Method, device and electronic device for generating medical draft based on large model

Abstract

The embodiments of the present disclosure disclose a method, device and electronic device for generating a medical draft based on a large model. A specific implementation of the method includes: using a large language model to generate a multi-database search formula for medical research directions; executing the search formula to obtain a preliminary search document set and a cross-database causal network; screening and extracting the preliminary search document set to obtain a draft data document set and a clinical heterogeneity factor set; performing dynamic evidence chain fusion to obtain a fusion effect size set and a conflict marker set; executing the heterogeneity judgment script at the same time to obtain a visual chart set; performing standardized conflict resolution based on the cross-database causal network and the conflict marker set to obtain a conflict resolution report; using a large language model to integrate documents and charts and generate a medical draft based on the medical draft template and the above-mentioned conflict resolution report. This implementation method automatically generates medical drafts using a large model, reduces the impact of manual intervention, and improves the integrity of the content of the medical draft.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the field of computer technology, and in particular to a method, device, and electronic device for generating a medical draft based on a large model. Background Art

[0002] Currently, the medical field widely uses meta-analysis (Meta-Analysis) methods to generate systematic medical drafts. This involves complex steps, including literature search, data extraction, statistical analysis, and structural interpretation, to produce a medical draft. The typical approach to producing a medical draft using the meta-analysis method is to rely entirely on manual, step-by-step execution to complete the draft.

[0003] However, when using the above method to obtain a medical draft, the following technical problems often occur:

[0004] First, traditional meta-analysis takes a long time to screen the literature. When included studies have conflicting conclusions, there's a lack of a quantitative arbitration mechanism. Conflicting conclusions are manually determined based on subjective experience, resulting in low credibility in preliminary medical manuscripts.

[0005] Second, manual operators are prone to deviate from medical reporting standards, resulting in omissions in the extraction of key data and a lack of integrity in the content of the medical draft.

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the inventive concept and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the Invention

[0007] The content of this disclosure is used to briefly introduce concepts that will be described in detail in the detailed description section below. The content of this disclosure is not intended to identify key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0008] Some embodiments of the present disclosure propose a method, apparatus, and electronic device for generating a medical draft based on a large model to solve one or more of the technical problems mentioned in the above background technology section.

[0009] In a first aspect, some embodiments of the present disclosure provide a method for generating a medical draft based on a large model, comprising: using a large language model to generate a multi-database search formula corresponding to a medical research direction; executing the multi-database search formula in a database to perform a cross-database search to obtain a preliminary search document set and a cross-database causal network; screening and extracting preliminary search documents in the preliminary search document set according to the large language model and a medical bias assessment tool to obtain a draft data document set and a clinical heterogeneity factor set; performing dynamic evidence chain fusion to determine weights based on the draft data document set and the clinical heterogeneity factor set to obtain a fusion effect size set and a conflict marker set; executing a heterogeneity judgment script to perform heterogeneity judgment based on the fusion effect size set and the clinical heterogeneity factor set to obtain a visualization chart set; performing standardized conflict resolution based on the cross-database causal network and the conflict marker set to obtain a conflict resolution report; and determining the results of the draft data document set and the visualization chart set based on the medical draft template and the conflict resolution report to generate a medical draft.

[0010] In a second aspect, some embodiments of the present disclosure provide a device for generating a medical draft based on a large model, comprising: a search formula generation unit, configured to generate a multi-database search formula corresponding to a medical research direction using a large language model; a document retrieval unit, configured to execute the multi-database search formula in a database to perform a cross-database search, and obtain a preliminary search document set and a cross-database causal network; a document screening unit, configured to screen and extract preliminary search documents in the preliminary search document set according to the large language model and a medical bias assessment tool, and obtain a draft data document set and a clinical heterogeneity factor set; a conflict marking unit, configured to screen and extract preliminary search documents in the preliminary search document set according to the large language model and a medical bias assessment tool, and obtain a draft data document set and a clinical heterogeneity factor set; a conflict marking unit, configured to screen and extract preliminary search documents in the preliminary search document set according to the preliminary search document set The set and the above-mentioned clinical heterogeneity factor set are dynamically fused with the evidence chain to determine the weights, thereby obtaining a fused effect size set and a conflict marker set; the chart generation unit is configured to execute the heterogeneity judgment script based on the above-mentioned fused effect size set and the above-mentioned clinical heterogeneity factor set to perform heterogeneity judgment, thereby obtaining a visual chart set; the report generation unit is configured to execute standardized conflict resolution based on the above-mentioned cross-database causal network and the above-mentioned conflict marker set, thereby obtaining a conflict resolution report; the draft generation unit is configured to determine the results of the above-mentioned draft data document set and the above-mentioned visual chart set using the above-mentioned large language model based on the medical draft template and the above-mentioned conflict resolution report, thereby generating a medical draft.

[0011] In a third aspect, some embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by one or more processors, the one or more processors implement the method described in any implementation manner in the first aspect.

[0012] In a fourth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method described in any implementation manner in the first aspect is implemented.

[0013] The above-described embodiments of the present disclosure have the following beneficial effects: Medical drafts generated by the large-scale model-based medical draft generation method of some embodiments of the present disclosure reduce manual intervention and improve the completeness of the medical draft content. Specifically, the reason for the incompleteness of medical draft content is that the medical field generally uses meta-analysis methods to generate systematic drafts, which relies entirely on manual labor to gradually complete the draft, which is time-consuming. Due to manual intervention, international standards are deviated from, resulting in incomplete content in the generated medical draft. Based on this, the large-scale model-based medical draft generation method of some embodiments of the present disclosure first utilizes a large language model to generate multi-database search formulas corresponding to medical research directions. This allows for automatic adaptation to different database syntaxes, avoiding potential biases in manual strategy design. Second, the multi-database search formula is executed in the database to perform a cross-database search, resulting in a preliminary search document set and a cross-database causal network. Thus, the parallel cross-database search improves the document retrieval speed, while the cross-database causal network is used to quantify the causal relationships between documents, providing a basis for tracing the causes of conflicting conclusions. Next, using the large language model and the medical bias assessment tool, the preliminary retrieved literature from the preliminary search set is screened and extracted to obtain a preliminary draft data set and a set of clinical heterogeneity factors. This screening and extraction process yields objective bias assessment results and standardized preliminary draft data. Subsequently, dynamic evidence chain fusion is performed based on the preliminary draft data set and the clinical heterogeneity factor set to determine weights, resulting in a fused effect size set and a conflict marker set. This allows for the automatic identification of conflicting conclusions exceeding a threshold, serving as a basis for subsequent revisions to the preliminary medical draft. Secondly, a heterogeneity assessment script is executed based on the fused effect size set and the clinical heterogeneity factor set to determine heterogeneity, generating a set of visual charts. This reduces the errors associated with manually selected effect models and achieves standardized chart generation. Finally, standardized conflict resolution is performed based on the cross-database causal network and the conflict marker set to generate a conflict resolution report. This generates a structured chain of evidence by analyzing and locating the source of conflict through graph path analysis. Finally, based on the medical draft template and the conflict resolution report, the large language model is used to determine the results of the draft data document set and the visualization chart set, generating a medical draft. Therefore, the medical draft template is used to limit the completeness of the chapters. Conflicting conclusions are forcibly corrected through conflict reports. In summary, using a large model to generate medical drafts reduces the risk of missing content in medical drafts due to human intervention. By constructing a causal network, conflicting conclusions in medical drafts are systematically resolved, transforming human subjective experience judgments into structured objective analysis, thereby improving the credibility of medical drafts. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The above and other features, advantages, and aspects of the various embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0015] Figure 1 is a flowchart of some embodiments of a method for generating a medical draft based on a large model according to the present disclosure;

[0016] Figure 2 is a schematic structural diagram of some embodiments of a large model-based medical draft generation device according to the present disclosure;

[0017] Figure 3 It is a structural diagram of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0018] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are for illustrative purposes only and are not intended to limit the scope of protection of the present disclosure.

[0019] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure may be combined with each other.

[0020] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0021] It should be noted that the modifications of "one" and "multiple" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, they should be understood as "one or more".

[0022] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0023] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0024] refer to Figure 1, shows a process 100 of some embodiments of the method for generating a medical draft based on a large model according to the present disclosure. The method for generating a medical draft based on a large model includes the following steps:

[0025] Step 101: Generate multi-database search expressions corresponding to medical research directions using a large language model.

[0026] In some embodiments, the execution entity (e.g., an electronic device) of the aforementioned large-model-based medical manuscript generation method can be an automated medical meta-analysis generation system deployed on a cloud computing platform, comprising the following core components: a large language model engine, a cross-repository search module, a causal network builder, a dynamic evidence fusion processor, and a conflict resolution engine. The large language model engine can be trained using a large language model (e.g., GPT-4 Turbo) as a base model, overlaid with a medical domain adapter, and fine-tuned using abstracts from the Cochrane Library of Systematic Reviews and PubMed. The cross-repository search module can be a module that initiates queries against medical databases (e.g., PubMed and Embase) to search and screen literature. The causal network builder can be a module that extracts corresponding elements (e.g., study subjects and interventions) from included literature and constructs causal logic chains between these elements. The dynamic evidence fusion processor can automatically adjust the weight of each study based on clinical heterogeneity factors (e.g., age bias in the study population and differences in intervention dosage). The conflict resolution engine can be a module that locates the source of conflict in the causal network and outputs an interpretable analysis of the conflict causes. The above-mentioned large model can be an artificial intelligence model with a large number of parameters and the ability to learn massive amounts of data, and can include a large language model and a multimodal large model. The above-mentioned medical research direction can be a systematic research topic that requires the generation of a medical draft. For example, "Analyze the correlation between the incidence of acute delirium after gynecological surgery and the anesthesia method and patient age." The above-mentioned multi-database search formula can be a standardized literature query command executed simultaneously in multiple academic databases. As an example, first, the medical knowledge encoding ability of the large language model is used to extract the elements in the medical research direction. Finally, the keywords within each element are connected using the database's "OR", and different elements are connected using the database's "AND" to obtain a multi-database search formula.

[0027] In some optional implementations of some embodiments, the execution entity may utilize a large language model to generate a multi-database search formula corresponding to a medical research direction, which may include the following steps:

[0028] The first step is to extract keywords from the aforementioned medical research directions to obtain research keywords. These research keywords can be core terms and phrases extracted from the medical research directions that represent the research subject. For example, a word segmentation tool (e.g., Jieba) can be used to segment the medical research directions to obtain research keywords.

[0029] In the second step, using the above large language model, perform the following generation steps:

[0030] The first sub-step is to generate a single database search formula based on the research keywords. The single database search formula can be a structured query command designed for a single academic database. For example, semantic recognition of the research keywords is performed using a large language model to obtain research elements, which are then connected according to database query rules to generate a database search formula.

[0031] In the second sub-step, the single-database search formula is subjected to syntax conversion based on the database conversion document to generate a multi-database search formula. The database conversion document may be a mapping file storing search syntax rules for different academic databases. The multi-database search formula may be a query statement corresponding to a structured query command format designed for multiple academic databases. As an example, the single-database search formula is extracted and processed to construct a syntax tree. The conversion rule library in the database conversion document is loaded using a large language model, and the syntax tree is converted to obtain a multi-database search formula.

[0032] Step 102: executing a multi-database search formula in the database to perform a cross-database search, and obtaining a preliminary search document set and a cross-database causal network.

[0033] In some embodiments, the execution entity may execute a multi-database search formula in a database to perform a cross-database search, thereby obtaining a preliminary search document set and a cross-database causal network. The preliminary search document set may be a collection of documents obtained through preliminary searches of various databases. The cross-database causal network may be a causal network diagram constructed by analyzing the relationships between different documents (e.g., citation relationships and thematic associations). Nodes represent research elements of a document, and edges represent causal relationships or strong associations between them.

[0034] For example, a multi-database search formula can be executed on a medical database. The resulting medical literature can be used as a preliminary search document set. By extracting the content information of the preliminary search documents, we can obtain the element information within the documents. We then compare the element information relationships between different documents to form the relationships between element information and documents. Using element information as nodes and the relationships between element information and documents as edges, we construct a graph network structure, resulting in a cross-database causal network.

[0035] In some optional implementations of some embodiments, the execution entity may execute the multi-database search formula in a database to perform a cross-database search to obtain a preliminary search document set and a cross-database causal network, which may include the following steps:

[0036] The first step is to use an asynchronous function to execute the multi-database search formula in parallel to obtain a search document set. The asynchronous function can be a non-blocking function, allowing the program to perform other tasks while waiting for I / O operations. The search document set can be a collection of raw document results returned from each database.

[0037] As an example, first, initialize the database connector to establish a cross-platform connection. Then, set up the asynchronous execution function of the asynchronous task scheduling mechanism to implement parallel retrieval of multiple data sources. Finally, call the asynchronous execution function to execute the multi-database search formula to retrieve and integrate documents from multiple data sources in real time to obtain the retrieved documents.

[0038] The second step is to perform multi-dimensional standard screening on each search document in the above search document set to generate preliminary search documents and obtain a preliminary search document set. The above multi-dimensional standard can be a standard for screening documents from multiple dimensions. For example, in the scenario of generating a medical draft, the multi-dimensional standard can be "the full text of the document is accessible and searchable. The document contains all the keywords extracted from the selected medical research direction. The document does not fall into the scope of gray literature (for example, degree theses and conference records). The research time (based on the time of document inclusion) is within three years." The above preliminary search documents can be search documents that meet the multi-dimensional standards. The above preliminary search document set can be a collection of relevant documents after screening.

[0039] As an example, first, based on the screening criteria, the content of the corresponding search criteria field in the searched documents is extracted. Finally, the field content is compared with the standard requirements to filter the documents.

[0040] The third step is to extract elements from each of the preliminary retrieved documents in the preliminary retrieved document set to obtain a document element information set. The document element information includes the study subjects, intervention measures, outcome variables, and study design in the document. The document element information set can be a collection of all document element information. The study subjects in the document can be the characteristics of the population involved in the study (e.g., age, gender, and disease status). The intervention measures can be the specific treatment methods and procedures implemented on the study subjects (e.g., drug intervention and surgical intervention). The outcome variables can be the primary indicators measured in the study (e.g., survival rate and complication rate). The study design can be the type of research method (e.g., cohort study: an observational study design used to analyze epidemiology).

[0041] As an example, first, based on the information fields of the search elements, the element information of the corresponding part in the document is extracted. Finally, the extracted element information is integrated to form a document element information set.

[0042] The fourth step is to determine the correspondence between each document element in the document element information set and the preliminary search document set to obtain a causal network correspondence set. The causal network correspondence set can be a set of association relationships between document elements and documents.

[0043] As an example, a mapping between document elements and documents is created, and the mapping relationship information is used as the association relationship between the document elements and the documents to obtain a corresponding set of causal networks.

[0044] The fifth step is to generate the cross-database causal network based on the document element information in the corresponding set of the causal network and the corresponding search documents. The cross-database causal network can be a graph structure representing the causal relationship between different documents.

[0045] As an example, a network graph is created with document element information as nodes and the relationship between document element information and documents as edges, which are added to the network graph to obtain a cross-database causal network.

[0046] Step 103 , based on the large language model and the medical bias assessment tool, the preliminary search documents in the preliminary search document set are screened and extracted to obtain a draft data document set and a clinical heterogeneity factor set.

[0047] In some embodiments, the execution entity may screen and extract the preliminary retrieved documents from the preliminary search literature set based on a large language model and a medical bias assessment tool to obtain a preliminary data document set and a clinical heterogeneity factor set. The medical bias assessment tool may be a standardized medical research quality assessment tool (e.g., Cochrane ROB2) for assessing the risk of bias in a study. The clinical heterogeneity factor set may be a set of heterogeneity factors corresponding to clinical study subjects, intervention regimens, and outcome measures. These factors include study subject heterogeneity factors (baseline population characteristics: age, gender; disease status: disease severity); intervention regimen heterogeneity factors (intervention type: medication type, surgical method; dosage and duration: medication dosage, treatment frequency); and outcome measurement heterogeneity factors (outcome measure selection: differences between primary and secondary outcomes, e.g., survival rate as the primary outcome in Study A and quality of life score as the primary outcome in Study B).

[0048] As an example, we can first construct a medical knowledge graph using a preliminary literature search. We then expand related concepts based on medical research topics and construct a semantic relationship network between concepts. Secondly, we can use the knowledge graph and medical bias assessment tools to conduct a bias assessment. Furthermore, we can identify key clinical heterogeneity factors and obtain a set of clinical heterogeneity factors. Finally, we can extract the characteristics of the overall literature based on the knowledge graph to extract the literature content and obtain a preliminary data document set.

[0049] In some optional implementations of some embodiments, the execution entity may screen and extract the preliminary search documents in the preliminary search document set based on the large language model and the medical bias assessment tool to obtain a draft data document set and a clinical heterogeneity factor set, which may include the following steps:

[0050] The first step is to obtain a pre-generated data extraction template. The data extraction template can be a structured table containing standard fields required for meta-analysis or a predefined framework in JSON format.

[0051] The second step was to use the aforementioned bias assessment tool to assess the bias of the literature retrieved from the initial search, and to obtain bias assessment results. The bias assessment results were used to score the quality of the literature in specific risk areas (e.g., randomization and data completeness).

[0052] As an example, according to the standards of medical bias assessment, the medical bias assessment tool is used to score the literature in turn to obtain the assessment results.

[0053] In the third step, in response to the bias assessment result reaching a preset value, content extraction is performed on the document content in the preliminary searched documents to obtain an extracted content set. The preset value may be a threshold for document screening. The content extraction may be a process of obtaining structured data from the documents. The extracted content set may be a collection of structured data extracted from the documents according to the template content. The content extraction method may be to extract corresponding content from the documents according to the template fields.

[0054] Step 4: For each extracted content in the above extracted content set, perform the following preprocessing operations:

[0055] In the first sub-step, the large language model is used to extract clinical heterogeneity factors from the extracted content to obtain at least one clinical heterogeneity factor. For example, the large language model can be used to summarize factors that cause heterogeneity, and then the literature content can be read and the heterogeneity factors of the literature can be integrated to obtain a set of clinical heterogeneity factors for the literature.

[0056] The second sub-step involves populating the extracted content into the data extraction template to produce a preliminary data document. The preliminary data document can be a standardized, analyzable, structured dataset suitable for meta-analysis. For example, according to the field specifications defined in the data extraction template, the corresponding field content is found in the extracted content and the template is populated.

[0057] Step 104 : Based on the draft data document set and the clinical heterogeneity factor set, dynamic evidence chain fusion is performed to determine weights, thereby obtaining a fusion effect size set and a conflict marker set.

[0058] In some embodiments, the execution entity may perform dynamic evidence chain fusion based on the draft data document set and the clinical heterogeneity factor set to determine weights, thereby obtaining a fused effect size set and a conflict marker set. The dynamic evidence chain fusion may be an evidence fusion method that dynamically adjusts weights based on the research characteristics and heterogeneity factors in the draft data document. The fused effect size set may be a set of comprehensive effect indicators after integrating the research characteristics. The conflict marker set may be a research structure identifier with significant disagreement. The weights may be used to measure the contribution of the research characteristics to the comprehensive effect size based on the research characteristics and clinical heterogeneity factors, and are numerical expressions of the quantitative influence of each document. The comprehensive effect indicator may be a quantitative indicator that characterizes the intervention measures after integrating the results of multiple documents. The research structure identifier may be a structured label used to mark significant disagreements or conflicts in the literature. For example, in the scenario of generating a medical draft, the research structure identifier may include the conflict type and conflict location.

[0059] As an example, first, extract study characteristics (e.g., age distribution, symptom proportion) from the draft data set. Feature encoding and weighting are performed on the clinical heterogeneity factor set. For example, age (weight 0.6) and function (weight 0.4). Second, determine the heterogeneity score within the literature based on the study effect sizes. For example, a study characteristic might be "age > 75, hypertension." The clinical heterogeneity factor set might be {"age threshold": 75, "blood pressure": "hypertension"}. The heterogeneity score might be "age threshold × 0.6 + blood pressure × 0.4" = 1 × 0.6 + 1 × 0.4 = 1.0. Finally, determine the fused effect size based on the heterogeneity score. When the fused effect size exceeds a preset threshold (e.g., 0.7), the corresponding clinical heterogeneity factor is used as a conflict flag. For example, the heterogeneity score might be 1.0 > 0.7. The conflict flag might be {"age threshold": 75, "blood pressure": "hypertension"}.

[0060] In some optional implementations of some embodiments, the execution entity may perform dynamic evidence chain fusion based on the draft data document set and the clinical heterogeneity factor set to determine weights, thereby obtaining a fusion effect size set and a conflict marker set, which may include the following steps:

[0061] The first step is to extract the structured data corresponding to the research features in the above-mentioned draft data document set to obtain the research feature content set. The above-mentioned research features can be the content of structured variables that describe the core attributes of the research. For example, in the medical draft generation scenario, the core attributes of the research include: research design, participants, intervention measures, outcome measures and time factors. The above-mentioned research feature content set can be a collection of specific research features extracted from the draft data document set. For example, in the medical draft generation scenario, the above-mentioned specific research features can be structured variables that can standardize the description of the core attributes of a medical study. For example, structured variables can include the subjects, methods, outcomes and time range of the study.

[0062] As an example, the corresponding content may be extracted from the draft document data by mapping the research characteristic fields to obtain the research characteristic content.

[0063] The second step is to perform standardization on each clinical heterogeneity factor in the clinical heterogeneity factor set to obtain a heterogeneity factor dataset. The standardization process may be to convert the numerical values ​​of different types of heterogeneity factors into a unified numerical representation. The heterogeneity factor dataset may be a structured data set after standardization.

[0064] For example, in the scenario of generating a medical manuscript, different types of clinical heterogeneity factors are processed using corresponding methods to obtain data corresponding to the heterogeneity factors. For example, continuous factors can be processed using numerical extraction methods, categorical factors can be processed using one-hot encoding methods, and text factors can be processed using vectorized representation methods. The data corresponding to the heterogeneity factors are normalized to obtain standardized heterogeneity factor data.

[0065] The third step is to negatively correlate the study feature content set with the heterogeneity factor dataset using a preset heterogeneity weight function to obtain a weight set corresponding to the study feature content set. The preset heterogeneity weight function can be a mathematical function used to quantify the degree to which study features affect heterogeneity. The negative correlation can be based on an inverse model of study features and heterogeneity factors (e.g., the more ideal the study feature, the lower the heterogeneity, and the higher the weight). The weight set corresponding to the study feature content set can be a set of relative importance scores for each study feature in the meta-analysis.

[0066] As an example, in the scenario of generating a medical manuscript, the above mathematical function can be ,in, It can be the weight of the i-th research feature. It can be the i-th heterogeneity factor data. It can be the heterogeneity sensitivity coefficient in the medical field.

[0067] The fourth step is to determine the research effect size based on the aforementioned medical research direction. This research effect size can be a statistical indicator that quantifies the effectiveness of the intervention in the medical research direction. For example, in the context of generating a medical manuscript, these statistical indicators might be odds ratios (death / survival), hazard ratios (cohort studies), and standardized mean differences (continuous outcomes such as blood pressure). In practice, statistical indicators can be determined by analyzing the element information within the medical research direction.

[0068] In the fifth step, the above research effect sizes are dynamically weighted averaged according to the above weight set to obtain the above fused effect size set. The above fused effect size set can be a set of structured data with statistical characteristics as fields and effect sizes as values.

[0069] Step 6: For each fusion effect size in the fusion effect size set, in response to the fusion effect size reaching a preset threshold, determining the research feature content corresponding to the fusion effect size as a conflict flag. The conflict flag can be the research feature corresponding to the fusion effect size exceeding the threshold, which is the source of the conflict.

[0070] As an example, based on the value of the fusion effect, the research characteristics in the corresponding literature are determined to obtain the conflict marker. For example, in the scenario of generating a medical manuscript, the fusion effect set can be [RR=0.72 (I 2 =45%), RR=1.15 (I 2 =75%), RR=0.68 (I 2 =30%)]. The corresponding study characteristics can be: Study 1: {age: <70 years, dose: 15 mg, design: RCT}, Study 2: {age: >75 years, dose: 25 mg, design: cohort study}, Study 3: {age: 70-75 years, dose: 18 mg, design: RCT}. The preset threshold can be I 2 If the rate is >50%, a conflicting flag is triggered in Study 2. The study characteristics of Study 2 (age: >75 years, dose: 25 mg, design: cohort study) serve as conflicting flags.

[0071] Step 105: Execute a heterogeneity judgment script based on the fusion effect size set and the clinical heterogeneity factor set to perform heterogeneity judgment and obtain a visualization chart set.

[0072] In some embodiments, the execution subject can execute a heterogeneity judgment script based on the fusion effect size set and the clinical heterogeneity factor set to perform heterogeneity judgment and obtain a set of visualization charts. The heterogeneity judgment script can be an algorithm program for automatically calculating heterogeneity indicators. The visualization chart set can be a chart required for medical research. For example, in the scenario of generating a medical draft, the visualization charts can include: a forest chart showing the effect size of each study, a heterogeneity bar chart showing the source of heterogeneity, and a funnel chart for detecting publication bias. As an example, the fusion effect size set and the clinical heterogeneity factor set can be used as data sources, and the corresponding medical chart generation tool can be used to generate charts for the content required for generating the medical draft.

[0073] In some optional implementations of some embodiments, the execution entity may execute a heterogeneity judgment script based on the fusion effect size set and the clinical heterogeneity factor set to perform heterogeneity judgment, and obtain a visualization chart set, which may include the following steps:

[0074] The first step is to determine the effect sizes of the quantitative research findings corresponding to the fused effect size set, thereby obtaining an effect size dataset. The effect sizes of the quantitative research findings can be expressed as specific numerical values ​​(e.g., hazard ratios, standardized mean differences). The effect size dataset can be a structured set of fused effect size values ​​in the form of an array or list. For example, the structured fields (quantitative research findings) in the fused effect size set can be parsed to directly extract the effect size numerical results.

[0075] The second step is to determine the heterogeneity index of the clinical heterogeneity factor corresponding to the clinical heterogeneity factor set to obtain a heterogeneity index dataset. The heterogeneity index can be a statistic that quantifies heterogeneity. The heterogeneity index dataset can be a structured set of heterogeneity indexes as numerical values.

[0076] For example, in the medical manuscript generation scenario, the clinical heterogeneity factors could be ["age difference", "dose difference", "study design difference"]. The heterogeneity index in the literature could be ["age difference": 0.58, "dose difference": 0.27, "study design difference": 0.15]. The heterogeneity index dataset could then be [0.58, 0.27, 0.15].

[0077] The third step is to execute the pre-generated heterogeneity assessment script based on the effect size dataset and the heterogeneity indicator dataset to obtain the statistical value and heterogeneity index. The heterogeneity assessment script can be a program that calculates the heterogeneity statistic. In practice, the code generation function in the large language model can be used to generate the script program based on the requirements of heterogeneity assessment. The statistical value can be a numerical value that describes the degree of variation in effect sizes between studies. The heterogeneity index can be a numerical value that describes the percentage of heterogeneity in the total variation.

[0078] The fourth step is to determine a heterogeneous effect model based on the heterogeneity index and the effect size dataset. The heterogeneous effect model can be a statistical model selected based on the degree of heterogeneity.

[0079] For example, in the scenario of generating medical manuscripts, the heterogeneity index (I 2 ) < 34%, then the heterogeneity effect model can be a fixed effect model. 2 )>=34%, then the heterogeneous effect model can be a random-effects model. The fixed-effects model is a statistical model used for meta-analysis that assumes that the true effect size of all included studies is the same and that differences between studies are due solely to random sampling error. The random-effects model is a statistical model used for meta-analysis that assumes that the true effect size of each study is inherently different and is normally distributed around the overall mean, influenced by factors such as study population, intervention method, and setting.

[0080] Step 5: Generate a heterogeneity forest plot and a heterogeneity bar chart based on the heterogeneity effect model and the effect size dataset. The heterogeneity forest plot can display the effect sizes and confidence intervals for each study. The heterogeneity bar chart can display the distribution of heterogeneity sources. In practice, corresponding functions in relevant image generation libraries can be used to generate these charts.

[0081] Step 6: Generate a significance test chart based on the statistical value and the effect size dataset. The significance test chart can be a chart that displays the statistical test structure. For example, in the scenario of generating a medical manuscript, the significance test chart can be a heat map.

[0082] Step 7: Determine the heterogeneity forest map, the heterogeneity bar map, and the significance test chart as the visualization chart set.

[0083] Step 106 : performing standardized conflict resolution based on the cross-database causal network and the conflict marker set to obtain a conflict resolution report.

[0084] In some embodiments, the execution entity may perform standardized conflict resolution based on the cross-database causal network and the conflict marker set to generate a conflict resolution report. Standardized conflict resolution may be a process that uniformly resolves conflicts according to predefined rules or algorithms to reach a consistent conclusion. The conflict resolution report may include a detailed description of the conflict, the resolution method, the resolution structure, and recommended corrections.

[0085] As an example, first, based on prior knowledge in the medical field, the rules for resolving conflicts are determined (for conflicts in studies of different quality, high-quality research results are selected. For studies of the same quality, the majority of consistent research results are selected. In addition, the conclusions of clinical guidelines are selected). Secondly, each conflict in the conflict marker set is resolved, and the corresponding conflict resolution rules are applied to obtain the resolution results. Finally, the cross-database causal network is modified based on the resolution results. The resolved causal relationships are added, and a conflict resolution report is generated. For example, in the scenario of generating a medical draft, the conflict resolution report can include: an overview of the number and type of conflicts, conflict elements, specific content of the conflicts, resolution methods, resolution results, and an updated causal network summary.

[0086] In the process of adopting technical solutions to solve the above-mentioned technical problem 2, the following problems often arise. When medical literature from multiple sources gives contradictory conclusions on the same clinical problem, traditional manual methods are difficult to quantify the source of the conflict and cannot extract key content, resulting in the omission of key evidence, causing the conclusions of the medical draft to be contradictory and some content to be missing.

[0087] Conventional solutions to these problems typically rely on medical guidelines for arbitration. However, the inventors considered that differences in research methodologies across different research papers could lead to misjudgment of reasonable errors. Therefore, we decided to adopt the following solution.

[0088] In some optional implementations of some embodiments, the execution entity may perform standardized conflict resolution based on the cross-database causal network and the conflict marker set to obtain a conflict resolution report, which may include the following steps:

[0089] The first step is to extract the document nodes and associated edges in the cross-database causal network to obtain graph-structured data. The document nodes can be network nodes representing individual documents in the cross-database causal network. The associated edges can be causal relationships between documents (e.g., citation relationships and subject element relationships). The graph-structured data can be a graph data structure represented by nodes and edges (e.g., an adjacency list or adjacency matrix). For example, extraction tools in graph databases (e.g., Neo4j) can be used to extract nodes and edges from the cross-database causal network.

[0090] The second step is to determine the research features corresponding to the conflict marker set and obtain the research feature content. The research features can be the literature research features (research subjects, intervention measures) corresponding to the conflict markers. The research feature content can be the specific feature content corresponding to the conflict markers in the literature.

[0091] As an example, based on the PIOCOS (Population, Intervention, Comparison, Outcome, Study design) elements, natural language processing technology can be used to extract research features in the literature corresponding to the conflict markers.

[0092] The third step is to normalize the graph structure data according to preset rules to obtain causal network structured data. The causal network structured data can be graph data that is easy to process after normalization.

[0093] As an example, according to the rules in medical guidelines, node names are unified into common medical terms and edges are unified into predefined types (e.g., “cause”, “inhibit”).

[0094] The fourth step is to construct hierarchical operational instructions based on the causal network structure of the cross-database causal network and the research feature content to obtain enhanced instructions. The hierarchical operational instructions can be operational commands organized according to a certain processing order and hierarchical structure. For example, in the scenario of generating a medical manuscript, the hierarchical operational instructions can be a top-down operational process. First, the data is processed, then the logic is processed, and finally the instructions are generated. The enhanced instructions can be an operational method used to guide feature enhancement and optimization.

[0095] For example, in the scenario of generating a medical draft, the research feature may be "spinal anesthesia". Then, a subgraph related to "spinal anesthesia" is generated and extracted, and the content related to "spinal anesthesia" is analyzed and extracted.

[0096] Step 5: Based on the initial search literature set, identify the data structure of subgraphs within the causal network structured data to obtain key subgraph structures. These subgraphs can be partial graph structures extracted from the causal network that are related to specific relevant topics. For example, in the context of generating a medical manuscript, these key subgraph structures can be partial graphs related to the content corresponding to the conflict marker. These specific relevant topics can be research features corresponding to the conflict marker.

[0097] For example, in the scenario of generating a medical manuscript, a subgraph containing the nodes of these documents can be extracted from the causal network structured data based on the serial type number of the preliminary retrieved documents.

[0098] In the sixth step, based on a pre-defined conflict resolution rule base, the aforementioned enhancement instructions are used to enhance the features of the key subgraph structure, thereby generating causal network data. The pre-defined conflict resolution rule base can be a set of pre-defined conflict resolution rules (selecting high-quality research results when conflicting between studies of varying quality). The causal network data can be the graph data generated by adding new relationships (e.g., node weights, edge weights) to the subgraph.

[0099] For example, in the scenario of generating a medical manuscript, you can use the rules in the preset conflict resolution rule base (for example, adding a literature quality score to a node, adding an association strength to an edge) and perform specific operations based on reinforcement instructions (for example, merging nodes).

[0100] Step 7: Match the nodes and connections corresponding to the research feature content in the causal network data to obtain adjacent document nodes and adjacent document edge weights. The adjacent document nodes can be document nodes directly connected to the document involved in the conflict marker. The adjacent document edge weights can be the weight values ​​of the edges connecting the document nodes.

[0101] As an example, we can first find the corresponding nodes in the causal network data based on the keywords of the research feature content. Finally, we can obtain the weights of the first-order adjacent nodes and connecting edges.

[0102] In the eighth step, based on the preset rules for graph path analysis, the correlation between the adjacent document nodes and the adjacent document edge weights in the causal network structured data is determined to obtain the cause of the conflict. The preset rules for graph path analysis may be steps for analyzing graph structure types. The cause of the conflict may be the cause of the conflicting conclusions in the draft. For example, in the scenario of generating a medical draft, the cause of the conflict may be differences in the research population or intervention measures. The preset rule for graph path analysis may be the shortest path.

[0103] As an example, in the scenario of generating medical manuscripts, graph algorithms (e.g., community detection) can be used to analyze the paths and weights between adjacent document nodes to infer the source of the conflict (e.g., two documents with opposite conclusions may belong to different communities).

[0104] The ninth step is to construct a conflict evidence chain based on the conflict causes, the research characteristics, and the causal network data. The conflict evidence chain can be a chain of evidence that connects the conflict causes, the characteristics of the relevant literature, and the network relationships.

[0105] As an example, in the scenario of generating a medical manuscript, the cause of the conflict can be linked to specific research characteristics (e.g., research subjects, intervention dose) and paths in the causal network data to form a chain of evidence (e.g., the conflicting conclusions between literature A and literature B are due to the different ages of the research subjects. Literature A targets <70 years old. Literature B targets >75 years old).

[0106] In the tenth step, the conflict evidence chain is converted into a structured conflict resolution report using a preset template inference engine. The preset template inference engine can be a pre-designed report template with a standard structure. The conflict resolution report can be a structured data result obtained by filling in the conflict resolution structure in a fixed format.

[0107] As an example, according to a preset template, the content in the conflict evidence chain can be filled into the corresponding position in the template to generate a report including the various parsed contents of the conflict.

[0108] The above-mentioned operation steps, as an inventive point of the present disclosure, solve the second technical problem mentioned in the background technology, "human operators are prone to deviate from the medical report specifications, resulting in omissions in the extraction of key data, and the lack of integrity of the content of the medical draft". In practice, conventional methods rely on medical guideline arbitration. The differences in research methods of different research documents lead to the misjudgment of reasonable errors, the inability to extract key content, the inability to connect the previous and subsequent logics in series, and the lack of integrity of the content of the medical draft. The present disclosure designs a solution for the dynamic evidence chain fusion and marking of conflicting conclusions, which fuses the conflicting conclusions through the dynamic evidence chain and performs objective attribution through a cross-database causal network. Finally, a structured conflict resolution report is generated. As the basis for the conclusion conflict, it solves the problem of missing content in the draft caused by manual deviation from the specifications, and ensures the integrity of the upper and lower evidence chains and content of the draft conclusions.

[0109] In the process of adopting technical solutions to solve the above-mentioned technical problem 2, the following problems often arise: conflicting conclusions are scattered in different documents, and it is difficult to manually construct a complete chain of evidence.

[0110] Conventional solutions to these problems typically use large language models to analyze the connection between literature and research topics. The inventors considered using large language models to analyze the relationship between literature and research directions. However, they discovered that the model's data, drawn from a wide range of sources, can easily deviate from the target literature set, leading to conflicting conclusions that are insufficiently relevant to the research topic. Therefore, they decided to adopt the following solution.

[0111] Optionally, the execution subject may construct hierarchical operation instructions based on the causal network structure of the cross-database causal network and the research feature content to obtain enhanced instructions, which may include the following steps:

[0112] The first step is to construct a subject-object relationship for the research features, generating a set of feature triples. This subject-object relationship can be a semantic relationship between the subject (executor) and the object (receiver). This feature triple set can be a structured relational data set (subject-relationship-object). For example, in the context of generating a medical manuscript, the semantic relationship might be "drug-effect-disease."

[0113] As an example, the research features in the research feature content may be parsed and constructed in the form of triples.

[0114] The second step is to determine the key node and edge weights of the causal network structure to obtain a set of quantitative indicators. This set of quantitative indicators can be a set of quantitative evaluation results for the nodes and edges. As an example, edge weights can be used as the basis for quantitative evaluation. For example, in the scenario of generating a medical draft, the edge weights might be {("neuraxial anesthesia", "delirium incidence"): 0.82, ("general anesthesia", "delirium incidence"): 0.76}. The quantitative indicators could then be 0.82, 0.76.

[0115] The third step is to perform the following instruction generation steps for each feature triple in the above feature triple set:

[0116] In the first sub-step, based on the quantitative indicator set, the characteristic triples are mapped to network nodes with quantitative indicators to obtain characteristic network nodes and conflict quantitative indicators. The characteristic network nodes may be nodes corresponding to the triples in the causal network structure. The conflict quantitative indicators may be conflict intensity indicators associated with the network nodes.

[0117] As an example, first, we can map the subject and object in a feature triple to nodes in a causal network to obtain a feature network node. Next, we can compare the relationship strength of the triple (for example, if the initial relationship strength is a strong association of "1") with the edge weights of the network to determine the degree of difference, which serves as a quantitative indicator of conflict.

[0118] The second sub-step is to extract subgraphs from the causal network structure based on the characteristic network nodes to obtain conflict subgraphs, wherein the conflict subgraphs may be local network structures including conflicting nodes.

[0119] As an example, we can first extract directly connected upstream and downstream nodes, centering on a characteristic network node, to obtain first-level nodes. Next, we extract second-level nodes that have causal connections with the characteristic network node. Finally, we generate a local network structure, including the characteristic network node, based on the first-level and second-level nodes.

[0120] In the third sub-step, based on a preset feature enhancement rule library, instructions are encapsulated for the feature network nodes, the conflict quantification indicators, and the conflict subgraph to obtain enhancement instructions. The preset feature enhancement rule library may be predefined rules for handling conflicts. The instruction encapsulation may be a process of creating executable operation commands.

[0121] As an example, first, preset rules can be matched based on conflict and type, resulting in rule types and resolution suggestions. Second, the conflicts and conflict subgraphs corresponding to the conflict quantification indicators are added to the instruction template. Finally, executable operation commands are generated. For example, in the scenario of generating a medical manuscript, the above rule types may include: supplementing data types for specific types of studies; adjusting the method type of the statistical model; and adding clinical types for subgroup analysis.

[0122] The above-mentioned operation steps, as an inventive point of the present disclosure, solve the second technical problem mentioned in the background technology and "conflicting conclusions are scattered in different documents, and it is difficult to manually construct a complete chain of evidence." In practice, using a large language model to analyze the relationship between documents and research directions may easily deviate from the target document set, resulting in insufficient relevance between the conflicting conclusions and the research topic. The present disclosure designs a solution for quantitative attribution of conflicts based on causal networks, which locates the nodes and associated edges in the causal network by constructing conflict triples. The subgraphs are extracted and matched according to the nodes to preset rules to generate executable instructions and encapsulate them. Therefore, by determining the document connections and executable instructions of the conflicting conclusions, the problems of fragmentation of conflicting conclusion documents and broken chains of evidence are solved.

[0123] Step 107 : Based on the medical draft template and the conflict resolution report, a large language model is used to determine the results of the draft data document set and the visualization chart set to generate a medical draft.

[0124] In some embodiments, the execution entity can use a large language model to determine the results of the draft data document set and the visual chart set based on the medical draft template and the conflict resolution report to generate a medical draft. As an example, first, the medical draft template and conflict resolution are analyzed using a large language model to obtain a structured medical draft template (e.g., imaging manifestations), and the conflict resolution report is cross-modally aligned with the structured medical draft template. The medical draft template includes semantic slots that can be filled (e.g., "method-sample size", "conclusion-contradiction") to facilitate the filling of the draft conclusion content and conflict content. Finally, keywords (e.g., disease entities) and related nodes (e.g., medical terms, experimental conclusions) are extracted from the draft data document set, the keywords are combined with the medical draft template, and the retrieval enhancement module in the large language model is used to generate the output medical draft and automatically insert objective analysis.

[0125] While employing technical solutions to address the aforementioned technical issue, the following issues often arise: Generating medical drafts using the retrieval enhancement module of a large language model often results in biases. The revised content loses semantic coherence with the context, leading to a decline in the overall quality of the medical draft.

[0126] Conventional solutions to these problems typically involve manual review and iterative optimization of the draft medical manuscript based on prompts. However, the inventors considered using prompts to optimize the draft medical manuscript as a whole. However, large language models might overfit the local standards of manual prompts, incorrectly correcting information that should be retained in the draft medical manuscript. This could lead to missing content and factual errors in the draft medical manuscript. We decided to adopt the following solution.

[0127] Optionally, in some optional implementations of some embodiments, the execution entity may determine the results of the draft data document set and the visualization chart set using the large language model based on the medical draft template and the conflict resolution report to generate the medical draft. The method may further include the following steps:

[0128] First, for the above medical draft, perform the following draft optimization steps:

[0129] In the first sub-step, in response to the problem that the medical draft does not contain modification suggestions fed back by the manual review terminal, conflict feature analysis is performed on the conflict resolution report to obtain a conflict feature information set.

[0130] The conflict feature information set may be a set of structured problem features extracted from feature reports and that can be processed by a machine.

[0131] For example, the natural language descriptions in conflict reports can be parsed to extract key features and store them in a structured format. For example, in a medical manuscript generation scenario, conflict feature information might be {"conflict type":"data contradiction","location":"RESULTS_SECTION_p3","related entities":["Dose A","1.2%"]}.

[0132] In the second sub-step, for the conflict feature information set, the following draft generation steps are performed:

[0133] Sub-step one: input the above-mentioned conflict feature information set, the above-mentioned medical draft and the adjustment prompt words for the medical draft into a medical draft optimization model pre-trained based on medical knowledge to obtain the medical draft modification position information set, the modified medical draft and the adjustment length ratio.

[0134] The adjustment prompt word may be an instruction for determining the direction of model modification. The medical draft optimization model may be a specialized model pre-trained based on medical knowledge. The medical draft modification location information set may be a set of location information of the content to be modified in the medical draft. The revised medical draft may be a medical draft after the modified content of the medical draft has been modified and optimized according to the adjustment prompt word. The adjusted length ratio may be the ratio of the modified content to the entire medical draft.

[0135] Sub-step two, based on the above-mentioned medical draft modification position information set, the modified medical draft, the above-mentioned medical draft and the local semantic adjustment prompt words, use the medical draft semantic verification model to generate local semantic verification information corresponding to the above-mentioned modified medical draft, wherein the above-mentioned medical draft optimization model and the above-mentioned medical draft semantic verification model are synchronously trained based on the training set of medical knowledge.

[0136] The local semantic adjustment prompt may be a modification guidance instruction focused on the context of the medical manuscript content corresponding to the modification location (e.g., "check whether the pronoun reference at the modification location is clear"). The local semantic verification information may be a semantic verification result of the context of the modified content of the medical manuscript. The medical manuscript semantic verification model may be based on the medical modification location information, extract the content at the local location corresponding to the modified medical manuscript and the medical manuscript, expand the content separately (e.g., context content expansion), and then compare the two expanded local manuscript contents to determine whether they are semantically identical.

[0137] As an example, the context text of the modified location of the medical manuscript can be extracted and input into the medical manuscript semantic verification model to determine whether the modified sentence is naturally connected with the context.

[0138] Sub-step three, in response to determining that the above-mentioned local semantic verification information indicates that the local semantic adjustment is correct and the above-mentioned adjustment length ratio is higher than the target ratio, based on the above-mentioned medical manuscript modification position information set, the revised medical manuscript, the above-mentioned medical manuscript and the overall semantic adjustment prompt words, the medical manuscript semantic verification model is used to generate the overall semantic verification information corresponding to the above-mentioned revised medical manuscript.

[0139] The target ratio may be a threshold that triggers the overall revision of the medical manuscript. The overall semantic adjustment prompt may be an instruction guiding the revision of the entire medical manuscript. The overall semantic verification information may be the semantic verification result of the revised medical manuscript.

[0140] Sub-step four: in response to determining that the overall semantic verification information indicates that the overall semantic adjustment is correct or the adjusted length ratio is not higher than the target ratio, a conflict resolution report of the revised medical manuscript is generated as a target conflict resolution report.

[0141] The target conflict analysis report may be a conflict report generated to evaluate the modification effect of the revised medical manuscript, and to detect whether there is any conflicting content that has not been modified and optimized.

[0142] As an example, a revised conflict resolution report may be obtained by traversing and comparing the revised parts in the revised medical manuscript with the conflict feature information set.

[0143] Sub-step 5: In response to determining that the target conflict resolution report indicates that there is no target conflict feature information set, the modified medical draft is sent to a target review terminal for manual review. The target review terminal may be a terminal where a review interface used by medical experts is located.

[0144] Sub-step 6: In response to receiving the review approval information sent by the target review terminal, the revised medical draft is determined as the medical draft. The review approval information can be the review opinion (e.g., text message, button confirmation signal) given by the medical expert after the review.

[0145] The second sub-step is to generate a target conflict feature information set in response to determining that the target conflict resolution report indicates the existence of a target conflict feature information set, wherein the target conflict feature information set may be unresolved conflict features in the revised medical manuscript.

[0146] As an example, the content of the target conflict feature information set can be traversed to determine whether it contains unresolved conflict features.

[0147] In the third sub-step, the target conflict feature information set is used as the conflict feature information set, the revised medical draft is determined as the medical draft, and the draft generation step is continued.

[0148] In the second step, in response to receiving the review annotation information sent by the target review terminal, the review annotation information is parsed to obtain a problem feature information set. The review annotation information may be revision comments provided by medical experts after reviewing and revising the medical manuscript. The problem feature information set may be a collection of structured problem feature information that can be processed by machines after parsing the review annotation information.

[0149] For example, natural language processing techniques can be used to parse the review comments to generate a parsed result. Finally, the parsed result is converted into a structured format for conflict features. For example, in a medical manuscript generation scenario, the review comment might be "Discuss the comparison of supplementary drug dosages of drug A." The question feature information might be {"Requirement Type": "Content Expansion", "Location": "DISCUSSION_SECTION"}.

[0150] In the third step, the above-mentioned problem feature information set is used as the conflict feature information set, the revised medical draft is determined as the medical draft, and the above-mentioned draft generation steps are continued.

[0151] Optionally, the above-mentioned execution entity can input the above-mentioned conflict feature information set, the above-mentioned medical draft and the adjustment prompt words for the medical draft into a medical draft optimization model pre-trained based on medical knowledge to obtain the medical draft modification position information set, the modified medical draft and the adjustment length ratio.

[0152] The medical draft optimization model can be based on a Transformer and medical knowledge graph encoder architecture, including a multimodal input layer, a medical knowledge enhancement layer, an instruction-following module, and an output layer. The multimodal input layer can be a network layer used to fuse text features (e.g., draft text, conflicting features) with structured data (e.g., conflicting locations, entity relationships). The medical knowledge enhancement layer can be a network layer that injects medical domain knowledge through a knowledge graph to resolve entity ambiguity. The instruction-following module can be a human feedback reinforcement learning optimization module that ensures the model accurately understands adjustment prompts. The output layer can be a network layer that generates modification suggestions (e.g., location information, replacement text) and determines the proportion of the modified text.

[0153] As an example, the conflicting feature information set and the first medical draft are input into the multimodal input layer. First, the multimodal input layer converts the conflicting feature information into a structured conflict feature vector, including the conflict type, study design, and effect size. The first medical draft is converted into a sequence of word vectors while preserving the semantic information of the original text. Second, the multimodal input layer uses a multi-head attention mechanism to determine the correlation between the text features in the word vector sequence and the conflicting feature vector, and adds absolute and relative position encodings. Finally, a fused feature vector is generated, which fuses the text features and the conflicting position information. This fused feature vector is then input into the medical knowledge enhancement layer. The medical knowledge enhancement layer uses the knowledge graph to identify medical entities in the first medical draft and the conflicting feature information set, and aligns the identified content information to resolve ambiguity. This generates knowledge enhancement features (each knowledge enhancement feature includes fused knowledge graph features and medical entity association information). The knowledge enhancement features and adjustment prompts for the first medical draft are input into the instruction following module. The instruction encoder can be used to convert the adjustment prompts into task vectors, and the prompts can be optimized by dynamically adjusting model parameters through human feedback reinforcement learning (e.g., RLHF). An optimized feature vector guided by the prompt word is obtained. This optimized feature vector is input to the output layer. A classifier can be used to predict the scope of the modified text to obtain the location of the modification. A decoder is then used to generate optimized content based on the context, resulting in a revised medical manuscript. The adjusted length ratio is determined by determining the ratio of the original text length to the modified text length.

[0154] Optionally, the execution entity may utilize a medical draft semantic verification model based on the set of modified medical draft location information, the modified medical draft, the medical draft, and the local semantic adjustment prompt words to generate local semantic verification information corresponding to the modified medical draft. The medical draft optimization model and the medical draft semantic verification model are synchronously trained based on a training set of medical knowledge.

[0155] The semantic verification model for medical manuscripts can be a comparative learning model using dual encoders (e.g., a Siamese Network), comprising a local encoder, a global encoder, a contrastive learning module, and a verification rule engine. The local encoder can be used to extract contextual features before and after the modification. The global encoder can be implemented using the BERT architecture to capture semantic features of the document. The contrastive learning module can be used to determine semantic similarity (e.g., cosine similarity) before and after the modification. The verification rule engine can be based on built-in medical text specifications and generate structured verification reports.

[0156] The above-mentioned shared data sets can be derived from the full text and revision records of PubMed medical literature, Cochrane systematic review reports and expert annotations, clinical research plans, and documents with modification suggestions given by the committee.

[0157] As an example, the set of revision location information for the medical manuscript, the revised medical manuscript, the medical manuscript, and local semantic adjustment prompts are preprocessed. Based on the revision location information, the revised local segments of the revised medical manuscript and the original content of the medical manuscript are extracted to obtain the original and revised local text segments. These segments are converted into text feature vectors. The local semantic adjustment prompts are also converted into feature vectors and fused with the text feature vectors to obtain local semantic prompt feature vectors. The original and revised local text segments are input into a local encoder. The local encoder can extract medical academic semantics using a pretrained BERT model and capture the semantic dependencies between the preceding and following sections using a self-attention mechanism to obtain feature vectors, thereby obtaining the original and revised local feature vectors. The revised medical manuscript and the medical manuscript are then input into a global encoder. The global encoder can encode the full text using a pretrained BERT model, obtain document-level semantic identifiers, and identify the semantic associations between the revision locations and other sections to obtain feature vectors, thereby obtaining the original and revised full text semantic vectors. The local semantic hint word vector, the original local feature vector, the modified local feature vector, the original full-text semantic vector, and the modified full-text semantic vector are input into the contrastive learning module to determine the similarity between the local semantic similarity and the global semantic consistency (for example, cosine similarity). The comprehensive semantic similarity and local feature similarity are obtained. The comprehensive semantic similarity, local feature similarity, and the set of medical draft modification position information are input into the verification rule engine. The verification rule engine can be an adaptive matching rule based on the built-in medical text specification. Based on the verification of the comprehensive semantic similarity and the rule threshold, combined with the corresponding medical draft modification position information, structured information containing the verification conclusion, deviation cause, and modification suggestion is generated to obtain local semantic verification information.

[0158] The above steps, as an inventive feature of this disclosure, address the first technical issue mentioned in the background art: "When conflicting research conclusions are included, there is a lack of a quantitative arbitration mechanism, and manual reliance on subjective experience is used to determine conflicts, resulting in low credibility of the conclusions in the draft medical manuscript." In practice, using the retrieval enhancement module of a large language model to generate draft medical manuscripts can lead to a break in the semantic coherence between the modified content and the context, resulting in a decline in the overall quality of the draft medical manuscript. This disclosure designs a three-level verification mechanism and a dynamic feedback draft optimization scheme. By structuring the conflict resolution report, the location of the modified content is determined. Modifications are made to the modified content and the semantic context is verified and judged. The percentage of the modified content in the full text is then determined. Based on a preset full-text percentage threshold, the full text of the modified medical manuscript is semantically verified to determine structural coherence. Drafts that pass semantic verification are then submitted for manual review, and continuous iterative optimization is performed based on the review comments and conflict reports to produce a standard draft medical manuscript. Therefore, to reduce the overfitting of the large model to the manual prompt words and the inadvertent deletion of correct information in the draft medical manuscript, this mechanism is used. A three-level verification mechanism and dynamic feedback are used to optimize the first draft. The contextual coherence of the modified content is verified through semantic analysis. The semantic logic verification of the full text is triggered based on the ratio threshold of the modified content. The medical first draft that passes the semantic logic verification of the full text is submitted to human workers for final review to iteratively optimize the medical first draft and obtain a complete medical first draft file.

[0159] The above-described embodiments of the present disclosure have the following beneficial effects: Medical drafts generated by the large-scale model-based medical draft generation method of some embodiments of the present disclosure reduce manual intervention and improve the completeness of the medical draft content. Specifically, the reason for the incompleteness of medical draft content is that the medical field generally uses meta-analysis methods to generate systematic drafts, which relies entirely on manual labor to gradually complete the draft, which is time-consuming. Due to manual intervention, international standards are deviated from, resulting in incomplete content in the generated medical draft. Based on this, the large-scale model-based medical draft generation method of some embodiments of the present disclosure first utilizes a large language model to generate multi-database search formulas corresponding to medical research directions. This allows for automatic adaptation to different database syntaxes, avoiding potential biases in manual strategy design. Second, the multi-database search formula is executed in the database to perform a cross-database search, resulting in a preliminary search document set and a cross-database causal network. Thus, the parallel cross-database search improves the document retrieval speed, while the cross-database causal network is used to quantify the causal relationships between documents, providing a basis for tracing the causes of conflicting conclusions. Next, using the large language model and the medical bias assessment tool, the preliminary retrieved literature from the preliminary search set is screened and extracted to obtain a preliminary draft data set and a set of clinical heterogeneity factors. This screening and extraction process yields objective bias assessment results and standardized preliminary draft data. Subsequently, dynamic evidence chain fusion is performed based on the preliminary draft data set and the clinical heterogeneity factor set to determine weights, resulting in a fused effect size set and a conflict marker set. This allows for the automatic identification of conflicting conclusions exceeding a threshold, serving as a basis for subsequent revisions to the preliminary medical draft. Secondly, a heterogeneity assessment script is executed based on the fused effect size set and the clinical heterogeneity factor set to determine heterogeneity, generating a set of visual charts. This reduces the errors associated with manually selected effect models and achieves standardized chart generation. Finally, standardized conflict resolution is performed based on the cross-database causal network and the conflict marker set to generate a conflict resolution report. This generates a structured chain of evidence by analyzing and locating the source of conflict through graph path analysis. Finally, based on the medical draft template and the conflict resolution report, the large language model is used to determine the results of the draft data document set and the visualization chart set, generating a medical draft. Therefore, the medical draft template is used to limit the completeness of the chapters. Conflicting conclusions are forcibly corrected through conflict reports. In summary, using a large model to generate medical drafts reduces the risk of missing content in medical drafts due to human intervention. By constructing a causal network, conflicting conclusions in medical drafts are systematically resolved, transforming human subjective experience judgments into structured objective analysis, thereby improving the credibility of medical drafts.

[0160] Further references Figure 2As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a device for generating a medical draft based on a large model. These device embodiments are similar to Figure 1 Corresponding to the method embodiments shown, the large model-based medical draft generation device can be specifically applied to various electronic devices.

[0161] like Figure 2 As shown, a large-scale model-based medical manuscript generation device 200 includes a search formula generation unit 201, a literature search unit 202, a literature screening unit 203, a conflict marking unit 204, a chart generation unit 205, a report generation unit 206, and a draft generation unit 207. The search formula generation unit 201 is configured to use a large language model to generate a multi-database search formula corresponding to a medical research direction. The literature search unit 202 is configured to execute the multi-database search formula in a database to perform a cross-database search, thereby obtaining a preliminary search document set and a cross-database causal network. The literature screening unit 203 is configured to screen and extract preliminary search documents from the preliminary search document set based on the large language model and a medical bias assessment tool, thereby obtaining a preliminary data document set and a clinical heterogeneity factor set. The conflict marking unit 204 is configured to perform dynamic evidence chain fusion based on the preliminary data document set and the clinical heterogeneity factor set to determine weights, thereby obtaining a fused effect size set and a conflict marking set. The chart generation unit 205 is configured to execute a heterogeneity assessment script based on the fused effect size set and the clinical heterogeneity factor set to determine heterogeneity, thereby generating a set of visualization charts. The report generation unit 206 is configured to perform standardized conflict resolution based on the cross-database causal network and the conflict marker set to generate a conflict resolution report. The draft generation unit 207 is configured to utilize the large language model to determine the results of the draft data document set and the visualization chart set based on the medical draft template and the conflict resolution report, thereby generating a medical draft.

[0162] It is understandable that the units recorded in the large model-based medical draft generation device 200 and the reference Figure 1 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the large model-based medical draft generation device 200 and the units contained therein, and will not be repeated here.

[0163] Reference below Figure 3 , which shows a structural schematic diagram of an electronic device (eg, an electronic device) 300 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0164] like Figure 3 As shown, electronic device 300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes based on programs stored in a read-only memory (ROM) 302 or programs loaded from a storage device 308 into a random access memory (RAM) 303. RAM 303 also stores various programs and data required for the operation of electronic device 300. Processing device 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to bus 304.

[0165] Typically, the following devices may be connected to the I / O interface 305: an input device 306 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 308 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 309. The communication device 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Figure 3 The electronic device 300 is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead. Figure 3 Each block shown in the figure may represent one device, or may represent multiple devices as needed.

[0166] In particular, according to some embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program comprising program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from a network via the communication device 309, or installed from the storage device 308, or installed from the ROM 302. When the computer program is executed by the processing device 301, the above-mentioned functions defined in the method of some embodiments of the present disclosure are performed.

[0167] It should be noted that in some embodiments of the present disclosure, the computer-readable medium described above may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. Computer-readable storage media may include, but are not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or components, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In some embodiments of the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device, or component. Furthermore, in some embodiments of the present disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. This propagated data signal may take a variety of forms, including, but not limited to, electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wire, optical cable, RF (radio frequency), or any suitable combination thereof.

[0168] In some embodiments, the client and server can communicate using any currently known or later developed network protocol, such as HTTP (HyperText Transfer Protocol), and can be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or later developed network.

[0169] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: generates a multi-database search formula corresponding to the medical research direction using a large language model; executes the multi-database search formula in the database to perform a cross-database search to obtain a preliminary search document set and a cross-database causal network; screens and extracts preliminary search documents in the preliminary search document set based on the large language model and the medical bias assessment tool to obtain a draft data document set and a clinical heterogeneity factor set; performs dynamic evidence chain fusion based on the draft data document set and the clinical heterogeneity factor set to determine weights to obtain a fused effect size set and a conflict marker set; executes a heterogeneity judgment script based on the fused effect size set and the clinical heterogeneity factor set to perform heterogeneity judgment to obtain a visualization chart set; performs standardized conflict resolution based on the cross-database causal network and the conflict marker set to obtain a conflict resolution report; and uses the large language model to determine the results of the draft data document set and the visualization chart set based on the medical draft template and the conflict resolution report to generate a medical draft.

[0170] Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0171] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0172] The units described in some embodiments of the present disclosure may be implemented in software or in hardware. The units described may also be provided in a processor. For example, they may be described as follows: a processor including a search formula generation unit, a document search unit, a document screening unit, a conflict marking unit, a chart generation unit, a report generation unit, and a draft generation unit. The names of these units do not, in some cases, constitute limitations on the units themselves. For example, the search formula generation unit may also be described as a "unit that utilizes a large language model to generate multi-database search formulas corresponding to medical research directions."

[0173] The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0174] The above descriptions are merely some preferred embodiments of the present disclosure and illustrate the underlying technical principles. Those skilled in the art should understand that the scope of the invention encompassed by the embodiments of the present disclosure is not limited to technical solutions formed by specific combinations of the aforementioned technical features. It also encompasses other technical solutions formed by any combination of the aforementioned technical features or their equivalents, without departing from the aforementioned inventive concept. For example, a technical solution formed by replacing the aforementioned features with (but not limited to) technical features with similar functions disclosed in the embodiments of the present disclosure.

Claims

1. A method for generating a medical draft based on a large model, comprising: Use a large language model to generate multi-database search expressions corresponding to medical research directions; Execute the multi-database search formula in the database to perform a cross-database search, and obtain a preliminary search document set and a cross-database causal network; Based on the large language model and the medical bias assessment tool, the preliminary search documents in the preliminary search document set are screened and extracted to obtain a draft data document set and a clinical heterogeneity factor set; According to the draft data document set and the clinical heterogeneity factor set, dynamic evidence chain fusion is performed to determine weights to obtain a fused effect size set and a conflict mark set, wherein the dynamic evidence chain fusion is performed to determine weights to obtain a fused effect size set and a conflict mark set according to the draft data document set and the clinical heterogeneity factor set, including: extracting structured data corresponding to the research features in the draft data document set to obtain a research feature content set; standardizing each clinical heterogeneity factor in the clinical heterogeneity factor set to obtain a heterogeneity factor data set; using a preset heterogeneity weight function, negatively correlating the research feature content set and the heterogeneity factor data set to obtain a weight set corresponding to the research feature content set; determining a research effect size according to the medical research direction; performing dynamic weighted averaging on the research effect size according to the weight set to obtain the fused effect size set; for each fused effect size in the fused effect size set, in response to the fused effect size reaching a preset threshold, determining that the research feature content corresponding to the fused effect size is a conflict mark; According to the fusion effect size set and the clinical heterogeneity factor set, executing a heterogeneity judgment script to perform heterogeneity judgment, and obtaining a visualization chart set; performing standardized conflict resolution based on the cross-database causal network and the conflict marker set to obtain a conflict resolution report, wherein the standardized conflict resolution is a process of uniformly handling conflicts according to predefined rules or algorithms to obtain a consistent conclusion, and the conflict resolution report is a report including a detailed description of the conflict, the resolution method, the resolution structure, and correction suggestions; According to the medical draft template and the conflict resolution report, the large language model is used to determine the results of the draft data document set and the visualization chart set to generate a medical draft.

2. The method according to claim 1, wherein The method utilizes a large language model to generate multi-database search formulas corresponding to medical research directions, including: Perform keyword extraction on the medical research direction to obtain research keywords; Using the large language model, the following generation steps are performed: Generate a single database search formula based on the research keywords; According to the database conversion document, the single database search formula is subjected to syntax conversion to generate a multi-database search formula.

3. The method according to claim 1, wherein The executing the multi-database search formula in the database to perform a cross-database search to obtain a preliminary search document set and a cross-database causal network includes: Utilizing an asynchronous function, executing the multi-database search formula in parallel to obtain a search document set; Performing multi-dimensional standard screening on each search document in the search document set to generate preliminary search documents, thereby obtaining a preliminary search document set; Extracting elements from each preliminary search document in the preliminary search document set to obtain a document element information set, wherein the document element information includes: research subjects, intervention measures, outcome variables, and research design in the document; For each document element information in the document element information set, determining a corresponding relationship between the document element information and the preliminary search document set to obtain a causal network corresponding set; The cross-database causal network is generated according to the document element information in the corresponding set of the causal network and the corresponding search documents.

4. The method according to claim 1, wherein The preliminary search literature in the preliminary search literature set is screened and extracted based on the large language model and the medical bias assessment tool to obtain a draft data literature set and a clinical heterogeneity factor set, including: Get pre-built data extraction templates; Using the medical bias assessment tool, the bias assessment is performed on the preliminary searched literature to obtain a bias assessment result; In response to the bias assessment result reaching a preset value, extracting the document content in the preliminary searched documents to obtain an extracted content set; For each extracted content in the extracted content set, perform the following pre-processing operations: extracting clinical heterogeneity factors from the extracted content using the large language model to obtain at least one clinical heterogeneity factor; Fill the extracted content into the data extraction template to obtain a draft data document.

5. The method according to claim 1, wherein The method further comprises executing a heterogeneity judgment script based on the fusion effect size set and the clinical heterogeneity factor set to perform heterogeneity judgment, thereby obtaining a visualization chart set, including: Determining the effect size of the quantitative research results corresponding to the fused effect size set to obtain an effect size data set; determining heterogeneity indices of clinical heterogeneity factors corresponding to the clinical heterogeneity factor set to obtain a heterogeneity index data set; Executing the heterogeneity evaluation script pre-generated according to the corresponding requirements based on the effect size data set and the heterogeneity index data set to obtain statistical values ​​and heterogeneity indices; Determining a heterogeneous effect model based on the heterogeneity index and the effect size data set; Generate a heterogeneity forest plot and a heterogeneity bar plot according to the heterogeneous effect model and the effect size data set; Generating a significance test chart according to the statistical value and the effect size data set; The heterogeneity forest plot, the heterogeneity bar chart, and the significance test chart are determined as the visualization chart set.

6. A device for generating a medical draft based on a large model, comprising: A search formula generation unit is configured to generate a multi-database search formula corresponding to a medical research direction using a large language model; A document retrieval unit is configured to execute the multi-database search formula in the database to perform a cross-database search, thereby obtaining a preliminary search document set and a cross-database causal network; a literature screening unit configured to screen and extract the preliminary search documents in the preliminary search document set based on the large language model and the medical bias assessment tool to obtain a draft data document set and a clinical heterogeneity factor set; The conflict marking unit is configured to perform dynamic evidence chain fusion to determine weights based on the draft data document set and the clinical heterogeneity factor set, and obtain a fused effect size set and a conflict marking set, wherein the dynamic evidence chain fusion to determine weights based on the draft data document set and the clinical heterogeneity factor set to obtain a fused effect size set and a conflict marking set includes: extracting structured data corresponding to the research features in the draft data document set to obtain a research feature content set; performing standardization on each clinical heterogeneity factor in the clinical heterogeneity factor set to obtain a heterogeneity factor data set; using a preset heterogeneity weight function to negatively correlate the research feature content set and the heterogeneity factor data set to obtain a weight set corresponding to the research feature content set; determining a research effect size based on the medical research direction; performing dynamic weighted averaging on the research effect size based on the weight set to obtain the fused effect size set; for each fused effect size in the fused effect size set, in response to the fused effect size reaching a preset threshold, determining that the research feature content corresponding to the fused effect size is a conflict mark; a chart generating unit configured to execute a heterogeneity judgment script to perform heterogeneity judgment based on the fusion effect size set and the clinical heterogeneity factor set, thereby obtaining a visualization chart set; a report generation unit configured to perform standardized conflict resolution based on the cross-database causal network and the conflict marker set to obtain a conflict resolution report, wherein the standardized conflict resolution is a process of uniformly handling conflicts according to predefined rules or algorithms to obtain a consistent conclusion, and the conflict resolution report is a report including a detailed description of the conflict, a resolution method, a resolution structure, and correction suggestions; The draft generation unit is configured to determine the results of the draft data document set and the visualization chart set using the large language model based on the medical draft template and the conflict resolution report to generate a medical draft.

7. An electronic device comprising: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 5.

8. A computer-readable medium having a computer program stored thereon, wherein: When the program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images