A cause-effect relationship determination system

Abstract

The present invention relates to a cause-effect relationship determination system comprising at least: a construction module for constructing a Bayesian network; a cause-effect library for establishing a historical original document library, characterized in that the construction module is configured to, in the case of a search failure for a relationship between a factor pair L m and L n , prompt a third party for expert argumentation on the relationship between the factor pair L m and L n .

Description

[0001] The application is a divisional application of application No. 201910034540.4, with the application date of January 15, 2019, the application type of invention, and the application name of a cause-effect relationship determination method based on a Bayesian network. TECHNICAL FIELD

[0002] The application belongs to the technical field of cause-effect relationship determination, and relates to a cause-effect relationship determination system. BACKGROUND

[0003] Cause-effect relationship determination refers to reasoning step by step to obtain the key reason and the root cause hidden in an unexpected event or an unexpected effect, and revealing the complex cause-effect relationship.

[0004] For example, a Chinese patent with the publication number CN109063253A discloses an aviation power supply system reliability modeling method based on a Bayesian network, which includes establishing an input-output relationship description table for each element of the aviation power supply system; constructing a corresponding Bayesian network node for each row of data of N sub-tables; determining the parent nodes of each node of the constructed Bayesian network; constructing the same-named parent nodes for the same-named Bayesian network nodes of the element C in step 2, determining the conditional probability distribution of each node on the basis of the determined state of each node of the constructed Bayesian network; determining the target node for the power busbar, and calculating the power supply reliability of the aviation power supply system; determining the corresponding target node for each power busbar, and calculating the power supply reliability of the aviation power supply system. The application improves the reliability calculation efficiency of the aviation power supply system. The method only involves a certain event of one object.

[0005] For example, a Chinese patent with the publication number CN105468703A discloses a cause-effect relationship determination method, which includes the following steps: initializing a cause-effect relationship knowledge base, the cause-effect relationship knowledge base including the abnormal phenomenon of a type of object, the reason causing the abnormal phenomenon, and the cause-effect relationship between the abnormal phenomenon and the reason; selecting the abnormal phenomenon of the current known state from the abnormal phenomenon list, forming a new cause-effect relationship knowledge base according to the cause-effect relationship in the cause-effect relationship knowledge base, and recording the traced reason; and outputting the traced reason as the result information.

[0006] In the prior art, the cause-effect relationship determination is for an abnormal effect or an unexpected effect of a certain entity object, but cannot be used for cause-effect relationship determination when at least two objects have at least one specific event. In addition, the cause-effect relationship determination method only involves qualitative determination. Therefore, how to determine the cause of the specific event when at least two objects have at least one specific event, and how to determine the cause of the specific event, need a system or a method to solve.

[0007] The prior art patent document with the publication number CN107563596A proposes a causal relationship determination system based on a Bayesian network, which determines the causal relationship between variables through conditional independence test, and the specific implementation method is as follows: one variable Xi in the top layer or the bottom layer is selected, and another node variable Xj connected with the Xi variable through a non-directed edge EAij in the middle layer is selected, the conditional independence between the variables Xi and Xj is tested, if there is another variable Xk, the variables Xi and Xj are conditionally independent given the variable Xk, then the non-directed edge EAij between the variables Xi and Xj is deleted, otherwise, the non-directed edge EAij is retained; the process is repeated until all variables in the top layer and the bottom layer have passed the conditional independence test; for those non-directed edges still retained and connected with the variables in the top layer or the bottom layer, the direction is that the top layer variable points to the middle layer variable or the middle layer variable points to the bottom layer variable; the nodes that have established directed edges are selected for conditional independence test, the direction of the non-directed edge between the two nodes is determined according to the causal discovery rule, the causal discovery rule is repeatedly applied until all existing non-directed edges pass the conditional independence test and are marked with a certain or possible direction; for the edges between variables that still cannot determine the direction after the above steps, the edges remain unchanged.

[0008] The prior art does not propose any solution to the problem of failure of causal relationship determination, and the present application determines the causal relationship of variables by constructing a module to prompt a third party to conduct expert argumentation on the relationship between factors and L m and L n , thereby eliminating unknown relationships in the relationship network. SUMMARY

[0009] In view of the deficiencies of the prior art, the present application provides a causal relationship determination system based on a Bayesian network, which comprises: a construction module for constructing a Bayesian network; a determination module for generating and outputting a causal relationship spectrum based on the requested Bayesian network; a causal library for establishing a historical original literature library; in the case that at least two objects have at least one non-specific event, the construction module analyzes the self factors of the objects causing the non-specific event to construct a factor set and constructs a non-directed graph structure constraint based on the factor set; and the construction module can correct the non-directed graph structure constraint based on the causal library to establish the Bayesian network; the determination module calculates a causal index of the self factors causing the non-specific event based on the Bayesian network and outputs a causal relationship spectrum of the non-specific event based on the causal index, thereby determining the key reason of the non-specific event.

[0010] According to a preferred embodiment, the constructing 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 constructing module prompts the third party to find evidence; in the case that the third party confirms that the exclusive factor exists objectively, the constructing module supplements the exclusive factor to the undirected graph structure constraint, and further modifies the undirected graph structure constraint.

[0011] According to a preferred embodiment, the constructing module retrieves the relationship between the factor pair L m and L n based on the causal library, and in the case of failure, the constructing module prompts the third party to conduct expert argumentation on the relationship between the factor pair L m and L n , and feeds back the result of the expert argumentation to the causal library, which conducts deep learning on the result of the expert argumentation to modify the causal library; in the case that the causal library conducts deep learning on the result of the expert argumentation, the constructing module modifies the relationship between the factor pair L m and L n based on the causal library according to the factor pair number L m and L n .

[0012] According to a preferred embodiment, the constructing module establishes a data set D based on the self-factor, divides the self-factor into a plurality of factor sets L m and L n according to the data set D as a unit and forms factor pairs L m and L n in a pairwise manner, and numbers the data set D and the factor pairs L m and L n ; the constructing module retrieves the relationship between the factor pairs L m and L n based on the causal library according to the factor pair number, and modifies the relationship between the factor pairs L m → L n , assigns a relationship confidence value to L n → L m , and assigns a relationship confidence value to L n ⊥ L m , to thereby construct the undirected graph structure constraint.

[0013] According to a preferred embodiment, based on Bayes' theorem, the construction module constructs a Bayesian network evaluation function based on the dataset D and the factor pair set L. This evaluation function is used to iteratively generate the Bayesian network with the highest evaluation index from several candidate Bayesian networks, based on the undirected graph structure constraints and when the construction module uses a heuristic search algorithm. Specifically, the construction module first constructs an initial first candidate Bayesian network based on the undirected graph structure constraints and evaluates it using the Bayesian network evaluation function to obtain a first evaluation index. Subsequently, the construction module uses the heuristic search algorithm to locally modify the second candidate Bayesian network based on the undirected graph structure constraints and evaluates it again using the Bayesian network evaluation function to obtain a second evaluation index. The construction module can obtain at least two candidate Bayesian networks and their corresponding evaluation indices through iterative iteration using the heuristic search algorithm. When at least two candidate Bayesian networks and at least two evaluation indices are obtained, the construction module outputs the final Bayesian grid with the optimal evaluation index as the output causal relationship spectrum of non-specific events.

[0014] According to a preferred embodiment, the determination module calculates the effect of each factor on L based on the final Bayesian grid and Pearl's principle. m and L n The causal indicators between factors L are used to output the causal relationship spectrum, wherein for factor L m The factors L are obtained by traversing the undirected graph structure constraints. m The connected nodes form its node set; and the relationship between each node and factor L is calculated sequentially. m The correlation between nodes is analyzed, and the nodes with the highest correlation are selected. Independence assumptions are then made, and nodes with the highest correlation in a given dataset D are removed. m Independent nodes are used to improve the decision-making efficiency of the decision-making module; Factor L n With factor L m The independence between them is measured by mutual information:

[0015]

[0016] When the mutual information exceeds the mutual information threshold, then factor L n With factor L m They are correlated but not very independent; if the mutual information does not exceed the threshold of mutual information, then factor L... n With factor L m They are not related and are independent.

[0017] According to a preferred embodiment, the cause-effect library is established in the following manner: the cause-effect library is based on a large number of relevant documents containing a plurality of historical attributes acquired in the technical field and classified to form a plurality of document units to construct the original document library, so as to mine the relationship confidence value between the attributes by data pattern mining; wherein the document layer in the cause-effect library 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 correlation strength of the words / phrases; the document layer constructs the correlation reduction coordinates of the documents, and classifies the relevant documents in the form of an iterative algorithm based on the correlation reduction coordinates and the classification function constructed based on the correlation strength of all the relevant documents to form a plurality of document units.

[0018] According to a preferred embodiment, in the case where the data layer in the cause-effect library acquires the document units, the data layer acquires the data set in the form of pairing of two historical attributes; the data layer extracts the relationship between two historical attributes in each relevant document in the form of syntactic analysis of natural language processing, so as to establish the relationship knowledge base of the two historical attributes, the relationship between the two historical attributes including positive relationship, reverse relationship and vertical relationship; and the data layer searches for the documents containing the two historical attributes in the document units based on the relationship knowledge table to acquire the relationship confidence value of the two historical attributes in a fusion manner to establish the relationship confidence value base of the two historical attributes, the relationship between the two historical attributes including positive relationship confidence value, reverse relationship confidence value and vertical relationship confidence value; thereby, the data layer constructs the historical data set based on the relationship knowledge base and the relationship confidence value base established in the form of pairing of all the historical attributes.

[0019] According to a preferred embodiment, the application further discloses a cause-effect relationship determination method based on a Bayesian network, the method comprising: a constructing module constructing a Bayesian network; a determining module generating and outputting a cause-effect relationship spectrum based on the Bayesian network; a cause-effect library establishing a historical original document library; in the case where at least one non-specific event occurs to at least two objects, the constructing module analyzes the self factors of the objects causing the non-specific event to construct a factor set and construct a constraint of undirected graph structure based on the factor set; and the constructing module can correct the constraint of undirected graph structure based on the cause-effect library to establish the Bayesian network; the determining module calculates a cause-effect index of the self factors causing the non-specific event based on the Bayesian network and outputs a cause-effect relationship spectrum of the non-specific event based on the cause-effect index, so as to determine the key reason of the non-specific event.

[0020] According to a preferred embodiment, the method further comprises: the constructing 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 constructing module prompts the third party to find evidence; in the case that the third party confirms that the exclusive factor is objectively existing, the constructing module supplements the exclusive factor to the undirected graph structure constraint, and further modifies the undirected graph structure constraint.

[0021] The advantage of the present application mainly lies in constructing the undirected graph constraint of the Bayesian network between factors of at least one non-specific event occurred between at least two objects, modifying the undirected graph constraint based on the literature knowledge base and the expert knowledge base, and then constructing the Bayesian network based on the undirected graph constraint by using a heuristic search algorithm. The structure learning of the Bayesian network is mainly used to reveal the qualitative relationship between variables, and also to reveal the quantitative relationship. However, there are many difficulties in constructing the Bayesian network from the data angle. Relying on the non-specific event between objects and its historical events can increase the correctness of the Bayesian network construction. Finally, based on the constructed Bayesian network, the causal analysis reasoning is realized. In summary, the fusion of object knowledge to construct a reasonable Bayesian network and causal analysis reasoning, and the determination of the causal relationship spectrum of the non-specific event and the key reason are the key problems to be solved by the present application. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 is a flowchart of a preferred embodiment of the determination method provided by the present application; and

[0023] Figure 2 is a preferred module schematic diagram of the determination system provided by the present application.

[0024] LIST OF REFERENCE NUMERALS

[0025] 1: constructing module 2: determination module

[0026] 3: causal library DETAILED DESCRIPTION

[0027] The present application will be described in detail below with reference to the accompanying drawings. Figure 1 and 2 will be described in detail.

[0028] In the description of the present application, the terms "first", "second", "third" and the like are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implying the number of the technical features indicated. Therefore, the features defined with "first", "second", "third" and the like can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise specifically limited.

[0029] Embodiment 1

[0030] The present embodiment provides a Bayesian network-based causal relationship determination system, which aims to output the causal relationship spectrum of at least one non-specific event occurring between at least two objects. The advantage of the present application mainly lies in constructing the undirected graph constraint of the Bayesian network between the factors of at least one non-specific event occurring between at least two objects, and correcting the undirected graph constraint based on the literature knowledge base and the expert knowledge base, and then constructing the Bayesian network based on the undirected graph constraint using a heuristic search algorithm. The structure learning of the Bayesian network is mainly used to reveal the qualitative relationship between variables, and at the same time to reveal the quantitative relationship. However, there are many difficulties in constructing the Bayesian network from the data alone. Relying on the non-specific events and their historical events between objects can increase the correctness of the construction of the Bayesian network. Finally, based on the constructed Bayesian network, causal analysis reasoning is realized. In summary, the fusion of object knowledge to construct a reasonable Bayesian network and causal analysis reasoning to determine the causal relationship spectrum of non-specific events and to determine the key reasons are the key problems to be solved by the present application.

[0031] Specifically, the system includes a construction module 1, a determination module 2 and a causal library 3. The construction module 1 is used to construct a Bayesian network. The determination module 2 is used to generate and output the causal relationship spectrum based on the requested Bayesian network. The causal library 3 is used to establish a historical original literature library. For the case of at least one non-specific event between 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 construct an undirected graph structure constraint based on the factor set. And the construction module 1 can correct the undirected graph structure constraint based on the causal library 3 to establish a Bayesian network. The determination module 2 calculates the causal index of the own factors causing the non-specific event based on the Bayesian network and outputs the causal relationship spectrum of the non-specific event based on the causal index, thereby determining the key reason of the non-specific event. In the present application, the non-specific event can be a real event, such as a collision between two cars, etc. The non-specific event can also be an event that researchers expect, such as a failure of a spacecraft docking, etc.

[0032] Preferably, the construction module 1 defines at least one domain based on non-specific events, and retrieves at least one other factor related to the non-specific events in the domain based on the causal database 3. If the other factor is not a self-factor, the construction module 1 prompts a third party to verify the evidence; if the third party confirms the objective existence of the other factor, the construction module 1 adds the other factor to the undirected graph structure constraint to further modify the undirected graph structure constraint.

[0033] Preferably, module 1 constructs factor pairs L based on the causal library 3. m and L n If the retrieval of the relationship between factors fails, Module 1 prompts a third party to check the relationship between factors and L. m and L n The relationships between factors are verified by experts, and the results of the expert verification are fed back to causal database 3. Causal database 3 then uses deep learning to refine its causal database based on the expert verification results. With causal database 3 using deep learning to refine its causal database based on the expert verification results, module 1 constructs modules based on causal database 3 to classify L according to factors. m and L n Numbering of factors to L m and L n The relationship between them needs to be corrected.

[0034] Preferably, module 1 establishes a dataset D based on its own factors, and divides its own factors into several factor sets L according to dataset D, forming factor pairs L by pairing them up in a pairwise manner. m and L n And for dataset D and factors on L m and L n Numbering is performed. Module 1, based on the causal library 3, assigns factor pairs to L based on their numbering. m and L n The relationship between L is searched, and correction factors are applied to L based on the search results. m and L n The relationship between L and m →L n Relationship reliability score, L n →L m Relationship reliability score and L n ⊥L m The relational confidence values ​​are assigned to construct the undirected graph structure constraints.

[0035] Preferably, according to Bayes rule, the constructing module 1 constructs a Bayes network evaluation function based on the data set D and the factor pair set L, which is used to iteratively generate a Bayes network with the highest evaluation index from several candidate Bayes networks based on the undirected graph structure constraint in the case that the constructing module 1 enables the heuristic search algorithm. Preferably, the constructing module 1 firstly constructs an initial first candidate Bayes network based on the undirected graph structure constraint and evaluates it to obtain a first evaluation index using the Bayes network evaluation function; then, the constructing module 1 enables the heuristic search algorithm to locally modify a second candidate Bayes network based on the undirected graph structure constraint, and evaluates the second candidate Bayes network again using the Bayes network evaluation function to obtain a second evaluation index; the constructing module 1 can obtain at least two candidate Bayes networks and corresponding evaluation indexes in an iterative loop manner based on the heuristic search algorithm. In the case that the constructing module 1 obtains at least two candidate Bayes networks and at least two evaluation indexes, it will output a final Bayes network with the optimal evaluation index as the output causal relationship spectrum of the non-specific event.

[0036] Preferably, the determining module 2 calculates the causal index between each factor L m and L n based on the final Bayes network and the Pearl principle, thereby outputting the causal relationship spectrum. Wherein, for the factor L m , the node set connected with the factor L m is obtained based on the undirected graph structure constraint in a traversal manner. And the correlation between each node and the factor L m is calculated successively, and the node with the largest correlation is selected to make the independence assumption, and the node independent of L m under the given data set D is deleted to improve the determination efficiency of the determining module 2. The independence between the factor L n and the factor L m is measured by mutual information:

[0037]

[0038] When the mutual information exceeds the threshold of mutual information, the factor L n and the factor L m have correlation and are not independent; when the mutual information does not exceed the threshold of mutual information, the factor L n and the factor L m have no correlation and are independent.

[0039] Preferably, the cause-effect library 3 is established in the following way: the cause-effect library 3 forms a number of literature units based on a number of relevant literatures containing a plurality of historical attributes acquired and classified to construct a raw literature library in the technical field to construct a relationship confidence value between attributes through data pattern mining; wherein the literature layer in the cause-effect library 3 counts the frequency of words / phrases in each literature, and obtains the joint occurrence probability of words / phrases according to the independence assumption; the literature layer calculates the relevance strength of words / phrases; the literature layer constructs the associated reduced coordinates of the literature, and classifies the relevant literatures in the form of an iterative algorithm according to a classification function constructed based on the associated reduced coordinates and the relevance strength of all relevant literatures to form a number of literature units.

[0040] Preferably, in the case of obtaining the literature unit in the cause-effect library 3, the data layer obtains the data set in the form of pairing two historical attributes; the data layer extracts the relationship between the two historical attributes in each relevant literature in the form of syntactic analysis of natural language processing to establish the relationship knowledge base of the two historical attributes, and the relationship between the two historical attributes includes positive relationship, negative relationship and vertical relationship; and the data layer searches for the literature containing two historical attributes in the literature unit based on the relationship knowledge table to obtain the relationship confidence value of the two historical attributes in a fusion manner to establish the relationship confidence value library of the two historical attributes, and the relationship between the two historical attributes includes positive relationship confidence value, negative relationship confidence value and vertical relationship confidence value; thereby, the data layer constructs the historical data set based on the relationship knowledge base and the relationship confidence value library established in the form of pairing two by two between all historical attributes.

[0041] Embodiment 2

[0042] The embodiment discloses a cause-effect relationship determination method based on Bayesian network, which is used for responsibility determination in medical disputes. The whole and / or part of the preferred embodiments of other embodiments can be supplemented as the embodiment. Preferably, the method can be realized by the method of the application and / or other alternative modules.

[0043] Medical trouble refers to the behavior of patients who are employed by medical disputes, together with patient families, to seriously hinder medical order, expand the situation, and cause negative impact on the hospital in various ways to exert pressure on the hospital and make profits. The direct consequence of medical trouble is to cause a large number of direct or indirect losses of medical staff in China, and to have very serious and adverse effects, which seriously affect the development of medical undertakings in China. When a medical dispute occurs, it is difficult to classify the causes of the medical dispute. Medical dispute refers to the dispute occurring in medical health, prevention and health care, medical beauty and other medical enterprises or institutions with legal qualifications. At present, medical disputes in China are particularly difficult to handle. The reason is that medical disputes are usually caused by medical errors and negligence. Medical negligence is the error of medical staff in the process of diagnosis and nursing. Medical error refers to the error of medical staff in medical activities such as diagnosis and treatment. These errors often lead to patient dissatisfaction or harm to patients, thus causing medical disputes. In addition to medical errors and negligence, sometimes the medical party has no negligence and errors in medical activities, and only due to the patient's unilateral dissatisfaction, disputes can also be caused. This kind of dispute can be caused by the patient's lack of basic medical knowledge, misunderstanding of correct medical treatment, natural change of disease and inevitable complications, and medical accidents, or it can be caused by the patient's unreasonable accusation. Some people also call it medical tort dispute, that is, the dispute between the provider and the recipient of medical services on whether the medical behavior and its consequences are tort and the responsibility for tort. Therefore, in order to give the doctor a comfortable and healthy working environment and a fair and just explanation to the patient, the causes of the medical dispute need to be presented in a transparent way, and transparency and fairness are achieved.

[0044] Therefore, the present embodiment provides a method for determining causality based on a Bayesian network, which aims to assist the judicial department to solve medical disputes. In the judicial practice of medical disputes, the adoption of the evidence or factors claimed by both parties is mostly from a legal perspective, and the relationship between the evidence or factors is mostly evaluated from a qualitative perspective, which is one of the reasons for the entanglement of the medical and patient parties and one of the reasons for the confusion of the judicial personnel in adjudication or judgment. In the case of entanglement of the medical and patient parties, both parties will question and even appeal the adjudication or judgment, occupying too many legal resources. In the era of advocating the value pursuit of "fairness and justice", solving the contradiction between the medical and patient parties through the way of "data speaking" to give the doctor a comfortable and easy working environment for saving lives, give the patient or his family a convincing explanation, and provide the judicial department with a scientific reference file, which is the important value of the present invention.

[0045] Specifically, the method mainly includes:

[0046] S1: constructing module 1: constructing a Bayesian network. Specifically, the constructing module 1 constructs an undirected graph structure constraint based on the factors claimed by the medical party and the patient party. And in the case of third party intervention, the constructing module 1 corrects the undirected graph structure constraint based on the causal library 3 to establish a Bayesian network.

[0047] S2: determining module 2: outputting a causal relationship spectrum based on the Bayesian network. The determining module 2 calculates the causal index between each factor pair based on the Bayesian network and outputs the causal relationship spectrum of the medical dispute based on the causal index, so that the causal relationship spectrum takes into account the factors claimed by both the medical party and the patient party and considers scientificity, which can effectively prevent the escalation of disputes between the medical party and the patient party and provide data support for the third party to make a ruling or judgment.

[0048] Preferably, the step S1 includes the following specific steps:

[0049] S11: establishing a data set D=(D1, D2……D i ) according to the factors claimed by the medical party and the patient party. L=(L1, L2……L n ) is a specific factor pair of a certain group of attributes. For example, the factors claimed by the medical party include patient illness time, hospitalization time, illness degree and illness type. Then the data set can be established with the time attribute and the disease attribute, i.e. D=(time attribute, disease attribute). And the specific factor pair L corresponds to the specific attributes of the illness time and the hospitalization time of the time factor, and is numbered.

[0050] S12: constructing an undirected graph structure constraint based on the relationship between the factors claimed by the medical party and the patient party. The relationship between each specific factor pair is determined by searching the relationship between the factors claimed by the medical party and the patient party, and the factor pairs are numbered. Preferably, the relationship between the factor pairs includes positive relationship, negative relationship and vertical relationship, i.e. attribute L m influences attribute L n , denoted as L m →L n . The relationship between attribute L m and attribute L n may be inverse relationship, i.e. attribute L n influences attribute L m , denoted as L n →L m . The relationship between attribute L m and attribute L n may be vertical relationship, i.e. attribute L m and attribute L n , mutually independent L n ⊥L mFor example, the patient believes that the drug component causes the cerebral palsy to aggravate, and the drug component affects the cerebral palsy to aggravate, so the drug component and the cerebral palsy to aggravate constitute a positive relationship, which is recorded as drug component→cerebral palsy to aggravate. Of course, it can also be considered that the cerebral palsy to aggravate and the drug component constitute a negative relationship, which is also recorded as drug component→cerebral palsy to aggravate.

[0051] S13: modifying the undirected graph structure constraint based on the causal library. In the causal library, the factor pair number is searched, and the relationship between L m and L n is modified, and the relationship confidence value of L m → L n , the relationship confidence value of L n → L m and the relationship confidence value of L n ⊥ L m are assigned. For example, the patient believes that the drug component causes the cerebral palsy to aggravate, but the search in the causal library finds that the drug component does not cause the cerebral palsy to aggravate, so the drug component→cerebral palsy to aggravate is modified to drug component⊥cerebral palsy to aggravate. For example, the drug component is L1, and the cerebral palsy to aggravate is L2, so the number of drug component→cerebral palsy to aggravate is recorded as 12.

[0052] Preferably, the construction module 1 defines at least one request field based on the factors claimed by the medical party and the patient party, and searches for at least one other factor related to the factors claimed by the medical party and the patient party based on the request field. In the case that the at least one factor is not in the factors claimed by the medical party and the patient party, the construction module 1 prompts the third party to find evidence. In the case that the third party determines that the at least one factor exists objectively, the construction module 1 supplements the at least one factor to the undirected graph structure constraint and further modifies the undirected graph structure constraint. For example, a newborn death event occurs, and the factors claimed by the medical party and the patient party include amniotic embolism, hypoxia and multiple births, so the construction module 1 defines the request field of women giving birth based on these factors, and then searches for other factors existing in the accident in the corresponding women giving birth field in the causal library 3, such as searching for thin uterine wall. However, the thin uterine wall does not appear in the factors claimed by the medical party and the patient party, so the construction module 1 prompts the third party to find out whether the puerpera has a thin uterine wall. If the condition exists objectively, the construction module 1 supplements the factor of thin uterine wall to further modify the undirected graph structure constraint. That is, in step S13, there is another possibility that at least one factor not claimed by either party is an important factor affecting the result, so the construction module 1 prompts the third party to find out the at least one factor. If the at least one factor exists objectively, the at least one factor needs to be added to the undirected graph structure constraint for modification, so as to increase the reliability and scientificity of the result and improve the fair judgment or ruling of the third party, which shows the rigor, fairness and responsibility of the third party.

[0053] Preferably, the relationship between some factor pair or some factor pairs and its relationship confidence value can not exist in the cause-effect library 3. In order to guarantee the factors claimed by both parties can be supported. That is, the building module 1 prompts the third party to conduct expert argumentation on the relationship between the factor pair L m and L n if the building module 1 fails to search the relationship between the factor pair L m and L n based on the cause-effect library 3. For example, the patient party claims that the head size of the newborn is not the cause of the death of the newborn. The relationship between the head size of the newborn and the death of the newborn does not exist in the cause-effect library 3. Then the building module 1 prompts the third party to conduct expert argumentation. And the result of the expert argumentation is fed back to the cause-effect library 3, and the cause-effect library 3 learns the result of the expert argumentation in depth to correct the cause-effect library 3. In the case that the cause-effect library 3 learns the result of the expert argumentation in depth, the building module 1 corrects the relationship between the factor pair L m and L n based on the cause-effect library 3 according to the factor pair L m and L n number.

[0054] Preferably, the building module 1 builds a Bayesian network evaluation function based on the cause-effect library 3, the data set D and the factor pair set L:

[0055] logP(G,D,K L )=logP(G)+logP(D|G)+logP(K L |G)

[0056] Preferably, the building module 1 builds a requested Bayesian network based on the Bayesian network evaluation function and the undirected graph structure constraint. In the formula, G is a Bayesian grid, whose value includes a directed acyclic graph in which a specific factor pair of a certain group of attributes L=(L1, L2…L n ) is combined as a node. Wherein, P(G) is a prior distribution. According to existing knowledge, the maximum value of logP(G)+logP(D|G) is equivalent to logP(G|D). logP(G|D) can be calculated according to the Bayesian information criterion BIC score. In the formula,

[0057]

[0058] Wherein, if any edge in the structure G is represented as L m →L n , then KL(L m →L n) is the relational confidence value. The summation in the formula is the summation of the document knowledge confidences of the positive relations corresponding to all the directed edges in the structure G. For a given dataset D, for any factor pair L m and L n , a Bayesian network is constructed based on the Bayesian network evaluation function and the undirected graph structure constraint. After determining the undirected graph structure constraint of the Bayesian network, a heuristic search algorithm such as the K2 algorithm can be performed to seek the network structure with the optimal score function. The general steps are: starting from an initial model, at each step of the search, first locally modify the current model using a search operator to obtain a series of candidate models, then calculate the score of each candidate model, and compare the optimal candidate model with the current model. If the score of the optimal candidate model is large, it is used as the next current model and the search continues; otherwise, the search is stopped and the current model is returned. According to the Bayesian principle, the candidate model with the largest score is the Bayesian network. Preferably, the Bayesian network evaluation function is constructed according to the established Bayesian network and the Bayesian rule. The Bayesian network evaluation function can be constructed according to classical heuristic structure learning algorithms, such as the K2 algorithm, the Max-Min Parents and Children algorithm, and the Markov Chain Monte Carlo search, etc.

[0059] Preferably, the determining module 2 calculates the causal index between each factor pair based on the requested Bayesian network and the Pearl principle, thereby outputting the causal relationship spectrum. The determining module 2 calculates the causal index between attributes based on data pattern mining, thereby being able to determine whether the attributes constitute a complication or a comorbidity according to the causal index. When the causal index is calculated, the determining module 2 calculates the causal index between attributes based on the Pearl principle and the Bayesian network structure. Pearl needs to calculate E(Y|do(X)) by intervening X to implement the X event when exploring whether event X is the cause of event Y, i.e. the average change of event Y under the intervention of X is greater than the significance level, then X is considered to be the cause of Y.

[0060] When the determining module 2 calculates the causal index between attributes based on data pattern mining, due to the huge amount of literature, the Bayesian grid is huge, therefore, the backdoor criterion is used to calculate the causal index. The backdoor criterion refers to that the Bayesian grid G is a directed acyclic graph, (L m , L n ) is a pair of nodes of G, and the node Z set is the backdoor of (L m , L n ), wherein all nodes in Z are not descendants of Z and Z blocks all connections L m to L m pointing to L n . Therefore, the causal relationship of the factor pair Lm and Ln can be inferred by the backdoor principle.

[0061] To be able to determine the module 2 by independence test to simplify the undirected graph constraints without affecting the causal relationship between factors. For example, the independence test can be used chi-square independence test. In the present invention, the independence test can also be used as follows: for the factor L m , by tabulating form based on the constructed undirected graph to obtain the node set consisting of L m connected nodes. And each node and the factor L m between the correlation is calculated, and from which the largest correlation node is selected to make independence assumptions, delete the given request subset D i independent nodes with L m In the present invention, the entropy is used to measure the uncertainty of the factor pair set L m . Given the factor L m , the uncertainty of the factor L n can be measured by conditional entropy as follows:

[0062]

[0063] The degree of correlation between the factor L n and L m can be measured by mutual information:

[0064]

[0065] If the mutual information exceeds the threshold of mutual information, it is considered that L n and L m have correlation. If the mutual information does not exceed the threshold of mutual information, it is considered that L n and L m have no correlation.

[0066] Preferably, the causal library 3 is established as follows: the causal library 3 is based on the obtained a large number of related literature containing a variety of historical attributes in the technical field and classifies them to form a number of literature units to construct the original literature library, to mine the relationship between historical attributes by data pattern. Among them, the literature layer of the causal library 3 counts the frequency of words / phrase in each literature, and obtains the joint occurrence probability of words / phrase according to the independence assumption. The literature layer calculates the correlation strength of words / phrase; the literature layer constructs the correlation reduction coordinates of the literature, and classifies the related literature according to the iterative algorithm form based on the correlation reduction coordinates and correlation strength of all related literature to form a number of literature units.

[0067] Preferably, the literature layer is based on a large number of relevant literatures containing various historical attributes. The literature layer classifies the relevant literatures into several literature units to form a raw literature database. The relevant literatures include medical records, research reports, conference literatures, journal literatures, books, academic papers and patents. In the case of such a large number of literatures, they need to be classified according to a certain method. The literature classification is to effectively observe the correlation between the historical attributes and reduce the load of the system. For example, it can be classified according to digestive diseases, cardiovascular diseases and neurological diseases. It can also be classified according to academic fields, such as rehabilitation and psychology. However, in the case of a large number of literatures, accurate and efficient classification will directly affect the differentiation of complications and comorbidities. Preferably, the literature classification can use Bayesian method, SVM method and k-NN method.

[0068] Preferably, the classification of the relevant literatures is carried out in the following manner: the literature layer counts the frequency of each word / phrase in each literature, and obtains the joint occurrence probability of the word / phrase according to the independence assumption. For example, for a specific literature, the joint occurrence probability distribution can be calculated according to the Naive Bayes method.

[0069] Preferably, the literature layer calculates the correlation strength of the word / phrase. Through the calculation of the correlation strength, the correlation of the word / phrase can be reflected, which is suitable for the classification of the literature. Preferably, in the classification, N is defined as a set of literature samples, V is a set of literature types, V i is a subset of the ith literature type. W is a set of words / phrases, W i is a subset of the ith word / phrase. S i is the number of samples in V j , and the correlation reduction coordinates T p of the pth sample is an n-dimensional array:

[0070]

[0071] where k i is the number of occurrences of the ith word in k is the normalization coefficient.

[0072] The correlation vector of V i is the average of the correlation reduction coordinates of all samples in V i , which reflects the correlation strength of the word / phrase in the literature, that is:

[0073]

[0074] Preferably, the document layer obtains the associated reduced coordinates of the documents, and classifies the associated documents in the form of an iterative algorithm based on the classification function constructed by the associated reduced coordinates of all the associated documents to form a plurality of document unit bodies. Preferably, the associated reduced coordinates of any document are:

[0075]

[0076] wherein q i is the number of occurrences of the i-th word in the document. In the classification, the distance between the document to be classified and the support points (b1, b2, …, b i ) of each category of documents V n is recorded as:

[0077]

[0078] According to the relevance intensity, a document classification function is constructed:

[0079]

[0080] wherein γ i is related to the relevance intensity.

[0081] Preferably, the iterative algorithm can adopt a minimization iterative algorithm, a minimum optimization iterative algorithm, and an expectation maximum iterative algorithm. Preferably, the classification function can perform deep learning based on the sample size of the associated documents, thereby enhancing the accuracy of the document layer.

[0082] Preferably, in the case where the data layer in the cause-effect library 3 obtains the document unit bodies, the data layer obtains the historical data set in the form of pairing of two historical attributes. The data layer extracts the relationship between the two historical attributes in each associated document in the form of syntactic analysis of natural language processing, to establish a relationship knowledge base of the two historical attributes, the relationship between the two historical attributes including a positive relationship, a reverse relationship, and a vertical relationship. Moreover, the data layer retrieves the documents containing the two historical attributes in the document unit bodies based on the relationship knowledge base to obtain the relationship confidence values of the two historical attributes in a fusion manner, to establish a relationship confidence value base of the two historical attributes, the relationship between the two historical attributes including a positive relationship confidence value, a reverse relationship confidence value, and a vertical relationship confidence value. Thus, the data layer constructs the historical data set based on the relationship knowledge base and the relationship confidence value base established by pairing of all the histories in pairs.

[0083] Preferably, the data layer can obtain the main characteristic parameters based on the literature unit and construct the historical data set based on the main characteristic parameters. In order to reduce the interference of a large number of characteristic parameters formed by a large number of related literatures on the causal relationship between historical attributes and improve the utilization value of the original literature library. Preferably, in the case of obtaining the literature unit in the data layer, the data layer obtains the historical data set in the form of pairing two historical attributes. The data layer extracts the relationship between two historical attributes in each related literature in the form of syntactic analysis of natural language processing, so as to establish the relationship knowledge base of two historical attributes, and the relationship between two historical attributes includes positive relationship, reverse relationship and vertical relationship. And, the data layer searches for the literature containing two historical attributes in the literature unit based on the relationship knowledge table to obtain the relationship confidence value of two historical attributes in a fusion manner to establish the relationship confidence value library of two historical attributes, and the relationship between two historical attributes includes positive relationship confidence value, reverse relationship confidence value and vertical relationship confidence value. Thus, the data layer constructs the historical data set based on the relationship knowledge base and the relationship confidence value library established in the form of pairing two by two between all historical attributes. For example, in the related literature, a plurality of historical attributes such as historical attribute L1, historical attribute L2, historical attribute L3 and historical attribute L4 are obtained. According to the relationship of the above historical attributes, the relationship knowledge base of historical attribute L1 and historical attribute L3, the relationship knowledge base of historical attribute L2 and historical attribute L3 can be established, and so on. Then, in the unit literature body, the relationship confidence value library is constructed based on the above relationship knowledge base according to the content in different literatures. Preferably, the sum of the positive relationship confidence value, the reverse relationship confidence value and the vertical relationship confidence value is normalized. That is, in the unit literature body, all the literatures are queried, and the positive relationship confidence value, the reverse relationship confidence value and the vertical relationship confidence value are weighted according to the frequency. The data layer inputs the above relationship knowledge base and relationship confidence value library into the judgment module 2 to perform the next step.

[0084] Preferably, for journal literature, the positive relationship confidence value of L1→L2 can also be defined as follows:

[0085]

[0086] Wherein, C(Xi) is the credibility of literature Xi, the formula is: C(Xi)=(IFi+1)×(CIi+1), Xi represents the i th literature, IFi is the standardized impact factor of the journal where the literature Xi is located, CIi is the standardized citation amount. If there is no literature existing the relationship between L1 and L2, then KL(L1→L2)=0, KL(L2→L1)=0, KL(L1⊥L2)=1. Other types of literature can be defined in the same way, for example, medical records can be defined according to the authority of doctors. For conference articles, they can be defined according to the authority of the conference, etc.

[0087] Embodiment 3

[0088] The embodiment discloses a Bayesian network-based causal relationship determination system, and the whole and / or part of the preferred implementation of other embodiments can be supplemented as the embodiment. Preferably, the method can be implemented by the method of the application and / or other alternative modules.

[0089] The embodiment provides a Bayesian network-based causal relationship determination method, which comprises: a construction module 1 constructing a Bayesian network; a determination module 2 generating and outputting a causal relationship spectrum based on the Bayesian network; and a causal database 3 establishing a historical original literature database. In the case that at least one non-specific event occurs in 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 construct a non-directed graph structure constraint based on the factor set; and the construction module 1 can correct the non-directed graph structure constraint based on the causal database 3 to establish the Bayesian network; the determination module 2 calculates a causal index of the self factors causing the non-specific event based on the Bayesian network and outputs a causal relationship spectrum of the non-specific event based on the causal index, so as to determine the key reason of the non-specific event.

[0090] Preferably, the construction module 1 defines at least one field based on the non-specific event, and retrieves at least one exclusive factor related to the non-specific event in the field based on the causal database 3. In the case that the exclusive factor does not belong to the self factor, the construction module 1 prompts a third party to find evidence. In the case that the third party confirms that the exclusive factor objectively exists, the construction module 1 supplements the exclusive factor to the non-directed graph structure constraint and further corrects the non-directed graph structure constraint.

[0091] Preferably, the construction module 1 used in the application is a server with a search engine and an operation function. The determination module 2 is a data server with an operation function. The causal database 3 is a server with a search engine, an operation function and storage. The construction module 1, the determination module 2 and the causal database 3 are connected in communication with each other through optical fibers, data lines, Bluetooth, wifi and / or 4G and the like wired and wireless modes.

[0092] It should be noted that the above specific embodiments are exemplary, and those skilled in the art can think of various solutions under the inspiration of the disclosure of the application, and these solutions also belong to the disclosed range of the application and fall within the protection scope of the application. Those skilled in the art should understand that the specification and drawings of the application are illustrative and do not constitute a limitation on the claims. The protection scope of the application is defined by the claims and their equivalents.

Claims

1. A causal relationship determination system, comprising at least: Module (1) is used to build Bayesian networks; Causal database (3) is used to establish a historical original document database. Its features are, In the case where at least two objects experience at least one non-specific event, the construction module (1) analyzes the self-factors of the objects that caused the non-specific events to construct a factor set and constructs an undirected graph structure constraint based on the factor set. The construction module (1) establishes a dataset D based on its own factors, and divides its own factors into several factor sets L according to the dataset D as the unit, and forms factor pairs L in a pairwise pairing manner. m and L n And for dataset D and factors on L m and L n Number them. Module (1) builds on the factor pair L based on the causal library (3). m and L n If a search for the relationship between factors L fails, a third-party prompt will be made to check the relationship between factors L. m and L n The relationship between them is verified by experts, and the results of the expert verification are fed back to the causal database (3). The causal database (3) performs deep learning on the results of the expert verification to correct the causal database (3). When the causal database (3) performs deep learning on the results of the expert argumentation, the construction module (1) performs L based on the causal database (3) according to the factors. m and L n Numbering the factors to L m and L n The relationship between them needs to be corrected.

2. The causal relationship determination system according to claim 1, characterized in that, The building module (1) can modify the undirected graph structure constraints based on the causal library (3) to build a Bayesian network.

3. The causal relationship determination system according to claim 2, characterized in that, The building module (1) defines at least one domain based on non-specific events, and retrieves at least one exclusive factor related to non-specific events in the domain based on the causal library (3).

4. The causal relationship determination system according to claim 3, characterized in that, Module (1) is built based on the causal library (3) according to the factor pair numbering pair factor pair L m and L n The relationship between L is searched, and correction factors are applied to L based on the search results. m and L n The relationship between L and m →L n Relationship reliability score, L n →L m Relationship reliability score and L n ⊥L m The relational confidence values ​​are assigned to construct the undirected graph structure constraints.

5. The causal relationship determination system according to claim 4, characterized in that, The building module (1) constructs a Bayesian network evaluation function based on the dataset D and the factor pair set L according to the Bayesian rule. The evaluation function is used to iteratively generate the Bayesian network with the highest evaluation index from several candidate Bayesian networks based on the undirected graph structure constraints and with the heuristic search algorithm enabled in the building module (1).

6. The causal relationship determination system according to claim 5, characterized in that, The causal relationship determination system also includes a determination module (2), which calculates the impact of each factor on L based on the final Bayesian grid and Pearl's principle. m and L n The causal indicators between them are used to output a causal relationship spectrum.

7. The causal relationship determination system according to claim 6, characterized in that, The decision module (2) obtains the factor L based on the undirected graph structure constraints through traversal. m The connected nodes constitute its node set.

8. The causal relationship determination system according to claim 7, characterized in that, The judgment module (2) calculates the relationship between each node and factor L sequentially. m The correlation between nodes is analyzed, and the nodes with the highest correlation are selected. Independence assumptions are then made, and nodes with the highest correlation in the given dataset D are removed. m Independent nodes.

Images