Large model medical term normalization system and method based on online retrieval enhancement

Abstract

The invention discloses a large-model medical term normalization system and method based on online retrieval enhancement, and relates to the technical field of monitoring analysis, and the system comprises a preliminary analysis module which is used for carrying out preliminary analysis on an input medical text, and recognizing primary medical terms corresponding to the medical text; the multi-level matching strategy module comprises a rule matching unit and an online retrieval unit, the rule matching unit is used for carrying out preliminary normalization processing on the recognized primary medical terms, and the online retrieval unit is used for carrying out online query on the primary medical terms subjected to the preliminary normalization processing and retrieving a plurality of matching items; the score sorting module is used for analyzing and processing the plurality of matching items, confirming scores corresponding to the matching items, sorting based on the scores corresponding to the matching items, and screening out an optimal matching item; and the integration output module is used for confirming standardization results corresponding to the primary medical terms of the medical text and integrating the standardization results. The method has the effect of improving the working efficiency.

Description

Technical Field

[0001] The present application relates to the field of monitoring and analysis technology, and in particular to a large-model medical term normalization system and method based on online retrieval enhancement. Background Art

[0002] In recent years, with the development of computing hardware, artificial intelligence services based on large language models have received more and more attention. Compared with traditional medical models, large language models have richer world knowledge, and their pre-training is expected to cover most of the corpus on the Internet. In addition, the ability to follow instructions and logical reasoning that emerges from large language models makes large models more powerful than traditional language models in many fields.

[0003] In the related technology, in actual medical scenarios, the same medical term may have different names, such as diagnostic terms and surgical terms. Such inconsistent medical data is not only not conducive to the standardized management of medical data, but also causes a lot of data preprocessing work for related research work based on clinical data, affecting work efficiency and increasing the burden on relevant staff. There is room for improvement. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present application provides a large-model medical terminology normalization system and method based on online retrieval enhancement.

[0005] In a first aspect, the present application provides a large-model medical term normalization system based on online retrieval enhancement, comprising: A preliminary parsing module is used to perform preliminary parsing of the input medical text based on the ERNIE-HEALTH model, extract term features, and then identify primary medical terms corresponding to the medical text; A multi-level matching strategy module, comprising a rule matching unit and an online retrieval unit, wherein the rule matching unit is used to perform preliminary normalization processing on the identified primary medical terms, and the online retrieval unit is used to perform online query on the primary medical terms after the preliminary normalization processing, and retrieve multiple matching items corresponding to the primary medical terms after the preliminary normalization processing; A scoring and sorting module, for analyzing and processing the multiple matching items, and confirming the scores corresponding to the matching items based on the results of the analysis and processing, and then sorting the matching items based on the scores corresponding to the matching items, and selecting the best matching item as the standardized result corresponding to the primary medical term; The integration output module is used to confirm the standardized results corresponding to the primary medical terms in the medical text, integrate the standardized results, and output based on the integrated results.

[0006] Preferably, the preliminary analysis module includes an anticipation processing unit, a feature extraction unit and a coarse-grained matching unit; The prediction processing unit is used to perform preprocessing operations on the input medical text, and the preprocessing operations include word segmentation processing, stop word removal processing and part-of-speech tagging processing; The feature extraction unit is used to extract the deep semantic features corresponding to the medical text after the preprocessing operation based on the preset ERNIE-HEALTH model; The coarse-grained matching unit is used to perform preliminary term matching according to the deep semantic features, and then identify the primary medical terms corresponding to the medical text.

[0007] Preferably, the rule matching unit includes a rule base and a matching engine subunit; The rule base is constructed by expert knowledge and historical data, and stores medical term variant data, synonym data, and abbreviation data; The matching engine subunit is used to match and normalize primary medical terms using a rule base.

[0008] Preferably, the online retrieval unit includes a retrieval strategy subunit and a data filtering subunit; The search strategy subunit is used to dynamically adjust the online search strategy, which includes selecting search sources, setting query parameters and optimizing search efficiency; The data filtering subunit is used to filter the retrieved multiple matching items based on semantic and context filtering algorithms.

[0009] Preferably, the scoring and ranking module includes a data identification unit, a data analysis unit and a ranking unit; The data identification unit is used to identify, from the multiple matching items, a semantic similarity score and a contextual relevance score corresponding to each matching item and the primary medical term, and to identify a frequency score and a professional score corresponding to each matching item; The data analysis unit is used to comprehensively analyze the semantic similarity score, context relevance score, and frequency score and professional score of each matching item and the primary medical term, and then determine the comprehensive score of each matching item; The sorting unit is used to sort the matching items according to the comprehensive scores corresponding to the matching items in a preset order, and select the best matching item as the standardized result corresponding to the primary medical term.

[0010] Preferably, a comprehensive analysis is performed on the semantic similarity score, context relevance score, and frequency score and professional score of each matching item corresponding to each matching item and the primary medical term, and then a comprehensive score corresponding to each matching item is determined, specifically including: By formula , confirm the comprehensive score corresponding to each matching item ; in, Indicates the number corresponding to each matching item. Expressed as primary medical terminology, Indicated as number The candidate standard terms corresponding to the matches of Represented as primary medical terms and numbered Semantic similarity scores and contextual relevance scores corresponding to candidate standard terms of matching items, Respectively indicated as number Frequency score and professionalism score corresponding to the candidate standard terms of the matching items, Respectively represented as preset weight coefficients; Obtain the comprehensive score corresponding to each matching item, and sort the comprehensive scores corresponding to each matching item in order from high to low, confirm the matching item with the highest comprehensive score, and use the candidate standard term corresponding to the matching item as the standardized result corresponding to the primary medical term.

[0011] Preferably, after outputting based on the integration result, the method further includes: In the preset time period, collect the real-time precision coefficients of the integrated results output, and use the function to convert the real-time precision coefficients into real-time precision coefficients according to the time series. express; By formula , confirm the change coefficient corresponding to the real-time precision coefficient ; in, Indicates the preset time period. Expressed as a preset standard precision coefficient, They are respectively expressed as the maximum precision coefficient and the minimum precision coefficient. Represented as a preset adjustment function; The change coefficient corresponding to the real-time precision coefficient The preset change threshold Make a comparison; If the change coefficient corresponding to the real-time precision coefficient , it is determined that the accuracy of the integrated result output in the future period is abnormal and needs to be adjusted.

[0012] In a second aspect, the present application provides a large-model medical term normalization method based on online retrieval enhancement, comprising the following steps: Based on the ERNIE-HEALTH model, the input medical text is preliminarily parsed and term features are extracted to further identify the primary medical terms corresponding to the medical text; A multi-level matching strategy includes a rule matching unit and an online retrieval unit, wherein the rule matching unit is used to perform preliminary normalization processing on the identified primary medical terms, and the online retrieval unit is used to perform online query on the primary medical terms after the preliminary normalization processing, and retrieve multiple matching items corresponding to the primary medical terms after the preliminary normalization processing; Analyze and process the multiple matching items, and confirm the scores corresponding to the matching items based on the results of the analysis and processing, and then sort the matching items based on the scores corresponding to the matching items, and select the best matching item as the standardized result corresponding to the primary medical term; Confirm the standardized results corresponding to the primary medical terms of the medical text, integrate the standardized results, and output based on the integrated results.

[0013] In a third aspect, the present application provides a computer-readable storage medium storing instructions, which, when executed on a computer, enable the computer to execute any one of the above-mentioned large-model medical terminology normalization systems based on online retrieval enhancement.

[0014] In summary, this application includes the following beneficial technical effects: The present application provides a large-model medical term normalization system based on online retrieval enhancement. By performing preliminary analysis on the input medical text, the primary medical term corresponding to the medical text is identified, and multiple matching items are retrieved through an online retrieval unit, and the scores corresponding to each matching item are confirmed. Then, the matching items are sorted based on the scores corresponding to each matching item, and the best matching item is screened out as the standardized result corresponding to the primary medical term, thereby effectively reducing the situation where the same medical term has different names that is not conducive to the standardized management of medical data, thereby effectively improving the work efficiency of the staff. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0016] Figure 1 It is a system schematic diagram of a large-model medical term normalization based on online retrieval enhancement in an embodiment of the present application.

[0017] Figure 2 It is a flow chart of the method for normalizing large-model medical terms based on online retrieval enhancement in an embodiment of the present application. DETAILED DESCRIPTION

[0018] The following is combined with Figure 1-2 This application is described in further detail.

[0019] Example 1

[0020] The embodiments of the present application disclose a large-model medical term normalization system based on online retrieval enhancement.

[0021] Reference Figure 1 , a large-model medical term normalization system based on online retrieval enhancement, including: A preliminary parsing module is used to perform preliminary parsing of the input medical text based on the ERNIE-HEALTH model, extract term features, and then identify primary medical terms corresponding to the medical text; A multi-level matching strategy module, comprising a rule matching unit and an online retrieval unit, wherein the rule matching unit is used to perform preliminary normalization processing on the identified primary medical terms, and the online retrieval unit is used to perform online query on the primary medical terms after the preliminary normalization processing, and retrieve multiple matching items corresponding to the primary medical terms after the preliminary normalization processing; A scoring and sorting module, for analyzing and processing the multiple matching items, and confirming the scores corresponding to the matching items based on the results of the analysis and processing, and then sorting the matching items based on the scores corresponding to the matching items, and selecting the best matching item as the standardized result corresponding to the primary medical term; The integration output module is used to confirm the standardized results corresponding to the primary medical terms in the medical text, integrate the standardized results, and output based on the integrated results.

[0022] Specifically, by combining the preliminary parsing capability of the ERNIE-HEALTH model and the dynamic updating capability of online retrieval, efficient normalization of medical terminology is achieved.

[0023] Further, the preliminary analysis module includes an anticipation processing unit, a feature extraction unit and a coarse-grained matching unit; The prediction processing unit is used to perform preprocessing operations on the input medical text, and the preprocessing operations include word segmentation processing, stop word removal processing and part-of-speech tagging processing; The feature extraction unit is used to extract the deep semantic features corresponding to the medical text after the preprocessing operation based on the preset ERNIE-HEALTH model; The coarse-grained matching unit is used to perform preliminary term matching according to the deep semantic features, and then identify the primary medical terms corresponding to the medical text.

[0024] It should be noted that the rule matching unit includes a rule base and a matching engine subunit; The rule base is constructed by expert knowledge and historical data, and stores medical term variant data, synonym data, and abbreviation data; The matching engine subunit is used to match and normalize primary medical terms using a rule base.

[0025] Specifically, the engine uses an efficient string matching algorithm that can quickly identify and process term variants, and through the combination of a rule base and a matching engine, it achieves rapid preliminary normalization of primary medical terminology.

[0026] Further, the online retrieval unit includes a retrieval strategy subunit and a data filtering subunit; The search strategy subunit is used to dynamically adjust the online search strategy, which includes selecting search sources, setting query parameters and optimizing search efficiency; by dynamically adjusting the search strategy, the timeliness and accuracy of the acquired terminology standardization data are ensured; The data filtering subunit is used to filter the multiple retrieved matching items based on semantic and context filtering algorithms, and the filtering process includes but is not limited to removing irrelevant or low-quality results.

[0027] Specifically, a filtering algorithm based on semantics and context is used to improve the reliability of retrieval results.

[0028] Furthermore, the scoring and ranking module includes a data identification unit, a data analysis unit and a ranking unit; The data identification unit is used to identify, from the multiple matching items, a semantic similarity score and a contextual relevance score corresponding to each matching item and the primary medical term, and to identify a frequency score and a professional score corresponding to each matching item; The data analysis unit is used to comprehensively analyze the semantic similarity score, context relevance score, and frequency score and professional score of each matching item and the primary medical term, and then determine the comprehensive score of each matching item; Specifically, the semantic similarity score, contextual relevance score, frequency score, and professionalism score can all be obtained by the deep learning model; The sorting unit is used to sort the matching items according to the comprehensive scores corresponding to the matching items in a preset order, and select the best matching item as the standardized result corresponding to the primary medical term.

[0029] It should be noted that the semantic similarity score, contextual relevance score, frequency score and professional score of each matching item are comprehensively analyzed with the primary medical term, and then the comprehensive score of each matching item is determined, including: By formula , confirm the comprehensive score corresponding to each matching item ; in, Indicates the number corresponding to each matching item. Expressed as primary medical terminology, Indicated as number The candidate standard terms corresponding to the matches of Represented as primary medical terms and numbered Semantic similarity scores and contextual relevance scores corresponding to candidate standard terms of matching items, Respectively indicated as number Frequency score and professionalism score corresponding to the candidate standard terms of the matching items, Respectively represented as preset weight coefficients; Obtain the comprehensive score corresponding to each matching item, and sort the comprehensive scores corresponding to each matching item in order from high to low, confirm the matching item with the highest comprehensive score, and use the candidate standard term corresponding to the matching item as the standardized result corresponding to the primary medical term.

[0030] Specifically, scoring and sorting the matches retrieved online can greatly improve the accuracy and efficiency of the system. In practical applications, medical terms often have polysemy, synonym variants, and contextual differences. Through the scoring and sorting mechanism, multiple candidate terms retrieved can be quantitatively evaluated to select standardized terms that best match the input semantics and context. The scoring mechanism comprehensively considers multiple factors such as semantic similarity, contextual relevance, frequency of occurrence of terms, and professionalism to ensure that the final selected terms are not only semantically close to the input terms, but also consistent with authoritative medical standards. At the same time, the sorting algorithm prioritizes the scoring results, allowing the system to quickly lock in the best matches and avoid the subjectivity and inefficiency of manual screening. In addition, by dynamically adjusting the weight parameters of the scoring model, it can adapt to the needs of different medical scenarios or fields, such as clinical diagnosis, drug information processing, or scientific research data analysis, thereby improving the versatility and robustness of the system. Overall, the scoring and sorting mechanism not only optimizes the accuracy of term normalization, but also significantly reduces the time cost of processing complex medical texts, providing reliable and efficient technical support for medical information processing.

[0031] Furthermore, after output based on the integration results, it also includes: In the preset time period, collect the real-time precision coefficients of the integrated results output, and use the function to convert the real-time precision coefficients into real-time precision coefficients according to the time series. express; By formula , confirm the change coefficient corresponding to the real-time precision coefficient ; in, Indicates the preset time period. Expressed as a preset standard precision coefficient, They are respectively expressed as the maximum precision coefficient and the minimum precision coefficient. Represented as a preset adjustment function; The change coefficient corresponding to the real-time precision coefficient The preset change threshold Make a comparison; If the change coefficient corresponding to the real-time precision coefficient , it is determined that the accuracy of the integrated result output in the future period is abnormal and needs to be adjusted.

[0032] Example 2

[0033] The embodiments of the present application also disclose a large-model medical term normalization method based on online retrieval enhancement.

[0034] Reference Figure 2 , a large-model medical term normalization method based on online retrieval enhancement, comprising the following steps: Based on the ERNIE-HEALTH model, the input medical text is preliminarily parsed and term features are extracted to further identify the primary medical terms corresponding to the medical text; A multi-level matching strategy includes a rule matching unit and an online retrieval unit, wherein the rule matching unit is used to perform preliminary normalization processing on the identified primary medical terms, and the online retrieval unit is used to perform online query on the primary medical terms after the preliminary normalization processing, and retrieve multiple matching items corresponding to the primary medical terms after the preliminary normalization processing; Analyze and process the multiple matching items, and confirm the scores corresponding to the matching items based on the results of the analysis and processing, and then sort the matching items based on the scores corresponding to the matching items, and select the best matching item as the standardized result corresponding to the primary medical term; Confirm the standardized results corresponding to the primary medical terms of the medical text, integrate the standardized results, and output based on the integrated results.

[0035] The above contents are merely examples and explanations of the concept of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the concept of the invention, they should all fall within the protection scope of the present invention.

[0036] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0037] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well.

Claims

1. A large-model medical term normalization system based on online retrieval enhancement, characterized by: include: A preliminary parsing module is used to perform preliminary parsing of the input medical text based on the ERNIE-HEALTH model, extract term features, and then identify primary medical terms corresponding to the medical text; A multi-level matching strategy module, comprising a rule matching unit and an online retrieval unit, wherein the rule matching unit is used to perform preliminary normalization processing on the identified primary medical terms, and the online retrieval unit is used to perform online query on the primary medical terms after the preliminary normalization processing, and retrieve multiple matching items corresponding to the primary medical terms after the preliminary normalization processing; A scoring and sorting module, for analyzing and processing the multiple matching items, and confirming the scores corresponding to the matching items based on the results of the analysis and processing, and then sorting the matching items based on the scores corresponding to the matching items, and selecting the best matching item as the standardized result corresponding to the primary medical term; The integration output module is used to confirm the standardized results corresponding to the primary medical terms in the medical text, integrate the standardized results, and output based on the integrated results.

2. The large-model medical term normalization system based on online retrieval enhancement according to claim 1 is characterized in that: The preliminary analysis module includes an anticipation processing unit, a feature extraction unit and a coarse-grained matching unit; The prediction processing unit is used to perform preprocessing operations on the input medical text, and the preprocessing operations include word segmentation processing, stop word removal processing and part-of-speech tagging processing; The feature extraction unit is used to extract the deep semantic features corresponding to the medical text after the preprocessing operation based on the preset ERNIE-HEALTH model; The coarse-grained matching unit is used to perform preliminary term matching according to the deep semantic features, and then identify the primary medical terms corresponding to the medical text.

3. The large-model medical term normalization system based on online retrieval enhancement according to claim 1 is characterized in that: The rule matching unit includes a rule base and a matching engine subunit; The rule base is constructed by expert knowledge and historical data, and stores medical term variant data, synonym data, and abbreviation data; The matching engine subunit is used to match and normalize primary medical terms using a rule base.

4. The large-model medical term normalization system based on online retrieval enhancement according to claim 1 is characterized in that: The online retrieval unit includes a retrieval strategy subunit and a data filtering subunit; The search strategy subunit is used to dynamically adjust the online search strategy, which includes selecting search sources, setting query parameters and optimizing search efficiency; The data filtering subunit is used to filter the retrieved multiple matching items based on semantic and context filtering algorithms.

5. The large-model medical term normalization system based on online retrieval enhancement according to claim 1 is characterized in that: The scoring and sorting module includes a data identification unit, a data analysis unit and a sorting unit; The data identification unit is used to identify, from the multiple matching items, a semantic similarity score and a contextual relevance score corresponding to each matching item and the primary medical term, and to identify a frequency score and a professional score corresponding to each matching item; The data analysis unit is used to comprehensively analyze the semantic similarity score, context relevance score, and frequency score and professional score of each matching item and the primary medical term, and then determine the comprehensive score of each matching item; The sorting unit is used to sort the matching items according to the comprehensive scores corresponding to the matching items in a preset order, and select the best matching item as the standardized result corresponding to the primary medical term.

6. The large-model medical term normalization system based on online retrieval enhancement according to claim 5 is characterized in that: The semantic similarity score, contextual relevance score, frequency score and professional score of each matching item and the primary medical term are comprehensively analyzed to determine the comprehensive score of each matching item, including: By formula , confirm the comprehensive score corresponding to each matching item ; in, Indicates the number corresponding to each matching item. Expressed as primary medical terminology, Indicated as number The candidate standard terms corresponding to the matches of Represented as primary medical terms and numbered Semantic similarity scores and contextual relevance scores corresponding to candidate standard terms of matching items, Respectively indicated as number Frequency score and professionalism score corresponding to the candidate standard terms of the matching items, Respectively represented as preset weight coefficients; Obtain the comprehensive score corresponding to each matching item, and sort the comprehensive scores corresponding to each matching item in order from high to low, confirm the matching item with the highest comprehensive score, and use the candidate standard term corresponding to the matching item as the standardized result corresponding to the primary medical term.

7. The large-model medical term normalization system based on online retrieval enhancement according to claim 1 is characterized in that: After the output is based on the integration results, it also includes: In the preset time period, collect the real-time precision coefficients of the integrated results output, and use the function to convert the real-time precision coefficients into real-time precision coefficients according to the time series. express; By formula , confirm the change coefficient corresponding to the real-time precision coefficient ; in, Indicates the preset time period. Expressed as a preset standard precision coefficient, They are respectively expressed as the maximum precision coefficient and the minimum precision coefficient. Represented as a preset adjustment function; The change coefficient corresponding to the real-time precision coefficient The preset change threshold Make a comparison; If the change coefficient corresponding to the real-time precision coefficient , it is determined that the accuracy of the integrated result output in the future period is abnormal and needs to be adjusted.

8. A large-model medical term normalization method based on online retrieval enhancement, applied to a large-model medical term normalization system based on online retrieval enhancement as described in any one of claims 1 to 7, characterized in that: The following steps are involved: Based on the ERNIE-HEALTH model, the input medical text is preliminarily parsed and term features are extracted to further identify the primary medical terms corresponding to the medical text; A multi-level matching strategy includes a rule matching unit and an online retrieval unit, wherein the rule matching unit is used to perform preliminary normalization processing on the identified primary medical terms, and the online retrieval unit is used to perform online query on the primary medical terms after the preliminary normalization processing, and retrieve multiple matching items corresponding to the primary medical terms after the preliminary normalization processing; Analyze and process the multiple matching items, and confirm the scores corresponding to the matching items based on the results of the analysis and processing, and then sort the matching items based on the scores corresponding to the matching items, and select the best matching item as the standardized result corresponding to the primary medical term; Confirm the standardized results corresponding to the primary medical terms of the medical text, integrate the standardized results, and output based on the integrated results.

9. A computer-readable storage medium, characterized in that: Instructions are stored, and when the instructions are executed on a computer, the computer executes the large-model medical term normalization system based on online retrieval enhancement as described in any one of claims 1 to 7.

Images