A large model-based intelligent field matching method, system and storage medium

Abstract

The application provides an intelligent field matching method and system based on a large model. Based on multi-dimensional medical data, features for similarity matching are extracted, the multi-dimensional medical data is divided into dictionary data and field data, and a quantitative index is calculated. The quantitative index includes field completeness rate, dictionary completeness rate, Java similarity matching rate, large model similarity matching rate, field complete equality rate, dictionary complete equality rate, Java runtime length influence rate, and large model runtime length influence rate. The quantitative index is weighted and scored to obtain a comprehensive score result. According to the comprehensive score result, a similarity matching method is selected, a task is split according to the smallest executable granularity and submitted to a forkjoin parallel framework for field matching. According to the quantitative index and a smoothing factor, the index weight is dynamically adjusted and normalized. The application has strong adaptability and high scalability, can significantly improve the field matching efficiency, and ensures the flexibility of the system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer technology and data processing technology, in particular to an intelligent field matching method and system based on a large model and a storage medium. BACKGROUND

[0002] With the rapid development of information technology, the amount of medical data is growing explosively, and the complexity of data governance and application is also increasing. Field matching, as a key step in the process of data integration, cleaning and analysis, plays a very important role in data governance. Traditional field matching methods rely on manual operation, which not only consumes a lot of time and human resources, but also cannot guarantee the accuracy and consistency of matching. Today, data-driven decision-making is becoming the core competence of hospitals, and this problem is particularly prominent.

[0003] Traditional field matching methods, such as Figure 1 As shown in the figure, for example, the java text similarity matching algorithm extracts the feature fields of the standard data field table, and establishes an index library by performing word segmentation on the feature fields. Then, the query object is obtained by performing word segmentation on the field to be matched. The matching result is obtained according to the index library and the query object. Based on the idea of inverted index, the standard field is automatically matched to achieve high matching accuracy with less cost.

[0004] Although the above method has a relatively fast response speed and relatively small resource cost, it lacks semantic understanding and can only achieve similarity matching through simple sentence segmentation, index creation and full-text retrieval. For example, "code" and "id" cannot be matched, and the text similarity is zero. However, under normal circumstances, the fields of the business library should be successfully matched with the data elements of the national standard data set, which requires a large amount of manual field matching and a tedious and inefficient process. In addition, the above method lacks growth ability and can only perform similarity matching based on fixed algorithms.

[0005] Therefore, in order to solve the above problems, the present application proposes an intelligent field matching method and system based on a large model and a storage medium. SUMMARY

[0006] The present application proposes the following technical solutions to address one or more technical defects in the prior art.

[0007] Based on the first aspect of the present application, an intelligent field matching method based on a large model is proposed, comprising:

[0008] S1: extracting features for similarity matching based on multi-dimensional medical data, dividing the multi-dimensional medical data into dictionary data and field data, and calculating a quantitative index;

[0009] The quantification indicators include a field completeness rate, a dictionary completeness rate, a Java similarity matching rate, a large model similarity matching rate, a field complete equality rate, a dictionary complete equality rate, a Java runtime length influence rate, and a large model runtime length influence rate.

[0010] S2: Perform weighted score calculation on the quantification indicators to obtain a comprehensive score result.

[0011] Score= ;

[0012] Score represents the comprehensive score result, represents a field completeness rate weight, represents a dictionary completeness rate weight, represents a Java similarity matching rate weight, represents a large model similarity matching rate weight, represents a field complete equality rate weight, represents a dictionary complete equality rate weight, represents a Java runtime length influence rate weight, represents a large model runtime length influence rate weight, represents a field completeness rate score, represents a dictionary completeness rate score, represents a Java similarity matching rate score, represents a large model similarity matching rate score, represents a field complete equality rate score, represents a dictionary complete equality rate score, represents a Java runtime length influence rate score, represents a large model runtime length influence rate score.

[0013] S3: Select a similarity matching method according to the comprehensive score result, split a task according to a minimum executable granularity, and submit to a forkjoin parallel framework for field matching.

[0014] S4: Dynamically adjust indicator weights according to the quantification indicators and a smoothing factor, and perform normalization processing on the indicator weights.

[0015] The present application can adaptively perform field matching through a dynamic weight adjustment mechanism and task parallel processing, realizes fine and multi-angle evaluation of field matching, not only considers matching accuracy, but also introduces a running efficiency factor, so that the selection of the field matching strategy is more intelligent and comprehensive, and significantly improves the automation level and comprehensive performance of field matching in a big data environment.

[0016] Further, the multi-dimensional medical data in step S1 includes table information of the data source, field information of the data source table, and country specification information;

[0017] The table information of the data source includes table English name and table note;

[0018] The field information of the data source table includes field English name, field note, and field data type;

[0019] The country specification information includes standard data set English name, data set Chinese name, data element English code, data element Chinese name, and data element data type;

[0020] The dictionary data is extracted for features: dictionary name, dictionary code, dictionary rule name, and dictionary rule code;

[0021] The field data is extracted for features: table name, table note, field name, field note, and field data type.

[0022] This step enhances the adaptability of the system to a complex data environment, improves the pertinence and effectiveness of feature extraction, and provides reliable data support for subsequent similarity matching.

[0023] Further, the calculation formula of the field completeness rate is:

[0024] ZDuan=ZDx / ZDall;

[0025] ZDuan represents the field completeness rate, including five field dimensions of table name, table note, field English name, field note, and field data type, ZDx represents the number of fields with values, and ZDall represents the total number of fields;

[0026] The calculation formula of the dictionary completeness rate is:

[0027] ZDian=ZDy / ZDiall;

[0028] ZDian represents the dictionary completeness rate, including four dictionary dimensions of dictionary name, dictionary code, dictionary rule name, and dictionary rule code, ZDy represents the number of dictionary items with values, and ZDiall represents the total number of dictionary items;

[0029] The calculation formula of the Java similarity matching rate is:

[0030] PPjava=mz / bzdall;

[0031] PPjava represents the Java algorithm matching rate, mz represents the number of matching hits, and bzdall represents the total number of table fields;

[0032] The calculation formula of the large model similarity matching rate is:

[0033] PPmode = mz / bzdall

[0034] Wherein, PPmode represents the large model matching rate, mz represents the number of matching hit fields, and bzdall represents the total number of table fields.

[0035] The quantifiable index evaluation of this step can quickly identify consistent fields and dictionary items, can objectively measure data quality and matching effect, and improves the explainability and controllability of the present application.

[0036] Further, the calculation formula of the field complete equality rate is:

[0037] AEduan = duanae / bzdall

[0038] Wherein, AEduan represents the field complete equality rate, duanae represents the number of completely equal fields, and bzdall represents the total number of table fields.

[0039] The calculation formula of the dictionary complete equality rate is:

[0040] AEdian = dianae / zdall

[0041] Wherein, AEdian represents the dictionary complete equality rate, dianae represents the number of completely equal dictionaries, and zdall represents the total number of dictionaries.

[0042] This step avoids unnecessary similarity calculation, saves computing resources, improves matching efficiency, and can be applied to fast matching in high consistency scenarios.

[0043] Further, the calculation formula of the Java runtime length influence rate is:

[0044] TimeJava = Min(100%, javatime / maxtime*100%)

[0045] Wherein, TimeJava represents the Java runtime length influence rate, javatime represents the actual Java runtime length, and maxtime represents the maximum allowed length.

[0046] The calculation formula of the large model runtime length influence rate is:

[0047] Timemode = Min(100%, modetime / maxtime*100%)

[0048] Wherein, Timemode represents the large model running time length influence rate, modetime represents the actual running time length of the large model, and maxtime represents the maximum allowed time length.

[0049] This step reflects the balance between field matching efficiency and accuracy, can reduce the influence of running time delay on matching results due to excessive pursuit of matching accuracy, can guarantee the system to match tasks within a limited time, and enhance the practicability and reliability of the system.

[0050] Further, according to the comprehensive score result, a similarity matching method is selected, specifically, if the score is greater than a preset highest threshold, a large model is selected for similarity matching;

[0051] If the comprehensive score result is less than a preset lowest threshold, a java algorithm is used for similarity matching;

[0052] Otherwise, a hybrid mode is used for similarity matching, a java algorithm is used for similarity matching of dictionary data, and a large model is used for similarity matching of fields.

[0053] The application dynamically selects a matching strategy through a comprehensive score result, guarantees field matching accuracy, and significantly improves the overall performance of the system.

[0054] Further, the update formula of the dynamically adjusted index weight is:

[0055] ;

[0056] Wherein, The smooth value of the change rate is calculated using the exponential weighted moving average (EWMA), Indicates the initial weight, Indicates the updated weight, f indicates the proportion factor in the weight adjustment formula, f∈(0.1,0.3], used to control the sensitivity of weight change, and the influence degree of the change rate on the maximum weight adjustment.

[0057] Further, the calculation formula of the smooth value is:

[0058] ;

[0059] Wherein, Indicates the current smooth value, Indicates the smooth value of the last moment, Indicates the original change rate at the current moment, and a indicates the smoothing coefficient, used to control the influence degree of the current change rate on the smooth value, a∈(0.1,0.3].

[0060] The step responds to real-time changes of system operation by accurately capturing and smoothing the weight trend, suppresses noise interference, ensures stability and reliability of weight adjustment, enhances robustness and adaptability of the system, and can continuously optimize according to actual conditions.

[0061] Based on the second aspect of the application, an intelligent field matching system based on a large model is also proposed, comprising:

[0062] The feature index module extracts features for similarity matching based on multi-dimensional medical data, divides the multi-dimensional medical data into dictionary data and field data, and calculates quantitative indexes.

[0063] The quantitative indexes include field completeness rate, dictionary completeness rate, Java similarity matching rate, large model similarity matching rate, field complete equality rate, dictionary complete equality rate, Java runtime length influence rate, and large model runtime length influence rate.

[0064] The comprehensive score module calculates the weighted score of the quantitative indexes to obtain a comprehensive score result.

[0065] Score= ;

[0066] Score represents the comprehensive score result, represents the field completeness rate weight, represents the dictionary completeness rate weight, represents the Java similarity matching rate weight, represents the large model similarity matching rate weight, represents the field complete equality rate weight, represents the dictionary complete equality rate weight, represents the Java runtime length influence rate weight, represents the large model runtime length influence rate weight, represents the field completeness rate score, represents the dictionary completeness rate score, represents the Java similarity matching rate score, represents the large model similarity matching rate score, represents the field complete equality rate score, represents the dictionary complete equality rate score, represents the Java runtime length influence rate score, represents the large model runtime length influence rate score.

[0067] Divide and conquer module: according to the comprehensive score result, select a similarity matching method, divide the multi-dimensional medical data into dictionary data and field data, split tasks according to the minimum executable granularity and submit to the forkjoin parallel framework for field matching;

[0068] Optimization module: dynamically adjust the index weight according to the quantitative index and smoothing factor, and normalize the index weight.

[0069] Based on the third aspect of the present application, a computer program product is also proposed, which has one or more computer programs thereon, when the computer programs are executed by a computer processor, the method as claimed in any one of the above is implemented.

[0070] The technical effect of the present application is that the present application optimizes the field similarity matching process in the medical field data standard sorting and mapping scene, especially the field matching between the national standard data set and multiple heterogeneous data sources, which can effectively process structured and unstructured data across different data sources, has strong adaptive ability, can automatically adjust the matching strategy, ensures the flexibility and scalability of the system, adapts to changes in data environment and technology upgrade, significantly improves the efficiency of generating database statements for data synchronization task in medical data governance, provides strong support for data-driven competition, not only improves the quality of medical services, but also promotes in-depth analysis and research of medical data, and promotes the innovation and development of the medical field. BRIEF DESCRIPTION OF DRAWINGS

[0071] Other features, objects and advantages of the present application will become more apparent through reading the detailed description of the non-limiting embodiments made with reference to the following drawings.

[0072] Figure 1 is a Java similarity matching flowchart provided by the prior art.

[0073] Figure 2 is a general flowchart of an intelligent field matching method based on a large model according to an embodiment of the present application.

[0074] Figure 3 is a module diagram of an intelligent field matching system based on a large model according to an embodiment of the present application.

[0075] Figure 4 is a flowchart of text similarity matching by an intelligent field matching system based on a large model according to an embodiment of the present application.

[0076] Figure 5 is an index result calculation flowchart according to an embodiment of the present application.

[0077] Figure 6A dynamic decision calling strategy flowchart is provided according to an embodiment of the present application.

[0078] Figure 7 A to-be-matched data graph for constructing a prompt is provided according to an embodiment of the present application.

[0079] Figure 8 A prompt CARE framework graph is constructed according to an embodiment of the present application.

[0080] Figure 9 A prompt example and construction process graph are provided according to an embodiment of the present application.

[0081] Figure 10 A data integration platform and large model interaction flowchart is provided according to an embodiment of the present application.

[0082] Figure 11 A patient address information uses a large model similarity matching result graph is provided according to an embodiment of the present application.

[0083] Figure 12 A structural schematic diagram of a computer system of an electronic device suitable for implementing an embodiment of the present application. DETAILED DESCRIPTION

[0084] The present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely intended to explain the related application, and not to limit the application. In addition, it should be noted that only parts related to the application are shown in the drawings for ease of description.

[0085] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and embodiments.

[0086] Figure 2 An intelligent field matching method based on a large model is shown, which comprises:

[0087] S1: extracting features for similarity matching based on multi-dimensional medical data, dividing the multi-dimensional medical data into dictionary data and field data, and calculating a quantitative index;

[0088] The quantitative index includes field completeness rate, dictionary completeness rate, Java similarity matching rate, large model similarity matching rate, field complete equality rate, dictionary complete equality rate, Java runtime length influence rate, and large model runtime length influence rate;

[0089] S2: weighted score calculation is performed on the quantitative index to obtain a comprehensive score result;

[0090] Score= ;

[0091] Score represents the overall score result. Indicates the field completeness weight. The weight represents the dictionary completeness rate. This represents the weight of the Java similarity matching rate. This represents the weight of the large model similarity matching rate. Indicates the weight of the field's complete equality rate. The weights represent the percentage of words that are completely identical. This indicates the weight of the Java runtime impact rate. Indicates the weights of the impact rate during the runtime of the large model. This indicates the field completeness score. This indicates the dictionary completeness score. This represents the Java similarity matching rate score. This represents the large model similarity matching rate score. This indicates the score for the complete equality rate of the field. This represents the score indicating the complete equality rate of the dictionary entries. This represents the Java runtime impact score. This represents the impact rate score of the large model runtime;

[0092] S3: Select a similarity matching method based on the comprehensive scoring results, split the task into the smallest executable granularity, and submit it to the forkjoin parallel framework for field matching;

[0093] S4: Dynamically adjust the indicator weights based on the quantification indicators and smoothing factors, and normalize the indicator weights.

[0094] It should be noted that this application, through a dynamic weight adjustment mechanism and parallel task processing, can adaptively perform field matching, achieving refined and multi-faceted evaluation of field matching. It not only considers matching accuracy but also introduces operational efficiency factors, making the selection of field matching strategies more intelligent and comprehensive, and significantly improving the automation level and overall performance of field matching in a big data environment.

[0095] It should be noted that the multi-dimensional medical data mentioned in step S1 includes table information of the data source, field information of the data source table, and national standard information;

[0096] The table information of the data source includes the English name of the table and table remarks;

[0097] The field information of the data source table includes the English name of the field, field remarks, and field data type;

[0098] The national standard information includes the English name of the standard dataset, the Chinese name of the dataset, the English code of the data element, the Chinese name of the data element, and the data element metadata type;

[0099] Extract features from dictionary data: dictionary name, dictionary code, dictionary rule name, and dictionary rule code;

[0100] Extract features from field data: table name, table comments, field names, field comments, and field data types.

[0101] It should be noted that dimensional feature analysis is performed on the standard value range and the data source dictionary. Dictionary data is compared for similarity using four dimensions: dictionary name, dictionary code, dictionary item name, and dictionary item code. The one-dimensional standard value range name corresponds to the dictionary name of the data source dictionary; the two-dimensional standard value range code corresponds to the dictionary code of the data source dictionary; the three-dimensional standard value range item name corresponds to the dictionary item name of the data source dictionary; and the four-dimensional standard value range item code corresponds to the dictionary item code of the data source dictionary. These four-dimensional features improve the accuracy of dictionary similarity matching.

[0102] The standard dataset and source table structure are analyzed using dimensional features. Field-type data are compared for similarity using five dimensions: table name, table comments, field name, field comments, and field data type. In the one-dimensional model, the dataset name of the standard dataset corresponds to the table name of the source table structure; in the two-dimensional model, the dataset code of the standard dataset corresponds to the table comments of the source table structure; in the three-dimensional model, the data element name of the standard dataset corresponds to the field comments of the source table structure; in the four-dimensional model, the data element code of the standard dataset corresponds to the field name of the source table structure; and in the five-dimensional model, the data element type of the standard dataset corresponds to the field data type of the source table structure. These five dimensions improve the accuracy of field similarity matching.

[0103] It should be noted that this step enhances the system's adaptability to complex data environments, improves the targeting and effectiveness of feature extraction, and provides reliable data support for subsequent similarity matching.

[0104] It should be noted that the formula for calculating the field completeness rate is as follows:

[0105] ZDuan = ZDx / ZDall;

[0106] ZDuan represents the field completeness rate, which includes five field dimensions: table name, table comments, field English name, field comments, and field data type; ZDx represents the number of values ​​in a field; and ZDall represents the total number of fields.

[0107] The formula for calculating the dictionary completeness rate is as follows:

[0108] ZDian = ZDy / ZDiall;

[0109] ZDian represents the dictionary completeness rate, which includes four dictionary dimensions: dictionary name, dictionary code, dictionary rule name, and dictionary rule code; ZDy represents the number of dictionary entries with values; and ZDiall represents the total number of dictionary entries.

[0110] The formula for calculating the Java similarity matching rate is as follows:

[0111] PPjava=mz / bzdall;

[0112] Where PPjava represents the Java algorithm matching rate, mz represents the number of fields matched, and bzdall represents the total number of table fields;

[0113] The formula for calculating the large model similarity matching rate is as follows:

[0114] PPmode=mz / bzdall;

[0115] Where PPmode represents the large model matching rate, mz represents the number of fields matched, and bzdall represents the total number of table fields.

[0116] It should be noted that the quantifiable metrics used in this step can quickly identify consistent fields and dictionary entries, objectively measure data quality and matching effectiveness, and improve the interpretability and controllability of this application.

[0117] It should be noted that the field completeness rate represents the data completeness of the five dimensions of the field. The higher the completeness of the field dimensions, the better the large model can utilize contextual understanding for semantic matching. The dictionary completeness rate represents the data completeness of the four dimensions of the dictionary. When the dictionary-level dimensions are missing, the Java algorithm is more stable for matching.

[0118] The Java similarity matching rate is the similarity score calculated by the Java algorithm in the matching process. It reflects the initial matching effect of the Java algorithm and is used as a reference. Using a large model for similarity matching is often preferred over the Java algorithm, especially when the matching rate is high.

[0119] It should be noted that the formula for calculating the complete equality rate of the fields is as follows:

[0120] AEduan=duanae / bzdall;

[0121] Where AEDuan represents the field equality rate, duanae represents the number of fields that are completely equal, and bzdall represents the total number of fields in the table.

[0122] The formula for calculating the complete equality rate of the dictionaries is:

[0123] AEdian = dianae / zdall;

[0124] Where AEdian represents the percentage of dictionaries that are completely identical, dianae represents the number of dictionaries that are completely identical, and zdall represents the total number of dictionaries.

[0125] It should be noted that the field complete equality rate represents the proportion of fields in table A and table B that are completely identical, while the dictionary complete equality rate refers to the proportion of entries in dictionary C that are completely identical to those in field D. When fields in table A and table B are completely identical, or when entries in dictionary C and field D are completely identical, the Java algorithm is sufficient to handle similarity matching.

[0126] It should be noted that this step avoids unnecessary similarity calculations, saves computing resources, improves similarity matching efficiency, and is suitable for fast matching in high consistency scenarios.

[0127] It should be noted that the formula for calculating the Java runtime impact rate is as follows:

[0128] TimeJava=Min(100%,javatime / maxtime×100%);

[0129] Among them, TimeJava represents the Java runtime impact rate, javatime represents the actual Java runtime, and maxtime represents the maximum allowed duration.

[0130] The formula for calculating the runtime impact rate of the large model is as follows:

[0131] Timemode=Min(100%,modetime / maxtime×100%);

[0132] Wherein, Timemode represents the impact rate of large model runtime, modetime represents the actual runtime of large model, and maxtime represents the maximum allowed runtime.

[0133] It should be noted that the Java runtime impact rate refers to the time required for the Java algorithm to process single-table comparisons, while the large model runtime impact rate refers to the time required for the large model to process single-table comparisons. The shorter the Java runtime, the more likely the Java algorithm will be chosen for similarity matching, while the longer the large model runtime, the more likely the large model will be called.

[0134] It should be noted that this step reflects a balance between field matching efficiency and accuracy. It can reduce runtime latency caused by excessive pursuit of matching accuracy, ensure that the system completes the matching task within a limited time, and enhance the system's practicality and reliability.

[0135] It should be noted that the similarity matching method is selected based on the comprehensive scoring result. Specifically, if the score is greater than the preset maximum threshold, a large model is selected for similarity matching.

[0136] If the overall score is less than the preset minimum threshold, a Java algorithm is used for similarity matching.

[0137] Otherwise, a hybrid approach is used for similarity matching: Java algorithms are used for similarity matching of dictionary data, while large models are used for similarity matching of fields.

[0138] It should be noted that this application dynamically selects the matching strategy based on the comprehensive scoring results, which not only ensures the accuracy of field matching but also significantly improves the overall performance of the system.

[0139] It should be noted that the update formula for the dynamically adjusted indicator weights is as follows:

[0140] ;

[0141] in, The smoothed value representing the rate of change is calculated using the exponentially weighted moving average (EWMA). Indicates the initial weights. The updated weight is represented by f, which represents the scaling factor in the weight adjustment formula. f∈(0.1,0.3] is used to control the sensitivity of the weight change and the degree of influence of the rate of change on the final weight adjustment.

[0142] It should be noted that the formula for calculating the smoothing value is:

[0143] ;

[0144] in, This represents the current smoothing value. This represents the smoothed value from the previous time step. The original rate of change at the current moment is represented by α, which represents the smoothing coefficient and is used to control the degree of influence of the current rate of change on the smoothed value. α ∈ (0.1, 0.3).

[0145] It should be noted that this step, through precise capture and smoothing of weight trends, not only responds to real-time changes in system operation but also suppresses noise interference, ensuring the stability and reliability of weight adjustments, enhancing the robustness and adaptability of the system, and enabling continuous optimization based on actual conditions.

[0146] The following is for reference. Figure 3 , Figure 3A smart field matching system based on a large model is also proposed, including feature index module a, comprehensive scoring module b, divide-and-conquer module c, and optimization module d.

[0147] In a specific practical example, the feature index module a is configured to: extract features for similarity matching based on multi-dimensional medical data, divide the multi-dimensional medical data into dictionary data and field data, and calculate quantitative indicators;

[0148] The quantitative indicators include field completeness rate, dictionary completeness rate, Java similarity matching rate, large model similarity matching rate, field exact equality rate, dictionary exact equality rate, Java runtime impact rate, and large model runtime impact rate.

[0149] In a specific practical example, the comprehensive scoring module b is configured to: perform weighted scoring calculation on the quantitative indicators to obtain a comprehensive scoring result;

[0150] Score= ;

[0151] Score represents the overall score result. Indicates the field completeness weight. The weight represents the dictionary completeness rate. This represents the weight of the Java similarity matching rate. This represents the weight of the large model similarity matching rate. Indicates the weight of the field's complete equality rate. The weights represent the percentage of words that are completely identical. This indicates the weight of the Java runtime impact rate. Indicates the weights of the impact rate during the runtime of the large model. This indicates the field completeness score. This indicates the dictionary completeness score. This represents the Java similarity matching rate score. This represents the large model similarity matching rate score. This indicates the score for the complete equality rate of the field. This represents the score indicating the complete equality rate of the dictionary entries. This represents the Java runtime impact score. This represents the impact rate score of the large model runtime.

[0152] In a specific practical example, the divide-and-conquer module c is configured to: select a similarity matching method based on the comprehensive scoring result, divide the multi-dimensional medical data into dictionary data and field data, split the task according to the smallest executable granularity, and submit it to the forkjoin parallel framework for field matching.

[0153] In a specific practical example, the optimization module d is configured to: dynamically adjust the index weights according to the quantification index and the smoothing factor, and normalize the index weights.

[0154] It should be noted that the flowchart for text similarity matching in the above system is as follows: Figure 4 As shown:

[0155] S41: Collect data to obtain table information and table structure field information of the data source;

[0156] Obtain national standards information;

[0157] The table information includes the English name of the table and table remarks; the field information includes the English name of the field, field remarks and field data type; and the national standard information includes the English name of the nominal dataset, the Chinese name of the dataset, the English code of the data element, the Chinese name of the data element and the data element data type.

[0158] S42 Indicator Result Calculation: Calculate the corresponding indicators based on the multi-dimensional data to be matched for similarity established in the previous step. The indicators include: field completeness rate weight, dictionary completeness rate weight, etc.

[0159] S43 Indicator Result Scoring: The score is calculated based on the indicator scoring model and the result values ​​of each indicator obtained in the previous step.

[0160] S44 Comprehensive Score Calculation: Based on the scores of each indicator obtained in the previous step, the dynamic weight values ​​of each indicator are used to calculate the final comprehensive score.

[0161] S45 Decision Matching Strategy: Based on the comprehensive score, decide whether to use the traditional Java algorithm or a large model for similarity matching;

[0162] S46 execution: Based on the data provided to the user by the quantile algorithm, the data to be matched for similarity is split into two parts: a dictionary and fields. The task is then divided and submitted to the forkjoin parallel computing framework.

[0163] S47 Dynamically Adjust Weight Values: After each task execution, the indicator value is re-evaluated, and the indicator weight value is dynamically adjusted based on the indicator value and smoothing factor. Finally, the weight value is normalized to ensure that the sum of all weights equals 1.

[0164] S48 Result Merging: In the final stage, all task execution result data are merged and summarized.

[0165] like Figure 5As shown, the dimensions of medical data analysis include single-table Java similarity matching (S51), single-table large-scale model similarity matching (S52), field content complete equality (S53), dictionary content complete equality rate (S54), calculation of dictionary four-dimensional data integrity (S55), and calculation of field five-dimensional data integrity (S56).

[0166] Single-table Java similarity matching S51 includes calculating the Java similarity matching rate S511 and the Java similarity runtime impact rate S512. Single-table large-model similarity matching S52 includes calculating the large-model similarity runtime impact rate S521 and the large-model similarity matching rate S522 to obtain a comprehensive score result. Then, based on the comprehensive score result, the traditional Java matching algorithm or the large model is selected for similarity matching.

[0167] It should be noted that, based on the divide-and-conquer approach, the data to be matched for similarity is split into fields and dictionaries. Then, the data to be matched in each part is further divided into tasks according to the smallest executable granularity and submitted to the forkjoin parallel framework for field matching. After each task is executed, the quantitative indicators are re-evaluated, and the weight values ​​of the quantitative indicators are dynamically adjusted based on the quantitative indicator values ​​and smoothing factors. The weight values ​​are normalized to ensure that the sum of all weights is always equal to 1. Finally, all execution results are merged and summarized.

[0168] In a specific implementation, the scoring range for each quantitative indicator is [20, 100], and the weight of the quantitative indicator is... The sum of these values ​​is 1, and the final comprehensive score range is... The scoring strategy is shown in Table 1. Assume the weights of the basic quantitative indicators are:

[0169] ;

[0170] Based on the comprehensive scoring results, a similarity matching method is selected, and a dynamic decision-making strategy is employed as follows: Figure 6 As shown, if the score is greater than the preset maximum threshold of 80, then the large model is selected for similarity matching S61, because the matching advantage of the large model is more significant when there is high-dimensional data integrity and high matching rate.

[0171] If the overall score is less than 50, the Java algorithm is used for similarity matching S62, because the Java algorithm is more efficient when low-dimensional completeness or complete equality occurs.

[0172] Otherwise, a hybrid mode for similarity matching (S63) can be used, employing a Java algorithm for similarity matching on dictionary data and a large model for similarity matching on fields, which can balance matching efficiency and matching accuracy.

[0173] Table 1 Scoring Strategy Table

[0174] ;

[0175] In a specific embodiment, the system collects multi-dimensional medical data from the hospital information system, including:

[0176] Data source table information {English name is PAT_MASTER, table remarks are patient master record table};

[0177] Field information of the data source table {Field English name is PATI_NAME, field remarks are patient name, field data type is VARCHAR;

[0178] Field English name is GENDER, field remarks are gender, field data type is CHAR;

[0179] Field English name is BIRTH_DATE, field remarks are date of birth, field data type is DATE;};

[0180] Data elements of national standard information {English code is PT_NAME, Chinese name is patient name, data type is string;

[0181] English code is GENDER, Chinese name is gender, data type is code;

[0182] English code is BIRTH_DATE, Chinese name is date of birth, data type is date;};

[0183] Calculate the following quantitative indicators:

[0184] Check the filling status of the five characteristic dimensions (table name, table remarks, field name, field remarks, field data type) of all fields in the PAT_MASTER table. Assume that this table has 20 fields and a total of 100 characteristic dimension values, among which 85 dimension values have values. The field completeness rate = 85 / 100 = 85%;

[0185] Check the four dimensions of the dictionary. Assume that there are 2 dictionary sub-items in the national standard value range, a total of 8 characteristic values, and all have values; the source dictionary has 2 dictionary sub-items, but the "dictionary code" dimension is missing, and a total of 7 characteristic values have values. The dictionary completeness rate = (8 + 7) / (8 + 8) = 15 / 16 ≈ 94%;

[0186] Use the traditional Java algorithm for preliminary matching. Assume that 5 fields are matched. Then the Java similarity matching rate = 5 / 20 = 25%;

[0187] Using a large model for matching, assuming the match hits 5 fields, the large model similarity match rate = 5 / 20 = 25%;

[0188] If the English name of the field GENDER is completely identical to the country code GENDER, and there are 3 such fields, then the field complete equality rate is 3 / 20 = 15%.

[0189] The (1, male) dictionary in the data source is not exactly the same as the standard (M, male). The number of dictionaries that are exactly the same is 0. Therefore, the dictionary equality rate is 0 / 2 = 0%.

[0190] The Java algorithm takes 3 seconds to process this table. The maximum allowed time (maxtime) set by the system is 10 seconds. Therefore, the impact of Java runtime is Min(100%, 3 / 10×100%) = 30%.

[0191] If the total processing time for the large model (including API calls) is 1 second, then the impact rate of the large model's runtime = Min(100%, 1 / 10×100%) = 10%;

[0192] According to Table 1, the quantitative indicators were scored as follows: field completeness rate score was 100, dictionary completeness rate score was 100, Java similarity matching rate score was 40, large model similarity matching rate score was 40, field complete equality rate score was 20, dictionary complete equality rate score was 20, Java runtime impact rate score was 40, and large model runtime impact rate score was 20.

[0193] The overall score is 20%×100+15%×100+15%×40+20%×40+10%×20+10%×20-5%×40-5%×20=50; which is equal to the preset minimum threshold of 50 points. A hybrid mode is used for similarity matching, with Java algorithm used for similarity matching of dictionary data and large model used for similarity matching of field data.

[0194] It should be noted that the data to be matched when constructing the prompt is as follows: Figure 7 As shown, when selecting a large model to perform similarity matching, the CARE framework is first used to construct the prompt word text, such as... Figure 8 As shown, the CARE framework includes a Context that provides background information, an Action that describes the large model's execution actions, a Result that specifies the model's output format and requirements, and an Example.

[0195] In specific embodiments, such as Figure 9As shown, the task context is set as Context{table name: BASEINFO_ADDRESS, table remarks: Personal Basic Information - Address Information Table, field English name: Addr_City, field remarks: Address - City (District)}, and the target data field context is {table name: ZEMRPATIENTADDRESS_INFO, table remarks: Patient Address Information Table, field English name: PATIENT_ID, field remarks: Patient ID……}. A data governance expert role is assigned to the large model, and the task is to compare the fields of two medical data sources, Table 1 and Table 2, and perform similarity comparisons based on the table field dimensions. Only successfully matched results are returned in the specified JSON format, and an example is provided: Example{"table1: Table 1 fields": "table2: Table 2 fields", "Similarity": The similarity score is output, giving the similarity matching score between the two fields. The specific return result is {"Addr_City":'CITY',"Addr_County':'DISTRICT',"Addr_House_No':'DETAILED_ADDRESS',"Addr_Province":'PROVINCE', ...};

[0196] In the integrated data integration platform 103, the source data and the data to be matched are input, and the large model 102 is selected to perform similarity matching to construct the prompt word text, as shown below. Figure 10 As shown, Python program 101 requests the large model 102 to package the prompt text into a request body, and transmits the request body to the large model 102 via the Open API protocol. Simultaneously, the field matching results of the large model 102 are cleaned, converted into JSON, and returned to the application for display on the integrated data platform 103. Patient address information is matched using the similarity results from the large model, as shown below. Figure 11 As shown, it can perform most text similarity matching, as well as semantic similarity matching at the meaning level through semantic understanding.

[0197] In a specific embodiment, the weight values ​​of each quantitative indicator are dynamically adjusted based on the quantitative indicators and the smoothing factor, and the original rate of change at the current time is... The value is 0.4, the smoothed value from the previous time step. The smoothing coefficient is 0.3. If the value is 0.2, then the smoothing value at this moment is... =0.2×0.4+0.8×0.3=0.32, initial weight If the value is 0.5 and the scaling factor f is 0.15, then the updated weights... The value is 0.5 × (1 - 0.15 × 0.32) = 0.476. By adjusting the weight of the quantitative indicators according to different data scenarios, the matching strategy can be automatically adjusted according to the specific domain dataset. The threshold for automatic adjustment of the quantitative weight is 50%.

[0198] It should be noted that if a significant amount of field dimension information is missing, and the field completeness rate is less than 50%, then the field completeness will be reduced based on a smoothing factor. The weights;

[0199] If a significant amount of dictionary dimension information is missing, resulting in a dictionary completeness rate of less than 50%, then the dictionary completeness is reduced based on a smoothing factor. The weights;

[0200] If the matching accuracy of the Java similarity algorithm is unstable, with a Java similarity matching rate of less than 50%, then a smoothing factor can be used to improve the Java similarity matching rate. The weights;

[0201] If the matching accuracy of the large model is unstable and the similarity matching rate of the large model is less than 50%, then the similarity matching rate of the large model should be reduced based on a smoothing factor. The weights;

[0202] If the field equality rate is low, with a matching rate of less than 50%, then the field equality rate will be reduced based on a smoothing factor. The weights;

[0203] If the dictionary equality rate is low, with a matching rate of less than 50%, then the dictionary equality rate is reduced based on a smoothing factor. The weights;

[0204] If Java runtime has a high impact, then reduce the impact rate of Java runtime based on a smoothing factor. The weights;

[0205] If the runtime of a large model has a significant impact, then the impact rate of the runtime of a large model can be increased based on a smoothing factor. The weights;

[0206] The adjusted weights are normalized according to the following formula, and the total adjusted weights are still 1.

[0207] The normalization formula is:

[0208] ;

[0209] .

[0210] It should be noted that this application, through a dynamic adjustment of weights using a smoothing factor and a scoring decision mechanism based on real-time calculation, adaptively adjusts the algorithm selection strategy for different data scenarios, effectively improving the flexibility and adaptability of algorithm selection. It unifies the scoring mechanism and weight normalization method for different dimensions, quantitative indicators, and weights, mapping complex multi-dimensional data into quantifiable comprehensive scores, thus making the complex similarity analysis system more scalable and operable. The dynamic algorithm selection mechanism based on comprehensive score thresholds balances accuracy and computational efficiency, enabling rapid response in different operating scenarios. It covers the entire process from quantitative modeling of data features to final algorithm selection in medical data standardization and mapping scenarios, comprehensively covering the technological innovation of the entire similarity matching system and providing complete technical support for the dynamic collaborative work of large models and traditional algorithms.

[0211] It should be noted that this application has demonstrated significant application value in the field of medical big data, achieving accurate field matching between national standard datasets such as public health and outpatient clinics and multiple heterogeneous data sources, greatly improving the efficiency and accuracy of data integration, providing strong data support for medical decision-making, being able to understand the meaning of fields at the semantic level to achieve more accurate matching, and also being able to combine domain knowledge and expert experience to build a knowledge base for auxiliary matching.

[0212] The following is for reference. Figure 12 It shows a schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application. Figure 12 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0213] like Figure 12 As shown, the computer system includes a central processing unit (CPU) 1201, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 1202 or programs loaded from storage section 1208 into random access memory (RAM) 1203. The RAM 1203 also stores various programs and data required for system operation. The CPU 1201, ROM 1202, and RAM 1203 are interconnected via bus 1204. An input / output (I / O) interface 1205 is also connected to bus 1204.

[0214] The following components are connected to I / O interface 1205: an input section 1206 including a keyboard, mouse, etc.; an output section 1207 including a liquid crystal display (LCD) and speakers, etc.; a storage section 1208 including a hard disk, etc.; and a communication section 1209 including a network interface card such as a LAN card and a modem, etc. The communication section 1209 performs communication processing via a network such as the Internet. A drive 1210 is also connected to I / O interface 1205 as needed. A removable medium 1211, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 1210 as needed so that computer programs read from it can be installed into storage section 1208 as needed.

[0215] Specifically, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 1209, and / or installed from removable medium 1211. When the computer program is executed by central processing unit (CPU) 1201, it performs the functions defined in the methods of this application. It should be noted that the computer-readable storage medium of this application can be a computer-readable signal medium or a computer-readable storage medium or any combination thereof. The computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable storage medium other than a computer-readable storage medium that can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. Program code contained on a computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0216] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0217] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0218] The modules described in the embodiments of this application can be implemented in software or in hardware.

[0219] In another aspect, this application also provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to: extract features for similarity matching based on multi-dimensional medical data, divide the multi-dimensional medical data into dictionary data and field data, and calculate quantitative indicators; the quantitative indicators include field completeness rate, dictionary completeness rate, Java similarity matching rate, large model similarity matching rate, field complete equality rate, dictionary complete equality rate, Java runtime impact rate, and large model runtime impact rate; perform weighted scoring calculation on the quantitative indicators to obtain a comprehensive score result; select a similarity matching method based on the comprehensive score result, split the task according to the smallest executable granularity and submit it to the forkjoin parallel framework for field matching; dynamically adjust the indicator weights according to the quantitative indicators and smoothing factors, and normalize the indicator weights.

[0220] Finally, it should be noted that the above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A smart field matching method based on a large model, characterized in that, include: S1: Based on multi-dimensional medical data, extract features for similarity matching, divide the multi-dimensional medical data into dictionary data and field data, and calculate quantitative indicators; The quantitative indicators include field completeness rate, dictionary completeness rate, Java similarity matching rate, large model similarity matching rate, field complete equality rate, dictionary complete equality rate, Java runtime impact rate, and large model runtime impact rate. S2: Perform weighted scoring calculations on the quantitative indicators to obtain a comprehensive score result; Score= ; Score represents the overall score result. Indicates the field completeness weight. The weight represents the dictionary completeness rate. This represents the weight of the Java similarity matching rate. This represents the weight of the large model similarity matching rate. Indicates the weight of the field's complete equality rate. The weights represent the percentage of words that are completely identical. This indicates the weight of the Java runtime impact rate. Indicates the weights of the impact rate during the runtime of the large model. This indicates the field completeness score. This indicates the dictionary completeness score. This represents the Java similarity matching rate score. This represents the large model similarity matching rate score. This indicates the score for the complete equality rate of the field. This represents the score indicating the complete equality rate of the dictionary entries. This represents the Java runtime impact score. This represents the impact rate score of the large model runtime; S3: Select a similarity matching method based on the comprehensive scoring results, split the task into the smallest executable granularity, and submit it to the forkjoin parallel framework for field matching; The similarity matching methods include large model similarity matching methods, Java algorithm similarity matching methods, and hybrid mode similarity matching methods; S4: Dynamically adjust the indicator weights based on the quantification indicators and smoothing factors, and normalize the indicator weights. The update formula for the dynamically adjusted indicator weights is as follows: ； in, The smoothed value representing the rate of change is calculated using the exponentially weighted moving average (EWMA). Indicates the initial weights. The updated weight is represented by f, which represents the scaling factor in the weight adjustment formula. f∈(0.1,0.3] is used to control the sensitivity of the weight change and the degree of influence of the rate of change on the final weight adjustment.

2. The method according to claim 1, characterized in that, The multi-dimensional medical data mentioned in step S1 includes table information of the data source, field information of the data source table, and national standard information; The table information of the data source includes the English name of the table and table remarks; The field information of the data source table includes the English name of the field, field remarks, and field data type; The national standard information includes the English name of the standard dataset, the Chinese name of the dataset, the English code of the data element, the Chinese name of the data element, and the data element metadata type; Extract features from dictionary data: dictionary name, dictionary code, dictionary rule name, and dictionary rule code; Extract features from field data: table name, table comments, field names, field comments, and field data types.

3. The method according to claim 1, characterized in that, The formula for calculating the field completeness rate is as follows: ZDuan = ZDx / ZDall; ZDuan represents the field completeness rate, which includes five field dimensions: table name, table comments, field English name, field comments, and field data type; ZDx represents the number of values ​​in a field; and ZDall represents the total number of fields. The formula for calculating the dictionary completeness rate is as follows: ZDian = ZDy / ZDiall; ZDian represents the dictionary completeness rate, which includes four dictionary dimensions: dictionary name, dictionary code, dictionary rule name, and dictionary rule code; ZDy represents the number of dictionary entries with values; and ZDiall represents the total number of dictionary entries. The formula for calculating the Java similarity matching rate is as follows: PPjava=mz / bzdall; Where PPjava represents the Java algorithm matching rate, mz represents the number of fields matched, and bzdall represents the total number of table fields; The formula for calculating the large model similarity matching rate is as follows: PPmode=mz / bzdall; Where PPmode represents the large model matching rate, mz represents the number of fields matched, and bzdall represents the total number of table fields.

4. The method according to claim 1, characterized in that, The formula for calculating the complete equality rate of the fields is: AEduan=duanae / bzdall; Where AEDuan represents the field equality rate, duanae represents the number of fields that are completely equal, and bzdall represents the total number of fields in the table. The formula for calculating the complete equality rate of the dictionaries is: AEdian = dianae / zdall; Where AEdian represents the percentage of dictionaries that are completely identical, dianae represents the number of dictionaries that are completely identical, and zdall represents the total number of dictionaries.

5. The method according to claim 1, characterized in that, The formula for calculating the Java runtime impact rate is as follows: TimeJava=Min(100%,javatime / maxtime×100%); Among them, TimeJava represents the Java runtime impact rate, javatime represents the actual Java runtime, and maxtime represents the maximum allowed duration. The formula for calculating the runtime impact rate of the large model is as follows: Timemode=Min(100%,modetime / maxtime×100%); Wherein, Timemode represents the impact rate of large model runtime, modetime represents the actual runtime of large model, and maxtime represents the maximum allowed runtime.

6. The method according to claim 1, characterized in that, The similarity matching method is selected based on the comprehensive score result. Specifically, if the score is greater than the preset maximum threshold, a large model is selected for similarity matching. If the overall score is less than the preset minimum threshold, a Java algorithm is used for similarity matching. Otherwise, a hybrid approach is used for similarity matching: Java algorithms are used for similarity matching of dictionary data, while large models are used for similarity matching of field data.

7. The method according to claim 1, characterized in that, The formula for calculating the smoothing value is: ； in, This represents the current smoothing value. This represents the smoothed value from the previous time step. The original rate of change at the current moment is represented by α, which represents the smoothing coefficient and is used to control the degree of influence of the current rate of change on the smoothed value. α ∈ (0.1, 0.3).

8. A smart field matching system based on a large model, characterized in that, include: Feature index module: Extracts features for similarity matching based on multi-dimensional medical data, divides the multi-dimensional medical data into dictionary data and field data, and calculates quantitative indicators; The quantitative indicators include field completeness rate, dictionary completeness rate, Java similarity matching rate, large model similarity matching rate, field complete equality rate, dictionary complete equality rate, Java runtime impact rate, and large model runtime impact rate. Comprehensive scoring module: performs weighted scoring calculations on the quantitative indicators to obtain a comprehensive scoring result; Score= ; Score represents the overall score result. Indicates the field completeness weight. The weight represents the dictionary completeness rate. This represents the weight of the Java similarity matching rate. This represents the weight of the large model similarity matching rate. Indicates the weight of the field's complete equality rate. The weights represent the percentage of words that are completely identical. This indicates the weight of the Java runtime impact rate. Indicates the weights of the impact rate during the runtime of the large model. This indicates the field completeness score. This indicates the dictionary completeness score. This represents the Java similarity matching rate score. This represents the large model similarity matching rate score. This indicates the score for the complete equality rate of the field. This represents the score indicating the complete equality rate of the dictionary entries. This represents the Java runtime impact score. This represents the impact rate score of the large model runtime; Divide and conquer module: Selects a similarity matching method based on the comprehensive scoring results, divides the multi-dimensional medical data into dictionary data and field data, splits the task according to the smallest executable granularity, and submits it to the forkjoin parallel framework for field matching; The similarity matching methods include large model similarity matching methods, Java algorithm similarity matching methods, and hybrid mode similarity matching methods; Optimization module: Dynamically adjusts the indicator weights based on the quantification indicators and smoothing factors, and normalizes the indicator weights. The update formula for the dynamically adjusted indicator weights is as follows: ； in, The smoothed value representing the rate of change is calculated using the exponentially weighted moving average (EWMA). Indicates the initial weights. The updated weight is represented by f, which represents the scaling factor in the weight adjustment formula. f∈(0.1,0.3] is used to control the sensitivity of the weight change and the degree of influence of the rate of change on the final weight adjustment.

9. A computer program product having one or more computer programs thereon, characterized in that, When the computer program is executed by a computer processor, the method described in any one of claims 1-7 is performed.

Images