Multi-role configuration and effective determination method based on semantic arbitration

Abstract

This invention discloses a multi-role configuration and effective judgment method based on semantic arbitration, belonging to the field of artificial intelligence technology. The method includes: setting up multiple organizational roles and an arbitration role; receiving trial description text and quantizing it, distributing it as the first clinical trial context vector to multiple organizations, randomly assigning a first speaker, and setting up an empty discussion pool; the first speaker generating speech content based on the clinical trial context vector and writing it into the discussion pool, calculating the similarity between the speech content vector and the clinical trial context vector, and calculating the arbitration score based on multi-organization attributes; the arbitration role performing a three-state judgment based on the arbitration score and parameters, generating an arbitration decision, and determining the effective speech content vector; fusing and updating the effective vector with the clinical trial context vector and writing it into the discussion pool; terminating the discussion according to preset conditions, outputting the effective judgment based on the final vector and the discussion pool, and recording the log; this method can effectively simulate multi-organization discussion scenarios, improving the efficiency and accuracy of clinical trial discussions.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence technology, specifically a multi-role configuration and effective judgment method based on semantic arbitration. Background Technology

[0002] In clinical trial discussions, it's common practice for multiple experts to discuss and provide recommendations based on investigational drugs or devices. However, existing discussion processes lack effective management and evaluation mechanisms. Experts' statements may stray from the topic, provide invalid or incorrect information, leading to inefficient discussions and difficulty in quickly and accurately arriving at effective treatment plans. Furthermore, current natural language processing technologies, when handling multi-role discussions, cannot accurately determine and arbitrate the validity of expert statements, addressing the specific needs of clinical trials. Therefore, there is an urgent need for a method that can semantically arbitrate multi-role discussions, rationally allocate roles, and accurately determine the validity of statements to improve the quality and efficiency of medical discussions. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes a multi-role configuration and effective judgment method based on semantic arbitration. This method sets up multiple organizational roles and an arbitration role; receives trial description text and quantizes it, distributing it as the first clinical trial context vector to multiple organizations; randomly assigns a first speaker and sets up an empty discussion pool; the first speaker generates speech content based on the clinical trial context vector and writes it into the discussion pool; the similarity between the speech content vector and the clinical trial context vector is calculated, and an arbitration score is calculated based on multi-organization attributes; the arbitration role performs a three-state judgment based on the arbitration score and parameters, generates an arbitration decision, and determines the effective speech content vector; the effective vector is merged and updated with the clinical trial context vector and written into the discussion pool; the discussion is terminated according to preset conditions, and an effective judgment is output based on the final vector and the discussion pool, with logs recorded; this method can effectively simulate multi-organization discussion scenarios, improving the efficiency and accuracy of clinical trial discussions.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] Multi-role configuration and effective determination methods based on semantic arbitration include:

[0006] Set up multiple organizational roles and arbitration roles, configure attribute tuples for each organizational role, and configure arbitration parameter sets for the arbitration role;

[0007] The system receives the trial description text, vectorizes it using a preset semantic embedding model to obtain the clinical trial description vector, and distributes it to all configured organizational roles as the first clinical trial context vector for discussion. It also randomly assigns the first speaker and sets up an empty discussion pool.

[0008] The current first speaker generates speech content based on the clinical trial context vector, vectorizes the speech content and writes it into the discussion pool, calculates the similarity between the speech content vector and the current clinical trial context vector, and calculates the arbitration score by combining the organizational role attribute information;

[0009] The arbitrator performs a three-state judgment based on the arbitration score and arbitration parameters, generates an arbitration decision, and broadcasts it to all organizational roles. At the same time, it determines the effective speech content vector based on the arbitration decision.

[0010] The effective speech content vector is fused with the current clinical trial context vector, the clinical trial context vector is updated, and the data is written into the discussion pool.

[0011] The discussion is terminated according to the preset termination conditions. The validity determination is output based on the final clinical trial context vector and discussion pool, and the entire process is logged.

[0012] Specifically, the process of setting up multiple organizational roles includes:

[0013] Load multi-organization information data tables from the information system and perform entity alignment on the multi-organization information data tables;

[0014] By mapping multi-organizational professional domains to semantic vectors through a pre-trained medical knowledge graph, multi-organizational professional vectors are obtained, and experience weights are set.

[0015] Historical clinical trial discussion records from multiple organizations were collected, and keyword distributions were extracted using TF-IDF. After normalization, style vectors were generated.

[0016] The professional vectors, experience weights, and style vectors from multiple organizations are concatenated into a text description, which is then input into a contrastive learning model to generate a role profile.

[0017] Each role is assigned an attention bias parameter, and the collaborative relationships between multiple organizations are modeled using a graph neural network to generate a role association matrix.

[0018] Based on the role association matrix, and combined with the preset role templates and role configuration files, each organizational role is instantiated.

[0019] Specifically, the process of configuring the arbitration parameter set for the arbitration role includes:

[0020] Set arbitration threshold range ,in, and These represent the minimum valid threshold and the maximum conflict threshold, respectively.

[0021] By analyzing historical data on arbitration decisions in historical clinical trial discussions, machine learning algorithms were used to train and obtain arbitration weight coefficients corresponding to different professional fields.

[0022] By combining a pre-constructed clinical trial complexity assessment index system, clinical trials are classified into difficulty levels. Based on the difficulty level of the clinical trial, arbitration parameters associated with the difficulty level of the clinical trial are generated. The arbitration parameters include arbitration threshold, disagreement threshold, speaking order rules, and discussion termination condition parameters.

[0023] Specifically, the implementation process of the clinical trial description vector includes:

[0024] The input test description text is preprocessed, and the BioBERT model is used to extract the deep semantic features of the preprocessed test description text to obtain different token semantic representation vectors.

[0025] A multi-head attention mechanism is used to weight and integrate the token semantic representation vectors to generate an initial clinical trial description vector;

[0026] The historical clinical trial vector library is called up, and the semantic distance between the current initial clinical trial description vector and the historical case vectors in the historical clinical trial vector library is calculated using cosine similarity. The historical case vectors with the highest similarity are selected as reference vectors based on the semantic distance.

[0027] The initial clinical trial description vector and the reference vector are weighted and summed according to similarity weights to generate an extended initial vector that includes historical references, which is the clinical trial description vector.

[0028] Specifically, the method for randomly assigning the first speaker includes:

[0029] The degree of each multi-organization node in the generated role association matrix is ​​calculated to obtain the centrality index of the multi-organization node, and the centrality index is normalized. The degree of the multi-organization node in the role association matrix is ​​the number of other multi-organization nodes directly connected to the multi-organization node. The other multi-organization nodes are multi-organization nodes that are not themselves.

[0030] Calculate the similarity between the professional vector and the clinical trial description vector for each organizational role, sort the organizational roles according to the similarity value, and generate an initial priority queue;

[0031] Weights are assigned to the normalized centrality index and the sorting position in the initial priority queue. For each organizational role, its comprehensive score is calculated, and the organizational roles are re-sorted according to the comprehensive score to obtain a candidate multi-organization set.

[0032] Check if there are any organizations with the same comprehensive score in the candidate multi-organization set. If there are organizations with the same comprehensive score, introduce a random perturbation factor to obtain the candidate multi-organization ranking result.

[0033] Set a protagonist rotation cycle parameter to record the identifier of the current discussion cycle. If the protagonist rotation cycle is reached, select the next organization role as the new first speaker candidate from the candidate multi-organization ranking results in order. If the current discussion cycle has not reached the protagonist rotation cycle, select the organization role ranked first from the candidate multi-organization ranking results as the first speaker candidate.

[0034] Specifically, the process by which the current first speaker generates speech content based on the clinical trial context vector includes:

[0035] The first speaker inputs the clinical trial context vector into a pre-trained LSTM-based medical dialogue generation model to extract the clinical trial time-series features;

[0036] The attribute tuples are mapped to attribute embedding vectors and then weighted and fused with clinical trial time-series features to obtain a fused feature representation constrained by attribute tuples.

[0037] The fused feature representation is input into the decoder part of the LSTM model. At each time step, the decoder generates the word probability distribution based on the current state and the fused feature representation.

[0038] At each time step, a word is selected for generation based on the probability distribution of the generated words using a reinforcement learning strategy, and the corresponding reward is calculated based on the generated word sequence. Then, the parameters of the medical dialogue generation model are adjusted based on the reward value. Through iterative training, a text sequence optimized by the reinforcement learning strategy is generated.

[0039] The text sequence optimized by the reinforcement learning strategy is post-processed to convert it into structured text containing diagnostic opinions, differential diagnoses, and treatment suggestions, thereby generating the speech content.

[0040] Specifically, the arbitration role performs a three-state determination based on the arbitration score, including comparing the arbitration score with the lowest valid threshold and the highest conflict threshold;

[0041] When the arbitration score is below the minimum valid threshold, the statement is deemed invalid.

[0042] When the arbitration score is higher than the highest conflict threshold, it is deemed a valid and unique statement;

[0043] When the arbitration score is within the arbitration threshold range, an arbitration decision is generated by combining the arbitration weight coefficient and the arbitration parameters through a decision tree model.

[0044] Specifically, determining the valid speech content vector based on the arbitration decision includes:

[0045] If the arbitration decision is that the statement is invalid, then the valid statement content vector will be set to empty.

[0046] If the arbitration decision is valid and there is only one statement, then it includes:

[0047] The speech content vectors that are determined to be valid and unique are denoised to obtain denoised speech content vectors; the orthogonal projection components of the speech content vectors in the clinical trial context vector direction are obtained by calculating the inner product and magnitude between the vectors; a gating mechanism is introduced to dynamically control the information fusion intensity according to the magnitude of the orthogonal projection components; time decay weighting is applied to the speech content vectors adopted in multiple rounds to generate a set of valid vectors with confidence labels;

[0048] If the arbitration decision is that multiple statements are combined to be effective, then the multiple statement content vectors are weighted and merged according to the attributes of the organizational roles to generate a valid statement content vector.

[0049] Specifically, the process of determining validity based on the final clinical trial context vector and discussion pool output includes:

[0050] The final clinical trial context vector is input into the clinical trial conclusion prediction model based on a convolutional neural network. The output of the clinical trial conclusion prediction model is then corrected by combining the statistical information of the valid speech content vectors in the discussion pool, and the final valid judgment result is generated.

[0051] Compared with the prior art, the beneficial effects of the present invention are:

[0052] 1. This invention proposes a multi-role configuration and effective judgment method based on semantic arbitration. By setting multiple organizational roles and arbitration roles, and configuring exclusive attribute tuples and arbitration parameter sets for different roles, a simulation framework that closely resembles a real medical discussion scenario is constructed. After receiving the trial description text, it is vectorized and distributed, and the first speaker is randomly assigned to start the discussion. This process design enables clinical trial discussions to proceed in an orderly and efficient manner, laying a good foundation for subsequent speech generation and arbitration decisions, and helping to explore clinical trials more comprehensively and in-depth.

[0053] 2. This invention proposes a multi-role configuration and effective judgment method based on semantic arbitration. During the discussion, the first speaker generates the speech content based on the clinical trial context vector and calculates the arbitration score. The arbitration role makes a three-state judgment based on the score and parameters to determine the effective speech content vector. Then, the effective vector is merged and updated with the clinical trial context vector. This series of operations not only realizes the scientific evaluation and screening of the speech content, but also continuously enriches the clinical trial context information, improves the quality of the discussion, and finally terminates the discussion according to preset conditions, outputs the effective judgment, and records the log, providing strong support for medical decision-making and facilitating subsequent review and analysis. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the multi-role configuration and effective determination method based on semantic arbitration of the present invention;

[0055] Figure 2 This is a flowchart illustrating the principle of the multi-role configuration and effective judgment method based on semantic arbitration of the present invention.

[0056] Figure 3 This is a flowchart illustrating the process of determining the first speaker in the semantic arbitration-based multi-role configuration and effective judgment method of the present invention. Detailed Implementation

[0057] Example 1

[0058] Please see Figure 1 and Figure 2 The present invention provides an embodiment of a multi-role configuration and effective determination method based on semantic arbitration, the method comprising steps S1 to S7, including the following steps:

[0059] S1: Set up multiple organizational roles and arbitration roles, configure attribute tuples for each organizational role, and configure arbitration parameter sets for the arbitration role;

[0060] In this invention, the multiple organizational roles refer to clinical investigators, contract research organizations, independent expert committees, regulatory agencies, and patient organizations; the arbitrator role is the sponsor.

[0061] Furthermore, the process of configuring attribute tuples for each organizational role includes:

[0062] (1) Extract the professional vector, experience weight, and style vector of the organizational role from the role configuration file;

[0063] (2) Associate the professional vector with the preset professional field knowledge base to obtain the detailed knowledge mastery index of the organization's role in each professional field, as a professional attribute;

[0064] (3) Based on the experience weight, combined with data such as years of service and number of clinical trials participated in from the multi-organization information data table, the experience value attribute is calculated;

[0065] (4) Perform cluster analysis on style vectors to determine the speaking style type of organizational roles, such as formal, lively, concise, etc., as style attributes;

[0066] (5) Combine professional attributes, experience value attributes and style attributes into attribute tuples, and assign a unique identifier to each attribute tuple.

[0067] S2: Receive the trial description text, vectorize it to obtain the clinical trial description vector, and distribute it to all configured organizational roles as the first clinical trial context vector for discussion. Randomly assign the first speaker and set an empty discussion pool.

[0068] Furthermore, the process of distributing the clinical trial description vector to all configured organizational roles as the first clinical trial context vector discussed includes:

[0069] (1) Establish a mapping relationship between the organizational role identifier and the clinical trial description vector;

[0070] (2) The clinical trial description vector is accurately distributed to the computing node corresponding to each organizational role through message queues or distributed communication mechanisms;

[0071] (3) During the distribution process, record information such as distribution time and receiving node to ensure the traceability of data transmission;

[0072] (4) After each organizational role receives the clinical trial description vector, it stores it as the first clinical trial context vector to be discussed.

[0073] Furthermore, after S2, the method further includes: performing anomaly detection on the clinical trial description vector, determining whether there is data noise or outliers by calculating the norm and distribution of each dimension of the vector; if an anomaly is detected, correcting the outliers using a median absolute deviation method to ensure the accuracy of the clinical trial context vector input to the organizational role. The median absolute deviation method is existing technology in the field and is not an inventive solution of this application, so it will not be described in detail here.

[0074] S3: The current first speaker generates speech content based on the clinical trial context, vectorizes the speech content and writes it into the discussion pool, calculates the similarity between the speech content vector and the current clinical trial context vector, and calculates the arbitration score by combining the organizational role attribute information;

[0075] Furthermore, the method for vectorizing the speech content includes:

[0076] (1) Obtain the text of the speech to be vectorized;

[0077] (2) The hierarchical attention mechanism is used to process the speech content text to be vectorized, and the text representation after hierarchical attention processing is obtained. The hierarchical attention mechanism usually includes two levels: word level and sentence level. At the word level, attention weights are assigned to words in each sentence to highlight important words; at the sentence level, attention weights are assigned to each sentence to focus on key sentences. Through this hierarchical processing, key information in long texts can be captured better, and important information can be avoided from being buried due to the length of the text.

[0078] (3) A predefined medical terminology dictionary and corresponding standard vector space are used to identify key medical terms from the text representation after hierarchical attention processing. Then, a linear transformation method is used to align and correct the vector representation of the key medical terms with the predefined standard vector space, so that the vector of the key medical terms has a more accurate semantic representation in the standard vector space, reducing semantic deviation caused by term ambiguity or different expression methods, and obtaining the text representation after vector space alignment and correction.

[0079] (4) Input the text representation after vector space alignment correction into a pre-trained neural network model, perform further feature extraction and encoding on the text through the model's encoding layer, and then extract a fixed-dimensional vector from the model's output as the speech semantic vector representation;

[0080] (5) Calculate the KL divergence between the speech semantic vector and the clinical trial context vector. KL divergence is an index used to measure the difference between two probability distributions. Here, it can be used to measure the degree of semantic deviation between the speech semantic vector and the clinical trial context vector. The smaller the KL divergence value, the closer the speech semantics are to the clinical trial context. The larger the KL divergence value, the higher the degree of semantic deviation. The KL divergence calculation formula is the prior art in this field and is not an inventive solution of this application. It will not be elaborated here.

[0081] (6) While storing the speech semantic vector, record the timestamp and role identifier of the speech, and output the speech semantic vector representation with timestamp and role identifier metadata. The timestamp is the time sequence information of the speech in the discussion process, and the role identifier is the role information of the multiple organizations in the speech, such as internal medicine multiple organizations, surgical multiple organizations. These metadata are of great significance for subsequent analysis of the time distribution of the speech and the speech characteristics of different roles and organizations.

[0082] Furthermore, the arbitration score is calculated as follows:

[0083] (1) Obtain the vectorized speech content vector and the current clinical trial context vector and perform standardization processing to eliminate the impact of vector dimension differences;

[0084] (2) The cosine similarity algorithm is used to calculate the basic similarity between the vectorized speech content vector and the clinical trial context vector. By calculating the dot product of the two vectors, an initial value reflecting the semantic association between the two is obtained. The value ranges from -1 to 1. The closer the value is to 1, the higher the semantic matching degree, and the closer it is to -1, the greater the semantic difference.

[0085] (3) Extract three core dimensions from the organizational role attribute information: professional matching degree, experience level, and credibility of historical statements. Among them, professional matching degree is determined by the cosine similarity between the professional field vector of the organizational role and the clinical trial field vector; experience level is divided into 1-5 levels according to the organizational role's years of service, number of clinical trial projects participated in, etc., corresponding to a quantitative score of 0.2-1.0; credibility of historical statements is calculated based on the frequency of past statements being adopted, and the formula is the number of valid historical statements divided by the total number of statements, resulting in a quantitative value between 0 and 1.

[0086] (4) Preset the weight coefficients for each attribute dimension according to the type and stage of the clinical trial. The weight of professional matching degree is usually set to 0.4-0.6, the weight of experience level is set to 0.2-0.3, and the weight of historical credibility is set to 0.2-0.3. The sum of the three weight coefficients is 1. Multiply the quantitative value of each attribute by the corresponding weight coefficient to obtain the weighted score of each attribute. Then add the three weighted scores to obtain the comprehensive weighted score of the organizational role attribute. The score range is between 0 and 1.

[0087] (5) Assign weights to the basic semantic similarity and the comprehensive weighted score of the organizational role attributes respectively, where the semantic similarity weight is set to 0.5-0.7 and the comprehensive weighted score of the attributes is set to 0.3-0.5, and the sum of the two weights is 1. Multiply the basic semantic similarity value by its corresponding weight, and add the comprehensive weighted score of the organizational role attributes by its corresponding weight to obtain the final arbitration score.

[0088] S4: The arbitration role performs a three-state judgment based on the arbitration score and arbitration parameters, generates an arbitration decision and broadcasts it to all organizational roles. At the same time, it determines the effective speech content vector based on the arbitration decision.

[0089] S5: Merge the effective speech content vector with the current clinical trial context vector, update the clinical trial context vector, and write it into the discussion pool;

[0090] Furthermore, the fusion of the effective speech content vector with the current clinical trial context vector is achieved through a gated loop unit structure. The effective speech content vector and the clinical trial context vector are used as inputs to update the clinical trial context vector. The gated loop unit structure is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0091] S6: Terminate the discussion according to the preset termination conditions, output a valid judgment based on the final clinical trial context vector and discussion pool, and record the entire process log.

[0092] Furthermore, terminating the discussion according to preset termination conditions includes:

[0093] Set a maximum discussion round threshold; the discussion will terminate when the number of discussion rounds reaches this threshold. Calculate the difference between the current clinical trial context vector and the previous clinical trial context vector; the discussion will terminate when the difference is less than a preset convergence threshold. Detect whether there are multiple consecutive rounds of invalid statements in the discussion pool; if so, the discussion will terminate.

[0094] Furthermore, the process of recording the entire process log includes: after each multi-organization speech, recording the organization role, speech content, speech time, and arbitration score; after arbitration by the arbitration role, recording the arbitration decision and the vector of valid speech content; during the discussion, recording the clinical trial context vector updated in each round; storing all recorded information in chronological order in a distributed log system and encrypting it.

[0095] Example 2

[0096] Please see Figure 3 The process of setting up multiple organizational roles as described in this embodiment includes:

[0097] Load multi-organization information data tables from the information system and perform entity alignment on the multi-organization information data tables;

[0098] By mapping multi-organizational professional domains to semantic vectors through a pre-trained medical knowledge graph, multi-organizational professional vectors are obtained, and experience weights are set.

[0099] Historical clinical trial discussion records from multiple organizations were collected, and keyword distributions were extracted using TF-IDF. After normalization, style vectors were generated. TF-IDF is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0100] The professional vectors, experience weights, and style vectors from multiple organizations are concatenated into a text description, which is then input into a contrastive learning model to generate a role profile.

[0101] Attention bias parameters are configured for each role, and the collaborative relationships between multiple organizations are modeled through graph neural networks to generate a role association matrix. The graph neural network is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0102] Based on the role association matrix, and combined with the preset role templates and role configuration files, each organizational role is instantiated.

[0103] Furthermore, the specific process of setting experience weights includes:

[0104] First, historical experience data from multiple organizations is collected and cleaned to identify core indicators such as years of participation and number of successful projects. Then, the indicators are quantified and graded, weights are assigned according to importance, and individual experience scores are calculated. Next, the individual experience scores are summed to obtain an initial comprehensive score, which is then normalized to eliminate dimensional differences. Finally, a time decay factor is used for dynamic correction to obtain the experience weights in the 0-1 range.

[0105] Furthermore, the individual experience score is obtained by multiplying the quantitative value of each organization's individual indicator by the corresponding weight.

[0106] Furthermore, a time decay factor is introduced to correct the normalized score, including: assigning a weight of 1.2 times to empirical data within the last 3 years, a weight of 1.0 times to data within the last 3-5 years, and a weight of 0.8 times to data older than 5 years, in order to highlight the reference value of recent experience and obtain the final empirical weight.

[0107] Furthermore, the methods for obtaining the preset character templates are as follows:

[0108] First, the job descriptions and behavioral guidelines for different positions were extracted from medical guidelines, hospital regulations, and industry standards to form the basic framework of the template. Then, the actual performance data of each role in historical case discussions were analyzed to summarize high-frequency behavioral patterns and effective interaction methods, which were then added to the template. Finally, experts in the medical field were invited to evaluate, revise, and improve the preliminary template to ensure that it meets the needs of clinical practice and professional standards, thus obtaining the role template.

[0109] The process of configuring the arbitration parameter set for the arbitration role includes:

[0110] Set arbitration threshold range ,in, and These represent the minimum valid threshold and the maximum conflict threshold, respectively.

[0111] By analyzing historical data on arbitration decisions in historical clinical trial discussions, machine learning algorithms are used to train and obtain arbitration weight coefficients corresponding to different professional fields. The machine learning algorithms are existing technology in this field and are not the inventive solution of this application, so they will not be described in detail here.

[0112] By combining a pre-constructed clinical trial complexity assessment index system, clinical trials are classified into difficulty levels. Based on the difficulty level of the clinical trial, arbitration parameters associated with the difficulty level of the clinical trial are generated. The arbitration parameters include arbitration threshold, disagreement threshold, speaking order rules, and discussion termination condition parameters.

[0113] Among them, the arbitration threshold is the critical value for judging the validity of a speech, including the minimum valid threshold and the maximum conflict threshold; the disagreement threshold is the critical value for judging whether there is a substantial conflict in the speeches of multiple roles. If it is lower than this value, it is considered that the opinions are basically consistent, and if it is higher than this value, in-depth discussion needs to be initiated; the speaking order rule is used to regulate the order of speeches of multiple roles to ensure that key roles participate in the discussion at the appropriate stage. This invention presets the speaking order according to the importance of roles; the discussion termination condition parameter is used to determine when to end the discussion, including the maximum number of rounds, the convergence similarity threshold, and the core opinion achievement rate.

[0114] Furthermore, the process of generating the arbitration threshold includes:

[0115] The effective speaking score ranges for trials of different difficulty levels are extracted from historical clinical trial data. For example, the effective speaking scores for simple trials are concentrated in the range of 0.6-0.9. Based on this, the basic intervals are set: the initial value of the minimum effective threshold for simple trials is 0.6, and the initial value of the maximum conflict threshold is 0.85; the initial value of the minimum effective threshold for complex trials is 0.7, and the initial value of the maximum conflict threshold is 0.9, thus obtaining the arbitration threshold.

[0116] Furthermore, the process of generating the divergence threshold includes:

[0117] Analyze the distribution of speech similarity under different difficulty levels in historical experiments, and calculate the critical similarity value for reaching consensus and generating disagreement in experiments of the same difficulty level. For example, the critical value for disagreement is 0.6 in simple experiments and 0.5 in complex experiments.

[0118] Furthermore, the termination condition discussed in this invention is set as follows:

[0119] The maximum number of rounds for the simple experiment is set to 3; the convergence similarity threshold is set to 0.8, and the experiment terminates if the similarity between two consecutive rounds is greater than or equal to 0.8; the core opinion achievement rate is greater than or equal to 80%. The maximum number of rounds for the complex experiment is set to 5; the convergence similarity threshold is set to 0.7; the core opinion achievement rate is greater than or equal to 60%.

[0120] Furthermore, the process of classifying the difficulty levels of clinical trials includes:

[0121] (1) Construct a multi-dimensional and multi-level clinical trial complexity assessment index system, including symptom manifestation dimension, symptom severity, symptom duration, and examination result dimension;

[0122] (2) Collect clinical trial data and perform preprocessing;

[0123] (3) Use the analytic hierarchy process (AHP) to determine the weights of each indicator, including:

[0124] The complexity assessment of clinical trials is taken as the target layer, symptom presentation, examination results, disease diagnosis, and treatment response are taken as the criteria layer, and the specific indicators under each criteria layer are taken as the protocol layer.

[0125] By comparing the importance of each indicator, a judgment matrix is ​​constructed. For example, in terms of the number of symptoms, the severity of symptoms, and the duration of symptoms under the symptom manifestation dimension, experts believe that the severity of symptoms is slightly more important than the number of symptoms, and the duration of symptoms is significantly more important than the number of symptoms.

[0126] By solving for the largest eigenvalue and the corresponding eigenvector of the judgment matrix, the weight vector of each indicator is obtained, and the consistency of the judgment matrix is ​​checked to ensure the rationality of the weights.

[0127] (4) Calculate the complexity score for each clinical trial based on the weight of each indicator and the preprocessed clinical trial data;

[0128] (5) Based on the complexity score of the clinical trial, cluster analysis is used to divide the clinical trial into different difficulty levels. For example, the clinical trial is divided into three levels: simple, medium and complex. Cluster analysis is prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0129] The implementation process of the clinical trial description vector includes:

[0130] The input test description text is preprocessed, and the BioBERT model is used to extract the deep semantic features of the preprocessed test description text to obtain different token semantic representations. The BioBERT model is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0131] A multi-head attention mechanism is used to weight and integrate the semantic representation of the token to generate an initial clinical trial description vector;

[0132] Furthermore, a multi-head attention mechanism is employed to integrate the semantic representation of the token, generating an initial clinical trial description vector, including:

[0133] (1) After extracting deep semantic features, a multi-head attention mechanism is used to process word vectors. The multi-head attention mechanism allows the model to focus on different parts of the text from multiple different representation subspaces, thereby better capturing long-distance dependencies and important semantic information in the text.

[0134] (2) The word vector sequence output by the BioBERT model is used as the input of the multi-head attention mechanism. The multi-head attention mechanism will perform a weighted summation on each word vector. The weights are determined by the semantic similarity between word vectors. Through the parallel computation of multiple attention heads, multiple different attention weight distributions are obtained. Then, these attention weight distributions are concatenated and linearly transformed to obtain a comprehensive vector representation.

[0135] (3) Further feature fusion and dimensionality reduction operations are performed on the vector after processing by the multi-head attention mechanism to finally generate a 768-dimensional initial clinical trial description vector. This 768-dimensional initial clinical trial description vector contains the main semantic information of the trial description text and can better represent the overall characteristics of the clinical trial.

[0136] The historical clinical trial vector library is called up, and the semantic distance between the current initial clinical trial description vector and the historical case vectors in the historical clinical trial vector library is calculated using cosine similarity. The historical case vectors with the highest similarity are selected as reference vectors based on the semantic distance.

[0137] It's important to explain that cosine similarity measures the degree of similarity between two vectors by calculating the cosine of the angle between them. For the current initial clinical trial description vector and the historical case vectors in the historical clinical trial vector database, their dot product and magnitude are calculated respectively, and then their similarity value is calculated using the cosine similarity formula. The similarity value ranges from -1 to 1. The closer the value is to 1, the more semantically similar the two vectors are; the closer the value is to -1, the less semantically similar they are; a value of 0 indicates that the two vectors are orthogonal, meaning there is no semantic association.

[0138] The initial clinical trial description vector and the reference vector are weighted and summed according to similarity weights to generate an extended initial vector that includes historical references, which is the clinical trial description vector.

[0139] The method for randomly assigning the first speaker includes:

[0140] The degree of each multi-organization node in the generated role association matrix is ​​calculated to obtain the centrality index of the multi-organization node, and the centrality index is normalized. The degree of the multi-organization node in the role association matrix is ​​the number of other multi-organization nodes directly connected to the multi-organization node. The other multi-organization nodes are multi-organization nodes that are not themselves.

[0141] Calculate the similarity between the professional vector and the clinical trial description vector for each organizational role, sort the organizational roles according to the similarity value, and generate an initial priority queue;

[0142] In this invention, cosine similarity is used to calculate the similarity between the professional vector of each organizational role and the clinical trial description vector. However, cosine similarity is a prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0143] Weights are assigned to the normalized centrality index and the sorting position in the initial priority queue. For each organizational role, its comprehensive score is calculated, and the organizational roles are re-sorted according to the comprehensive score to obtain a candidate multi-organization set.

[0144] Furthermore, the comprehensive score is obtained by converting the ranking position of the organizational role in the initial priority queue into a corresponding score, such as the higher the ranking, the higher the score. Then, the normalized organizational role centrality index, that is, the centrality value with a uniform value range after eliminating the influence of the unit of measurement, and the corresponding score of the role are used as two core evaluation dimensions. Finally, the normalized centrality index of each organizational role is multiplied by its corresponding weight, and the ranking score of the role in the initial priority queue is multiplied by its corresponding weight. The sum of the two products is the comprehensive score of the organizational role.

[0145] Check if there are any organizations with the same overall score in the candidate multi-organization set. If so, introduce a random perturbation factor to obtain the candidate multi-organization ranking result; the range of the random perturbation factor is... ;

[0146] Set a protagonist rotation cycle parameter to record the identifier of the current discussion cycle. If the protagonist rotation cycle is reached, select the next organization role as the new first speaker candidate from the candidate multi-organization ranking results in order. If the current discussion cycle has not reached the protagonist rotation cycle, select the organization role ranked first from the candidate multi-organization ranking results as the first speaker candidate.

[0147] The process by which the current first speaker generates speech content based on the clinical trial context vector includes:

[0148] The first speaker inputs the clinical trial context vector into a pre-trained LSTM-based medical dialogue generation model to extract the clinical trial temporal features. The Long Short-Term Memory (LSTM) network is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0149] Among them, the LSTM model, with its internal memory unit and gating mechanism, can effectively capture the temporal dependencies in the context vector of clinical trials, perform layer-by-layer processing and feature extraction on the input vector, and output the temporal feature representation of clinical trials. The LSTM model is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0150] The attribute tuples are mapped to attribute embedding vectors and then weighted and fused with clinical trial time-series features to obtain a fused feature representation constrained by attribute tuples.

[0151] In this method, the attribute embedding vector is weighted and fused with the time-series features of clinical trials. This fusion method enables the model to consider both the time-series information of clinical trials and the subjective attribute information of multiple organizations when generating text.

[0152] For example, if the multi-organization's expertise is in the field of cardiovascular disease, then when generating disease-related vocabulary, the probability of cardiovascular disease-related vocabulary will increase; if the multi-organization's experience is inclined towards conservative treatment, then when generating treatment plans, the probability of vocabulary related to aggressive treatment plans will decrease; if the multi-organization possesses advanced diagnostic technology capabilities, then when generating diagnostic opinions, the probability of vocabulary citing advanced diagnostic technologies will increase. In this way, the generated speech content is made to conform to the multi-organization's expertise, experience, and capability characteristics.

[0153] The fused feature representation is input into the decoder part of the LSTM model. At each time step, the decoder generates the word probability distribution based on the current state and the fused feature representation.

[0154] At each time step, a word is selected for generation based on the probability distribution of the generated words using a reinforcement learning strategy, and the corresponding reward is calculated based on the generated word sequence. Then, the parameters of the medical dialogue generation model are adjusted based on the reward value. Through iterative training, a text sequence optimized by the reinforcement learning strategy is generated.

[0155] The reward function is designed based on the quality, accuracy, completeness, and compliance with medical standards of the generated speech content. For example, a higher reward is given if the generated speech content contains accurate diagnostic opinions, reasonable differential diagnoses, and effective treatment suggestions; a lower reward or penalty is given if the speech content contains errors or is incomplete. Then, reinforcement learning algorithms, such as policy gradient algorithms, are used to adjust the parameters of the medical dialogue generation model based on the reward values, making the model more inclined to select words that can obtain high rewards. The policy gradient algorithm is prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0156] The text sequence optimized by the reinforcement learning strategy is post-processed to convert it into structured text containing diagnostic opinions, differential diagnoses, and treatment suggestions, thereby generating the speech content.

[0157] Furthermore, the optimized text sequence undergoes post-processing to convert it into structured text containing diagnostic opinions, differential diagnoses, and treatment recommendations. Specific steps include:

[0158] (1) Key information such as disease name, symptom description, diagnostic basis, list of differential diagnoses and corresponding treatment suggestions are extracted from the text sequence using the relation extraction method. The relation extraction method is the prior art in this field and is not an inventive solution of this application. It will not be described in detail here.

[0159] (2) Organize the extracted key information. For example, list the diagnostic opinions separately, sort the differential diagnoses according to their probability, and present the treatment suggestions in different categories such as drug treatment and surgical treatment to form a structured text.

[0160] The arbitration role performs a three-state determination based on the arbitration score, including comparing the arbitration score with the lowest valid threshold and the highest conflict threshold.

[0161] When the arbitration score is below the minimum valid threshold, the statement is deemed invalid.

[0162] When the arbitration score is higher than the highest conflict threshold, it is deemed a valid and unique statement;

[0163] When the arbitration score is within the arbitration threshold range, an arbitration decision is generated by combining the arbitration weight coefficient and the arbitration parameters through a decision tree model.

[0164] The determination of the valid speech content vector based on the arbitration decision includes:

[0165] If the arbitration decision is that the statement is invalid, then the valid statement content vector will be set to empty.

[0166] If the arbitration decision is valid and there is only one statement, then it includes:

[0167] The speech content vectors that are determined to be valid and unique are denoised to obtain denoised speech content vectors; the orthogonal projection components of the speech content vectors in the clinical trial context vector direction are obtained by calculating the inner product and magnitude between the vectors; a gating mechanism is introduced to dynamically control the information fusion intensity according to the magnitude of the orthogonal projection components; time decay weighting is applied to the speech content vectors adopted in multiple rounds to generate a set of valid vectors with confidence labels;

[0168] If the arbitration decision is that multiple statements are combined to be effective, then the multiple statement content vectors are weighted and merged according to the attributes of the organizational roles to generate a valid statement content vector.

[0169] The process for determining validity based on the final clinical trial context vector and discussion pool output includes:

[0170] The final clinical trial context vector is input into the clinical trial conclusion prediction model based on a convolutional neural network. The output of the clinical trial conclusion prediction model is then corrected by combining the valid speech content vectors in the discussion pool to generate the final valid judgment result. The convolutional neural network is prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0171] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the present invention. All of these variations are within the protection scope of the present invention.

[0172] If the technical solution disclosed herein involves personal information, the product using this technical solution has clearly informed the user of the personal information processing rules and obtained the user's voluntary consent before processing the personal information. If the technical solution disclosed herein involves sensitive personal information, the product using this technical solution has obtained the user's separate consent before processing the sensitive personal information, and also meets the requirement of "express consent". For example, at personal information collection devices such as cameras, clear and prominent signs are set up to inform users that they have entered the scope of personal information collection and that personal information will be collected. If an individual voluntarily enters the collection scope, it is deemed that they have agreed to the collection of their personal information; or on the personal information processing device, with clear signs / information informing users of the personal information processing rules, authorization is obtained from the individual through pop-up information or by asking the individual to upload their personal information; wherein, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the types of personal information processed.

Claims

1. A multi-role configuration and effective judgment method based on semantic arbitration, characterized in that, include: Set up multiple organizational roles and arbitration roles, configure attribute tuples for each organizational role, and configure arbitration parameter sets for the arbitration role; The system receives the trial description text, vectorizes it using a preset semantic embedding model to obtain the clinical trial description vector, and distributes it to all configured organizational roles as the first clinical trial context vector for discussion. It also randomly assigns the first speaker and sets up an empty discussion pool. The current first speaker generates speech content based on the clinical trial context vector, vectorizes the speech content and writes it into the discussion pool, calculates the similarity between the speech content vector and the current clinical trial context vector, and calculates the arbitration score by combining the organizational role attribute information; The arbitrator performs a three-state judgment based on the arbitration score and arbitration parameters, generates an arbitration decision, and broadcasts it to all organizational roles. At the same time, it determines the effective speech content vector based on the arbitration decision. The effective speech content vector is fused with the current clinical trial context vector, the clinical trial context vector is updated, and the data is written into the discussion pool. The discussion is terminated according to the preset termination conditions. The validity determination is output based on the final clinical trial context vector and discussion pool, and the entire process is logged.

2. The multi-role configuration and effective determination method based on semantic arbitration as described in claim 1, characterized in that, The process of setting up multiple organizational roles includes: Load multi-organization information data tables from the information system and perform entity alignment on the multi-organization information data tables; By mapping multi-organizational professional domains to semantic vectors through a pre-trained medical knowledge graph, multi-organizational professional vectors are obtained, and experience weights are set. Historical clinical trial discussion records from multiple organizations were collected, and keyword distributions were extracted using TF-IDF. After normalization, style vectors were generated. The professional vectors, experience weights, and style vectors from multiple organizations are concatenated into a text description, which is then input into a contrastive learning model to generate a role profile. Each role is assigned an attention bias parameter, and the collaborative relationships between multiple organizations are modeled using a graph neural network to generate a role association matrix. Based on the role association matrix, and combined with the preset role templates and role configuration files, each organizational role is instantiated.

3. The multi-role configuration and effective determination method based on semantic arbitration as described in claim 2, characterized in that, The process of configuring the arbitration parameter set for the arbitration role includes: Set arbitration threshold range ,in, and These represent the minimum valid threshold and the maximum conflict threshold, respectively. By analyzing historical data on arbitration decisions in historical clinical trial discussions, machine learning algorithms were used to train and obtain arbitration weight coefficients corresponding to different professional fields. By combining a pre-constructed clinical trial complexity assessment index system, clinical trials are classified into difficulty levels. Based on the difficulty level of the clinical trial, arbitration parameters associated with the difficulty level of the clinical trial are generated. The arbitration parameters include arbitration threshold, disagreement threshold, speaking order rules, and discussion termination condition parameters.

4. The multi-role configuration and effective determination method based on semantic arbitration as described in claim 3, characterized in that, The implementation process of the clinical trial description vector includes: The input test description text is preprocessed, and the BioBERT model is used to extract the deep semantic features of the preprocessed test description text to obtain different token semantic representation vectors. A multi-head attention mechanism is used to weight and integrate the token semantic representation vectors to generate an initial clinical trial description vector; The historical clinical trial vector library is called up, and the semantic distance between the current initial clinical trial description vector and the historical case vectors in the historical clinical trial vector library is calculated using cosine similarity. The historical case vectors with the highest similarity are selected as reference vectors based on the semantic distance. The initial clinical trial description vector and the reference vector are weighted and summed according to similarity weights to generate an extended initial vector that includes historical references, which is the clinical trial description vector.

5. The multi-role configuration and effective determination method based on semantic arbitration as described in claim 4, characterized in that, The method for randomly assigning the first speaker includes: The degree of each multi-organization node in the generated role association matrix is ​​calculated to obtain the centrality index of the multi-organization node, and the centrality index is normalized. The degree of the multi-organization node in the role association matrix is ​​the number of other multi-organization nodes directly connected to the multi-organization node. The other multi-organization nodes are multi-organization nodes that are not themselves. Calculate the similarity between the professional vector and the clinical trial description vector for each organizational role, sort the organizational roles according to the similarity value, and generate an initial priority queue; Weights are assigned to the normalized centrality index and the sorting position in the initial priority queue. For each organizational role, its comprehensive score is calculated, and the organizational roles are re-sorted according to the comprehensive score to obtain a candidate multi-organization set. Check if there are any organizations with the same comprehensive score in the candidate multi-organization set. If there are organizations with the same comprehensive score, introduce a random perturbation factor to obtain the candidate multi-organization ranking result. Set a protagonist rotation cycle parameter to record the identifier of the current discussion cycle. If the protagonist rotation cycle is reached, select the next organization role as the new first speaker candidate from the candidate multi-organization ranking results in order. If the current discussion cycle has not reached the protagonist rotation cycle, select the organization role ranked first from the candidate multi-organization ranking results as the first speaker candidate.

6. The multi-role configuration and effective determination method based on semantic arbitration as described in claim 5, characterized in that, The process by which the current first speaker generates speech content based on the clinical trial context vector includes: The first speaker inputs the clinical trial context vector into a pre-trained LSTM-based medical dialogue generation model to extract the clinical trial time sequence features; The attribute tuples are mapped to attribute embedding vectors and then weighted and fused with clinical trial time-series features to obtain a fused feature representation constrained by attribute tuples. The fused feature representation is input into the decoder part of the LSTM model. At each time step, the decoder generates the word probability distribution based on the current state and the fused feature representation. At each time step, a word is selected for generation based on the probability distribution of the generated words using a reinforcement learning strategy, and the corresponding reward is calculated based on the generated word sequence. Then, the parameters of the medical dialogue generation model are adjusted based on the reward value. Through iterative training, a text sequence optimized by the reinforcement learning strategy is generated. The text sequence optimized by the reinforcement learning strategy is post-processed to convert it into structured text containing diagnostic opinions, differential diagnoses, and treatment suggestions, thereby generating the speech content.

7. The multi-role configuration and effective determination method based on semantic arbitration as described in claim 6, characterized in that, The arbitration role performs a three-state determination based on the arbitration score, including comparing the arbitration score with the lowest valid threshold and the highest conflict threshold. When the arbitration score is below the minimum valid threshold, the statement is deemed invalid. When the arbitration score is higher than the highest conflict threshold, it is deemed a valid and unique statement; When the arbitration score is within the arbitration threshold range, an arbitration decision is generated by combining the arbitration weight coefficient and the arbitration parameters through a decision tree model.

8. The multi-role configuration and effective determination method based on semantic arbitration as described in claim 7, characterized in that, The determination of the valid speech content vector based on the arbitration decision includes: If the arbitration decision is that the statement is invalid, then the valid statement content vector will be set to empty. If the arbitration decision is valid and there is only one statement, then it includes: The speech content vectors that are determined to be valid and unique are denoised to obtain denoised speech content vectors; the orthogonal projection components of the speech content vectors in the clinical trial context vector direction are obtained by calculating the inner product and magnitude between the vectors; a gating mechanism is introduced to dynamically control the information fusion intensity according to the magnitude of the orthogonal projection components; time decay weighting is applied to the speech content vectors adopted in multiple rounds to generate a set of valid vectors with confidence labels; If the arbitration decision is that multiple statements are combined to be effective, then the multiple statement content vectors are weighted and merged according to the attributes of the organizational roles to generate a valid statement content vector.

9. The multi-role configuration and effective determination method based on semantic arbitration as described in claim 8, characterized in that, The process for determining validity based on the final clinical trial context vector and discussion pool output includes: The final clinical trial context vector is input into the clinical trial conclusion prediction model based on a convolutional neural network. The output of the clinical trial conclusion prediction model is then corrected by combining the statistical information of the valid speech content vectors in the discussion pool, and the final valid judgment result is generated.

Images