A method and system for generating a rational drug use report combined with artificial intelligence

Abstract

The present invention relates to the technical field of data processing and report generation, and particularly relates to a method and system for generating a reasonable medication report combined with artificial intelligence. The method includes the following steps: selecting a report category; constructing a report parsing model and parsing report data; using a domain knowledge graph to enhance the parsing of report data; generating a structured report; optimizing the algorithms for report parsing and report generation according to the feedback information of users; using two rich text editing areas to edit the data part and the conclusion part of the report; eliminating the report content that does not need to be displayed; and exporting the final edited result. The present invention combines artificial intelligence technology to automatically parse report data and generate a report, thereby improving the parsing efficiency and accuracy; by introducing a self-learning mechanism, optimizing the algorithms for report parsing and generation, and enhancing the quality of report generation; by providing a friendly interface and rich editing functions, adapting to the personalized needs of different users, and having a wide application prospect in the medical field.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing and report generation, and particularly to a method and system for generating a rational drug use report combined with artificial intelligence. Background Art

[0002] In the information age, reports are crucial for the smoothness of business and the quality of decision-making. In the medical field, a large number of complex reports are generated every day, but the traditional manual generation method is time-consuming and error-prone, especially inefficient when dealing with a large amount of complex data. Therefore, realizing the automated and intelligent generation of reports to improve efficiency and accuracy has become a problem that urgently needs to be solved.

[0003] In the prior art, reports are usually generated by using preset templates and fixed rules. Such methods often lack intelligent and automated processing capabilities, do not have the ability to automatically analyze complex report data and understand its connotations, and cannot dynamically generate reports with clear structures and accurate content based on the specific content of the data. In addition, such methods often lack generality and are difficult to flexibly adapt to the operating habits of different users and the report generation requirements. For example, in the medical field, different doctors or departments may need to generate reports with different formats and contents according to their own needs, while the existing report generation methods often cannot meet these personalized needs. Therefore, there is an urgent need to propose a method that can automatically analyze report data, generate structured reports, and has intelligent and automated capabilities.

[0004] Aiming at the deficiencies of the prior art, the present invention proposes a method for generating a rational drug use report combined with artificial intelligence. By using artificial intelligence technology, this method can automatically analyze report data and generate reports, thereby improving the analysis efficiency and accuracy; by introducing a self-learning mechanism, it optimizes the report analysis and generation algorithms to improve the quality of report generation; by providing a friendly interface and rich editing functions, it adapts to the personalized needs of different users and meets the wide range of needs in the medical field. Summary of the Invention

[0005] Aiming at the defects in the prior art, the present invention provides a method and system for generating a rational drug use report combined with artificial intelligence.

[0006] In a first aspect, the present invention provides a method for generating a rational drug use report in combination with artificial intelligence, comprising: selecting a report category for a report to be generated; constructing a report parsing model based on the report category; parsing report data through the report parsing model to obtain basic parsing data; using a domain knowledge graph to perform enhanced parsing on the basic parsing data to obtain enhanced parsing data; generating a structured report based on the enhanced parsing data; based on the structured report, optimizing the report parsing and report generation algorithms according to user feedback information to obtain a structured improved report; using two rich text editing areas to edit the data part and the conclusion part of the structured improved report to obtain an editing result, the rich text editing area supports pre-editing and saving rich text content as a template, the template including a private template and a shared template; eliminating content that does not need to be displayed in the editing result to obtain a final editing result; exporting the final editing result to obtain a rational drug use report in combination with artificial intelligence. The present invention can generate high-quality structured reports by accurately parsing report data and strengthening the analysis by using domain knowledge graphs; the algorithm is continuously optimized through user feedback, and the report content is further improved to meet diverse needs; the editing convenience and personalized customization capabilities of the report are enhanced by introducing two rich text editing areas and supporting flexible configuration of private and shared templates; the final editing results are refined and accurate by eliminating content that does not need to be displayed; the rational drug use report obtained by the export function can not only effectively integrate artificial intelligence technology, but also show a high degree of readability and practicality, thereby providing support for clinical decision-making and promoting the improvement of rational drug use levels.

[0007] Optionally, the construction of a report parsing model based on the report category includes: preprocessing the report data based on the report category to obtain optimized report data; constructing a report parsing model based on the optimized report data, wherein the report parsing model includes a report intelligence model and a report deep analysis model. The present invention obtains optimized report data with clear structure and accurate content by preprocessing the report data based on the report category, thereby providing a reliable basis for subsequent analysis; by using the optimized report data to construct a report parsing model, the integrated application of the report intelligence model and the report deep analysis model is realized. The report intelligence model can accurately identify and understand the key information in the report, while the report deep analysis model can deeply explore the intrinsic connection and potential value between the data, thereby comprehensively improving the depth and breadth of report analysis.

[0008] Optionally, parsing the report data through the report parsing model to obtain basic parsing data includes: parsing the text data in the optimized report data through the report intelligent language model to obtain basic text parsing data, where the text parsing includes semantic analysis, named entity recognition, and context relationship understanding; parsing the numerical data in the optimized report data through the report in-depth analysis model to obtain basic numerical parsing data, where the numerical parsing includes feature extraction, pattern recognition, and predictive analysis. The present invention parses the text data in the optimized report data through the report intelligent language model to obtain basic text parsing data containing rich semantic information, accurate named entities, and clear context relationships, providing support for the comprehensive understanding of the report content; parses the numerical data in the optimized report data through the report in-depth analysis model, extracts key features, identifies data patterns, and conducts predictive analysis of potential associations, thereby obtaining basic numerical parsing data, providing a basis for the quantitative analysis and trend prediction of the data.

[0009] Optionally, the report intelligent language model satisfies the following expression:

[0010]

[0011] Wherein, represents the explanatory text of the report data, represents the natural language parsing function, represents the original text data in the report, represents the lexical features of the data in the report, represents the syntactic tree, represents the noise reduction function, is the context information of the report, is the domain knowledge; the report in-depth analysis model satisfies the following expression:

[0012]

[0013] Wherein, represents the deep parsing result, represents the deep learning model function, represents the original data set in the report, represents the feature extraction function, represents the data preprocessing function, represents the unstructured data, Indicates the data type. Through the report intelligent language model of the present invention, the original text data in the report is converted into accurate explanatory text, making full use of the in-depth analysis of lexical features and syntactic trees by natural language parsing functions, and combining context information and domain knowledge to reduce noise interference, so as to ensure that the report content is accurately and comprehensively understood and expressed; through the report in-depth analysis model, key information is extracted from the original report dataset to construct a deep analysis result. This process relies on the learning ability of the deep learning model, improves data quality through feature extraction functions, and balances data preprocessing functions, combined with an optimized neural network architecture and training parameters, to achieve precise parsing and in-depth mining of report data.

[0014] Optionally, generating a structured report based on the enhanced parsed data includes: selecting a report template according to the enhanced parsed data; filling the enhanced parsed data based on the report template; verifying the enhanced parsed data to obtain a verification result; and generating a structured report through the verification result. Through the enhanced parsed data and selection of a report template, the present invention ensures the matching of the report structure and data content, which not only improves the professionalism and readability of the report, but also enables the report to more accurately reflect the information and trends behind the data; filling the enhanced parsed data based on the selected report template can achieve precise positioning and effective display of the data; verifying the enhanced parsed data can obtain a reliable verification result, providing a guarantee for generating a structured report.

[0015] Optionally, optimizing the algorithms for report parsing and report generation based on the structured report according to the user's feedback information to obtain a structured refined report includes: collecting the user's feedback information based on the structured report; forming a feedback analysis report according to the feedback information; determining the algorithm optimization requirements for report parsing and report generation through the feedback analysis report to form a list of algorithm optimization requirements; designing an algorithm optimization plan according to the list of algorithm optimization requirements; obtaining the optimized algorithms for report parsing and report generation by using the algorithm optimization plan; and obtaining a structured refined report through the optimized algorithms. By collecting the user's feedback information, the present invention obtains the direct evaluation and suggestions of the user on the effects of report parsing and report generation, providing a basis for algorithm optimization; forming a feedback analysis report based on the feedback information and determining the optimization requirements of the algorithms for report parsing and report generation accordingly, clarifying the specific goals and directions for algorithm improvement, and forming a clear list of algorithm optimization requirements, thereby improving the pertinence and efficiency of the optimization work; by designing an algorithm optimization plan and obtaining the optimized algorithms for report parsing and report generation, the optimized algorithms improve the accuracy of report parsing and the intelligence of report generation by making targeted improvements to the problems feedback by the user while retaining the advantages of the original algorithms; the structured refined report generated by applying the optimized algorithms improves the quality and practicality of the report.

[0016] In a second aspect, a rational drug use report generation system integrated with artificial intelligence provided by the present invention includes a report selection module, a report parsing module, a report generation module, a self-learning module, a rich text editing module, a reset and exclusion module, an export module, and a template management module; the report parsing module is connected to the report selection module, the self-learning module, and the report generation module; the report generation module is connected to the report parsing module, the self-learning module, the rich text editing module, the reset and exclusion module, the export module, and the template management module; the report selection module is used to select the report category for which a report needs to be generated; the report parsing module is used to receive the report category information transmitted by the report selection module and transmit the parsed report data to the report generation module and the self-learning module; the report generation module is used to receive the parsed data and generate a final report in combination with the rich text content edited by the user, template information, and reset instructions; the self-learning module is used to collect user feedback and operation data to improve the parsing and generation algorithms; the rich text editing module is used to transmit the rich text content edited by the user to the report generation module; the reset and exclusion module is used to delete specific content in the report according to user instructions; the export module is used to export the generated report as a PDF or Word document, and a custom watermark can be added to the PDF document; the template management module is used to provide rich text information pre-edited and saved by the user. By using intelligent report parsing and automatic generation technology, the present invention can ensure the accuracy and standardization of drug use reports, accelerate the report generation process, and improve its quality; by means of the self-learning module, the system can continuously optimize the parsing and generation algorithms based on user feedback and actual operation data, achieve self-adjustment and improvement, and thus continuously improve the intelligence level of report generation; through comprehensive editing and export functions, the rich text content can be freely edited, templates can be flexibly selected, and specific parts of the report can be reset or deleted immediately, meeting the diverse needs of users and enhancing the practicality of the report and the convenience of operation.

[0017] Optionally, the report parsing module includes a data preprocessing unit, a feature extraction unit, and a semantic understanding unit; the data preprocessing unit is used to clean and format the report data; the feature extraction unit uses deep learning technology to extract key information in the report; the semantic understanding unit uses natural language processing technology to analyze the logical relationship and context meaning between data. Through the data preprocessing unit, the present invention can efficiently clean and format the report data, ensuring the accuracy and consistency of the data, laying a foundation for subsequent feature extraction and semantic understanding, and improving the efficiency and accuracy of the entire report parsing process; through the feature extraction unit, the key information in the report can be accurately extracted, and the key information is crucial for the rational drug use decision-making; through the semantic understanding unit, the logical relationship and context meaning between data can be deeply understood, so as to accurately analyze the complex information in the report, avoid decision-making mistakes caused by information misunderstanding or omission, provide a more reliable and intelligent report parsing service for rational drug use, and enhance the practicality and intelligent level of the system.

[0018] Optionally, the self-learning module adopts a reinforcement learning algorithm and is optimized through the following steps:

[0019] Monitor and record the feedback behaviors of the user during use, including editing and modification, content deletion, and report regeneration requests; analyze the feedback behaviors to identify the key factors affecting the report quality and user experience, and obtain an analysis result; according to the analysis result, adjust the deep learning model parameters and natural language processing strategies. The present invention can continuously monitor and record various feedback behaviors of the user during use, including editing and modification, content deletion, and report regeneration requests, so as to accumulate rich user interaction data and provide a basis for subsequent analysis and optimization; by deeply analyzing these feedback behaviors, the key factors affecting the report quality and user experience, such as information missing, logical errors, or format non-standardization, can be accurately identified, and the corresponding analysis result can be obtained, providing a direction and basis for optimizing the report generation algorithm; through the analysis result, the deep learning model parameters and natural language processing strategies can be automatically adjusted, continuously optimizing the algorithms for report parsing and generation, thereby improving the accuracy and readability of the report, meeting the growing personalized needs of users, and enhancing the intelligent level and user satisfaction of the system.

[0020] Optionally, the template management module includes a version control unit and a permission management unit; the version control unit allows users to track and compare templates of different versions; the permission management unit sets access and editing permissions for private templates and shared templates according to user roles. Through the version control unit of the present invention, templates of different versions can be easily tracked and compared, the iterative process of templates can be effectively managed, the accuracy and consistency of template content can be ensured, and it is also convenient to trace back historical versions, quickly restore or draw on previous designs, improving work efficiency and the flexibility of template management; through the permission management unit, strict access and editing permissions are set according to user roles to protect the security and privacy of private templates, prevent unauthorized access and modification, and promote the reasonable utilization of shared templates. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a flowchart of a method for generating a rational drug use report combined with artificial intelligence according to an embodiment of the present invention;

[0022] Figure 2 is a schematic structural diagram of a rational drug use report generation system combined with artificial intelligence according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] The specific embodiments of the present invention will be described in detail below. It should be noted that the embodiments described here are only for illustrative purposes and are not used to limit the present invention. In the following description, a large number of specific details are set forth in order to provide a thorough understanding of the present invention. However, it will be apparent to those of ordinary skill in the art that: the present invention does not have to be practiced with these specific details. In other instances, well-known circuits, software, or methods have not been specifically described in order to avoid obscuring the present invention.

[0024] Throughout the specification, references to "one embodiment", "an embodiment", "one example", or "an example" mean that a particular feature, structure, or characteristic described in connection with the embodiment or example is included in at least one embodiment of the present invention. Thus, the phrases "in one embodiment", "in an embodiment", "one example", or "an example" appearing throughout the specification do not necessarily all refer to the same embodiment or example. Furthermore, the particular features, structures, or characteristics may be combined in any suitable combination and / or sub-combination in one or more embodiments or examples. In addition, those of ordinary skill in the art should understand that the diagrams provided herein are for illustrative purposes only and are not necessarily drawn to scale.

[0025] Please refer to Figure 1 , an embodiment of the present invention provides a method for generating a rational drug use report combined with artificial intelligence, and the method includes the following steps:

[0026] S1. Select the report category for which the report is to be generated.

[0027] In one embodiment, the report category to be generated is selected through the following steps:

[0028] S11. Define the purpose and requirements of the report.

[0029] In this embodiment, first, define the use of the report, whether it is for internal decision-making, external reporting, or other specific requirements.

[0030] Furthermore, understand who the audience of the report is and what data and information they will be interested in.

[0031] S12. Evaluate the applicability of the report category.

[0032] In this embodiment, review the existing reports on rational drug use, and understand their categories, structures, and contents.

[0033] Furthermore, determine the report category to be generated according to the purpose of the report and the needs of the audience.

[0034] S13. Select the specific report template and format.

[0035] In this embodiment, refer to the standard report templates in the industry.

[0036] Furthermore, customize the format and layout of the report according to specific requirements, ensure that the report is easy to read and understand, and at the same time contains all necessary information.

[0037] S14. Verify and optimize the report category.

[0038] In this embodiment, review the selected report to ensure the accuracy and integrity of the report.

[0039] Furthermore, collect feedback from the audience, and understand their satisfaction with the report and suggestions for improvement.

[0040] Furthermore, optimize and adjust the report category according to the feedback.

[0041] S2. Based on the report category, construct a report parsing model.

[0042] Among them, S2 further includes the following steps:

[0043] S21. Preprocess the report data based on the report category to obtain optimized report data.

[0044] In one embodiment, based on the report category selected in step S1, preprocess the report data, and the process of the preprocessing is as follows:

[0045] S211. Define the preprocessing objective.

[0046] In this embodiment, according to the specific requirements of the report category, the data quality requirements are clarified, including data integrity, accuracy, consistency, and timeliness.

[0047] Furthermore, according to the data quality requirements, the data preprocessing tasks that need to be carried out are identified, such as missing value processing, outlier processing, and data format conversion.

[0048] S212. Data cleaning.

[0049] In this embodiment, the missing values in the data are processed.

[0050] Specifically, identify and count the positions and quantities of the missing values; according to business logic or statistical methods, select appropriate filling strategies, such as using the average value, median, mode, or model-based predicted values to fill the missing values; for the missing values that cannot be reasonably filled, consider deleting the relevant records or performing other processing.

[0051] Furthermore, the outlier values in the data are processed.

[0052] Specifically, use statistical methods or business logic to identify the outlier values; according to the actual situation, correct, delete, or handle the outlier values as special cases.

[0053] S213. Data conversion and formatting.

[0054] In this embodiment, the data is converted and formatted.

[0055] Specifically, ensure that the data types of all fields are consistent with the requirements of the report category, such as converting the string type to the numerical type, or converting the date type to a unified format; perform format standardization processing on date, time, and currency fields to ensure data consistency and readability; for categorical variables, consider using one-hot encoding, label encoding methods for encoding processing.

[0056] S214. Data integration and consolidation.

[0057] In this embodiment, multiple data sources are merged. Among them, when merging the data sources, duplicate records are removed to ensure data uniqueness.

[0058] Furthermore, consistency checks are performed on the merged data to ensure that there are no conflicts or contradictions between the data from different data sources.

[0059] S215. Data verification and validation.

[0060] In this embodiment, it is checked whether the data is complete, including whether all necessary fields are filled.

[0061] Further, verify the accuracy of the data by comparing historical data, business logic, or external data sources.

[0062] Further, check whether the data between different fields is consistent, such as whether the relationship between the drug prescription amount and the drug prescription quantity is reasonable.

[0063] S216. Generate optimized report data.

[0064] In this embodiment, the preprocessed data is exported in a format suitable for report generation or data analysis, such as CSV, Excel, or database files.

[0065] Further, according to the requirements of the report category, use a report generation tool or software to generate the final report.

[0066] Further, review the generated report to ensure that the data is accurate, the format is standardized, and the information is complete.

[0067] S22. Build a report parsing model based on the optimized report data.

[0068] In one embodiment, based on the optimized report data, a report parsing model is built using a basic large model expressed in mathematics, such as the Zero inference model. The report parsing model includes a report intelligent language model and a report in-depth analysis model.

[0069] Specifically, the report intelligent language model satisfies the following expression:

[0070]

[0071] Where represents the explanatory text of the report data, represents the natural language parsing function, represents the original text data in the report, represents the lexical features of the data in the report, represents the syntactic tree, represents the noise reduction function, which is used to reduce noise, and its result is used as one of the inputs of the function to generate the explanatory text of the report data , is the context information of the report, is the domain knowledge; the report in-depth analysis model satisfies the following expression:

[0072]

[0073] Where represents the in-depth analysis result, represents the deep learning model function, Represents the original data set in the report, Represents the feature extraction function, Represents the data preprocessing function, Represents unstructured data, Represents the data type. Represents the report data set Is preprocessed, and the preprocessed data is processed by the feature extraction function Finally, the results of feature extraction, unstructured data And the data type or task type Are passed as inputs to the deep learning model function To obtain the deep parsing result . Such a logical process is more in line with the conventional processes of data processing and machine learning models. The report Zhiyan model can receive the original text data in the report, utilize natural language parsing functions, and combine lexical features and syntactic trees to generate an explanatory text for the report data. At the same time, through the noise reduction function, combined with the context information and domain knowledge of the report, the accuracy and relevance of the explanatory text are ensured. The report deep analysis model focuses on in-depth analysis of the original data set in the report. It enhances the expressiveness of data features through the feature enhancement factor and uses a deep learning model, combined with data preprocessing functions, unstructured data, and training parameters, to output a structured data set. This data set can provide clearer and more valuable data analysis results to assist decision-makers in making more accurate decisions.

[0074] In another embodiment, first deploy the local version of Tongyi Qianwen's large language model and fine-tune the report structure and content based on the optimized report data.

[0075] Furthermore, by designing different prompt engineering, the report Zhiyan model and the report deep analysis model are respectively constructed.

[0076] S3. Through the report parsing model, parse the report data to obtain the basic parsing data.

[0077] Among them, S3 further includes the following steps:

[0078] S31. Through the report Zhiyan model, perform text parsing on the text data in the optimized report data to obtain the basic text parsing data.

[0079] In one embodiment, through the report Zhiyan model constructed in step S2, natural language technology is utilized to parse the text data in the optimized report data to obtain the basic text parsing data, and the text parsing includes semantic analysis, named entity recognition, and context relationship understanding.

[0080] Specifically, the semantic analysis includes the following steps:

[0081] First, establish a vocabulary library that contains information such as the definitions, synonyms, and antonyms of each word;

[0082] Furthermore, use a word embedding algorithm, such as Word2Vec, to map the vocabulary to a low-dimensional vector space and calculate the similarity between them to capture the correlations between words;

[0083] Furthermore, perform syntactic analysis to break down a sentence into different chunks, such as the subject, predicate, and object, to achieve an understanding of the sentence structure;

[0084] Furthermore, perform semantic role labeling to associate each word in a sentence with its semantic role, where the semantic roles include the agent, patient, time, and location;

[0085] Furthermore, perform semantic relation extraction to extract the relationships and connections between different entities from the text.

[0086] Among them, the semantic analysis model includes a word embedding algorithm, a syntactic analyzer, a semantic role labeler, and a relation extraction algorithm. The objective function of the word embedding algorithm is to minimize the following loss function:

[0087] ,

[0088] where, represents the loss function, represents the total number of words in the text, represents the size of the context window, represents the current word, represents the word in the context, represents given the current word when, the context word appears probability.

[0089] The named entity recognition includes the following steps:

[0090] First, determine the boundaries of the entities in the text, that is, word segmentation. It should be noted that in Chinese text, since there is no space as a separator between words, word segmentation is a crucial step in named entity recognition.

[0091] Furthermore, identify individual entities and determine the categories to which the entities belong, such as person names, place names, and organization names.

[0092] Furthermore, identify compound entities, that is, entities composed of multiple words.

[0093] Furthermore, common named entity recognition models include conditional random fields, support vector machines, hidden Markov models, and deep learning models such as recurrent neural networks, long short-term memory networks, and transformer models. Among them, the objective function of the conditional random field is usually to maximize the following log-likelihood function:

[0094] ,

[0095] where, is the state feature function, is the transition feature function, is the input feature function, denotes the input sequence, denotes the output sequence, denotes the length of the sequence, denotes the score of the forward pass, is the transition score, representing the probability of transitioning from state to state . is the emission score, representing the probability of observing given state .

[0096] The context relationship understanding includes the following steps:

[0097] First, analyze the context information in the text to understand the specific meaning and context of the entity;

[0098] Furthermore, use the context model and attention mechanism method to achieve context understanding to provide more accurate and complete semantic analysis results.

[0099] Among them, the context relationship understanding model includes a context model and an attention mechanism.

[0100] The context vector is usually used to represent the context information of a certain position or entity in the text. It is obtained by weighted summation of the words in the context or using a more complex neural network structure. The expression of the context vector is as follows:

[0101] ,

[0102] where, denotes the context vector of position , denotes the weight of the word at position for position , denotes the word embedding vector of the word at position , denotes the text function.

[0103] Attention weights are used to measure the correlation or importance between different parts of a text. They are obtained by calculating the similarity score between a query vector and a context vector. The expression is as follows:

[0104]

[0105] Where, represents the position of the query vector and the attention weight between the word vector at position . is a similarity calculation function representing the text function.

[0106] Context relation extraction refers to identifying and extracting the relationships between entities or the associations between events from a text. This usually involves further processing and analysis of the context vector.

[0107] For entity relation extraction, a relation classifier can be defined to receive the context vectors of two entities as input and output the relation category between them.

[0108]

[0109] Where, represents the relation category between two entities represents the upper context vector of the entity represents the lower context vector of the entity.

[0110] S32. Through the report in-depth analysis model, numerical analysis is performed on the numerical data in the optimized report data to obtain basic numerical analysis data. The numerical analysis includes feature extraction, pattern recognition, and predictive analysis.

[0111] In one embodiment, through the report in-depth analysis model constructed in step S2, deep learning technology is utilized to perform numerical analysis on the numerical data in the optimized report data to obtain basic numerical analysis data. The numerical analysis includes feature extraction, pattern recognition, and predictive analysis.

[0112] Specifically, the feature extraction is a key step in deep learning, aiming to extract useful information or features from the original data for subsequent analysis and prediction. In numerical data, feature extraction may involve data preprocessing, transformation, or dimensionality reduction.

[0113] Suppose we have a numerical data set where, is the feature vector of the th data point. Feature extraction can be expressed as:

[0114]

[0115] Among them, represents the extracted feature vector, represents the feature extraction function, which may be a linear transformation, a non-linear transformation, principal component analysis (PCA), an autoencoder, or other complex deep learning models. In the linear transformation, the following conditions are satisfied:

[0116]

[0117] Among them, represents the weight matrix, which determines the importance of each original feature, represents the bias vector, which is used to adjust the baseline of the output features.

[0118] The pattern recognition is another important step in deep learning, aiming to identify potential patterns or regularities from the extracted features. It usually involves classification, clustering, or regression tasks.

[0119] Assume that we have already extracted the features , the pattern recognition can be expressed as:

[0120]

[0121] Among them, represents the data point 's predicted label or class, represents the pattern recognition function, which may be a classifier, a clustering algorithm, or a regression model.

[0122] The prediction analysis involves using a trained model to make predictions on new data. It is usually achieved by passing the new data through the feature extraction and pattern recognition stages.

[0123] For the new data point , the prediction analysis can be expressed as:

[0124]

[0125] Among them, represents the predicted probability value, represents the conversion function, which is used to convert the output into a probability distribution, , represents the bias vector, , represents the weight matrix, represents the non-linear activation function, which is used to introduce non-linearity so that the model can learn complex feature representations; is used to adjust the baseline of the extracted features, Meet the following conditions:

[0126]

[0127] Among them, is the number of categories (for classification tasks) or the number of output variables (for regression tasks); The element determines the influence degree of each dimension in the original feature vector on the extracted features, Meet the following conditions:

[0128]

[0129] Among them, represents the dimension. It should be noted that for regression tasks, the function can be removed and a linear output or non - linear transformation can be directly used.

[0130] S4. Utilize the domain knowledge graph to perform enhanced parsing on the basic parsed data to obtain enhanced parsed data.

[0131] In one embodiment, first, collect and organize domain knowledge related to rational drug use, including drug information, disease information, and relevant medical terms and concepts.

[0132] Furthermore, use this knowledge to construct a domain knowledge graph, where the nodes in the graph represent entities and the edges represent the relationships between entities.

[0133] Furthermore, use the domain knowledge graph to perform enhanced parsing on the basic parsed data.

[0134] Specifically, through matching and reasoning, identify and supplement the key information that may be missing in the report, such as automatically supplementing the dosage and usage information according to the drug name. At the same time, utilize the relationship information in the graph to analyze the logical relationships and context meanings between the data, improving the accuracy and depth of the parsing.

[0135] Furthermore, obtain enhanced parsed data containing richer and more accurate information.

[0136] S5. Generate a structured report based on the enhanced parsed data.

[0137] In one embodiment, first, clarify the problems to be solved by the report, the information to be provided, or the decisions to be supported; understand who the main readers of the report are, their information needs, interests, and background knowledge.

[0138] Furthermore, sort out the enhanced parsing results, including reviewing the text parsing results, summarizing the numerical parsing results, and integrating the information.

[0139] Specifically, the reviewed text parsing results include key findings of semantic analysis, named entity recognition, and context relationship understanding; the summarized numerical parsing results include main conclusions of feature extraction, pattern recognition, and predictive analysis; the integrated information includes combining the results of text parsing and numerical parsing to form a coherent and consistent information system.

[0140] Further, design the report structure, including determining the report framework, dividing chapters, designing charts, and visualization.

[0141] Specifically, the determination of the report framework includes: designing the overall framework of the report according to the report objectives and audience needs; the overall framework includes an introduction, main body, conclusion, and recommendation sections; the division of chapters includes: dividing the report content into several chapters or subsections, with each chapter or subsection focusing on a specific topic or issue; the design of charts and visualization includes: selecting appropriate charts, graphs, and visualization tools to display key data and findings, enhancing the readability and attractiveness of the report.

[0142] Further, automatically generate a structured report.

[0143] S6. Based on the structured report, according to the user feedback information, optimize the algorithms for report parsing and report generation to obtain a structured refined report.

[0144] Among them, S6 further includes the following steps:

[0145] S61. Based on the structured report, collect user feedback information.

[0146] In one embodiment, based on the structured report, design multiple feedback collection channels through online questionnaires, email surveys, phone interviews, or face-to-face meetings to ensure that users can conveniently provide feedback.

[0147] Further, in the feedback collection channels, clearly list the content that users may need to provide feedback on, such as the accuracy, readability, integrity, practicality, format aesthetics of the report, as well as specific suggestions or opinions from users.

[0148] Further, set a fixed feedback collection period, such as monthly or quarterly, organize and analyze the collected user feedback to form a preliminary summary of user feedback.

[0149] S62. According to the feedback information, form a feedback analysis report.

[0150] In one embodiment, classify and summarize the user feedback according to the dimensions of content, importance, and frequency of occurrence to form a preliminary framework of the feedback analysis report.

[0151] Further, for quantifiable feedback, such as the accuracy of reports and readability scores, a quantitative evaluation is conducted to form specific quantitative indicators.

[0152] Further, based on the results of classification and induction and quantitative evaluation, a feedback analysis report is written to elaborate in detail the overall situation, main problems, and suggestions of user feedback.

[0153] S63. Through the feedback analysis report, determine the optimization requirements for the report parsing and report generation algorithms, and form a list of optimization requirements.

[0154] In one embodiment, according to the feedback analysis report, identify the key points and specific problems that need to be optimized in the report parsing and report generation algorithms.

[0155] Further, for the identified optimization requirements, formulate clear optimization goals, such as improving the accuracy, readability, and generation efficiency of reports.

[0156] Further, organize the optimization goals and corresponding optimization requirements into a list of algorithm optimization requirements, providing a basis for the subsequent design of optimization solutions.

[0157] S64. According to the list of optimization requirements, design an algorithm optimization solution.

[0158] In one embodiment, research the optimization methods and successful cases of relevant algorithms in the industry, and draw on their experiences and lessons to provide reference for the design of the algorithm optimization solution.

[0159] Further, according to the list of optimization requirements and research results, design specific algorithm optimization strategies, such as improving data preprocessing methods, optimizing feature extraction algorithms, and adjusting report generation templates.

[0160] Further, formulate a detailed implementation plan for the algorithm optimization solution, including optimization steps, time nodes, and personnel division of labor.

[0161] S65. Utilize the algorithm optimization solution to obtain optimized algorithms for report parsing and report generation.

[0162] In one embodiment, according to the algorithm optimization solution, perform coding implementation and optimization on the original report parsing and report generation algorithms.

[0163] Further, conduct sufficient testing and verification on the optimized algorithm to ensure that it can meet the optimization goals and requirements, while ensuring the stability and reliability of the algorithm.

[0164] Further, according to the test and verification results, perform iterative optimization on the algorithm until the best effect is achieved.

[0165] S66. Through the optimized algorithm, obtain a structured refinement report.

[0166] In one embodiment, a new structured improvement report is generated by using an optimized report parsing and report generation algorithm.

[0167] Furthermore, the structured improvement report is provided to some users for evaluation, and their feedback and suggestions are collected.

[0168] Furthermore, according to the evaluation feedback of the users, the algorithm is continuously improved and optimized to ensure that the structured improvement report can continuously meet the needs and expectations of the users.

[0169] S7. Use two rich text editing areas to edit the data part and the conclusion part of the structured improvement report to obtain an editing result.

[0170] Among them, S7 further includes the following steps:

[0171] S71. Determine two rich text editing areas.

[0172] In one embodiment, by setting two rich text editing areas, they are respectively used to edit the data part and the conclusion part of the structured improvement report. The rich text editing areas support common rich text formats, such as bold, italic, underline, color, font size, paragraph alignment, list, and picture insertion; they also support pre-editing and saving rich text content as templates, and the templates include private templates and shared templates; the private templates pre-edit and save specific rich text content as private templates according to the personalized needs of the users. For example, users can create templates containing specific titles, subtitles, paragraph formats, etc. according to their own report styles. The shared templates provide a set of standard and reusable rich text templates for all users, covering common report structures and format requirements, such as title pages, tables of contents, main texts, conclusions, and references.

[0173] S72. Use the two rich text editing areas to edit the data part and the conclusion part of the structured improvement report to obtain an editing result.

[0174] In one embodiment, when a user needs to edit the structured improvement report, first enter the rich text editing area; the system provides a template selection interface, allowing the user to select a suitable private template or shared template from the template library as the starting point for editing.

[0175] Furthermore, in the rich text editing area of the data part, the user can input or paste the data content of the report according to the structure and format requirements of the selected template; the user can use the rich text editing function to format and typeset the data content to ensure the clarity and readability of the data.

[0176] Further, in the rich text editing area of the conclusion part, the user inputs or pastes the corresponding conclusion content according to the data analysis results and conclusions of the report. Similarly, the user can use the rich text editing function to format and typeset the conclusion content to highlight key information and conclusions.

[0177] Further, after completing the editing, the user should save the editing result. The system should provide a save button or shortcut key to facilitate the user to save the edited rich text content to the database. At the same time, the system should support the user to save the currently edited rich text content as a new private template or shared template for subsequent reuse.

[0178] S8. Remove the content that does not need to be displayed from the editing result to obtain the final editing result.

[0179] In one embodiment, open the editing result and review the content item by item to determine whether each item needs to be retained. If a certain item does not meet the removal criteria, that is, it needs to be retained, then mark it as "retained"; if a certain item meets the removal criteria, that is, it does not need to be displayed, then mark it as "removed".

[0180] Further, according to the review result, perform the removal operation. For the content marked as "removed", the deletion function of the rich text editor can be used to remove it from the editing result; when performing the removal operation, it should be ensured that the content that needs to be retained is not accidentally deleted, and the format and structure of the editing result should be kept unchanged as much as possible.

[0181] Further, after completing the removal operation, review and verify the editing result. Ensure that all content that does not need to be displayed has been correctly removed and that the content that needs to be retained has not been accidentally deleted. At the same time, check whether the format and structure of the editing result are still clear, consistent, and easy to understand.

[0182] Further, according to the results of the review and verification, perform the final sorting and typesetting of the editing result to obtain the final editing result. It should be noted that this step needs to ensure that all retained content is arranged in logical order and format requirements.

[0183] S9. Export the final editing result to obtain a rational drug use report combined with artificial intelligence.

[0184] In one embodiment, first open the rich text editor or document processing software where the final editing result is located.

[0185] Further, select the "Save As" or "Export" option in the "File" menu.

[0186] Further, in the popped-up dialog box, select the Word document (.docx or.doc) as the save type. Specify the save location and click the "Save" button.

[0187] In another embodiment, first open the software where the final editing result is located.

[0188] Further, select the "Print" option in the "File" menu. It should be noted that "Print" here actually means exporting to PDF rather than true printing.

[0189] Further, select the installed PDF converter or printer driver in the printer list.

[0190] Further, click the "Properties" or "Settings" button and adjust the PDF export options as needed.

[0191] Further, click "OK" to return to the print dialog box, and then click the "Print" or "Export to PDF" button.

[0192] Further, specify the save location and file name, and click the "Save" button.

[0193] In yet another embodiment, first open the PDF document using a PDF editor.

[0194] Further, in the software, select the "Tools" or "Edit" menu and find the "Watermark" or "Header and Footer" option.

[0195] Further, click the "Add" or "Create" button to enter the watermark editing interface.

[0196] Further, in the watermark editing interface, select "Text" or "Image" as the watermark type and input or select the watermark content.

[0197] Further, adjust the properties such as the position, size, transparency, rotation angle, etc. of the watermark to meet the requirements. If necessary, set the watermark to be applied to specific pages or the entire document of the document.

[0198] Please refer to Figure 2 , Figure 2 which is a schematic structural diagram of a rational drug use report generation system combined with artificial intelligence in an embodiment of the present invention. The system includes an input module, a processing module, an output module, and a storage module. The input module, the processing module, the output module, and the storage module are interconnected. Among them, the storage module is used to store a computer program, the computer program includes program instructions, the processing module is configured to call the program instructions, and the system uses the rational drug use report generation method combined with artificial intelligence as described above.

[0199] The input module includes a report selection module, a rich text editing module, a reset and exclusion module, and a template management module; the processing module includes a report parsing module and a self-learning module; the output module includes a report generation module and an export module: the storage module is characterized by fast read and write speed, large capacity, and high reliability, and is mainly used to store the data input by the input module and the result data processed by the processing module, and can meet the needs of storing a large amount of data.

[0200] The report parsing module is connected to the report selection module, the self-learning module, and the report generation module; the report generation module is connected to the report parsing module, the self-learning module, the rich text editing module, the reset and exclusion module, the export module, and the template management module.

[0201] The report selection module is used to select the report category for which a report needs to be generated.

[0202] The report parsing module is used to receive the report category information passed by the report selection module and pass the parsed report data to the report generation module and the self-learning module.

[0203] The report generation module is used to receive the parsed data and generate a final report in combination with the rich text content edited by the user, template information, and reset instructions.

[0204] The self-learning module is used to collect user feedback and operation data to improve the parsing and generation algorithms.

[0205] The rich text editing module is used to pass the rich text content edited by the user to the report generation module.

[0206] The reset and exclusion module is used to delete specific content in the report according to the user's instructions.

[0207] The export module is used to export the generated report in PDF or Word format and can add a custom watermark to the PDF.

[0208] The template management module is used to provide the template information edited and saved by the user for the report generation module to reference when generating a report.

[0209] The report parsing module includes a data preprocessing unit, a feature extraction unit, and a semantic understanding unit; the data preprocessing unit is used to clean and format the report data; the feature extraction unit uses deep learning technology to extract key information in the report; the semantic understanding unit uses natural language processing technology to parse the logical relationship and context meaning between the data.

[0210] The self-learning module adopts a reinforcement learning algorithm and realizes optimization through the following steps: monitoring and recording the feedback behaviors of users during use, including editing modifications, content deletion, and report regeneration requests; analyzing the feedback behaviors to identify key factors affecting report quality and user experience and obtaining analysis results; and adjusting the parameters of the deep learning model and natural language processing strategies according to the analysis results.

[0211] The template management module includes a version control unit and a permission management unit; the version control unit allows users to track and compare different versions of templates; the permission management unit sets access and editing permissions for private templates and shared templates according to user roles.

[0212] In summary, the present invention can effectively solve the problems of low efficiency and easy errors in the traditional report generation method, insufficient intelligence, automation ability and poor versatility of the existing system, and inability to flexibly meet the requirements of large data volume, complex classification and diversified report generation in the medical field. By using artificial intelligence technology to automatically parse and generate reports, the efficiency and accuracy of report generation are improved, and the error rate is reduced; through continuous optimization of the self-learning mechanism, the system can continuously learn and evolve to adapt to the personalized needs of different users; by providing a user-friendly human-computer interaction interface with flexible editing and export functions, the report generation requirements in different scenarios are met; through the template management function, common report templates are preset to improve work efficiency, and at the same time, both private and shared modes are supported to enhance the flexibility of use.

[0213] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered by the scope of the claims and the description of the present invention.

Claims

1. A method for generating a rational drug use report combined with artificial intelligence, characterized in that The method includes the following steps: Select the report category for which the report is to be generated; Based on the report category, construct a report parsing model; Through the report parsing model, parse the report data to obtain basic parsed data; Utilize the domain knowledge graph to perform enhanced parsing on the basic parsed data to obtain enhanced parsed data; Generate a structured report based on the enhanced parsed data; Based on the structured report, according to the user's feedback information, optimize the algorithms for report parsing and report generation to obtain a structured refined report; Utilize two rich text editing areas to edit the data part and conclusion part of the structured refined report to obtain an editing result. The rich text editing areas support pre-editing and saving rich text content as templates, and the templates include private templates and shared templates; Eliminate the content that does not need to be displayed in the editing result to obtain the final editing result; Export the final editing result to obtain a rational drug use report combined with artificial intelligence.

2. The method for generating a reasonable medication report combined with artificial intelligence according to claim 1, wherein The constructing a report parsing model based on the report category includes: Based on the report category, preprocess the report data to obtain optimized report data; Based on the optimized report data, construct a report parsing model, and the report parsing model includes a report intelligent language model and a report in-depth analysis model; The report intelligent language model satisfies the following expression: , Among them, represents the explanatory text of the report data, represents the natural language parsing function, represents the original text data in the report, represents the lexical features of the data in the report, represents the syntactic tree, represents the noise reduction function, is the context information of the report, is the domain knowledge; the report in-depth analysis model satisfies the following expression: , Among them, represents the in-depth parsing result, represents the deep learning model function, represents the original data set in the report, represents the feature extraction function, represents the data preprocessing function, represents unstructured data, represents the data type.

3. The method for generating a reasonable medication use report combined with artificial intelligence according to claim 1, characterized in that The obtaining basic parsed data by parsing the report data through the report parsing model includes: Through the report intelligent language model, perform text parsing on the text data in the optimized report data to obtain basic text parsing data, and the text parsing includes semantic analysis, named entity recognition, and context relationship understanding; Through the report in-depth analysis model, perform numerical parsing on the numerical data in the optimized report data to obtain basic numerical parsing data, and the numerical parsing includes feature extraction, pattern recognition, and predictive analysis.

4. A method for generating a reasonable medication report combined with artificial intelligence according to claim 1, characterized in that, The generating a structured report according to the enhanced parsed data includes: Select a report template according to the enhanced parsed data; Based on the report template, fill in the enhanced parsed data; Verify the enhanced parsed data to obtain a verification result; Generate a structured report through the verification result.

5. A method for generating a reasonable medication use report combined with artificial intelligence according to claim 1, characterized in that, The optimizing the algorithms for report parsing and report generation based on the structured report according to the user's feedback information to obtain a structured refined report includes: Based on the structured report, collect the user's feedback information; Form a feedback analysis report according to the feedback information; Through the feedback analysis report, determine the algorithm optimization requirements for report parsing and report generation to form an algorithm optimization requirement list; Design an algorithm optimization plan according to the algorithm optimization requirement list; Utilize the algorithm optimization plan to obtain optimized algorithms for report parsing and report generation; Obtain a structured refined report through the optimized algorithms.

6. A rational drug use report generation system combined with artificial intelligence, the system uses a rational drug use report generation method according to any one of claims 1 to 5, characterized in that, The system includes a report selection module, a report parsing module, a report generation module, a self-learning module, a rich text editing module, a reset and exclusion module, an export module, and a template management module; the report parsing module is connected to the report selection module, the self-learning module, and the report generation module; the report generation module is connected to the report parsing module, the self-learning module, the rich text editing module, the reset and exclusion module, the export module, and the template management module; The report selection module is used to select the report category for which a report needs to be generated; The report parsing module is used to receive the report category information passed by the report selection module and pass the parsed report data to the report generation module and the self-learning module; The report generation module is used to receive the parsed data and generate a final report in combination with the rich text content, template information, and reset instructions edited by the user; The self-learning module is used to collect user feedback and operation data to improve the parsing and generation algorithms; The rich text editing module is used to pass the rich text content edited by the user to the report generation module; The reset and exclusion module is used to delete specific content in the report according to the user's instructions; The export module is used to export the generated report as a PDF or Word document, and the PDF document can be added with a custom watermark; The template management module is used to provide the rich text information pre-edited and saved by the user.

7. The rational drug use report generation system combined with artificial intelligence according to claim 6, characterized in that, The report parsing module includes a data preprocessing unit, a feature extraction unit, and a semantic understanding unit; The data preprocessing unit is used to clean and format the report data; The feature extraction unit uses deep learning technology to extract key information in the report; The semantic understanding unit uses natural language processing technology to parse the logical relationship and context meaning between data.

8. The rational drug use report generation system combined with artificial intelligence according to claim 6, characterized in that The self-learning module adopts a reinforcement learning algorithm and is optimized through the following steps: Monitor and record the feedback behaviors of the user during use, including editing modifications, content deletions, and report regeneration requests; Analyze the feedback behaviors, identify the key factors affecting the report quality and user experience, and obtain the analysis results; According to the analysis results, adjust the deep learning model parameters and natural language processing strategies.

9. The rational drug use report generation system combined with artificial intelligence according to claim 6, characterized in that The template management module includes a version control unit and a permission management unit; The version control unit allows the user to track and compare different versions of the template; The permission management unit sets the access and editing permissions for private templates and shared templates according to the user role.

Images