Cruise ship material procurement whole-process collaborative management method and system combined with multi-dimensional data

Abstract

The application provides a cruise ship material procurement whole-process collaborative management and control method and system combining multi-dimensional data, relates to the procurement management and control technical field, and the method comprises the following steps: converting an external procurement platform push into a semantic feature vector, and generating a time efficiency influence factor; performing similarity measurement on the semantic feature vector, judging a transaction data packet state, and calling a historical invoice index; when the transaction data packet state is a semantic fuzzy state, generating a natural language change summary; generating a supplier agreement file according to the natural language change summary, and embedding the semantic feature vector and the similarity measurement result into the agreement file as hidden watermark information. Through the application, the structural defects of inconsistent master-slave table data in the prior art can be solved, accurate differentiation and consistency processing of multiple pushes under the same procurement number can be realized, and the technical effects of eliminating inconsistent defects, improving agreement generation and price effectiveness, and enhancing whole-link reliability and traceability can be achieved.

Description

Technical Field

[0001] This application relates to the field of procurement control technology, and in particular to a collaborative control method and system for the entire procurement process of cruise ship supplies that combines multi-dimensional data. Background Technology

[0002] As the scale of cruise ship supply procurement continues to expand, the frequency of transaction data exchange between external procurement platforms and internal procurement systems is increasing. The consistency and reliability of transaction result data during cross-system transmission have become key factors affecting procurement execution efficiency. Currently, existing external procurement platform transaction data reception solutions typically rely solely on the purchase number or interface serial number as the sole criterion. When the external platform pushes transaction data with the same purchase number again due to network jitter, duplicate submissions by operators, or corrections to the transaction result, the internal system cannot effectively distinguish whether the push is a duplicate or a correction push carrying changes to business content. It often resorts to simply rejecting the data or partially overwriting certain fields to address this issue.

[0003] In summary, existing technologies suffer from structural defects due to the lack of effective identification and comparison mechanisms for transaction data. This leads to the inability to accurately distinguish and consistently process multiple pushes under the same purchase number. When the same data is repeatedly entered into the database or when corrected data is only partially covered, inconsistencies in master-slave table data can easily occur. This further affects the accuracy of automatically generated downstream procurement agreements and the validity of supplier agreement prices, thereby reducing the reliability and traceability of the entire data flow in cruise ship material procurement. Summary of the Invention

[0004] The purpose of this application is to provide a collaborative management and control method and system for the entire process of cruise ship material procurement that combines multi-dimensional data. This system aims to solve the structural defects in existing technologies, such as the lack of an effective identification and comparison mechanism for the business content of transaction data. These defects result in multiple pushes under the same procurement number being unable to be accurately distinguished and processed consistently. When the same data is repeatedly entered into the database or when the corrected data is only covered by some fields, inconsistencies in master and slave table data can easily occur. This further affects the accuracy of the automatic generation of downstream procurement agreements and the validity of supplier agreement prices, thereby reducing the reliability of data flow and business traceability in the entire cruise ship material procurement chain.

[0005] In view of the above problems, this application provides a method and system for collaborative management and control of the entire process of cruise ship material procurement that combines multi-dimensional data.

[0006] Firstly, this application provides a method for collaborative management and control of the entire process of cruise ship material procurement that combines multidimensional data. This method is implemented through a collaborative management and control system for the entire process of cruise ship material procurement that combines multidimensional data. The method includes: acquiring transaction data packages pushed by an external procurement platform; converting structured numerical fields and unstructured text description fields into semantic feature vectors and generating a timeliness impact factor; inputting the semantic feature vectors into a historical transaction vector library for similarity measurement; determining the status of the transaction data package by adjusting a judgment threshold based on the timeliness impact factor; and retrieving the historical document index corresponding to the historical vector closest to the semantic feature vector; when the transaction data package status is semantically ambiguous, performing retrieval enhancement based on the historical context corresponding to the semantic feature vector and the historical document index, and generating a natural language change summary; generating a corresponding supplier agreement file through a rule engine mapping based on the natural language change summary, and embedding the semantic feature vectors and similarity measurement results as hidden watermark information into the agreement file.

[0007] Preferably, the method for collaborative management and control of cruise ship material procurement combining multi-dimensional data further includes: quantifying the spatial proximity of the semantic feature vector with each historical vector stored in the historical transaction vector library, and calculating the original value of the highest historical vector proximity; dynamically shifting and correcting the preset standard discrimination interval based on the timeliness impact factor to generate a dynamic discrimination interval; comparing the original value of the highest historical vector proximity with the dynamic discrimination interval to mark the transaction data package as a highly similar state, a semantically ambiguous state, or a significantly different state; and retrieving the historical documents associated with the corresponding historical vector by reverse indexing according to the original value of the highest historical vector proximity to obtain the historical document index.

[0008] Preferably, the cruise ship material procurement collaborative management method combining multidimensional data further includes: if the original value of the highest historical vector proximity is higher than the upper limit of the dynamic discrimination interval, it is marked as a highly approximate state; if the original value of the highest historical vector proximity is between the upper and lower limits of the dynamic discrimination interval, it is marked as a semantically ambiguous state; if the original value of the highest historical vector proximity is lower than the lower limit of the dynamic discrimination interval, it is marked as a significantly different state.

[0009] Preferably, the cruise ship material procurement collaborative management method combining multi-dimensional data further includes: extracting the business occurrence time from the transaction data package; calculating the difference between the business occurrence time and the current time to obtain the original time interval value measured in a preset time unit; mapping the business occurrence time to one of multiple preset timeliness levels based on the original time interval value, and calculating the timeliness impact factor by combining the preset basic attenuation coefficient corresponding to each timeliness level.

[0010] Preferably, the cruise ship material procurement collaborative management method combining multi-dimensional data further includes: the multiple preset time-sensitive levels include at least a real-time transaction level, a recent retrospective level, and a long-term archiving level.

[0011] Preferably, the cruise ship material procurement collaborative management method combining multi-dimensional data further includes: jointly decoding the semantic feature vector and the historical semantic vector corresponding to the historical document index, extracting the target field, change value, change direction, and external influencing factors related to the current voyage business scenario that change between the two, to obtain the extracted content; calling a pre-trained retrieval enhancement generation model to combine the extracted content into descriptive text as the natural language change summary.

[0012] Preferably, the cruise ship material procurement collaborative management method combining multi-dimensional data further includes: when the transaction data package status is highly similar, directly extracting the historical approval conclusions and agreement templates corresponding to the historical document index, and archiving and storing the transaction data package as a time-sensitive version of the historical data in a lightweight manner.

[0013] Preferably, the cruise ship material procurement collaborative management method combining multi-dimensional data further includes: when the transaction data packet status is significantly different, it is determined that the transaction data packet has no valid historical reference, the transaction data packet is directly written into the master data table as a brand new business record, and a mandatory manual review task is generated and pushed to the approval terminal at the same time. After receiving the review approval instruction, it enters the rule engine mapping stage.

[0014] Preferably, the cruise ship material procurement collaborative management method combining multi-dimensional data further includes: performing digital signature verification on the transaction data packet, and after the verification is passed, checking each key field in the data packet for null values ​​or format type errors, and generating an initial screening verification result; if the initial screening verification result is unsuccessful, the current process is terminated and an alarm log is generated.

[0015] Secondly, this application also provides a cruise ship material procurement full-process collaborative management and control system combining multi-dimensional data, used to execute the cruise ship material procurement full-process collaborative management and control method combining multi-dimensional data as described in the first aspect, including: a timeliness impact factor generation module, used to obtain transaction data packages pushed by an external procurement platform, convert structured numerical fields and unstructured text description fields into semantic feature vectors, and generate timeliness impact factors; a historical document index retrieval module, used to input the semantic feature vectors into a historical transaction vector library for similarity measurement, combine the timeliness impact factors to determine the status of the transaction data package by adjusting the judgment threshold, and retrieve the historical document index corresponding to the historical vector closest to the semantic feature vectors; a natural language change summary generation module, used to perform retrieval enhancement based on the historical context corresponding to the semantic feature vectors and the historical document indexes when the transaction data package status is semantically ambiguous, and generate a natural language change summary; and a supplier agreement file generation module, used to generate a corresponding supplier agreement file through rule engine mapping based on the natural language change summary, and embed the semantic feature vectors and similarity measurement results as hidden watermark information into the agreement file.

[0016] The technical solution provided in this application has at least the following technical effects or advantages: by realizing the technical goal of constructing a multi-dimensional identification and dynamic comparison framework for transaction data business content to support accurate differentiation and consistency processing of multiple pushes under the same purchase number, it achieves the technical effects of eliminating the structural defects of inconsistent master and slave table data, improving the accuracy of automatic generation of downstream procurement agreements and the effectiveness of supplier agreement prices, and enhancing the reliability of data flow and business traceability of the entire cruise ship material procurement chain.

[0017] The above description is merely an overview of the technical solution of this application. To enable a clearer understanding of the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1This is a flowchart illustrating the collaborative management method for the entire process of cruise ship material procurement that incorporates multidimensional data, as described in this application.

[0020] Figure 2 This is a schematic diagram of the structure of the cruise ship material procurement collaborative management system that combines multi-dimensional data, as proposed in this application.

[0021] Figure labeling: Timeliness impact factor generation module 1, historical document index retrieval module 2, natural language change summary generation module 3, supplier agreement document generation module 4. Detailed Implementation

[0022] This application provides a collaborative management method and system for the entire cruise ship material procurement process, integrating multi-dimensional data. It addresses existing technologies that suffer from structural defects due to the lack of effective identification and comparison mechanisms for transaction data. These defects include the inability to accurately distinguish and consistently process multiple submissions under the same procurement number, and the potential for inconsistencies between master and slave tables when identical data is repeatedly entered into the database or when corrected data is only partially covered. This inconsistency further affects the accuracy of automatically generated downstream procurement agreements and the validity of supplier agreement prices, thereby reducing the reliability and traceability of the entire cruise ship material procurement data flow. The application aims to construct a multi-dimensional identification and dynamic comparison framework for transaction data to support accurate differentiation and consistent processing of multiple submissions under the same procurement number. This achieves the technical effects of eliminating structural defects in master-slave table data inconsistencies, improving the accuracy of automatically generated downstream procurement agreements and the validity of supplier agreement prices, and enhancing the reliability and traceability of the entire cruise ship material procurement data flow.

[0023] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0024] Example 1, please refer to the appendix. Figure 1 This application provides a method for collaborative management and control of the entire process of cruise ship material procurement that combines multidimensional data, and applies it to a collaborative management and control system for the entire process of cruise ship material procurement that combines multidimensional data. Specifically, it includes the following steps: S1: Obtain the transaction data package pushed by the external procurement platform, transform the structured numerical fields and unstructured text description fields into semantic feature vectors, and generate the timeliness impact factor.

[0025] Furthermore, this application also includes: performing digital signature verification on the transaction data packet, and after the signature is verified, checking each key field in the data packet for null values ​​or format type errors, and generating an initial screening verification result; if the initial screening verification result is unsuccessful, terminating the current process and generating an alarm log record.

[0026] Furthermore, this application also includes: extracting the business occurrence time from the transaction data packet; calculating the difference between the business occurrence time and the current time based on the time span to obtain the original time interval value measured in a preset time unit; mapping the business occurrence time to one of multiple preset timeliness levels based on the original time interval value, and calculating and generating the timeliness impact factor by combining the preset basic attenuation coefficient corresponding to each timeliness level.

[0027] Furthermore, this application also includes: the plurality of preset time-sensitive levels include at least a real-time transaction level, a recent retrospective level, and a long-term archiving level.

[0028] Specifically, after receiving the transaction data package pushed by the external procurement platform, a digital signature verification operation is performed on the transaction data package. The external procurement platform refers to a third-party business platform that operates independently of the cruise ship's material procurement system and is used to complete the inquiry, quotation, and transaction result generation. The transaction data package refers to a data set issued by the external procurement platform that contains the main procurement transaction information, supplier information, material details, and attachment information. The digital signature verification operation is the process of verifying the digital signature attached to the transaction data package using an asymmetric encryption algorithm. This process confirms that the transaction data package has not been tampered with during transmission and indeed originates from the claimed external procurement platform. After the digital signature verification operation is successful, each key field in the transaction data package is checked for empty values ​​or format type errors. Key fields are necessary data items that must be included in the transaction data package and used for subsequent business processing, including the purchase number, supplier code, material code, quotation amount, and validity period. Empty values ​​refer to situations where no data value is filled in the key field. Format type errors refer to situations where the data type filled in the key field is inconsistent with the default data type, such as filling a numeric field with a text string. After completing each check, an initial screening verification result is generated based on the check results. This result indicates whether the transaction data packet passes the initial screening for data integrity and format correctness. If the initial screening verification result is negative, the current process is terminated and an alarm log is generated. The current process refers to the complete processing from receiving the transaction data packet to completing the initial data screening. Terminating the current process means stopping any subsequent processing operations on the transaction data packet, including field conversion, similarity measurement, and protocol generation. The alarm log is a log file that writes the reason for the initial screening verification failure, the source identifier of the transaction data packet, and the check time to a persistent storage medium. This log file is used for subsequent auditing and troubleshooting. After generating the alarm log, no further processing steps for the transaction data packet are executed, thus preventing invalid data from entering the system and causing data pollution. If the signature verification passes and the fields are complete and correct, the semantic feature vector extraction step is allowed. Table 1 shows the detailed history of the initial screening verification of the transaction data packet.

[0029] Table 1: Detailed Record of Initial Screening and Verification History of Transaction Data Packets

[0030] Next, the structured numerical fields and unstructured text description fields are jointly transformed into semantic feature vectors. Structured numerical fields refer to data items in the transaction data package stored in a predetermined format and possessing clear numerical meaning, including the numerical sequence corresponding to the purchase number, the quoted amount, quantity, tax rate, and validity period timestamp. Unstructured text description fields refer to data items in the transaction data package existing in natural language form without fixed format constraints, including material names, specifications, supplier remarks, and descriptions of transaction conditions. The semantic feature vector maps these two types of fields to a fixed-dimensional numerical array using a pre-defined encoding model. This array represents the distribution of the original fields in the semantic space. During the transformation process, structured numerical fields are normalized and converted into numerical vector components, while unstructured text description fields are segmented and word-embedded and then converted into text vector components. The two components are concatenated to form a complete semantic feature vector. This semantic feature vector is used for subsequent similarity comparison with historical transaction vectors to determine the degree of correlation between the current transaction data package and historical records.

[0031] Extract the transaction occurrence time from the transaction data package. The transaction occurrence time refers to the original point in time when the procurement transaction result corresponding to the transaction data package was actually generated or confirmed in the external procurement platform. It is usually recorded as a timestamp in the header field or business body field of the transaction data package. Read the timestamp value from the predetermined field position in the transaction data package and convert it into the system's unified time format for subsequent calculations.

[0032] The difference between the transaction occurrence time and the current time is calculated to obtain the original time interval value measured in a preset time unit. The current time refers to the moment indicated by the system clock when the difference calculation is performed; this moment is dynamically obtained each time a transaction data packet is processed. The time span refers to the absolute duration between the transaction occurrence time and the current time. The difference calculation is a mathematical operation that subtracts the two moments to obtain the duration value. The preset time unit refers to the time measurement benchmark pre-specified in the system configuration; this benchmark can be minutes, hours, or days, used to uniformly represent the size of the time interval. The original time interval value is the duration expressed in the preset time unit after the difference calculation; this value provides a quantitative basis for subsequent timeliness level mapping. For example, a set of time parameters from an actual processing process is selected for illustrative purposes. The transaction occurrence time is taken from the transaction confirmation moment recorded in the transaction data packet, specifically 10:30:00 on June 13, 2024. The current time used for difference calculation is obtained by reading the system clock when receiving transaction data packets, specifically 14:20:00 on June 26, 2024. The default time unit in this example is minutes, meaning the time span is measured in integers of sixty seconds. The difference calculation operation involves subtracting the absolute number of seconds corresponding to the current time from the absolute number of seconds corresponding to the transaction occurrence time. The difference is then divided by the sixty seconds contained in one minute to obtain the quantity value measured in minutes. From 10:30:00 on June 13, 2024 to 14:20:00 on June 26, 2024, 13 complete days have passed. Each complete day contains 24 hours, and each hour contains 60 minutes. The 13 complete days translate to 13 multiplied by 24 and then by 60, resulting in 18720 minutes. From 10:30:00 on June 26, 2024 to 14:20:00 on June 26, 2026, a total of 3 hours and 50 minutes elapsed on the time record. 3 hours translates to 180 minutes, and adding 50 minutes gives 230 minutes. Adding the minutes from the two time periods together, 18720 minutes plus 230 minutes equals 18950 minutes. This value is the original time interval measured in minutes as the preset time unit.

[0033] The system maps the time of a business transaction to one of several preset timeliness levels based on the original time interval values. Combining this with preset basic attenuation coefficients corresponding to each timeliness level, a timeliness impact factor is calculated. Preset timeliness levels refer to several time interval categories pre-divided by the system, each corresponding to a different level of business timeliness sensitivity. These categories are specifically customized by those skilled in the art based on actual circumstances. The multiple preset timeliness levels include at least a real-time transaction level, a recent retrospective level, and a long-term archiving level. The real-time transaction level is the level corresponding to the shortest time interval among the preset timeliness levels, used to collect transaction data packets with a small time span between the business transaction time and the current time. Transaction data packets belonging to the real-time transaction level typically indicate that the procurement transaction result was recently generated by the external procurement platform and has not yet experienced a long circulation delay, thus exhibiting the highest timeliness. The real-time transaction level has the highest base decay coefficient among all levels, ensuring that transaction data packets belonging to this level retain a high weight in the dynamic offset correction of the subsequent similarity judgment threshold. The system tends to perform the most accurate duplicate identification and coverage judgment on these transaction data packets to ensure that the immediacy of price and supplier information is not excessively weakened by time factors. The recent retrospective level is a level category in the preset timeliness level corresponding to a medium time interval, used to include transaction data packets whose time span between the transaction occurrence time and the current time is in the middle range. Transaction data packets belonging to the recent retrospective level usually indicate that the purchase transaction results have occurred for some time, but have not yet reached the state of needing to be archived for a long time, and still have a certain degree of business reference value. The base attenuation coefficient associated with the recent retrospective level is smaller than that of the real-time transaction level, but larger than that of the long-term archive level. This results in transaction data packets belonging to this level receiving a moderate weight in the dynamic offset correction of the subsequent similarity judgment threshold. When processing these transaction data packets, a balance needs to be struck between timeliness and historical reference value; earlier transaction records cannot be completely ignored, nor can they be given the same weight as real-time data in influencing the judgment results. The long-term archive level is the level category corresponding to the longest time interval in the preset timeliness levels, used to collect transaction data packets with a large time span between the transaction occurrence time and the current time. Transaction data packets belonging to the long-term archive level typically indicate that the procurement transaction results occurred a long time ago, and the current reference value of their prices and supplier information has significantly decreased. Their main purpose has shifted to historical record archiving and audit traceability rather than directly influencing current procurement decisions. The base attenuation coefficient associated with the long-term archive level is the smallest among all levels, so that the transaction data packets belonging to this level are given the lowest weight in the dynamic offset correction of the subsequent similarity determination threshold. When processing these transaction data packets, the influence of time factors on the determination threshold is reduced, so as to prevent overly old transaction records from improperly interfering with the accuracy of current duplicate identification and correction coverage.

[0034] The base attenuation coefficient is a weighted value pre-assigned to each preset time-sensitivity level. This weighted value reflects the influence of different time-sensitivity levels on the similarity judgment threshold. The closer the time-sensitivity levels are, the larger the base attenuation coefficient is; the farther the time-sensitivity levels are, the smaller the base attenuation coefficient is. Mapping refers to comparing the original time interval value with the time interval specified by each preset time-sensitivity level to determine which level's interval the original time interval value falls into. The time-sensitivity impact factor is the final coefficient value obtained by using the base attenuation coefficient corresponding to the mapped time-sensitivity level as the initial value, and further fine-tuning it based on the relative position of the original time interval value within its respective level's interval. This final coefficient value is used for subsequent dynamic offset correction of the judgment threshold, so that earlier transaction data is given a lower weight in similarity judgment, while later transaction data is given a higher weight, thereby improving the time-sensitivity accuracy of duplicate identification and correction coverage.

[0035] S2: Input the semantic feature vector into the historical transaction vector library for similarity measurement, combine the timeliness impact factor to determine the status of the transaction data packet by adjusting the judgment threshold, and retrieve the historical document index corresponding to the historical vector that is closest to the semantic feature vector.

[0036] Furthermore, this application also includes: quantifying the spatial proximity of the semantic feature vector with each historical vector stored in the historical transaction vector library, and calculating the original value of the highest historical vector proximity; dynamically shifting and correcting the preset standard discrimination interval based on the timeliness impact factor to generate a dynamic discrimination interval; comparing the original value of the highest historical vector proximity with the dynamic discrimination interval to mark the transaction data package as a highly similar state, a semantically ambiguous state, or a significantly different state; and retrieving the historical documents associated with the corresponding historical vector by reverse indexing according to the original value of the highest historical vector proximity to obtain the historical document index.

[0037] Furthermore, this application also includes: if the original value of the highest historical vector proximity is higher than the upper limit of the dynamic discrimination interval, it is marked as a highly approximate state; if the original value of the highest historical vector proximity is between the upper limit and the lower limit of the dynamic discrimination interval, it is marked as a semantically ambiguous state; if the original value of the highest historical vector proximity is lower than the lower limit of the dynamic discrimination interval, it is marked as a significantly different state.

[0038] Specifically, the historical transaction vector library refers to a pre-established and continuously maintained set of vectors in persistent storage media. Each historical vector in this set corresponds to a semantic feature vector generated from a previously successfully processed and stored transaction document. The semantic feature vector is then compared with each historical vector stored in the historical transaction vector library using a pre-defined distance metric function, such as cosine similarity, to mathematically calculate the geometric proximity between the semantic feature vector and each historical vector, thus obtaining a numerical value representing their similarity. Each historical vector undergoes one quantization calculation, resulting in a proximity value corresponding to its semantic feature vector. All proximity values ​​corresponding to historical vectors form a numerical set. The highest original proximity value of the historical vector is then calculated, i.e., the maximum value is taken from this numerical set as the highest original proximity value of the historical vector. This original value reflects the semantic proximity level between the current transaction data packet and the most similar historical transaction record. For example, the semantic feature vector is represented by A, and each historical vector stored in the historical transaction vector library is represented by B. i Let , where i is used to distinguish different historical vectors in the historical transaction vector library, and the value of i ranges from 1 to n, where n represents the total number of historical vectors stored in the historical transaction vector library. The spatial proximity is quantified using cosine similarity, expressed by the formula sim(A,B). i )=(A·B i ) / (||A||×||B i ||), where A·B i Represents semantic feature vector A and history vector B i The dot product operation refers to multiplying the values ​​at corresponding positions of two vectors and then summing the products to obtain a scalar value. ||A|| represents the magnitude of the semantic feature vector A, which is the square root of the sum of the squares of the values ​​at each position of the vector. ||B i || represents the historical vector B i The modulus is calculated in the same way as ||A||. This is achieved by comparing the semantic feature vector A with each historical vector B. i The calculations are performed separately, and each calculation yields a proximity value sim(A,B). i The proximity value ranges from -1 to 1. A value closer to 1 indicates a higher degree of directional consistency and geometric closeness between the semantic feature vector and the historical vector in the semantic space. Conversely, a value closer to -1 indicates a higher degree of directional opposition and geometric closeness between the semantic feature vector and the historical vector in the semantic space. Each historical vector B stored in the historical transaction vector library... iAfter sequential quantization calculations, n proximity values ​​corresponding to the semantic feature vector A are obtained, namely sim(A,B1) to sim(A,B1). n The complete numerical sequence of proximity data is used, where each value in the sequence represents the degree of similarity between the semantic features of the current transaction data packet and the corresponding historical transaction record. After all proximity values ​​are calculated, the sequence is from sim(A,B1) to sim(A,B1). n The maximum value is taken from the complete numerical sequence of ), and this maximum value is the original value of the highest historical vector proximity, denoted by the symbol S. max S indicates that max =max{sim(A,B i S | i=1,2,...,n} max It reflects the semantic proximity level between the current transaction data packet and the most similar one among all historical transaction records, providing a quantitative comparison benchmark for subsequent landing point comparison with dynamic discrimination intervals.

[0039] Dynamic offset correction is applied to a preset standard discrimination interval based on the timeliness impact factor to generate a dynamic discrimination interval. The preset standard discrimination interval refers to a pair of numerical boundaries predetermined during the offline configuration phase. This pair of boundaries includes an upper limit and a lower limit, used for preliminary classification of spatial proximity before adjustment for timeliness factors. Specific values ​​are customized by those skilled in the art based on actual conditions. Dynamic offset correction involves applying the timeliness impact factor to the upper and lower limits of the preset standard discrimination interval using a preset calculation method, causing both to shift numerically in the same direction and magnitude, thereby generating a new pair of numerical boundaries after timeliness adjustment, i.e., the dynamic discrimination interval. The generation of dynamic discrimination intervals means that the judgment threshold is no longer a fixed static value, but can adaptively change according to the timeliness of transaction data. Newer transaction data obtains a wider or narrower discrimination interval, while older transaction data receives the opposite adjustment direction, thereby improving the adaptability of repeated identification and correction coverage across different timeliness scenarios.

[0040] The highest historical vector proximity value refers to the maximum value in the sequence obtained after quantifying the spatial proximity between the semantic feature vector and all historical vectors in the historical transaction vector library. It represents the semantic closeness between the current transaction data packet and the most similar historical transaction record. If the highest historical vector proximity value exceeds the upper limit of the dynamic discrimination interval, the current transaction data packet is marked as highly similar. Highly similarity is a classification label indicating that the business content described by the current transaction data packet and the corresponding historical vector in the historical transaction vector library is substantially consistent. Substantial consistency means that there are no substantial differences between the two data packets in key business fields such as purchase number, supplier code, material code, quotation amount, and validity period that would affect the procurement execution decision. In the highly similar state, it is confirmed that the currently pushed transaction data packet is a duplicate of an existing historical record; no new write or overwrite update operation is required, nor is it necessary to trigger the subsequent search enhancement summary generation step.

[0041] When the original value of the highest historical vector proximity is between the upper and lower limits of the dynamic discrimination interval, meaning the original value of the highest historical vector proximity is greater than the lower limit and less than the upper limit, i.e., falling within the intermediate value band defined by the dynamic discrimination interval, the current transaction data packet is marked as semantically ambiguous. Semantically ambiguous state is a classification labeling result used to indicate that there is an identifiable business logic change between the current transaction data packet and the most similar historical record in the historical transaction vector library. An identifiable business logic change refers to differences in dimensions such as supplier identity, material specifications, transaction price, or effective period that can be perceived by the system, but the direction and magnitude of the change have not exceeded the historical evolution pattern. The historical evolution pattern refers to the time-series trend characteristics such as price fluctuation range, supplier replacement frequency, and material specification change patterns statistically derived from the historical transaction vector library. In the semantically ambiguous state, it is impossible to determine whether the current data should be ignored as duplicate data or overwritten as corrected data based solely on the single value of spatial proximity. Therefore, it is necessary to trigger the subsequent search enhancement summary generation step, which combines the historical context pointed to by the historical document index to perform in-depth analysis of the changed content, in order to clarify the specific fields involved and the direction of the change.

[0042] A value below the lower limit of the dynamic discrimination interval means that the original value of the highest historical vector proximity is less than the lower limit, i.e., it falls within the low-value region defined by the dynamic discrimination interval. When the original value of the highest historical vector proximity is below the lower limit of the dynamic discrimination interval, the current transaction data package is marked as having a significant difference. A significant difference is a classification labeling result used to indicate that the semantic distance between the current transaction data package and all historical vectors in the historical transaction vector library is large, and there is no similarity basis between the current data and any historical record sufficient to trigger repeated identification or correction coverage. A significant difference indicates that the current transaction data package represents a completely new procurement business scenario. A completely new procurement business scenario refers to a procurement transaction result newly generated in an external procurement platform involving supplier combinations, material category structures, or pricing models that have never appeared in the historical transaction vector library, or there is an irreconcilable conflict between the current data and historical records. An irreconcilable conflict means that the procurement number in the current data is the same as in the historical records, but the supplier code and material details have fundamentally changed simultaneously, making it impossible to ensure data consistency by simply overwriting historical records. In cases of significant differences, it is determined that there is no need to retrieve the historical document index. Retrieving the historical document index means locating the storage location identifier in the database based on the historical vector corresponding to the original value of the highest historical vector proximity. Since there are no historical records with reference value, the current transaction data package is directly used as the new business baseline. The new business baseline refers to a completely new data base version that is independent of any historical records. This version serves as the original basis for the generation of subsequent procurement agreements and the initialization of material prices. The current data is written to the master data table in an independent processing manner without being associated with any historical records.

[0043] Based on the inverse index of the highest historical vector proximity value, the historical documents associated with the corresponding historical vector are retrieved. That is, the historical vector corresponding to the highest historical vector proximity value is used as the query keyword to locate the complete storage record to which the historical vector belongs in the historical transaction vector database, thus obtaining the historical document index. A historical document refers to a set of transaction data that has been received, verified, saved, and synchronized with the protocol in previous processing flows. This set includes transaction master table records, supplier detail records, material detail records, and attachment records. A historical document index is a locator used to uniquely identify the storage location of the historical document in the database. This locator can be a primary key identifier or a storage path string.

[0044] S3: When the transaction data packet status is semantically ambiguous, perform retrieval enhancement based on the semantic feature vector and the historical context corresponding to the historical document index to generate a natural language change summary.

[0045] Furthermore, this application also includes: jointly decoding the semantic feature vector and the historical semantic vector corresponding to the historical document index to extract the target field, the changed value, the change direction, and the external influencing factors related to the current voyage business scenario that are changing between the two, so as to obtain the extracted content; calling a pre-trained retrieval enhancement generation model to combine the extracted content into descriptive text as the natural language change summary.

[0046] Furthermore, this application also includes: when the transaction data packet status is highly similar, directly extracting the historical approval conclusions and agreement templates corresponding to the historical document index, and storing the transaction data packet as a time-sensitive version of the historical data in a lightweight manner.

[0047] Furthermore, this application also includes: when the transaction data packet status is a significantly different status, it is determined that the transaction data packet has no valid historical reference basis, and the transaction data packet is directly written into the master data table as a brand new business record, and a mandatory manual review task is generated and pushed to the approval terminal at the same time. After receiving the review approval instruction, it enters the rule engine mapping stage.

[0048] Specifically, after applying a semantically ambiguous state marker to the current transaction data packet, a joint decoding operation is initiated to resolve the details of the differences between the current data and historical records. The joint decoding operation refers to the process of inputting the semantic feature vector and the historical semantic vector corresponding to the historical document index into the same decoding network for collaborative processing. The semantic feature vector represents the distribution position of the current transaction data packet in the semantic space, while the historical semantic vector refers to the semantic feature vector generated and saved when the historical transaction record pointed to by the historical document index is entered into the database. During the joint decoding process, the decoding network performs a dimension-by-dimensional comparison of the two vectors. This dimension-by-dimensional comparison involves performing a numerical difference operation between each component of the semantic feature vector and the component at the same position in the historical semantic vector. The result of the difference operation reflects the degree of deviation between the current data and the historical record in that semantic dimension. Based on the difference operation result, the target field, the changed value, the direction of change, and the external influencing factors related to the current voyage's business scenario are extracted. These four items together constitute the extracted content. The target field refers to the name identifier of the field in the business fields contained in the transaction data packet that has undergone numerical or content changes, such as the purchase number field, supplier code field, material code field, quotation amount field, or validity period field. The changed value refers to the specific data content of the original value of the target field recorded in the historical record and the new value recorded in the current transaction data package. The direction of change refers to the trend type of the new value relative to the original value, including different categories such as value increase, value decrease, supplier change, material specification upgrade, or validity extension. External influencing factors related to the current voyage's business scenario refer to auxiliary information independent of the internal fields of the transaction data package but capable of explaining the cause of the change. The current voyage's business scenario refers to the specific time period, route area, and resupply port conditions of the cruise ship's material procurement during this voyage operation. External influencing factors include seasonal price fluctuations, port rate adjustments, changes in supplier inventory levels, or changes in the supply and demand relationship in the shipping market. The extracted content serves as the output of joint decoding, providing structured difference description material for subsequent generation of natural language change summaries.

[0049] After obtaining the extracted content, a pre-trained retrieval enhancement and generation model is invoked to combine the extracted content into descriptive text, which serves as a natural language change summary output. The pre-trained retrieval enhancement and generation model refers to a language model that has been pre-trained on a large-scale text corpus and can dynamically retrieve external knowledge bases during the generation process. This model possesses the ability to convert structured input into fluent natural language output. When the retrieval enhancement and generation model is invoked, the target field names and their corresponding historical original values ​​and current new values, change direction categories, and descriptions of external influencing factors from the extracted content are used as model input. Simultaneously, auxiliary information is retrieved from the historical context corresponding to the historical document index. This auxiliary information includes historical approval conclusions, historical agreement price effective records, and historical supplier evaluation records. Under the combined effect of the input and auxiliary information, the retrieval enhancement and generation model reorganizes the various difference details in the extracted content according to chronological order and business logic causal relationships, combining them into a descriptive text with a complete narrative structure. The description text contains a complete logical chain of change deduction and an explanation of the expected impact. The logical chain of change deduction refers to the coherent deduction process that starts from the original state of the historical records, describing in sequence what external factors caused the field to change in what direction, the magnitude of the change, and whether the change is reasonable in business or in line with historical evolution patterns. The explanation of the expected impact refers to the description of the ripple effect of the new price predicted based on the change value and direction on subsequent procurement budgets, supplier agreement terms, and inventory replenishment plans. The combined description text output is the natural language change summary, which is read and understood by human reviewers at the approval terminal. This allows reviewers to quickly grasp the changes in the current transaction data package relative to the historical records, as well as the business reasons and impact chain behind the changes, without having to compare the original data one by one.

[0050] When the transaction data package status is marked as highly similar, an extraction operation is directly performed to obtain the historical approval conclusions and agreement templates corresponding to the historical document index. Historical approval conclusions refer to the final approval records generated after manual review or automatic rule verification of historical documents during previous processing. These conclusions record whether the historical transaction data was approved for entry into the procurement execution chain and the approval remarks attached upon approval. Agreement templates refer to the standardized contract structure documents on which the supplier agreements corresponding to historical documents were generated. These templates contain preset elements such as the agreement clause framework, price entry locations, applicable tax rate rules, and validity period format. Based on the historical document index, the complete content of historical approval conclusions and agreement templates is read from the approval record table and agreement template storage area of ​​the database, respectively, providing necessary reference information for subsequent archiving processing.

[0051] After extracting historical approval conclusions and agreement templates, the current transaction data package is stored as a lightweight archived version of the historical data. Historical data refers to the complete set of previously successfully processed and stored transaction records pointed to by the historical document index. The subsidiary time-sensitive version is a new sub-record relative to the main version of the historical data. This sub-record shares the same core business identifier but is appended with a timestamp of the current processing time, used to indicate duplicate or highly similar transaction data received at different points in time within the same business context. Lightweight archived storage refers to the process of writing the current transaction data package into the archived storage area in a simplified record format instead of the master data table. The simplified record format means that only the source identifier, reception time, data fingerprint summary, and association reference with the historical document index of the transaction data package are saved, without saving the complete transaction master table fields, supplier details set, and material details set. Executing lightweight archived storage also means terminating the current process. Terminating the current process means that after completing the archive write operation, the system will no longer execute any subsequent processing steps, including no longer triggering subsequent summary generation operations and no longer triggering subsequent agreement remapping operations. The summary generation operation refers to the process described above, where, under semantically ambiguous conditions, the retrieved content is combined into descriptive text by calling the retrieval enhancement generation model. The agreement remapping operation refers to the automatic synchronization process of converting transaction data into supplier agreements and agreement details. Under highly similar conditions, since the current data is essentially consistent with historical data, there is no need to regenerate change summaries to explain the differences, nor is it necessary to regenerate agreement files to overwrite existing valid agreements. Therefore, the system achieves traceable recording of duplicate data through lightweight archiving and storage, while avoiding unnecessary consumption of computing resources and the risk of agreement data overwriting.

[0052] When a transaction data packet is marked as "significantly different," it is determined that the current transaction data packet has no valid historical reference. A valid historical reference refers to the set of historical records in the historical transaction vector library that can establish a meaningful correspondence with the current transaction data packet in terms of core business fields. A meaningful correspondence means that the historical record matches the current data in key identifiers such as purchase number, supplier code, or material code, and can provide reference value for the processing decisions of the current data. In the "significantly different" state, since the semantic proximity of the current data to all historical records is below the lower limit of the dynamic discrimination interval, it is determined that there is no historical record in the historical transaction vector library that can serve as a valid reference for determining whether the current data is a duplicate, correction, or continuation. Therefore, the current data cannot be classified as any variant or subsidiary version of the existing historical record.

[0053] After determining that there are no valid historical references, the current transaction data package is directly written into the master data table as a brand new business record. A brand new business record refers to a newly added data entity independent of all existing records in the historical transaction vector library. This data entity has no duplication or overwriting relationship with historical records in terms of the combination of purchase number, supplier code, and material code. The master data table is the core persistent storage structure in the database used to store transaction master information, supplier details, material details, and ancillary information. This table is the basic data source for subsequent purchase agreement generation, price synchronization, and execution chain traceability. The write operation refers to inserting the values ​​of each business field in the current transaction data package into the corresponding database columns according to the preset field mapping relationship of the master data table, forming a complete new data row record. After the master data table write is completed, the brand new business record obtains a globally unique data identifier assigned by the system for reference and association in subsequent stages.

[0054] Simultaneously with the completion of the master data table write, a mandatory manual review task is generated and pushed to the approval terminal. The mandatory manual review task is a manual review unit that cannot be bypassed by automated rules. This task encapsulates all business content of the current transaction data packet, the data identifier written to the master data table, and the reasons for the judgment that there is no valid reference to the historical transaction vector database. The approval terminal refers to a computing or mobile device with a graphical user interface. This device runs an approval application and provides task viewing, data browsing, and opinion submission functions to business personnel with approval permissions. The push operation refers to transmitting the mandatory manual review task from the system backend service to the notification queue of the approval terminal via network protocol, enabling the approval terminal to display the number and summary information of pending tasks in real time.

[0055] After the mandatory manual review task is pushed to the approval terminal, it enters a waiting state until a review approval instruction is received before proceeding to the rule engine mapping stage. The review approval instruction is a confirmation signal sent to the system backend by the operator at the approval terminal after reviewing all the contents of the current transaction data package included in the mandatory manual review task, by clicking the approval button or entering an approval password. This indicates that the operator acknowledges the legality and accuracy of the current data as a new business record. The rule engine mapping stage is the automated processing step that converts supplier information, material details, transaction price, and validity period in the transaction data into supplier agreement documents and agreement detail records. This stage is a crucial transformation node for transaction data to enter the procurement execution chain.

[0056] S4: Generate the corresponding supplier agreement file by mapping the natural language change summary through the rule engine, and embed the semantic feature vector and similarity measurement results as hidden watermark information into the agreement file.

[0057] Specifically, a rule engine is a decision-making system pre-configured with conditional decision branches and action execution sequences, capable of converting key elements in natural language descriptions into structured protocol generation instructions. A supplier agreement document is a formal contract document recording the agreement between the purchaser and supplier regarding specific material types, transaction prices, applicable tax rates, and validity periods. This document is stored as a structured electronic file in the system's file storage area. The mapping operation involves mapping the changed values ​​and directions of the target fields in the natural language change summary to the price entry positions, applicable tax rate items, and validity period start and end dates in the agreement document according to the rule engine's built-in conversion rules. Simultaneously, unchanged fields retain their historical values ​​or use system default values. After the mapping operation is complete, a complete supplier agreement document is output. This document contains all necessary fields such as agreement number, agreement type, supplier identification, detailed material price list, tax rate configuration, effective date, and expiration date, thus providing a legally binding contractual basis for subsequent procurement execution.

[0058] After the supplier agreement document is generated, semantic feature vectors and similarity measurement results are embedded as hidden watermark information into the agreement document. The semantic feature vector is a fixed-dimensional numerical array mapped by a preset encoding model from structured numerical fields and unstructured text description fields. This array represents the distribution position of the current transaction data packet in the semantic space. The similarity measurement result refers to all values ​​or key statistics obtained after quantifying the spatial proximity between the semantic feature vector and each historical vector stored in the historical transaction vector library, including the original value of the highest historical vector proximity and the corresponding historical vector index. The hidden watermark information refers to digital marker content embedded in the agreement document in an invisible manner. This content does not change the readable text and layout of the agreement document, but can be extracted from the agreement document for verification using specialized detection tools. The embedding operation involves using a digital watermarking algorithm to disperse and hide the binary encoding of the semantic feature vector and similarity measurement results in the redundant data space of the agreement document. The redundant data space includes custom fields reserved in the file format specification, the least significant bit of numerical fields, or the space encoding gaps in text fields. By embedding, the agreement file carries not only the supplier agreement terms but also the semantic fingerprint of the original transaction data on which the agreement was generated, as well as the degree of similarity with historical records. This allows any subsequent agreement file exported or transmitted from outside the system to be traced back to its data source and processing decision basis through watermark extraction, preventing the agreement file from being illegally replaced or tampered with without being detected. At the same time, it provides auditors with technical clues to infer the business scenario to which the transaction data package belongs from the agreement file.

[0059] In summary, the cruise ship material procurement collaborative management method combining multi-dimensional data provided in this application has the following technical effects: by realizing the technical goal of constructing a multi-dimensional identification and dynamic comparison framework for transaction data business content to support accurate differentiation and consistency processing of multiple pushes under the same procurement number, it achieves the technical effects of eliminating structural defects in master-slave table data inconsistency, improving the accuracy of automatic generation of downstream procurement agreements and the effectiveness of supplier agreement prices, and enhancing the reliability of data flow and business traceability of the entire cruise ship material procurement chain.

[0060] Example 2: Based on the same inventive concept as the cruise ship supply procurement collaborative management method combining multi-dimensional data in the foregoing examples, this application also provides a cruise ship supply procurement collaborative management system combining multi-dimensional data. Please refer to the appendix. Figure 2 The system includes: a timeliness impact factor generation module 1, used to acquire transaction data packages pushed by an external procurement platform, convert structured numerical fields and unstructured text description fields into semantic feature vectors, and generate timeliness impact factors; a historical document index retrieval module 2, used to input the semantic feature vectors into a historical transaction vector library for similarity measurement, combine the timeliness impact factors to determine the status of the transaction data package by adjusting the judgment threshold, and retrieve the historical document index corresponding to the historical vector closest to the semantic feature vectors; a natural language change summary generation module 3, used to perform retrieval enhancement based on the historical context corresponding to the semantic feature vectors and the historical document indexes when the transaction data package status is semantically ambiguous, and generate a natural language change summary; and a supplier agreement file generation module 4, used to generate a corresponding supplier agreement file through rule engine mapping based on the natural language change summary, and embed the semantic feature vectors and similarity measurement results as hidden watermark information into the agreement file.

[0061] Furthermore, the cruise ship material procurement collaborative management system combining multi-dimensional data is also used for: quantifying the spatial proximity of the semantic feature vector with each historical vector stored in the historical transaction vector library, and calculating the original value of the highest historical vector proximity; dynamically shifting and correcting the preset standard discrimination interval based on the timeliness impact factor to generate a dynamic discrimination interval; comparing the original value of the highest historical vector proximity with the dynamic discrimination interval to mark the transaction data package as highly similar, semantically ambiguous, or significantly different; and retrieving the historical documents associated with the corresponding historical vector by reverse indexing according to the original value of the highest historical vector proximity to obtain the historical document index.

[0062] Furthermore, the cruise ship material procurement collaborative management system that combines multi-dimensional data is also used to: mark a highly approximate state if the original value of the highest historical vector proximity is higher than the upper limit of the dynamic discrimination interval; mark a semantically ambiguous state if the original value of the highest historical vector proximity is between the upper and lower limits of the dynamic discrimination interval; and mark a significantly different state if the original value of the highest historical vector proximity is lower than the lower limit of the dynamic discrimination interval.

[0063] Furthermore, the cruise ship material procurement collaborative management system that combines multi-dimensional data is also used to: extract the business occurrence time from the transaction data package; calculate the difference between the business occurrence time and the current time to obtain the original time interval value measured in a preset time unit; map the business occurrence time to one of multiple preset timeliness levels based on the original time interval value, and calculate and generate the timeliness impact factor by combining the preset basic attenuation coefficient corresponding to each timeliness level.

[0064] Furthermore, the cruise ship material procurement collaborative management system that combines multi-dimensional data is also used to: include at least a real-time transaction level, a recent retrospective level, and a long-term archiving level among the multiple preset time-sensitive levels.

[0065] Furthermore, the cruise ship material procurement full-process collaborative management and control system that combines multi-dimensional data is also used to: jointly decode the semantic feature vector and the historical semantic vector corresponding to the historical document index, extract the target fields, change values, change directions, and external influencing factors related to the current voyage business scenario that have changed between the two, and obtain the extracted content; call the pre-trained retrieval enhancement generation model to combine the extracted content into descriptive text, as the natural language change summary.

[0066] Furthermore, the cruise ship material procurement collaborative management system that combines multi-dimensional data is also used to: when the status of the transaction data package is highly similar, directly extract the historical approval conclusions and agreement templates corresponding to the historical document index, and archive and store the transaction data package as a time-sensitive version of the historical data in a lightweight manner.

[0067] Furthermore, the cruise ship material procurement full-process collaborative management and control system that combines multi-dimensional data is also used to: when the status of the transaction data packet is in a significantly different state, determine that the transaction data packet has no valid historical reference basis, directly write the transaction data packet as a brand new business record into the master data table, and simultaneously generate a mandatory manual review task to push to the approval terminal. After receiving the review approval instruction, it enters the rule engine mapping stage.

[0068] Furthermore, the cruise ship material procurement collaborative management system that combines multi-dimensional data is also used to: perform digital signature verification on the transaction data package, and after the verification is passed, check whether there are empty values ​​or format type errors in the key fields of the data package, and generate an initial screening verification result; if the initial screening verification result is unsuccessful, the current process is terminated and an alarm log is generated.

[0069] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The cruise ship material procurement collaborative management method and specific examples combining multi-dimensional data in the aforementioned embodiment one are also applicable to the cruise ship material procurement collaborative management system combining multi-dimensional data in this embodiment. Through the foregoing detailed description of the cruise ship material procurement collaborative management method combining multi-dimensional data, those skilled in the art can clearly understand the cruise ship material procurement collaborative management system combining multi-dimensional data in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0070] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0071] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A collaborative management method for the entire process of cruise ship material procurement that combines multi-dimensional data, characterized in that: include: Obtain transaction data packages pushed by external procurement platforms, transform structured numerical fields and unstructured text description fields into semantic feature vectors, and generate timeliness impact factors; The semantic feature vector is input into the historical transaction vector library for similarity measurement. The status of the transaction data packet is determined by adjusting the judgment threshold in combination with the timeliness impact factor. The historical document index corresponding to the historical vector that is closest to the semantic feature vector is retrieved. When the transaction data packet is in a semantically ambiguous state, the retrieval is enhanced based on the semantic feature vector and the historical context corresponding to the historical document index, and a natural language change summary is generated. The corresponding supplier agreement file is generated by mapping the natural language change summary through the rule engine, and the semantic feature vector and similarity measurement results are embedded in the agreement file as hidden watermark information.

2. The cruise ship material procurement collaborative management method combining multi-dimensional data as described in claim 1, characterized in that, The semantic feature vector is input into the historical transaction vector database for similarity measurement. The status of the transaction data packet is determined by adjusting the judgment threshold based on the timeliness impact factor. The historical document index corresponding to the historical vector closest to the semantic feature vector is retrieved, including: The semantic feature vector is compared with each historical vector stored in the historical transaction vector library to perform a quantitative calculation of the spatial proximity, and the original value of the highest historical vector proximity is calculated. Based on the aforementioned timeliness impact factor, the preset standard discrimination interval is dynamically offset and corrected to generate a dynamic discrimination interval; The original value of the proximity of the highest historical vector is compared with the dynamic discrimination interval to mark the transaction data packets as highly similar, semantically ambiguous or significantly different. Based on the reverse index of the original value of the highest historical vector proximity, the historical documents associated with the corresponding historical vector are retrieved to obtain the historical document index.

3. The cruise ship material procurement collaborative management method combining multi-dimensional data as described in claim 2, characterized in that, The original value of the highest historical vector proximity is compared with the dynamic discrimination interval to mark the transaction data packets as highly similar, semantically ambiguous, or significantly different, including: If the original value of the proximity of the highest historical vector is higher than the upper limit of the dynamic discrimination interval, it is marked as a highly approximate state; If the original value of the proximity of the highest historical vector is between the upper and lower limits of the dynamic discrimination interval, it is marked as a semantically ambiguous state; If the original value of the proximity of the highest historical vector is lower than the lower limit of the dynamic discrimination interval, it is marked as a state of significant difference.

4. The cruise ship material procurement collaborative management method combining multi-dimensional data as described in claim 1, characterized in that, Generate timeliness impact factors, including: Extract the time of the transaction from the transaction data packet; The difference between the time of the business occurrence and the current time is calculated to obtain the original time interval value measured in a preset time unit. The time of occurrence of the service is mapped to one of multiple preset timeliness levels based on the original time interval value. The timeliness impact factor is calculated and generated by combining the preset basic attenuation coefficient corresponding to each timeliness level.

5. The cruise ship material procurement collaborative management method combining multi-dimensional data as described in claim 4, characterized in that, The multiple preset time-sensitive levels include at least a real-time transaction level, a recent retrospective level, and a long-term archiving level.

6. The cruise ship material procurement collaborative management method combining multi-dimensional data as described in claim 1, characterized in that, When the transaction data packet status is semantically ambiguous, retrieval enhancement is performed based on the semantic feature vector and the historical context corresponding to the historical document index to generate a natural language change summary, including: The semantic feature vector and the historical semantic vector corresponding to the historical document index are jointly decoded to extract the target fields, change values, change directions, and external influencing factors related to the current voyage business scenario that are changing between the two, thus obtaining the extracted content. The pre-trained retrieval enhancement generative model is invoked to combine the extracted content into descriptive text, which serves as the natural language change summary.

7. The cruise ship material procurement collaborative management method combining multi-dimensional data as described in claim 1, characterized in that, When the transaction data packet status is highly similar, the historical approval conclusions and agreement templates corresponding to the historical document index are directly extracted, and the transaction data packet is archived and stored as a time-sensitive version of the historical data in a lightweight manner.

8. The cruise ship material procurement collaborative management method combining multi-dimensional data as described in claim 1, characterized in that, When the transaction data packet status is "significant difference", it is determined that there is no valid historical reference for the transaction data packet. The transaction data packet is directly written into the master data table as a brand new business record, and a mandatory manual review task is generated and pushed to the approval terminal. After receiving the review approval instruction, it enters the rule engine mapping stage.

9. The cruise ship material procurement collaborative management method combining multi-dimensional data as described in claim 1, characterized in that, Before converting structured numerical fields and unstructured text description fields into semantic feature vectors, the following steps are also included: The transaction data packet is digitally signed and verified. After the verification is successful, the key fields in the data packet are checked one by one to see if there are any empty values ​​or format type errors, and the initial screening verification results are generated. If the initial screening verification result is unsuccessful, the current process will be terminated and an alarm log will be generated.

10. A cruise ship supplies procurement end-to-end collaborative management system integrating multi-dimensional data, characterized in that: The steps for implementing the cruise ship material procurement collaborative management method combining multidimensional data as described in any one of claims 1 to 9 include: The timeliness impact factor generation module is used to obtain transaction data packages pushed by external procurement platforms, transform structured numerical fields and unstructured text description fields into semantic feature vectors, and generate timeliness impact factors. The historical document index retrieval module is used to input the semantic feature vector into the historical transaction vector library for similarity measurement, combine the timeliness impact factor to determine the status of the transaction data packet by adjusting the judgment threshold, and retrieve the historical document index corresponding to the historical vector that is closest to the semantic feature vector. The Natural Language Change Summary Generation Module is used to generate a Natural Language Change Summary by performing retrieval enhancement based on the semantic feature vector and the historical context corresponding to the historical document index when the transaction data packet status is semantically ambiguous. The supplier agreement file generation module is used to generate a corresponding supplier agreement file by mapping the natural language change summary through a rule engine, and to embed the semantic feature vector and similarity measurement results as hidden watermark information into the agreement file.

Images