Intelligent identification method for subject data logic conflict based on thinking chain technology

Abstract

The invention relates to the technical field of data analysis and recognition, in particular to an intelligent recognition method for subject data logic conflicts based on a thinking chain technology, which comprises the following steps: converting original data of a subject into a logic proposition sequence with a time sequence mark, the original data comprising numeric data and semantic data, the logic proposition sequence comprises a numeric proposition unit and a semantic proposition unit; inputting the logic proposition sequence into a thinking chain reasoning model, and generating a dynamic logic association network comprising a main reasoning path and an auxiliary reasoning path; and based on the dynamic logical association network, executing bidirectional reasoning verification and rule template matching, and outputting a verification result set comprising conflict type marks and confidence rating. According to the method, the reasoning network adaptability is improved, the implicit logic contradiction recognition capability is remarkably improved, and the method is particularly suitable for conflict tracking in chronic disease evolution paths and multi-stage test processes.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis and recognition, and in particular to a method for intelligently identifying logical conflicts in subject data based on thought chain technology. Background Art

[0002] With the widespread application of numerical data from medical examinations and digital questionnaires in clinical research, chronic disease management, and individual health monitoring, the acquisition and analysis of multimodal data from subjects has become a key topic in the field of intelligent healthcare. Currently, subject data generally consists of two types: high-frequency, structured physiological indicator data (such as heart rate, blood pressure, and sleep data); and unstructured or semi-structured text data (such as subjective symptom questionnaires and medical records). This multimodal data provides a rich source of information for medical analysis, but it exhibits significant differences in semantic structure, temporal expression, and logical consistency.

[0003] Existing medical data quality control methods mostly focus on filling missing fields, detecting outliers, and standardized modeling, with less attention paid to logical consistency verification across data. This is particularly true in scenarios highly sensitive to data temporal and semantic relationships, such as clinical trials and chronic disease management. Inconsistencies such as "questionnaires describing 'persistent dizziness' while device data show stable blood pressure" are common. These cross-modal conflicts are not only easily overlooked by existing algorithms but can also mislead medical conclusions and even impact treatment strategies.

[0004] Traditional logic verification methods usually use rule template matching or ontology reasoning mechanisms, but they have the following technical limitations:

[0005] Unable to handle logical jumps between different modalities, such as the causal inconsistency between "behavioral description" and "physiological fluctuations";

[0006] Lack of dynamic reasoning capabilities for time series data, unable to capture hidden contradictions in the evolution process;

[0007] The lack of modeling for semantic uncertainty and confidence makes it difficult to support multi-level conflict judgment. Summary of the Invention

[0008] The present invention provides an intelligent identification method for logical conflicts in subject data based on thought chain technology. This method combines time, semantics, and modal three-dimensional information and has multi-path reasoning and conflict level assessment capabilities to achieve systematic discovery of potential logical conflicts in subject data and risk warnings.

[0009] A method for intelligently identifying logical conflicts in subject data based on thought chain technology includes the following steps:

[0010] S1, logical proposition sequence generation: converting the subject's original data into a logical proposition sequence with time sequence markers, wherein the original data includes numerical data and semantic data, and the logical proposition sequence includes numerical proposition units and semantic proposition units;

[0011] S2, dynamic reasoning chain construction: inputting the logical proposition sequence into the thinking chain reasoning model to generate a dynamic logical association network including a main reasoning path and an auxiliary reasoning path;

[0012] S3, multi-dimensional conflict verification: based on the dynamic logic association network, perform bidirectional reasoning verification and rule template matching, and output a verification result set including conflict type tags and confidence ratings.

[0013] Optionally, the S1 specifically includes:

[0014] S11, data preprocessing: perform outlier removal and time alignment on the collected numerical data, and perform entity recognition and time element extraction on the questionnaire text semantic data;

[0015] S12, Numerical proposition generation: For the pre-processed numerical data, divide it into basic proposition segments according to the preset time window, calculate the standard deviation and trend slope of the data in each proposition segment as the fluctuation characteristic value, and form a numerical proposition with time weight. ;

[0016] S13, Semantic Proposition Conversion: Analyze semantic data through natural language processing models, identify key medical entities and behavioral descriptions, and convert unstructured sentences into semantic propositions based on standard logic. , where the time parameter t is determined according to the questionnaire submission time or the occurrence time of the text description event.

[0017] Optionally, the S1 also includes time series tag fusion: the numerical proposition and semantic propositions Sorted by time axis, the propositions with missing timestamps are supplemented with time stamps using adjacent event interpolation method to generate a logical proposition sequence with unified temporal coding.

[0018] Optionally, the S2 specifically includes:

[0019] S21, main reasoning path construction: Based on the time axis alignment mechanism, the consecutive propositions in the logical proposition sequence are connected in chronological order to establish a causal reasoning chain;

[0020] S22, generating auxiliary reasoning paths, wherein the auxiliary reasoning paths include at least a first auxiliary path, a second auxiliary path, and a third auxiliary path;

[0021] S23, dynamic network generation: Calculate the interaction weights between the main reasoning path and the auxiliary reasoning path through the influence factor transfer matrix to generate a dynamic logical association network including multi-layer connection relationships.

[0022] Optionally, in S22:

[0023] The first auxiliary path performs cross-modal counterfactual reasoning, hypothetically verifying the association between numerical propositions and semantic propositions;

[0024] The second auxiliary path implements reverse reasoning under rule constraints and reversely detects proposition contradictions according to preset medical logic rules;

[0025] The third auxiliary path constructs a cross-cycle verification channel to conduct correlation analysis between the current research cycle propositions and historical research data.

[0026] Optionally, the weight value of each connection edge in the dynamic logical association network is calculated as:

[0027] ;in, Proposition node and The comprehensive logical weight between represents the time decay function, represents the modal compatibility function, is the time decay factor, is the modal correlation factor.

[0028] Optionally, in the causal reasoning chain, the input weight of each proposition node is determined by the temporal closeness and modal correlation factor of the preceding node.

[0029] Optionally, the S3 specifically includes:

[0030] S31, bidirectional reasoning verification execution:

[0031] Forward reasoning verification: along the main reasoning path of the dynamic logic association network, the rationality score of the causal relationship between propositions is calculated node by node in chronological order. When the fluctuation characteristic value of adjacent proposition nodes deviates from the medical logic rule by more than the first conflict judgment threshold, it is marked as a time axis contradiction;

[0032] Reverse reasoning verification: Starting from the current verification node, traverse the reverse detection channel in the auxiliary reasoning path to reversely infer the logical consistency of the historical proposition. When the confidence difference between the reverse conclusion and the existing proposition exceeds the second conflict judgment threshold, it is marked as a logical chain contradiction;

[0033] S32, dynamic matching of rule templates: calling the medical logic rule template in the rule library, and performing multi-level rule matching between the proposition nodes in the dynamic logic association network and the template preset conditions;

[0034] S33, verification result synthesis:

[0035] Aggregate the primary path verification results, auxiliary path verification results, and rule matching results, and calculate the comprehensive conflict confidence level according to the conflict type weight formula. , the conflict levels are divided according to the confidence interval, and a structured verification result set including conflict type marking, conflict node location coding and confidence rating is generated.

[0036] Optionally, the multi-level rule matching in S32 specifically includes:

[0037] Calculate the similarity matrix between semantic propositions and rule predicates. When the maximum matching degree in the similarity matrix falls below the dynamic confidence threshold, generate a data consistency alarm.

[0038] Joint rule verification is performed on cross-modal associated propositions. When the fluctuation range of the numerical proposition and the probability of occurrence of the medical event described by the semantic proposition do not meet the preset compatibility conditions, it is marked as a modal conflict.

[0039] Optionally, the comprehensive conflict confidence Calculated as:

[0040] ;in, are the scores of forward reasoning verification, reverse reasoning verification and rule matching, is the dynamic weight coefficient.

[0041] Beneficial effects of the present invention:

[0042] The present invention constructs a logical proposition sequence that integrates numerical propositions and semantic propositions, uses fluctuation characteristics such as standard deviation and trend slope to quantify physiological data, and then combines the natural language processing model to extract confidence metrics to quantify semantic triples, thereby achieving a unified temporal expression of unstructured questionnaires and numerical data. By supplementing missing information through time interpolation and modal compatibility functions, the present invention effectively solves the problem of "device timestamp and text expression being out of sync" in medical scenarios, laying a comparable and traceable temporal logic foundation for subsequent logical reasoning.

[0043] The present invention proposes a "one main and three auxiliary" reasoning chain architecture: the main path is used for temporal causal verification, and the three auxiliary paths correspond to cross-modal counterfactual reasoning, rule-constrained reverse reasoning, and cross-cycle historical consistency verification, respectively, to achieve a comprehensive verification mechanism for forward tracing and reverse correction of the logical chain. Furthermore, a dynamic logical association network is constructed through the influencing factor transfer matrix, and a time attenuation factor and a modal compatibility factor are introduced to integrate static rules and dynamic data features, thereby improving the adaptability of the reasoning network and significantly enhancing the ability to identify implicit logical contradictions. It is particularly suitable for conflict tracking in the evolutionary path of chronic diseases and multi-stage experimental processes.

[0044] The present invention designs a dynamic weight adjustment mechanism by integrating the main path verification score, reverse reasoning consistency and rule matching results. The weight coefficient can be dynamically updated according to time decay, path depth and rule version, forming a conflict confidence scoring model with stable structure and strong interpretability. The logical confidence is converted into clinical risk level (such as level I to level III), supporting real-time monitoring of subject data quality and risk warning function of medical decision-making support system, avoiding false positive conflict judgments caused by ambiguous descriptions or data delays, and improving applicability and credibility in real medical environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0046] Figure 1 Schematic diagram of a method flow in an embodiment of the present invention;

[0047] Figure 2 Schematic diagram of the thought chain reasoning model of an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Those skilled in the art may also implement some known technologies in other alternative ways. The accompanying drawings are only for describing the embodiments in more detail and are not intended to limit the present invention in any specific way.

[0049] like Figure 1-Figure 2 As shown, a method for intelligently identifying logical conflicts in subject data based on thought chain technology includes the following steps:

[0050] S1, logical proposition sequence generation: converting the subject's original data into a logical proposition sequence with time sequence markers. The original data includes numerical data and semantic data, and the logical proposition sequence includes numerical proposition units and semantic proposition units.

[0051] S2, dynamic reasoning chain construction: input the logical proposition sequence into the thinking chain reasoning model to generate a dynamic logical association network including the main reasoning path and auxiliary reasoning path;

[0052] S3, multi-dimensional conflict verification: Based on a dynamic logical association network, it performs bidirectional reasoning verification and rule template matching, and outputs a verification result set including conflict type labels and confidence ratings.

[0053] S1 specifically includes:

[0054] S11, data preprocessing: perform outlier removal and timestamp alignment on the collected numerical data of the subjects; perform named entity recognition (NER) and time element extraction on the questionnaire text data to extract the time expressions related to the event.

[0055] S12, numerical proposition generation: The pre-processed physiological data is divided into several proposition segments according to the dynamic time window, and the fluctuation characteristic value of each segment is calculated: ;in, Indicates the type of physiological indicator (heart rate, blood pressure, etc.), For time period The standard deviation of the internal indicator values, which measures the fluctuation range, It is the linear trend slope of the indicator within the time period, that is, the slope of the fitted straight line, which reflects the trend change. is the time interval corresponding to the proposition segment, expressed as Specifically, let the original physiological data time series be: , using dynamic time window The following indicators are calculated for the data in the window:

[0056] S121. Standard deviation (Fluctuation Range): ,in, is the standard deviation within the time period, which represents the intensity of data fluctuation. is the mean of the data in the window, is the number of samples in the window;

[0057] S122. Trend slope k, (changing trend): The slope is estimated using the least squares method:

[0058] ; The slope reflects the time interval The trend change direction and rate of endogenous physiological parameters.

[0059] S123. Output proposition format for each window: .

[0060] S13, semantic proposition conversion: parse the questionnaire text through the natural language processing model, extract medical-related semantic units, and convert unstructured sentences into standard logical proposition expressions. , perform the following steps through the pre-trained natural language processing model:

[0061] S131. Named Entity Recognition (NER) and Relation Extraction:

[0062] ,in, Indicates the subject of the proposition, such as "patient", Indicates predicates, such as "report", "experience", "take" and other behavioral descriptions, For objects, such as "headache", "insomnia", "certain medicine" and other medical entities, is the confidence of the semantic proposition, which represents the credibility score of the proposition extracted from the original text;

[0063] S132. Semantic Confidence calculate: ,in, is the sentence vector representation, W and 𝑏 are the scoring network parameters, and the output value is in the range [0,1].

[0064] S133. Time parameter t extraction: ;

[0065] S134. Standardized semantic proposition expression: .

[0066] S14, Time series tag fusion: All numerical propositions and semantic propositions According to the time parameter Sort and construct a unified sequence of temporal logic propositions: ; For propositions with missing timestamps , use the interpolation method of adjacent marked propositions to estimate the time: ; This interpolation method is adjusted based on medical knowledge (drug taking time, delayed symptom manifestation time) to improve the accuracy of time marking.

[0067] S2, based on the unified temporal logic proposition sequence, constructs a main path and three auxiliary paths to generate a dynamic logical association network, specifically including:

[0068] S21, Main reasoning path construction: Connect logical propositions according to the time axis sequence to form a temporal causal main path. The connection weights between adjacent proposition nodes in the path are determined by the following dual weight mechanism:

[0069] ;in, On the main path and The causal connection weights between propositions, represents the temporal closeness function, defined as the negative exponential decay of the time interval: ; is the modal correlation factor function, which determines the degree of coupling between proposition types (numerical-semantic). are the time series weight coefficient and the modal association weight coefficient, Time decay coefficient.

[0070] S22, auxiliary reasoning path generation: Based on the main path logic, the following three types of supplementary reasoning paths are constructed:

[0071] The first auxiliary path: cross-modal counterfactual reasoning: for proposition pairs ,like For semantic type, For numeric type, construct the following hypothesis verification logic: ; If the verification fails, it is marked as a potential conflict pair.

[0072] The second auxiliary path: reverse reasoning under rule constraints: Assume that the set of medical logic rules is , for the terminal proposition , perform reverse expansion: ; If any generated pre-order proposition is inconsistent with the main path, it is recorded as a structural contradiction.

[0073] The third auxiliary path: cross-cycle verification channel: Assume that the historical research data proposition set is , construct a periodic mapping function: ,in, is the proposition semantics / indicator matching function. If similar propositions form a logical closed-loop conflict in history, it is marked as a long-term latent contradiction.

[0074] S23, dynamic logical association network generation: The main path and three auxiliary paths form a multi-layer heterogeneous graph network, and the connection edge weights are calculated using the following general transfer weight formula: ;in, Proposition node and The comprehensive logical weight between represents the time decay function, represents the modal compatibility function, which is defined as follows:

[0075] ;

[0076] in, It is a modality matching function driven by medical knowledge graph. To control the time and modal factor weights, is the global time decay rate, represents the modal compatibility constant.

[0077] S3 is based on a dynamic logical association network and sequentially performs bidirectional reasoning verification of the main path and auxiliary path, medical rule matching, and conflict level confidence synthesis calculation. Specifically, it includes:

[0078] S31, bidirectional reasoning verification execution:

[0079] S311, forward reasoning verification: along the main reasoning path Perform causal plausibility computation in chronological order: for each pair of adjacent proposition nodes , calculate the rationality score:

[0080] ;

[0081] When the following conditions are met, it is marked as "timeline contradiction": ;in, is the fluctuation characteristic value corresponding to the proposition ( 、 , is a reasonable value predicted according to medical rules, is the first conflict determination threshold, ranging from 0.15 to 0.25. is the causal rationality score of the current node pair, with a value range of [0,1].

[0082] S312, reverse reasoning verification: in the auxiliary path, verify the proposition from the current Looking back on the historical proposition sequence , derive the conclusion of the reverse proposition , and compare the confidence with the current proposition:

[0083] ; When the difference satisfies: , it is marked as "logical chain contradiction", where is the actual proposition confidence, is the second conflict determination threshold, ranging from 0.2 to 0.3. is the reverse consistency score, with a value range of [0,1], is the confidence level of the proposition obtained by reverse reasoning. The reverse reasoning is as follows:

[0084] a Select the target proposition (e.g., an observation or symptom) as a starting point for verification;

[0085] bCalling rule base Perform reverse logic reasoning (e.g., "If X happens, then Y happens first" can be reversed to "If X has already happened, then Y should have also happened"): ;

[0086] c is derived from the , calculate its confidence based on its: semantic matching with existing propositions in the main path, confidence score of the rules used, and support strength of the current context (consistency of adjacent propositions). .

[0087] S32, rule template matching:

[0088] S321, rule predicate matching score matrix: for semantic proposition set , build its predicate and rule template set The semantic similarity matrix of: ;in, For the The predicate of a proposition, Indicates the rule template predicates, is the word vector semantic similarity function (cosine similarity), if , then trigger the data consistency alarm, is a dynamic confidence threshold function, which is determined by the proposition confidence Adaptive adjustment, The bigger, Higher: ; is the relaxation factor of the lower limit of confidence, which is set to 0.07 to avoid being too strict. Semantic Proposition in Item obtained.

[0089] S14, Time series tag fusion: All numerical propositions and.

[0090] S322, Cross-modal joint rule verification: For cross-modal proposition pairs , according to the joint rules Verify compatibility:

[0091] ; If Valid=0, mark as "modal conflict".

[0092] S33, Verification result synthesis and conflict confidence calculation: Integrate the three types of conflict verification results and use the weighted average method to synthesize the comprehensive conflict confidence :

[0093] ;in, are the scores of forward verification, reverse verification and rule matching respectively, is the dynamic weight coefficient, defined as follows:

[0094] ;

[0095] in, is the time difference between the proposition time and the current moment, Indicates the number of reverse reasoning paths available for the current node. is the version number of the current rule template, are basic weight coefficients, taking values of 1.0, 0.8, and 0.6 respectively. is the time decay coefficient.

[0096] Conflict level classification and result set generation: based on comprehensive confidence Map the conflict levels to generate a structured verification result set:

[0097] ;

[0098] Result output format: .

[0099] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0100] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for intelligently identifying logical conflicts in subject data based on thought chain technology, characterized in that: The following steps are involved: S1, logical proposition sequence generation: converting the subject's original data into a logical proposition sequence with time sequence markers, wherein the original data includes numerical data and semantic data, and the logical proposition sequence includes numerical proposition units and semantic proposition units; S2, dynamic reasoning chain construction: inputting the logical proposition sequence into the thinking chain reasoning model to generate a dynamic logical association network including a main reasoning path and an auxiliary reasoning path; S3, multi-dimensional conflict verification: based on the dynamic logic association network, perform bidirectional reasoning verification and rule template matching, and output a verification result set including conflict type tags and confidence ratings.

2. The method for intelligently identifying logical conflicts in subject data based on thought chain technology according to claim 1, characterized in that: Said S1 specifically includes: S11, data preprocessing: perform outlier removal and time alignment on the collected numerical data, and perform entity recognition and time element extraction on the questionnaire text semantic data; S12, Numerical proposition generation: For the pre-processed numerical data, divide it into basic proposition segments according to the preset time window, calculate the standard deviation and trend slope of the data in each proposition segment as the fluctuation characteristic value, and form a numerical proposition with time weight. ; S13, Semantic Proposition Conversion: Analyze semantic data through natural language processing models, identify key medical entities and behavioral descriptions, and convert unstructured sentences into semantic propositions based on standard logic. , where the time parameter t is determined according to the questionnaire submission time or the occurrence time of the text description event.

3. The method for intelligently identifying logical conflicts in subject data based on thought chain technology according to claim 1, characterized in that: The S1 also includes time series tag fusion: the numerical proposition and semantic propositions Sorted by time axis, the propositions with missing timestamps are supplemented with time stamps using adjacent event interpolation method to generate a logical proposition sequence with unified temporal coding.

4. The method for intelligently identifying logical conflicts in subject data based on thought chain technology according to claim 1, characterized in that: The S2 specifically includes: S21, main reasoning path construction: Based on the time axis alignment mechanism, the consecutive propositions in the logical proposition sequence are connected in chronological order to establish a causal reasoning chain; S22, generating auxiliary reasoning paths, wherein the auxiliary reasoning paths include at least a first auxiliary path, a second auxiliary path, and a third auxiliary path; S23, dynamic network generation: Calculate the interaction weights between the main reasoning path and the auxiliary reasoning path through the influence factor transfer matrix to generate a dynamic logical association network including multi-layer connection relationships.

5. The method for intelligently identifying logical conflicts in subject data based on thought chain technology according to claim 4, characterized in that: In said S22: The first auxiliary path performs cross-modal counterfactual reasoning, hypothetically verifying the association between numerical propositions and semantic propositions; The second auxiliary path implements reverse reasoning under rule constraints and reversely detects proposition contradictions according to preset medical logic rules; The third auxiliary path constructs a cross-cycle verification channel to conduct correlation analysis between the current research cycle propositions and historical research data.

6. The method for intelligently identifying logical conflicts in subject data based on thought chain technology according to claim 5, characterized in that: The weight value of each connection edge in the dynamic logical association network is calculated as: ;in, Proposition node and The comprehensive logical weight between represents the time decay function, represents the modal compatibility function, is the time decay factor, is the modal correlation factor.

7. The method for intelligently identifying logical conflicts in subject data based on thought chain technology according to claim 6, characterized in that: In the causal reasoning chain, the input weight of each proposition node is determined by the temporal closeness and modal correlation factor of the preceding node.

8. The method for intelligently identifying logical conflicts in subject data based on thought chain technology according to claim 1, characterized in that: The S3 specifically includes: S31, bidirectional reasoning verification execution: Forward reasoning verification: along the main reasoning path of the dynamic logic association network, the rationality score of the causal relationship between propositions is calculated node by node in chronological order. When the fluctuation characteristic value of adjacent proposition nodes deviates from the medical logic rule by more than the first conflict judgment threshold, it is marked as a time axis contradiction; Reverse reasoning verification: Starting from the current verification node, traverse the reverse detection channel in the auxiliary reasoning path to reversely infer the logical consistency of the historical proposition. When the confidence difference between the reverse conclusion and the existing proposition exceeds the second conflict judgment threshold, it is marked as a logical chain contradiction; S32, dynamic matching of rule templates: calling the medical logic rule template in the rule library, and performing multi-level rule matching between the proposition nodes in the dynamic logic association network and the template preset conditions; S33, verification result synthesis: Aggregate the primary path verification results, auxiliary path verification results, and rule matching results, and calculate the comprehensive conflict confidence level according to the conflict type weight formula. , the conflict levels are divided according to the confidence interval, and a structured verification result set including conflict type marking, conflict node location coding and confidence rating is generated.

9. The method for intelligently identifying logical conflicts in subject data based on thought chain technology according to claim 8, characterized in that: The multi-level rule matching in S32 specifically includes: Calculate the similarity matrix between semantic propositions and rule predicates. When the maximum matching degree in the similarity matrix falls below the dynamic confidence threshold, generate a data consistency alarm. Joint rule verification is performed on cross-modal associated propositions. When the fluctuation range of the numerical proposition and the probability of occurrence of the medical event described by the semantic proposition do not meet the preset compatibility conditions, it is marked as a modal conflict.

10. The method for intelligently identifying logical conflicts in subject data based on thought chain technology according to claim 8, characterized in that: The comprehensive conflict confidence Calculated as: ;in, are the scores of forward reasoning verification, reverse reasoning verification and rule matching, is the dynamic weight coefficient.

Images