System, method and terminal for dynamic entity recognition based on data quality awareness

Abstract

This invention provides a dynamic entity recognition system, method, and terminal based on data quality awareness. The system uses a data quality awareness and analysis module to assess the quality of raw input data for specific medical entity types, generating feature vectors that accurately reflect the data quality status. A composable processing chain repository module pre-sets multiple processing chains with different processing strategies for each medical entity type. Each processing chain consists of multiple processors arranged in a specific execution order to meet diverse data processing needs. A dynamic processing chain decision module queries a strategy knowledge base based on the data quality feature vectors, accurately outputting suitable processing chain identifiers, each uniquely corresponding to a processing chain. A chained execution engine module retrieves the corresponding processing chain from the dynamic processing chain decision module based on the processing chain identifier and drives each processor in the processing chain to process the raw input data sequentially, ultimately outputting accurate recognition results. This invention can intelligently select and execute the processing chain most suitable for the quality characteristics of each input data, achieving optimal allocation of computing resources and maximizing the success rate of recognition tasks.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, and in particular to a dynamic entity recognition system, method and terminal based on data quality perception. Background Technology

[0002] In healthcare information systems, entity recognition is a crucial foundational technology for matching patient master indexes and normalizing master data such as drugs and diseases. However, the hard-coded, single-algorithm model in current mainstream entity recognition system architectures has significant drawbacks. This method directly encodes specific recognition logic, such as rule-based, dictionary-based, or single machine learning models, into the system core. This leads to extreme system rigidity; any changes in business logic or algorithm upgrades necessitate modification, testing, and redeployment of the core code—a lengthy and risky process. Furthermore, a fixed model struggles to handle both high-quality data (such as data with complete information) and low-quality data (such as data with missing key fields), often only managing a compromise performance level for all situations, resulting in either wasted resources or decreased recognition accuracy.

[0003] Static configuration-based rule engines or processing pipelines are an improvement over hard-coded single algorithm models. They allow for the pre-assembly of fixed processing chains for different "entity types" through external configuration files, such as configuring a "full value matching -> synonym matching -> AI matching" chain for the "pharmaceutical" type. However, this improvement has a core problem: its configuration is static and type-granular, rather than dynamic and data-granular. The system cannot detect significant quality differences between different input data of the same type. Taking "pharmaceutical" data as an example, for high-quality data with complete and standardized "approval numbers," the complete chain forces it to undergo all subsequent time-consuming fuzzy matching calculations unnecessarily, severely wasting computational resources; while for low-quality data with missing "approval numbers" but non-standard "generic names," the fixed chain may fail to effectively identify the data due to a lack of sufficiently strong deep analysis steps (such as being skipped by short-circuit analysis), ultimately leading to matching failure.

[0004] In summary, existing technologies have significant shortcomings in entity recognition, given the problems of the two mainstream architectures mentioned above. Currently, there is a lack of systems and methods capable of real-time sensing of input data quality characteristics at the single-request granularity, and dynamically and automatically selecting and adapting to the optimal processing flow accordingly. This makes it difficult for the system to achieve a fundamental balance between processing efficiency and recognition accuracy, failing to meet the requirements of medical information systems for efficient and accurate entity recognition, and consequently affecting the overall level and quality of medical data information processing. Summary of the Invention

[0005] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a dynamic entity recognition system, method and terminal based on data quality perception, which is used to solve the technical problems of entity recognition technology, such as the lack of a system and method to perceive the quality characteristics of input data in real time at the granularity of a single request and dynamically and automatically adapt to the optimal processing flow, which makes it difficult to balance processing efficiency and recognition accuracy, and thus fails to meet the needs of medical informatization and affects the level of data informatization processing.

[0006] To achieve the above and other related objectives, this invention provides a dynamic entity recognition system based on data quality awareness. The system includes: a data quality awareness and analysis module for assessing the quality of raw input data of a medical entity type and generating a data quality feature vector; a composable processing chain repository module for pre-setting multiple processing chains with different strategies for each medical entity type; wherein each processing chain is composed of multiple processors arranged in execution order; a dynamic processing chain decision module for querying a strategy knowledge base based on the data quality feature vector and outputting an appropriate processing chain identifier; wherein each processing chain identifier corresponds to a processing chain; and a chained execution engine module for loading the corresponding processing chain from the dynamic processing chain decision module based on the processing chain identifier, driving each processor in the processing chain to operate on the raw input data, and outputting the final recognition result.

[0007] In one embodiment of the present invention, the step of performing quality assessment on the original input data of a medical entity type and generating a data quality feature vector includes: extracting the corresponding quality assessment rules based on the medical entity type of the original input data; and performing data quality assessment on each key attribute dimension of the original input data based on the extracted quality assessment rules to obtain the data quality assessment results of each key attribute dimension, thereby generating a data quality feature vector.

[0008] In one embodiment of the present invention, a processing chain adapted to different data quality levels is pre-configured for each type of medical entity.

[0009] In one embodiment of the present invention, the types of processing chains include: a precise identification chain for processing high-quality data, a semantic fusion chain for processing medium-quality data, and a deep exploration chain for processing low-quality data.

[0010] In one embodiment of the present invention, the step of querying the strategy knowledge base to output the appropriate processing chain identifier based on the data quality feature vector includes: determining the data quality level of the current original input data based on the data quality feature vector, and querying the strategy knowledge base to output the appropriate processing chain identifier based on the corresponding medical entity type and data quality level.

[0011] In one embodiment of the present invention, each processor in the driving processing chain operates on the original input data and outputs the final recognition result, including: if a precise identification chain is loaded, then each processor in the precise identification chain is driven to perform precise query based on the original input data to output a unique standard entity; if a semantic fusion chain is loaded, then each processor in the semantic fusion chain is driven to perform text analysis and arbitration based on the original input data to output a high-confidence inferred entity; if a deep exploration chain is loaded, then each processor in the deep exploration chain is driven to perform knowledge reasoning and AI prediction to obtain the predicted entity with the highest matching degree.

[0012] In one embodiment of the present invention, the system further determines whether the predicted entity matching degree obtained by the deep exploration chain is greater than the matching degree threshold. If it is not greater, the predicted entity is sent to the manual review terminal for review.

[0013] In one embodiment of the present invention, the data quality perception and analysis module is used to evaluate the completeness, standardization, validity, and logical anomalies of the original input data.

[0014] To achieve the above and other related objectives, this invention provides a dynamic entity recognition method based on data quality awareness, applied to the aforementioned dynamic entity recognition system based on data quality awareness. The method includes: assessing the quality of original input data of a medical entity type to generate a data quality feature vector; querying a strategy knowledge base based on the data quality feature vector to output an appropriate processing chain identifier; loading the corresponding processing chain from a dynamic processing chain decision module based on the processing chain identifier, driving each processor in the processing chain to operate on the original input data, and outputting the final recognition result.

[0015] To achieve the above and other related objectives, the present invention provides an electronic terminal, comprising: one or more memories and one or more processors; the one or more memories are used to store a computer program; the one or more processors are connected to the memories and are used to run the computer program to execute the dynamic entity recognition system based on data quality perception.

[0016] As described above, this invention is a dynamic entity recognition system, method, and terminal based on data quality awareness, which has the following beneficial effects: The data quality awareness and analysis module of this invention assesses the quality of raw input data for specific medical entity types, generating feature vectors that accurately reflect the data quality status. The composable processing chain repository module pre-sets multiple processing chains with different processing strategies for each medical entity type. Each processing chain consists of multiple processors arranged in a specific execution order to meet diverse data processing needs. The dynamic processing chain decision module queries the strategy knowledge base based on the data quality feature vector and accurately outputs the appropriate processing chain identifier, with each identifier uniquely corresponding to a processing chain. The chained execution engine module obtains the corresponding processing chain from the dynamic processing chain decision module based on the processing chain identifier and drives each processor in the processing chain to process the raw input data sequentially, ultimately outputting an accurate recognition result. This invention can intelligently select and execute the processing chain most suitable for the quality characteristics of each input data, achieving optimal allocation of computing resources and maximizing the success rate of recognition tasks. Attached Figure Description

[0017] Figure 1 The diagram shown is a structural schematic of a dynamic entity recognition system based on data quality awareness according to an embodiment of the present invention.

[0018] Figure 2 The diagram shown is a flowchart of a dynamic entity recognition system based on data quality awareness according to an embodiment of the present invention.

[0019] Figure 3 The diagram shown is a flowchart illustrating a method for identifying a medical drug standard library according to an embodiment of the present invention.

[0020] Figure 4 The diagram shown is a flowchart of a dynamic entity recognition method based on data quality awareness according to an embodiment of the present invention.

[0021] Figure 5 The diagram shown is a structural schematic of an electronic terminal according to an embodiment of the present invention. Detailed Implementation

[0022] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0023] It should be noted that in the following description, reference is made to the accompanying drawings, which illustrate several embodiments of the invention. It should be understood that other embodiments may also be used, and changes in mechanical composition, structure, electrical system, and operation may be made without departing from the spirit and scope of the invention. The following detailed description should not be considered limiting, and the scope of the embodiments of the invention is defined only by the claims of the published patents. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. Spatially related terms, such as “upper,” “lower,” “left,” “right,” “below,” “below,” “lower part,” “above,” “upper part,” etc., may be used herein to illustrate the relationship between one element or feature shown in the figures and another element or feature.

[0024] Throughout this specification, when it is said that a part is "connected" to another part, this includes not only "direct connection" but also "indirect connection" by placing other elements in between. Furthermore, when it is said that a part "includes" a certain constituent element, unless otherwise stated otherwise, this does not exclude other constituent elements, but rather means that other constituent elements may also be included.

[0025] The terms "first," "second," and "third," etc., used herein are for the purpose of describing various parts, components, regions, layers, and / or segments, but are not limiting. These terms are used only to distinguish one part, component, region, layer, or segment from others. Therefore, the "first part," "component," "region," "layer," or "segment" described below may refer to a "second part," "component," "region," "layer," or "segment" without departing from the scope of this invention.

[0026] Furthermore, as used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context indicates otherwise. It should be further understood that the terms “comprising,” “including,” indicate the presence of the stated feature, operation, element, component, item, kind, and / or group, but do not preclude the presence, occurrence, or addition of one or more other features, operations, elements, components, items, kinds, and / or groups. The terms “or” and “and / or” as used herein are interpreted as inclusive, or mean any one or any combination thereof. Thus, “A, B, or C” or “A, B, and / or C” means “any one of: A; B; C; A and B; A and C; B and C; A, B, and C.” Exceptions to this definition arise only when combinations of elements, functions, or operations are inherently mutually exclusive in some manner.

[0027] This invention provides a dynamic entity recognition system based on data quality awareness. A data quality awareness and analysis module assesses the quality of raw input data for specific medical entity types, generating feature vectors that accurately reflect the data quality status. A composable processing chain repository module pre-sets multiple processing chains with different processing strategies for each medical entity type. Each processing chain consists of multiple processors arranged in a specific execution order to meet diverse data processing needs. A dynamic processing chain decision module queries a strategy knowledge base based on the data quality feature vectors, accurately outputting suitable processing chain identifiers, each uniquely corresponding to a processing chain. A chained execution engine module retrieves the corresponding processing chain from the dynamic processing chain decision module based on the processing chain identifier and drives each processor in the processing chain to process the raw input data sequentially, ultimately outputting accurate recognition results. This invention can intelligently select and execute the processing chain most suitable for the quality characteristics of each input data, achieving optimal allocation of computing resources and maximizing the success rate of recognition tasks.

[0028] The present invention will now be described in detail with reference to the accompanying drawings, so that those skilled in the art can readily implement it. The present invention can be embodied in many different forms and is not limited to the embodiments described herein.

[0029] like Figure 1 This diagram illustrates the structure of a dynamic entity recognition system based on data quality awareness, as described in an embodiment of the present invention.

[0030] The system includes:

[0031] The Data Quality Awareness and Analysis Module 1 is used to assess the quality of raw input data for a specific medical entity type, generating a data quality feature vector. The medical entity types involved are broadly encompassing, including but not limited to common categories such as drugs (covering detailed information such as drug name, specifications, dosage form, and manufacturer), patients (containing rich content such as basic patient information, medical records, diagnostic results, and treatment processes), and diseases (involving professional knowledge such as disease name, classification, symptoms, and pathogenesis). Through accurate quality assessment of this raw data, a solid and reliable quality basis is provided for subsequent data processing flows, ensuring that the entire system operates based on high-quality data.

[0032] The composable processing chain repository module 2 is used to pre-configure multiple processing chains with differentiated strategies for each type of medical entity. For each type of medical entity, multiple processing chains with differentiated strategies are pre-configured. Each processing chain is constructed by combining multiple processors with different functions and cooperating with each other according to a preset, strict execution order. Considering the significant differences in data characteristics, processing requirements, and business logic among different types of medical entities such as patients and drugs, this module pre-plans multiple processing chains. These processing chains have different strategic focuses: some emphasize data cleaning and preprocessing to remove noisy data and correct erroneous data; some emphasize data transformation and integration, uniformly formatting and associating data from different sources; and some emphasize data mining and analysis to discover hidden potential information and patterns in the data. At the same time, they also vary in algorithm complexity, ranging from simple rule matching algorithms to complex machine learning algorithms, meeting the requirements for data processing accuracy and efficiency in different scenarios.

[0033] Each processing chain is essentially an ordered execution plan based on processors. These processors have clearly defined input / output interfaces and functional definitions, ensuring the stability and scalability of the processing chain. Furthermore, each processor possesses high flexibility, supporting free combination according to actual business needs and data characteristics. When facing new data processing scenarios or changes in business requirements, processors can be easily added or removed, quickly adjusting the structure and function of the processing chain to perfectly adapt to the diverse needs of different data processing scenarios, providing strong support for the system's flexibility and adaptability.

[0034] The dynamic processing chain decision module 3, connected to the data quality perception and analysis module 1, is used to query the built-in strategy knowledge base based on the data quality feature vector and output a suitable processing chain identifier; each processing chain identifier uniquely corresponds to one processing chain. Internally, the module maintains a rich and well-structured strategy knowledge base. This knowledge base contains multiple sets of carefully designed "condition-action" mapping rules, formulated based on extensive practical experience and in-depth analysis of the data quality characteristics of different medical entity types. Simultaneously, to improve the intelligence and adaptability of decision-making, the knowledge base may also deploy advanced intelligent decision-making models, such as machine learning-based classification models or decision tree models. The module takes the data quality feature vector as input and, through a complex matching calculation process with the rules or models in the strategy knowledge base, comprehensively considers various factors of data quality as well as the applicable scenarios and advantages of different processing chains, ultimately outputting the processing chain identifier most suitable for the current data quality status, providing precise guidance for subsequent data processing procedures.

[0035] The chain execution engine module 4, connected to the dynamic processing chain decision module 3, the data quality perception and analysis module 1, and the composable processing chain repository module 2, loads the corresponding processing chain from the composable processing chain repository module 2 based on the processing chain identifier. Then, strictly following the preset execution order of the processing chain, it drives each processor in the processing chain to process the original input data sequentially. During the execution of each processor, the processing status and data changes are monitored in real time to ensure the stability and reliability of the processing process. Simultaneously, to improve the system's fault tolerance and processing efficiency, preset short-circuit conditions are set. Once these short-circuit conditions are triggered during processing, such as encountering serious data errors or processing timeouts, the chain execution engine module 4 will promptly stop the execution of the current processing chain and take corresponding error handling measures, such as recording error information and rolling back data processing operations. Finally, when the processing chain is completed or the preset short-circuit conditions are triggered, the module will promptly output the target data identification result, bringing the system's data processing task to a successful conclusion and providing accurate and reliable data support for subsequent business applications.

[0036] In one embodiment, the data quality perception and analysis module 1 is used to conduct a comprehensive evaluation of the completeness, standardization, validity, and logical anomalies of the original input data. The specific evaluation dimensions are as follows:

[0037] Completeness Assessment: Completeness is one of the fundamental elements of data quality. This module rigorously verifies the existence of key business fields in the original input data. As the core information carriers of data records, key business fields play a decisive role in accurately reflecting business facts. For example, in patient information data, the ID number is a crucial field uniquely identifying the patient; its absence makes it impossible to accurately link the patient's various medical records. In drug information data, the drug approval number is a key credential for the legal marketing of drugs; its absence significantly reduces the completeness and credibility of drug information. Therefore, this module uses preset rules and algorithms to check the existence of each key business field in the data records, accurately determining whether any key information is missing, and promptly marking the location and records of missing fields, providing clear guidance for subsequent data completion and processing.

[0038] Standardization Assessment: This module aims to ensure that data conforms to unified standards and specifications in terms of format, encoding, and value range, thereby improving data consistency and readability. It meticulously validates the content of each field, checking whether it matches preset format standards, encoding rules, or value ranges. For example, different business systems may have different requirements for date representation, such as "YYYY-MM-DD" or "MM / DD / YYYY". This module validates date fields in the data according to preset unified date format standards, correcting or marking dates that do not conform to the standards. For administrative division codes, it checks their accuracy according to national standard administrative division coding rules to avoid data association errors caused by coding mistakes. Furthermore, for fields with specific value ranges, such as gender fields which typically only accept values ​​like "male," "female," or "unknown," this module strictly verifies whether the field values ​​are within the preset value range to ensure data standardization and accuracy.

[0039] Validity Assessment: Validity assessment focuses on determining whether the values ​​of each field fall within a reasonable range according to business logic, ensuring the rationality and usability of the data at the business level. Each business field has its specific business meaning and value range; data outside this range is often incorrect or invalid. For example, the age field is an important indicator reflecting a patient's physiological state, and its value should meet the requirement of being greater than 0 and less than 150. If an age is negative or exceeds the reasonable upper limit, it clearly does not conform to business logic. This module will promptly identify and mark these invalid data, reminding data processing personnel to conduct further verification and correction. By assessing the validity of each field's values, obviously erroneous data can be effectively filtered out, improving data quality and reliability.

[0040] Logical Anomaly Assessment: Logical anomaly assessment is a crucial method for discovering potential contradictions and errors in data. This module employs advanced algorithms and logical rules to identify logical conflicts or pattern anomalies within the data. In complex business data, different fields often possess certain logical relationships. If these relationships are disrupted, logical anomalies will occur. For example, a gender field might be labeled as male, but the corresponding disease diagnosis might be cervicitis. This is clearly a logical contradiction, as cervicitis is a common female reproductive system disease. This module analyzes and compares the logical relationships between fields in the data to promptly identify such logical anomalies, recording and marking them in detail. This provides data processing personnel with targeted investigation directions, ensuring the logical consistency and accuracy of the data.

[0041] In one embodiment, the quality assessment of the original input data of a medical entity type to generate a data quality feature vector includes:

[0042] Based on the medical entity type corresponding to the original input data (covering common types such as drugs, patients, and medical records), specific quality assessment rules (including dimension definitions, verification logic, and quantitative standards) are extracted from the pre-built rule base to adapt to the entity type.

[0043] Based on the extracted quality assessment rules, the data quality assessment results for each key attribute dimension of the original input data are obtained by evaluating its completeness, standardization, validity, and logical anomalies, and a data quality feature vector is generated.

[0044] In the integrity assessment phase, the system verifies the existence of each key business field one by one, leaving no potentially missing information unchecked. For example, if the ID number field in patient information is not found in the data record or its value is empty, it is considered incomplete. The standardization assessment focuses on whether the field content conforms to preset format standards, coding rules, or value ranges. Taking the date field in medical records as an example, the system checks whether it is recorded according to the uniform "YYYY-MM-DD" format, marking dates that do not meet the format requirements. The validity assessment aims to determine whether the field values ​​are within a reasonable business logic range. For example, in the patient age field, if the value is less than 0 or greater than 150, it clearly does not conform to business logic and is judged as invalid data. The logical anomaly assessment uses complex algorithms and logical rules to identify whether there are logical conflicts or pattern anomalies in the data. For example, if a patient's gender is labeled as male, but the diagnosis result is gynecological disease, this presents a clear logical contradiction, and the system will promptly capture this anomaly.

[0045] In one embodiment, a processing chain adapted to different data quality levels is pre-configured for each type of medical entity.

[0046] In one embodiment, the type of the processing chain includes:

[0047] Precise Identification Chain: This chain targets high-quality data, which is typically complete, accurate, and standardized. Key business fields are intact, data formats conform to standards, values ​​fall within reasonable business ranges, and there are no obvious logical contradictions. The core objective of this processing chain is to achieve rapid and accurate matching and identification. By directly utilizing high-quality data, the Precise Identification Chain can quickly and accurately match the data with existing knowledge bases and databases, rapidly identifying the corresponding medical entities and providing efficient support for subsequent data analysis and applications.

[0048] Semantic Fusion Chain: Designed for medium-quality data, this type of data may have some localized issues, such as missing identifiers, incomplete field formatting, and minor logical ambiguities. The Semantic Fusion Chain focuses on resolving semantic alignment issues arising from missing or ambiguous identifiers. Through semantic analysis and fusion of the data, it uncovers the underlying information, unifying the semantic representation of data from different sources and formats, thus achieving data alignment and integration.

[0049] DeepMindChain: Used to process low-quality data, which typically suffers from serious problems such as missing key fields, disorganized data formats, prominent logical contradictions, and even novel or highly anomalous entity patterns. DeepMindChain is dedicated to identifying and addressing novel or highly anomalous entity patterns. Through in-depth mining and analysis of low-quality data, it uncovers hidden valuable information, explores new medical entities and patterns, and provides new perspectives and ideas for medical research and decision-making.

[0050] In one embodiment, the strategy knowledge base stores processing chain identifiers for different medical entity types and data quality levels; the step of querying the strategy knowledge base to output the appropriate processing chain identifier based on the data quality feature vector includes:

[0051] The data quality level of the current original input data is determined based on the data quality feature vector, and the appropriate processing chain identifier is queried from the strategy knowledge base based on the corresponding medical entity type and data quality level.

[0052] Specifically, the data quality feature vector consists of quantitative evaluation results across four dimensions: completeness, standardization, validity, and logical anomalies. For example, the completeness dimension might include the rate of missing key fields; the standardization dimension might include the format compliance rate; the validity dimension might include the rate of reasonable value selection; and the logical anomaly dimension might include the proportion of logical conflicts. The system generates a comprehensive quality score by weighting and summing the scores of each dimension using a pre-defined weighting model. The current data quality level is determined based on a score range (e.g., ≥85 points indicates high quality, 60-85 points indicates medium quality, and <60 points indicates low quality) or a threshold rule (e.g., downgrading if any dimension falls below a specific threshold). Based on the determined data quality level and the medical entity type (e.g., disease, drug, examination item) of the original input data, query conditions are constructed. These query conditions are input into the strategy knowledge base, and a retrieval mechanism searches for matching processing chain identifiers.

[0053] In one embodiment, such as Figure 2 Each processor in the driving processing chain operates on the original input data, and the final recognition result output includes:

[0054] If a precise identifier chain is loaded, the processors within the chain are driven to perform precise queries based on the original input data and output unique standard entities. Specifically, the precise identifier chain includes a field cleaner, a key full-value matcher, and a result outputter. The field cleaner performs basic cleaning on the original input, removing redundant symbols (such as spaces and newlines), standardizing capitalization, and correcting obvious spelling errors. The key full-value matcher uses the cleaned fields as query keys and performs strict full-value matching in a standard knowledge base (such as a disease classification database or a drug classification database). If a completely identical entry exists in the knowledge base, the corresponding standard entity and its code are returned directly. The result outputter encapsulates the matched standard entity, its code, and the matching confidence score into a structured result.

[0055] If a semantic fusion chain is loaded, the processors within the chain are driven to perform text analysis and arbitration based on the original input data, outputting high-confidence inferred entities. Specifically, the semantic fusion chain includes a field cleaner, a text normalizer, a synonym matcher, an edit distance matcher, and an evidence fusion arbitrator. The field cleaner performs basic cleaning on the original input, removing redundant symbols (such as spaces and line breaks), standardizing capitalization, and correcting obvious spelling errors. The text normalizer converts the cleaned text into standard medical terms (e.g., "heart attack" → "myocardial infarction"), standardizes abbreviations (e.g., "coronary heart disease" → "coronary atherosclerotic heart disease"), and parses compound terms (e.g., "hypertension with diabetes" is split into "hypertension" and "diabetes"). The synonym matcher searches for equivalent expressions of standardized terms in the thesaurus (e.g., synonyms for "myocardial infarction" include "heart attack" and "heart attack"), generating a candidate entity list. The edit distance matcher calculates the edit distance between unmatched terms and knowledge base entries (e.g., the edit distance between "diabetes" and "diabetes" is 1), filtering candidate entities with similarity higher than a threshold (e.g., ≥80%). The evidence fusion arbitrator integrates evidence from multiple sources (such as synonym matching results, edit distance ranking, and contextual information) to arbitrate and select the most likely standard entity. If a conflict occurs, it is marked as "low confidence" or triggers manual review.

[0056] If a deep exploration chain is loaded, it drives the processors within the chain to perform knowledge reasoning and AI prediction to obtain the predicted entity with the highest matching degree. Specifically, the deep exploration chain includes a feature extractor, a machine learning model predictor, and a candidate interpreter. The feature extractor extracts key features (such as symptom combinations, disease patterns, and drug component descriptions) from the original input and converts them into numerical representations that the model can process. The intelligent model predictor performs classification or regression prediction on the numerical features based on pre-trained models (such as deep neural networks and graph models), outputting candidate entities and their probability scores. The candidate interpreter performs interpretability analysis on the model prediction results, generates reasoning paths (such as "because the typical manifestations of symptom A and disease C have a similarity of 95%, the prediction is C"), and marks low-confidence predictions for human review.

[0057] In one embodiment, it is determined whether the predicted entity matching degree obtained by the deep exploration chain is greater than a matching degree threshold. If it is not greater, the predicted entity is sent to the manual review end for review. The prediction result generated by the candidate interpreter includes a clear matching confidence level (e.g., 85%), which is calculated by combining the model prediction probability and the inference path support. The system presets a matching degree threshold (e.g., 80%). If the overall confidence level of the predicted entity is greater than or equal to the threshold, the result is directly output; if it is less than the threshold, the manual review process is triggered.

[0058] Through this mechanism, DeepChain maintains the efficiency of intelligent reasoning while ensuring that low-confidence prediction results are rigorously reviewed by human experts, and the final output results meet medical-grade standards in terms of accuracy and reliability.

[0059] To better describe the dynamic entity recognition system based on data quality awareness, the following specific embodiments are provided.

[0060] Example 1: Method for identifying a medical drug standard library. For example... Figure 3 This is a flowchart illustrating the medical drug standard library identification method in this embodiment.

[0061] This embodiment takes the identification and matching of the standard entity "drug" as an example, aiming to accurately match inconsistent drug "raw material" data collected from different Hospital Information Systems (HIS) into a unified drug master data (standard library). Drug raw material data typically includes the following key attribute dimensions: drug name, specifications, manufacturer, packaging, and dosage form.

[0062] Step 1.1: Define the data quality assessment dimensions and calculation rules.

[0063] We have carefully defined quality assessment rules for these five core dimensions of drug data.

[0064] To check the completeness of the drug name, check if the "Drug Name" field is empty. If the field is empty, the quality is judged as "poor"; if the generic name is present, such as "Aspirin", the quality is "medium"; if the complete brand name is present, such as "Bayer Aspirin Enteric-coated Tablets", the quality is judged as "excellent".

[0065] Regarding the standardization of key labels, check whether the combination of "specification" and "dosage form" constitutes a valid and resolvable label, such as "0.1g*30 tablets / box". If it can be clearly resolved into a "content-quantity-unit / packaging" structure, the quality is "excellent"; if the structure is ambiguous, it is "medium"; if it is missing, it is "poor".

[0066] To assess the credibility of vendor information, we check if the "vendor" field exists in a pre-defined whitelist of authoritative vendors. If it is on the list and the name matches exactly, the quality is "Excellent"; if it is an abbreviation or has a common alias, the quality is "Medium"; if it is an unfamiliar name or is missing, the quality is "Poor".

[0067] To ensure overall information consistency, we will perform cross-field logical validation on the "packaging" description. For example, if the dosage form is "tablets" but the packaging unit is described as "tube", it will be marked as "information conflict".

[0068] Step 1.2: Build a multi-strategy processing chain repository.

[0069] For drug matching scenarios, three processing chains with different strategies and complexities are pre-defined. Chain A is the precise identification chain, whose processing flow is: key field cleaner - specification - dosage form full value matcher - manufacturer name precise matcher. It is suitable for processing "high-quality" data, relying on highly standardized "specification + dosage form" as the primary key, combined with manufacturer information for fast and accurate database lookup. Chain B is the semantic fusion chain, whose processing flow is: drug name standardizer - generic name synonym matcher - specification fuzzy parsing matcher - multi-evidence confidence arbitrator. It mainly handles "medium-quality" data. When key identifiers are not standardized, it parses the semantics of drug names, such as comparing "cefotaxime sodium" and "rocephin", performs fault-tolerant parsing of specifications, such as the conversion between "2g" and "2000mg", and integrates evidence from multiple dimensions for comprehensive judgment. Chain C is the deep exploration chain, whose processing flow is: multi-field feature vector generator - drug knowledge graph relationship finder - AI-based similarity ranking model - manual review and early warning system. It is suitable for processing data of "poor" quality or "conflicting information". When the data is extremely incomplete or contradictory, it is converted into feature vectors, and potential correlations are searched in the drug knowledge graph. Alternatively, a complex AI model can be invoked to make the most probable inferences, and a human final confirmation can be prompted.

[0070] Step 1.3: Configure the dynamic decision strategy library.

[0071] A simple decision-making logic was established, mapping quality assessment results to processing chains. Rule R1 stipulates that if the key identifier standardization is "Excellent" and the manufacturer information credibility is "Excellent / Medium", then chain A (precise identifier chain) is selected. Rule R2 indicates that if the key identifier standardization is "Medium" and the drug name completeness is "Excellent / Medium", then chain B (semantic fusion chain) is selected. Rule R3 states that if there is a conflict in overall information consistency, or if two of the first three core quality dimensions are assessed as poor, then chain C (deep exploration chain) is selected.

[0072] During runtime dynamic matching, assume the system receives two sets of pharmaceutical raw material data from different sources and with significantly different quality. For request A (high-quality data - triggering the precise chain), its raw material data is {Drug Name: “Bayer Aspirin Enteric-coated Tablets”, Specification: “100mg 30 tablets”, Manufacturer: “Bayer Pharmaceuticals”, Packaging: “Box”, Dosage Form: “Enteric-coated Tablets”}. After the quality perception analysis module runs, a quality profile is generated: the completeness of the drug name is excellent (complete brand name), the standardization of key identifiers is excellent (“100mg 30 tablets” and “enteric-coated tablets” can be clearly parsed), the credibility of the manufacturer information is excellent (“Bayer Pharmaceuticals” is in the whitelist), and the overall information consistency is normal. The dynamic decision module matches rule R1 (excellent identifier and credible manufacturer) and outputs decision selection chain A. During the intelligent execution phase, the engine loads chain A, the key field cleaner unifies "specification" to "100mg*30 tablets / box", the specification-dosage form full value matcher directly finds the unique corresponding item in the standard library, and the manufacturer name precise matcher passes the verification. The entire process is completed within 10 milliseconds without activating any synonyms, fuzzy matching, or AI models, achieving optimal resource utilization.

[0073] For request B (low-quality and conflicting data - triggering a deep exploration chain), its raw material data is {Drug Name: "Saline", Specification: "500 ml", Manufacturer: "A Pharmaceutical Company", Packaging: "Bottle", Dosage Form: "Injection"}. Quality perception generates a quality profile: Drug name completeness is poor ("Saline" is a vague colloquial term), key identifier standardization is medium ("500 ml" is parsable, but not combined with the dosage form "Injection" to form a unique identifier), manufacturer information credibility is poor ("A Pharmaceutical Company" is not in the whitelist), and overall information consistency is conflicting (the dosage form is "Injection," but "Saline" usually refers to "Sodium Chloride Injection" or "Glucose Injection," which is ambiguous). The dynamic decision module matches rule R3 (information conflict and multi-dimensional quality poor), outputting decision selection chain C. In the intelligent execution phase, the engine loads chain C, and the multi-field feature vector generator converts "saline, 500 ml, injection, bottle" into features. The drug knowledge graph relationship lookup tool finds that both "sodium chloride injection" and "glucose injection" are "injection solutions," and common specifications include "500 ml / bottle." The AI ​​similarity ranking model integrates all weak signals of the raw materials and may calculate that "0.9% sodium chloride injection" has the highest matching probability. Finally, the manual review and warning device submits the matching result (such as "0.9% sodium chloride injection (500 ml: bottle)") along with low confidence markers and conflict explanations to a human pharmacist for final review and confirmation. The system automatically uses the most powerful and complex analysis capabilities for this problematic data, instead of hastily matching or directly abandoning it.

[0074] This embodiment demonstrates how the present invention successfully transforms abstract quality perception into concrete, differentiated processing strategies. Regarding precise resource allocation, the lowest-cost, precise query is used for high-quality, complete information like "Bayer Aspirin," effectively freeing up system resources. For intelligent problem escalation, the processing intensity is automatically increased for low-quality, conflicting "saline" data, utilizing knowledge graphs and AI models to transform the problem into a manageable, human-attentive task, rather than a simple "matching failure." In terms of transparency in the decision-making process, from quality profiling (clearly presenting why the data quality is considered poor), to the decision basis (clearly explaining why chain C was chosen), and then to the intermediate results of each processor within the chain (such as which candidates the knowledge graph found), the entire decision chain is clear, explainable, and auditable. This mechanism ensures that the system maintains high efficiency while handling massive, heterogeneous medical data, and possesses powerful processing capabilities for complex, ambiguous, long-chain problems—the core advantage of this invention compared to static processes. Furthermore, this architecture can be seamlessly extended to matching scenarios for various master data in the medical field, such as patient, surgical, and diagnostic data.

[0075] Similar to the principles of the above embodiments, the present invention provides a dynamic entity recognition method based on data quality awareness.

[0076] The following specific embodiments are provided in conjunction with the accompanying drawings:

[0077] like Figure 4 This document illustrates a flowchart of a dynamic entity recognition method based on data quality awareness, as described in an embodiment of the present invention. Applied to the aforementioned dynamic entity recognition system based on data quality awareness, the method includes:

[0078] Step S1: Perform quality assessment on the original input data of a medical entity type and generate a data quality feature vector.

[0079] In one embodiment, the quality assessment rules corresponding to the medical entity types based on the original input data are extracted; based on the extracted quality assessment rules, data quality assessments are performed on each key attribute dimension of the original input data to obtain the data quality assessment results for each key attribute dimension, and a data quality feature vector is generated.

[0080] In one embodiment, the original input data is evaluated for completeness, normalization, validity, and logical anomalies.

[0081] Step S2: Based on the data quality feature vector, query the strategy knowledge base to output the appropriate processing chain identifier.

[0082] In one embodiment, the step of querying the strategy knowledge base to output the appropriate processing chain identifier based on the data quality feature vector includes: determining the data quality level of the current original input data based on the data quality feature vector, and querying the strategy knowledge base for the appropriate processing chain identifier based on the corresponding medical entity type and data quality level.

[0083] In one embodiment, a processing chain adapted to different data quality levels is pre-configured for each type of medical entity.

[0084] In one embodiment, the processing chain types include: a precise identification chain for processing high-quality data, a semantic fusion chain for processing medium-quality data, and a deep exploration chain for processing low-quality data.

[0085] Step S3: Based on the processing chain identifier, drive each processor in the processing chain to operate on the original input data and output the final recognition result.

[0086] In one embodiment, each processor in the driving processing chain operates on the original input data and outputs the final recognition result, including: if a precise identification chain is loaded, then each processor in the precise identification chain is driven to perform precise queries based on the original input data to output a unique standard entity; if a semantic fusion chain is loaded, then each processor in the semantic fusion chain is driven to perform text analysis and arbitration based on the original input data to output a high-confidence inferred entity; if a deep exploration chain is loaded, then each processor in the deep exploration chain is driven to perform knowledge reasoning and AI prediction to obtain the predicted entity with the highest matching degree.

[0087] In one embodiment, it is determined whether the predicted entity matching degree obtained by the deep exploration chain is greater than the matching degree threshold. If it is not greater, the predicted entity is sent to the manual review terminal for review.

[0088] Since the implementation principle of the dynamic entity recognition method based on data quality awareness has been described in the foregoing embodiments, it will not be repeated here.

[0089] The dynamic entity recognition system based on data quality awareness provided in this invention can be implemented on the terminal side or the server side. Regarding the hardware structure of the electronic terminal, please refer to [link to relevant documentation]. Figure 5 This is a schematic diagram of an optional hardware structure of an electronic terminal 1000 provided in an embodiment of the present invention. The terminal 1000 can be a mobile phone, computer device, tablet device, personal digital processing device, factory back-end processing device, etc. The terminal 1000 includes: at least one processor 1001, a memory 1002, at least one network interface 10010, and a user interface 1009. The various components in the device are coupled together through a bus system 1005. It is understood that the bus system 1005 is used to realize the connection and communication between these components. In addition to a data bus, the bus system 1005 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 5 All buses are labeled as bus systems.

[0090] The user interface 1009 may include a monitor, keyboard, mouse, trackball, clicker, button, touchpad, or touch screen.

[0091] It is understood that memory 1002 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM) or programmable read-only memory (PROM), which serves as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM) and synchronous static random access memory (SSRAM). The memories described in the embodiments of this invention are intended to include, but are not limited to, these and any other suitable categories of memory.

[0092] In this embodiment of the invention, the memory 1002 is used to store various types of data to support the operation of the terminal 1000. Examples of this data include: any executable program for operation on the terminal 1000, such as the operating system 10021 and application programs 10022; the operating system 10021 includes various system programs, such as the framework layer, core library layer, driver layer, etc., for implementing various basic services and handling hardware-based tasks. The application program 10022 may include various applications, such as media players, browsers, etc., for implementing various application services. The implementation of the data quality-aware dynamic entity recognition system provided in this embodiment of the invention can be included in the application program 10022.

[0093] The methods disclosed in the above embodiments of the present invention can be applied to or implemented by the processor 1001. The processor 1001 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in the processor 1001 or by instructions in the form of software. The processor 1001 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor 1001 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor 1001 may be a microprocessor or any conventional processor, etc. The steps of the accessory optimization method provided in the embodiments of the present invention can be directly reflected as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium, which is located in a memory. The processor reads the information in the memory and combines it with its hardware to complete the steps of the aforementioned method.

[0094] In an exemplary embodiment, the terminal 1000 may be used to execute the aforementioned method by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), or complex programmable logic devices (CPLDs).

[0095] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented using computer program-related hardware. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0096] In the embodiments provided in this application, the computer-readable and writable storage medium may include read-only memory, random access memory, EEPROM, CD-ROM or other optical disc storage devices, disk storage devices or other magnetic storage devices, flash memory, USB flash drive, portable hard drive, or any other medium capable of storing desired program code in the form of instructions or data structures and accessible by a computer. Additionally, any connection may be appropriately referred to as a computer-readable medium. For example, if instructions are transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. However, it should be understood that computer-readable and writable storage media and data storage media do not include connections, carrier waves, signals, or other transient media, but are intended for non-transient, tangible storage media. The disks and optical discs used in the application include compact optical discs (CDs), laser optical discs, optical discs, digital multifunction optical discs (DVDs), floppy disks, and Blu-ray discs, where disks typically copy data magnetically, while optical discs use lasers to copy data optically.

[0097] Compared with the prior art, the present invention has the following advantages:

[0098] 1. Significant improvement in resource efficiency.

[0099] This invention transforms a one-size-fits-all, fixed process into a dynamic, tailored selection, enabling the system to precisely allocate valuable computing resources (especially expensive AI / ML computations) to the low-quality data that needs them most. For the abundant high-quality data, the system can complete the process quickly at extremely low cost (such as simple rule matching), thereby significantly reducing the average processing time and increasing overall throughput. Experimental data shows that in medical data matching scenarios, this invention can reduce the processing time for high-quality data (60%) by more than 70%, and increase overall throughput by approximately 40%.

[0100] 2. Improved recognition accuracy and system robustness.

[0101] The dynamic selection mechanism of this invention ensures that every piece of data, regardless of its quality, receives the appropriate processing intensity. Low-quality data is not processed hastily, but is guided to a processing chain containing more in-depth analysis algorithms, increasing the probability of successful matching and effectively reducing the missed match rate caused by poor-quality data.

[0102] 3. A leap forward in system intelligence and self-adaptation capabilities.

[0103] This invention transforms the system from passively executing preset processes into an intelligent agent that actively perceives, analyzes, and makes decisions. This perception-decision capability is a key foundation for building self-optimizing and adaptive systems.

[0104] 4. Excellent architectural flexibility and maintainability.

[0105] This invention features a clear modular design. The "Quality Perception Module," "Decision Module," and "Processing Chain Repository" can all be independently expanded and replaced. For example, the decision module can be easily upgraded from rule-based to machine learning model-based without modifying other parts of the system.

[0106] In summary, the dynamic entity recognition system, method, and terminal based on data quality awareness of the present invention assesses the quality of raw input data for specific medical entity types through a data quality awareness and analysis module, generating feature vectors that accurately reflect the data quality status. The composable processing chain repository module pre-sets multiple processing chains with different processing strategies for each medical entity type. Each processing chain consists of multiple processors arranged in a specific execution order to meet diverse data processing needs. The dynamic processing chain decision module queries the strategy knowledge base based on the data quality feature vectors and accurately outputs the appropriate processing chain identifier, with each identifier uniquely corresponding to one processing chain. The chained execution engine module retrieves the corresponding processing chain from the dynamic processing chain decision module based on the processing chain identifier and drives each processor in the processing chain to process the raw input data sequentially, ultimately outputting an accurate recognition result. This invention can intelligently select and execute the processing chain most suitable for the quality characteristics of each input data, achieving optimal allocation of computing resources and maximizing the success rate of recognition tasks. Therefore, this invention effectively overcomes the various shortcomings of existing technologies and has high industrial application value.

[0107] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A dynamic entity recognition system based on data quality awareness, characterized in that, include: The data quality perception and analysis module is used to assess the quality of raw input data of a medical entity type and generate a data quality feature vector. The composable processing chain repository module is used to pre-configure multiple processing chains with different strategies for each type of medical entity; each processing chain consists of multiple processors arranged in execution order. The dynamic processing chain decision module is used to query the strategy knowledge base based on the data quality feature vector and output the appropriate processing chain identifier; wherein, each processing chain identifier corresponds to a processing chain. The chain execution engine module is used to load the corresponding processing chain from the dynamic processing chain decision module based on the processing chain identifier, drive each processor in the processing chain to operate on the original input data, and output the final recognition result.

2. The dynamic entity recognition system based on data quality awareness as described in claim 1, characterized in that, The process of performing quality assessment on the original input data of a medical entity type and generating a data quality feature vector includes: Based on the medical entity types in the original input data, extract the corresponding quality assessment rules; Based on the extracted quality assessment rules, data quality assessments are performed on each key attribute dimension of the original input data to obtain the data quality assessment results for each key attribute dimension, and a data quality feature vector is generated.

3. The dynamic entity recognition system based on data quality awareness as described in claim 1, characterized in that, Pre-configure processing chains for each type of medical entity to suit different data quality levels.

4. The dynamic entity recognition system based on data quality awareness as described in claim 3, characterized in that, The types of processing chains include: a precise identification chain for processing high-quality data, a semantic fusion chain for processing medium-quality data, and a deep exploration chain for processing low-quality data.

5. The dynamic entity recognition system based on data quality awareness according to claim 4, characterized in that, The processing chain identifiers that are adapted to the data quality feature vector query strategy knowledge base output include: The data quality level of the current original input data is determined based on the data quality feature vector, and the appropriate processing chain identifier is queried from the strategy knowledge base based on the corresponding medical entity type and data quality level.

6. The dynamic entity recognition system based on data quality awareness according to claim 4, characterized in that, Each processor in the driving processing chain operates on the original input data, and the final recognition result output includes: If a precise identifier chain is loaded, then each processor in the precise identifier chain will perform a precise query based on the original input data to output a unique standard entity. If a semantic fusion chain is loaded, then each processor in the semantic fusion chain is driven to perform text analysis and arbitration based on the original input data, and output a high-confidence inferred entity. If a deep exploration chain is loaded, it drives each processor in the deep exploration chain to perform knowledge reasoning and AI prediction to obtain the predicted entity with the highest matching degree.

7. The dynamic entity recognition system based on data quality awareness as described in claim 6, characterized in that, The system also determines whether the matching degree of the predicted entity obtained by the deep exploration chain is greater than the matching degree threshold. If it is not greater, the predicted entity is sent to the manual review end for review.

8. The dynamic entity recognition system based on data quality awareness as described in claim 1, characterized in that, The data quality perception and analysis module is used to evaluate the completeness, standardization, validity, and logical anomalies of the original input data.

9. A dynamic entity recognition method based on data quality awareness, characterized in that, The method, applied to any one of claims 1 to 8, for a data quality-aware dynamic entity recognition system, comprises: Perform quality assessment on the original input data of a medical entity type and generate a data quality feature vector; Based on the data quality feature vector query strategy knowledge base, the corresponding processing chain identifier is output; The corresponding processing chain is loaded based on the processing chain identifier, and each processor in the processing chain is driven to operate on the original input data to output the final recognition result.

10. An electronic terminal, characterized in that, include: One or more memories and one or more processors; The one or more memories are used to store computer programs; The one or more processors are connected to the memory and are used to run the computer program to perform the method as described in claim 9.

Images