A digital syndrome differentiation system for Treatise on Febrile Diseases based on knowledge graph

Abstract

The present application relates to a digital syndrome differentiation system of Treatise on Febrile Diseases based on knowledge graph. The system includes a knowledge layer, a syndrome differentiation layer and a user layer; the knowledge layer is used to extract the five aspects of knowledge of Treatise on Febrile Diseases, such as disease, syndrome, symptom, method and prescription, from the original Treatise on Febrile Diseases and related works under the guidance of experts in classical prescriptions, and to construct a knowledge graph of Treatise on Febrile Diseases; the syndrome differentiation layer is used to construct a syndrome differentiation knowledge base of Treatise on Febrile Diseases based on the knowledge graph of Treatise on Febrile Diseases in the knowledge layer, to perform knowledge reasoning on weighted symptom vectors, and to generate a syndrome differentiation result set; the user layer is used to normalize the symptom questionnaire filled out by the user, to form a weighted symptom vector, and to return the syndrome differentiation result set to the user after sorting it according to credibility. The digital syndrome differentiation system of Treatise on Febrile Diseases constructed based on knowledge graph, credibility reasoning, generative reasoning and other technologies has a high accuracy and execution efficiency in the digital syndrome differentiation of six meridian diseases and their variants, and provides a new idea for the development of the field of intelligent auxiliary diagnosis of Treatise on Febrile Diseases.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a digital syndrome differentiation system of Treatise on Febrile Diseases based on knowledge graph. Background Art

[0002] Since a large number of subjective qualitative words are used in syndrome differentiation theory and disease description, the results of Six Meridians syndrome differentiation are highly subjective. With the vigorous development of information science, especially artificial intelligence technology, expert systems based on intelligent reasoning have been deeply studied and widely used in the field of TCM auxiliary diagnosis and treatment. Summary of the invention

[0003] Based on this, it is necessary to provide a digital syndrome differentiation system of Treatise on Febrile Diseases based on knowledge graph to address the above technical issues.

[0004] A digital syndrome differentiation system for Treatise on Febrile Diseases based on knowledge graph, the system comprising:

[0005] The knowledge layer is used to construct a conceptual classification system and a conceptual relationship description framework of the Treatise on Febrile Diseases based on the knowledge of diseases, syndromes, symptoms, methods and prescriptions of the Treatise on Febrile Diseases extracted from the original Treatise on Febrile Diseases and related materials, and to construct a knowledge graph of the Treatise on Febrile Diseases based on the conceptual classification system and the conceptual relationship description framework of the Treatise on Febrile Diseases; it is also used to receive the syndrome differentiation results that meet the evaluation requirements from the syndrome differentiation layer, and add the syndrome differentiation results to the knowledge graph of the Treatise on Febrile Diseases.

[0006] The dialectical layer is used to build a Treatise on Febrile Diseases rule base based on the Treatise on Febrile Diseases knowledge graph, convert the weighted symptom vectors from the user layer into a weighted fact set, and add the weighted fact set to the comprehensive database, and use a credibility hybrid reasoning method to perform dialectical reasoning based on the weighted fact set and the Treatise on Febrile Diseases rule base, and perform quality assessment on the dialectical result set obtained by reasoning, and pass the dialectical results that meet the assessment requirements to the knowledge layer.

[0007] The user layer is used to realize the interaction between the user and the system, normalize the symptom questionnaire filled in by the user, form a weighted symptom vector and submit it to the dialectical layer, and receive the dialectical result set and reasoning explanation sequence returned by the dialectical layer, and present it to the user in a visual form.

[0008] The digital syndrome differentiation system of Treatise on Febrile Diseases based on knowledge graphs includes a knowledge layer, a syndrome differentiation layer and a user layer; the knowledge layer contains five aspects of knowledge (stored in the form of knowledge graphs) of Treatise on Febrile Diseases, including diseases, syndromes, symptoms, methods and prescriptions extracted from the original Treatise on Febrile Diseases, national planning textbooks, modern literature and other materials under the guidance of experts in classical prescriptions, and is responsible for constructing the knowledge graph of Treatise on Febrile Diseases; the syndrome differentiation layer constructs the syndrome differentiation knowledge base of Treatise on Febrile Diseases based on the knowledge graph of Treatise on Febrile Diseases in the knowledge layer, performs knowledge reasoning on weighted symptom vectors, and generates a syndrome differentiation result set; the user layer is responsible for normalizing the symptom questionnaires filled out by users, forming weighted symptom vectors, and returning the syndrome differentiation result set to users after sorting them according to credibility. The digital syndrome differentiation system of Treatise on Febrile Diseases, which is constructed based on artificial intelligence technologies such as knowledge graphs, credibility reasoning and generative reasoning, has a high accuracy and execution efficiency in the digital syndrome differentiation of six meridian diseases and their variants, and provides a new idea for the development of the field of intelligent auxiliary diagnosis of Treatise on Febrile Diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 It is a block diagram of the composition of the digital syndrome differentiation system of Treatise on Febrile Diseases based on the knowledge graph in one embodiment;

[0010] Figure 2 The present invention is a block diagram of the composition of the digital syndrome differentiation system of Treatise on Febrile Diseases based on the knowledge graph in one embodiment. DETAILED DESCRIPTION

[0011] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0012] In one embodiment, Figure 1 As shown, a digital syndrome differentiation system of Treatise on Febrile Diseases based on knowledge graph is provided, and the system includes:

[0013] The knowledge layer is used to construct the concept classification system and concept relationship description framework of the Treatise on Febrile Diseases based on the knowledge of diseases, syndromes, symptoms, methods and prescriptions of the Treatise on Febrile Diseases extracted from the original treatise and related materials, and to construct the knowledge graph of the Treatise on Febrile Diseases based on the concept classification system and concept relationship description framework of the Treatise on Febrile Diseases; it is also used to receive the syndrome differentiation results that meet the evaluation requirements from the syndrome differentiation layer, and add the syndrome differentiation results to the knowledge graph of the Treatise on Febrile Diseases.

[0014] Specifically, the data sources of the knowledge layer are the ancient book "Treatise on Febrile Diseases" of the Song Dynasty, "Treatise on Febrile Diseases with Annotations" by Liu Duzhou, the national planned textbook "Selected Readings from Treatise on Febrile Diseases (New Century 4th Edition)", the 2009 national standard "Terms of Clinical Diagnosis and Treatment of Traditional Chinese Medicine (Revised Edition)" and other materials. The knowledge extraction is carried out under the guidance of experts in the classic prescriptions of "Treatise on Febrile Diseases". Based on the concept classification system and concept relationship description framework of Treatise on Febrile Diseases, a top-down approach is used to construct the knowledge graph of Treatise on Febrile Diseases. Meeting the evaluation requirements means that the credibility of the inference conclusion (i.e., the result of syndrome differentiation) reaches or exceeds the threshold (for example, 0.8, and 100% credibility is 1).

[0015] The syndrome differentiation layer is used to build the Treatise on Febrile Diseases rule base based on the Treatise on Febrile Diseases knowledge graph, convert the weighted symptom vectors from the user layer into a weighted fact set, and add the weighted fact set to the comprehensive database. It also uses a credibility hybrid reasoning method to perform dialectical reasoning based on the weighted fact set and the Treatise on Febrile Diseases rule base, and conducts quality assessment on the dialectical result set obtained by reasoning, and passes the dialectical results that meet the assessment requirements to the knowledge layer.

[0016] Specifically, the syndrome differentiation layer is the core of the digital syndrome differentiation system of Treatise on Febrile Diseases based on the knowledge graph. This layer mainly completes three tasks: First, the construction of the knowledge base of Treatise on Febrile Diseases. This layer constructs a set of rules based on the knowledge graph of Treatise on Febrile Diseases in the knowledge layer, and calculates the evidence in each rule and the credibility value assigned to the rule itself; Second, knowledge reasoning. This layer converts the weighted symptom vector transmitted from the user layer into a set of weighted facts, adds it to the comprehensive database, and then uses the credibility hybrid reasoning method for dialectical reasoning. In this process, in order to speed up the pattern matching process and avoid the problem of matching combination explosion, a comparison table of necessary symptoms of the six meridians of Treatise on Febrile Diseases is constructed, and the selection of the hypothetical target in the reverse reasoning process is completed based on it; Third, knowledge update. This layer conducts a quality assessment on the syndrome differentiation result set obtained by the knowledge reasoning process, and passes the syndrome differentiation results with a quality score of not less than 80 to the knowledge layer in the form of a triple set to complete the dynamic incremental update of the knowledge graph of Treatise on Febrile Diseases.

[0017] The user layer is used to realize the interaction between the user and the system. It normalizes the symptom questionnaire filled out by the user, forms a weighted symptom vector and submits it to the dialectical layer. It also receives the dialectical result set and reasoning explanation sequence returned by the dialectical layer and displays them to the user in a visual form.

[0018] Specifically, the user layer mainly completes two tasks: First, under the guidance of the experts of Shanghan Lun Jingfang, this layer designs the user symptom questionnaire according to the symptom classification method of Shanghan Lun described in Table 1 to express the user's query request, and then normalizes the questionnaire to form a weighted symptom vector and submits it to the syndrome differentiation layer; Second, receive the syndrome differentiation results and their reasoning explanation sequence returned by the syndrome differentiation layer, and present them to the user in a visual form.

[0019] Table 1 Classification of symptoms in Treatise on Febrile Diseases

[0020]

[0021]

[0022] In order to facilitate the normalization of symptom names, user requests are submitted in the form of symptom questionnaires after data preprocessing. The sample of the symptom questionnaire is shown in Table 2. In the symptom questionnaire, each symptom comes from the symptom classification table in Table 1, so as to achieve the uniformity of symptom description. In addition, in order to accurately describe the severity of each symptom, four levels are set for it, namely mild, general, severe, and serious (with credibility values ​​of 0.2, 0.6, 0.8, and 1, respectively).

[0023] Table 2 Symptom questionnaire sample

[0024]

[0025]

[0026] Therefore, each symptom in the syndrome differentiation request is expressed in the form of (standard term, credibility value), and the syndrome differentiation request is expressed by the symptom vector and its weight vector. For example, the existing syndrome differentiation request is as follows: (pulse floating, general), (aversion to cold, more severe), (fever, more severe), (sweating, general), (nasal congestion, mild), then the symptom vector and its weight vector obtained after normalization processing are: q 症状 =(pulse floating, aversion to cold, fever, sweating, nasal congestion), q 可信度 =(0.6, 0.8, 0.8, 0.6, 0.2).

[0027] In the above-mentioned digital syndrome differentiation system of Treatise on Febrile Diseases based on knowledge graph, the system includes knowledge layer, syndrome differentiation layer and user layer; the knowledge layer contains five aspects of knowledge (stored in the form of knowledge graph) of Treatise on Febrile Diseases, such as disease, syndrome, symptom, method and prescription extracted from the original Treatise on Febrile Diseases, national planning textbooks, modern literature and other materials under the guidance of classical prescription experts, and is responsible for constructing the knowledge graph of Treatise on Febrile Diseases; the syndrome differentiation layer constructs the syndrome differentiation knowledge base of Treatise on Febrile Diseases based on the knowledge graph of Treatise on Febrile Diseases in the knowledge layer, performs knowledge reasoning on weighted symptom vectors, and generates a syndrome differentiation result set; the user layer is responsible for normalizing the symptom questionnaire filled out by the user, forming a weighted symptom vector, and returning the syndrome differentiation result set to the user after sorting according to credibility. The digital syndrome differentiation system of Treatise on Febrile Diseases constructed based on artificial intelligence technologies such as knowledge graph, credibility reasoning and generative reasoning has high accuracy and execution efficiency in digital syndrome differentiation of six meridian diseases and their variants, and provides a new idea for the development of intelligent auxiliary diagnosis field of Treatise on Febrile Diseases.

[0028] In one of the embodiments, the conceptual classification system of Treatise on Febrile Diseases includes the conceptual classification system of diseases, syndromes, symptoms, methods, and prescriptions of Treatise on Febrile Diseases; the conceptual relationship description framework is based on the conceptual classification system of Treatise on Febrile Diseases, and uses conceptual relationships and attribute relationships to define the relationships between concepts for description, wherein the conceptual relationship is represented in the form of a triple <concept A, relationship R, concept B>, and the attribute relationship is represented in the form of a triple <concept A, attribute B, attribute value C>, and the type of the attribute value is specified to be a number or a string.

[0029] The knowledge base of Treatise on Febrile Diseases is constructed based on the knowledge graph of Treatise on Febrile Diseases. The construction of the antecedents and consequents of the rules in the knowledge base depends on the concepts and entities of the five aspects of disease, syndrome, symptom, method and prescription in the knowledge graph. To this end, it is necessary to establish the concept set of the five aspects of disease, syndrome, symptom, method and prescription.

[0030] Extract 15 disease names from the original text of Treatise on Febrile Diseases (i.e., disease set D = {d 1 , d 2 , d 3 , …, d 15}), 299 syndrome types (i.e., syndrome type set P = {p 1 , p 2 , p 3 ,…,p 299}), 561 symptoms (i.e., symptom set S = {s 1 ,s 2 ,s 3 ,…,s 561}), 12 treatment methods (i.e., the treatment method set T = {t 1 , t 2 , t 3 ,…,t 12}), 112 prescriptions (i.e., prescription set F = {f 1 , f 2 , f 3 , …, f 112}).

[0031] In one embodiment, if Figure 2 As shown, the syndrome differentiation layer includes a rule base construction module, a comprehensive database construction module, an inference engine, an interpreter, and a quality assessment module; the rule base construction module is used to construct and store the Treatise on Febrile Diseases rule base based on the Treatise on Febrile Diseases knowledge graph; the Treatise on Febrile Diseases rule base stores all the rules of the Treatise on Febrile Diseases, including the six-channel disease outline syndrome reasoning rules, prescription syndrome reasoning rules, treatment method prohibited misuse reasoning rules, and common knowledge reasoning rules.

[0032] Since the 398 original articles in the middle 10 chapters of "Treatise on Febrile Diseases" discuss how to diagnose a certain soup syndrome according to the patient's current symptoms, the number of production rules extracted and formulated by these articles accounts for a high proportion of the entire knowledge base. On the other hand, in the process of syndrome differentiation, the contribution of each symptom to syndrome differentiation is different. Some symptoms reflect the pathological essence of this stage, while some symptoms have little effect on the identification of syndrome types. Therefore, for the original content of syndrome differentiation involved in the knowledge extraction process of the original text of "Treatise on Febrile Diseases", this embodiment ignores those minor symptoms (called concurrent symptoms) and only retains those symptoms that mainly manifest the current condition and reflect the pathological essence of the current stage (called main symptoms). For example, Article 12 of the original text of "Treatise on Febrile Diseases" states, "When the sun is hit by a stroke, the yang is floating and the yin is weak. If the yang is floating, the heat will be spontaneous; if the yin is weak, the sweat will come out. If the patient is afraid of cold, afraid of wind, has a fever, and has nasal congestion and dry retching, Guizhi Decoction is the main one." From the original text knowledge extraction method, we can get symptoms such as floating and slow pulse, aversion to cold, aversion to wind, fever, nasal congestion, and dry retching. However, since the symptoms of nasal congestion and dry retching are not the pathological essence of solar stroke syndrome, the corresponding rules of this original text will be described in the following form:

[0033] (type? x syndrome), (manifestationOf sweating? x), (manifestationOf fever? x), (manifestationOf aversion to cold? x), (manifestationOf aversion to wind? x), (manifestationOf headache? x), (manifestationOf floating pulse? x)=>(isPattern? x sunstroke syndrome)

[0034] In this rule, the symptoms of nasal congestion and retching are omitted, and the main symptoms such as floating pulse, aversion to cold, aversion to wind, and fever are used to construct the antecedent of the rule. By eliminating the interference of secondary symptoms in syndrome differentiation, this reasoning method that relies on main symptoms to identify syndrome types can significantly reduce the complexity of digital syndrome differentiation.

[0035] The comprehensive database building module is used to transform the weighted symptom vectors from the user layer into weighted fact sets and store them in the comprehensive database.

[0036] The inference engine is used to perform dialectical reasoning based on the weighted fact set in the comprehensive database and the rule base of Treatise on Febrile Diseases, adopt a credibility hybrid reasoning method, obtain a dialectical result set, and store the new facts obtained by reasoning in the comprehensive database.

[0037] The interpreter is used to record the reasoning process of the inference engine and feed it back to the user layer.

[0038] The quality assessment module is used to conduct quality assessment on the dialectical result set and pass the dialectical results that meet the assessment requirements to the knowledge layer.

[0039] In one of the embodiments, the facts in the antecedents and consequents of the rules in the Treatise on Febrile Diseases rule base are represented in the form of quadruple with credibility values, and the rules themselves are represented in the form of productions with credibility values; the quadruple with credibility values ​​is in the form of: a quadruple consisting of object 1, relationship, object 2 and credibility value, or a quadruple consisting of object, attribute, attribute value and credibility value.

[0040] In one of the embodiments, the rule for determining the credibility value in the quadruple with credibility value representing the fact is: when the fact is a fact from the knowledge graph of Treatise on Febrile Diseases, its credibility value is set to 1; when the fact is a new fact generated in the reasoning process, if the new fact is an initial fact, the credibility value of the fact is given by the user who provides the evidence; if the new fact is an intermediate fact, the credibility of the fact is the credibility of the intermediate conclusion; if the new fact is the negation of the original fact, the credibility of the new fact is the opposite of the credibility of the original fact; if the new fact is a conjunctive fact, the credibility of the new fact is the minimum value of the credibility of all sub-facts, and if the new fact is a disjunctive fact, the credibility of the new fact is the maximum value of the credibility of all sub-facts.

[0041] In one embodiment, the rules in the Shanghan Lun rule base are expressed as follows:

[0042] E=>F CF(H,E)

[0043] Among them: E is the premise evidence, F is the knowledge conclusion, CF(H,E) is the rule credibility, and E and F are facts represented by four-tuples with credibility values.

[0044] In one embodiment, the inference engine includes a pattern matching module, a rule selection module and a rule execution module; the pattern matching module is used to use the Rete algorithm and the symptom vector memory learning method according to the weighted fact set from the user layer, and construct a comparison table of necessary symptoms for the six meridian diseases to significantly narrow the search space, match the weighted fact set with the antecedent of each rule in the rule library, and obtain the currently available rule set; the rule selection module is used to sort the currently available rule set in descending order according to the rule credibility value, and for several rules with equal rule credibility values, sort them in descending order according to the size of their evidence credibility values ​​to obtain a multi-level queue of available rule sets; the rule execution module is used to take out the queue head rule with the highest priority from the multi-level queue of available rule sets, execute the rule, obtain the inference result, and process according to the inference result to obtain the dialectical result set.

[0045] Specifically, efficient pattern matching algorithms are the core of production rule engines. Rete algorithm is a very popular pattern matching algorithm. Its main idea is to organize the antecedents of productions into a discriminant network for pattern matching, so as to achieve the effect of trading space for time. Since the pattern matching process uses the discriminant network to match the facts in the existing fact base with the antecedents of each rule in the rule base, the matching efficiency decreases significantly with the increase of facts and rules. In order to reduce the impact of the scale of facts and rules on the pattern matching efficiency, and to further speed up the matching process, this embodiment adopts memory learning and the comparison table of necessary symptoms of six meridian diseases to achieve this purpose.

[0046] Since the rule credibility reflects the diagnosis and treatment methods of "Treatise on Febrile Diseases" for a certain disease, it is directional and holistic compared to the evidence in the rule antecedent, so it is more important than the evidence credibility and has a higher priority. Therefore, this paper will sort the currently available rule set in descending order according to the rule credibility value. For several rules with equal rule credibility values, they will be sorted in descending order according to the size of their evidence credibility values, so as to complete the priority sorting of the entire available rule set. The detailed description of the rule selection method based on credibility is as follows.

[0047] The construction of the multi-level queue of available rule sets can be divided into the initial stage and the running stage. In the initial stage, the available rule set is obtained by pattern matching based on the user request symptom vector and its weight vector, so it is necessary to create a multi-level queue of available rule sets from scratch; in the running stage, every time a rule is taken out from the current rule set for execution, a new fact set can be inferred, and a new rule set can be obtained by pattern matching based on it, so the original multi-level queue of available rule sets needs to be updated.

[0048] Initial stage:

[0049] Calculate the evidence credibility value CF(E) of each rule in the currently available rule set, and remove the rules whose CF(E) is less than the dialectical result threshold ε, and obtain the rule set R current = {r 1 , r 2 , r 3 , ..., r t};

[0050] According to the order of the rule credibility values ​​CF(H, E), based on R current Rule set to establish multi-level queue Q current Rules in the same queue have the same CF(H, E), and queues are sorted from high to low according to CF(H, E);

[0051] For multi-level queues Q current For each queue, sort the rules in the queue in descending order according to the evidence credibility value CF(E). The algorithm ends.

[0052] Operation phase:

[0053] Calculate the evidence credibility value CF(E) of each rule in the new rule set, and remove the rules whose CF(E) is less than the dialectical result threshold ε, and obtain the rule set R new = {r 1 , r 2 , ..., r k};

[0054] From the rule set R new Take out a rule in turn, insert it into a queue according to the credibility CF(H, E) of the rule (if there is no queue with CF(H, E) as the credibility value of the rule to be inserted in the existing multi-level queue, create a new queue), and then insert it into the correct position of the queue according to the evidence credibility CF(E) of the rule, so that the queue elements after insertion are still arranged in order according to CF(E). Repeat this step until the rule set R new Is empty. The algorithm ends.

[0055] When the available rule set multi-level queue Q current After the construction is completed, the first rule of the queue with the highest priority is taken out each time the rule is executed, and the rule is added to the historical rule set.

[0056] In one of the embodiments, processing is performed according to the inference result to obtain a dialectical result set, including: if the inference result is a new fact, then checking whether the fact is a certificate name or disease name whose credibility is not less than the dialectical result threshold ε, if so, adding the certificate name or disease name to the dialectical result set; otherwise, adding the new fact to the comprehensive database, and matching the new fact with the rule base, and updating the available rule set; if the inference result is not a new fact, then judging whether the current available rule set multi-level queue is empty; if the rule set multi-level queue is not empty, then continuing to take out the next priority rule from the available rule set multi-level queue, and executing it; if the rule set multi-level queue is empty, checking whether the dialectical result set is empty, if so, taking all other rules in the original rule base that do not contain the current available rule set as a new production pattern matching rule set, and matching the new production pattern matching rule set; if not, returning the dialectical result set, and the inference ends.

[0057] In one of the embodiments, a comparison table of necessary symptoms for six meridian diseases is constructed to significantly narrow the search space, including: dividing the entire rule base into a main syndrome rule set and a variant syndrome rule set, wherein the main syndrome rule set is further divided into a Taiyang disease rule set, a Yangming disease rule set, a Shaoyang disease rule set, a Taiyin disease rule set, a Shaoyin disease rule set, and a Jueyin disease rule set; based on the main pulse symptoms and outline symptoms of the six meridian diseases, 2 to 3 symptoms that are representative and statistically significant for each meridian syndrome are extracted as a guide for heuristic search, and a comparison table of necessary symptoms for six meridian diseases is constructed based on this; based on the current requested symptoms, the comparison table of necessary symptoms for six meridian diseases is constructed ... vector, look up the comparison table of necessary symptoms of six-channel disease syndromes, and determine whether the current request may be the main syndrome or a variant syndrome; if the comparison table of necessary symptoms of six-channel disease syndromes determines that the syndrome is the main syndrome, the rule set corresponding to the main syndrome is used as the rule matching range of the Rete algorithm; otherwise, the rule set corresponding to the variant syndrome is used as the rule matching range of the Rete algorithm; in the above two cases, if the syndrome differentiation result with a credibility greater than the threshold set by the system cannot be inferred, the rule set that has not yet participated in the pattern matching in the above case is used as the rule matching range of the next round of the Rete algorithm, and the Rete algorithm is executed again for production pattern matching.

[0058] Specifically, the comparison table of necessary symptoms for six meridian diseases is shown in Table 3.

[0059] Table 3 Comparison table of necessary symptoms of six meridian diseases

[0060]

[0061] In one of the embodiments, the symptom vector memory learning method achieves the purpose of learning by memorizing and evaluating the information provided by the external environment; the query request received by the user layer and its dialectical result are stored in the knowledge base, and when the same query request is received in the future, the dialectical result can be directly retrieved from the knowledge base without having to reason and solve again; whenever a new query symptom vector is received, the similarity between the current request symptom vector and the historical symptom vector is calculated; if the obtained similarity is greater than a preset threshold, the dialectical result corresponding to the currently compared historical symptom vector is stored in the candidate result set; otherwise, the similarity between the current request symptom vector and the next historical symptom vector is continued to be calculated until all historical symptom vectors are scanned; the similarity calculation formula is:

[0062]

[0063] Where a represents the request symptom vector, and its vector a element includes two parts. One part is the set of vector elements that are the same in vector a and vector b, denoted as (a 1 , a 2 , ..., a t ), and the other part is the set of vector elements that are different between vector a and vector b, denoted as (at+1 , a t+2 , ..., a m ), the weight vectors corresponding to these two parts are normalized to (w a1 , w a2 , ..., w at ) and (w a(t+1) , w a(t+2) , ..., w am ); b represents the historical symptom vector, where the vector b element consists of two parts, one of which is the set of vector elements that are the same as those in vector b and vector a, denoted as (b 1 , b 2 , ..., b t ), the other part is the set of vector elements that are different between vector b and vector a, denoted as (b t+1 , b t+2 , ..., b n ), (w b1 , w b2 ,…,w bt ) and (w b(t+1) , w b(t+2) , ..., w bn ) is the normalization result of the weight vectors corresponding to the two parts of the b vector; Represents a vector (w a1 , w a2 , ..., w at ) and (w b1 , w b2 , ..., w bt ) between ; is a vector (w a(t+1) , w a(t+2) , ..., w am ) and (wb (t+1) , w b(t+2) , ..., w bn ) is the sum of the harmonic means of .

[0064] If the credibility of the current reasoning dialectical result is greater than the preset threshold, the dialectical result and its symptom vector will be stored in the knowledge base. The scale of the entire historical result is 1 / 10 of the number of rules in the knowledge base, and it is sorted according to the time of most recent use, so that the most recent symptom vector request and its dialectical result are ranked first, and the historical symptom vector and its dialectical result that have not been used for more than the preset time length will be removed from the memory bank.

[0065] For example, the existing symptom vector a = (headache, fever, chills, floating pulse, nasal congestion, vomiting), its weight vector w a=(0.8, 0.8, 0.8, 1, 0.2, 0.2); symptom vector b = (headache, fever, chills, floating pulse, cough), its weight vector w b =(0.8, 0.8, 1, 0.8, 0.2). Vector w a The normalized values ​​are (0.2105, 0.2105, 0.2105, 0.2632, 0.0526, 0.0526), ​​and the vector w b The normalized values ​​are (0.2222, 0.2222, 0.2778, 0.2222, 0.0556), and then the similarity between symptom vector a and symptom vector b is calculated according to formula 4-3, and the value of S(a, b) is 0.8261. This value is greater than the symptom vector similarity threshold δ (δ = 0.8), so it can be considered that symptom vector a and symptom vector b are essentially the same, and the conditions they describe are basically the same.

[0066] Specifically, memory learning is also called mechanical learning, which achieves the purpose of learning by memorizing and evaluating the information provided by the external environment. In order to improve the vector matching speed, both the current request symptom vector and the historical symptom vector stored in the knowledge base are arranged in the order of Chinese pinyin.

[0067] The working process of the digital syndrome differentiation system of Shanghan Lun based on knowledge graph: after receiving the user's request symptom vector and its weight vector, the system initializes the comprehensive database accordingly, and searches the historical symptom vector set based on the symptom vector memory learning method. If the current request symptom vector is already in the historical symptom vector set, it means that the system has processed the request (or a request with extremely high similarity), and then the historical reasoning results (several syndrome differentiation results) are returned to the user layer; otherwise, starting from the comprehensive database, generative reasoning is performed based on the credibility and hybrid reasoning strategies. The specific steps are as follows:

[0068] ① Perform inclusion operation on the facts in the comprehensive database and the comparison table of necessary symptoms of six meridian diseases and syndromes, determine the type of the syndrome (one or more) or variant syndrome that may be inferred by the current request, and use the rule set of the corresponding category as the production pattern matching rule set;

[0069] ②Build a Rete network based on the above production pattern matching rule set;

[0070] ③Perform pattern matching according to the Rete algorithm and create (or update) the available rule set;

[0071] ④ Use a credibility-based rule selection method to sort the currently available rule sets and create (or update) a multi-level queue of available rule sets;

[0072] ⑤ Take the queue-first rule with the highest priority from the available rule set multi-level queue, execute the rule, and process the inference result as follows:

[0073] If the inference result is not a new fact, then determine whether the multi-level queue of the currently available rule set is empty. If it is not empty, go to step ⑤; otherwise, it means that there are no available rules and the inference cannot be performed, so go to step ⑥.

[0074] If the inference result is a new fact, check whether the fact is a syndrome name (or disease name) with a credibility not less than the syndrome differentiation result threshold ε. If so, add the syndrome name (or disease name) to the syndrome differentiation result set; otherwise, add the new fact to the comprehensive database and go to ③ to update the available rule set.

[0075] ⑥ Check whether the dialectical result set is empty. If so, it means that the current production pattern matching rule set cannot deduce the dialectical result, so all other rules in the original knowledge base that do not contain the current production pattern matching rule set are used as new production pattern matching rule sets, and go to ②; if not, it means that the dialectical result has been derived based on the current production pattern matching rule set, so the dialectical result set is returned and the reasoning ends.

[0076] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0077] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A digital syndrome differentiation system of Treatise on Febrile Diseases based on knowledge graph, characterized in that: The system comprises: The knowledge layer is used to construct the concept classification system and concept relationship description framework of the Treatise on Febrile Diseases based on the knowledge of diseases, syndromes, symptoms, methods and prescriptions of the Treatise on Febrile Diseases extracted from the original Treatise on Febrile Diseases and related materials, and to construct the knowledge graph of the Treatise on Febrile Diseases based on the concept classification system and concept relationship description framework of the Treatise on Febrile Diseases; it is also used to receive the syndrome differentiation results that meet the evaluation requirements from the syndrome differentiation layer, and add the syndrome differentiation results to the knowledge graph of the Treatise on Febrile Diseases; The syndrome differentiation layer is used to construct a rule base of the Treatise on Febrile Diseases based on the Treatise on Febrile Diseases knowledge graph, convert the weighted symptom vectors from the user layer into a weighted fact set, add the weighted fact set into the comprehensive database, and perform syndrome differentiation reasoning using a credibility hybrid reasoning method based on the weighted fact set and the Treatise on Febrile Diseases rule base, and perform quality assessment on the syndrome differentiation result set obtained by reasoning, and pass the syndrome differentiation results that meet the assessment requirements to the knowledge layer; The user layer is used to realize the interaction between the user and the system, normalize the symptom questionnaire filled in by the user, form a weighted symptom vector and submit it to the syndrome differentiation layer, and receive the syndrome differentiation result set and reasoning explanation sequence returned by the syndrome differentiation layer, and display them to the user in a visual form; The dialectical layer includes: an inference engine, which is used to perform dialectical reasoning based on the weighted fact set in the comprehensive database and the Shanghan Lun rule base, using a credibility hybrid reasoning method to obtain a dialectical result set, and store the new facts obtained by reasoning in the comprehensive database; The inference engine includes: a pattern matching module, which is used to use the Rete algorithm and the symptom vector memory learning method according to the weighted fact set from the user layer, and to construct a comparison table of necessary symptoms of six meridian diseases to significantly narrow the search space, and to match the weighted fact set with the antecedent of each rule in the Shanghan Lun rule base to obtain the current available rule set; Among them, a comparison table of necessary symptoms of six meridian diseases is constructed to significantly narrow the search space, including: The entire rule base is divided into the main syndrome rule set and the variable syndrome rule set, wherein the main syndrome rule set is further divided into the Taiyang disease rule set, Yangming disease rule set, Shaoyang disease rule set, Taiyin disease rule set, Shaoyin disease rule set and Jueyin disease rule set; According to the main pulse symptoms and outline symptoms of the six meridian diseases, 2 to 3 symptoms that are representative and statistically significant for each meridian syndrome are extracted as a guide for heuristic search, and a comparison table of necessary symptoms for the six meridian diseases is constructed based on this. According to the current request symptom vector, search the necessary symptom comparison table of six meridian diseases and syndromes to determine whether the current request may be the main syndrome or a variant syndrome; If the result of the Six-Channel Disease Necessary Symptom Comparison Table is the current syndrome, the rule set corresponding to the current syndrome is used as the rule matching range of the Rete algorithm; otherwise, the rule set corresponding to the variant syndrome is used as the rule matching range of the Rete algorithm; In the above two cases, if no dialectical result with a credibility greater than the threshold set by the system can be inferred, the rule set that has not yet participated in the pattern matching in the above case will be used as the rule matching range of the next round of Rete algorithm, and the Rete algorithm will be executed again for production pattern matching.

2. The system according to claim 1, characterized in that The concept classification system of Treatise on Febrile Diseases includes the concept classification system of diseases, syndromes, symptoms, methods and prescriptions of Treatise on Febrile Diseases; The concept relationship description framework is based on the concept classification system of Treatise on Febrile Diseases. It uses concept relationship and attribute relationship to define the relationship between concepts. The concept relationship is represented by a triple <concept A, relationship R, concept B>, and the attribute relationship is represented by a triple <concept A, attribute B, attribute value C>, and the type of the attribute value is specified as a number or a string.

3. The system according to claim 1, characterized in that The dialectical layer also includes a rule base construction module, a comprehensive database construction module, an interpreter, and a quality assessment module; The rule base construction module is used to construct and store the rule base of Treatise on Febrile Diseases based on the knowledge graph of Treatise on Febrile Diseases; the rule base of Treatise on Febrile Diseases stores all the rules of Treatise on Febrile Diseases, including the reasoning rules of the outline of six meridian diseases, the reasoning rules of prescriptions and syndromes, the reasoning rules of the prohibited and misused treatment methods, and the reasoning rules of common knowledge; The comprehensive database construction module is used to convert the weighted symptom vectors from the user layer into a weighted fact set and store it in the comprehensive database; The interpreter is used to record the reasoning process of the inference engine and feed it back to the user layer; The quality assessment module is used to perform quality assessment on the dialectical result set and pass the dialectical results that meet the assessment requirements to the knowledge layer.

4. The system according to claim 3, characterized in that The facts in the rule antecedents and consequents in the Shanghan Lun rule base are expressed in the form of four-tuples with credibility values, and the rules themselves are expressed in productions with credibility values; The four-tuple with the credibility value is in the form of a four-tuple consisting of object 1, relationship, object 2 and credibility value, or a four-tuple consisting of object, attribute, attribute value and credibility value.

5. The system according to claim 4, characterized in that The determination rule of the credibility value in the quadruple with credibility value representing the fact is: When the fact is a fact from the Shanghan Lun knowledge graph, its credibility value is set to 1; When a fact is a new fact generated in the reasoning process, if the new fact is an initial fact, the credibility value of the fact is given by the user who provides the evidence; if the new fact is an intermediate fact, the credibility of the fact is the credibility of the intermediate conclusion; if the new fact is the negation of the original fact, the credibility of the new fact is the opposite of the credibility of the original fact; if the new fact is a conjunctive fact, the credibility of the new fact is the minimum value of the credibility of all sub-facts; if the new fact is a disjunctive fact, the credibility of the new fact is the maximum value of the credibility of all sub-facts.

6. The system according to claim 3, characterized in that The representation form of the rules in the Shanghan Lun rule base is: E=>F CF(H,E) Among them: E is the premise evidence, F is the knowledge conclusion, CF(H,E) is the rule credibility, and E and F are facts represented by four-tuples with credibility values.

7. The system according to claim 1, characterized in that The inference engine also includes a rule selection module and a rule execution module; The rule selection module is used to sort the currently available rule set in descending order according to the rule credibility value, and for several rules with equal rule credibility values, sort them in descending order according to the size of their evidence credibility values ​​to obtain a multi-level queue of available rule sets; The rule execution module is used to take out the queue head rule with the highest priority from the multi-level queue of the available rule set, execute the rule, obtain the inference result, and process according to the inference result to obtain the dialectical result set.

8. The system according to claim 7, characterized in that Processing is performed according to the inference results to obtain a dialectical result set, including: If the inference result is a new fact, then check whether the fact is a syndrome name or disease name whose credibility is not less than the syndrome differentiation result threshold ε. If so, then add the syndrome name or disease name to the syndrome differentiation result set; otherwise, add the new fact to the comprehensive database, and match the new fact with the rule base to update the available rule set; If the reasoning result is not a new fact, determine whether the currently available rule set multi-level queue is empty; if the rule set multi-level queue is not empty, continue to take out the next priority rule from the available rule set multi-level queue and execute it; if the rule set multi-level queue is empty, check whether the dialectical result set is empty. If so, all other rules in the original rule base that do not contain the currently available rule set are used as new production pattern matching rule sets, and rule matching is performed on the new production pattern matching rule set; if not, return the dialectical result set and the reasoning ends.

9. The system according to claim 7, characterized in that The symptom vector memory learning method achieves the purpose of learning by memorizing and evaluating the information provided by the external environment; The query request and its identification result received by the user layer are stored in the knowledge base. When the same query request is received in the future, the identification result can be directly retrieved from the knowledge base without having to reason and solve again. Whenever a new query symptom vector is received, the similarity between the current request symptom vector and the historical symptom vector is calculated; if the obtained similarity is greater than the preset threshold, the dialectical result corresponding to the currently compared historical symptom vector is stored in the candidate result set; otherwise, the similarity between the current request symptom vector and the next historical symptom vector is continued to be calculated until all historical symptom vectors are scanned; the similarity calculation formula is: Where a represents the request symptom vector, and its vector a element consists of two parts. One part is the set of vector elements that are the same as those in vector a and vector b, denoted as (a1, a2, …, a t ), and the other part is the set of vector elements that are different between vector a and vector b, denoted as (a t+1 ,a t+2 ,…,a m ), the weight vectors corresponding to these two parts are normalized to (w a1 ,w a2 ,…,w at ) and (w a(t+1) ,w a(t+2) ,…,w am ); b represents the historical symptom vector, where the vector b element consists of two parts, one of which is the set of vector elements that are the same as those in vector b and vector a, denoted as (b1, b2, …, b t ), the other part is the set of vector elements that are different between vector b and vector a, denoted as (b t+1 ,b t+2 ,…,b n ), (w b1 ,w b2 ,…,w bt ) and (w b(t+1) ,w b(t+2) ,…,w bn ) is the normalization result of the weight vectors corresponding to the two parts of the b vector; Represents a vector (w a1 ,w a2 ,…,w at ) and (w b1 ,w b2 ,…,w bt ) between ; Vector (w a(t+1) ,w a(t+2) ,…,w am ) and (w b(t+1) ,w b(t+2) ,…,w bn ) is the sum of the harmonic means of If the credibility of the current reasoning dialectical result is greater than the preset threshold, the dialectical result and its symptom vector will be stored in the knowledge base. The scale of the entire historical result is 1 / 10 of the number of rules in the knowledge base, and it is sorted according to the time of most recent use, so that the most recent symptom vector request and its dialectical result are ranked first, and the historical symptom vector and its dialectical result that have not been used for more than the preset time length will be removed from the memory bank.

Images