Knowledge reasoning-based medicine access report automatic writing system

Abstract

The invention relates to the technical field of medicine market admission and artificial intelligence, and discloses a knowledge reasoning-based medicine admission report automatic writing system, which comprises a data integration module used for collecting multi-dimensional data and constructing a medicine admission knowledge base; the antagonistic rhetorical strategy module is used for formulating a demonstration blueprint containing a demonstration strategy for the knowledge base; the knowledge reasoning module performs analysis reasoning according to blueprint guidance and outputs a result; the evaluation module is used for judging a quasi-input condition according to a reasoning result and generating a comprehensive evaluation conclusion; and the report generation module is used for automatically writing and outputting a medicine access report in combination with the evaluation conclusion and the argument blueprint. Through cooperative work of multiple modules, automation and intellectualization of the whole report writing process are achieved, and the problems that in the prior art, manual work is relied on, efficiency is low, and deep intelligent evaluation ability is lacked are effectively solved.

Description

Technical Field

[0001] This invention relates to the fields of pharmaceutical market access and artificial intelligence technology, and in particular to an automatic pharmaceutical market access report writing system based on knowledge reasoning. Background Technology

[0002] In the pharmaceutical industry, market access for new drugs is a crucial step in realizing their commercial value. This process involves a comprehensive analysis of multi-dimensional information, including the legal and regulatory framework of the target market, clinical value, and economic viability, in order to write a professional access report. This process is extremely complex and requires a high level of expertise.

[0003] Currently, the writing of pharmaceutical access reports mainly relies on the manual labor of market access specialists. Specialists typically use tools such as literature databases and policy websites to retrieve information, then manually screen, integrate, and analyze it, and finally complete the report writing manually.

[0004] This traditional model has significant limitations. First, report writing relies heavily on manual labor, leading to inefficiency throughout the process. Second, the selection and interpretation of evaluation criteria are easily influenced by the personal experience of specialists, making the evaluation process susceptible to subjective factors and lacking objectivity and consistency. More importantly, existing databases and other auxiliary tools are functionally limited, making it difficult to adapt to the complex entry requirements of different markets for strategic adjustments, and lacking in-depth, intelligent, and comprehensive evaluation capabilities.

[0005] Therefore, this invention proposes an automatic writing system for medical access reports based on knowledge reasoning to address the shortcomings of existing technologies. Summary of the Invention

[0006] The purpose of this invention is to provide an automatic pharmaceutical access report writing system based on knowledge reasoning, which solves the problems of existing technologies where pharmaceutical access report writing mainly relies on manual labor, resulting in low efficiency, the evaluation process being easily affected by subjective factors, and existing auxiliary tools being difficult to adapt to the complex access conditions of different markets and lacking in-depth intelligent comprehensive evaluation capabilities.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] The first aspect of this invention provides a knowledge-based reasoning-based automatic pharmaceutical access report writing system, comprising:

[0009] The data integration module is used to collect multi-dimensional data sources and, based on the target market and target drug information input by the user, integrate the collected information related to the target market, laws and regulations, policy documents and market data, in order to build and update the pharmaceutical access knowledge base;

[0010] The adversarial rhetoric strategy module is used to develop an argumentation blueprint, including argumentation strategies, for the aforementioned medical access knowledge base.

[0011] The knowledge reasoning module is used to analyze and reason about the information in the medical access knowledge base according to the guidance of the argumentation blueprint, and output the analysis and reasoning results.

[0012] The evaluation module is used to determine whether the target drug meets the access conditions based on the analysis and reasoning results, and to generate a comprehensive evaluation conclusion.

[0013] The report generation module is used to automatically write and output a pharmaceutical access report based on the comprehensive evaluation conclusions and the demonstration blueprint.

[0014] In a specific embodiment, the above system is further defined as follows:

[0015] The specific workflow of the data integration module includes the following steps:

[0016] Collect data from multiple dimensions, including structured databases and unstructured documents;

[0017] Structured data is collected from the structured database through data query, and information is collected from the unstructured documents through optical character recognition technology, named entity recognition model and relation extraction model to form knowledge triples;

[0018] Based on user-inputted target market and target drug information, we integrate relevant laws, regulations, policy documents, and market data from the collected structured data and knowledge triplets.

[0019] The medical access knowledge base is constructed and updated by mapping the integrated data and knowledge triples to the graph structure of a graph database. The construction process creates new nodes and edges in the graph database for newly added entities and relations; the update process updates the corresponding nodes, edges, or their attributes in the graph database for existing entities and relations.

[0020] The specific workflow of the adversarial rhetoric strategy module includes the following steps:

[0021] The adversarial review profile generator dynamically constructs an adversarial review profile containing multiple weighted potential points of contention based on information in the medical access knowledge base.

[0022] The dynamic argument framework builder selects the optimal argument framework from a pre-set rhetorical strategy library based on the adversarial review profile.

[0023] By using the evidence chain and weakness hedging planner, the optimal argument framework is parsed into specific execution instructions, forming the argument blueprint.

[0024] The adversarial review profile generated by the adversarial review profile generator is a set of multiple weighted potential challenge points, and the formal representation of the set of multiple weighted potential challenge points is as follows:

[0025]

[0026] In the formula, P represents the adversarial review profile; n is the total number of potential points of contention; i is an integer index from 1 to n; Q i This represents a potential point of contention identified by index i; Representing the potential point of contention Q i The weight.

[0027] The dynamic argument framework constructor selects the optimal argument framework by integrating the utility function.

[0028] The calculation of the comprehensive utility function uses the adversarial review profile P as input, and the formula for calculating the comprehensive utility function includes:

[0029] U(F,P)=w S ·S H (F,P)+w C ·C M (F,P)-w R ·R E (F,P);

[0030] In the formula, F represents the argument framework to be evaluated; P represents the adversarial review profile; U(F,P) represents the overall utility value of the argument framework F relative to the adversarial review profile P; S H (F,P) represents the success rate score based on historical data; C M (F,P) is a matching score for the compatibility between the argument framework F and the adversarial review profile P; R E (F,P) is a risk score used to assess the evidence required to employ the aforementioned argumentation framework F; w S w C and w R These are the preset weighting coefficients.

[0031] The knowledge reasoning module employs a hybrid reasoning technology, which includes rule-based reasoning and machine learning-based reasoning. Its specific workflow includes the following steps:

[0032] The analysis of the aforementioned argument blueprint guides the differentiation between explicit arguments that require direct logical judgment and strategic arguments that require multi-step in-depth analysis.

[0033] The rule-based reasoning is invoked to process the explicit arguments in the medical access knowledge base;

[0034] The machine learning-based reasoning is invoked to process the strategic arguments in the medical access knowledge base;

[0035] The results of the rule-based reasoning and the machine learning-based reasoning are integrated to form and output the analysis and reasoning results.

[0036] The specific workflow of the evaluation module includes the following steps:

[0037] The analysis and reasoning results are mapped to multiple preset evaluation dimensions, each with its own weight.

[0038] Based on the results of this mapping, a compliance score is calculated for each evaluation dimension;

[0039] All compliance scores are weighted and aggregated to obtain a comprehensive score;

[0040] The comprehensive score is compared with one or more preset admission thresholds to determine whether the target drug meets the admission criteria, and the comprehensive evaluation conclusion is generated based on the judgment and the comprehensive score.

[0041] The specific workflow of the report generation module includes the following steps:

[0042] The aforementioned argument blueprint is analyzed into a report structure outline containing multiple chapters and core arguments;

[0043] For each chapter in the report structure outline, a natural language generation model is invoked, guided by the core arguments corresponding to that chapter, and the comprehensive evaluation conclusions, analysis and reasoning results are extracted as content materials to generate the text of that chapter;

[0044] All generated chapter texts are combined and formatted to output a complete medical access report.

[0045] A second aspect of this invention provides a method for automatically generating medical access reports based on knowledge reasoning, the method comprising the following steps:

[0046] S1. Collect multi-dimensional data sources, and based on the target market and target drug information input by the user, integrate the collected information related to the target market, laws, regulations, policy documents and market data to build and update the pharmaceutical access knowledge base;

[0047] S2. For the aforementioned pharmaceutical access knowledge base, develop an argumentation blueprint including argumentation strategies;

[0048] S3. Based on the guidance of the aforementioned argumentation blueprint, analyze and reason about the information in the pharmaceutical access knowledge base, and output the analysis and reasoning results;

[0049] S4. Based on the analysis and reasoning results, determine whether the target drug meets the access conditions and generate a comprehensive evaluation conclusion;

[0050] S5. Based on the comprehensive evaluation conclusions and the demonstration blueprint, automatically generate and output a pharmaceutical access report.

[0051] In summary, the present invention has at least one of the following beneficial technical effects:

[0052] 1. This invention constructs a medical access knowledge base and employs a hybrid technology based on knowledge-based deep reasoning to achieve systematic integration and analysis of multidimensional information. The system can utilize structured knowledge graphs for logically rigorous judgments while simultaneously uncovering deep-seated relationships and potential trends among data. This combined analytical approach effectively overcomes the reliance on personal experience in traditional manual assessments, significantly reduces assessment bias caused by subjective factors, and thus greatly enhances the objectivity and depth of access analysis conclusions.

[0053] 2. This invention, through an adversarial rhetoric strategy module, can dynamically construct an adversarial review profile to anticipate potential points of contention and select the optimal argumentation framework based on a comprehensive utility function. This design ensures that the final medical access report possesses a systematic argumentation strategy and weakness mitigation plan before output. Therefore, the resulting document is no longer a simple display of facts, but a highly logical and strategic argumentation system, effectively enhancing the persuasiveness of the content and its robustness in complex review environments.

[0054] 3. This invention significantly improves overall efficiency by implementing an integrated process from data integration to final content generation. In particular, its report generation module can automatically organize and format text content based on the argumentation blueprint and comprehensive evaluation conclusions, using a natural language generation model. This writing mechanism not only drastically shortens the time required by traditional manual methods but also ensures that each document adheres to unified, high-quality structured standards and argumentation logic, thereby achieving high efficiency and standardization in the report production process. Attached Figure Description

[0055] Figure 1 This is an architecture diagram of an automatic pharmaceutical access report writing system based on knowledge reasoning, according to an embodiment of the present invention.

[0056] Figure 2This is a flowchart of a method for automatically generating a medical access report based on knowledge reasoning, according to an embodiment of the present invention.

[0057] Figure 3 This is a flowchart illustrating the knowledge base construction and argumentation planning process according to an embodiment of the present invention.

[0058] Figure 4 This is a flowchart illustrating the analysis execution and report generation process according to an embodiment of the present invention.

[0059] The modules include: 100, Data Integration Module; 150, Medical Access Knowledge Base; 200, Adversarial Rhetoric Strategy Module; 300, Knowledge Reasoning Module; 400, Evaluation Module; and 500, Report Generation Module. Detailed Implementation

[0060] The following is in conjunction with the appendix Figure 1 - Appendix Figure 4 The present invention will be further described in detail below.

[0061] This invention provides an automatic writing system for medical access reports based on knowledge reasoning.

[0062] See attached document Figure 1 , Figure 1 This is an architecture diagram of a knowledge-reasoning-based automatic pharmaceutical access report writing system according to an embodiment of the present invention. The system provided by the present invention may include: a data integration module 100, an adversarial rhetoric strategy module 200, a knowledge reasoning module 300, an evaluation module 400, and a report generation module 500. In one embodiment, the system further includes a pharmaceutical access knowledge base 150 for data storage.

[0063] In the initial stage of system operation, users input basic information about the target market and target drugs into the system through a user interface that is not shown in the figure.

[0064] The data integration module 100 receives basic information input by the user at its input end and is connected to an external multidimensional data source. This module collects information from the multidimensional data source and, in conjunction with the user input, integrates relevant laws, regulations, policy documents, and market data related to the target market. The output end of this module is connected to the pharmaceutical access knowledge base 150, whereby the integrated information is stored, thereby enabling the construction or updating of the knowledge base.

[0065] The adversarial rhetoric strategy module 200 has its input connected to the medical access knowledge base 150. This module reads information from the knowledge base 150 to formulate an argument blueprint that includes specific argumentation strategies. This argument blueprint, as the output of this module, is transmitted to the knowledge reasoning module 300 and the report generation module 500.

[0066] The knowledge reasoning module 300 has its input terminals connected to the output terminals of the adversarial rhetoric strategy module 200 and the medical access knowledge base 150, respectively. This module receives the argument blueprint generated by module 200 and, guided by the blueprint, retrieves and processes relevant information from the knowledge base 150, performing analysis and reasoning. The module outputs the results of the analysis and reasoning to the evaluation module 400.

[0067] The evaluation module 400 has its input connected to the output of the knowledge reasoning module 300. This module receives the analysis and reasoning results and, based on these results, determines whether the target drug meets the access criteria, ultimately generating a comprehensive evaluation conclusion. The output of this module is connected to the report generation module 500 for transmitting the comprehensive evaluation conclusion.

[0068] The report generation module 500 has its input terminals connected to the output terminals of the adversarial rhetoric strategy module 200 and the evaluation module 400, respectively. This module simultaneously receives the argumentation blueprint and the comprehensive evaluation conclusion, and based on these two inputs, automatically writes and ultimately outputs a structured and complete pharmaceutical access report.

[0069] See attached document Figure 1 and Figure 3 , Figure 4 This is a flowchart of the knowledge base construction and argumentation planning process according to an embodiment of the present invention. In a specific embodiment, the specific workflow of the data integration module 100 is as follows.

[0070] First, the data integration module 100 performs the data collection step, acquiring raw information from preset or dynamically accessed multidimensional data sources. These data sources include two categories: one is structured databases, such as relational databases storing drug clinical trial data and market sales data; the other is unstructured documents, such as government documents, regulations, medical journals, and industry research reports in PDF or image format. This step provides raw materials for subsequent pharmaceutical access analysis.

[0071] Next, the data integration module 100 processes the two types of data sources in parallel. For structured databases, the system executes predefined data query instructions, such as SQL queries, to directly extract structured data records. For unstructured documents, the system initiates an information extraction pipeline. If the document is in image format, it is first converted into machine-readable text using an Optical Character Recognition (OCR) engine. Subsequently, a Named Entity Recognition (NER) model, such as one employing a BiLSTM-CRF architecture, processes the text to identify and label predefined entity categories, such as drug names, organization names, disease names, and policy identifiers.

[0072] After named entity recognition is complete, relation extraction models, such as classification models based on the Transformer architecture, analyze the identified entity pairs in the text to determine whether predefined semantic relationships exist between them, such as approval, applicable, and side effects. Each identified entity, relation, and entity combination is formatted as a knowledge triple, with the standard form being (head entity, relation, tail entity). This process forms the basis for building a knowledge-based analysis system, transforming unstructured information into structured knowledge units.

[0073] After information extraction and structured data collection are completed, the data integration module 100 performs the integration step. Based on the target market and target drug information initially input by the user, it filters and selects all collected structured data and knowledge triples. This step ensures that only laws, regulations, policy documents, and market data directly relevant to the current access analysis task are retained, thus providing an accurate and highly relevant dataset for subsequent reasoning and evaluation.

[0074] Finally, the data integration module 100 maps the integrated data and knowledge triples to a graph database to construct or update the medical access knowledge base 150. This mapping process follows these rules: the head and tail entities in each knowledge triple correspond to nodes in the graph database, and the relation corresponds to the directed edge connecting the two nodes. When constructing the knowledge base, if the entity or relation from the triple does not yet exist in the graph database, the system creates a new node or edge. When updating the knowledge base, if the entity or relation already exists, the system checks and updates the attributes of the corresponding node or edge, such as updating the effective date of a policy or the value of a market data point. In this way, the system can effectively automate report writing while ensuring the accuracy and timeliness of the data foundation.

[0075] See attached document Figure 1 and Figure 3 In one specific embodiment, the adversarial rhetoric strategy module 200 develops an argument blueprint to guide subsequent steps through a serialization process that includes multiple sub-functional components.

[0076] First, the adversarial rhetoric strategy module 200 analyzes information in the pharmaceutical access knowledge base 150 through its internal adversarial review profile generator to dynamically construct adversarial review profiles. This generator retrieves and identifies potential negative information related to the target drug or points of contention that may be encountered during the review process from the knowledge base 150 using preset rules or classification models. Examples include known side effects, higher pricing than similar drugs, and weaknesses in efficacy comparison data. In one embodiment, the preset rules are based on structured queries using a knowledge graph. For example, a rule can be defined as generating a relevant point of contention when the difference between a certain attribute value (such as price) of the target drug and the average value of the corresponding attribute of similar drugs in the knowledge base exceeds a preset threshold. In another embodiment, the classification model is a pre-trained text classifier that takes text fragments extracted from the knowledge base 150 as input and outputs a classification result indicating whether the text fragment constitutes a potential point of contention. Through this step, the system can establish a quantitative and targeted strategic foundation for subsequent pharmaceutical access analysis.

[0077] The adversarial review profile generated by the adversarial review profile generator is a set of multiple weighted potential challenge points, which are formally represented as follows:

[0078]

[0079] In the formula, P represents the adversarial review profile; n is the total number of potential points of contention; i is an integer index from 1 to n; Q i This represents a potential point of contention identified by index i; Representing the potential point of contention Q i The weight is a positive real number.

[0080] Next, module 200, through its internal dynamic argument framework constructor, selects the optimal argument framework from a pre-defined rhetorical strategy library based on the adversarial review profile P generated in the previous step. This rhetorical strategy library contains various pre-stored argument frameworks, such as cost-benefit priority frameworks, clinical innovation highlighting frameworks, and safety and risk control frameworks. The selection process is achieved by calculating a comprehensive utility function U(F,P) for each argument framework F to be evaluated; the argument framework with the highest utility value is determined as the optimal framework. This method is the core of knowledge-based strategy formulation, ensuring a high degree of match between the selected strategy and the current challenge. The formula for calculating the comprehensive utility function is:

[0081] U(F,P)=w S ·S H (F,P)+w C ·C M(F,P)-w R ·R E (F,P);

[0082] In the formula, F represents the argument framework to be evaluated; P represents the adversarial review profile; U(F,P) represents the overall utility value of the argument framework F relative to the adversarial review profile P; S H (F,P) represents the success rate score based on historical data. This score is calculated by querying 150 historical cases in the knowledge base that are similar to the current adversarial review profile, and statistically determining the proportion of successful applications using the argumentation framework F to be evaluated; C M (F,P) is a matching score for the compatibility between the argument framework F and the adversarial review profile P; R E (F,P) is a risk score used to assess the evidence required to employ the aforementioned argumentation framework F; w S w C and w R These are preset weighting coefficients used to adjust the system's emphasis on the three dimensions of historical success rate, problem relevance, and evidence risk when selecting strategies.

[0083] Finally, the adversarial rhetoric strategy module 200, through its internal evidence chain and weakness hedging planner, parses the optimal argument framework selected in the previous step into a series of specific, executable instructions, and combines these instructions to form the final argument blueprint. This argument blueprint is a structured data object that explicitly defines the chapter structure of the final report, the core arguments of each chapter, the list of evidence required to support each argument (i.e., pointers to specific nodes in the knowledge base 150), and response strategies for specific weaknesses. This argument blueprint will be output to the knowledge reasoning module 300 and the report generation module 500 to guide subsequent analysis and automated writing.

[0084] See attached document Figure 1 and Figure 4 In one specific embodiment, the knowledge reasoning module 300 is used to perform specific analytical tasks specified by the argument blueprint. At the core of the knowledge reasoning module 300 is a hybrid reasoning technology that includes both rule-based reasoning and machine learning-based reasoning.

[0085] For a specific report generation task, the proportion of rule-based reasoning in the hybrid reasoning technology is defined as a preset first proportion, and the proportion of machine learning-based reasoning is defined as a preset second proportion. Here, the preset proportions do not refer to fixed, globally unchanging values, but rather to the fact that when the argument blueprint is generated and transmitted to the knowledge reasoning module 300, the composition of arguments within the argument blueprint is already determined, thus making the task volume ratio of the two reasoning technologies predetermined for this execution process. Specifically, the first proportion is the ratio of the number of explicit arguments in the argument blueprint to the total number of arguments; the second proportion is specifically the ratio of the number of strategic arguments in the argument blueprint to the total number of arguments.

[0086] As a concrete example, suppose a generated argument blueprint contains 10 arguments to be processed. Seven of these arguments are marked as explicit arguments (e.g., verifying whether the target drug has received FDA approval, querying the pricing cap regulations for this type of drug in the target market), and three arguments are marked as strategic arguments (e.g., predicting the market share trend of the target drug over the next three years, assessing the probability of success of adopting a strategy that highlights clinical uniqueness). In this scenario, for this task, the first proportion is 70% (7 / 10), and the second proportion is 30% (3 / 10).

[0087] To achieve the above hybrid reasoning, the knowledge reasoning module 300 outputs the final analysis results by performing the following steps:

[0088] First, the knowledge reasoning module 300 parses the argument blueprint input by the adversarial rhetoric strategy module 200. By reading the type label of each argument to be processed in this blueprint, the system can distinguish between explicit arguments that require direct logical judgment and strategic arguments that require multi-step in-depth analysis. This parsing and differentiation step is the foundation for realizing hybrid reasoning. It decomposes a macro-level writing task into two sub-tasks of different natures and is a key link in realizing knowledge-based automated analysis.

[0089] Next, for all subtasks identified as explicit arguments, the knowledge reasoning module 300 invokes the rule-based reasoning engine for processing. In one embodiment, the reasoning engine translates such arguments into query instructions in one or more structured query languages ​​(e.g., SPARQL) and executes them on the graph database of the medical access knowledge base 150 to obtain accurate and unambiguous logical judgment results.

[0090] Simultaneously or subsequently, for all subtasks identified as strategic arguments, the knowledge reasoning module 300 invokes the machine learning-based reasoning engine for processing. In one embodiment, the reasoning engine may employ a pre-trained graph neural network (GNN) model to perform tasks such as link prediction or node classification by learning the embedding representations of nodes in the knowledge base 150, thereby generating quantitative, probabilistic analysis results for these complex arguments.

[0091] Finally, the knowledge reasoning module 300 integrates the deterministic result output by the rule-based reasoning with the probabilistic result output by the machine learning-based reasoning. This integration step uniformly fills the results from both sources into a predefined structured data object, forming a complete analysis and reasoning result containing multi-dimensional information. This result provides comprehensive data support for subsequent medical access assessment and automated report writing, and is transmitted to the assessment module 400 as the final output of the knowledge reasoning module 300.

[0092] See attached document Figure 1 and Figure 4 In one specific embodiment, the evaluation module 400 receives the analysis and reasoning results output by the knowledge reasoning module 300, and performs a multi-step quantitative evaluation process based on the results to generate a final comprehensive evaluation conclusion.

[0093] First, the evaluation module 400 maps the analysis and reasoning results to multiple pre-defined evaluation dimensions, each with its own weight. These evaluation dimensions are predefined in the knowledge base 150 and may include, for example, clinical value, cost-effectiveness, innovativeness, and regulatory compliance. Each evaluation dimension j is assigned a weight w. j This weight value reflects the relative importance of that dimension in the market access review system. The mapping process involves assigning each data item in the structured data object of the analysis and reasoning results to its corresponding evaluation dimension. For example, clinical trial efficacy data is mapped to the clinical value dimension, while drug pricing information is mapped to the economic dimension.

[0094] Next, based on the mapping results from the previous step, the evaluation module 400 calculates a compliance score C for each evaluation dimension. j The calculation method for this score depends on the nature of the dimension. For quantifiable dimensions, such as economy, the compliance score can be obtained by normalizing the cost-benefit ratio. For qualitative dimensions, such as innovativeness, the compliance score can be obtained directly from the classification confidence score output by a machine learning model, or by scoring the results of multiple rule-based checks. This step transforms complex analytical results into standardized numerical values, which is the foundation for achieving objective reasoning and evaluation based on knowledge.

[0095] After calculating the compliance scores for all evaluation dimensions, the evaluation module 400 performs a weighted aggregation of all compliance scores to obtain a comprehensive score S. final In one embodiment, the weighted aggregation is calculated using the following formula:

[0096]

[0097] In the formula, S final Represents the overall score; m is the total number of preset evaluation dimensions; j is an integer index from 1 to m used to traverse all dimensions; w j This represents the weight of the j-th evaluation dimension, and the sum of the weights of all dimensions is 1. C j This represents the compliance score for the j-th evaluation dimension.

[0098] Finally, the evaluation module 400 will calculate the overall score S. final It compares the drug with one or more preset admission thresholds to determine whether the target drug meets the admission criteria.

[0099] For example, the system can preset two thresholds T. low and T high If S final ≥T high If T is positive, it is considered a recommended admission; if T is negative, it is considered a recommended admission. l ow≤S final <T high If S, then it is judged as a suggestion for review; if S f inal <T low If the result is negative, the system will determine that admission is "not recommended". The system will then generate a comprehensive evaluation conclusion, which includes this judgment result and a comprehensive score S. final and the scores C for each dimension j The structured data objects. The comprehensive evaluation conclusions provide core, quantitative criteria for the subsequent automatic generation of medical access reports, and are transmitted to the report generation module 500 as the final output of the evaluation module 400.

[0100] See attached document Figure 1 and Figure 4 In one specific embodiment, the report generation module 500 is the final execution unit of the system described in this invention. It receives the argument blueprint output by the adversarial rhetoric strategy module 200 and the comprehensive evaluation conclusion output by the evaluation module 400, and based on these two inputs, executes a multi-step process to output the final medical access report. Specifically, the report generation module 500 is used for:

[0101] First, the argument blueprint is analyzed into a report structure outline containing multiple chapters and core arguments. This analysis step transforms the logical strategies and arguments in the argument blueprint into a concrete, hierarchical document directory. This outline clarifies the chapter divisions, chapter order, and the core arguments to be elaborated in each chapter of the final report, laying the structural foundation for the subsequent automated report writing.

[0102] Secondly, for each chapter in the report's structural outline, a natural language generation model is invoked to generate the text for that chapter. In this step, the system uses the core argument corresponding to the chapter as a high-level guide to ensure that the generated text's thematic direction aligns with the argumentation blueprint. Simultaneously, the system extracts the comprehensive evaluation conclusion and the analysis and reasoning results from the outputs of the previous knowledge reasoning module 300 and evaluation module 400 as specific material to fill in the chapter content. In one embodiment, the natural language generation model is a pre-trained language model based on the Transformer architecture, which takes the core argument and content material as input and outputs a logically coherent and detailed paragraph. This process is the core of knowledge-based content creation, ensuring the report's professionalism and accuracy.

[0103] Finally, all the generated chapter texts are combined and formatted to output a complete pharmaceutical access report. In this step, the system assembles all the independently generated chapter texts into a complete document according to the order specified in the report structure outline. Subsequently, the system applies a preset formatting template to format the document, including setting fonts, margins, adding headers and footers, and generating a table of contents. After formatting is completed, a structurally complete and formatted pharmaceutical access report is finally output, thus completing the entire knowledge-reasoning-based automated pharmaceutical access report writing process.

[0104] See attached document Figure 2 This invention also provides a method for automatically generating medical access reports based on knowledge reasoning, the method comprising the following steps:

[0105] S1. Collect multi-dimensional data sources and, based on user-inputted target market and target drug information, integrate relevant laws, regulations, policy documents, and market data from the collected information to construct and update the pharmaceutical access knowledge base. In this step, the system first collects raw data from data sources including structured databases and unstructured documents. Then, through data querying and information extraction techniques (including optical character recognition, named entity recognition, and relation extraction), the raw data is uniformly processed into structured data records and knowledge triples. Finally, the system maps this data related to the user's goals to a graph database, completing the construction and updating of the pharmaceutical access knowledge base by creating or updating nodes and edges in the graph. This step provides the data foundation for subsequent knowledge-based analysis processes.

[0106] S2. For the aforementioned pharmaceutical access knowledge base, an argumentation blueprint, including argumentation strategies, is developed. In this step, the system first dynamically generates an adversarial review profile containing multiple weighted potential points of contention by analyzing information in the knowledge base. Next, the system uses a comprehensive utility function to match an optimal argumentation framework for this profile from a pre-defined rhetorical strategy library. Finally, the system parses this optimal argumentation framework into a series of specific execution instructions, forming a structured argumentation blueprint. This step provides macro-level strategic guidance for the entire pharmaceutical access report writing process.

[0107] S3. Guided by the aforementioned argument blueprint, the system analyzes and reasons about the information in the pharmaceutical access knowledge base, and outputs the analysis and reasoning results. In this step, the system first parses the argument blueprint, distinguishing between explicit arguments and strategic arguments. Subsequently, the system employs a hybrid reasoning technique, calling a rule-based reasoning engine to process explicit arguments and a machine learning-based reasoning engine to process strategic arguments. Finally, the system integrates the outputs of the two reasoning engines to form a comprehensive analysis and reasoning result.

[0108] S4. Based on the analysis and reasoning results, determine whether the target drug meets the admission criteria and generate a comprehensive evaluation conclusion. In this step, the system first maps the analysis and reasoning results output in the previous step to multiple preset, weighted evaluation dimensions. Then, it calculates a compliance score for each dimension and performs weighted aggregation to obtain a comprehensive score. Finally, the system compares this comprehensive score with a preset admission threshold, makes an admission judgment, and generates a comprehensive evaluation conclusion based on this judgment and the scores. This conclusion is the core basis for achieving automated report generation.

[0109] S5. Based on the comprehensive evaluation conclusions and the argumentation blueprint, automatically write and output the pharmaceutical access report. In this step, the system first parses the argumentation blueprint into a report structure outline. Then, for each chapter in the outline, the system calls a natural language generation model, guided by the core arguments of the chapter, and extracts the comprehensive evaluation conclusions, analysis, and reasoning results as content material to generate chapter text. Finally, the system combines and formats all the generated chapter texts to output a complete pharmaceutical access report with detailed content and rigorous logic, thus completing the final writing task.

[0110] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A knowledge-based reasoning-based automatic pharmaceutical access report writing system, characterized in that, include: The data integration module is used to collect multi-dimensional data sources and, based on the target market and target drug information input by the user, integrate the collected information related to the target market, laws and regulations, policy documents and market data, in order to build and update the pharmaceutical access knowledge base; The adversarial rhetoric strategy module is used to develop an argumentation blueprint, including argumentation strategies, for the aforementioned medical access knowledge base. The knowledge reasoning module is used to analyze and reason about the information in the medical access knowledge base according to the guidance of the argumentation blueprint, and output the analysis and reasoning results. The evaluation module is used to determine whether the target drug meets the access conditions based on the analysis and reasoning results, and to generate a comprehensive evaluation conclusion. The report generation module is used to automatically write and output a pharmaceutical access report based on the comprehensive evaluation conclusions and the demonstration blueprint.

2. The automatic writing system for medical access reports based on knowledge reasoning according to claim 1, characterized in that, The data integration module is specifically used for: Collect data from multiple dimensions, including structured databases and unstructured documents; Structured data is collected from the structured database through data query, and information is collected from the unstructured documents through optical character recognition technology, named entity recognition model and relation extraction model to form knowledge triples; Based on the target market and target drug information input by the user, laws, regulations, policy documents and market data related to the target market are integrated from the collected structured data and knowledge triples. The integrated data and knowledge triples are mapped to the graph structure of the graph database to construct and update the medical access knowledge base; Specifically, constructing the medical access knowledge base involves creating new nodes and edges in the graph database for the newly added entities and relationships in the integrated data. Updating the medical access knowledge base specifically involves updating the corresponding nodes, edges, or their attributes in the graph database for existing entities and relationships in the integrated data.

3. The automatic writing system for medical access reports based on knowledge reasoning according to claim 1, characterized in that, The adversarial rhetoric strategy module is specifically used for: The adversarial review profile generator dynamically constructs an adversarial review profile containing multiple weighted potential points of contention based on information in the medical access knowledge base. The dynamic argumentation framework builder selects the optimal argumentation framework from a pre-set rhetorical strategy library based on the adversarial review profile. By using the evidence chain and weakness hedging planner, the optimal argument framework is parsed into specific execution instructions, forming an argument blueprint.

4. The automatic writing system for medical access reports based on knowledge reasoning according to claim 3, characterized in that, The adversarial review profile generated by the adversarial review profile generator is a set of multiple weighted potential challenge points, and the formal representation of the set of multiple weighted potential challenge points is as follows: P={(Q i ,w Qi )∣i=1,2,…,n}; In the formula, P represents the adversarial review profile; n is the total number of potential points of contention; i is an integer index from 1 to n; Q i Represents a potential point of contention identified by index i; w Qi Representing the potential point of contention Q i The weight.

5. The automatic writing system for medical access reports based on knowledge reasoning according to claim 4, characterized in that, The dynamic argument framework constructor selects the optimal argument framework by integrating the utility function. The calculation of the comprehensive utility function uses the adversarial review profile P as input, and the formula for calculating the comprehensive utility function is as follows: include: U(F,P)=w S ·S H (F,P)+w C ·C M (F,P)-w R ·R E (F,P); In the formula, F represents the argument framework to be evaluated; P represents the adversarial review profile; U(F,P) represents the overall utility value of the argument framework F relative to the adversarial review profile P; S H (F,P) represents the success rate score based on historical data; C M (F,P) is a matching score for the compatibility between the argument framework F and the adversarial review profile P; R E (F,P) is a risk score used to assess the evidence required to employ the aforementioned argumentation framework F; w S w C and w R These are the preset weighting coefficients.

6. The automatic writing system for medical access reports based on knowledge reasoning according to claim 1, characterized in that, The knowledge reasoning engine employs hybrid reasoning technology, which includes rule-based reasoning at a preset first proportion and machine learning-based reasoning at a preset second proportion.

7. The automatic writing system for medical access reports based on knowledge reasoning according to claim 6, characterized in that, The knowledge reasoning module is specifically used for: The analysis of the aforementioned argument blueprint guides the differentiation between explicit arguments that require direct logical judgment and strategic arguments that require multi-step in-depth analysis. The rule-based reasoning is invoked to process the explicit arguments in the medical access knowledge base; The machine learning-based reasoning is invoked to process the strategic arguments in the medical access knowledge base; The results of the rule-based reasoning and the machine learning-based reasoning are integrated to form and output the analysis and reasoning results.

8. The automatic writing system for medical access reports based on knowledge reasoning according to claim 1, characterized in that, The evaluation module is specifically used for: The analysis and reasoning results are mapped to multiple preset evaluation dimensions, each with its own weight. Based on the mapping results, a compliance score is calculated for each of the evaluation dimensions. All the aforementioned compliance scores are weighted and aggregated to obtain a comprehensive score; Finally, the comprehensive score is compared with one or more preset admission thresholds to determine whether the target drug meets the admission criteria, and the comprehensive evaluation conclusion is generated based on the judgment and the comprehensive score.

9. The automatic writing system for medical access reports based on knowledge reasoning according to claim 1, characterized in that, The report generation module is specifically used for: The aforementioned argument blueprint is analyzed into a report structure outline containing multiple chapters and core arguments; For each chapter in the report structure outline, a natural language generation model is invoked, guided by the core arguments corresponding to the chapter, and the comprehensive evaluation conclusions and the analysis and reasoning results are extracted as content materials to generate the text of the chapter; Combine and format all the generated text from the aforementioned chapters to output the complete pharmaceutical access report.

10. A method for automatically generating pharmaceutical access reports based on knowledge reasoning, applied to a pharmaceutical access report automatic generation system based on knowledge reasoning as described in any one of claims 1-9, characterized in that, Includes the following steps: Collect multi-dimensional data sources and, based on user-inputted target market and target drug information, integrate the collected information related to the target market, laws, regulations, policy documents, and market data to construct and update the pharmaceutical access knowledge base; For the aforementioned pharmaceutical access knowledge base, an argumentation blueprint including argumentation strategies was developed; Guided by the aforementioned argumentation blueprint, the information in the aforementioned pharmaceutical access knowledge base is analyzed and reasoned, and the analysis and reasoning results are output. Based on the analysis and reasoning results, determine whether the target drug meets the access conditions and generate a comprehensive evaluation conclusion; Based on the comprehensive evaluation conclusions and the demonstration blueprint, a pharmaceutical access report is automatically generated and output.

Images