A Guided Assisted Assessment Information Processing Method and System

Abstract

This invention belongs to the field of auxiliary data processing, and particularly relates to a guided auxiliary assessment information processing method and system. This method constructs a matching decision parameter set between an initial target and a standard knowledge base, and gradually optimizes the user profile and decision parameters by combining a dynamic step-by-step guided question-and-answer strategy. It uses a probability sequence iterative update mechanism to dynamically adjust the question-and-answer guidance frequency. Based on a decision discrimination model, it calculates the confidence sequence of the matching target and achieves multi-round feedback optimization through threshold comparison: when the confidence level meets the standard, the optimal strategy is output; when it does not meet the standard, step-by-step enhancement is executed repeatedly until a preset round threshold is reached, ultimately achieving accurate decision-making in medical assessment. This application, through the collaborative mechanism of the knowledge base and question-and-answer strategy, adopts a confidence-driven dynamic discrimination process, significantly improving the accuracy and reliability of assessment results in complex target scenarios, while also possessing an abnormal round control mechanism to ensure the integrity of the processing flow.

Description

Technical Field

[0001] This invention belongs to the field of auxiliary data processing, and in particular relates to a guided auxiliary evaluation information processing method and system. Background Technology

[0002] While existing technologies have made some progress in the field of intelligent medical consultation, they still suffer from common shortcomings. First, existing methods largely rely on pre-set question-and-answer templates (CN111159369B) or fixed-process text analysis (CN113921149B), lacking the ability to dynamically guide complex objectives and failing to adapt to the diverse needs of clinical consultation. Second, most systems employ single-modal data processing (CN117423480A, CN118352004A), failing to integrate multi-dimensional medical data, thus limiting the accuracy of objective identification and the comprehensiveness of treatment recommendations. Third, existing technologies suffer from a lag in updating medical knowledge bases (CN119181480A), failing to integrate the latest medical research findings and clinical guidelines in real time, affecting the timeliness of diagnosis. Furthermore, traditional methods suffer from rigid human-computer interaction modes (CN111159369B, CN118352004A), lacking personalized guidance strategies, resulting in low patient participation and a tendency to overlook key objectives. These shortcomings collectively result in significant deficiencies in the clinical applicability, diagnostic accuracy, and user experience of existing intelligent consultation systems. There is an urgent need to develop intelligent diagnosis and treatment methods that can achieve dynamic multi-round guidance, multimodal data fusion, and self-learning capabilities. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes a guided auxiliary assessment information processing method and system. This method constructs a matching decision parameter set between an initial target and a standard knowledge base, and progressively optimizes user profiles and decision parameters using a dynamic, step-by-step guided question-and-answer strategy. It dynamically adjusts the question-and-answer guidance frequency using a probability sequence iterative update mechanism. Based on a decision discrimination model, it calculates the confidence sequence of the matching target and achieves multi-round feedback optimization through threshold comparison: when the confidence level meets the standard, the optimal strategy is output; when it does not meet the standard, step-by-step enhancement is executed repeatedly until a preset round threshold is reached, ultimately achieving accurate decision-making in medical assessment. This application, through the collaborative mechanism of the knowledge base and question-and-answer strategy, and employing a confidence-driven dynamic discrimination process, effectively solves the problems of insufficient information collection and single decision-making basis in traditional medical assistance, significantly improving the accuracy and reliability of assessment results in complex target scenarios. Simultaneously, it possesses an abnormal round control mechanism to ensure the integrity of the processing flow.

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

[0005] A guided, assisted assessment information processing method includes:

[0006] Based on the initial question-and-answer information and a pre-set standard knowledge base, a preliminary matching is performed to obtain the initial matching target sequence decision parameter set;

[0007] Based on the initial matching target sequence decision parameter set, combined with the preset step-by-step guided question-and-answer assistance strategy and synchronously calling the standard knowledge base, a user profile and the corresponding step-by-step enhanced decision parameter set for the matching target sequence are obtained.

[0008] The step-by-step guided question-and-answer assistance strategy constructs the question-and-answer frequency of the matching target sequence in the next round by step-by-step enhancing the initial discrimination probability sequence corresponding to each matching target in the decision parameter set after each round of question-and-answer update.

[0009] Based on the user profile and the step-by-step enhanced decision parameter set of the corresponding matching target sequence, combined with the preset decision discrimination model, the confidence sequence corresponding to the matching target sequence is obtained.

[0010] The confidence sequence corresponding to the matching target sequence is compared with the preset confidence threshold. If there is at least one matching target whose confidence is greater than the preset confidence threshold, the parameter strategy space corresponding to the matching target with the highest confidence is output.

[0011] If the confidence of all matching targets is less than the preset confidence threshold, then determine whether the number of rounds of the current step-by-step guided question-and-answer assistance process is less than the preset question round threshold. If it is less, then repeat the step-by-step guided question-and-answer assistance and decision-making process. If, until the number of question rounds is greater than or equal to the preset question round threshold, all the confidence of the corresponding matching target sequence is still less than or equal to the preset confidence threshold, then reply with a question-and-answer error.

[0012] Specifically, the process of obtaining the initial matching target sequence decision parameter set includes:

[0013] Based on a pre-set multimodal preloading model combined with a pre-set historical information database and the user's initial input text information, an initial target ternary information sequence is obtained;

[0014] The initial target ternary information sequence includes the target name, the matching target text parameters, and the initial correlation matching degree between the target name and the matching target text parameters;

[0015] Based on the initial target ternary information sequence, combined with the preset standard knowledge base and the preset extended phrase correction model, the initial matching target sequence decision parameter set is obtained in descending order of the initial association matching degree.

[0016] Specifically, the initial matching target sequence decision parameter set includes the initial matching target and the corresponding standard target parameter text feature set, the correction accuracy of the target keyword group corresponding to each matching target, and the initial association matching degree sequence corresponding to the initial matching target sequence;

[0017] The extended phrase correction model is obtained by training a matching confidence algorithm combined with a preset target extended phrase library. It is used to standardize and correct non-standard target keyword phrases in the initial target ternary information sequence and obtain the correction accuracy of the corresponding target keyword phrases.

[0018] The target extended phrase library is composed of a standard target name, a non-standard target keyword phrase containing at least one key character from the standard target name, and associated aliases.

[0019] Specifically, 4. The process of obtaining user profiles and corresponding step-by-step enhanced decision parameter sets includes:

[0020] Based on the initial matching target sequence decision parameter set, combined with the preset target parameter-profile association mapping space and the first supplementary parameter information, a user profile and an initial enhanced decision parameter space are obtained.

[0021] The first supplementary parameter information includes the basic information and location information of the user corresponding to the matching target;

[0022] The user profile includes the same parameters as the first supplementary parameter information;

[0023] The target parameter-profile association mapping space is obtained by combining the user profile with the standard knowledge base and historical information database, and pre-training the association matching algorithm.

[0024] The initial enhanced decision parameter space includes user profiles, a descending sequence of initial matching targets, and corresponding initial discrimination probability sequences and matching target text parameters;

[0025] Based on the numerical ratio of the initial discrimination probabilities corresponding to all matching targets in the initial augmented decision parameter space, the frequency of second supplementary information queries corresponding to each matching target in the initial augmented decision parameter space is obtained.

[0026] Specifically, the process of obtaining user profiles and corresponding step-by-step enhanced decision parameter sets also includes:

[0027] Based on the initial enhanced decision parameter space and the standard knowledge base, a second cyclic question and answer text set is generated by combining the frequency of second supplementary information queries corresponding to each matching target with a preset text random generation model. The standard knowledge base and the preset extended phrase correction model are called simultaneously to correct the generated second cyclic question and answer text set.

[0028] The initial augmented decision parameter space is updated based on the corrected second-cycle question-and-answer text set to obtain the second initial discrimination probability corresponding to each matching target after the update.

[0029] Specifically, the process of obtaining user profiles and corresponding step-by-step enhanced decision parameter sets also includes:

[0030] Based on the updated second initial discrimination probability corresponding to each matching target and the initial association mapping threshold, the initial matching target sequence is filtered to obtain the filtered and updated second matching target sequence and the corresponding augmented decision parameter space. The above process of obtaining the second matching target sequence and the corresponding augmented decision parameter space is repeated until all initial discrimination probabilities in the updated matching target sequence are greater than the initial association mapping threshold and the matching target corresponding to the maximum initial discrimination probability remains unchanged. The target sequence and the corresponding augmented decision parameter space after step-by-step augmentation are then obtained.

[0031] Specifically, the target parameter-profile association mapping space includes a three-level mapping; the three-level mapping includes a BMI index-target mapping layer, an age-target mapping layer, and a gender-target mapping layer;

[0032] The construction process of the BMI indicator-target mapping layer includes:

[0033] Given a fixed BMI index, the correlation analysis algorithm is used to combine the frequency of occurrence and target parameter information of each level of BMI index in different regions with the corresponding target within a unit of time to obtain the correlation degree between each level of BMI index in each region and different targets.

[0034] The BMI index-target mapping layer is constructed based on the BMI index at each level in each region, the target name, and the correlation between the BMI index at each level and different targets.

[0035] Specifically, the construction process of the age-target mapping layer includes:

[0036] Based on the correlation between the BMI index at each level in each region and different targets as a weighted weight, combined with the age level information in each region, the correlation between each age level in each region and different targets is obtained through correlation analysis.

[0037] The age levels are defined as a series of 5-year intervals.

[0038] The age-target mapping layer is constructed based on the age level, target name, and correlation between different targets corresponding to each age level in each region;

[0039] The construction process of the gender-target mapping layer includes:

[0040] Based on the correlation between each age group and different targets in each region as a weighted weight, combined with the gender ratio information under each age group in each region, the correlation between each gender and different targets in each region is obtained, and the gender-target layer is obtained.

[0041] Based on the BMI indicator-target mapping layer, age-target mapping layer, and gender-target mapping layer, the target parameter-profile association mapping space is obtained through a cascading method.

[0042] Specifically, the preset historical information database is constructed from user history and real-time question-and-answer results information through knowledge graphs and graph databases;

[0043] The historical information database and the standard knowledge base are connected through the extended phrase correction model and the index information constructed based on the correlation between question-and-answer information and standard target parameter information.

[0044] A guided auxiliary assessment information processing system includes: an initial matching module, a step-by-step question-and-answer module, and a discrimination module;

[0045] The discrimination module includes a confidence unit, a secondary discrimination unit, and an inquiry adjustment unit;

[0046] The initial matching module is used to obtain initial question and answer information, and to perform preliminary matching based on the initial question and answer information and a preset standard knowledge base to obtain an initial matching target sequence decision parameter set;

[0047] The step-by-step question-answering module is used to obtain a user profile and a corresponding step-by-step enhanced decision parameter set for the matching target sequence based on the initial matching target sequence decision parameter set combined with a preset step-by-step guided question-answering assistance strategy and synchronously calling the standard knowledge base;

[0048] The confidence unit obtains the confidence sequence corresponding to the matching target sequence based on the user profile and the step-by-step enhanced decision parameter set of the corresponding matching target sequence, combined with a preset decision discrimination model.

[0049] The secondary discrimination unit is used to compare the confidence sequence corresponding to the matching target sequence with a preset confidence threshold. If there is at least one matching target whose confidence is greater than the preset confidence threshold, the parameter strategy space corresponding to the matching target with the highest confidence is output.

[0050] If the confidence scores of all matching targets are less than the preset confidence threshold, the query adjustment unit determines whether the number of queries in the current step-by-step guided question-and-answer assistance process is less than the preset query round threshold. If it is less, the step-by-step guided question-and-answer assistance and decision-making process is repeated. If, until the number of queries is greater than or equal to the preset query round threshold, all the confidence scores of the corresponding matching target sequence are still less than or equal to the preset confidence threshold, then a question-and-answer anomaly is reported.

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

[0052] This invention addresses the shortcomings of existing technologies by constructing a collaborative mechanism between a multimodal preloading model and a historical case database. This transforms the user's initial input text into an initial target ternary information sequence containing the target name, text parameters, and correlation matching degree. Combined with an extended phrase correction model, non-standard target words are standardized, effectively solving the problems of target expression diversity and knowledge base matching deviation. Based on a dynamic weighting mechanism of a three-level target parameter-profile association mapping space, and through regionalized correlation degree calculation and a hierarchical progressive mapping strategy, the invention achieves accurate quantification of the correlation between user multidimensional profile features and the target. Finally, it employs a step-by-step guided question-answering strategy based on the discriminant probability ratio. The dynamic question-answering frequency allocation method, combined with the iterative optimization mechanism of the text random generation model and the extended phrase correction model, achieves targeted and semantically standardized supplementary information collection. Through a multi-round iterative discrimination model driven by confidence thresholds, and utilizing the probability sequence filtering and dynamic screening mechanism of the target sequence during the parameter space update process, the fault tolerance and convergence efficiency of the decision-making process are enhanced. In addition, the index connection architecture based on the knowledge graph case database and the standard knowledge base, combined with the semantic expansion capability of the target extended phrase database, forms a bidirectional mapping channel covering the differences between standard terms and user expressions, thereby improving the robustness of target matching and knowledge base compatibility from the perspective of data collaboration. Attached Figure Description

[0053] Figure 1 This is a flowchart of a guided auxiliary assessment information processing method according to Embodiment 1 of the present invention;

[0054] Figure 2 This is a block diagram of a guided auxiliary evaluation information processing system according to Embodiment 2 of the present invention. Detailed Implementation

[0055] Example 1

[0056] Existing intelligent consultation technologies (such as template question-answering trees, BERT target standardization, and large language models) have significant limitations in intelligent consultation systems, telemedicine, and chronic disease management: static models cannot dynamically optimize the consultation path based on real-time patient feedback (such as fuzzy targets and multimodal data), leading to missed diagnoses of atypical cases; the interactive interface lacks dynamic visualization guidance (such as probabilistic graph iteration) and patient interaction mechanisms, resulting in incomplete information collection; and the management of historical medical data is rigid and disconnected from real-time consultations, making it difficult to support personalized decision-making. These problems restrict the efficiency and accuracy of scenarios such as clinical auxiliary diagnosis and community health management. Therefore, please refer to... Figure 1 The present invention provides an embodiment of a guided assisted assessment information processing method, comprising the following steps:

[0057] S1. Initialize user consultation information, obtain initial question and answer information, perform preliminary matching and word standardization calibration based on the initial question and answer information and a preset standard knowledge base, and obtain the initial matching target sequence decision parameter set;

[0058] Furthermore, in this embodiment, the matching target is the disease information matched during the pre-diagnosis stage;

[0059] S2. Based on the initial matching target sequence decision parameter set, combined with the preset step-by-step guided question-and-answer assistance strategy and simultaneously calling the standard knowledge base, obtain the user profile and the corresponding step-by-step enhanced decision parameter set of the matching target sequence, and use the question-and-answer information after standardization of each step phrase to update the step-by-step enhanced decision parameter set of the matching target sequence in real time.

[0060] The step-by-step guided question-and-answer assistance strategy constructs the question-and-answer frequency of the matching target sequence in the next round by step-by-step enhancing the initial discrimination probability sequence corresponding to each matching target in the decision parameter set after each round of question-and-answer update.

[0061] S3. Based on the user profile and the step-by-step enhanced decision parameter set of the corresponding matching target sequence, and combined with the preset decision discrimination model, obtain the confidence sequence corresponding to the matching target sequence;

[0062] Furthermore, the decision discrimination model in this embodiment is constructed by combining a confidence function with a decision tree algorithm.

[0063] S4. Compare the confidence sequence corresponding to the matching target sequence with the preset confidence threshold. If there is at least one matching target whose confidence is greater than the preset confidence threshold, output the parameter strategy space corresponding to the matching target with the highest confidence.

[0064] Furthermore, the parameter strategy space in this embodiment includes the pathological parameters corresponding to the target (disease) with the highest confidence level and the corresponding matching diagnosis and treatment methods and rehabilitation suggestions.

[0065] S5. If the confidence of all matching targets is less than the preset confidence threshold, determine whether the number of rounds of the current step-by-step guided question-and-answer assistance process is less than the preset question round threshold. If it is less, repeat the step-by-step guided question-and-answer assistance and decision-making process. If the confidence of all matching target sequences is still less than or equal to the preset confidence threshold until the number of question rounds is greater than or equal to the preset question round threshold, then reply with a question-and-answer error.

[0066] To further illustrate the implementation process of the above method more clearly, a specific example is given here, which includes the following stages:

[0067] Initialization and target acquisition:

[0068] User information initialization:

[0069] Initialize and obtain information about the current user's specified patient (including the patient's historical medical records and reference information from hospitals / physical examination institutions; load the local image recognition and embedding model on the device; high-performance mobile phones or terminals can choose to load the local knowledge base to speed up the diagnosis).

[0070] The user initiates a consultation request and provides a preliminary description of their symptoms. This information is then processed by the knowledge base, and the Agent is invoked to guide the user to supplement and confirm key diagnostic symptoms based on the user's symptom matching degree from high to low. This process generates an initial set of matching symptom sequence decision parameters based on the user's current symptom description.

[0071] Based on the content of the initial matching symptom sequence decision parameter set, guide the user to supplement basic information in the first step and determine whether the basic information has been completed: height, weight, age, gender, etc. If the information is complete, a diagnostic probability set (from high to low) is generated. Based on the content of the decision parameter set, the system focuses on asking about the pathological diagnosis with the highest probability and guides the user to gradually supplement symptom information in the second step or directly confirm the symptoms asked by the AI ​​in multiple rounds / combinations. If not, the system returns to force supplementation.

[0072] The second step involves gradually supplementing the symptom information, including: duration, location and severity of pain, mental state, questions asked, and stool samples. Once the information is complete, the user is guided to the third step to selectively fill in more auxiliary diagnostic information, including hospital history diagnoses, hospital lab reports, CT scans, MRI scans, and other records.

[0073] Furthermore, during the multi-round / combined confirmation process here, the underlying standard knowledge base and the preset extended phrase correction model are used simultaneously to correct the user input in real time.

[0074] Knowledge base pre-diagnosis: Based on the complete symptom description provided by the user, the knowledge base is invoked in the form of a parameter set to generate knowledge base-level pre-diagnosis information;

[0075] Furthermore, in this embodiment, the process of updating the step-by-step augmentation decision parameter set of the matching target sequence in real time using the standardized question-and-answer information of each step phrase is as follows:

[0076] By utilizing the disease information gradually supplemented in each step of the question-and-answer process, combined with the standardized question-and-answer information from the standard knowledge base, the step-by-step enhancement decision parameter set corresponding to the matching target sequence of the previous step is updated and enhanced.

[0077] Large-scale model follow-up: The model combines the pre-diagnosis output from the knowledge base with the content of the user's diagnostic decision parameter set for verification, and outputs the decision confidence parameter of the current matched disease. The backend determines whether the decision confidence parameter of the preferred pathology is greater than the preset confidence threshold. If it is greater, the diagnostic content is output based on the preferred pathology, including diagnostic information, disease name, and suggested treatment / medication. At the same time, the user is guided to choose expert consultation, online medicine purchase, or go to the hospital to register for consultation, etc. Finally, the user is guided to score the diagnosis and the process ends.

[0078] If the number of queries is less than the threshold, it is determined whether the number of queries has been greater than or equal to the threshold. If not, the pre-diagnosis and follow-up diagnosis stages are repeated. When the number of queries is greater than or equal to the threshold, if the confidence level corresponding to the pathological decision confidence parameter is still less than the preset confidence threshold, an abnormality is output, i.e., diagnosis is not possible. The user is advised to seek medical diagnosis, and recommended hospital information is provided. Finally, the user is guided to rate the diagnosis and the process ends.

[0079] Furthermore, the process of obtaining the initial matching target sequence decision parameter set in this embodiment includes:

[0080] Based on a preset multimodal preloading model, combined with a preset historical information database and the user's initial input text information, an initial target ternary information sequence is obtained; in this embodiment, the historical information database is the information database corresponding to the patient's historical diagnostic cases.

[0081] The initial target ternary information sequence includes the target name, the matching target text parameters, and the initial correlation matching degree between the target name and the matching target text parameters; the target name is the name of the disease, and the same applies to the following description.

[0082] Based on the initial target ternary information sequence, combined with the preset standard knowledge base and the preset extended phrase correction model, the initial matching target sequence decision parameter set is obtained in descending order of the initial association matching degree.

[0083] Furthermore, in this embodiment, the preset historical information database is constructed from user history and real-time diagnostic information through a knowledge graph and a graph database;

[0084] Furthermore, in this embodiment, the historical information base and the standard knowledge base are connected through the extended phrase correction model and the index information constructed based on the correlation between the queried pathological information and the standard pathological information.

[0085] Furthermore, in this embodiment, the multimodal preloading model is constructed by integrating a pre-trained Chinese BERT model, a pre-trained ResNet-18 algorithm, and an LSTM algorithm. The pre-trained Chinese BERT model is used to process text data in the pre-loaded case information database and the user's initial input consultation information, the pre-trained ResNet-18 algorithm is used to process image data in the pre-loaded case information database, and the LSTM algorithm is used to process time-series numerical data in the pre-loaded case information database.

[0086] Furthermore, the exemplary process of multimodal preloading model construction and data processing in this embodiment includes:

[0087] Input mode:

[0088] Text data: User symptom descriptions (e.g., "headache for three days, accompanied by nausea"), medical record texts.

[0089] Image data: Medical images (CT scans, X-rays), scanned copies of laboratory reports.

[0090] Time series data: historical timeline of medical visits (such as blood pressure monitoring records and medication cycles).

[0091] Model ensemble strategy:

[0092] Chinese BERT (Text Processing):

[0093] Use a pre-trained BERT-Base-Chinese model (12-layer Transformer, 768-dimensional hidden layers).

[0094] After inputting the symptom text, the 768-dimensional vector of the CLS tag is extracted as the semantic feature of the symptom.

[0095] Example: User inputs "throbbing pain in temples" → Output feature vector;

[0096] ResNet-18 (Image Processing):

[0097] Load the ResNet-18 model pre-trained on ImageNet, remove the fully connected layers, and retain the output of the last convolutional layer (512-dimensional features).

[0098] Image preprocessing: Normalized to 224×224 resolution, RGB channels normalized.

[0099] Example: CT image → Output feature vector;

[0100] LSTM (Time Series Processing):

[0101] A two-layer LSTM network (128-dimensional hidden layer) is used as input to standardized time series data (such as blood pressure value series).

[0102] Time alignment: Pad zeros to the maximum length of unequal-length sequences and extract the hidden state of the last time step.

[0103] Example: Blood pressure records [120 / 80, 130 / 85, ...] → Output feature vector;

[0104] The semantic features, image features, and numerical sequence features of the above-mentioned diseases are fused and input into the graph database (Neo4j) of the case information database. Similarity retrieval is performed, and the output is: triple <disease name, matching pathology, association matching degree>.

[0105] For example, the initial matching process includes:

[0106]

[0107] Furthermore, the initial matching target sequence decision parameter set in this embodiment includes the initial matching target and the corresponding standard pathological text feature set, the correction accuracy of the target keyword group corresponding to each matching target, and the initial association matching degree sequence corresponding to the initial matching target sequence;

[0108] The extended phrase correction model is obtained by training a matching confidence algorithm combined with a preset target extended phrase library. It is used to standardize and correct non-standard target keyword phrases in the initial target ternary information sequence and obtain the correction accuracy of the corresponding target keyword phrases.

[0109] The target extended phrase library is obtained by combining a standard target name, a non-standard target keyword phrase containing at least one key character from the standard target name, and associated aliases. Furthermore, in this embodiment, the target extended phrase library is a disease extended phrase library, and the other targets in the embodiment also represent the meaning of diseases.

[0110] Furthermore, in this embodiment, the preset historical information database is constructed from user history and real-time question-and-answer results information through a knowledge graph and a graph database;

[0111] Furthermore, in this embodiment, the historical information base and the standard knowledge base are connected through the extended phrase correction model and the index information constructed based on the correlation between question-and-answer information and standard target parameter information.

[0112] Furthermore, the core implementation principle of the extended phrase correction model in this embodiment includes:

[0113] Model architecture:

[0114] Input: User's original symptom phrase (e.g., "headache" or "shortness of breath").

[0115] Processing flow:

[0116] Keyword segmentation: Extract core words for symptoms ("head", "asthma") using medical word segmentation tools (such as LTP).

[0117] Extended Library Search: Search for non-standard words containing the core words and their corresponding standard words in the extended term library for the disease symptoms.

[0118] Example library entries:

[0119]

[0120] Confidence calculation: The matching score of the candidate standard word is calculated by weighting the edit distance (Levenshtein Distance) and word vector similarity (Word2Vec). For example: Confidence = 0.6 × (1 - edit distance / maximum length) + 0.4 × word vector cosine similarity;

[0121] Threshold filtering: retain candidate words with confidence scores > 0.7, and output the final correction results in order of score.

[0122] Training data construction:

[0123] Extract doctor-patient dialogue records from electronic medical records and manually annotate the mapping relationship between non-standard symptoms and standard terminology.

[0124] Expand the non-standard lexicon using data augmentation techniques (such as replacing synonyms and adding noise).

[0125] Real-time processing steps and examples:

[0126] Step 1: Input processing and initial triple generation;

[0127] User input: "Headache for three days, occasional shortness of breath";

[0128] Multimodal preloading model processing:

[0129] Text modality: Extracting disease semantic vectors (CLS vectors) using Chinese BERT.

[0130] Image modality: Skip this step if there is no image input.

[0131] Time series data: Analyze historical heart rate records (e.g., [72,85,78] bpm), and generate time series features using LSTM.

[0132] Output initial triples:

[0133]

[0134]

[0135] Step 2: Expansion phrase correction and parameter set construction;

[0136] Non-standard word correction:

[0137] The input word "head swelling" matches the standard word "headache" in the expansion library, with a confidence of 0.85;

[0138] The input word "difficulty breathing" is corrected to "dyspnea", with a confidence of 0.92.

[0139] Update triples:

[0140]

[0141] Generate decision parameter set:

[0142]

[0143] <00003​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​Anemia (similarity 0.70), benign paroxysmal positional vertigo (beta-ocular block) (similarity 0.65);

[0156] Corrected matching:

[0157] Benign paroxysmal positional vertigo (BPPV) (similarity 0.65 × 0.88 = 0.572);

[0158] Meniere's disease (new match, similarity 0.75 × 0.88 = 0.66);

[0159] Result: Meniere's disease became the primary candidate target.

[0160] Furthermore, the process of obtaining the user profile and the corresponding step-by-step enhanced decision parameter set in this embodiment includes:

[0161] Based on the initial matching target sequence decision parameter set, combined with the preset target parameter-profile association mapping space and the first supplementary parameter information, a user profile and an initial enhanced decision parameter space are obtained.

[0162] Furthermore, the target parameter in this embodiment is the standard pathological parameter corresponding to each disease. This standard pathological parameter is obtained from a professional medical information database and is organized and completed by professionals in the field to obtain a standard knowledge base (i.e., a standard medical knowledge base).

[0163] The first supplementary parameter information includes the basic information and location information of the user corresponding to the matching target; wherein, the basic information includes height, weight, age, gender and location information;

[0164] The user profile includes the same parameters as the first supplementary parameter information;

[0165] The target parameter-profile association mapping space is obtained by combining the user profile with the standard knowledge base and historical information database, and pre-training the association matching algorithm.

[0166] The initial enhanced decision parameter space includes user profiles, a descending sequence of initial matching targets, and corresponding initial discrimination probability sequences and matching target text parameters;

[0167] Based on the numerical ratio of the initial discrimination probabilities corresponding to all matching targets in the initial augmented decision parameter space, the frequency of second supplementary information queries corresponding to each matching target in the initial augmented decision parameter space is obtained.

[0168] Based on the initial enhanced decision parameter space and the standard knowledge base, a second cyclic question and answer text set is generated by combining the frequency of second supplementary information queries corresponding to each matching target with a preset text random generation model. The standard knowledge base and the preset extended phrase correction model are called simultaneously to correct the generated second cyclic question and answer text set.

[0169] The initial augmented decision parameter space is updated based on the corrected second-cycle question-and-answer text set to obtain the second initial discrimination probability corresponding to each matching target after the update.

[0170] Based on the updated second initial discrimination probability corresponding to each matching target and the initial association mapping threshold, the initial matching target sequence is filtered to obtain the filtered and updated second matching target sequence and the corresponding enhanced decision parameter space.

[0171] Repeat the above process of obtaining the second matching target sequence and the corresponding enhanced decision parameter space until all initial discrimination probabilities in the updated matching target sequence are greater than the initial association mapping threshold and the matching target corresponding to the maximum initial discrimination probability remains unchanged, and obtain the target sequence and the corresponding enhanced decision parameter space after step-by-step enhancement.

[0172] Furthermore, in this embodiment, to better illustrate the process of obtaining the step-by-step enhanced decision parameter set, the following specific implementation process is given, which includes:

[0173] Example Background:

[0174] A 35-year-old female patient, 160 cm tall and weighing 72 kg (BMI 28, classified as obese), residing in Shanghai, presented with throbbing headache accompanied by nausea. Initially matched candidate pathological targets were migraine, hypertensive encephalopathy, and increased intracranial pressure, with initial probabilities of 55%, 25%, and 20%, respectively.

[0175] Step 1: User profile and initial parameter space construction;

[0176] The system dynamically adjusts its mapping based on the patient's basic information (age, gender, BMI, and geographic location) and a pre-defined target parameter-profile association mapping space. This mapping space is generated based on historical case data from the Shanghai area and reflects the correlation patterns between different population characteristics and diseases.

[0177] Obesity association correction: Obesity (BMI=28) increased the association with migraine by 15% and with hypertensive encephalopathy by 30%.

[0178] Age correction: 35 years old is a high-risk age for migraines, with a 10% increased probability; while the probability of hypertensive encephalopathy is reduced by 5% in younger patients.

[0179] Gender correction: Women have an additional 20% increased risk of developing migraines.

[0180] After multi-dimensional corrections, the probability of migraine increased from 55% to 79.8%, the probability of hypertensive encephalopathy was slightly adjusted to 31.3%, and the probability of increased intracranial pressure remained unchanged at 20%. The system generates an initial enhanced parameter space containing the corrected probabilities, user profiles, and target descriptions, providing a benchmark for subsequent question-and-answer guidance.

[0181] For example, the initial enhancement parameter space includes:

[0182]

[0183] Step 2: Dynamic question-answer generation and semantic correction;

[0184] Based on the corrected probability ratios, the system automatically allocates the frequency of questions for each pathology in each round of Q&A. Migraine has the highest probability (79.8%), accounting for the majority of the 6 questions in this round, while hypertensive encephalopathy and increased intracranial pressure are each allocated 2 questions.

[0185] Questions related to migraines (6 questions):

[0186] Focus on typical features, such as "Did you experience visual auras such as flashes of light or scotomas before the headache?" and "Is the pain concentrated on one side and accompanied by photophobia?"

[0187] Questions about hypertensive encephalopathy (2 questions):

[0188] Pay attention to acute symptoms, such as "Has your blood pressure suddenly risen to above 180 / 120 mmHg recently?";

[0189] Questions about increased intracranial pressure (2 questions):

[0190] Ask about specific symptoms, such as "Did you experience blurred vision or projectile vomiting?";

[0191] The frequency of inquiries mentioned above is just an assumption. In actual applications, inquiries should be focused on those with a higher frequency ratio.

[0192] After the user answers, the system uses an extended phrase correction model to convert colloquial expressions into standard medical terms. For example, the patient's description of "flashes in front of eyes" is corrected to "visual aura," and "blood pressure is a little high" is standardized to "stage 2 hypertension (160 / 100 mmHg)."

[0193] Step 3: Parameter update and target selection;

[0194] The system adjusts the confidence probabilities of each pathology based on the user's responses:

[0195] Supportive responses (such as confirming visual aura): The probability of migraine increases by 18%, reaching 97.8%.

[0196] Negative answers (e.g., denying a sudden increase in blood pressure): The probability of hypertensive encephalopathy decreases by 15%, to 16.3%; the probability of increased intracranial pressure decreases to 5% due to denying blurred vision.

[0197] Based on a preset probability threshold (0.5), the system screens out hypertensive encephalopathy and increased intracranial pressure, retaining only migraine as the sole candidate target. The updated parameter space focuses on in-depth validation of migraine, including new symptom parameters (such as visual aura confirmation and blood pressure recording) and an adjusted high confidence level (97.8%).

[0198] Step 4: Iteration Termination and Result Output;

[0199] Since only migraine remains in the current matching target sequence and its probability far exceeds the threshold, and this target has maintained its maximum probability for two consecutive iterations, the system determines that the convergence condition has been met and terminates the question-and-answer process. The final output is an enhanced decision parameter set.

[0200] Furthermore, the target parameter-profile association mapping space in this embodiment includes a three-level mapping; the three-level mapping includes a BMI index-target mapping layer, an age-target mapping layer, and a gender-target mapping layer;

[0201] Furthermore, the construction process of the BMI indicator-target mapping layer in this embodiment includes:

[0202] Given a fixed BMI index, the correlation analysis algorithm is used to combine the frequency of (illness) occurrence and target parameter information of each level of BMI index and corresponding different targets in different regions within a unit of time to obtain the correlation degree between each level of BMI index and different targets in each region.

[0203] The BMI index-target mapping layer is constructed based on the BMI index at each level in each region, the target name, and the correlation between the BMI index at each level and different targets.

[0204] Furthermore, to more clearly illustrate the three-level mapping of the target parameter-image association mapping space, this embodiment provides some corresponding examples, specifically including:

[0205] BMI Indicator - Target Mapping Layer Construction Example:

[0206] Scenario: Constructing a BMI indicator-target mapping layer for East China;

[0207] Data source: 100,000 electronic medical records from top-tier hospitals in East China over the past 5 years, statistically analyzing the disease distribution of patients with different BMI levels.

[0208] BMI grading and related medical conditions:

[0209] BMI ≥ 28 (obese):

[0210] Hypertension: prevalence rate 32% (correlation 0.82);

[0211] Type 2 diabetes: prevalence rate 18% (association 0.75);

[0212] Osteoarthritis: prevalence rate 12% (correlation 0.68);

[0213] BMI 24-27.9 (overweight):

[0214] Fatty liver: 25% of cases (correlation 0.70);

[0215] Sleep apnea: frequency of occurrence 10% (correlation 0.60);

[0216] BMI 18.5-23.9 (normal):

[0217] Functional gastrointestinal disorders: 15% of cases (correlation 0.55);

[0218] BMI < 18.5 (underweight):

[0219] Anemia: 20% of cases (correlation 0.65);

[0220] The mapping result is as follows:

[0221]

[0222]

[0223] Furthermore, the construction process of the age-target mapping layer includes:

[0224] Based on the correlation between the BMI index at each level in each region and different targets as a weighted weight, combined with the age level information in each region, the correlation between each age level in each region and different targets is obtained through correlation analysis.

[0225] Furthermore, in this embodiment, the age levels are defined as one level every 5 years;

[0226] The age-target mapping layer is constructed based on the age level, target name, and correlation between different targets corresponding to each age level in each region;

[0227] Example of building an age-target mapping layer:

[0228] Scenario: Based on the BMI-target layer in East China, construct an age-target mapping layer for the 35-39 age group;

[0229] Data weighting rule: The correlation between BMI and disease risk is adjusted by combining age distribution.

[0230] Calculation example:

[0231] The association between hypertension and age group 35-39 was adjusted:

[0232] For the BMI ≥ 28 group: correlation coefficient 0.82 × 30% obesity rate in this age group = 0.246;

[0233] For the BMI group of 24-27.9: correlation coefficient 0.70 × overweight percentage 40% = 0.280;

[0234] Overall correlation coefficient: 0.246 + 0.280 = 0.526 (0.75 after standardization);

[0235] Mapping layer structure:

[0236]

[0237] Furthermore, the construction process of the gender-target mapping layer in this embodiment includes:

[0238] Based on the correlation between each age group and different targets in each region as a weighted weight, combined with the gender ratio information under each age group in each region, the correlation between each gender and different targets in each region is obtained, and the gender-target layer is obtained.

[0239] Furthermore, the gender-target mapping layer construction example in this embodiment includes:

[0240] Scenario: Based on the 35-39 age group in East China, construct a gender segmentation mapping;

[0241] Data source: Male-to-female patient ratio (45% male, 55% female) and gender distribution of the disease in this age group.

[0242] Calculation example:

[0243] Correction for gender association in hypertension:

[0244] Male patients account for 60% → increased correlation: 0.75 × (1 + 0.6) = 1.20;

[0245] Female patients account for 40% → correlation decay: 0.75 × (1 - 0.2) = 0.60;

[0246] Gender-related correction for anxiety disorders:

[0247] Female patients account for 70% → increased correlation: 0.58 × (1 + 0.7) = 0.986;

[0248] The mapping layer structure includes:

[0249]

[0250]

[0251] Based on the BMI indicator-target mapping layer, age-target mapping layer, and gender-target mapping layer, the target parameter-profile association mapping space is obtained through a cascading method.

[0252] Furthermore, examples of dynamic diagnostic decisions for the aforementioned target parameter-profile association mapping space include:

[0253] System behavior:

[0254] Initial matching: The user's complaint of "headache" initially matched with migraine (probability 0.65) and anxiety (probability 0.50).

[0255] Image correction:

[0256] BMI = 29 → Probability of hypertension + 0.82 × regional weight 0.3 → Increased to 0.65 + 0.25 = 0.90;

[0257] For women aged 38 and above, the probability of anxiety disorder increases by 1.7 to 0.50 × 1.7 = 0.85.

[0258] Q&A guidance:

[0259] For anxiety disorders: Ask, "Does this also involve insomnia or palpitations?" (Frequency rate: 40%)

[0260] For migraines: Ask questions such as "Is the headache unilateral? Are there any visual auras?" (50% frequency)

[0261] User responses: Insomnia confirmed, unilateral headache denied → Confidence for anxiety disorder increased to 0.95, and confidence for migraine decreased to 0.60.

[0262] Output results: Anxiety disorder is the preferred diagnosis, and psychological assessment and anti-anxiety treatment are recommended.

[0263] In summary, this embodiment first achieves heterogeneous feature space mapping of textual disease descriptions, medical image features, and physiological temporal data by integrating a multimodal preloaded model of BERT, ResNet-18, and LSTM. In the textual modality, the BERT model captures the contextual semantic associations of disease descriptions through a deep attention mechanism (such as temporal feature matching between "throbbing pain in the temples" and "pulsating headache"). In the image modality, the ResNet-18 convolutional kernel extracts local pathological features of medical images (such as the spatial distribution pattern of temporal lobe morphological abnormalities in CT images). In the temporal modality, the LSTM iterative... The ring network structure analyzes the dynamic evolution of physiological parameters such as blood pressure and heart rate; the three are spliced ​​at the feature level to form a high-dimensional joint representation vector, breaking through the limitations of traditional single-modal analysis, so that the disease matching process simultaneously covers the evidence chain of semantic, spatial and temporal dimensions. Furthermore, by extending the hybrid similarity calculation (edit distance + word vector cosine similarity) of the word correction model and the hierarchical mapping mechanism of the disease expansion lexicon, the patient's colloquial expression (such as "head swelling") is corrected to standard medical terminology (such as "headache"), eliminating the semantic gap caused by expression differences and achieving accurate alignment between unstructured input and standardized knowledge base;

[0264] Secondly, based on a three-level cascaded correction mechanism (BMI-target layer, age-target layer, gender-target layer) of the target parameter-profile association mapping space, the system dynamically adjusts the probability weights of the initial matching targets. For example, in the BMI-target layer, through regional statistical modeling (e.g., the weighting coefficient for increased migraine incidence in obese individuals in Shanghai), combined with age-level risk attenuation factors (e.g., a reduction in the baseline probability of meningitis in 35-year-old patients) and gender-specific weighting terms (e.g., the migraine risk multiplier for women), a dynamic quantitative model of the association between multidimensional user profiles and diseases is formed. This mechanism enables the discrimination probability of the initial matching target sequence to reflect individualized risk characteristics, rather than relying on a static probability distribution. Based on this, the step-by-step guidance strategy uses a Markov decision process to generate dynamic question frequency allocation: based on the probability ratio of each target in the current round (e.g., migraine 79.8% accounts for 60.9% of the total probability of 1.311), question resources are adaptively allocated, prioritizing targeted inquiries for key discriminant symptoms of high-probability pathologies (e.g., visual aura of migraine). Meanwhile, the text randomization model, combined with standard knowledge base constraints, ensures the medical standardization and logical coherence of the generated questions (e.g., "Is the pain unilateral?" conforms to the ICD-11 diagnostic criteria), avoiding irrelevant information interference caused by open-ended questions. This probability-driven question generation strategy significantly shortens the path to collecting effective evidence and reduces the number of redundant interactions.

[0265] Third, a dynamic decision-making closed loop is constructed through confidence threshold comparison and multi-round reinforcement iterations. After each round of question-and-answer, the decision discrimination model uses an attention-weighted multi-source evidence fusion algorithm to calculate the updated confidence level by combining text matching degree, image evidence strength, and temporal pattern consistency (such as the progressive aggravation feature of migraine). For supportive answers (such as confirming visual aura), a linear gain strategy is used to increase the confidence level of the corresponding pathology; for negative answers (such as denying a sudden increase in blood pressure), a probability decay penalty is applied. After the confidence sequence is filtered by the threshold, the system automatically removes low-probability targets (such as removing targets with an intracranial pressure increase probability that drops to 5%), narrowing the range of matching targets for the next round, forming a progressive decision focus. In addition, the design of the iteration termination condition (the highest probability target is stable and all targets exceed the threshold) prevents the risk of misjudgment caused by premature convergence and avoids the waste of resources caused by infinite loops. This mechanism based on evidence accumulation and dynamic shrinkage of the parameter space enables the system to maintain stable decision directionality when facing ambiguous symptoms or multiple pathological competition.

[0266] Fourth, the historical information database and the standard knowledge base are connected bidirectionally via a knowledge graph and a graph database (Neo4j). The symptom-pathology mapping relationship constructed by the extended phrase correction model serves as a semantic bridge, supporting traceability queries from non-standard terms to ICD codes (e.g., mapping "angina pectoris" to "angina pectoris" and associating it with the ICD-10 code I20 for coronary heart disease). During the diagnosis process, the system writes the corrected symptom parameters, user profiles, and question-and-answer records to the graph database nodes in real time, establishing a symptom-pathology-profile association network through an attribute graph model. This structured data storage method not only provides a traceable chain of evidence for the generation of decision parameter sets (e.g., demonstrating the derivation path of "visual aura → migraine"), but also optimizes the statistical model of the historical case database by continuously accumulating real-time consultation data, forming a data-driven self-evolutionary capability. For example, when a region adds new epidemiological characteristics, the BMI index-target mapping layer can automatically update the correlation coefficient, ensuring the timeliness of the profile correction factor.

[0267] Fifth, the preset query round threshold and confidence anomaly detection mechanism provide dual protection: when the confidence threshold cannot be reached after multiple iterations, the system proactively terminates the process and suggests referral, avoiding delays caused by algorithm limitations. Simultaneously, the design of localizing image recognition and embedding models on high-performance terminals balances response speed and energy consumption through dynamic allocation of computing resources (e.g., loading the local knowledge base only on high-performance devices), ensuring the smoothness of mobile applications. This layered processing strategy maintains the accuracy of the core algorithm while expanding the system's applicability in different hardware environments.

[0268] Example 2

[0269] Please see Figure 2Another embodiment of the present invention provides: a guided auxiliary evaluation information processing system, comprising: an initial matching module, a step-by-step question-and-answer module, and a discrimination module;

[0270] The discrimination module includes a confidence unit, a secondary discrimination unit, and an inquiry adjustment unit;

[0271] The initial matching module is used to obtain initial question and answer information, and to perform preliminary matching based on the initial question and answer information and a preset standard knowledge base to obtain an initial matching target sequence decision parameter set;

[0272] The step-by-step question-answering module is used to obtain a user profile and a corresponding step-by-step enhanced decision parameter set for the matching target sequence based on the initial matching target sequence decision parameter set combined with a preset step-by-step guided question-answering assistance strategy and synchronously calling the standard knowledge base;

[0273] The discrimination module is used to perform re-examination discrimination and adjust the diagnosis results based on the user profile and the corresponding step-by-step enhanced decision parameter set combined with the preset decision discrimination model;

[0274] The confidence unit obtains the confidence sequence corresponding to the matching target sequence based on the user profile and the step-by-step enhanced decision parameter set of the corresponding matching target sequence, combined with a preset decision discrimination model.

[0275] The inquiry adjustment unit is used to count the number of pre-diagnosis inquiry rounds and, in the case of no results from follow-up visits, adjust the follow-up visit output results based on the count of pre-diagnosis inquiry rounds and the inquiry round threshold.

[0276] The secondary discrimination unit is used to compare the confidence sequence corresponding to the matching target sequence with a preset confidence threshold. If there is at least one matching target whose confidence is greater than the preset confidence threshold, the parameter strategy space corresponding to the matching target with the highest confidence is output.

[0277] If the confidence scores of all matching targets are less than the preset confidence threshold, the query adjustment unit determines whether the number of queries in the current step-by-step guided question-and-answer assistance process is less than the preset query round threshold. If it is less, the step-by-step guided question-and-answer assistance and decision-making process is repeated. If, until the number of queries is greater than or equal to the preset query round threshold, all the confidence scores of the corresponding matching target sequence are still less than or equal to the preset confidence threshold, then a question-and-answer anomaly is reported.

[0278] Example 3

[0279] Another embodiment of the present invention provides: a guided auxiliary assessment information processing platform, which is used to implement a guided auxiliary assessment information processing method, including: a login interface, an output interface, an initial matching module, a step-by-step question and answer module, and a discrimination module; the discrimination module includes a confidence unit, a secondary discrimination unit, and a query adjustment unit;

[0280] First, enter the login interface and check if you are logged in. If you are not logged in, proceed with the login process. If you are logged in, load the historical information database, standard knowledge base, and extended phrase correction model through the initial matching module, and generate a greeting.

[0281] Second, during the login process, confirm whether to bind patient information. If binding is confirmed, the patient is bound. If not, proceed with the binding process. Once binding is complete, a patient information card is generated. Based on the patient information card, confirm the selection of the patient.

[0282] Third, based on the confirmed patient selection, the step-by-step question-and-answer module, combined with the loaded historical information database, standard knowledge base, and extended phrase correction model, conducts multiple rounds of guided pre-diagnosis inquiries to obtain user profiles and corresponding step-by-step enhanced decision parameter sets.

[0283] Furthermore, a specific exemplary process for multi-round guided pre-diagnosis questioning in this embodiment includes:

[0284] The patient begins to ask questions, such as "I have a cough." Based on the questions asked, the initial matching target sequence decision parameter set is determined by combining the historical information database, the standard knowledge base, and the extended phrase correction model.

[0285] The initial matching target sequence decision parameter set includes the patient's height, weight, age, mental state, body temperature, duration and duration of cough, presence of phlegm, and pain in other parts of the body.

[0286] Based on the confirmed initial matching target sequence decision parameter set, the parameters are completed through a step-by-step question answering module, and a missing parameter check is performed on the parameters completed in each step. After M rounds of completion and passing the missing parameter check, the user profile and the corresponding step-by-step enhanced decision parameter set are obtained.

[0287] Fourth, based on user profiles and corresponding step-by-step enhanced decision parameter sets, the confidence level of the matching target is obtained through confidence units;

[0288] Fifth, based on the confidence level of the matching target, a secondary discrimination unit determines whether there is a matching target with a confidence level greater than the preset threshold. If so, the output interface outputs the diagnostic parameters and corresponding diagnostic strategies corresponding to the target with the highest confidence level and scores the corresponding diagnostic process. If not, the query adjustment unit determines whether the current step-by-step guided pre-diagnosis query rounds are less than the preset query rounds threshold. If they are less, the above distributed guided pre-diagnosis and decision discrimination process is repeated. If the confidence level of the corresponding matching target is still less than or equal to the preset threshold until the number of query rounds is greater than or equal to the preset query threshold, the system replies that a diagnosis cannot be made and recommends the optimal medical information, while simultaneously scoring the corresponding diagnostic process.

[0289] Furthermore, to more clearly illustrate the multi-round inquiry process of the aforementioned guided assisted assessment information processing platform, the following specific example is provided in this embodiment:

[0290] Agent Medical Assistant: Hello! To better assist you, please first tell me some basic information about yourself, such as: gender, age, height, weight, etc.

[0291] Patient: I am a 45-year-old male, 175cm tall and weigh 80kg.

[0292] Agent Medical Assistant: Thank you for your information! Next, could you tell me what your main symptoms are? For example: fever, cough, headache, etc.

[0293] Patient: I've been having headaches lately, and they seem to be getting worse, especially at night.

[0294] Medical Assistant: Understood. Please describe your headache symptoms further: How long has it lasted? Are there any other accompanying symptoms, such as nausea, vomiting, or blurred vision?

[0295] Patient: It's been going on for about three days. The headache is intermittent, I feel a little nauseous occasionally, but I haven't vomited. My eyes are also a little uncomfortable.

[0296] Medical Agent: Okay, thank you for your detailed description. Next, please tell me if there is any supporting diagnostic information, such as: any past medical history, or any tests that have been done before? For example, blood pressure, blood sugar, etc.

[0297] Patient: I have a history of high blood pressure, and I've been taking medication to control it for almost five years. I recently had my blood pressure checked, and the doctor said everything was normal.

[0298] Agent Medical Assistant: Thank you for sharing. Based on your symptom description and medical history, I will guide you to provide more information to help us make a better diagnosis. Please tell me if your symptoms include any of the following: fever, chills, fatigue, chest tightness, etc.

[0299] Patient: No fever, no chills, no weakness, and no chest discomfort.

[0300] Agent Medical Assistant: Okay, thank you for your patient reply. Next, I need you to provide more detailed diagnostic information. Please tell me the specific time and frequency of your symptoms: for example, are they most severe in the morning, noon, or evening?

[0301] Patient: The symptoms are mostly more severe at night and milder when I wake up in the morning.

[0302] Medical Agent: Okay, I understand. Next, I need you to provide some additional information, such as whether you have had any key tests done in your medical history, such as body temperature and blood pressure.

[0303] Patient: As I've already mentioned, I have a history of hypertension, but my recent blood pressure tests have all been within the normal range.

[0304] Medical Agent: Thank you. Based on the information we have gathered so far, I will provide a preliminary diagnostic suggestion. Considering your symptoms and medical history, it is likely a headache related to high blood pressure. We recommend that you go to the hospital for further examination, especially a brain CT or MRI scan, to rule out other possibilities.

[0305] Medical Agent: If needed, I can help you schedule an appointment with the nearest specialist or recommend a medical facility. Would you like that?

[0306] Patient: Yes, could you help me schedule an appointment with a neurologist?

[0307] Agent Medical Assistant: Okay, I will schedule an appointment for you with a neurologist. Please wait a moment.

[0308] Medical Agent: Your appointment has been completed. Please confirm your information. Your appointment time is tomorrow at 2 PM, and the location is the Neurology Department of XX Hospital. Is there anything else I can help you with?

[0309] Patient: No, thank you!

[0310] Agent Medical Assistant: You're welcome. Wishing you a speedy recovery! Please feel free to contact me if you need any assistance.

[0311] Example 4

[0312] An electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement a guided auxiliary evaluation information processing method.

[0313] A computer-readable storage medium having computer instructions stored thereon, which, when executed, perform a guided auxiliary evaluation information processing method.

[0314] 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, under the guidance of the present invention, can make changes, modifications, substitutions and variations to the above embodiments without departing from the spirit and scope of the claims. All of these variations are within the protection scope of the present invention.

[0315] 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 guided auxiliary assessment information processing method, characterized in that, include: Based on the initial question-and-answer information and a pre-set standard knowledge base, a preliminary matching is performed to obtain the initial matching target sequence decision parameter set; Based on the initial matching target sequence decision parameter set, combined with the preset step-by-step guided question-and-answer assistance strategy and synchronously calling the standard knowledge base, a user profile and the corresponding step-by-step enhanced decision parameter set of the matching target sequence are obtained. The step-by-step guided question-and-answer assistance strategy constructs the question-and-answer frequency of the matching target sequence in the next round by step-by-step enhancing the initial discrimination probability sequence corresponding to each matching target in the decision parameter set after each round of question-and-answer update. Based on the user profile and the step-by-step enhanced decision parameter set of the corresponding matching target sequence, combined with the preset decision discrimination model, the confidence sequence corresponding to the matching target sequence is obtained. The confidence sequence corresponding to the matching target sequence is compared with the preset confidence threshold. If there is at least one matching target whose confidence is greater than the preset confidence threshold, the parameter strategy space corresponding to the matching target with the highest confidence is output. If the confidence of all matching targets is less than the preset confidence threshold, then determine whether the number of rounds of the current step-by-step guided question-and-answer assistance process is less than the preset question round threshold. If it is less, then repeat the step-by-step guided question-and-answer assistance and decision-making process. If, until the number of question rounds is greater than or equal to the preset question round threshold, all the confidence of the corresponding matching target sequence is still less than or equal to the preset confidence threshold, then reply with a question-and-answer error. The process of obtaining the user profile and the corresponding step-by-step enhanced decision parameter set includes: Based on the initial matching target sequence decision parameter set, combined with the preset target parameter-profile association mapping space and the first supplementary parameter information, a user profile and an initial enhanced decision parameter space are obtained. The first supplementary parameter information includes the basic information and location information of the user corresponding to the matching target; The user profile includes the same parameters as the first supplementary parameter information; The target parameter-profile association mapping space is obtained by combining the user profile with the standard knowledge base and historical information database, and pre-training the association matching algorithm. The initial enhanced decision parameter space includes user profiles, a descending sequence of initial matching targets, and corresponding initial discrimination probability sequences and matching target text parameters; Based on the numerical ratio of the initial discrimination probabilities corresponding to all matching targets in the initial augmented decision parameter space, the frequency of second supplementary information queries corresponding to each matching target in the initial augmented decision parameter space is obtained.

2. The guided auxiliary evaluation information processing method as described in claim 1, characterized in that, The process of obtaining the initial matching target sequence decision parameter set includes: Based on a pre-set multimodal preloading model combined with a pre-set historical information database and the user's initial input text information, an initial target ternary information sequence is obtained; The initial target ternary information sequence includes the target name, the matching target text parameters, and the initial correlation matching degree between the target name and the matching target text parameters; Based on the initial target ternary information sequence, combined with the preset standard knowledge base and the preset extended phrase correction model, the initial matching target sequence decision parameter set is obtained in descending order of the initial association matching degree.

3. The guided auxiliary evaluation information processing method as described in claim 2, characterized in that, The initial matching target sequence decision parameter set includes the initial matching target and the corresponding standard target parameter text feature set, the correction accuracy of the target keyword group corresponding to each matching target and the initial association matching degree sequence corresponding to the initial matching target sequence; The extended phrase correction model is obtained by training a matching confidence algorithm combined with a preset target extended phrase library. It is used to standardize and correct non-standard target keyword phrases in the initial target ternary information sequence and obtain the correction accuracy of the corresponding target keyword phrases. The target extended phrase library is obtained by combining standard target names, non-standard target keyword phrases containing at least one key character from the standard target name, and associated aliases.

4. The guided auxiliary evaluation information processing method as described in claim 3, characterized in that, The process of obtaining the user profile and the corresponding step-by-step enhanced decision parameter set also includes: Based on the initial enhanced decision parameter space and the standard knowledge base, a second cyclic question and answer text set is generated by combining the frequency of second supplementary information queries corresponding to each matching target with a preset text random generation model. The standard knowledge base and the preset extended phrase correction model are called simultaneously to correct the generated second cyclic question and answer text set. The initial augmented decision parameter space is updated based on the corrected second-cycle question-and-answer text set to obtain the second initial discrimination probability corresponding to each matching target after the update.

5. The guided auxiliary evaluation information processing method as described in claim 4, characterized in that, The process of obtaining the user profile and the corresponding step-by-step enhanced decision parameter set also includes: Based on the updated second initial discrimination probability corresponding to each matching target and the initial association mapping threshold, the initial matching target sequence is filtered to obtain the filtered and updated second matching target sequence and the corresponding enhanced decision parameter space. Repeat the above process of obtaining the second matching target sequence and the corresponding enhanced decision parameter space until all initial discrimination probabilities in the updated matching target sequence are greater than the initial association mapping threshold and the matching target corresponding to the maximum initial discrimination probability remains unchanged, and obtain the target sequence and the corresponding enhanced decision parameter space after step-by-step enhancement.

6. The guided auxiliary evaluation information processing method as described in claim 5, characterized in that, The target parameter-profile association mapping space includes a three-level mapping; the three-level mapping includes a BMI index-target mapping layer, an age-target mapping layer, and a gender-target mapping layer; The process of constructing the BMI indicator-target mapping layer includes: Given a fixed BMI index, the correlation analysis algorithm is used to combine the frequency of occurrence and target parameter information of each level of BMI index in different regions with the corresponding target within a unit of time to obtain the correlation degree between each level of BMI index in each region and different targets. The BMI index-target mapping layer is constructed based on the BMI index at each level in each region, the target name, and the correlation between the BMI index at each level and different targets.

7. The guided auxiliary evaluation information processing method as described in claim 6, characterized in that, The construction process of the age-target mapping layer includes: Based on the correlation between the BMI index at each level in each region and different targets as a weighted weight, combined with the age level information in each region, the correlation between each age level in each region and different targets is obtained through correlation analysis. The age levels are defined as a series of 5-year intervals. The age-target mapping layer is constructed based on the age level, target name, and correlation between different targets corresponding to each age level in each region; The construction process of the gender-target mapping layer includes: Based on the correlation between each age group and different targets in each region as a weighted weight, combined with the gender ratio information under each age group in each region, the correlation between each gender and different targets in each region is obtained, and the gender-target layer is obtained. Based on the BMI indicator-target mapping layer, age-target mapping layer, and gender-target mapping layer, the target parameter-profile association mapping space is obtained through a cascading method.

8. The guided auxiliary evaluation information processing method as described in claim 7, characterized in that, The preset historical information database is constructed from user history and real-time question and answer results information through knowledge graphs and graph databases; The historical information database and the standard knowledge base are connected through the extended phrase correction model and the index information constructed based on the correlation between question-and-answer information and standard target parameter information.

9. A guided auxiliary assessment information processing system, used to implement the guided auxiliary assessment information processing method according to any one of claims 1-8, characterized in that, include: The system comprises an initial matching module, a step-by-step question-answering module, and a discrimination module; the discrimination module includes a confidence unit, a secondary discrimination unit, and a query adjustment unit. The initial matching module is used to obtain initial question and answer information, and to perform preliminary matching based on the initial question and answer information and a preset standard knowledge base to obtain an initial matching target sequence decision parameter set; The step-by-step question-answering module is used to obtain a user profile and a corresponding step-by-step enhanced decision parameter set for the matching target sequence by combining the initial matching target sequence decision parameter set with a preset step-by-step guided question-answering assistance strategy and synchronously calling the standard knowledge base. The confidence unit obtains the confidence sequence corresponding to the matching target sequence based on the user profile and the step-by-step enhanced decision parameter set of the corresponding matching target sequence, combined with a preset decision discrimination model. The secondary discrimination unit is used to compare the confidence sequence corresponding to the matching target sequence with a preset confidence threshold. If there is at least one matching target whose confidence is greater than the preset confidence threshold, the parameter strategy space corresponding to the matching target with the highest confidence is output. If the confidence scores of all matching targets are less than the preset confidence threshold, the query adjustment unit determines whether the number of queries in the current step-by-step guided question-and-answer assistance process is less than the preset query round threshold. If it is less, the step-by-step guided question-and-answer assistance and decision-making process is repeated. If, until the number of queries is greater than or equal to the preset query round threshold, all the confidence scores of the corresponding matching target sequence are still less than or equal to the preset confidence threshold, then a question-and-answer anomaly is reported.

Images