CN-115470334-B - Improved DS evidence theory method based on liver disease consultation discussion model

CN115470334BCN 115470334 BCN115470334 BCN 115470334BCN-115470334-B

Abstract

The invention relates to an improved DS evidence theory method based on liver disease consultation study model, which comprises the steps of firstly establishing a dispute table and a dialogue tree model for consultation, establishing an identification frame, obtaining initial evaluation values of all nodes, rule relations among the nodes and corresponding strength, obtaining evaluation values of all evidence nodes for conclusion nodes according to CEM matrix, calculating PIGNISTIC probability distances and conflict coefficients in a fusion process, synthesizing to obtain conflict distances, scaling the distances while considering conflict consistency among evidence sources in the synthesis process, calculating the credibility of all evidence sources again, carrying out weighted average correction on the evidence sources according to the credibility of all nodes, and carrying out fusion by using traditional DS evidence combination rules to obtain the final evaluation value of the nodes. In an example, the method can effectively solve the conflict problem without changing DS combination rules, and has better anti-interference capability and convergence.

Inventors

XI XUGANG
ZHOU YU
LI LIHUA
Fan Panhui
WANG TING
MENG MING

Assignees

杭州电子科技大学

Dates

Publication Date: 20260508
Application Date: 20220913

Claims (5)

1. An improved DS evidence theory method based on liver disease consultation discussion model is characterized by comprising the following steps: step one, creating a liver disease consultation discussion model, determining an identification framework, converting a consultation scene into a dispute table, and creating a dialogue tree; step two, calculating the evaluation value of the evidence node to the theoretical node according to the initial evaluation value of the evidence of each dispute, the rule relation and response strength of the evidence and the conclusion; The first step specifically comprises the following steps: Step 1.1 determining an identification framework Identifying frame power sets The method comprises three propositions, namely trust, distrust and unknowing, wherein the trust, the distrust and the unknowing of a statement are respectively represented by a doctor, and the evaluation value of each node is represented by a vector w= (b, d, i); Step 1.2, converting all doctor views in a consultation scene into a dispute table through natural language understanding, wherein each dispute contains four parts of contents, namely evidence, conclusion, initial evaluation value of the evidence, rule relation and response strength of the evidence and the conclusion, taking the evidence of the dispute as a child node and the conclusion as a father node to form a dialogue tree, and the rule relation of the evidence and the conclusion is divided into four types: ; where p represents evidence, h represents a conclusion, the former two represent evidence or a negative support conclusion of evidence, and the latter two represent evidence or a negative objection conclusion of evidence; The intensity of the response of evidence and conclusions is expressed as: ; the value range is 0-1; Thirdly, calculating PIGNISTIC probability distance matrixes P among n nodes; Step four, calculating a conflict coefficient matrix K among n nodes; The fourth step specifically comprises: calculating a conflict coefficient matrix K between n nodes: ; Wherein the method comprises the steps of The conflict coefficient between the node i and the node j is represented by the following specific formula: ; Wherein the method comprises the steps of Representing the probability distribution function of the i-th node, And Respectively representing propositions of i node and j node under the identification framework; Step five, calculating a conflict distance matrix D among n nodes; The fifth step specifically comprises the following steps: step 5.1, calculating evidence conflict/consistency among n nodes : ; Representing evidence conflict/consistency for inodes and j-nodes, where And The evidence conflict quantity and the evidence consistent quantity of the i node and the j node are respectively represented, and the specific formulas are as follows: ; ; Step 5.2, calculating a conflict distance matrix D between n nodes: ; Wherein the method comprises the steps of The collision distance between the node i and the node j is represented by the following specific formula: ; Wherein the method comprises the steps of Representing the distance between node i and node j; step six, calculating the credibility Crd (i) of n nodes; And step seven, modifying the evidence source by taking the credibility as a weight, and carrying out traditional DS evidence combination on the modification result.
2. The method for improving DS evidence theory based on liver disease consultation discussion model of claim 1, wherein said step two specifically comprises: Step 2.1, for a dispute A, obtaining a complete evidence mapping matrix CEM (A) according to the rule relation and response intensity of evidence and conclusions, ; Step 2.2 calculating a conclusion evaluation value w (h) from the initial evaluation value w (p) of evidence p of dispute A and the CEM matrix 。
3. The method for improving DS evidence theory based on liver disease consultation discussion model of claim 1, wherein said step three specifically comprises: Step 3.1, for a certain node in the dialogue tree, the evaluation value of the node by the child node of the node can be obtained through the steps, n-1 related child nodes (n is more than or equal to 1) are shared by the node, n evaluation value vectors w 1 ,···,w n (n is more than or equal to 1) can be obtained by adding the node, m i (b),m i (d),m i (a) is used for representing the belief assignment of the propositions b, d and a under the probability distribution function m i of the ith node; step 3.2, PIGNISTIC probability functions of n node evaluation value vectors are calculated, wherein the PIGNISTIC probability function calculation formula of the ith node is as follows: ; Where |B| represents the potential of B subset B, i.e., the number of elements contained in set B; Step 3.3, PIGNISTIC probability distance matrix P among n nodes is calculated; ; Wherein the method comprises the steps of The specific formula for representing the distance between the node i and the node j is as follows: 。
4. The method for improving DS evidence theory based on liver disease consultation discussion model of claim 1, wherein said step six specifically comprises: step 6.1, calculating a similarity matrix S among n nodes, ; Wherein the method comprises the steps of Representing the similarity between node i and node j, the similarity and distance are the opposite concepts, so the similarity can be expressed as: ; step 6.2, calculating the support degree Sup (i) of n nodes: ; summing all elements except the similarity of each row in the similarity matrix, and reflecting the supporting degree of the node by all other nodes by the method; step 6.3, normalizing the n node support degrees to obtain a reliability Crd (i): 。
5. The method for improving DS evidence theory based on liver disease consultation discussion model of claim 1, wherein said step seven specifically comprises: Step 7.1, modifying an evidence source, and carrying out weighted average modification by taking the credibility Crd (i) of each node as a weight; ; And 7.2, synthesizing the modified n nodes for n-1 times by using a traditional DS evidence combination rule, wherein the synthesis formula is as follows: 。

Description

Improved DS evidence theory method based on liver disease consultation discussion model Technical Field The invention relates to an improved DS evidence theory method based on a liver disease consultation discussion model, and belongs to the technical field of dialect discussion models. Background With the continuous development of science and technology, dialectics have become an important research direction in the field of artificial intelligence, and in a liver disease consultation system of medicine, dialectics are usually performed in a dialogue manner, and each doctor proposes his own perspective for a certain diagnosis or treatment scheme. Through natural language understanding, the evidence can be converted into a dispute table, wherein disputes comprise evidence, conclusion, initial evaluation value of the evidence and rule relation and response strength of the evidence and the conclusion. The evidence of the dispute is taken as a child node, the conclusion is taken as a father node, and finally the whole debate scene can be converted into a dialogue tree. Whether each node in the dialogue tree is trusted depends on the initial evaluation value of the node and the evaluation of other evidence nodes on the node, so that the initial evaluation of the node is required to be fused with the objective evaluation of other nodes on the node to obtain a comprehensive evaluation value of all doctor views so as to judge whether the node is trusted. In practical debate, both the presentation of disputes and the relationship between disputes have uncertainty due to information uncertainty and imperfection, while DS evidence theory itself has the ability to deal with uncertainty problems, so that the theory can be well applied in debate discussion models. Although DS evidence theory has the ability to deal with uncertain problems, when the collision of various evidence sources is serious, the use of traditional DS evidence combination rules often results in an out-of-logic result, and even in some extreme cases, such as Zadeh counterexamples, the result of evidence fusion is contrary to the intuitive judgment of people. Therefore, aiming at the problem that the traditional DS evidence theory cannot deal with the defect of high conflict evidence, many students at home and abroad make improvements. The improved method mainly focuses on modifying the combination rules and modifying the evidence sources. Disclosure of Invention The invention aims to provide an improved DS evidence theory method based on conflict evidence weighting correction in a evidence source modification mode aiming at the defects of the existing method. The evaluation value of a node in a dialogue tree of the liver disease consultation system and the evaluation value of other related child nodes on the node are obtained. The recognition framework consists of a triplet including trusted, untrusted and unknowable propositions. Each proposition will have its own node and the evaluation value of the node by various other related child nodes. And taking all the evaluation values as basic probability distribution values, fusing by the DS evidence theory improvement method, and finally judging whether the node is credible or not by using a fusion result. In order to achieve the above purpose, the technical scheme of the invention is as follows: an improved DS evidence theory method based on liver disease consultation discussion model comprises the following steps: Step one, creating a liver disease consultation study model, determining an identification framework, converting a consultation scene into a dispute table, and creating a dialogue tree, wherein the specific contents are as follows Step 1.1, determining an identification framework Θ, wherein the identification framework power set 2 Θ = { Belief, disbelief, ignorance }, which is used by the invention, comprises three propositions of trust, distrust and unknowing, and respectively represents trust, distrust and unclear degree of a doctor on the statement. The evaluation value of each node is represented by a vector w= (b, d, i). And 1.2, converting all doctor views in the consultation scene into a dispute table through natural language understanding, wherein each dispute contains four parts of contents, namely evidence, conclusion, initial evaluation value of the evidence, rule relation and response strength of the evidence and the conclusion, taking the evidence of the dispute as a child node, and taking the conclusion as a father node to form a dialogue tree. The rule relation between evidence and conclusion is divided into four types: where p represents evidence, h represents a conclusion, the former two represent evidence or a negative support conclusion of evidence, and the latter two represent evidence or a negative objection conclusion of evidence. The intensity of the response of evidence and conclusions is expressed as The value range is 0-1. Step two, calculating the evaluation value