AI health consultation optimization method and system based on multi-expert fusion

Abstract

The invention relates to the technical field of artificial intelligence, and discloses an AI health consultation optimization method and system based on multi-expert fusion, and the system comprises a data collection module, a preprocessing module, an expert fusion module, a consultation generation module, an optimization processing module, an evaluation module, a parameter adjustment module, and a report output module. When health consultation suggestions are generated, knowledge fusion processing is carried out on standardized health data through a multi-expert fusion algorithm, and weight proportions of different expert opinions can be dynamically adjusted in combination with an expert weight distribution and consistency verification mechanism, so that the problem that the reference value of the consultation suggestions is reduced due to expert opinion divergence is avoided; according to the method, scientificity and accuracy of health consultation suggestions are guaranteed, reliability of personalized consultation is improved, timely correction is carried out when the suggestions have deviations, and real-time performance and adaptability of health consultation are guaranteed.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to an AI-based health consultation optimization method and system based on multi-expert fusion. Background Technology

[0002] Artificial intelligence is a new technological science that studies and develops theories, methods, technologies, and application systems for simulating, extending, and expanding human intelligence.

[0003] Currently, because health consultation involves multi-dimensional health data and multi-source expert knowledge, when generating personalized health consultation suggestions, the integrated multi-expert knowledge base may have differing opinions when analyzing specific user health issues. This can reduce the reference value of the consultation suggestions and make it impossible to guarantee the accuracy of health consultations.

[0004] Therefore, we propose an AI-based health consultation optimization method and system based on multi-expert fusion to solve the above problems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an AI-based health consultation optimization method and system based on multi-expert fusion, which solves the problem of the inability to dynamically adjust the weight allocation of different expert opinions as mentioned in the background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an AI-based health consultation optimization method and system based on multi-expert fusion, wherein the method includes the following steps:

[0007] S1. Collect user health data and multi-expert knowledge data to generate a raw health consultation dataset;

[0008] S2. Perform data preprocessing on the original health consultation dataset, including data cleaning, missing value imputation and feature standardization, to generate standardized health data;

[0009] S3. Based on the multi-expert fusion algorithm, the standardized health data is subjected to knowledge fusion processing, and combined with expert weight allocation and consistency verification, fused knowledge data is generated.

[0010] S4. Using the AI ​​consultation model, preliminary health consultation suggestions are generated based on the fused knowledge data, and personalized adjustments are made based on the user profile.

[0011] S5. Receive real-time feedback data from users, dynamically optimize the preliminary health consultation suggestions, and generate optimized suggestion data;

[0012] S6. Evaluate the effectiveness of the optimization suggestions data using evaluation indicators, and generate consultation effectiveness evaluation results;

[0013] S7. Based on the consultation effect evaluation results, adjust the parameters and expert fusion weights of the AI ​​consultation model to generate optimized parameter data;

[0014] S8. Based on the optimization suggestion data and optimization parameter data, generate the final health consultation report and output it to the user terminal.

[0015] Preferably, generating the original health consultation dataset in S1 includes the following steps:

[0016] S11. Collect real-time health data of users through wearable devices, mobile application interfaces and medical databases, including physiological parameters, medical history records and lifestyle data, and generate a subset of user health data;

[0017] S12. Extract expert consultation rules, clinical guidelines and case data from the multi-expert knowledge base to generate a subset of multi-expert knowledge data;

[0018] S13. Integrate the user health data subset and the multi-expert knowledge data subset, remove redundant information, and generate the original health consultation dataset.

[0019] Preferably, the generation of standardized health data in step S2 includes the following steps:

[0020] S21. Clean the original health consultation dataset by using an anomaly detection algorithm to identify and remove noisy data, and generate cleaned data.

[0021] S22. The cleaned data is processed for missing values, and the missing values ​​are filled using interpolation to generate complete data.

[0022] S23. Perform feature standardization processing on the complete data, and convert the data into a unified dimension using the Z-score standardization method to generate standardized health data.

[0023] Preferably, the generation of fused knowledge data in S3 includes the following steps:

[0024] S31. Based on expert reputation and historical performance, assign dynamic weights to each expert and generate expert weight data.

[0025] S32. Use a weighted average method to perform knowledge fusion on the standardized health data to generate initial fused data;

[0026] S33. Perform consistency verification on the initial fusion data, eliminate expert opinions through conflict detection mechanism, and generate consistent fusion knowledge data.

[0027] Preferably, generating preliminary health consultation suggestions in S4 includes the following steps:

[0028] S41. Construct an AI consultation model, which is based on a deep learning network, takes the fused knowledge data as input, and outputs preliminary health consultation suggestions;

[0029] S42. Based on user profile data, including age, gender, and health preferences, the preliminary health consultation suggestions are adjusted to generate personalized consultation suggestions;

[0030] S43. Verify the rationality of the personalized consultation suggestions through simulation testing, and generate verified suggestion data.

[0031] Preferably, the generation of optimization suggestion data in S5 includes the following steps:

[0032] S51. Collect real-time user feedback data through the user interface, including satisfaction ratings and suggestions for improvement;

[0033] S52. Using a reinforcement learning algorithm, the personalized consultation suggestions are iteratively optimized based on the real-time user feedback data to generate dynamic optimization suggestions;

[0034] S53. Perform a compliance check on the dynamic optimization suggestions to ensure they comply with medical standards, and generate optimization suggestion data.

[0035] Preferably, the generation of consultation effectiveness evaluation results in S6 includes the following steps:

[0036] S61. Define evaluation metrics, including accuracy, response time, and user satisfaction index;

[0037] S62. Evaluate the effectiveness of the optimization suggestions using the test dataset, calculate the scores of each indicator, and generate preliminary evaluation results.

[0038] S63. Based on expert review, revise the preliminary assessment results to generate a consultation effectiveness assessment result.

[0039] Preferably, the generation of optimized parameter data in S7 includes the following steps:

[0040] S71. Analyze the consultation effect evaluation results to identify parameter bias and insufficient expert weight in the AI ​​consultation model;

[0041] S72. Adjust the model parameters using the gradient descent method to generate updated model parameters;

[0042] S73. Recalculate the expert weights based on the evaluation results and generate optimized expert weight data;

[0043] S74. Integrate the updated model parameters and optimization expert weight data to generate optimization parameter data.

[0044] Preferably, the step S8 of generating the final health consultation report and outputting it to the user includes the following steps:

[0045] S81. Combine the optimization suggestion data and optimization parameter data to generate the original report data;

[0046] S82. Convert the original report data into readable text using natural language generation technology to generate a preliminary report;

[0047] S83. Add visualization charts and summaries to generate the final health consultation report, and push it to the user's terminal through the cloud platform.

[0048] Preferably, the system includes:

[0049] The data acquisition module collects user health data through the user interface unit, acquires multi-expert knowledge data through the knowledge base integration unit, and outputs the original health consultation dataset through the data verification unit.

[0050] The preprocessing module receives the original health consultation dataset, removes noise through the data cleaning unit, completes the data through the missing value processing unit, and generates standardized health data through the standardization unit.

[0051] The expert fusion module receives the standardized health data, assigns expert weights through the weight calculation unit, integrates knowledge using the fusion algorithm unit, and outputs fused knowledge data through the consistency verification unit.

[0052] The consultation generation module receives the fused knowledge data, generates preliminary consultation suggestions through the AI ​​model unit, combines them with user profile data through the personalization adjustment unit, and outputs personalized consultation suggestions through the verification unit.

[0053] The optimization processing module receives real-time user feedback data, parses the feedback information through the feedback analysis unit, adjusts suggestions using the dynamic optimization unit, and outputs optimization suggestion data through the compliance check unit.

[0054] The evaluation module receives the optimization suggestion data, evaluates the effect through the indicator calculation unit, corrects the result through the expert review unit, and outputs the consultation effect evaluation result through the result generation unit.

[0055] The parameter adjustment module receives the consultation effect evaluation results, adjusts the AI ​​model through the model parameter optimization unit, recalculates the expert weights through the weight update unit, and outputs the optimized parameter data through the integration unit.

[0056] The report output module integrates the optimization suggestion data and optimization parameter data, creates report content through the text generation unit, adds charts through the visualization unit, and outputs the final health consultation report through the push unit.

[0057] Compared with existing technologies, this invention provides an AI-based health consultation optimization method and system based on multi-expert fusion, which has the following beneficial effects:

[0058] 1. In this invention, when generating health consultation suggestions, a multi-expert fusion algorithm is used to perform knowledge fusion processing on standardized health data. Combined with expert weight allocation and consistency verification mechanisms, the weight ratio of different expert opinions can be dynamically adjusted to avoid the problem of reduced reference value of consultation suggestions due to disagreements among experts, thus ensuring the scientificity and accuracy of health consultation suggestions and improving the reliability of personalized consultation.

[0059] 2. In this invention, when optimizing health consultation suggestions, the initial health consultation suggestions are dynamically optimized by receiving real-time feedback data from users and using reinforcement learning algorithms. This allows for real-time adjustment of the suggestion content based on changes in user needs, preventing the consultation suggestions from becoming disconnected from the user's actual situation. Furthermore, it ensures timely correction of any deviations in the suggestions, guaranteeing the real-time nature and adaptability of health consultations.

[0060] 3. In this invention, when evaluating the effectiveness of health consultation, by defining multi-dimensional evaluation indicators including accuracy, response time, and user satisfaction index, and combining expert review to conduct a comprehensive evaluation of the optimization suggestions data, a comprehensive quantitative analysis of consultation services can be achieved. This avoids the problem of insufficient optimization basis due to a single evaluation indicator, thereby improving the overall quality of health consultation and user experience. Attached Figure Description

[0061] Figure 1 This is a flowchart of the AI ​​health consultation optimization method based on multi-expert fusion according to the present invention;

[0062] Figure 2 This is a diagram illustrating the architecture of the AI-based health consultation optimization system based on multi-expert fusion, as described in this invention. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] Please see Figure 1 - Figure 2The AI-based health consultation optimization method and system, based on multi-expert fusion, includes the following steps:

[0065] S1. Collect user health data and multi-expert knowledge data to generate a raw health consultation dataset;

[0066] S2. Perform data preprocessing on the original health consultation dataset, including data cleaning, missing value imputation, and feature standardization, to generate standardized health data;

[0067] S3. Based on the multi-expert fusion algorithm, standardized health data is processed for knowledge fusion, and combined with expert weight allocation and consistency verification, fused knowledge data is generated.

[0068] S4. Utilize AI consultation models to generate preliminary health consultation suggestions based on integrated knowledge data, and make personalized adjustments based on user profiles;

[0069] S5. Receive real-time feedback data from users, dynamically optimize preliminary health consultation suggestions, and generate optimized suggestion data;

[0070] S6. Evaluate the effectiveness of the optimization suggestions data using evaluation indicators, and generate consultation effectiveness evaluation results;

[0071] S7. Based on the consultation effectiveness evaluation results, adjust the parameters and expert fusion weights of the AI ​​consultation model to generate optimized parameter data;

[0072] S8. Based on the optimization suggestion data and optimization parameter data, generate the final health consultation report and output it to the user.

[0073] Generating the original health consultation dataset in S1 includes the following steps:

[0074] S11. Collect real-time health data of users through wearable devices, mobile application interfaces and medical databases, including physiological parameters, medical history records and lifestyle data, and generate a subset of user health data;

[0075] S12. Extract expert consultation rules, clinical guidelines and case data from the multi-expert knowledge base to generate a subset of multi-expert knowledge data;

[0076] S13. Integrate the subset of user health data and the subset of multi-expert knowledge data, remove redundant information, and generate the original health consultation dataset.

[0077] The generation of standardized health data in S2 includes the following steps:

[0078] S21. Clean the original health consultation dataset by using an anomaly detection algorithm to identify and remove noisy data, generating cleaned data. The specific implementation process is as follows:

[0079] First, it is necessary to clarify the definition and types of noise data. In the context of health consultation, noise data usually refers to outliers that deviate from the normal range, do not conform to medical common sense, and are different from most data patterns, caused by equipment error, human input error, transmission interference, and atypical cases.

[0080] Next, we move on to the selection and implementation phase of anomaly detection algorithms. Given the complex characteristics often present in health data, a combined algorithm strategy is typically adopted rather than a single method.

[0081] Statistical detection methods: For numerical physiological parameters, Z-score is first used for preliminary screening. This method identifies outliers far from the main data by calculating the standard deviation multiple of the data point from the mean of the overall data distribution.

[0082] Distance-based detection methods: For high-dimensional health feature vectors, normal data points should cluster in high-density areas, while abnormal points should be far away from most other points. Points that are too far away are judged as noise.

[0083] Density-based detection methods: For datasets with uneven distribution and different density regions, points with a density lower than their neighborhood are identified as outliers. This method can identify local anomalies rather than global anomalies.

[0084] After the algorithm is executed, the identified noisy data needs to be carefully removed and corrected. Finally, the cleaned data will enter the subsequent missing value imputation and feature standardization process.

[0085] S22. Perform missing value processing on the cleaned data, use interpolation to fill in the missing values, and generate complete data. The specific implementation process is as follows:

[0086] First, missing value analysis and pattern identification are required. The system will identify the location, number, and distribution pattern of missing values ​​in the dataset. This identification is crucial because interpolation is best suited for handling randomly missing data. For non-random missing data, special processing is required by combining causal analysis and expert rules.

[0087] Secondly, select an appropriate interpolation algorithm based on the characteristics of the data:

[0088] For time series data: when health data is collected in chronological order, time series interpolation is the most effective method;

[0089] Linear interpolation: This is the most commonly used and intuitive method. It assumes that the change in missing values ​​is linear between two adjacent known data points. The system calculates the slope of the line connecting these two points to determine the value of the missing point. This method is computationally efficient and suitable for physiological parameters with relatively stable trends.

[0090] Spline interpolation: When data changes have smooth curve characteristics, spline interpolation can provide more accurate estimates. It connects all known data points by constructing segments of low-order polynomial curves and ensuring the smoothness of the connection points, thereby better fitting complex trends.

[0091] For non-time series numerical data: when there is no clear temporal order between data points, but there is some spatial and characteristic proximity relationship;

[0092] K-nearest neighbor interpolation: This method is based on the "like attracts like" principle. The system first finds the K most similar complete records in the feature space that are most similar to the record containing the missing value. Then, the missing value is filled in by the mean of the corresponding features of these K nearest neighbors. This is especially useful when processing multi-dimensional features in user profiles.

[0093] Finally, after completing the interpolation, the system will evaluate and verify the interpolation effect.

[0094] S23. Perform feature standardization on the complete data. Use the Z-score standardization method to convert the data into a unified dimension and generate standardized health data. The specific process is as follows:

[0095] First, the mean and standard deviation of the feature values ​​are calculated. The system will traverse each numerical feature in the standardized health dataset and calculate the arithmetic mean and standard deviation of each feature. The mean represents the central tendency of the feature data, while the standard deviation measures the dispersion of the data.

[0096] Secondly, the Z-score transformation formula is applied for standardization. For each original data value in the dataset, the system will calculate according to the following formula:

[0097] ;

[0098] Through this transformation, the original data is converted into a new data distribution with a mean of 0 and a standard deviation of 1;

[0099] Finally, the standardized results are output and verified. After the calculation is completed, the system generates the transformed standardized health data and performs basic verification.

[0100] The generation of fused knowledge data in S3 includes the following steps:

[0101] S31. Based on expert reputation and historical performance, assign dynamic weights to each expert and generate expert weight data.

[0102] Specifically, the dynamic allocation of expert weights is achieved using the entropy weight method, as follows:

[0103] The first step is to construct an evaluation matrix: For each expert, a scoring matrix is ​​generated based on their historical accuracy, authority in their professional field, relevance to the current consultation question, and historical effectiveness.

[0104] The second step is to calculate the entropy value: using the formula... ,in Here, m represents the standardized value, and m is the number of experts. Calculate the entropy value for each indicator.

[0105] The third step is to determine the weights: weight coefficients. And perform normalization to ensure that the sum is 1;

[0106] S32. Use a weighted average method to perform knowledge fusion on standardized health data to generate initial fused data. The specific process is as follows:

[0107] The first step is to determine the expert weights: this is the foundation and key to the weighted average method;

[0108] Step 2: Obtain standardized outputs from each expert:

[0109] Each expert will independently analyze the same set of standardized health data processed by the S2 step and output their quantitative knowledge judgments. These judgments themselves must be standardized, quantifiable numerical values ​​and vectors.

[0110] Step 3: Perform weighted average calculation:

[0111] This is the core computational step of the algorithm: the system multiplies the quantitative output of each expert by its corresponding weight, and then sums all the products to obtain the final fusion result.

[0112] Its basic formula is:

[0113] ;

[0114] Through this calculation, the opinions of highly weighted experts will have a greater influence on the final result, thereby improving the overall quality and reliability of the integrated knowledge.

[0115] Step 4: Perform consistency verification and output:

[0116] After obtaining the initial weighted average result, the system will enter the consistency verification step of S33 in weight 4. After passing the verification, high-quality fused knowledge data will be generated, providing a solid data foundation for the AI ​​consultation model to generate suggestions in the subsequent S4 step.

[0117] S33. Perform consistency verification on the initial fused data, eliminate expert opinions through conflict detection mechanism, and generate consistent fused knowledge data.

[0118] Generating preliminary health consultation recommendations in S4 includes the following steps:

[0119] S41. Construct an AI consultation model. The specific process is as follows:

[0120] First, clarify the core architecture and input / output of the model. The AI ​​consultation model is essentially an intelligent decision-making system based on deep learning. Its core input is the fused knowledge data generated through step S3, which integrates quantitative knowledge from multiple experts. Its core output is the preliminary health consultation suggestions from step S4. The goal of the model is to establish a complex mapping relationship from fused knowledge to personalized suggestions.

[0121] Secondly, the model structure is designed. Given the complexity of health consultation issues, the model typically adopts a layered deep learning network structure.

[0122] Feature extraction layer: The bottom layer of the model uses convolutional networks to perform deep feature extraction and dimensionality reduction on the input fused knowledge data, capturing the complex patterns and relationships hidden in the data;

[0123] Inference Decision Layer: Above feature extraction, an inference mechanism is built, which is a deep neural network whose last layer uses the Softmax function to generate preliminary, standardized consultation suggestions;

[0124] ;

[0125] in, The output of the multi-head attention function, For querying the matrix, The key matrix, For value matrices, For splicing operations, For the output of the 1st to Nth attention heads, This is the weight matrix;

[0126] ;

[0127] in, For the first The output of each attention head, For the index of attention head, For the first The query weight matrix for each head, For the first The key weight matrix of the head, For the first The value weight matrix of each head;

[0128] ;

[0129] in, Based on the attention function, The softmax function normalizes the input into a probability distribution and applies it to the rows of a matrix. To query the dot product of the key, Scaling factor It is the transpose operator. It is a value matrix;

[0130] Personalized Adjustment Layer: This is the key innovation of the model. After generating the initial suggestions, the model does not output directly, but needs to integrate the user profiles mentioned in step S4. This process can be achieved through the attention mechanism. The model learns to assign different attention weights to different user profile features, thereby weighting and adjusting the initial suggestions to generate truly personalized consultation suggestions.

[0131] Finally, the training and optimization of the model are completed. Model construction cannot be separated from training on large-scale, high-quality data:

[0132] Training data preparation: Historical fusion knowledge data, corresponding user profiles, and health advice that has been verified by experts and ultimately proven to be effective are used as training samples;

[0133] Training process: The backpropagation algorithm and gradient descent optimizer are used to iteratively adjust all parameters in the model by minimizing the difference between the model's output suggestions and the actual effective suggestions;

[0134] Validation and Testing: The model performance is evaluated through independent validation and test sets to ensure its generalization ability. Only models that pass rigorous testing will be deployed in actual consulting.

[0135] The AI ​​consultation model adopts the Transformer architecture with the following parameters: the input layer dimension is 256, the hidden layer dimension is 512, the number of attention heads is set to 8, during training, the Adam optimizer is used, the loss function is cross-entropy, the number of iterations is 1000 rounds, and the model output generates a suggestion probability distribution through the Softmax function. This model is based on a deep learning network, inputs fused knowledge data, and outputs preliminary health consultation suggestions.

[0136] S42. Based on user profile data, including age, gender, and health preferences, personalize the initial health consultation suggestions to generate personalized consultation suggestions;

[0137] S43. Verify the rationality of personalized consultation suggestions through simulation tests, and generate verified suggestion data.

[0138] The steps involved in generating optimization suggestion data in S5 are as follows:

[0139] S51. Collect real-time user feedback data through the user interface, including satisfaction ratings and suggestions for improvement;

[0140] S52. Using reinforcement learning algorithms, personalized consultation suggestions are iteratively optimized based on real-time user feedback data to generate dynamically optimized suggestions. The specific process is as follows:

[0141] First, we define the key elements of reinforcement learning, which forms the basis for building an optimization framework;

[0142] State: The state space needs to comprehensively describe the context of the current consultation, which usually includes the content of the current personalized consultation suggestion, the user's real-time health data, user profile, and historical interaction records. A typical state is represented as a feature vector of this information.

[0143] Action: The action space is defined as the adjustment operations that the AI ​​model performs on the current consultation suggestion. This is the specific parameters in the suggestion that are fine-tuned, and it is also a module of the suggestion that is replaced.

[0144] Rewards: The reward function is the engine driving model optimization. It needs to quantify real-time user feedback data, including explicit user feedback. At the same time, the system will also mine implicit feedback and transform it into reward values. The design of the reward function is crucial, as it needs to guide the model to optimize in the direction of improving user satisfaction and health effects.

[0145] The reward function of the reinforcement learning algorithm is designed as follows:

[0146] ;

[0147] in The weighting coefficient for satisfaction. Weighting coefficients for the degree of improvement in health indicators;

[0148] Secondly, the strategy optimization and iteration loop begins. After defining the above elements, the reinforcement learning algorithm starts to work.

[0149] Policy network decision-making: The AI ​​consultation model, as a policy network, will select an action that it believes will yield the highest cumulative reward based on the current state;

[0150] Interaction and Reward Acquisition: The system pushes the adjusted new suggestions to users and collects the real-time feedback data generated by users. This feedback is calculated by the reward function to obtain a quantified reward value.

[0151] Policy update: The algorithm evaluates the merits of previous actions based on the rewards obtained and updates the parameters of the policy network accordingly. When an adjustment action brings high rewards, the policy network will strengthen the probability of choosing such actions in similar future states, and vice versa. This process realizes a closed loop of "practice-feedback-learning-improvement".

[0152] Finally, to ensure the robustness and security of the optimization process, security mechanisms must be introduced, as it is directly related to user health.

[0153] The balance between exploration and utilization: The algorithm needs to strike a balance between "utilizing" the best known suggestions and "exploring" better new suggestions. Initially, more exploration is encouraged, but as the learning progresses, it gradually leans towards utilizing proven effective suggestions.

[0154] Constraints and Boundaries: All suggested adjustments must be made within a safety boundary defined by the medical knowledge base;

[0155] Combining offline and online learning: In order to reduce the risk of trial and error with real users, the system will first use historical interaction data to conduct large-scale offline reinforcement learning to pre-train the model. In actual deployment, it will then conduct careful online learning fine-tuning with a smaller learning rate.

[0156] S53. Conduct compliance checks on dynamic optimization suggestions to ensure they comply with medical standards and generate optimization suggestion data.

[0157] The steps involved in generating consultation effectiveness evaluation results in S6 are as follows:

[0158] S61. Define evaluation metrics, including accuracy, response time, and user satisfaction index;

[0159] S62. Evaluate the effectiveness of the optimization suggestions using the test dataset, calculate the scores of each indicator, and generate preliminary evaluation results.

[0160] S63. Based on expert review, revise the preliminary evaluation results to generate the consultation effectiveness evaluation results.

[0161] The steps involved in generating optimization parameter data in S7 are as follows:

[0162] S71. Analyze the results of the consultation effectiveness evaluation to identify parameter biases and insufficient expert weights in the AI ​​consultation model;

[0163] S72. Adjust the model parameters using gradient descent to generate updated model parameters. The specific process is as follows:

[0164] First, a loss function is defined to quantify the error in consultation effectiveness. The loss function is a mathematical expression that measures the gap between the suggestions generated by the model and the expected effect. In the context of health consultation, this function is usually constructed based on the consultation effectiveness evaluation results generated in the S6 steps, and its mathematical form can be expressed as:

[0165] ;

[0166] in: It is a loss function that depends on the model parameters. , It is the number of data samples used for evaluation. The model is for the first Input data The output, i.e., the predicted consultation effect, It is the first The actual consultation effectiveness evaluation results corresponding to each sample;

[0167] Secondly, the gradient is calculated to determine the parameter update direction. The gradient is a vector that indicates the direction in which the loss function increases the fastest at the current position. Therefore, the negative gradient direction is the direction in which the function value decreases the fastest. The gradient of the model parameters is calculated by adjusting the loss function... Regarding parameters Taking the partial derivative, we get:

[0168] ;

[0169] in, For gradient operators, and The partial derivative sign indicates right The partial derivative of each component results in a derivative with... Vectors of the same dimension;

[0170] This gradient value quantifies the magnitude and direction of each parameter's "contribution" to the total error;

[0171] Finally, the model parameters are iteratively updated until convergence. After obtaining the gradient, the core update rule of gradient descent is used to adjust the model parameters:

[0172] ;

[0173] in: These are the updated parameters. These are the parameters before the update. The learning rate is a crucial hyperparameter that controls the step size of each parameter update. A learning rate that is too large can lead to oscillations or even failure to converge, while a learning rate that is too small will result in slow convergence. In order to be in The gradient vector of the loss function calculated at the specified location;

[0174] Learning rate in gradient descent Set to 0.01, the convergence threshold is a change in the loss function value of less than 0.001. Example iteration: Initial parameters. =[0.5, 0.3], optimized after 100 iterations to =[0.8, 0.1], resulting in a 90% reduction in losses;

[0175] S73. Recalculate the expert weights based on the evaluation results and generate optimized expert weight data;

[0176] S74. Integrate the updated model parameters and optimization expert weight data to generate optimization parameter data.

[0177] Generating and exporting the final health consultation report to the user in S8 includes the following steps:

[0178] S81. Combine the optimization suggestion data and optimization parameter data to generate the raw data for the report;

[0179] S82. Convert the raw report data into readable text using natural language generation technology to generate a preliminary report;

[0180] S83. Add visualization charts and summaries to generate the final health consultation report, and push it to the user's terminal via the cloud platform. The system includes:

[0181] The data acquisition module collects user health data through the user interface unit, acquires multi-expert knowledge data through the knowledge base integration unit, and outputs the original health consultation dataset through the data verification unit.

[0182] The preprocessing module receives the raw health consultation dataset, removes noise through the data cleaning unit, completes the data through the missing value handling unit, and generates standardized health data through the standardization unit.

[0183] The expert fusion module receives standardized health data, assigns expert weights through the weight calculation unit, integrates knowledge using the fusion algorithm unit, and outputs fused knowledge data through the consistency verification unit.

[0184] The consultation generation module receives integrated knowledge data, generates preliminary consultation suggestions through the AI ​​model unit, combines user profile data through the personalization adjustment unit, and outputs personalized consultation suggestions through the verification unit.

[0185] The optimization processing module receives real-time user feedback data, parses the feedback information through the feedback analysis unit, adjusts suggestions using the dynamic optimization unit, and outputs optimization suggestion data through the compliance check unit.

[0186] The evaluation module receives optimization suggestion data, evaluates the effect through the indicator calculation unit, corrects the results through the expert review unit, and outputs the consultation effect evaluation results through the result generation unit.

[0187] The parameter adjustment module receives the consultation effect evaluation results, adjusts the AI ​​model through the model parameter optimization unit, recalculates the expert weights through the weight update unit, and outputs the optimized parameter data through the integration unit.

[0188] The report output module integrates optimization suggestion data and optimization parameter data, creates report content through the text generation unit, adds charts using the visualization unit, and outputs the final health consultation report through the push unit.

[0189] The AI-based health consultation optimization method and system operation steps based on multi-expert fusion are as follows:

[0190] I. Principles of Data Acquisition and Integration

[0191] This method first collects user health data and expert knowledge data from multiple sources to form a raw dataset. The specific principle is as follows: the system obtains users' physiological parameters, medical history records and lifestyle data in real time through wearable devices, mobile application interfaces and medical databases. At the same time, it extracts consultation rules, clinical guidelines and case data from multiple expert knowledge bases. During data integration, the system removes redundant information to ensure the integrity and consistency of the data, laying the foundation for subsequent processing. The core of this step is to ensure the diversity and real-time nature of data sources to comprehensively reflect the user's health status.

[0192] II. Data Preprocessing Principles

[0193] The preprocessing stage improves data quality by cleaning, imputing missing values, and standardizing the raw data. In principle, the system uses anomaly detection algorithms to identify and remove noisy data to avoid outlier interference. For missing values, interpolation is used to fill in the missing values ​​to ensure data continuity. Finally, Z-score standardization converts the data into a uniform dimension to eliminate dimensional differences between features. The principle behind this process is to make the data more suitable for AI model analysis and reduce bias through mathematical transformations and algorithmic processing.

[0194] III. Principles of Multi-Expert Knowledge Fusion

[0195] Knowledge fusion is the core innovation of this method. Its principle is based on dynamic weight allocation and consistency verification mechanism. The system will assign dynamic weights to each expert based on the expert's reputation and historical performance factors, and use the weighted average method to perform preliminary fusion of standardized data. In principle, the opinions of experts with high weights have a greater impact on the final result, but the system will also verify consistency through conflict detection mechanism to eliminate differences of opinion. This principle ensures the reliability and objectivity of consultation suggestions.

[0196] IV. Principles of AI Consultation Suggestion Generation

[0197] The AI ​​consultation model, built using deep learning networks, generates preliminary health advice based on fused knowledge data. In principle, the model captures hidden patterns in the data through a feature extraction layer, and then makes personalized adjustments based on user profiles. After generation, the system verifies the rationality of the advice through simulation tests to ensure that it conforms to common medical knowledge. The principle of this step is to transform multi-expert knowledge into an actionable personalized solution.

[0198] V. Dynamic Optimization Processing Principle

[0199] The system optimizes suggestions in real time through reinforcement learning algorithms. In principle, it defines states, actions, and reward functions, and optimizes the suggestion content through iterative learning. At the same time, the system performs compliance checks to ensure that the suggestions comply with medical standards. This principle reflects adaptive learning capabilities and improves the real-time nature and adaptability of consultations.

[0200] VI. Principles of Effect Evaluation and Parameter Adjustment

[0201] The evaluation phase quantifies the consultation effect through multi-dimensional indicators. In principle, it combines test datasets and expert reviews for comprehensive correction. Parameter adjustment is based on the gradient descent method to optimize AI model parameters and recalculate expert weights. This principle achieves closed-loop optimization and ensures continuous system improvement.

[0202] VII. Report Output Principles

[0203] The final report generation utilizes natural language generation technology to convert optimized data into readable text, and adds visualization charts and summaries. In principle, the system integrates suggestions and parameter data and pushes them to the user end through the cloud platform to ensure the report's understandability and usability.

[0204] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0205] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An AI-based health consultation optimization method based on multi-expert fusion, characterized in that, The method includes the following steps: S1. Collect user health data and multi-expert knowledge data to generate a raw health consultation dataset; S2. Perform data preprocessing on the original health consultation dataset, including data cleaning, missing value imputation, and feature standardization, to generate standardized health data; S3. Based on the multi-expert fusion algorithm, the standardized health data is subjected to knowledge fusion processing, and combined with expert weight allocation and consistency verification, fused knowledge data is generated. S4. Using the AI ​​consultation model, preliminary health consultation suggestions are generated based on the fused knowledge data, and personalized adjustments are made based on the user profile. S5. Receive real-time feedback data from users, dynamically optimize the preliminary health consultation suggestions, and generate optimized suggestion data; S6. Evaluate the effectiveness of the optimization suggestions data using evaluation indicators, and generate consultation effectiveness evaluation results; S7. Based on the consultation effect evaluation results, adjust the parameters and expert fusion weights of the AI ​​consultation model to generate optimized parameter data; S8. Based on the optimization suggestion data and optimization parameter data, generate the final health consultation report and output it to the user terminal.

2. The AI ​​health consultation optimization method based on multi-expert fusion according to claim 1, characterized in that, The generation of the original health consultation dataset in S1 includes the following steps: S11. Collect real-time health data of users through wearable devices, mobile application interfaces and medical databases, including physiological parameters, medical history records and lifestyle data, and generate a subset of user health data; S12. Extract expert consultation rules, clinical guidelines and case data from the multi-expert knowledge base to generate a subset of multi-expert knowledge data; S13. Integrate the user health data subset and the multi-expert knowledge data subset, remove redundant information, and generate the original health consultation dataset.

3. The AI ​​health consultation optimization method based on multi-expert fusion according to claim 1, characterized in that, The generation of standardized health data in S2 includes the following steps: S21. Clean the original health consultation dataset by using an anomaly detection algorithm to identify and remove noisy data, and generate cleaned data. S22. The cleaned data is processed for missing values, and the missing values ​​are filled using interpolation to generate complete data. S23. Perform feature standardization processing on the complete data, and convert the data into a unified dimension using the Z-score standardization method to generate standardized health data.

4. The AI ​​health consultation optimization method based on multi-expert fusion according to claim 1, characterized in that, The generation of fused knowledge data in S3 includes the following steps: S31. Based on expert reputation and historical performance, assign dynamic weights to each expert and generate expert weight data. S32. Use a weighted average method to perform knowledge fusion on the standardized health data to generate initial fused data; S33. Perform consistency verification on the initial fusion data, eliminate expert opinions through conflict detection mechanism, and generate consistent fusion knowledge data.

5. The AI ​​health consultation optimization method based on multi-expert fusion according to claim 1, characterized in that, The process of generating preliminary health consultation recommendations in S4 includes the following steps: S41. Construct an AI consultation model, which is based on a deep learning network, takes the fused knowledge data as input, and outputs preliminary health consultation suggestions; S42. Based on user profile data, including age, gender, and health preferences, the preliminary health consultation suggestions are adjusted to generate personalized consultation suggestions; S43. Verify the rationality of the personalized consultation suggestions through simulation testing, and generate verified suggestion data.

6. The AI ​​health consultation optimization method based on multi-expert fusion according to claim 5, characterized in that, The generation of optimization suggestion data in S5 includes the following steps: S51. Collect real-time user feedback data through the user interface, including satisfaction ratings and suggestions for improvement; S52. Using a reinforcement learning algorithm, the personalized consultation suggestions are iteratively optimized based on the real-time user feedback data to generate dynamic optimization suggestions; S53. Perform a compliance check on the dynamic optimization suggestions to ensure they comply with medical standards, and generate optimization suggestion data.

7. The AI ​​health consultation optimization method based on multi-expert fusion according to claim 1, characterized in that, The generation of consultation effectiveness evaluation results in S6 includes the following steps: S61. Define evaluation metrics, including accuracy, response time, and user satisfaction index; S62. Evaluate the effectiveness of the optimization suggestions using the test dataset, calculate the scores of each indicator, and generate preliminary evaluation results. S63. Based on expert review, revise the preliminary assessment results to generate a consultation effectiveness assessment result.

8. The AI ​​health consultation optimization method based on multi-expert fusion according to claim 1, characterized in that, The generation of optimization parameter data in S7 includes the following steps: S71. Analyze the consultation effect evaluation results to identify parameter bias and insufficient expert weight in the AI ​​consultation model; S72. Use gradient descent to adjust the model parameters and generate updated model parameters; S73. Recalculate the expert weights based on the evaluation results and generate optimized expert weight data; S74. Integrate the updated model parameters and optimization expert weight data to generate optimization parameter data.

9. The AI ​​health consultation optimization method based on multi-expert fusion according to claim 1, characterized in that, The process of generating the final health consultation report and outputting it to the user in step S8 includes the following steps: S81. Combine the optimization suggestion data and optimization parameter data to generate the original report data; S82. Convert the original report data into readable text using natural language generation technology to generate a preliminary report; S83. Add visualization charts and summaries to generate the final health consultation report, and push it to the user's terminal through the cloud platform.

10. An AI health consultation optimization system based on multi-expert fusion, used to implement the AI ​​health consultation optimization method based on multi-expert fusion as described in any one of claims 1-9, characterized in that, The system includes: The data acquisition module collects user health data through the user interface unit, acquires multi-expert knowledge data through the knowledge base integration unit, and outputs the original health consultation dataset through the data verification unit. The preprocessing module receives the original health consultation dataset, removes noise through the data cleaning unit, completes the data through the missing value processing unit, and generates standardized health data through the standardization unit. The expert fusion module receives the standardized health data, assigns expert weights through the weight calculation unit, integrates knowledge using the fusion algorithm unit, and outputs fused knowledge data through the consistency verification unit. The consultation generation module receives the fused knowledge data, generates preliminary consultation suggestions through the AI ​​model unit, combines them with user profile data through the personalization adjustment unit, and outputs personalized consultation suggestions through the verification unit. The optimization processing module receives real-time user feedback data, parses the feedback information through the feedback analysis unit, adjusts suggestions using the dynamic optimization unit, and outputs optimization suggestion data through the compliance check unit. The evaluation module receives the optimization suggestion data, evaluates the effect through the indicator calculation unit, corrects the result through the expert review unit, and outputs the consultation effect evaluation result through the result generation unit. The parameter adjustment module receives the consultation effect evaluation results, adjusts the AI ​​model through the model parameter optimization unit, recalculates the expert weights through the weight update unit, and outputs the optimized parameter data through the integration unit. The report output module integrates the optimization suggestion data and optimization parameter data, creates report content through the text generation unit, adds charts through the visualization unit, and outputs the final health consultation report through the push unit.

Images