Hepatitis B research path generation method and device based on CR-HepB platform and electronic equipment

Abstract

The invention discloses a hepatitis B research path generation method and device based on a CR-HepB platform and electronic equipment, and belongs to the technical field of data processing.The hepatitis B research path generation method based on the CR-HepB platform comprises the steps that multi-dimensional field data related to hepatitis B is obtained from the CR-HepB platform; performing data integration processing on the multi-dimensional field data to obtain multi-source input data; and multi-source input data is input into the trained Qwen2 model, a hepatitis B research path is obtained, and the trained Qwen2 model is a model in which a dynamic weight adjustment mechanism is introduced. According to the method, a customized training large language model is combined with multi-dimensional data resources of a CR-HepB platform, and a hepatitis B research path and a research method are intelligently generated. The innovative research method of multi-dimensional data integration and logic consistency is generated, personalized scientific research path customization and dynamic optimization are supported, and the scientific research efficiency and innovation ability are remarkably improved.

Description

Technical Field

[0001] The present application belongs to the field of data processing technology, and specifically relates to a method, device and electronic equipment for generating a hepatitis B research pathway based on a CR-HepB platform. Background Art

[0002] Hepatitis B is a global public health problem. Chronic diseases caused by hepatitis B virus (HBV) infection may lead to severe liver damage, including cirrhosis and hepatocellular carcinoma. Currently, about 257 million people are infected with HBV worldwide, resulting in more than 800,000 deaths each year. Despite significant progress in vaccination and antiviral treatment, many research gaps remain to be filled, including viral mutation mechanisms, drug resistance issues, new treatment options, and optimization of personalized medicine.

[0003] In the field of hepatitis B, the core challenges of research mainly include: existing research has accumulated a large amount of literature, clinical trial data and experimental data, but these data are scattered and lack systematic organization; researchers often need to spend a lot of time to find research gaps from massive literature and data, and this method of relying on manual analysis is inefficient and easy to miss potential important research directions; research hotspots in the field of hepatitis B are constantly changing with the progress of clinical practice and science and technology, but traditional research path planning lacks dynamic update capabilities. Therefore, a research path generation method is urgently needed to dynamically adapt to changes in the scientific research field. Summary of the invention

[0004] The purpose of this application is to provide a method, device and electronic equipment for generating a hepatitis B research path based on the CR-HepB platform to dynamically adapt to changes in the scientific research field to generate a research path.

[0005] According to a first aspect of an embodiment of the present application, a method for generating a hepatitis B research pathway based on a CR-HepB platform is provided, and the method may include:

[0006] Obtain multi-dimensional data related to hepatitis B from the CR-HepB platform;

[0007] Perform data integration processing on multi-dimensional field data to obtain multi-source input data;

[0008] The multi-source input data is input into the trained Qwen2 model to obtain the hepatitis B research path. The trained Qwen2 model is a model that introduces a dynamic weight adjustment mechanism.

[0009] In some optional embodiments of the present application, multi-dimensional field data related to hepatitis B is obtained from the CR-HepB platform, including:

[0010] Obtain hepatitis B-related research literature included in the CR-HepB platform;

[0011] Access to electronic medical records and follow-up records in the CR-HepB platform;

[0012] Obtain the gene sequencing database of the CR-HepB platform;

[0013] Access disease progression and monitoring data collated in the CR-HepB platform.

[0014] In some optional embodiments of the present application, data integration processing is performed on multi-dimensional domain data to obtain multi-source input data, including:

[0015] Integrate multi-dimensional field data based on knowledge graph to obtain integrated data;

[0016] The integrated data is preprocessed to obtain multi-source input data.

[0017] In some optional embodiments of the present application, multi-dimensional domain data is integrated based on the knowledge graph to obtain integrated data, including:

[0018] Extract key content from multi-dimensional domain data as standard entities;

[0019] Establish relationship edges between standard entities to represent semantic associations between data items;

[0020] Align data of different modalities into a unified semantic space to obtain fused data;

[0021] Organize the extracted entities and relationships into a dynamic knowledge graph, which is updated as the data changes;

[0022] Perform data verification and optimization on the knowledge graph to obtain integrated data.

[0023] In some optional embodiments of the present application, preprocessing is performed on the integrated data to obtain multi-source input data, including:

[0024] Generate dynamic summaries and perform multi-level semantic annotation on literature data;

[0025] Perform multi-dimensional time series alignment and dynamic prediction of clinical data;

[0026] Real-time functional annotation expansion and multi-dimensional feature extraction of genomic data;

[0027] Clean the integrated data to obtain multi-source input data.

[0028] In some optional embodiments of the present application, the trained Qwen2 model is obtained by training using the following method:

[0029] Extract keywords from literature data, clinical data, and genomic data;

[0030] Perform word frequency statistics on keywords to obtain the frequency of occurrence of keywords;

[0031] Calculate the weight of each keyword based on its frequency of occurrence;

[0032] According to the semantic importance of keywords, the semantic weights of keywords are dynamically adjusted during the training of the Qwen2 model.

[0033] In some optional embodiments of the present application, the weight calculation formula is:

[0034]

[0035] Among them, w i is the weight; is the common word weight; Freq(k) is the keyword frequency.

[0036] According to a second aspect of an embodiment of the present application, a device for generating a hepatitis B research path based on a CR-HEPB platform is provided, and the device may include:

[0037] The acquisition module is used to obtain multi-dimensional field data related to hepatitis B from the CR-HepB platform;

[0038] The integration processing module is used to perform data integration processing on multi-dimensional field data to obtain multi-source input data;

[0039] The research path generation module is used to input multi-source input data into the trained Qwen2 model to obtain the hepatitis B research path. The trained Qwen2 model is a model that introduces a dynamic weight adjustment mechanism.

[0040] According to a third aspect of an embodiment of the present application, an electronic device is provided, which may include:

[0041] processor;

[0042] a memory for storing processor-executable instructions;

[0043] Wherein, the processor is configured to execute instructions to implement the hepatitis B research pathway generation method based on the CR-HEPB platform as shown in any one of the embodiments of the first aspect.

[0044] According to the fourth aspect of the embodiments of the present application, a storage medium is provided. When the instructions in the storage medium are executed by a processor of an information processing device or a server, the information processing device or the server implements a hepatitis B research path generation method based on the CR-HEPB platform as shown in any one of the embodiments of the first aspect.

[0045] The above technical solution of the present application has the following beneficial technical effects:

[0046] The method of the embodiment of this application uses a large language model with customized training combined with the multi-dimensional data resources of the CR-HepB platform to intelligently generate hepatitis B research paths and research methods. Through customized field training, it is possible to accurately parse scientific research literature, explore research gaps, and generate a dynamically adapted research roadmap, covering specific content such as experimental design and data analysis paths. In addition, this method realizes the generation of innovative research methods that are consistent with multi-dimensional data integration and logic, while supporting personalized scientific research path customization and dynamic optimization, significantly improving scientific research efficiency and innovation capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a flow chart of a method for generating a hepatitis B research pathway of the CR-HepB platform in an exemplary embodiment of the present application;

[0048] Figure 2 It is a schematic diagram of the structure of a hepatitis B research pathway generating device of the CR-HepB platform in an exemplary embodiment of the present application;

[0049] Figure 3 is a schematic structural diagram of an electronic device in an exemplary embodiment of the present application;

[0050] Figure 4 It is a schematic diagram of the hardware structure of an electronic device in an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in conjunction with specific implementations and with reference to the accompanying drawings. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present application. In addition, in the following description, the description of known structures and technologies is omitted to avoid unnecessary confusion of the concepts of the present application.

[0052] The accompanying drawings show schematic diagrams of layer structures according to embodiments of the present application. These figures are not drawn to scale, and some details are magnified and some details may be omitted for the purpose of clarity. The shapes of various regions and layers shown in the figures and the relative sizes and positional relationships therebetween are only exemplary, and may deviate in practice due to manufacturing tolerances or technical limitations, and those skilled in the art may additionally design regions / layers with different shapes, sizes, and relative positions according to actual needs.

[0053] Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0054] In the description of the present application, it should be noted that the terms "first", "second" and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0055] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0056] In conjunction with the accompanying drawings, the hepatitis B research path generation method, device and electronic device based on the CR-HepB platform provided in the embodiments of the present application are described in detail through specific embodiments and their application scenarios.

[0057] like Figure 1 As shown, in a first aspect of the embodiment of the present application, a method for generating a hepatitis B research pathway based on the CR-HepB platform is provided, and the method may include:

[0058] S110: Obtain multi-dimensional data related to hepatitis B from the CR-HepB platform;

[0059] S120: performing data integration processing on multi-dimensional domain data to obtain multi-source input data;

[0060] S130: Input the multi-source input data into the trained Qwen2 model to obtain the hepatitis B research path. The trained Qwen2 model is a model that introduces a dynamic weight adjustment mechanism.

[0061] The method of this embodiment uses a large language model with customized training combined with the multi-dimensional data resources of the CR-HepB platform to intelligently generate hepatitis B research paths and research methods. Through customized field training, it is possible to accurately parse scientific research literature, explore research gaps, and generate a dynamically adapted research roadmap, covering specific content such as experimental design and data analysis paths. In addition, this method realizes the generation of innovative research methods that integrate multi-dimensional data and are consistent with logic, while supporting personalized scientific research path customization and dynamic optimization, significantly improving scientific research efficiency and innovation capabilities.

[0062] The CR-HepB platform covers multi-dimensional field data, which can ensure the comprehensiveness and scientificity when generating research paths.

[0063] In some embodiments, multi-dimensional data related to hepatitis B are obtained from the CR-HepB platform, including:

[0064] Obtain hepatitis B-related research literature included in the CR-HepB platform;

[0065] Access to electronic medical records and follow-up records in the CR-HepB platform;

[0066] Obtain the gene sequencing database of the CR-HepB platform;

[0067] Access disease progression and monitoring data collated in the CR-HepB platform.

[0068] Among them, hepatitis B related research literature can include core data of related papers such as hepatitis B treatment progress, drug resistance mechanism, personalized medicine, clinical trial design, etc.

[0069] Example:

[0070] Title of the article: "Progress in research on drug resistance mechanisms in HBV antiviral therapy"

[0071] Literature content: Describes the mechanism of entecavir resistance, involving the A1762T and G1764A mutations of the virus, and the treatment strategy based on this.

[0072] Electronic medical records and follow-up records can include basic patient information (age, gender), diagnosis results, treatment process records, follow-up data (viral load, drug effects, etc.).

[0073] Example:

[0074] {

[0075] "Patient ID": "12345",

[0076] "Gender": "Male",

[0077] "Age": 45,

[0078] "Viral Load": "1e5",

[0079] "Medication": "Entecavir",

[0080] "Follow-up records": [

[0081] {"Date":"2024-01-01","Viral Load":"5e4"},

[0082] {"Date":"2024-02-01","Viral Load":"2e4"} ]

[0084] }

[0085] The gene sequencing database may include HBV viral genome sequences, drug resistance-related mutation sites (such as A1762T, G1764A), and genotype information related to efficacy.

[0086] Example:

[0087] {

[0088] "Variation site": ["A1762T","G1764A"],

[0089] "Notes": ["may affect viral replication", "associated with drug resistance"]

[0090] }

[0091] Disease progression and surveillance data can include cirrhosis degree, hepatocellular carcinoma (HCC) risk assessment, regional infection rates, etc.

[0092] Example:

[0093] {

[0094] "Liver cirrhosis score": "F2 (moderate fibrosis)",

[0095] "HCC risk score": "moderate risk",

[0096] "Regional infection rate": {

[0097] "Asia": "5.7%",

[0098] "Africa": "8.0%"

[0099] }

[0100] }

[0101] In some embodiments, data integration processing is performed on multi-dimensional domain data to obtain multi-source input data, including:

[0102] Integrate multi-dimensional field data based on knowledge graph to obtain integrated data;

[0103] The integrated data is preprocessed to obtain multi-source input data.

[0104] In this embodiment, literature data, clinical data, and genomic data can be deeply integrated, and a large model can be used to achieve intelligent optimization of the entire process.

[0105] Exemplary:

[0106] Literature data preprocessing, dynamic summary generation and multi-level semantic annotation.

[0107] Objective: To break through the limitations of traditional keyword extraction and achieve semantic understanding and structured processing of literature data based on a large model. Use a generative model with multiple rounds of prompts to extract research questions, hypotheses, and conclusions from the literature in a hierarchical manner:

[0108] First-level semantic extraction: extract paragraph-level semantic content (such as description of drug resistance mechanism).

[0109] Secondary semantic extraction: extract detailed information, such as experimental methods and result data points, and mark them as specific fields.

[0110] Construct hierarchical extraction logic, combine the model's understanding of the document's semantic hierarchy, and automatically generate structured summaries. Perform semantic correction on the document content and detect logical consistency in the extracted information:

[0111] Check whether there are any contradictions between the research conclusions and the hypotheses;

[0112] Check the consistency between experimental data and conclusions.

[0113] Context comparison technology is used to verify the consistency of document content and reduce redundant and conflicting information. For multiple documents with similar topics, semantic similarity is calculated through semantic embedding to generate dynamic topic summaries. The model dynamically generates a comprehensive summary of multiple documents on "drug resistance mutations" to directly extract the core conclusions. Semantic aggregation and dynamic multi-topic summaries are achieved to provide a more efficient path for the integration of multiple documents.

[0114] Use large models to extract research questions, hypotheses and conclusions from literature and generate dynamic summaries, avoiding the limitations of traditional static keyword extraction.

[0115] Example:

[0116] Original text: Detailed description of drug resistance mechanisms from multiple experimental studies.

[0117] Dynamic summary: "The A1762T site of HBV may significantly reduce the efficacy of entecavir."

[0118] Clinical data preprocessing, multi-dimensional time series alignment and dynamic prediction. Achieve time series standardization and completion of clinical data, solve problems such as inconsistent data formats and missing values. Use time-aware networks to process patient follow-up records and normalize different time points (such as January, March, and April) into a unified timeline:

[0119] Input: Patient medication time point and viral load data;

[0120] Processing: Normalize by "treatment cycle" to align different patients to a unified time dimension.

[0121] The time-aware model is used to intelligently align patient follow-up time points to form a standardized time series across patients. Missing data points in follow-up records (such as viral load and drug use records) are dynamically supplemented through trend predictions of the generative model:

[0122] Predict reasonable values ​​for missing points based on historical data and patient characteristics;

[0123] The model automatically generates a completion report and annotates the confidence interval of the completion value.

[0124] Combine the generative model with time series completion to improve the scientificity and accuracy of missing value completion. Combine the patient's viral load, gene mutation records, medication records and other data to generate a multi-dimensional dynamic feature matrix:

[0125] The model generates dynamic features based on the interactive relationship between multi-source data, such as the causal chain of "medication → viral load decrease → gene mutation function".

[0126] Through feature matrix generation, a global view of multidimensional time series is formed, providing high-dimensional input for subsequent analysis.

[0127] Multi-dimensional time series alignment: For follow-up data, deep learning algorithms are used to standardize the treatment time points of different patients for unified analysis.

[0128] Example:

[0129] Original data: Patient 1 was followed up in January and March; Patient 2 was followed up in February and April.

[0130] Alignment result: The monthly sequence is unified into the 1st and 2nd months.

[0131] Genomic data preprocessing, real-time functional annotation expansion and multi-dimensional feature extraction. Use large models to dynamically expand genomic annotations and combine literature and clinical data to enhance functional interpretation capabilities. Models parse genomic data, extract mutation sites and their preliminary functional annotations:

[0132] Automatically identify key mutations (such as A1762T, G1764A) and their possible functions;

[0133] Generate an initial annotation such as "associated with increased viral load."

[0134] Combine the genome sequence and the functional annotation database to generate semantic labels for mutations as input for subsequent analysis. The model parses new literature and automatically updates mutation functional annotations. For example:

[0135] When new literature finds that a certain mutation “affects immune escape,” the model automatically adds it to the existing annotations.

[0136] Functionally expand the impact path of the mutation site and generate high-dimensional annotations:

[0137] For example, a multi-layer relationship chain of “G1764A→virus replication→drug resistance” is dynamically generated.

[0138] Real-time function expansion combined with contextual reasoning provides dynamic, high-dimensional annotations for genomic data. The model generates a multi-dimensional feature vector of mutations based on the annotations:

[0139] Generate a high-dimensional vector representation of the biological function, drug action pathway, and clinical relevance of the mutation site.

[0140] Combined with annotation extension, multidimensional mutation signatures are generated for further model input or analysis.

[0141] Automatically update variant functional annotations of genomic data based on the latest research trends.

[0142] Example:

[0143] Original data: G1764A has no relevant annotations.

[0144] Updated results: Combined with new literature, marked as "resistance mutation sites that may affect the efficacy of entecavir."

[0145] Data cleaning, intelligent cleaning and rule adaptive optimization. Solve data anomalies and redundancy problems through model-driven cleaning rule generation and feedback optimization. The model automatically generates cleaning rules through global data analysis, for example:

[0146] The upper and lower limits of viral load are detected (if it is lower than 100 or higher than 1e9, it is marked as abnormal).

[0147] Timestamp anomaly detection (such as follow-up time duplication or out of reasonable range).

[0148] Automatically generate cleaning rules to dynamically adapt to the characteristics of different data sets. Cleaning results are evaluated in real time through the model feedback mechanism:

[0149] Detect mislabeling (e.g. marking normal values ​​as abnormal);

[0150] Adjust rule thresholds and ranges based on feedback, and gradually optimize the cleaning strategy.

[0151] Adaptive rule optimization based on feedback reduces manual intervention and improves cleaning efficiency. For repeated or redundant data, the model automatically merges them through semantic similarity calculations; removes low-relevance fields, retains high-value features, and improves the overall quality of data. Combining semantic analysis with rule optimization, efficient cleaning and feature screening are achieved.

[0152] Through multi-level semantic parsing and consistency correction, literature data preprocessing has achieved dynamic and efficient aggregation. Time-aware modeling combined with dynamic completion technology integrates clinical data into standardized multi-dimensional time series. Genomic data annotation forms high-dimensional feature representation and multi-level paths through dynamic updates and functional expansion. Intelligent cleaning rules combined with feedback mechanisms achieve dynamic optimization of rules and efficient processing of redundant data.

[0153] The data preprocessing method of this embodiment not only realizes the integration of multi-source data, but also fully utilizes the large model capabilities to improve processing efficiency and innovation.

[0154] In some embodiments, the multi-dimensional domain data is integrated based on the knowledge graph to obtain integrated data, including:

[0155] Extract key content from multi-dimensional domain data as standard entities;

[0156] Establish relationship edges between standard entities to represent semantic associations between data items;

[0157] Align data of different modalities into a unified semantic space to obtain fused data;

[0158] Organize the extracted entities and relationships into a dynamic knowledge graph, which is updated as the data changes;

[0159] Perform data verification and optimization on the knowledge graph to obtain integrated data.

[0160] Specifically, the steps of data integration based on knowledge graph may include:

[0161] 1. Data entity extraction: extract key content from different types of data (literature, clinical records, genomic data, etc.) into standardized "entities" to form the node basis of the knowledge graph.

[0162] For text data such as literature and case reports, we use the semantic parsing capabilities of large-scale pre-trained models to decompose paragraph content into multiple fields (such as research questions, hypotheses, and conclusions). We design guidance prompts to allow the model to extract core entities based on the context. For example:

[0163] Gene mutations in the literature (such as “A1762T”) were extracted as “gene mutation entities”.

[0164] Patient characteristics in cases (such as “viral load changes”) are extracted as “patient dynamic attributes”.

[0165] Fields in the database (such as genome sequences and treatment records) are matched to generate "standard entities" that conform to the knowledge graph structure. Redundant fields are merged to ensure uniqueness. For example, "variation sites" and "functional annotations" in genomic data are integrated into one entity node. Semantic abstraction is performed on diverse entities to classify similar but different data (such as "drug resistance mutations" and "drug resistance mechanisms") into a unified type. Semantic similarity is evaluated through the model to generate a consistent labeling system.

[0166] 2. Data relationship construction: establish relationship edges between entities to represent the semantic associations between data items (such as “patient→drug→efficacy”, “gene→mutation→function”).

[0167] Identify potential relationships through semantic analysis of documents and structured data. For example:

[0168] Extract the co-occurrence relationship between drug names and gene mutations.

[0169] Extract the causal relationship between "time series changes" and "treatment options" from patient medical records.

[0170] Generate descriptive language for each relationship to annotate the semantics of the relationship. The prompts generated by the model based on the context, such as "the patient's viral load decreased after taking the drug", will be converted into "the positive correlation between the drug and the viral load". Dynamically assign weights to relationships based on data credibility or semantic strength. For example:

[0171] The relationship of “gene mutation → drug resistance” mentioned many times in the literature will be given a higher weight.

[0172] Inconsistent records in clinical data are marked as low-weight relationships.

[0173] Dynamically update the weights and edge types of relationships based on new data or correction feedback. For example, a new relationship of “drug → toxicity” may be added as new literature is added.

[0174] 3. Multimodal data alignment: align data of different modalities (text, tables, images, etc.) into a unified semantic space to achieve cross-domain fusion.

[0175] Parse text data into semantic vectors and generate standardized embedding representations. For genomic sequence or time series data, use similarity mapping techniques to generate semantic vectors that are aligned with the text modality.

[0176] Use cross-modal learning methods to semantically align different data types, for example, matching feature points in patient imaging data with disease descriptions in medical records. The model automatically adjusts the distance between modalities during the alignment process to ensure semantic consistency in the embedding space. Verify the consistency of the alignment results. For example, ensure that the annotations between the imaging data, genomic data, and medical records of the same patient match. Detect contradictory relationships (such as incorrect time points or repeated fields) during the alignment process and make correction suggestions.

[0177] 4. Knowledge graph generation and update: Organize the extracted entities and relationships into a dynamic knowledge graph and automatically update it as the data changes.

[0178] The initial graph is constructed using entity nodes and relationship edges, representing the contents of all data sources as a directed graph. Each entity node carries multidimensional attributes (such as functional annotations of gene mutations, treatment time series of patients, etc.). Using incremental learning technology, the model automatically expands the graph when new data is identified. For example, when a new document mentions "new drug → gene mutation → efficacy", the model automatically generates new nodes and relationship edges. Graph expansion supports both entity merging and relationship revision to avoid redundancy. For new data, the model automatically adjusts the weights of entities and relationships based on their context and source. The credibility of new relationships is verified through multiple rounds of feedback mechanisms, and the graph structure is gradually optimized.

[0179] 5. Data verification and optimization to ensure the integrity, consistency and timeliness of data in the knowledge graph.

[0180] Design a set of validation rules (such as time sequence, causal logic) to automatically verify whether there are contradictions in the graph. For example:

[0181] If a medical record shows a drop in viral load, but the profile shows the medication wasn’t effective, the model would flag a potential error.

[0182] The model scores the integrity of the graph, marks low-quality nodes and edges, and prompts users to revise. Redundant or ambiguous entities are merged to improve the clarity of the graph. The model automatically learns based on the user's operation records on the graph (such as modifying or supplementing data) and updates the subsequent data processing flow. The edge weights and types of the knowledge graph are updated in real time according to the new data, and have the ability to learn dynamically and adjust adaptively. Unlike traditional static graphs, this method supports multi-dimensional optimization of relationships. Not only does it perform semantic mapping on text and structured data, but it also performs semantic alignment on images and time series data to achieve multi-domain integration. The knowledge structure is automatically generated through large model prompt words without the need for manual definition of rules, which greatly improves the efficiency of graph generation. The model dynamically generates semantic labels and entity relationships based on the context, solving the problem of lack of flexibility in rules in traditional methods. The self-learning and updating of the graph is driven by user interaction records, supporting the dynamic synchronous development of the graph and research.

[0183] In some embodiments, preprocessing the integrated data to obtain multi-source input data includes:

[0184] Generate dynamic summaries and perform multi-level semantic annotation on literature data;

[0185] Perform multi-dimensional time series alignment and dynamic prediction of clinical data;

[0186] Real-time functional annotation expansion and multi-dimensional feature extraction of genomic data;

[0187] Clean the integrated data to obtain multi-source input data.

[0188] In some embodiments, the trained Qwen2 model is trained by the following method:

[0189] Extract keywords from literature data, clinical data, and genomic data;

[0190] Perform word frequency statistics on keywords to obtain the frequency of occurrence of keywords;

[0191] Calculate the weight of each keyword based on its frequency of occurrence;

[0192] According to the semantic importance of keywords, the semantic weights of keywords are dynamically adjusted during the training of the Qwen2 model.

[0193] In this embodiment, the large model customized training (Qwen2) is used to optimize Qwen2 and perform dynamic domain fine-tuning. Qwen2 performs well in general semantic understanding, but its semantic understanding ability in the hepatitis B field is insufficient. By introducing a dynamic weight adjustment mechanism, the semantic weight of keywords in the hepatitis B field is improved during training.

[0194] In some embodiments, the weight is calculated as follows:

[0195]

[0196] Among them, w i is the weight; is the common word weight; Freq(k) is the keyword frequency.

[0197] The core of the dynamic weight adjustment mechanism is to dynamically enhance the weight of keywords according to the semantic importance of domain keywords, so that the model pays more attention to the key content of the domain during training and reasoning. By adjusting the attention allocation and embedding vectors, domain-related information occupies a higher semantic priority in the model.

[0198] Specifically, extract high-frequency words related to the field of hepatitis B (such as "HBV", "drug resistance", "gene mutation"). Extract common drugs and symptoms (such as "entecavir", "viral load"). Extract mutation sites (such as "A1762T", "G1764A").

[0199] Perform word frequency statistics on the extracted keywords and record the frequency of each word in the domain data. Use the semantic embedding of the model to calculate the semantic importance of each keyword and determine its contribution to the domain task. The model parses the input text and dynamically calculates the weight based on the frequency of keyword occurrence and contextual relationship: high-frequency keywords (such as "HBV" and "drug resistance mechanism") obtain higher semantic weights. The weight of low-frequency keywords is kept at a low level to avoid semantic interference. Dynamic weight matrix generation: Generate a dynamic semantic weight matrix based on the weight formula, covering all keywords in the training corpus.

[0200] Weight Matrix Example

[0201] W=[HBV A1762T resistance mechanism entecavir 1.8 2.53.02.0]

[0203] Based on the data from the CR-HepB platform, semantic labels (such as "gene mutation" and "therapeutic drugs") are annotated for keywords. The annotated data is used as domain corpus and input into Qwen2 for fine-tuning training.

[0204] During the training process, the model dynamically adjusts the semantic weights of keywords:

[0205] Increase the embedding vector weight of domain keywords so that they are given priority in semantic computing. Optimize the attention allocation mechanism of Transformer and increase the attention weight of domain keywords. Optimize attention allocation and modify the weight allocation formula of the multi-head attention layer:

[0206] A new =A original +W dynamic

[0207] A new : The adjusted attention matrix.

[0208] A original : The original attention matrix.

[0209] W dynamic : Dynamic keyword weight matrix.

[0210] In the early stages of fine-tuning training, keyword weights are gradually enhanced to avoid over-reliance on keywords and interference with general semantics. In the later stages of training, the weight matrix is ​​fixed to ensure model stability.

[0211] The CR-HepB domain-specific test set is used to evaluate Qwen2’s ability to understand domain texts: domain keyword extraction accuracy; semantic relevance coverage.

[0212] If the semantic weight of the model for certain keywords is insufficient, adjust the α coefficient in the weight formula for optimization.

[0213] Case analysis verification:

[0214] Input domain text and observe whether the model output can accurately understand and generate relevant domain conclusions:

[0215] Example input: "Effects of A1762T and G1764A mutations on the efficacy of entecavir."

[0216] Sample output: "A1762T and G1764A mutations may lead to increased drug resistance and require further experimental verification."

[0217] Different from the traditional model training of fixed word vectors, the dynamic weight adjustment mechanism is introduced to enable the model to flexibly adapt to the semantic importance of domain keywords. By dynamically adjusting the embedding weights of keywords, the Transformer attention allocation mechanism is optimized to improve the coverage of domain semantics. During the fine-tuning process, the keyword weight is dynamically enhanced to balance domain adaptation capabilities and general semantic understanding capabilities to avoid overfitting. During the model training and testing phases, dynamic feedback is provided through semantic coverage and keyword weight allocation to continuously optimize domain performance.

[0218] The specific model application process may include the following steps:

[0219] 1. Use the big model to parse the input content and identify the research objectives and related keywords.

[0220] Example prompt words:

[0221] "Extract the research objectives and core keywords entered below."

[0222] Input: "Objective: Drug resistance research"

[0223] Output: "Keywords: drug resistance, mutation, entecavir"

[0224] Convert the literature content into a structured summary and extract the core research hypotheses and conclusions.

[0225] Prompt: "Extract hypotheses and conclusions from the following literature and standardize the output."

[0226] Clinical and genomic data are transformed into a unified semantic space through an embedding model.

[0227] 2. Data feature extraction and association, multi-source data feature extraction:

[0228] Literature data: Extract the relationship between mutations and drugs (such as "drug-resistance mutations and entecavir").

[0229] Clinical data: Extract time series features (such as “viral load change trend”).

[0230] Genomic data: extracting functional annotations and gene effects.

[0231] The model establishes interactive relationships between data through semantic mapping.

[0232] Output example:

[0233] "The A1762T mutation may lead to elevated viral loads."

[0234] 3. Generate a solution path for the research problem, including experimental steps and data analysis methods.

[0235] Generate experimental design plans based on research objectives and input data. Provide targeted data analysis suggestions and output optimal methods and variables.

[0236] Output Format

[0237] Output in structured JSON:

[0238] {

[0239] "Research Path": [

[0240] "Step 1: Collect genomic data and locate drug-resistant mutation sites.",

[0241] "Step 2: Analysis of viral load changes to verify drug resistance mechanisms.",

[0242] "Step 3: Personalize treatment based on results."

[0243] ],

[0244] "Experimental Design": {

[0245] "Hypothesis": "The drug resistance site A1762T significantly reduces drug efficacy.",

[0246] "Variable": ["Viral load", "Drug resistance mutation"],

[0247] "Method": "Linear regression analysis"

[0248] },

[0249] "Analysis Suggestions": {

[0250] "Method": "Regression Analysis",

[0251] "Reason": "The linear relationship between viral load and mutation needs further verification."

[0252] }

[0253] }

[0254] The output generation model combination may include the following steps:

[0255] 1. Research path generation

[0256] Model training phase:

[0257] During the fine-tuning process, add the research path generation task:

[0258] Input: "Literature content, clinical data, genomic data"

[0259] Output: "Step 1: locate mutations; Step 2: analyze changes in viral load; Step 3: optimize treatment options"

[0260] Reasoning phase:

[0261] The model dynamically generates paths based on user input:

[0262] Tips:

[0263] "Generates a research path based on the following inputs, with specific actions for each step."

[0264] Input: "Objective: drug resistance analysis; Genomic data: A1762T; Clinical data: elevated viral load"

[0265] Output: "Step 1: Genomic data analysis to locate mutations; Step 2: Changes in viral load to verify drug resistance mechanisms"

[0266] 2. Experimental Design Generation

[0267] Model training phase:

[0268] Use the domain experiment design template to strengthen the model's understanding of experimental logic:

[0269] Input: "Hypothesis: A1762T affects efficacy"

[0270] Output: "Variables: viral load, drug-resistant mutation; Method: regression analysis"

[0271] Reasoning phase:

[0272] Model generation experimental design:

[0273] Tips:

[0274] "Generate an experimental design, output variables and methods based on the following assumptions."

[0275] Input: "Hypothesis: A1762T mutation reduces efficacy"

[0276] Output: "Variables: viral load, drug-resistant mutation; Method: regression analysis"

[0277] 3. Analysis and suggestion generation

[0278] Model training phase:

[0279] During training, add the analysis suggestion generation task:

[0280] Input: "Variables: viral load, drug resistance mutation"

[0281] Output: "Suggestion: Use regression analysis to verify the relationship between variables"

[0282] Reasoning stage:

[0283] Generate suggestions based on input variables:

[0284] Tips:

[0285] "Generate analysis recommendations for the following variables and provide justifications."

[0286] Input: "Variables: viral load, drug resistance mutation"

[0287] Output: "Recommendation: Use linear regression to analyze the relationship between viral load and mutations"

[0288] Dynamically generate research paths by combining input data with training models, avoiding the limitations of traditional fixed processes and ensuring real-time adaptability of results. Use domain templates and data to strengthen training models so that they can accurately generate experimental designs and improve scientificity and operability. Optimize analysis method suggestions through domain data and generate dynamic analysis strategies based on user input. Achieve cross-modal semantic mapping of literature, clinical data, and genomic data, providing technical support for deep integration of input and output.

[0289] The above embodiments not only enhance the logic of input and output, but also improve the innovation and applicability of the technology through the dynamic generation capability of model training.

[0290] It should be noted that the execution subject of the method for generating a hepatitis B research path based on the CR-HEPB platform provided in the embodiment of the present application may be a hepatitis B research path generation device based on the CR-HEPB platform, or a control module in the hepatitis B research path generation device based on the CR-HEPB platform for executing the method for generating a hepatitis B research path based on the CR-HEPB platform. In the embodiment of the present application, the hepatitis B research path generation device based on the CR-HEPB platform executes the method for generating a hepatitis B research path based on the CR-HEPB platform as an example to illustrate the device for generating a hepatitis B research path based on the CR-HEPB platform provided in the embodiment of the present application.

[0291] like Figure 2 As shown, in a second aspect of the embodiment of the present application, a hepatitis B research path generation device based on the CR-HEPB platform is provided, and the device may include:

[0292] An acquisition module 210 is used to acquire multi-dimensional field data related to hepatitis B from the CR-HepB platform;

[0293] The integration processing module 220 is used to perform data integration processing on multi-dimensional domain data to obtain multi-source input data;

[0294] The research path generation module 230 is used to input multi-source input data into the trained Qwen2 model to obtain the hepatitis B research path. The trained Qwen2 model is a model that introduces a dynamic weight adjustment mechanism.

[0295] The hepatitis B research path generation device based on the CR-HEPB platform in the embodiment of the present application can be a device, or a component, an integrated circuit, or a chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. Exemplarily, the mobile electronic device can be a mobile phone, a tablet computer, a laptop computer, a PDA, an in-vehicle electronic device, a wearable device, an ultra-mobile personal computer (ultra-mobile personal computer, UMPC), a netbook or a personal digital assistant (personal digital assistant, PDA), etc., and the non-mobile electronic device can be a server, a network attached storage (Network Attached Storage, NAS), a personal computer (personal computer, PC), a television (television, TV), a teller machine or a self-service machine, etc., which is not specifically limited in the embodiment of the present application.

[0296] The hepatitis B research path generation device based on the CR-HEPB platform in the embodiment of the present application can be a device with an operating system. The operating system can be an Android operating system, an iOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.

[0297] The hepatitis B research path generation device based on the CR-HEPB platform provided in the embodiment of the present application can achieve Figure 1 To avoid repetition, the various processes implemented by the method embodiment are not described here.

[0298] Alternatively, if Figure 3 As shown, an embodiment of the present application also provides an electronic device 300, including a first processor 301, a first memory 302, and a program or instruction stored in the first memory 302 and executable on the first processor 301. When the program or instruction is executed by the first processor 301, each process of the above-mentioned embodiment of the hepatitis B research path generation method based on the CR-HEPB platform is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0299] It should be noted that the electronic devices in the embodiments of the present application include the mobile electronic devices and non-mobile electronic devices mentioned above.

[0300] Figure 4 A schematic diagram of the hardware structure of an electronic device to implement an embodiment of the present application.

[0301] The hardware structure 400 of the electronic device includes but is not limited to: a radio frequency unit 401, a network module 402, an audio output unit 403, an input unit 404, a sensor 405, a display unit 406, a user input unit 407, an interface unit 408, a second memory 409, and a second processor 410 and other components.

[0302] Those skilled in the art will appreciate that the hardware structure 400 of the electronic device may also include a power source (such as a battery) for supplying power to each component, and the power source may be logically connected to the second processor 410 through a power management system, thereby implementing functions such as managing charging, discharging, and power consumption management through the power management system. Figure 4 The electronic device structure shown in the figure does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently, which will not be described in detail here.

[0303] It should be understood that in the embodiment of the present application, the input unit 404 may include a graphics processor (Graphics Processing Unit, GPU) 4041 and a microphone 4042, and the graphics processor 4041 processes the image data of the static picture or video obtained by the image capture device (such as a camera) in the video capture mode or the image capture mode. The display unit 406 may include a display panel 4061, and the display panel 4061 may be configured in the form of a liquid crystal display, an organic light emitting diode, etc. The user input unit 407 includes a touch panel 4071 and other input devices 4072. The touch panel 4071 is also called a touch screen. The touch panel 4071 may include two parts, a touch detection device and a touch controller. Other input devices 4072 may include but are not limited to a physical keyboard, a function key (such as a volume control button, a switch button, etc.), a trackball, a mouse, and a joystick, which will not be repeated here. The second memory 409 can be used to store software programs and various data, including but not limited to applications and operating systems. The second processor 410 can integrate an application processor and a modem processor, wherein the application processor mainly processes an operating system, a user interface, and applications, etc., and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the second processor 410 .

[0304] The embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, each process of the above-mentioned embodiment of the hepatitis B research path generation method based on the CR-HEPB platform is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0305] The processor is a processor in the electronic device described in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0306] An embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned embodiment of the hepatitis B research path generation method based on the CR-HEPB platform, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0307] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0308] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0309] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0310] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A method for generating a hepatitis B research pathway based on the CR-HepB platform, characterized in that: include: Obtain multi-dimensional data related to hepatitis B from the CR-HepB platform; Performing data integration processing on the multi-dimensional field data to obtain multi-source input data; The multi-source input data is input into the trained Qwen2 model to obtain the hepatitis B research path. The trained Qwen2 model is a model that introduces a dynamic weight adjustment mechanism.

2. The method for generating a hepatitis B research pathway based on the CR-HepB platform according to claim 1, characterized in that: The multi-dimensional data related to hepatitis B obtained from the CR-HepB platform include: Obtain hepatitis B-related research literature included in the CR-HepB platform; Access to electronic medical records and follow-up records in the CR-HepB platform; Obtain the gene sequencing database of the CR-HepB platform; Access disease progression and monitoring data collated in the CR-HepB platform.

3. The method for generating a hepatitis B research pathway based on the CR-HepB platform according to claim 1, characterized in that: The step of performing data integration processing on the multi-dimensional domain data to obtain multi-source input data includes: Integrate the multi-dimensional domain data based on the knowledge graph to obtain integrated data; The integrated data is preprocessed to obtain multi-source input data.

4. The method for generating a hepatitis B research pathway based on the CR-HepB platform according to claim 3, characterized in that: The step of integrating the multi-dimensional domain data based on the knowledge graph to obtain integrated data includes: Extracting key contents from the multi-dimensional domain data as standard entities; Establishing relationship edges between the standard entities to represent semantic associations between data items; Align data of different modalities into a unified semantic space to obtain fused data; Organizing the extracted entities and relationships into a dynamic knowledge graph, which is updated as the data changes; The knowledge graph is subjected to data verification and optimization to obtain integrated data.

5. The method for generating a hepatitis B research pathway based on the CR-HepB platform according to claim 4, characterized in that: The preprocessing of the integrated data to obtain multi-source input data includes: Generate dynamic summaries and perform multi-level semantic annotation on literature data; Perform multi-dimensional time series alignment and dynamic prediction of clinical data; Real-time functional annotation expansion and multi-dimensional feature extraction of genomic data; The integrated data is cleaned to obtain multi-source input data.

6. The method for generating a hepatitis B research pathway based on the CR-HepB platform according to claim 1, characterized in that: The trained Qwen2 model is obtained by training using the following method: Extract keywords from literature data, clinical data, and genomic data; Performing word frequency statistics on the keywords to obtain the occurrence frequency of the keywords; Calculate the weight of each keyword according to the occurrence frequency; According to the semantic importance of the keyword, the semantic weight of the keyword is dynamically adjusted during the training of the Qwen2 model.

7. The method for generating a hepatitis B research pathway based on the CR-HepB platform according to claim 6, characterized in that: The calculation formula of the weight is: Among them, w i is the weight; is the common word weight; Freq(k) is the keyword frequency.

8. A hepatitis B research path generation device based on the CR-HEPB platform, characterized in that: include: The acquisition module is used to obtain multi-dimensional field data related to hepatitis B from the CR-HepB platform; An integration processing module, used to perform data integration processing on the multi-dimensional domain data to obtain multi-source input data; The research path generation module is used to input the multi-source input data into the trained Qwen2 model to obtain the hepatitis B research path. The trained Qwen2 model is a model that introduces a dynamic weight adjustment mechanism.

9. An electronic device, characterized in that: include: A processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the method for generating a hepatitis B research pathway based on the CR-HEPB platform as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the steps of the method for generating a hepatitis B research pathway based on the CR-HEPB platform as described in any one of claims 1 to 7 are implemented.

Images