CN114091676A - Causal relationship determination method - Google Patents
Causal relationship determination method Download PDFInfo
- Publication number
- CN114091676A CN114091676A CN202111336126.2A CN202111336126A CN114091676A CN 114091676 A CN114091676 A CN 114091676A CN 202111336126 A CN202111336126 A CN 202111336126A CN 114091676 A CN114091676 A CN 114091676A
- Authority
- CN
- China
- Prior art keywords
- library
- factor
- causal
- cause
- bayesian network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N7/00—Computing arrangements based on specific mathematical models
- G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
- G06N5/022—Knowledge engineering; Knowledge acquisition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Computational Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Algebra (AREA)
- Probability & Statistics with Applications (AREA)
- Computational Linguistics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a causal relationship determination method, which at least comprises the following steps: constructing a Bayesian network by using the construction module; establishing a historical original document library by using a cause and effect library; feeding back the results demonstrated by the experts to the cause and effect library; and performing deep learning on the result demonstrated by the expert to correct the cause and effect library.
Description
The invention relates to a division application with the application number of 201910034540.4, the application date of 2019, 01, 15 and the application type of invention, and the application name of the division application is a causal relationship determination method based on a Bayesian network.
Technical Field
The invention belongs to the technical field of cause tracing and relates to a causal relationship judgment method.
Background
The reason tracing refers to that in the case of an unexpected event or an unexpected effect, reasoning is carried out step by step to obtain the key reason and the root reason hidden in the event or the effect, and the complex causal relationship is revealed.
For example, chinese patent publication No. CN109063253A discloses a bayesian network-based reliability modeling method for an aviation power supply system, which includes establishing an input-output relationship description table for each element of the aviation power supply system; constructing corresponding Bayesian network nodes for each row of data of the N sub-tables; determining a father node of each node of the established Bayesian network; constructing a homonymous father node for the homonymous Bayesian network node of the element C in the step 2, and determining the conditional probability distribution of each node on the basis of determining the state of each node of the established Bayesian network; determining a target node for the power supply bus bar, and calculating the power supply reliability of the aviation power supply system; and determining corresponding target nodes for each power supply bus bar, and calculating the power supply reliability of the aviation power supply system. The invention improves the reliability calculation efficiency of the aviation power supply system. The method only involves a certain event occurring in one object.
For example, a reason tracing method disclosed in chinese patent publication No. CN105468703A includes the following steps: initializing a causal relationship knowledge base, wherein the causal relationship knowledge base comprises abnormal phenomena of a class of objects, reasons causing the abnormal phenomena and causal relationships between the abnormal phenomena and the reasons; selecting the abnormal phenomena in the current known state from the abnormal phenomenon list, forming a new causal relationship knowledge base according to the causal relationship in the causal relationship knowledge base, and recording the traced reasons; the traced reason is output as result information.
In the prior art, the reason tracing aims at the case that an abnormal or unexpected effect occurs on a certain entity object, but the reason tracing cannot be performed when at least one specific event occurs on at least two objects; furthermore, the cause tracing method involves only qualitative determination. Therefore, how to determine the reason of at least one specific event occurring in at least two objects needs to be solved.
A prior art patent document CN107563596A proposes a bayesian network-based causal relationship determination system, which checks and determines causal relationships among variables through conditional independence, and the specific implementation method is as follows: selecting one variable Xi in the top layer or the bottom layer, selecting another node variable Xj connected with the node of the Xi variable through an undirected edge EAij in the middle layer, checking the condition independence between the variables Xi and Xj, if another variable Xk exists, and the variable Xi and the variable Xj are independent under the condition of the given variable Xk, deleting the undirected edge EAij between the variable Xi and the variable Xj, and otherwise, keeping the undirected edge EAij; repeating the process until all variables in the top and bottom layers have been tested for conditional independence; for the non-directional edges still reserved and connected with the variables in the top layer or the bottom layer, the direction of the non-directional edges is that the top layer variable points to the middle layer variable or the middle layer variable points to the bottom layer variable; selecting nodes with established directed edges to carry out conditional independence test, judging the direction of the undirected edges between the two nodes according to a causal discovery rule, and repeatedly applying the causal discovery rule until all the existing undirected edges pass the conditional independence test and mark determined or possible directions; the edges between the variables for which the direction still cannot be determined through the above steps remain unchanged.
According to the method, after a certain unknown causal relationship is met for the first time, the causal relationship is determined through expert demonstration, then the result demonstrated by the expert is fed back to the causal library, and the causal library carries out deep learning on the result demonstrated by the expert so as to correct the causal library. When the causal relationship is encountered again, the causal relationship can be judged based on the corrected causal database without expert authentication. The above-mentioned prior art does not provide a technical solution for correcting the causal library, and if an unknown relationship is not determined by the above-mentioned prior art, any event having the unknown relationship cannot be determined by the above-mentioned prior art when the causal relationship is determined by using the above-mentioned prior art.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a Bayesian network-based causal relationship determination system, which comprises: the building module is used for building a Bayesian network; a decision module for generating and outputting a causal relationship spectrum based on the request Bayesian network; the cause and effect library is used for establishing a historical original document library; in the case of at least one non-specific event occurring for at least two objects, the construction module analyzes the object's own factors causing the non-specific event to construct a factor set and constructs an undirected graph structure constraint based on the factor set; and the build module is capable of revising the undirected graph structure constraints based on the causal library to establish the bayesian network; the decision module calculates a causal indicator of the non-specific event caused by the self factor based on the Bayesian network and outputs a causal relationship spectrum of the non-specific event based on the causal indicator, thereby determining a key cause of the non-specific event.
According to a preferred embodiment, said construction module defines at least one domain based on said non-specific events and retrieves at least one exclusive factor related to said non-specific events in said domain based on said causal library; in the case that the exclusive factor does not belong to the self factor, the construction module prompts the third party for evidence finding; and in the case that the third party determines that the exclusive factor is objectively present, the construction module adds the exclusive factor into the undirected graph structure constraint to further modify the undirected graph structure constraint.
According to a preferred embodiment, the construction module is based on the causal library pair factor pairs LmAnd LnThe construction module prompts the third party to correct the factor pair L under the condition that the retrieval of the relationship between the factor pair L and the factor pair L failsmAnd LnPerforming expert demonstration on the relationship between the two, and feeding back the result of the expert demonstration to the cause and effect library, wherein the cause and effect library performs deep learning on the result of the expert demonstration to correct the cause and effect library; in a case where the cause and effect library deeply learns the results demonstrated by the experts, the construction module may perform L-learning on the basis of the cause and effect library according to the factorsmAnd LnNumber pair said factor pair LmAnd LnThe relationship between them is corrected.
According to a preferred embodiment, the construction module establishes a data set D based on the self-factor, divides the self-factor into a plurality of factor sets L according to the data set D as a unit, and forms a factor pair L in a pairwise pairing mannermAnd LnAnd for said data set D and said factor pairs LmAnd LnNumbering; the construction module pairs factor pairs L according to factor pairs numbers based on the causal library pairsmAnd LnThe relation between the two is searched, and the correction factor is carried out on the L according to the search resultmAnd LnAnd for Lm→LnA relation certainty value of (1), Ln→LmIs a relation certainty value and Ln⊥LmAnd assigning the relation certainty value to construct the undirected graph structure constraint.
According to a preferred embodiment, according to bayesian law, the construction module constructs a bayesian network evaluation function based on the data set D and the factor pair set L, where the evaluation function is used to iteratively generate a bayesian network with a highest evaluation index from a plurality of candidate bayesian networks based on the undirected graph structure constraint under the condition that the construction module enables a heuristic search algorithm, and the construction module first constructs an initial first candidate bayesian network based on the undirected graph structure constraint and evaluates the first candidate bayesian network by using the bayesian network evaluation function to obtain a first evaluation index; then, the building module enables the heuristic search algorithm to locally modify the second candidate Bayesian network based on the undirected graph structure constraint, and evaluates the second candidate Bayesian network again by using the Bayesian network evaluation function to obtain a second evaluation index; the building module can obtain at least two candidate Bayesian networks and corresponding evaluation indexes in an iterative loop mode based on the heuristic search algorithm; the construction module outputs a final Bayesian grid which has the optimal evaluation index and serves as a causal relationship spectrum of the output non-specific events under the condition that at least two candidate Bayesian networks and at least two evaluation indexes are obtained.
According to a preferred embodiment, the decision module calculates the factor pairs L based on the final bayesian grid and Pearl principlemAnd LnThereby outputting the causal relationship spectrum, wherein for a causeHormone LmConstraint obtaining and factor L based on the undirected graph structure through a traversal formmThe connected nodes form a node set of the nodes; and successively calculating each node and factor LmThe node with the maximum correlation is selected for independence assumption, and the node with the maximum correlation under a given data set D and the node L are deletedmThe independent nodes are used for improving the judgment efficiency of the judgment module; factor LnAnd factor LmThe independence between them is measured by mutual information:
factor L when the mutual information exceeds a threshold value of the mutual informationnAnd factor LmThe method has correlation and is not very independent; factor L if the mutual information does not exceed the threshold value of mutual informationnAnd factor LmHas no correlation and is independent.
According to a preferred embodiment, the cause and effect library is built up in the following way: the cause and effect library is based on a plurality of acquired related documents containing various historical attributes and is classified to form a plurality of document unit bodies according to the technical field so as to construct the original document library, and a relation confidence value between the attributes is mined through a data mode; the method comprises the following steps that a document layer in a cause and effect library is used for counting the frequency of words/phrases in each document, and the joint occurrence probability of the words/phrases is obtained according to an independence hypothesis; the literature layer calculates the relevance strength of the words/phrases; the literature layer constructs the related reduced coordinates of the literature, and classifies the related literature according to an iterative algorithm form based on the related reduced coordinates of all the related literatures and a classification function constructed by the relevance strength so as to form a plurality of literature unit bodies.
According to a preferred embodiment, in the case that a data layer in the cause and effect library acquires the literature unit cell, the data layer acquires the data set in a mode that two historical attributes are paired; the data layer extracts the relationship between two historical attributes of each relevant document in a syntactic analysis mode of natural language processing so as to establish a relationship knowledge base of the two historical attributes, wherein the relationship between the two historical attributes comprises a forward relationship, a reverse relationship and a vertical relationship; the data layer searches documents containing the two histories in the document unit body based on a relational knowledge table to acquire relational reliability values of the two history attributes in a fusion mode so as to establish a relational reliability value library of the two history attributes, wherein the relation between the two history attributes comprises a forward relational reliability value, a reverse relational reliability value and a vertical relational reliability value; thus, the data layer constructs the historical data set based on the relational knowledge base and the relational certainty value base which are established between all the histories in a pairwise matching manner.
According to a preferred embodiment, the invention further discloses a bayesian network-based causal relationship determination method, which includes: the construction module constructs a Bayesian network; the judgment module generates and outputs a causal relationship spectrum based on the request Bayesian network; establishing a historical original document library by the cause and effect library; in the case of at least one non-specific event occurring for at least two objects, the construction module analyzes the object's own factors causing the non-specific event to construct a factor set and constructs an undirected graph structure constraint based on the factor set; and the build module is capable of revising the undirected graph structure constraints based on the causal library to establish the bayesian network; the decision module calculates a causal indicator of the non-specific event caused by the self factor based on the Bayesian network and outputs a causal relationship spectrum of the non-specific event based on the causal indicator, thereby determining a key cause of the non-specific event.
According to a preferred embodiment, the method further comprises: the construction module defines at least one domain based on the non-specific event and retrieves at least one exclusive factor related to the non-specific event in the domain based on the causal library; in the case that the exclusive factor does not belong to the self factor, the construction module prompts the third party for evidence finding; and in the case that the third party determines that the exclusive factor is objectively present, the construction module adds the exclusive factor into the undirected graph structure constraint to further modify the undirected graph structure constraint.
The method has the advantages that the undirected graph constraint of the Bayesian network among the factors is constructed aiming at least one non-specific event occurring between at least two objects, the undirected graph constraint is corrected based on the literature knowledge base and the expert knowledge base, and then the Bayesian network is constructed based on the undirected graph constraint by adopting a heuristic search algorithm. The structure learning of the Bayesian network is mainly used for revealing qualitative relationships among variables and simultaneously revealing quantitative relationships. However, there are many difficulties in constructing a bayesian network purely from a data perspective. Relying on nonspecific events among objects and historical events thereof can increase the correctness of the Bayesian network construction. And finally, based on the constructed Bayesian network, causal analysis reasoning is realized. In conclusion, the key problems to be solved by the invention are to construct a reasonable Bayesian network and cause-effect analysis reasoning by fusing object knowledge, determine a cause-effect relationship spectrum of non-specific events and determine key reasons.
Drawings
FIG. 1 is a schematic flow chart diagram of a preferred embodiment of a decision method provided by the present invention; and
fig. 2 is a schematic diagram of a preferred module of the decision system provided by the present invention.
List of reference numerals
1: and constructing a module 2: decision module
3: cause and effect library
Detailed Description
This is described in detail below with reference to figures 1 and 2.
In the description of the present invention, the terms "first", "second", "third" and "third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implying any number of technical features indicated. Thus, features defined as "first," "second," "third," and so forth may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality" means two or more unless specifically defined otherwise.
Example 1
The embodiment provides a bayesian network-based causal relationship determination system, which aims at outputting a causal relationship spectrum when at least one non-specific event occurs to at least two objects. The method has the advantages that the undirected graph constraint of the Bayesian network among the factors is constructed aiming at least one non-specific event occurring between at least two objects, the undirected graph constraint is corrected based on the literature knowledge base and the expert knowledge base, and then the Bayesian network is constructed based on the undirected graph constraint by adopting a heuristic search algorithm. The structure learning of the Bayesian network is mainly used for revealing qualitative relationships among variables and simultaneously revealing quantitative relationships. However, there are many difficulties in constructing a bayesian network purely from a data perspective. Relying on nonspecific events among objects and historical events thereof can increase the correctness of the Bayesian network construction. And finally, based on the constructed Bayesian network, causal analysis reasoning is realized. In conclusion, the key problems to be solved by the invention are to construct a reasonable Bayesian network and cause-effect analysis reasoning by fusing object knowledge, determine a cause-effect relationship spectrum of non-specific events and determine key reasons.
Specifically, the system comprises a construction module 1, a determination module 2 and a cause and effect library 3. The building module 1 is used for building a bayesian network. The decision module 2 is used for generating and outputting a causal relationship spectrum based on the request Bayesian network. The cause and effect library 3 is used to build a historical original document library. In the case of at least one non-specific event for at least two objects, the construction module 1 analyzes the own factors of the object causing the non-specific event to construct a factor set and constructs an undirected graph structure constraint based on the factor set. And the construction module 1 can modify the undirected graph structure constraints based on the causal library 3 to build a bayesian network. The judging module 2 calculates a causal index of a non-specific event caused by a self factor based on the Bayesian network and outputs a causal relationship spectrum of the non-specific event based on the causal index, thereby determining a key cause of the non-specific event. In the present invention, the unspecific event may be an actually occurring event, such as a collision between two cars, or the like. Non-specific events may also be events envisioned by scientific researchers, such as a spacecraft docking failure, etc.
Preferably, the construction module 1 defines at least one domain based on the non-specific event and retrieves at least one other factor related to the non-specific event in the domain based on the cause and effect library 3. In the case that the exclusive factor does not belong to the own factor, the construction module 1 prompts a third party to find out evidence; in the case that the third party determines that the exclusion factor is objectively present, the construction module 1 adds the exclusion factor to the undirected graph structure constraint to further correct the undirected graph structure constraint.
Preferably, the construction module 1 is based on the cause and effect library 3 for the factor pairs LmAnd LnUnder the condition that the retrieval of the relation between the factor pair L and the factor pair L fails, the construction module 1 prompts a third party to carry out factor pair LmAnd LnThe relationship between the two is subjected to expert demonstration, the result of the expert demonstration is fed back to the cause and effect library 3, and the cause and effect library 3 carries out deep learning on the result of the expert demonstration to correct the cause and effect library 3. In the case where the cause and effect library 3 deeply learns the results demonstrated by the experts, the construction module 1 factor-by-factor the L based on the cause and effect library 3mAnd LnNumber pair factor pair LmAnd LnThe relationship between them is corrected.
Preferably, the building module 1 builds a data set D based on the self-factor, divides the self-factor into a plurality of factor sets L according to the data set D as a unit, and forms factor pairs L in a pairwise matching mannermAnd LnAnd for the data set D and the factor pair LmAnd LnNumbering is performed. The construction module 1 is based on the cause and effect library 3, and the factor pairs L are numbered according to the factor pairsmAnd LnThe relation between the two is searched, and the correction factor is carried out on the L according to the search resultmAnd LnAnd for Lm→LnA relation certainty value of (1), Ln→LmIs a relation certainty value and Ln⊥LmAnd assigning the relation certainty value to construct an undirected graph structure constraint.
Preferably, according to bayesian rules, the building module 1 builds a bayesian network evaluation function based on the data set D and the factor pair set L, where the evaluation function is used to iteratively generate a bayesian network with the highest evaluation index from a plurality of candidate bayesian networks based on undirected graph structure constraints under the condition that the building module 1 enables a heuristic search algorithm. Preferably, the building module 1 firstly builds an initial first candidate bayesian network based on the undirected graph structure constraint and evaluates the first candidate bayesian network by adopting a bayesian network evaluation function to obtain a first evaluation index; then, the building module 1 starts a heuristic search algorithm to locally modify the second candidate Bayesian network based on the undirected graph structure constraint, and evaluates the second candidate Bayesian network again by using a Bayesian network evaluation function to obtain a second evaluation index; the building module 1 can obtain at least two candidate bayesian networks and corresponding evaluation indexes in an iterative loop mode based on a heuristic search algorithm. The construction module 1 outputs a final bayesian grid which has an optimal evaluation index and is used as a causal relationship spectrum for outputting non-specific events under the condition that at least two candidate bayesian networks and at least two evaluation indexes are obtained.
Preferably, the decision module 2 calculates the factor pairs L based on the final bayesian grid and the Pearl principlemAnd LnAnd outputting a causal relationship spectrum. Wherein for the factor LmConstraint acquisition and factor L based on undirected graph structure in a traversal formmThe connected nodes constitute their node set. And successively calculating each node and factor LmThe node with the maximum correlation is selected for independence assumption, and the node with the maximum correlation under a given data set D and the node L are deletedmAnd the independent nodes are used for improving the judgment efficiency of the judgment module 2. Factor LnAnd factor LmThe independence between them is measured by mutual information:
when the mutual information exceeds the threshold value of the mutual information, then the factor LnAnd factor LmThe method has correlation and is not very independent; if the mutual information does not exceed the threshold value of the mutual information, then the factor LnAnd factor LmHas no correlation and is independent.
Preferably, the cause and effect library 3 is built as follows: the cause and effect library 3 is based on a plurality of acquired relevant documents containing various historical attributes and classifies the documents to form a plurality of document unit bodies according to the technical field so as to construct an original document library, and relationship confidence values among the attributes are mined through a data mode; the document layer in the cause and effect library 3 counts the frequency of words/phrases in each document, and obtains the joint occurrence probability of the words/phrases according to the independence hypothesis; the document layer calculates the relevance strength of the words/phrases; and constructing related reduced coordinates of the documents by the document layer, and classifying the related documents according to an iterative algorithm form based on the related reduced coordinates of all the related documents and a classification function constructed by the relevance strength to form a plurality of document unit bodies.
Preferably, in the case that the data layer in the cause and effect library 3 acquires a document unit cell, the data layer acquires a data set in a manner that two historical attributes are paired; the data layer extracts the relation between two historical attributes of each relevant document in a syntactic analysis mode of natural language processing so as to establish a relation knowledge base of the two historical attributes, wherein the relation between the two historical attributes comprises a forward relation, a reverse relation and a vertical relation; the data layer retrieves documents containing two historical attributes in a document unit body based on a relational knowledge table to acquire relational certainty values of the two historical attributes in a fusion mode so as to establish a relational certainty value library of the two historical attributes, wherein the relation between the two historical attributes comprises a forward relational certainty value, a reverse relational certainty value and a vertical relational certainty value; therefore, the data layer constructs a historical data set based on a relation knowledge base and a relation certainty value base which are established between all histories in a pairwise matching mode.
Example 2
The embodiment discloses a causal relationship determination method based on a bayesian network, which is used for determining responsibility in medical disputes, and under the condition of not causing conflict or contradiction, the whole and/or part of contents of the preferred embodiments of other embodiments can be used as a supplement of the embodiment. Preferably, the method may be implemented by the method of the present invention and/or other alternative modules.
The medical alarm refers to the act of forcing and making profit from the hospital in the form of seriously hampering the medical order, enlarging the situation and causing negative effects to the hospital by adopting various ways together with the family members of the patients who are hired in medical disputes. The direct consequence of medical alarm is that a great amount of medical staff in China directly or indirectly run off, and serious adverse effects are generated, and the development of medical career in China is seriously affected. When medical disputes occur, the division of the causes of the medical disputes is difficult to define. Medical disputes refer to disputes that occur in medical enterprises, public institutions or institutions with legal qualifications, such as medical sanitation, preventive care, medical cosmetology, etc., and the current Chinese medical disputes are particularly unmanageable matters. As a result, medical disputes are usually caused by medical mistakes and mistakes. Medical errors are errors made by medical personnel during the course of diagnostic care. The medical mistake refers to the mistake of medical staff in medical activities such as diagnosis and treatment. These mistakes often result in dissatisfaction or harm to the patient, thereby causing medical disputes. In addition to medical disputes caused by medical mistakes and mistakes, sometimes the medical party does not have any negligence or mistake in the medical activities, and disputes are caused only by the unilateral dissatisfaction of the patients. Such disputes may be caused by a lack of basic medical knowledge of the patient, an inability to understand proper medical treatment, natural outcome and inevitable complications of the disease, and medical accidents, or may be caused by an unscrupulous liability of the patient. It is also known as medical infringement dispute, i.e. dispute between the provider and the recipient of medical services as to whether or not medical action and its consequences infringe and liability for infringement. Therefore, in order to provide a comfortable and healthy working environment for the medical staff and to say that the medical staff is fair and fair, the reason for the medical dispute needs to be presented in a transparent manner, so as to achieve transparency and fairness.
Therefore, the embodiment of the invention provides a causal relationship determination method based on a bayesian network, aiming at assisting the judicial department in solving medical disputes. In the practice of medical dispute judicial practice, the evidence for proving the double-hair claims or the factors for the claims are mostly collected from the legal perspective, and the relationships among the evidence or the factors are mostly evaluated from the qualitative perspective, which is one of the reasons for the confusion of doctors and patients and also one of the reasons for perplexing the judgment or judgment of the judicial staff; under the condition that the doctor and the patient are not entangled clearly, the doctor and the patient can ask for questions or even complain about the judgment, and excessive legal resources are occupied. In the times of pursuing the legal value advocating 'fair sense', the contradiction between doctors and patients is solved by 'data speaking', a doctor is also given a light working environment for rescuing and supporting injuries, a convincing method is given to patients or family members of the patients, and a scientific reference file is provided for judicial authorities, namely the invention has important value.
Specifically, the method mainly comprises the following steps:
s1: constructing a module 1: and constructing a Bayesian network. Specifically, the construction module 1 constructs the undirected graph structure constraint based on factors claimed by the doctor and the patient. And in the case of third party intervention, the construction module 1 modifies the undirected graph structure constraints based on the causal library 3 to build a bayesian network.
S2: and a judging module 2: and outputting a causal relationship spectrum based on the Bayesian network. The judgment module 2 calculates causal indexes between the factor pairs based on the bayesian network and outputs a causal relationship spectrum of the medical dispute based on the causal indexes, so that the causal relationship spectrum takes the factors claimed by the doctor and the patient into consideration and considers scientificity, the dispute between the doctor and the patient can be effectively prevented from being upgraded, and data support for adjudication or judgment is provided for a third party.
Preferably, the specific step of step S1 includes:
s11: establishing a data set D (D) according to the factors claimed by the doctor and the patient1，D2……Di) Are a number of sets of attributes. L ═ L (L)1，L2……Ln) Of a certain set of attributesThe concrete factors are combined. For example, the factors claimed by the medical practitioner include the time of the patient's illness, the time of the visit, the degree of illness, and the kind of illness. The dataset may establish a dataset D with a time attribute and a condition attribute (time attribute, condition attribute). And the specific factors combine the specific attributes of the illness time and the hospitalization time corresponding to the L time factors and number the illness time and the hospitalization time.
S12: and constructing an undirected graph structure constraint based on the relationship between the factors claimed by the doctor and the patient. The relationship between each specific factor pair is determined by retrieving the relationship between the factors claimed by the doctor and the patient, and the factor pairs are numbered. Preferably, the relationship between the factor pairs includes a positive relationship, a negative relationship, and a vertical relationship, i.e., the attribute LmInfluence Attribute LnIs marked as Lm→Ln. Attribute LmAnd attribute LnThe relationship that can occur is possibly the inverse relationship, i.e. the attribute LnInfluence Attribute LmIs marked as Ln→Lm. Attribute LmAnd attribute LnThe relationship that appears may be a vertical relationship, i.e., attribute LmAnd attribute LnDo not affect each other Ln⊥Lm. For example, if the patient considers that the drug component causes the aggravation of cerebral palsy, the drug component affects the aggravation of cerebral palsy, and the drug component and the aggravation of cerebral palsy form a positive relationship, and is labeled as drug component → aggravation of cerebral palsy. Of course, it is also considered that the cerebral palsy aggravation has a negative relationship with the pharmaceutical ingredient, and is also referred to as pharmaceutical ingredient → cerebral palsy aggravation.
S13: and correcting the undirected graph structure constraint based on the causal library. The factor pair numbers are searched in the cause and effect library, and L is selected according to the searched factor pair numbersmAnd LnIs corrected for the relationship of (A), and Lm→LnA relation certainty value of (1), Ln→LmIs a relation certainty value and Ln⊥LmAnd assigning the relation certainty value of (2). For example, the patient considers that the drug component causes aggravation of cerebral palsy, but if the search of the cause and effect library finds that the drug component does not cause aggravation of cerebral palsy, the drug component → aggravation of cerebral palsy is corrected to the drug component ≠ aggravation of cerebral palsy. For example, the pharmaceutical ingredient is L1Brain of humanThe paralysis is aggravated to L2Then, the number of the drug component → the cerebral palsy exacerbation is marked as 12.
Preferably, the construction module 1 defines at least one request field based on the factors claimed by the doctor and the patient, and retrieves at least one other factor related to the factors claimed by the doctor and the patient based on the request field. In the event that the at least one factor is not within the factors claimed by the medical and patient parties, the building module 1 prompts the third party for evidence finding. In the case that the third party determines that the at least one factor is objectively present, the construction module 1 adds the at least one factor to the undirected graph structure constraint to further modify the undirected graph structure constraint. For example, when a neonatal death event occurs, the factors claimed by the doctor and the patient include amniotic fluid embolism, hypoxia and multiple fetuses, the building module 1 will search for female production based on the request field defined by the factors, the building module 1 will search for other factors existing in the accident in the corresponding female production field in the cause and effect library 3, such as thin uterine wall which does not appear in the factors claimed by the doctor and the patient, the building module 1 will prompt the third party to find out whether the parturient has thin uterine wall, and if the situation is objectively existed, the building module 1 will add the factor of the thin uterine wall to further correct the situation without the graph structural constraint. That is, in step S13, there is another possibility that at least one factor that is not proved by any one of the two parties is an important factor affecting the result, the building module 1 prompts the third party to find out the at least one factor, and if the at least one factor occurs objectively, the at least one factor needs to be modified by adding an undirected graph structure constraint to increase the reliability and scientificity of the result, and improve the fairness judgment or arbitration of the third party, thereby showing the rigor, fairness and accountability of the third party.
Preferably, the cause and effect library 3 may not have a certain factor pair or a relationship between certain factor pairs and its relationship certainty value. In order to ensure that the factors claimed by both parties can be supported. Namely, the building module 1 is based on the cause and effect library 3 to pair the factors LmAnd LnWhen the search of the relationship between the building blocks fails, the building module 1 providesShowing third party to factor pair LmAnd LnThe relationship between them is subject to expert demonstration. For example, the patient advocates that the size of the head of the neonate is not the cause of death of the neonate. There is no relationship between neonatal head size and neonatal death in cause and effect library 3. The building block 1 will prompt a third party for expert demonstration. And the results demonstrated by the experts are fed back to the cause and effect library 3, and the cause and effect library 3 performs deep learning on the results demonstrated by the experts to modify the cause and effect library 3. In the case where the cause and effect library 3 deeply learns the results demonstrated by the experts, the construction module 1 factor-by-factor the L based on the cause and effect library 3mAnd LnNumber pair factor pair LmAnd LnThe relationship between them is corrected.
Preferably, the construction module 1 constructs a bayesian network evaluation function based on the cause and effect library 3, the data set D and the factor pair set L:
logP(G,D,KL)＝logP(G)+logP(D|G)+logP(KL|G)
preferably, the construction module 1 constructs the request bayesian network based on a bayesian network evaluation function and an undirected graph structure constraint. Wherein G is a Bayesian grid whose values include L ═ L (L)1，L2……Ln) The specific factors of a certain set of attributes are paired into a directed acyclic graph of nodes. Where P (G) is a prior distribution. From the prior knowledge, the maximum value of logP (G) + logP (D | G) is equivalent to logP (G | D). logP (G | D) may be scored according to bayesian information criteria BIC. In the formula (I), the compound is shown in the specification,
wherein, any one side in the structure G is represented as Lm→LnThen, then
KL(Lm→Ln) I.e. a relationship certainty value. The sum in the formula is to the structure GAnd summing the document knowledge credibility of the forward relation corresponding to the directional edge. For a given data set D, any of the factors in D pair LmAnd LnAnd constructing the Bayesian network based on the Bayesian network evaluation function and the undirected graph structure constraint. After the undirected graph structure constraint of the Bayesian network is determined, a heuristic search algorithm, such as K2 algorithm, can be executed to find a network structure with an optimal scoring function. The general steps are as follows: the method comprises the steps of starting searching from an initial model, locally modifying a current model by using a search operator at each step of searching to obtain a series of candidate models, then calculating the score of each candidate model, and comparing the optimal candidate model with the current model. If the score of the optimal candidate model is large, the optimal candidate model is used as the next current model and is searched continuously; otherwise, stopping searching and returning to the current model. And according to the Bayes principle, the candidate model with the largest score is the Bayes network. Preferably, the Bayesian network evaluation function is constructed according to the established Bayesian network and Bayesian rules. The Bayesian network evaluation function can be constructed according to a classical heuristic structure learning algorithm, such as a K2 algorithm, a Max-Min places and Children algorithm, a Markov chain Monte Carlo search and the like
Preferably, the decision module 2 calculates a causal index between each pair of factors based on the requesting bayesian network and the Pearl principle, thereby outputting a causal relationship spectrum. The decision module 2 is based on mining causal indicators between attributes in a pass-through data pattern, so that it is possible to determine whether complications or complications are formed between causal indicator attributes. In case of causal indexes, the decision module 2 calculates causal indexes between attributes based on the Pearl principle and the bayesian network structure. In order to find out whether the event X is the cause of the event Y, Pearl needs to execute the X event through the intervention X, and calculate E (Y | do (X)), that is, if the event Y changes more than the significance level on average in the case of the intervention X, X is considered to be the cause of Y.
When the decision module 2 is based on mining causal indexes between attributes in a pass-through data pattern, the causal indexes are calculated using the back-gate criterion because the number of documents is large, which results in a large bayesian grid. The back door criterion means that the Bayesian gridG is a directed acyclic graph, (L)m，Ln) Is a pair of nodes of G, and the set of nodes Z is (L)m，Ln) Wherein all nodes in Z are not descendants of Z and Z blocks all pointers LmIs connected to LmTo LnThe path of (2). Thus, the causal relationship of factors to Lm and Ln can be inferred through backgate principles.
In order to be able to simplify the undirected graph constraint by independence checking without affecting the causal relationship between pairs of factors, the decision module 2 is designed to perform a deterministic function. For example, the independence test may employ a chi-square independence test. In the present invention, the independence test can also be performed as follows: for the factor LmConstruction-based undirected graph acquisition and L by way of columnarmThe connected nodes constitute their node set. And successively calculating each node and factor LmThe node with the maximum correlation is selected for independence assumption, and the node with the maximum correlation in the given request subset D is deletediLower and LmAn independent node. In the present invention, entropy is used to measure the set L of factor pairsmUncertainty of (2). At a given factor LmIn the case of (2), factor LnThe uncertainty of (c) can be measured in the following way with conditional entropy:
factor LnAnd LmThe degree of correlation between them can be measured by mutual information:
if the mutual information exceeds the threshold value of the mutual information, L is considerednAnd LmHas relevance. If the mutual information does not exceed the threshold value of the mutual information, L is considerednAnd LmThere is no correlation.
Preferably, the cause and effect library 3 is built as follows: the cause and effect library 3 is based on a plurality of acquired relevant documents containing a plurality of historical attributes and classifies the documents to form a plurality of document unit bodies according to the technical field so as to construct an original document library, and a relation certainty value between the historical attributes is mined through a data pattern. The document layer in the cause and effect library 3 counts the frequency of words/phrases in each document, and obtains the joint occurrence probability of the words/phrases according to the independence assumption. The document layer calculates the relevance strength of the words/phrases; and constructing related reduced coordinates of the documents by the document layer, and classifying the related documents according to an iterative algorithm form based on the related reduced coordinates of all the related documents and a classification function constructed by the relevance strength to form a plurality of document unit bodies.
Preferably, the literature layer is based on a plurality of acquired relevant literature having a plurality of historical attributes. The literature layer classifies related literatures to form a plurality of literature unit bodies so as to construct an original literature base. The relevant documents include medical records for medical visits, research reports, conference documents, journal documents, books, academic papers, and patents. In the case of such a large number of documents, they need to be classified in a certain way. The document classification is performed in order to enable effective observation of the correlation between history attributes and reduction of the load on the system. For example, the classification can be made according to digestive tract diseases, cardiovascular diseases, and neurological diseases. Classification can also be made according to academic fields, such as rehabilitation and psychology, among others. However, in a severe form of the large body of literature, its accurate and efficient classification directly affects the differentiation of complications and complications. Preferably, the literature score can employ Bayesian, SVM, and k-NN methods.
Preferably, the classification of the relevant documents is performed as follows: and the document layer counts the frequency of words/phrases in each document and acquires the joint occurrence probability of the words/phrases according to the independence assumption. For example, for a particular document, its joint occurrence probability distribution can be computed according to a naive bayes method.
Preferably, the document layer calculates the strength of association of words/phrases. The relevance of the word/phrase can be reflected by the calculation of the relevance strength, and the relevance is suitable for the classification of the literature. Preferably, in classifying, the definitionN is a set of document samples, V is a set of document types, ViIs a subset of the ith document type. W is a set of words/phrases, WiIs a subset of the ith word/phrase. At ViIn which contains SjSamples, wherein the associated reduced coordinate T of the p-th samplepIs an n-dimensional array:
wherein k isi(i-1, 2,3, … n) wherein the number of occurrences of the ith word,
and normalizing the coefficients.
At ViIs all ViThe average of the reduced coordinates of the middle sample association reflects the strength of association of the word/phrase in the document, namely:
preferably, the literature layer acquires the relevant reduced coordinates of the literature, and classifies the relevant literature according to a classification function constructed based on the relevant reduced coordinates of all the relevant literatures in an iterative algorithm form to form a plurality of literature unit bodies. Preferably, for any document its associated reduced coordinates are:
in the formula, qiIs the number of occurrences of the ith word in the document. In the classification, the documents to be classified and each class of documents ViSupport point (b) of1,b2,…,bn) The distance of (d) is noted as:
and constructing a document classification function according to the relevance strength:
in the formula, gammaiIs related to the strength of association.
Preferably, the iterative algorithm may employ a minimum iterative algorithm, a minimum optimized iterative algorithm, and a desired maximum iterative algorithm. Preferably, the classification function is capable of deep learning based on the sample size of the relevant documents, thereby enhancing the accuracy of the document layer.
Preferably, in the case where the data layer in the cause and effect library 3 acquires a document unit cell, the data layer acquires a history data set in such a manner that two history attributes are paired. The data layer extracts the relation between two historical attributes of each relevant document in a syntactic analysis mode of natural language processing so as to establish a relation knowledge base of the two historical attributes, wherein the relation between the two historical attributes comprises a forward relation, a reverse relation and a vertical relation. And the data layer retrieves the documents containing the two histories in the document unit body based on the relational knowledge table to acquire the relational certainty values of the two historical attributes in a fusion mode so as to establish a relational certainty value library of the two historical attributes, wherein the relation between the two historical attributes comprises a forward relational certainty value, a reverse relational certainty value and a vertical relational certainty value, and therefore the data layer establishes a historical data set based on the relational knowledge base and the relational certainty value library which are established between all the histories in a pairwise matching mode.
Preferably, the data layer is capable of obtaining the principal feature parameters based on the document unit volumes and constructing a historical data set based on the principal feature parameters. So as to reduce the interference of a plurality of characteristic parameters formed by a plurality of related documents on the causal relationship between the historical attributes and improve the utilization value of the original document library. Preferably, in the case where the data layer acquires a document unit cell, the data layer acquires the historical data set in such a way that two historical attributes are paired. The data layer is in natural language for each relevant documentThe syntactic analysis mode of the theory extracts the relationship between two historical attributes to establish a relationship knowledge base of the two historical attributes, wherein the relationship between the two historical attributes comprises a forward relationship, a reverse relationship and a vertical relationship. And the data layer searches the document containing the two historical attributes in the document unit body based on the relational knowledge table to acquire the relational certainty values of the two historical attributes in a fusion mode so as to establish a relational certainty value library of the two historical attributes, wherein the relation between the two historical attributes comprises a forward relational certainty value, a reverse relational certainty value and a vertical relational certainty value. Therefore, the data layer constructs a historical data set based on a relation knowledge base and a relation certainty value base which are established between all historical attributes in a pairwise matching mode. For example, in the related document, the history attribute L is acquired1History attribute L2History attribute L3And a history attribute L4Etc. several historical attributes. According to the relation of the history attributes, the history attribute L can be constructed1And a history attribute L3Relational knowledge base of (1), historical attribute L2And a history attribute L3The relational knowledge base of (2) and so on. And then, in the unit document body, constructing a relational certainty value library based on the relational knowledge library according to contents in different documents. Preferably, the sum of the forward relation certainty value, the backward relation certainty value and the vertical relation certainty value is processed according to normalization. In other words, in the unit document body, all documents are subjected to traversal query, and the forward relation certainty value, the backward relation certainty value and the vertical relation certainty value are weighted according to frequency. And the data layer establishes a historical data set by the relational knowledge base and the relational certainty value base and inputs the historical data set into the judgment module 2 to carry out the next step.
Preferably, for journal literature, L1→L2The forward relationship certainty value of (a) may also be defined as follows:
where C (Xi) is the confidence level of document Xi, the formula is: (xi) × (IFi +1) × (C)Ii +1), Xi denotes the i document, IFi is the normalized influence factor in the journal of the document Xi, and CIi is the normalized reference amount. If there is no L in the literature1And L2In the relation of (1), then KL (L)1→L2)＝0，KL(L2→L1)＝0，KL(L1⊥L2) 1. Other types of documents may be defined in the same manner, for example, medical records may be defined based on the authority of a physician. For a meeting article, the definition can be made according to the authority of the meeting, and the like.
Example 3
The embodiment discloses a causal relationship determination system based on a bayesian network, and under the condition of no conflict or contradiction, the whole and/or part of the contents of the preferred embodiments of other embodiments can be used as a supplement of the embodiment. Preferably, the method may be implemented by the method of the present invention and/or other alternative modules.
The embodiment provides a causal relationship determination method based on a bayesian network, which comprises the following steps: the building block 1 builds a bayesian network. The decision module 2 generates and outputs a causal relationship spectrum based on the requesting bayesian network. The cause and effect library 3 builds a historical original document library. In the case that at least one non-specific event occurs to at least two objects, the construction module 1 analyzes the self factors of the objects causing the non-specific event to construct a factor set and constructs an undirected graph structure constraint based on the factor set; the construction module 1 can correct the undirected graph structure constraint based on the cause and effect library 3 to establish a Bayesian network; the judging module 2 calculates a causal index of a non-specific event caused by a self factor based on the Bayesian network and outputs a causal relationship spectrum of the non-specific event based on the causal index, thereby determining a key cause of the non-specific event.
Preferably, the construction module 1 defines at least one domain based on the non-specific event and retrieves at least one exclusive factor related to the non-specific event in the domain based on the cause and effect library 3. In the case where the exclusive factor does not belong to the own factor, the construction module 1 prompts a third party for evidence finding. In the case that the third party determines that the exclusion factor is objectively present, the construction module 1 adds the exclusion factor to the undirected graph structure constraint to further correct the undirected graph structure constraint.
Preferably, the building block 1 employed in the present invention is a server having a search engine and having an arithmetic function. The judgment module 2 is a data server having an arithmetic function. The cause and effect library 3 searches for a server of the engine and has an arithmetic function and storage. The construction module 1, the judgment module 2 and the cause and effect library 3 are in communication connection with each other in a wired or wireless mode through optical fibers, data lines, Bluetooth, wifi and/or 4G and the like.
It should be noted that the above-mentioned embodiments are exemplary, and that those skilled in the art, having benefit of the present disclosure, may devise various arrangements that are within the scope of the present disclosure and that fall within the scope of the invention. It should be understood by those skilled in the art that the present specification and figures are illustrative only and are not limiting upon the claims. The scope of the invention is defined by the claims and their equivalents.
Claims (10)
1. A causal relationship determination system, comprising at least:
a construction module (1) for constructing a Bayesian network;
a cause and effect library (3) for building a historical original document library,
characterized in that the construction module (1) is configured to feed back the results of the expert demonstration to the causality library (3), the causality library (3) performing a deep learning on the results of the expert demonstration to modify the causality library (3).
2. A causal relationship determination system, comprising at least:
a construction module (1) for constructing a Bayesian network;
a cause and effect library (3) for building a historical original document library,
characterized in that the construction module (1) is configured to define at least one domain based on non-specific events and to retrieve at least one exclusive factor related to a non-specific event in the domain based on the cause and effect library (3);
in the case that the exclusive factor does not belong to the own factor, the construction module (1) prompts a third party to find out evidence; and under the condition that the exclusion factor is objectively determined by a third party, the construction module (1) adds the exclusion factor into the undirected graph structure constraint to further modify the undirected graph structure constraint.
3. The causal relationship decision system according to claim 1 or 2, characterized in that the construction module (1) is based on the causal library (3) numbering pairs of factor pairs L according to factor pairsmAnd LnThe relation between the two is searched, and the correction factor is carried out on the L according to the search resultmAnd LnAnd for Lm→LnA relation certainty value of (1), Ln→LmIs a relation certainty value and Ln⊥LmAnd assigning the relation certainty value to construct an undirected graph structure constraint.
4. A causal relation determination system according to claim 3, wherein the causal library (3) is configured to build an original document library by classifying a plurality of document cells based on the obtained related documents with multiple historical attributes according to the technical field, so as to mine the relationship certainty value between the attributes through the data pattern.
5. A causal relation determination system according to claim 4, wherein the literature layers in the causal library (3) count the frequency of words/phrases in each literature, and obtain the joint occurrence probability of words/phrases according to the independence assumption.
6. A causal relationship determination system as claimed in claim 5, wherein the literature layer calculates the strength of association of words/phrases; and constructing related reduced coordinates of the documents by the document layer, and classifying the related documents according to an iterative algorithm form based on the related reduced coordinates of all the related documents and a classification function constructed by the relevance strength to form a plurality of document unit bodies.
7. A causal relation determination system according to claim 6, wherein the data layer of the causal relation database (3) retrieves documents containing two histories in a document unit based on the relational knowledge table to obtain the relational certainty values of the two history attributes in a fusion manner so as to establish the relational certainty value database of the two history attributes, so that the data layer establishes the historical data set based on the relational knowledge database and the relational certainty value database which are established in a pairwise manner between all histories.
8. The causal relationship determination system of claim 7, wherein at ViIs all ViThe average of the reduced coordinates of the middle sample association reflects the strength of association of the word/phrase in the document, namely:
9. a causal relationship determination method, comprising at least:
constructing a Bayesian network by using the construction module (1);
establishing a historical original document library by using the cause and effect library (3);
feeding back the results demonstrated by the experts to a cause and effect library (3);
the results demonstrated by the expert are subjected to deep learning to modify the cause and effect library (3).
10. A causal relationship determination method, comprising at least:
constructing a Bayesian network;
establishing a historical original document library by using a cause and effect library;
defining at least one domain based on the non-specific event;
retrieving at least one exclusive factor associated with a non-specific event in the domain based on a causality library;
prompting a third party to find evidence under the condition that the exclusive factor does not belong to the own factor;
and in the case that the third party determines that the exclusive factor is objectively present, adding the exclusive factor into the undirected graph structure constraint to further correct the undirected graph structure constraint.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111336126.2A CN114091676A (en) | 2019-01-15 | 2019-01-15 | Causal relationship determination method |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111336126.2A CN114091676A (en) | 2019-01-15 | 2019-01-15 | Causal relationship determination method |
| CN201910034540.4A CN109800881B (en) | 2019-01-15 | 2019-01-15 | Bayesian network-based causal relationship determination method |
Related Parent Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910034540.4A Division CN109800881B (en) | 2019-01-15 | 2019-01-15 | Bayesian network-based causal relationship determination method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114091676A true CN114091676A (en) | 2022-02-25 |
Family
ID=66558923
Family Applications (4)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111336050.3A Pending CN114091675A (en) | 2019-01-15 | 2019-01-15 | Traceability reasoning method and system |
| CN202111336127.7A Pending CN114091677A (en) | 2019-01-15 | 2019-01-15 | Causal relationship determination system |
| CN201910034540.4A Active CN109800881B (en) | 2019-01-15 | 2019-01-15 | Bayesian network-based causal relationship determination method |
| CN202111336126.2A Pending CN114091676A (en) | 2019-01-15 | 2019-01-15 | Causal relationship determination method |
Family Applications Before (3)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111336050.3A Pending CN114091675A (en) | 2019-01-15 | 2019-01-15 | Traceability reasoning method and system |
| CN202111336127.7A Pending CN114091677A (en) | 2019-01-15 | 2019-01-15 | Causal relationship determination system |
| CN201910034540.4A Active CN109800881B (en) | 2019-01-15 | 2019-01-15 | Bayesian network-based causal relationship determination method |
Country Status (1)
| Country | Link |
|---|---|
| CN (4) | CN114091675A (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111597307A (en) * | 2020-05-18 | 2020-08-28 | 山西大学 | Judicial judgment reasoning method based on interpretable causal model |
| CN112102156B (en) * | 2020-09-09 | 2023-06-09 | 重庆大学 | Controllable automobile image synthesis method based on causal flow model |
| CN112185583B (en) * | 2020-10-14 | 2022-05-31 | 天津之以科技有限公司 | Data mining quarantine method based on Bayesian network |
| CN113657599B (en) * | 2021-08-20 | 2024-05-28 | 北京航空航天大学 | Accident cause and effect reasoning method, device, electronic equipment and readable storage medium |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103729553A (en) * | 2013-12-19 | 2014-04-16 | 浙江工商大学 | Classification control method for urban safety complex events on basis of Bayesian network learning |
| CN105608634A (en) * | 2015-11-09 | 2016-05-25 | 国网新疆电力公司 | Bayesian network based electrical network risk early-warning evaluation model |
| CN105468703B (en) * | 2015-11-18 | 2019-05-21 | 章斌 | A kind of reason retroactive method |
| CN108320040B (en) * | 2017-01-17 | 2021-01-26 | 国网重庆市电力公司 | Acquisition terminal fault prediction method and system based on Bayesian network optimization algorithm |
| CN107563596A (en) * | 2017-08-03 | 2018-01-09 | 清华大学 | A kind of evaluation index equilibrium state analysis method based on Bayes's causal network |
-
2019
- 2019-01-15 CN CN202111336050.3A patent/CN114091675A/en active Pending
- 2019-01-15 CN CN202111336127.7A patent/CN114091677A/en active Pending
- 2019-01-15 CN CN201910034540.4A patent/CN109800881B/en active Active
- 2019-01-15 CN CN202111336126.2A patent/CN114091676A/en active Pending
Also Published As
| Publication number | Publication date |
|---|---|
| CN114091675A (en) | 2022-02-25 |
| CN109800881B (en) | 2021-11-23 |
| CN114091677A (en) | 2022-02-25 |
| CN109800881A (en) | 2019-05-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109800881B (en) | Bayesian network-based causal relationship determination method | |
| US8055603B2 (en) | Automatic generation of new rules for processing synthetic events using computer-based learning processes | |
| US20090287503A1 (en) | Analysis of individual and group healthcare data in order to provide real time healthcare recommendations | |
| CN112151130B (en) | Decision support system based on literature retrieval and construction method | |
| US20080294692A1 (en) | Synthetic Events For Real Time Patient Analysis | |
| CN112151170A (en) | Method for calculating a score of a medical advice for use as a medical decision support | |
| US20200265955A1 (en) | Identifying Valid Medical Data for Facilitating Accurate Medical Diagnosis | |
| CN114639479A (en) | Intelligent diagnosis auxiliary system based on medical knowledge map | |
| CN113362959A (en) | Sudden respiratory infectious disease risk prediction model for regional epidemic prevention and control | |
| Jung et al. | Essential properties and explanation effectiveness of explainable artificial intelligence in healthcare: A systematic review | |
| González et al. | Case-based reasoning in intelligent health decision support systems | |
| Diaz-Garelli et al. | DataGauge: a practical process for systematically designing and implementing quality assessments of repurposed clinical data | |
| Anggrawan et al. | Application of KNN Machine Learning and Fuzzy C-Means to Diagnose Diabetes | |
| Ahmad | Mining health data for breast cancer diagnosis using machine learning | |
| Deja et al. | Mining clinical pathways for daily insulin therapy of diabetic children | |
| Kobrinskii et al. | Expert system for the diagnosis of orphan diseases | |
| CN117012374B (en) | Medical follow-up system and method integrating event map and deep reinforcement learning | |
| Zou et al. | An intelligent adverse delivery outcomes prediction model based on the fusion of multiple obstetric clinical data | |
| Thirumahal et al. | Semantic integration of heterogeneous healthcare data based on hybrid root linked health record ontology | |
| Maryum et al. | Hospital management society: A framework based on fuzzy logic and association rule mining towards well-being society | |
| Ghosh et al. | Markov logic networks in health informatics | |
| Yu | Analysis and Prediction of Heart Disease Based on Machine Learning Algorithms | |
| Chen et al. | Research on the Aided Diagnosis Method of Diseases Based on Domain Semantic Knowledge Bases | |
| Abikoye et al. | Adaptive Neuro-Fuzzy Inference System for HIV/AIDS Diagnosis, Clinical Staging and Regimen Prescription. | |
| Andreasik et al. | ControlSem–distributed decision support system based on semantic web technologies for the analysis of the medical procedures |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |