standardized thesaurus and automated word segmentation matching rules, ensuring unified recognition standards and a high degree of automation. This not only overcomes the drawbacks of manual recognition, such as reliance on experience, strong subjectivity, and low efficiency, but also possesses strong scenario adaptability and stability. It can efficiently and in batches complete the initial screening of cargo attributes in freight document data, providing accurate and reliable preliminary identification basis for subsequent risk review.

[0038] S130. Based on the cargo control party number, cargo flow data, and standard text of cargo name, a search is performed in the pre-built historical freight document database, and a pre-built multi-algorithm weighted fusion method is used to perform similarity matching calculation to obtain the associated historical freight document data corresponding to the freight document data to be identified.

[0039] Specifically, the associated historical freight document data can be understood as historical freight document records stored in a pre-built historical freight document database that share highly similar characteristics with the current freight document data to be identified in terms of cargo control entity, transportation flow scenario, and cargo attributes. These records contain cargo attribute markers from historical business operations, which can be identified by recognizing the "Cargo Nature" field in the corresponding records. This provides an important reference for the risk review of the current freight document. The core search dimensions are the standardized cargo control entity number, cargo flow data, and standard cargo name text. Precise retrieval and similarity matching calculations are performed in the historical freight document database. By comprehensively comparing the consistency of cargo control entity, similarity of transportation flow characteristics, and matching degree of cargo name attributes, historical freight document data that meets the preset similarity threshold is selected and thus identified as the associated historical freight document data corresponding to the current freight document data to be identified. The pre-constructed multi-algorithm weighted fusion method can be understood as pre-selecting multiple text similarity algorithms and completing model building. Combining the freight document business scenario, the weight coefficients corresponding to each algorithm are pre-set. When performing document similarity matching, each algorithm independently calculates the similarity score of the corresponding fields between the freight document to be identified and the candidate historical freight documents. Then, the similarity scores of each group are weighted and summed according to the preset weights to obtain the fused comprehensive similarity result, which is used as the basis for determining the document correlation.

[0040] Specifically, the cargo control party number, cargo flow data, and standardized cargo name text are extracted from the freight documents to be identified. Based on these three key retrieval elements, an initial search is performed in a pre-built historical freight document database to select a set of candidate documents. Then, a pre-built multi-algorithm weighted fusion model is called to perform similarity matching operations on the candidate documents and the documents to be identified. The final similarity value is obtained by weighting and combining the results of different text similarity algorithms. Based on this, the associated historical freight document data corresponding to the freight document data to be identified is obtained.

[0041] In this embodiment, based on the cargo controller number, cargo flow data, and standard text of cargo name, a search is performed in a pre-built historical freight document database. A pre-built multi-algorithm weighted fusion model is called for similarity matching, which significantly improves the accuracy and adaptability of historical data association retrieval and effectively avoids problems such as mismatch, omission, and invalid matching. At the same time, similarity calculation is performed based on standardized fields to ensure the uniformity and objectivity of matching rules. It can efficiently and stably mine historical freight document data of similar goods, similar transport entities, and similar transport scenarios, providing comprehensive and effective historical data support for subsequent secondary verification and validation based on historical cargo attribute marking information, and significantly improving the comprehensiveness and reliability of overall cargo attribute identification.

[0042] S140. Based on the initial cargo attribute identification results and the cargo attribute tagging information corresponding to the associated historical freight document data, determine the target cargo attribute identification result of the freight document data to be identified.

[0043] Specifically, based on the initial intelligent cargo attribute recognition results obtained from cargo name segmentation and matching, the cargo attribute marking information retained in historical freight document data that is highly similar to the current freight document scenario and has been verified by business is cross-checked and jointly recognized. This process mutually verifies and corrects possible fuzzy recognition, boundary ambiguity, and misjudgment in a single recognition step, and finally accurately outputs the target cargo attribute recognition result corresponding to the freight document data to be identified.

[0044] In this embodiment, a two-layer verification mechanism, consisting of intelligent keyword preliminary judgment and historical real business record review, is used to determine the target cargo attribute identification result of the freight document data to be identified. This breaks the limitations of relying solely on keyword matching for identification and effectively compensates for the inaccuracy of preliminary judgment results in complex scenarios such as similar product names, ambiguous category boundaries, and special mixed cargo. It fully utilizes historical real business risk control experience, significantly reduces the probability of missed or misjudged cargo attributes and the risk of concealment, and significantly improves the accuracy, rigor, and scenario adaptability of the overall cargo attribute identification results. This makes the identification results more in line with the actual maritime freight document business risk standards, providing a reliable and traceable decision-making basis for freight document transportation risk control review.

[0045] Optionally, the target cargo attribute identification result of the freight document data to be identified is determined based on the initial cargo attribute identification result and the cargo attribute marking information corresponding to the associated historical freight document data, including: if the initial cargo attribute identification result is a first attribute, then the first attribute is determined as the target cargo attribute identification result of the freight document data to be identified; if the initial cargo attribute identification result is a second attribute, and the cargo attribute marking information of the associated historical freight document data is a first attribute identifier, then the target cargo attribute identification result of the freight document data to be identified is determined to be a suspected concealed document; if the initial cargo attribute identification result is a second attribute, and the cargo attribute marking information of the associated historical freight document data is a second attribute identifier, then the second attribute is determined as the target cargo attribute identification result of the freight document data to be identified.

[0046] The first attribute and the second attribute are two initial identification results obtained after performing cargo attribute identification on the freight document to be identified. The first attribute refers to dangerous goods, and the second attribute refers to general goods. The first attribute identifier and the second attribute identifier are annotation information pre-recorded in the associated historical freight document data. The first attribute identifier is used to indicate that the cargo in the historical document is dangerous goods, and the second attribute identifier is used to indicate that the cargo in the historical document is general goods. The first attribute and the first attribute identifier, and the second attribute and the second attribute identifier, respectively form a one-to-one attribute definition relationship, thereby realizing cross-comparison and verification between the model identification output results and the historical document filing annotation information.

[0047] Specifically, the target cargo attribute identification result is determined collaboratively by combining the initial cargo attribute identification result with the cargo attribute marker information carried by the associated historical freight documents. When the initial cargo attribute identification result output by the model is the first attribute, the first attribute is directly used as the final target cargo attribute identification result. When the initial cargo attribute identification result is the second attribute, the cargo attribute marker information of the associated historical freight documents is further compared. If the historical document marker is the first attribute identifier, it indicates that there is a significant conflict between the document identification conclusion and the historical freight record. Based on this, the freight document to be identified is determined to be a suspected concealed document. If the historical document marker is the second attribute identifier, the initial identification conclusion is accepted, and the second attribute is determined as the target cargo attribute identification result.

[0048] In this embodiment, a two-layer closed-loop verification system consisting of intelligent initial judgment and historical review is constructed to determine the cargo attribute identification result of the freight document data to be identified. Through clear standardized rules, explicit cargo attributes can be quickly identified, and implicit concealment risks that cannot be identified by single product name matching can be mined based on historical real business data. This effectively solves the pain points of traditional identification that relies solely on product name keyword identification, cannot identify implicit risks, and misses or misidentifies concealment behaviors. It avoids the limitations of identification based on a single identification dimension. The entire identification logic is clear, the rules are unified, and it can be automated. This not only improves the accuracy of cargo attribute identification but also strengthens the ability to investigate hidden illegal freight document risks, comprehensively improving the completeness, accuracy, and risk control of cargo attribute risk identification in freight documents.

[0049] Optionally, the method further includes: when the target cargo attribute identification result of the freight document data to be identified meets the target screening conditions, generating a cargo flow route risk distribution map based on the freight document data to be identified and the corresponding target cargo attribute identification result, which is used to display high-risk routes, wherein the target screening conditions are set based on risk attribute information.

[0050] Specifically, the target screening criteria can be understood as pre-configured screening rules based on risk attribute information. These rules are set by combining risk attribute information such as cargo attributes and suspected concealment judgment results of documents. They are used to screen document samples with risk characteristics from all unidentified freight documents. For example, screening thresholds and judgment rules are configured based on risk attribute information of regulatory concern. Suspected concealment labels, dangerous goods attributes, etc., can be selected individually or in combination as screening benchmarks to form a configurable screening logic. In this embodiment, data corresponding to the first attribute of the target cargo or the suspected concealment document can be judged as meeting the target screening criteria. The cargo flow route risk distribution map can be understood as a visual map constructed based on the screened risk document data. This distribution map uses the route information such as the origin, destination, and transportation path recorded in the freight documents to be identified as its spatial basis. It is generated by integrating the target cargo attribute identification results of each document, assigning risk levels, and superimposing statistics to visually display the transportation routes where various risks are concentrated. For example, the geographical information of the flight routes in the selected risk documents can be extracted, the number and risk level of risk documents for the same flight route can be summarized and statistically analyzed, and the statistically obtained risk quantification indicators can be mapped to the flight route spatial layer to generate a cargo flow route risk distribution map that can reflect the degree of risk clustering.

[0051] Specifically, when the system detects that the target cargo attribute identification result corresponding to the cargo document data to be identified meets the target screening conditions preset based on risk attribute information, the system combines the cargo document data to be identified and its target cargo attribute identification result to construct a cargo flow route risk distribution map to intuitively present high-risk routes.

[0052] In this embodiment, freight document data with risk characteristics can be automatically screened out, and route risk visualization modeling can be completed based on the transportation information and risk assessment results contained in the documents. Discrete document risk information is transformed into a spatial route risk map, which clearly identifies the transportation routes where risk behaviors such as concealment of dangerous goods occur. This makes it easier for regulatory personnel to quickly grasp the risk distribution pattern, realize targeted control of key routes, and improve the targeting and overall control efficiency of freight safety supervision.

[0053] In one specific embodiment, a Streamlit web interface is built, integrating configuration functions such as file import, similarity threshold adjustment, and enhanced algorithm start / stop; it supports searching by carrier keywords and manually filtering risk records by multiple selections; it generates a bar chart of cargo flow route risk distribution to intuitively display high-risk routes; it displays the results table online (including cargo controller number, carrier, new and old product names, new and old cargo properties, and similarity), and supports downloading both filtered results and all comparison results in Excel file byte streams for easy manual review and archiving. The complete business execution process for the freight document cargo attribute identification and dangerous goods concealment identification tool developed based on Streamlit is as follows: After the Streamlit web interface application starts, it initializes the page. Users can choose to directly load the local default PKL historical database, or import and verify custom historical data by uploading an Excel file or entering a local file path. Then, the system reads the new freight document data to be identified by either uploading an Excel file or entering a local path. After clicking the "Start Comparison" button on the page, the system completes the association and matching of new and old data by combining the CCP (Cargo Control Party) number with the Trade Lane cargo flow route, and enters the similarity calculation sub-process. The process includes: standardizing text preprocessing to remove noise from the cargo name; extracting core cargo keywords; detecting cargo attribute identifiers and applying score deduction penalties; and integrating the token set matching algorithm Token SetRatio, the token sorting and editing matching algorithm Token Sort Ratio, and the partial sliding matching algorithm Partial. The system calculates the final comprehensive similarity using three weighted matching algorithms: Ratio. It then determines whether the similarity reaches a preset threshold. If the similarity is below the threshold, the corresponding record is directly excluded. If the similarity reaches the threshold, the system further verifies the hazardous attributes of the goods in the old and new records. If only the hazardous attributes of the two records are inconsistent, the record is identified as a suspicious concealment and stored in the Session state cache. The page then loads and displays all comparison results. Users can choose to filter data by Shipper and generate Trade Lane route statistics charts and display the filtered tables. Finally, users can choose to download a subset of the filtered results or an Excel file containing all suspicious results. If no download is made, the process ends directly.

[0054] Optionally, before performing cargo attribute identification, multiple rule libraries can be initialized and configured, including a noise regularization pattern library, a preset cargo category keyword library, and a preset cargo attribute keyword library. The noise regularization pattern library is used to remove redundant descriptions in cargo names that are irrelevant to the actual cargo category, improving the purity and accuracy of subsequent text matching. For example, it can cover matching rules for: quantity words (e.g., "100 CTNS"), packaging units ("CARTONS", "PALLETS"), weight / volume ("500KG", "2CBM"), dimensions ("100×50×30"), general freight descriptions ("FREIGHT", "CARGO"), weight markings ("NW", "GW"), etc. The preset cargo category keyword library stores data categorized by chemical, metal, electronics, textile, food, and furniture / hardware, directly anchoring to the actual cargo category. The preset cargo attribute keyword library includes United Nations Dangerous Goods Numbers (UNxxxx) and International Maritime Dangerous Goods Code (IMDG). High-risk keywords such as "Class" (flammable / corrosive / toxic) are used; brand and model filtering regular expressions are configured to avoid brand differences interfering with category judgment. Additionally, a historical freight document database needs to be built, reading historical Excel data of freight documents within a preset historical time period, automatically validating the required fields for cargo control party number, cargo flow, cargo description, cargo nature, and carrier; deduplication is performed using cargo control party number, cargo flow, cargo description, and cargo nature as key fields to eliminate completely duplicate freight document records; the Python serialization module Pickle is used to persistently save the deduplicated data as a local default database, supporting one-click loading and use, and also allowing one-click updates and replacements of the database by uploading new data; it also supports both small file web page uploads and large file local path reading import modes.

[0055] The technical solution of this embodiment involves acquiring the freight document data to be identified, performing standardized preprocessing on the freight document data to obtain standardized freight document data, wherein the standardized freight document data includes at least the cargo control party number, cargo flow data, and standard text of cargo name; retrieving a preset cargo category keyword library and a preset cargo attribute keyword library, identifying cargo attributes on the standard text of cargo name, and determining the initial cargo attribute identification result corresponding to the freight document data to be identified; based on the cargo control party number, cargo flow data, and standard text of cargo name, performing similarity matching in a pre-built historical freight document database to obtain the associated historical freight document data corresponding to the freight document data to be identified; and determining the target cargo attribute identification result of the freight document data to be identified based on the initial cargo attribute identification result and the cargo attribute tagging information corresponding to the associated historical freight document data. This solution achieves standardized, intelligent, and dual-verification of cargo attribute identification in freight documents. It ensures data accuracy from the source through data preprocessing, enables rapid and accurate identification of cargo names based on a dedicated keyword database, and completes result verification through historical data similarity matching. This effectively solves the problems of low efficiency, strong subjectivity, high rates of missed and false judgments, and concealment of cargo attributes associated with traditional manual identification. It significantly improves the accuracy, comprehensiveness, and efficiency of cargo attribute identification in freight documents. Furthermore, based on a multi-dimensional data linkage verification mechanism, it strengthens the reliability and credibility of the identification results, providing standardized and traceable data support for freight document security risk control.

[0056] Example 2 Figure 2 This is a flowchart of a maritime cargo attribute identification method based on multi-algorithm weighted fusion provided in Embodiment 2 of the present invention. The method in this embodiment is a further optimization of the method in the above embodiments. Optionally, a search is performed in a pre-built historical freight document database based on the cargo control party number and cargo flow data to obtain freight document data retrieval results. If the freight document data retrieval results contain at least one historical freight document record, the historical cargo name text is extracted from each historical freight document record, and the similarity between the historical cargo name text and the standard cargo name text is calculated using a pre-built multi-algorithm weighted fusion method to obtain similarity data corresponding to each historical freight document record. The historical freight document records corresponding to the similarity data that meet the preset similarity threshold in the similarity data corresponding to each historical freight document record are determined as the associated historical freight document data corresponding to the freight document data to be identified. Figure 2 As shown, the method includes: S210. Obtain the freight document data to be identified, perform standardized preprocessing on the freight document data to be identified, and obtain standardized freight document data. The freight document data to be identified includes booking data and / or bill of lading data. The standardized freight document data includes at least the cargo control party number, cargo flow data, and standard text of cargo name.

[0057] S220. Retrieve the preset cargo category keyword library and the preset cargo attribute keyword library, perform cargo attribute recognition on the standard text of cargo name, and determine the initial cargo attribute recognition result corresponding to the freight document data to be recognized.

[0058] S230. Based on the cargo control party number and cargo flow data, a search is performed in the pre-built historical freight document database to obtain the freight document data search results.

[0059] Specifically, based on the standardized cargo control party number and cargo flow data, a precise dimensional search is performed in the pre-built historical freight document database. The cargo control party number is used to accurately locate historical freight document records of the same cargo control entity. Then, the search scope is constrained and filtered in combination with the corresponding transportation and cargo flow characteristics. Irrelevant historical data with inconsistent entities or large differences in transportation scenarios are eliminated, and finally, accurate and reliable freight document data search results are obtained.

[0060] In this embodiment, the dual-dimensional joint retrieval mode based on cargo control party number and cargo flow data can effectively narrow the retrieval scope and filter invalid interference data compared to single keyword retrieval, avoiding the problems of broad retrieval data and distorted matching objects. It can accurately locate historical freight document samples that are consistent with the current freight document business entity and similar to the transportation scenario, providing a clean and effective dataset foundation for subsequent similarity matching and risk review, significantly improving the accuracy of subsequent data matching and the overall reliability of risk identification. At the same time, the retrieval rules are standardized and highly automated, which can adapt to the rapid retrieval needs of large batches of freight document data and effectively improve the overall identification operation efficiency.

[0061] S240. If the freight document data retrieval results contain at least one historical freight document record, then extract the historical cargo name text from each historical freight document record, and use a pre-constructed multi-algorithm weighted fusion method to calculate the similarity between the historical cargo name text and the standard cargo name text, thereby obtaining the similarity data corresponding to each historical freight document record.

[0062] Among them, the pre-constructed multi-algorithm weighted fusion method integrates a variety of preset fuzzy matching algorithms, which may include word segmentation set matching algorithm, word segmentation sorting and editing matching algorithm, and local sliding matching algorithm, etc.

[0063] Specifically, based on the retrieved freight document data, a pre-constructed multi-algorithm weighted fusion method is used to further calculate the similarity of product names for the retrieved results with valid historical freight document records. Specifically, when there is at least one valid historical freight document record in the retrieved freight document data, the historical product name text stored in each historical freight document record is extracted one by one, and its similarity is compared with the current standardized preprocessed standard product name text. The similarity data corresponding to each historical freight document record is then quantified.

[0064] In this embodiment, based on the cargo control party number and cargo flow retrieval screening, a more refined similarity quantification assessment is conducted from the cargo name dimension. The numerical similarity results replace the traditional subjective and fuzzy comparison, making the data matching process more objective, standardized, and quantifiable. This can accurately distinguish the degree of cargo similarity between different historical freight document records and current freight document data, effectively avoiding the problems of inaccurate cargo attribute matching and insufficient scenario fit caused by relying solely on subject and cargo flow screening. This provides accurate and reliable quantitative basis for subsequent screening of highly correlated historical freight document data and cross-verification of cargo attributes, further improving the overall precision and accuracy of risk identification.

[0065] Optionally, a pre-constructed multi-algorithm weighted fusion method is used to calculate the similarity between historical cargo name text and standard cargo name text, obtaining similarity data corresponding to each historical freight document record. This includes: calculating multiple basic similarity scores between historical cargo name text and standard cargo name text based on multiple preset fuzzy matching algorithms, including word segmentation set matching algorithm, word segmentation sorting and editing matching algorithm, and local sliding matching algorithm; determining the number of intersections and unions of keyword sets between historical cargo name text and standard cargo name text, and determining the keyword matching similarity score based on the number of intersections and unions of keyword sets; and performing weighted fusion based on multiple basic similarity scores and keyword matching similarity scores to obtain the similarity between historical cargo name text and standard cargo name text.

[0066] The system includes multiple pre-configured fuzzy matching algorithms, which are multi-dimensional text similarity calculation algorithms pre-configured to address business characteristics such as non-standard word descriptions, inconsistent word order, and redundant local modifications in freight document cargo name texts. These algorithms include, but are not limited to, word segmentation set matching algorithms, word segmentation sorting and editing matching algorithms, and local sliding matching algorithms, which are used for differential comparison from three dimensions: word set overlap, word segmentation order difference, and local key text similarity. Multiple complementary fuzzy matching algorithms can be selected, adapted, and solidified based on common difference features of ocean freight name texts to form a comprehensive set of text comparison operators. The basic similarity score is a single-item similarity value calculated independently by each pre-configured fuzzy matching algorithm for historical cargo name texts and standard cargo name texts. This can be achieved by inputting two sets of cargo name texts one by one into each pre-configured fuzzy matching algorithm, performing text comparison and quantification according to the corresponding algorithm rules, thereby obtaining independent basic similarity scores across multiple dimensions. Keyword matching similarity score is a specialized quantitative result used to characterize the degree of overlap of core business keywords between two sets of product names. It can be obtained by extracting the core keyword sets of historical product name text and standard product name text, counting the number of keyword intersections and unions, and calculating the keyword matching similarity score based on the ratio of the number of intersections to the number of unions. This provides a detailed quantitative basis for subsequent multi-dimensional score weighted fusion and obtaining high-precision product name similarity results.

[0067] Specifically, multiple preset fuzzy matching algorithms, including word segmentation set matching algorithm, word segmentation sorting and editing matching algorithm, and local sliding matching algorithm, are invoked to compare historical product name texts with standard product name texts, outputting basic similarity scores across multiple dimensions. Simultaneously, the number of intersections and unions of keyword sets between the two types of product name texts is counted. Based on the quantitative relationship between intersections and unions, the keyword matching similarity score is calculated. Finally, all basic similarity scores and keyword matching similarity scores are weighted and fused together to obtain the final product name text similarity result after integrating multi-dimensional feature weights.

[0068] For example, a multi-dimensional fusion scoring mechanism is adopted to accurately calculate the similarity of product names. The overall similarity score range is set from 0 to 100. The specific calculation logic includes two dimensions: multi-algorithm fuzzy matching and core keyword precise matching. The final comprehensive score is obtained by weighting the results with fixed weights. The system synchronously calls the TokenSet Ratio word segmentation matching algorithm, the Token Sort Ratio word segmentation sorting and editing matching algorithm, and the Partial Ratio text fuzzy matching algorithm to complete the basic similarity calculation. The TokenSet Ratio decomposes the product name text into an unordered set of words, ignores word order and repeated words, and calculates the similarity based on the proportion of word intersection and union. It can achieve full score matching for texts with interchangeable word orders. The Token Sort Ratio calculates the edit distance similarity by sorting and concatenating the words alphabetically, which is suitable for scoring in scenarios with fine-tuning of word order. The Ratio method employs a local sliding window matching approach, accurately adapting to local similarity recognition in scenarios involving both long and short texts. Based on this, the system extracts two sets of core keywords, A and B, for comparative product names and assigns specific similarity scores. When both keyword sets are not empty, a percentage score is calculated using the ratio of the keyword intersection to the union; that is, if both keyword sets are not empty, |A∩B| / |A∪B|×100 is calculated. When only one side has a core keyword, 70% of the minimum score of Token Set Ratio and Token Sort Ratio is taken as the keyword matching score. When neither side has a core keyword, the average score of the three fuzzy matching algorithms is taken as the keyword matching score. The final comprehensive similarity score is calculated as follows: core keyword matching score 40%, Token Set Ratio score 30%, Token Sort Ratio score 20%, and Partial similarity score 20%. The Ratio score is calculated using a fixed weighting of 10%, which takes into account the generalization ability of unordered text matching, word order fine-tuning matching, and partial inclusion matching, as well as the accuracy of core category keyword matching, effectively improving the fault tolerance and accuracy of product name similarity matching.

[0069] Optionally, a fixed similarity score penalty mechanism can be set to adjust the similarity between the text of historical cargo names and the standard text of cargo names in the new freight document data. This mechanism applies to the comparison records between new and historical freight document data after matching the cargo control party number and cargo flow information. If the cargo names on the new and historical freight documents have a high text similarity, but the cargo nature is inconsistent (one is labeled as dangerous goods and the other as non-dangerous goods), a fixed deduction will be applied to the calculated basic similarity score. A fixed deduction of 20 points will be applied to each compliance scenario. For example, if the historical dangerous goods name is "LITHIUM ION BATTERY" and the new freight document's non-dangerous goods name is "LITHIUM". The two items, "BATTERY" and "BATTERY", have a base similarity score of 90. After deducting points through this penalty mechanism, the final similarity score is 70. It should be clarified that the final similarity score after correction by this mechanism is only used to represent the credibility that the goods corresponding to the two shipping documents belong to the same real category. The higher the score, the higher the similarity of the text descriptions of the two product names. It does not directly represent the degree of danger of the goods. The actual dangerous attributes of the goods are identified and marked separately by the independent cargo nature field, realizing the logical decoupling of category similarity matching and dangerous goods attribute identification, and effectively avoiding the problem of mismatch of categories caused by similar product names and inconsistent dangerous attributes.

[0070] In this embodiment, multiple fuzzy matching algorithms are used to complement and adapt to different scenarios of product name text differences. This approach can take into account the overlap of word segmentation sets, differences in word order editing, and coverage of local key text similarity features. At the same time, a keyword set matching dimension is introduced to strengthen the comparison weight of core business features and eliminate interference from invalid text features. Furthermore, a weighted fusion mechanism is used to scientifically integrate multi-dimensional scores, effectively improving the comprehensiveness, stability, and accuracy of product name similarity calculation. This approach can accurately adapt to complex scenarios in freight document business, such as inconsistent product name expression formats, word order differences, and different local modifications. It avoids the distortion problem in similarity calculation caused by a single algorithm and provides high-precision and quantifiable similarity data support for subsequent screening of historical freight document data and review of cargo attribute risks.

[0071] S250. Determine the associated historical freight document data corresponding to the freight document data to be identified from the similarity data corresponding to each historical freight document record that meets the preset similarity threshold.

[0072] Specifically, after obtaining the product name similarity data corresponding to each historical freight document record, all similarity data are compared and filtered. The similarity data that meets the preset similarity threshold is extracted, and the historical freight document record corresponding to the similarity data that meets the preset similarity threshold is taken as the associated historical freight document data with the best matching degree with the current freight document data to be identified, and the closest cargo attributes and freight document scenarios.

[0073] In this embodiment, the quantified similarity value obtained by multi-algorithm fusion calculation is used as the core screening criterion. The screening criteria are objective, unified, quantifiable, and free from human subjective bias. It can accurately identify historical freight document samples with the best business scenario matching degree that meet the preset similarity threshold. This effectively avoids the problem of messy matching references and lack of prominent effective information caused by the mixing of multiple sets of historical data. It ensures that the historical data used for risk review has reference value and scenario adaptability that meet the preset similarity threshold. This provides the most appropriate and reliable historical business basis for the cross-validation of subsequent cargo attribute identification results and the identification of hidden risks, further improving the accuracy and credibility of the overall cargo attribute identification results.

[0074] Optionally, the method also includes: if the similarity data corresponding to each historical freight document record is lower than the preset similarity threshold, it is determined that there is no associated historical freight document data corresponding to the freight document data to be identified; if there are multiple historical freight document records whose similarity data meets the preset similarity threshold, the similarity reward and punishment mechanism is invoked to adjust the corresponding similarity data.

[0075] The preset similarity threshold is a critical quantitative standard used to determine whether historical freight document data has effective reference value. It is used to filter out relevant historical data that is highly compatible with the current freight document data scenario and has significance for risk verification, while filtering out invalid historical samples with excessive differences between cargo attributes and transportation scenarios and extremely low reference value. It is a core constraint parameter to ensure the effectiveness of historical data verification and avoid mismatch interference. By combining the similarity distribution characteristics of product names of massive real maritime freight document samples, the business accuracy requirements of cargo attribute recognition, and the difference patterns of different cargo names, through a large number of sample training, simulation testing, and scenario iteration verification, the lower limit range of similarity of effective matching historical data is statistically determined. Combined with manual business experience, the fixed critical threshold obtained after calibration and optimization can ensure that effective historical data is not missed, while eliminating invalid matching data with low similarity and high interference to the greatest extent. This provides a unified, stable, and business scenario-adaptable quantitative recognition basis for similarity screening and fallback recognition logic.

[0076] Specifically, after obtaining the similarity data corresponding to each historical freight document record, the similarity data that meets the preset similarity threshold is compared with the preset similarity threshold. If all similarity data are below the preset similarity threshold, it is determined that there is no valid matching historical freight document data with similar scenarios for the current freight document data to be identified, and there is no historical risk control reference information that can be used for cross-verification. Under this premise, the system directly generates the final cargo attribute identification result of the freight document data to be identified based on the cargo attribute identification result obtained by matching the standard text of the cargo name. If the similarity data corresponding to multiple historical freight document records meets the preset similarity threshold, the similarity reward and punishment mechanism is invoked to adjust the corresponding similarity data.

[0077] In this embodiment, by setting a fixed similarity threshold to strictly control the effective admission conditions of historical data, invalid historical freight document data with low similarity, large scene differences, and extremely low reference value can be automatically filtered out, avoiding the reverse interference of biased historical data with the recognition results. At the same time, independent fallback recognition rules are set for unpopular goods, newly added goods, and new transportation scenarios without valid historical data, solving the technical blind spot caused by relying solely on historical data verification, which makes it impossible to identify new scenarios and scenarios without samples. This ensures that the entire cargo attribute recognition process can run stably and effectively in various scenarios with and without historical samples, taking into account both the rigor of the recognition results and the universality of the algorithm scenarios.

[0078] Optionally, if it is determined that there is no associated historical freight document data corresponding to the freight document data to be identified, the method further includes: determining the target cargo attribute identification result of the freight document data to be identified based on the cargo attribute identification result.

[0079] Specifically, if it is determined that there is no associated historical freight document data corresponding to the freight document data to be identified, the cargo attribute identification result is directly determined as the target cargo attribute identification result of the freight document data to be identified.

[0080] Optionally, if multiple historical freight document records have similarity data that meets a preset similarity threshold, a similarity reward and punishment mechanism is invoked to adjust the corresponding similarity data. This includes: extracting dangerous goods marking information corresponding to each of the multiple historical freight document records; if the dangerous goods marking information of each historical freight document record is consistent, the similarity data remains unchanged; if the dangerous goods marking information of at least one historical freight document record is inconsistent with the other records, the similarity data is reduced according to a preset step size.

[0081] Specifically, when the similarity data corresponding to multiple historical freight document records reaches a preset similarity threshold, a similarity reward and punishment mechanism is activated for similarity correction. First, the dangerous goods marking information carried by each of the multiple historical freight document records is extracted for consistency verification. If the dangerous goods marking information of all historical freight document records is consistent, the corresponding similarity data remains unchanged. However, if there is inconsistency in the dangerous goods marking information within multiple historical freight document records, the corresponding similarity data is reduced according to a preset step size. It should be noted that if the dangerous goods marking information carried by multiple historical freight document records is not completely consistent, that is, if the information in the carriageature field corresponding to the dangerous goods marking information includes both general cargo and dangerous goods, then the current similarity data needs to be adjusted. The similarity can be reduced according to a preset step size to complete the operation of adjusting the current similarity data.

[0082] For example, a matching record refers to the same comparison record in the historical database after the new booking is associated with the cargo controller number and cargo flow. For instance, if the historical cargo name is "LITHIUM ION BATTERY" (cargo attribute) and the new booking cargo name is "LITHIUM BATTERY" (non-cargo attribute), the basic similarity score is 90 points. Because one is a cargo attribute and the other is a non-cargo attribute, 20 points are deducted, resulting in a final similarity score of 70 points. The final similarity score is used to characterize the credibility of "whether the cargo in the two booking records is of the same true category." A higher score indicates a closer similarity in cargo name descriptions, rather than directly indicating the degree of danger. The degree of danger is separately identified by the "Cargo Nature (Cargo Attribute / Non-Cargo Attribute)" field.

[0083] S260. Based on the initial cargo attribute identification results and the cargo attribute tagging information corresponding to the associated historical freight document data, determine the target cargo attribute identification result of the freight document data to be identified.

[0084] The technical solution of this embodiment collects the original data of the freight documents to be screened and performs standardized preprocessing to clean the data and remove noise from the product names, generating standardized freight document data containing the cargo controller number, cargo flow data, and standard text of the cargo name, thus achieving the structuring and standardization of the original messy data. Then, it retrieves a preset cargo category keyword library and a preset cargo attribute keyword library to intelligently match and judge the standardized cargo name text, outputting the corresponding initial cargo attribute identification results, completing the preliminary identification of cargo risks. Subsequently, based on the dual dimensions of the cargo controller number and cargo flow data, it accurately retrieves the corresponding freight document data search results in a pre-built historical freight document database, quickly identifying the same... The system uses historical freight document samples from the same transportation scenario. When valid historical freight document records exist in the search results, the historical cargo name text of each historical freight document record is extracted one by one. The similarity data between each historical record and the current standard cargo name is quantified through a multi-algorithm fusion similarity calculation method. Historical freight document records corresponding to similarity data that meet the preset similarity threshold are selected as the associated historical freight document data of the current freight document data to be identified, ensuring that the matching historical samples are highly consistent with the current business scenario. Finally, the system combines the preliminary cargo attribute identification results from the previous step with the cargo attribute marking information inherent in the associated historical freight document data for comprehensive judgment, and accurately outputs the target cargo attribute identification results of the freight document data to be identified. This solution achieves fully automated identification throughout the entire process, including standardized data preprocessing, intelligent initial judgment, accurate matching of historical samples, and dual-layer cross-verification. It addresses the issues of inconsistent and disorganized original data by standardizing source data, ensures unified, efficient, and objective initial judgment standards based on a dual-keyword database, improves the accuracy of historical data association through dual-dimensional retrieval of main entities and cargo flows, and optimizes matching based on multi-algorithm similarity. Finally, it achieves dual verification by combining intelligent initial judgment with historical real business markers. This effectively overcomes the shortcomings of traditional manual identification, such as low efficiency, strong subjectivity, susceptibility to missed and false judgments, and difficulty in identifying concealed cargo attributes. It significantly improves the automation, accuracy, and scenario adaptability of cargo attribute risk identification in ocean freight documents, comprehensively covering the identification of explicit cargo attributes and the investigation of implicit concealment risks, providing accurate, reliable, and traceable intelligent technical support for freight document and freight security risk control.

[0085] Example 3 Figure 3 This is a schematic diagram of the structure of a maritime cargo attribute identification system based on multi-algorithm weighted fusion provided in Embodiment 3 of the present invention. Figure 3 As shown, the system includes: The standardized freight document data determination module 310 is used to acquire freight document data to be identified, perform standardized preprocessing on the freight document data to be identified, and obtain standardized freight document data. The freight document data to be identified includes booking data and / or bill of lading data. The standardized freight document data includes at least the cargo control party number, cargo flow data, and standard text of cargo name. The initial cargo attribute identification result determination module 320 is used to retrieve the preset cargo category keyword library and the preset cargo attribute keyword library, perform cargo attribute identification on the standard text of cargo name, and determine the initial cargo attribute identification result corresponding to the freight document data to be identified. The associated historical freight document data determination module 330 is used to search in the pre-built historical freight document database based on the cargo control party number, cargo flow data and cargo name standard text, and to perform similarity matching calculation using a pre-built multi-algorithm weighted fusion method to obtain the associated historical freight document data corresponding to the freight document data to be identified. The target cargo attribute identification result determination module 340 is used to determine the target cargo attribute identification result of the freight document data to be identified based on the initial cargo attribute identification result and the cargo attribute tag information corresponding to the associated historical freight document data.

[0086] The technical solution of this embodiment involves obtaining the freight document data to be identified through a standardized freight document data determination module, performing standardized preprocessing on the freight document data to be identified to obtain standardized freight document data. The freight document data to be identified includes booking data and / or bill of lading data, and the standardized freight document data includes at least the cargo control party number, cargo flow data, and standard text of the cargo name. An initial cargo attribute identification result determination module retrieves a preset cargo category keyword library and a preset cargo attribute keyword library to identify the cargo attributes of the standard text of the cargo name, determining the initial cargo attribute identification result corresponding to the freight document data to be identified. An associated historical freight document data determination module searches a pre-built historical freight document database based on the cargo control party number, cargo flow data, and standard text of the cargo name, and performs similarity matching calculation using a pre-built multi-algorithm weighted fusion method to obtain the associated historical freight document data corresponding to the freight document data to be identified. A target cargo attribute identification result determination module determines the target cargo attribute identification result of the freight document data to be identified based on the initial cargo attribute identification result and the cargo attribute tag information corresponding to the associated historical freight document data. This solution achieves standardized, intelligent, and dual-verification of cargo attribute identification in freight documents. It ensures data accuracy from the source through data preprocessing, enables rapid and accurate identification of cargo names based on a dedicated keyword database, and completes result verification through historical data similarity matching. This effectively solves the problems of low efficiency, strong subjectivity, high rates of missed and false judgments, and concealment of dangerous goods in traditional cargo attribute identification methods. It significantly improves the accuracy, comprehensiveness, and efficiency of cargo attribute identification in freight documents. Furthermore, based on a multi-dimensional data linkage verification mechanism, it strengthens the reliability and credibility of the identification results, providing standardized and traceable data support for freight document security risk control.

[0087] Based on the above embodiments, optionally, the standardized freight document data determination module 310 is used to perform data cleaning on each field information in the freight document data to be identified, and obtain cleaned field information; for the goods name field in the cleaned field information, a pre-built noise regular expression pattern library is called, and redundant descriptive text in the goods name field is removed by regular expression matching to obtain the standard text of the goods name corresponding to the goods name field; the content of the corresponding field in the cleaned field information is replaced with the standard text of the goods name corresponding to the goods name field to obtain the standardized freight document data.

[0088] Optionally, the initial cargo attribute identification result determination module 320 is specifically used to perform word segmentation processing on the standard text of cargo name to obtain multiple word segmentation units; perform keyword matching on the preset cargo category keyword library based on the multiple word segmentation units to obtain cargo type matching results; perform matching on the preset cargo attribute keyword library based on the multiple word segmentation units to obtain cargo attribute matching results; and determine the initial cargo attribute identification result corresponding to the freight document data to be identified based on the cargo type matching result and the cargo attribute matching result.

[0089] Optionally, the associated historical freight document data determination module 330 is specifically used to search the pre-built historical freight document database based on the cargo control party number and cargo flow data to obtain freight document data retrieval results; if the freight document data retrieval results contain at least one historical freight document record, then the historical cargo name text in each historical freight document record is extracted, and the similarity between the historical cargo name text and the standard cargo name text is calculated using a pre-built multi-algorithm weighted fusion method to obtain the similarity data corresponding to each historical freight document record. The pre-built multi-algorithm weighted fusion method includes multiple preset fuzzy matching algorithms; the historical freight document records corresponding to the similarity data that meet the preset similarity threshold in the similarity data corresponding to each historical freight document record are used to determine the associated historical freight document data corresponding to the freight document data to be identified.

[0090] Optionally, the historical freight document data determination module 330 is specifically used to calculate multiple basic similarity scores between the historical freight name text and the standard freight name text based on multiple preset fuzzy matching algorithms. Among them, the multiple preset fuzzy matching algorithms include word segmentation set matching algorithm, word segmentation sorting and editing matching algorithm, and local sliding matching algorithm; determine the number of intersections and unions of keyword sets between the historical freight name text and the standard freight name text, and determine the keyword matching similarity score based on the number of intersections and unions of keyword sets; and perform weighted fusion based on multiple basic similarity scores and keyword matching similarity scores to obtain the similarity between the historical freight name text and the standard freight name text.

[0091] Optionally, the associated historical freight document data determination module 330 is also specifically used to determine that if the similarity data corresponding to each historical freight document record is lower than the preset similarity threshold, there is no associated historical freight document data corresponding to the freight document data to be identified; if there are multiple historical freight document records whose similarity data meets the preset similarity threshold, the similarity reward and punishment mechanism is invoked to adjust the corresponding similarity data.

[0092] Optionally, if it is determined that there is no associated historical freight document data corresponding to the freight document data to be identified, the system is also specifically used to determine the target freight attribute identification result of the freight document data to be identified based on the initial freight attribute identification result.

[0093] Optionally, the associated historical freight document data determination module 330 is also specifically used to extract dangerous goods marking information corresponding to multiple historical freight document records. If the dangerous goods marking information of each historical freight document record is consistent, the similarity data remains unchanged; if the dangerous goods marking information of at least one historical freight document record is inconsistent with the other records, the similarity data is reduced according to the preset step size.

[0094] Optionally, the target cargo attribute identification result determination module 340 is specifically used to determine the first attribute as the target cargo attribute identification result of the freight document data to be identified if the initial cargo attribute identification result is the first attribute; if the initial cargo attribute identification result is the second attribute, and the cargo attribute marking information associated with the historical freight document data is the first attribute identifier, then the target cargo attribute identification result of the freight document data to be identified is identified as a suspected concealed document; if the initial cargo attribute identification result is the second attribute, and the cargo attribute marking information associated with the historical freight document data is the second attribute identifier, then the second attribute is determined as the target cargo attribute identification result of the freight document data to be identified.

[0095] The maritime cargo attribute identification system based on multi-algorithm weighted fusion provided in this embodiment of the invention can execute the maritime cargo attribute identification method based on multi-algorithm weighted fusion provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0096] Example 4 This embodiment provides a visual UI interface for identifying maritime cargo attributes using a multi-algorithm weighted fusion method. It is integrated into intelligent cargo detection terminals, warehouse management backends, and logistics sorting control systems to intuitively complete operations such as data import, intelligent attribute identification, cargo property comparison, data query, and result display. Figure 4 , Figure 5 The diagram shows a visual UI for a cargo nature comparison tool. This tool uses a left-right split layout, divided into two main areas: a left sidebar (data import, parameter configuration, and operation execution area) and a right main content area (calculation prompts, result filtering, visualization charts, data tables, and export function). Its core purpose is to read historical booking data and current new booking data, automatically match records with the same shipper and route, and filter out suspicious documents with highly similar cargo names but historically declared dangerous goods while currently declared ordinary goods, for business personnel to conduct risk verification. Detailed visualization page information is shown below: I. Complete Explanation of Left Sidebar Controls 1. Use the default database (1) Interface location: at the top of the "Data Files and Parameters" section in the sidebar, displayed after the program detects the local historical database file.

[0097] (2) Default database definition: The default database is a complete booking history of the past three years after deduplication, retaining only core matching fields such as CCP ID, Trade Lane, Cargo Description, Cargo Nature, and Shipper. The system automatically generates and saves the historical Excel file after the first upload, eliminating the need to repeatedly upload large historical files every time the tool is opened.

[0098] (3) Functional logic: Check: Directly load the locally saved historical database without manually uploading historical data; Uncheck: The historical Excel file must be manually uploaded every time the tool is used.

[0099] 2. Update the default database (1) Display conditions: It will only appear after the "Use default database" checkbox is checked.

[0100] (2) Explanation of meaning: The old version of the default history database already exists locally. When the historical data is updated, the original historical file has errors, or the history needs to be completely replaced, check this option, upload a brand new complete historical Excel file, and after the calculation is completed, it will directly overwrite and replace the old local file to generate a new version of the default database.

[0101] (3) Applicable scenarios: annual update of three years of historical booking data, and reset of damaged historical database data.

[0102] 3. New booking data retrieval method (radio button, containing two options) (1) Option 1: Upload an Excel file, applicable to small monthly new booking documents ≤200MB, uploaded directly through a web browser, which is the daily routine usage mode.

[0103] (2) Option 2: Read from local path, applicable to Excel files larger than 200MB. There is a size limit for uploading from the webpage. Manually enter the complete absolute path of the file on the local disk and the program will directly read the file on the local disk without file size limit.

[0104] (3) Function: It is compatible with business data sources of different sizes and solves the problem of file upload restrictions on web pages.

[0105] 4. Product Name Similarity Threshold (1) Control type: Numeric slider, range 50~100, default value 75.

[0106] (2) Meaning of terms: The system will calculate a similarity score of 0-100 for the names of the new and old goods after cleaning. Only when the document similarity score is greater than or equal to the set threshold will it be judged as a suspicious record with highly similar names.

[0107] (3) Numerical rules: The higher the threshold, the stricter the matching, and the fewer data are filtered out; the lower the threshold, the looser the matching, and the more suspicious documents are generated.

[0108] (4) Business purpose: Business personnel can freely adjust the screening criteria according to the degree of risk control.

[0109] 5. Enable enhanced similarity calculation (1) Control type: Checkbox, enabled by default.

[0110] (2) Meaning: When enabled, the optimized weighted similarity algorithm compute_similarity is called; when disabled, only basic text matching is used, and there are no business-specific optimization rules.

[0111] (3) The enhancement algorithm includes automatic cleaning of noisy content; weighting of core cargo keywords; and multi-dimensional weighted fusion.

[0112] 6. Start comparison (Execute button) The system's core trigger button, when clicked, automatically executes the complete business process: Load / update the historical database → Read and verify new booking data → Match new and old documents by CCP ID + Trade Lane → Batch calculate product name similarity → Filter high-risk underreporting records → Cache the results on the page and render them in the right-hand display area.

[0113] II. Explanation of Functional Modules on the Right-Side Main Page 7. Filter by Shipper (1) Overall meaning: All suspicious risk records are grouped by shipper, and the shipper can be selected by custom selection. Only the risk documents corresponding to the selected shipper are displayed, so as to realize the targeted shipper verification.

[0114] (2) Definition of terms: Shipper refers to the enterprise that actually submits the booking application to the shipping company and provides the goods. Shipper (New): The shipper corresponding to the current new booking document; Shipper (History): The shipper corresponding to the historical booking document.

[0115] Supporting controls within the module: ①Shipper search box: Enter company keywords to filter the list of shippers; ②Shipper multi-select component: Supports single selection, multi-selection, and selection of all shippers; ③Shipper statistics text: Displays the total number of all shippers and the number of shippers currently selected.

[0116] 8. Trade Lane (1) Definition of term: Trade Lane refers to a fixed transportation route from the port of origin to the port of destination of goods (such as China-Southeast Asia, East China-Europe and America routes).

[0117] (2) Matching rules: The system only compares the similarity of product names for new and old booking documents with the same CCP ID and the same Trade Lane. Documents from different routes are not included in the matching.

[0118] 9. Number of Trade Lane hits Within the current range of risk data filtered by Shipper, the total number of suspicious dangerous goods declarations for each route is counted, sorted in descending order of quantity, and the top 20 routes are displayed by default, generating a visual bar chart.

[0119] By analyzing the number of hits, high-risk routes can be identified intuitively, facilitating targeted and focused supervision.

[0120] 10. Percentage of Cargo Nature New results in the screening results (1) Basic term breakdown: Cargo Nature: nature of goods; Cargo Nature (History): nature of goods declared in the past, DG represents Dangerous Goods; Cargo Nature (New): nature of goods declared in the new past, GC represents General Cargo.

[0121] (2) Percentage meaning: This tool has built-in filtering rules that retain only documents for “historical DG dangerous goods and current month GC general goods”. After filtering, all new goods recorded are GC. This indicator is used to count the percentage of general goods documents in the filtered sample to all filtered data, and intuitively shows the overall structure of the risk sample.

[0122] 11. Meaning of the bottom results table The table is the final risk verification checklist, and the business meaning of each column field is as follows:

[0123] Overall business significance of the table: All documents in the list are high-risk and suspicious. The same shipper and the same route have a long history of declaring dangerous goods. However, this month, the goods with highly similar names are declared as ordinary goods, which poses a risk of concealing dangerous goods. Business personnel will review and verify each document according to the table.

[0124] 12. Explanation of the Excel Download Button (1) Download the filtered results as an Excel file: Only export the subset of data filtered for the current Shipper (shipper); (2) Download all results as Excel: Export all suspicious records matched in this operation, without being restricted by the Shipper (shipper) filtering conditions.

[0125] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for identifying maritime cargo attributes based on multi-algorithm weighted fusion, characterized in that, include: Obtain the freight document data to be identified, and perform standardized preprocessing on the freight document data to be identified to obtain standardized freight document data. The freight document data to be identified includes booking data and / or bill of lading data, and the standardized freight document data includes at least the cargo control party number, cargo flow data, and standard text of cargo name. Retrieve a preset cargo category keyword library and a preset cargo attribute keyword library, perform cargo attribute recognition on the standard text of the cargo name, and determine the initial cargo attribute recognition result corresponding to the freight document data to be identified; Based on the cargo control party number, the cargo flow data, and the standard text of the cargo name, a search is performed in the pre-built historical freight document database, and a pre-built multi-algorithm weighted fusion method is used to perform similarity matching calculation to obtain the associated historical freight document data corresponding to the freight document data to be identified. Based on the initial cargo attribute identification result and the cargo attribute tagging information corresponding to the associated historical freight document data, the target cargo attribute identification result of the freight document data to be identified is determined.

2. The method according to claim 1, characterized in that, The standardization preprocessing of the freight document data to be identified to obtain standardized freight document data includes: Data cleaning is performed on each field of the freight document data to be identified to obtain cleaned field information; For the product name field in the cleaned field information, a pre-built noise regular expression pattern library is called to remove redundant descriptive text in the product name field through regular expression matching, so as to obtain the standard text of the product name corresponding to the product name field. The standardized freight document data is obtained by replacing the content of the corresponding field in the cleaned field information with the standard text of the goods name field.

3. The method according to claim 1, characterized in that, The process of retrieving a preset cargo category keyword library and a preset cargo attribute keyword library, performing cargo attribute recognition on the standard text of the cargo name, and determining the initial cargo attribute recognition result corresponding to the freight document data to be identified includes: The standard text of the goods name is segmented into multiple segmentation units. Based on the multiple word segmentation units, keyword matching is performed in the preset goods category keyword library to obtain goods type matching results; Based on the multiple word segmentation units, the preset cargo attribute keyword library is matched to obtain cargo attribute matching results; Based on the cargo type matching result and the cargo attribute matching result, the initial cargo attribute identification result corresponding to the freight document data to be identified is determined.

4. The method according to claim 1, characterized in that, The process involves retrieving cargo control party numbers, cargo flow data, and standard text of cargo names from a pre-built historical freight document database, and then performing similarity matching calculations using a pre-built multi-algorithm weighted fusion method to obtain associated historical freight document data corresponding to the freight document data to be identified, including: Based on the cargo control party number and the cargo flow data, a search is performed in the pre-built historical freight document database to obtain freight document data search results. If the freight document data retrieval results contain at least one historical freight document record, then the historical cargo name text in each historical freight document record is extracted, and the similarity between the historical cargo name text and the standard cargo name text is calculated using the pre-constructed multi-algorithm weighted fusion method to obtain the similarity data corresponding to each historical freight document record. The pre-constructed multi-algorithm weighted fusion method includes multiple preset fuzzy matching algorithms. The historical freight document records corresponding to the freight document data to be identified are determined by identifying the historical freight document data that meet the preset similarity threshold among the similarity data corresponding to each historical freight document record.

5. The method according to claim 4, characterized in that, The method employs the pre-constructed multi-algorithm weighted fusion method to calculate the similarity between the historical cargo name text and the standard cargo name text, obtaining similarity data corresponding to each historical freight document record, including: Multiple basic similarity scores between the historical goods name text and the standard goods name text are calculated based on a variety of preset fuzzy matching algorithms. The various preset fuzzy matching algorithms include a word segmentation set matching algorithm, a word segmentation sorting and editing matching algorithm, and a local sliding matching algorithm. Determine the number of intersections and unions of the keyword sets of the historical product name text and the standard product name text, and determine the keyword matching similarity score based on the number of intersections and unions of the keyword sets; The similarity between the historical product name text and the standard product name text is obtained by weighted fusion based on the multiple basic similarity scores and the keyword matching similarity scores.

6. The method according to claim 4, characterized in that, The method also includes: If the similarity data corresponding to each historical freight document record is lower than the preset similarity threshold, it is determined that there is no associated historical freight document data corresponding to the freight document data to be identified. If multiple historical freight document records have similarity data that meet the preset similarity threshold, the similarity reward and punishment mechanism will be invoked to adjust the corresponding similarity data.

7. The method according to claim 6, characterized in that, If it is determined that there is no associated historical freight document data corresponding to the freight document data to be identified, the method further includes: Based on the initial cargo attribute identification results, the target cargo attribute identification results of the freight document data to be identified are determined.

8. The method according to claim 6, characterized in that, If multiple historical freight document records have similarity data that meets a preset similarity threshold, then a similarity reward and punishment mechanism is invoked to adjust the corresponding similarity data, including: Extract the dangerous goods marking information corresponding to each of the multiple historical freight document records. If the dangerous goods marking information of each of the historical freight document records is consistent, then the similarity data remains unchanged. If the dangerous goods marking information of at least one of the historical freight document records is inconsistent with the other records, then the similarity data is reduced according to a preset step size.

9. The method according to claim 1, characterized in that, The step of determining the target cargo attribute identification result of the freight document data to be identified based on the initial cargo attribute identification result and the cargo attribute tag information corresponding to the associated historical freight document data includes: If the initial cargo attribute identification result is the first attribute, then the first attribute is determined as the target cargo attribute identification result of the freight document data to be identified; If the initial cargo attribute identification result is the second attribute, and the cargo attribute marking information of the associated historical freight document data is the first attribute identifier, then the target cargo attribute identification result of the freight document data to be identified is identified as a suspected concealed document. If the initial cargo attribute identification result is the second attribute, and the cargo attribute tagging information of the associated historical freight document data is the second attribute identifier, then the second attribute is determined as the target cargo attribute identification result of the freight document data to be identified.

10. A maritime cargo attribute identification system based on multi-algorithm weighted fusion, characterized in that, include: The standardized freight document data determination module is used to acquire freight document data to be identified, perform standardized preprocessing on the freight document data to be identified, and obtain standardized freight document data. The freight document data to be identified includes booking data and / or bill of lading data, and the standardized freight document data includes at least cargo control party number, cargo flow data, and standard text of cargo name. The initial cargo attribute identification result determination module is used to retrieve a preset cargo category keyword library and a preset cargo attribute keyword library, perform cargo attribute identification on the standard text of the cargo name, and determine the initial cargo attribute identification result corresponding to the freight document data to be identified. The module for determining associated historical freight document data is used to search in a pre-built historical freight document database based on the cargo control party number, the cargo flow data, and the standard text of the cargo name, and to perform similarity matching calculation using a pre-built multi-algorithm weighted fusion method to obtain the associated historical freight document data corresponding to the freight document data to be identified. The target cargo attribute identification result determination module is used to determine the target cargo attribute identification result of the freight document data to be identified based on the initial cargo attribute identification result and the cargo attribute tag information corresponding to the associated historical freight document data.

Images