Method for extracting disease and symptoms and inducements thereof in medical text

Abstract

The invention discloses a method for extracting a disease and symptoms and inducements thereof from a medical text. The method comprises the following steps: acquiring the medical text; extracting symptom group information and inducement chain information from the medical text and outputting the information; performing sentence pattern rule matching on all sentences in the medical text according to a preset sentence pattern rule, and segmenting the sentences meeting the preset sentence pattern rule to obtain a clause group containing effective information and a clause group not containing the effective information; sentence pattern rule matching is carried out on clauses in the clause group containing the effective information, information extraction is carried out on each matched clause, and target information of symptom expression and inducement is extracted; selecting target information meeting an impurity removal rule from the extracted target information according to the impurity removal rule; according to a marking rule, a subject is matched for each clause in the target information conforming to the impurity removal rule; and outputting the subject and the target information conforming to the impurity removal rule. The method is high in adaptability to processing complex texts and high in information extraction precision.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular to a method for automatically extracting a disease and its symptom manifestations and inducements. Background Art

[0002] With the advancement of natural language processing technology, medical texts contain a large number of descriptions of diseases, symptoms, causes, etc. How to effectively extract relevant information from these unstructured texts has become one of the important issues to improve the intelligence and automation of medical information systems. At present, information extraction methods usually rely on rule matching, keyword recognition and pattern analysis. These methods have shown good results when processing structured or standardized medical texts due to their high interpretability and efficiency. However, these methods often face the challenges of insufficient extraction accuracy and coverage when faced with the complexity and variability of texts.

[0003] With the continuous advancement of natural language processing (NLP) technology, deep learning-based models (such as BERT, GPT, etc.) are gradually being applied to medical information extraction tasks. These models can extract more detailed and comprehensive features with the support of large-scale corpora by automatically learning text semantics, significantly improving recognition effects. However, in actual medical scenarios, due to the lack of sufficient annotated training data, deep learning-based extraction models still face challenges in terms of accuracy and generalization, especially when dealing with rare cases or field-specific medical terms, deep learning models often show weak adaptability. Therefore, in actual scenarios with limited data and complex medical texts, rule matching methods are still effective and operational technical means that can efficiently extract disease and symptom information under specific conditions.

[0004] In medical applications, rule matching technology can effectively extract structured information through pre-defined rules and keywords, especially when dealing with more standardized medical texts. However, existing rule matching methods still face problems such as insufficient extraction accuracy and incomplete information coverage when dealing with long texts, complex medical terminology combinations, and disease descriptions under multiple conditions. In addition, some descriptions in medical texts involve ambiguous medical terms, or require a combination of multiple conditions to judge the disease and its symptoms, which further increases the complexity of rule matching.

[0005] The invention with application number 202210500531.1 discloses a method and device for joint extraction of medical entities and relationships based on medical records, the method comprising: semantically segmenting the medical record text to obtain multiple texts to be recognized; sequentially combining the segmented words in the multiple texts to be recognized to obtain the sequence of words to be recognized of the multiple texts to be recognized; the sequence of words to be recognized is composed of one segmented word or multiple continuous segmented words in the corresponding text to be recognized; using the entity relationship joint extraction model, medical entity recognition is performed on the sequence of words to be recognized of any text to be recognized, and multiple medical entities and entity types of multiple medical entities are obtained, and after determining the head entity and the tail entity of the two medical entities in the text to be recognized, based on the entity representation of the head entity and the tail entity, the nearest tail entity distance and the tail entity interval, it is determined whether there is an entity relationship to be confirmed between the two medical entities. The above invention uses semantic segmentation technology to segment medical record texts, but when dealing with complex medical texts, especially texts containing multiple diseases, symptoms and their causes, the semantic segmentation method is difficult to fully capture these complex relationships, which in turn causes information loss or context confusion. In addition, the word segmentation sequence combination method fails to fully solve the ambiguity of terms or polysemous words, which leads to entity recognition errors or relationship misjudgment. When using an entity relationship joint extraction model to identify medical entities and their relationships, the model's adaptability will be insufficient when faced with complex medical terms or descriptions of new diseases, and the accuracy of the model will also be affected when training data is scarce. The recognition method based on the relationship between the head entity and the tail entity is difficult to handle complex sentence structures, especially when it involves multiple disease descriptions or unclear relationships between symptoms. In addition, the generalization ability of the model is limited by the breadth of the training data. For rare diseases or symptoms not covered in the training set, the model's recognition and extraction capabilities will be limited. Summary of the invention

[0006] The present invention proposes a method for extracting diseases and their symptoms and causes from medical texts to solve the technical problem that existing methods have insufficient extraction accuracy when processing long texts, complex medical term combinations and disease descriptions under multiple conditions.

[0007] In order to achieve the above object, the technical solution of the present invention is achieved as follows: A method for extracting diseases and their symptoms and causes from medical texts, comprising the following steps: S1. Obtain and store medical texts containing diseases and their related symptoms and causes; S2, extracting symptom group information and cause chain information from the medical text obtained in step S1 and outputting them; S3, according to the preset sentence rules, perform a first round of sentence rule matching on all the sentences in the medical text obtained in step S1, segment the sentences that meet the preset sentence rules, and obtain clause groups containing valid information and clause groups that do not contain valid information; S4, discarding the clause group that does not contain valid information, performing a second round of sentence rule matching on the clauses in the clause group that contains valid information according to the preset sentence rule, extracting information from each matched clause, and extracting target information on symptom manifestations and inducements; S5, selecting target information that meets the screening rules from the extracted target information according to the screening rules; S6, matching a subject for each clause in the target information obtained in step S5 and meeting the screening rule according to the marking rule; S7, output the target information that meets the screening rules obtained in step S5 and the subject matched in step S6.

[0008] Preferably, the method for extracting symptom group information and causative chain information is: based on symptom group keywords and causative chain keywords, performing relationship analysis on sentences in the medical text, identifying the logical relationship between sentences, and obtaining and extracting symptom group and causative chain information.

[0009] Preferably, in the process of extracting symptom cluster information and causative chain information, if multiple consecutive sentences contain marker words of symptom clusters or causative chains, the multiple sentences are regarded as a whole and marked as symptom clusters or causative chains of the corresponding diseases.

[0010] Preferably, the method for segmenting sentences that conform to preset sentence rules is: using the punctuation mark closest to the sentence rule marker and the punctuation mark closest to the sentence rule marker as the segmentation boundary, and splitting the screened sentences that conform to the preset sentence rules.

[0011] Preferably, the method for extracting information from each matched clause is: when extracting information from each matched clause, a corresponding operation strategy is adopted according to the sentence pattern rule matched by the clause, as follows: If a clause matches only one sentence rule, information extraction is performed directly according to the sentence rule; If a clause matches multiple sentence rules at the same time, first, check whether there are the same sentence markers but different numbers of sentence markers in these matching rules. If so, discard the sentence rule with fewer sentence markers, and select the sentence rule with more sentence markers to take effect and extract the target information; secondly, if the sentence marker is an independent single sentence rule marker, multiple sentence rules are allowed to take effect at the same time; finally, when a sentence rule contains two or more sentence markers, determine whether the extraction information corresponding to the sentence markers contains an inclusion relationship, discard the sentence rules whose corresponding extraction information does not have an inclusion relationship, and select the sentence rules whose corresponding extraction information has an inclusion relationship to take effect and extract the target information.

[0012] Preferably, the extracted symptom manifestations and causes are simple entities or composite entities, the composite entity includes a simple entity and additional information, and the additional information includes a part word, a quantifier, an adverb of degree and an adverb of direction.

[0013] Preferably, the symptom manifestations include clinical symptoms at the onset of the disease and other information related to the disease manifestations, and the other information related to the disease manifestations include one or more combinations of the disease-affecting population, the onset time of the disease, the severity of the disease, the onset site of the disease, and complications.

[0014] Preferably, the subject is a literal word with a clear semantic expression function, and the subject is a disease or symptom manifestation.

[0015] Preferably, the method for matching the subject of each clause in the target information that meets the screening rules obtained in step S5 according to the marking rules is as follows: first, the positions of all disease entities in the medical text obtained in step S1 are marked to provide a reference for subject identification. Then, for each piece of screened target information, the text fragment before the sentence rule marker word is checked to determine whether the text fragment meets the conditions for being the subject of the target information; if it does, this text fragment is determined to be the subject of the target information; if it does not, according to specific rules, the disease entity closest to the current sentence rule marker word is searched among all previously marked disease entities as the subject of the target information. If the sentence rule requires that the subject be a symptom manifestation or a inducement, and a qualifying subject cannot be found in the current clause, a qualifying text fragment is searched in the immediately preceding clause as the subject.

[0016] Preferably, the medical text is derived from professional medical textbooks and / or professional medical websites. One method for obtaining the medical text is: scanning and extracting professional medical textbooks using optical character recognition technology; collecting medical text from professional medical websites using web crawler technology.

[0017] Compared with the prior art, the present invention has the following beneficial effects: Improved accuracy and coverage of information extraction: The present invention combines the technical solutions of rule matching and multiple information screening. Through multiple rounds of matching and screening processing, it effectively reduces the extraction errors caused by the complexity of medical terminology and non-standard descriptions, and improves the accuracy and coverage of disease and symptom manifestation and cause information extraction.

[0018] The complexity of medical texts is fully considered: When processing disease, symptom and cause information, the present invention is not limited to simple entity extraction, but also specially designs the recognition and processing scheme for complex entities. Compared with the existing method of only recognizing simple entities, the present invention has more advantages in fine-grained medical information extraction.

[0019] Taking into account the accuracy and universality of data sources: Unlike methods that partially rely on electronic medical record data, the data of this invention comes from reviewed medical textbooks and professional medical websites, thereby reducing the influence of doctors' subjective descriptions and ensuring the authority and accuracy of the data. Therefore, this invention performs better in data credibility and universality, which helps to be more widely used in different medical scenarios.

[0020] Reduce the matching error rate: During the matching process, the present invention avoids the inaccuracy caused by crossing punctuation marks by matching only within a single independent sentence. By limiting the scope of the matching operation, the present invention effectively reduces the error rate in the matching process and improves the processing effect of long or complex sentences.

[0021] Flexible sentence pattern matching: For different text descriptions, the present invention supports the situation where a sentence matches multiple sentence pattern rules, and provides a further processing mechanism for overlapping marker words. This flexibility enables the system to adapt to diverse sentence expressions and improves the processing capability and adaptability of complex texts.

[0022] Efficient extraction of symptoms and causes: By adopting a step-by-step segmentation and multi-round screening method, the present invention achieves efficient information extraction in complex medical texts and ensures the integrity and relevance of the extracted information. Compared with traditional single-round matching or simple rules, the present invention can better retain the key details in the disease description and improve the practicality of the extraction results.

[0023] Combining the advantages of structured and unstructured text processing: When processing relatively standardized medical texts, the method of the present invention fully utilizes the efficiency and interpretability of rule matching; and when processing complex or non-standard texts, through multiple rounds of screening and composite entity extraction, it overcomes the shortcomings of traditional rule-based methods in terms of coverage and flexibility, and achieves a better overall effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0025] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] A method for extracting diseases, their symptoms and causes from medical texts, such as Figure 1 As shown, the following steps are included: S1. Acquire and store medical texts containing diseases and their related symptoms and causes.

[0028] The medical texts are derived from professional medical textbooks and professional medical websites rather than electronic medical record data. By using standardized data sources, the interference of doctors' subjectivity can be effectively reduced, thereby ensuring the accuracy, authority and consistency of the extracted data.

[0029] In this embodiment, the method for obtaining medical text is: using optical character recognition (OCR) technology to scan and extract professional medical textbooks, so as to obtain high-quality medical texts. Professional medical textbooks are rich in systematic medical knowledge and are an important source of information for the study of diseases and their related symptoms. Relevant medical information covering the latest disease diagnosis and treatment guidelines and clinical case analysis is collected from professional medical websites through web crawler technology to obtain high-quality medical texts. In order to ensure the accuracy and reliability of the data, the data obtained by the above method needs to be cleaned before use to remove noise and redundant information. The specific operations are: correcting OCR recognition errors, automatically correcting incorrect characters and terms using medical vocabulary; removing duplicate and redundant content through text similarity algorithm to ensure data uniqueness; at the same time, standardizing the text format, segmenting and marking different types of medical information; unifying medical terms to reduce ambiguity from texts from different sources; eliminating outdated or erroneous information to ensure that the data complies with the latest medical guidelines; desensitizing patient information involving privacy to ensure data security. After cleaning, the medical text is more consistent and reliable, laying a high-quality data foundation for subsequent information extraction and analysis.

[0030] S2. Extract symptom group information and causal chain information from the medical text obtained in step S1 and output them.

[0031] Furthermore, the symptom cluster refers to a collection of multiple symptoms that exist simultaneously and are interrelated, such as the physical symptom cluster of depression, which includes sleep disorders, appetite disorders, loss of energy, etc. The causal chain refers to a series of interrelated factors or events that lead to the occurrence of the disease. These factors or events may be connected to each other in chronological order or causally to form a complete chain, such as a cold leading to a fever, and a fever leading to a patient's headache.

[0032] The method for extracting symptom group information and causal chain information is as follows: based on symptom group keywords and causal chain keywords, the sentences in the medical text are analyzed for relationship, the logical association between the sentences is identified, and the symptom group and causal chain information is identified and extracted. In particular, when multiple consecutive sentences contain the marker words of the symptom group or causal chain, these sentences are regarded as a whole, and are marked as the symptom group or causal chain of the corresponding disease. For example, "mainly manifested as increased vaginal discharge, purulent, purulent or serous, with a fishy smell, and those with vaginitis often have vaginal pain" is a symptom group, which meets the continuous symptom group marker words "manifested as", "presented", "with", "accompanied" appearing in multiple sentences, and another example is "the patient's heart failure is caused by long-term uncontrolled hypertension, and the occurrence of hypertension is closely related to atherosclerosis. Further analysis shows that the fundamental cause of atherosclerosis is long-term dyslipidemia." The continuous marker words of the causal chain such as "originating from", "related to...", "fundamental cause" appear in the causal chain.

[0033] The present invention extracts overall information from medical text data in the initial stage, identifies and extracts symptom groups and causal chain information, and compensates for the problem of information loss that may be caused by subsequent focus on the internal relationship of a single sentence, thereby ensuring the integrity and accuracy of information extraction.

[0034] S3. According to the preset sentence rules, a first round of sentence rule matching is performed on all the sentences in the medical text obtained in step S1, and the sentences that meet the preset sentence rules are segmented to obtain clause groups containing valid information and clause groups that do not contain valid information.

[0035] It should be noted that the clauses in the clause group containing valid information are the smallest clauses with only one punctuation mark, that is, they cannot be divided any further. For example, "xxx," and "bbb?" are the smallest clauses, but "xxx, yyy." is not the smallest clause, because "xxx, yyy." can also be divided into "xxx," and "yyy.".

[0036] Furthermore, sentence pattern matching is first performed on all sentences in the medical text in turn, and sentences that meet the preset sentence pattern rules are screened out (for example, verbs such as "manifests as" and "may appear" can indicate the boundaries of the symptom entity to be identified, and verbs such as "caused by..." and "caused by..." can indicate the boundaries of the predisposing entity to be identified. Among them, "manifests as" and "may appear" are sentence pattern rule markers). If the screened out sentence that meets the preset sentence pattern rules is not the smallest clause, the nearest punctuation mark before the sentence pattern rule marker and the nearest punctuation mark after the sentence pattern rule marker are used as segmentation boundaries to split the screened out sentence that meets the preset sentence pattern rules into clauses containing target information. This punctuation-based segmentation strategy can effectively split the sentence into clauses containing target information, avoiding the inefficient method of performing cyclic judgment word by word, thereby greatly improving processing efficiency, reducing unnecessary time waste, and ensuring that key information can be quickly and accurately extracted during sentence analysis. If the sentence that meets the preset sentence rules is the smallest clause and contains sentence rule marker words, there is no need to segment it and the subsequent step S4 is directly performed.

[0037] S4. Discard the clause groups that do not contain valid information, perform a second round of sentence rule matching on the clauses in the clause groups that contain valid information according to the preset sentence rules, extract information from each matched clause, and extract the target information of symptom manifestations and inducements.

[0038] Furthermore, in the clauses containing valid information used for matching, the smallest unit that meets the sentence matching conditions is an independent sentence in the text, that is, the sentence obtained by segmenting using the punctuation-based segmentation strategy mentioned in step S3. In the sentence rule matching process, the matching operation is limited to a single sentence to avoid matching across punctuation marks, thereby effectively reducing the matching error rate.

[0039] Furthermore, when extracting information from each matched clause, a corresponding operation strategy needs to be adopted according to the sentence pattern rules matched by the clause, as follows: If a clause matches only one sentence rule, information extraction is performed directly according to the sentence rule; If a clause matches multiple sentence rules at the same time, first, check whether there are the same sentence markers but different numbers of sentence markers in these matching rules. If so, discard the sentence rule with fewer sentence markers, select the sentence rule with more sentence markers to take effect and extract the target information; secondly, if the sentence marker is an independent single sentence rule marker, multiple sentence rules are allowed to take effect at the same time; finally, when a sentence rule contains two or more sentence markers, determine whether the extraction information corresponding to the sentence marker contains an inclusion relationship, discard the sentence rule whose corresponding extraction information does not contain an inclusion relationship, select the sentence rule whose corresponding extraction information has an inclusion relationship to take effect and extract the target information. Through the above multi-level judgment and screening, it is ensured that effective information can be accurately extracted in complex sentences, thereby improving the accuracy and robustness of overall information extraction.

[0040] Furthermore, the extracted symptom manifestations and causes are simple entities or composite entities, the composite entity includes a simple entity and additional information, and the additional information includes a part word, a quantifier, an adverb of degree and an adverb of direction.

[0041] Furthermore, the symptom manifestations include clinical symptoms at the onset of the disease and other information related to the disease manifestations, and the other information related to the disease manifestations include one or more combinations of the disease-affecting population, the onset time of the disease, the severity of the disease, the onset site of the disease, and complications.

[0042] S5. Select target information that meets the screening rules from the extracted target information according to the screening rules to exclude interference information and improve the accuracy and reliability of the final information extraction result.

[0043] Furthermore, the method of selecting target information that meets the screening rules from the extracted target information according to the screening rules is: judging whether the extracted target information contains the corresponding marked landmark words (for example, if it is marked as "cause" or "symptom manifestation", it should contain disease-related words such as "disease", "inflammation" and "tumor"), or whether it meets the corresponding word formation rules (such as "symptom manifestation" and "cause" must satisfy the "location + disease keyword" structure, such as "chest pain"), or whether it follows a specific grammatical structure (such as expressed in the sentence pattern of "subject + predicate + disease description" or "time + subject + symptom manifestation", such as "the patient has developed fever symptoms recently" or "chest pain yesterday"). If the extracted target information meets any of the above conditions, it is retained, otherwise it is discarded.

[0044] S6. Match a subject for each clause in the target information obtained in step S5 and meeting the screening rule according to the marking rule.

[0045] Furthermore, the subject is a literal word with a clear semantic expression function, such as a noun.

[0046] Specifically, the subject is a disease or a symptom, and the specific choice of disease or symptom depends on the sentence structure rules.

[0047] Furthermore, the method for matching the subject of each clause in the target information that meets the screening rule obtained in step S5 according to the marking rule is: First, the positions of all disease entities in the medical text obtained in step S1 are marked to provide a reference for subsequent subject identification. Then, for each selected target information, the text fragment before the sentence rule marker word is checked to determine whether the text fragment meets the conditions for being the subject of the target information; if it meets the conditions, the text fragment is identified as the subject of the target information; if it does not meet the conditions, the disease entity closest to the current sentence rule marker word is found in all previously marked disease entities according to specific rules as the subject of the target information. If the sentence rule requires the subject to be a symptom manifestation or inducement, and a qualified subject cannot be found in the current clause, a qualified text fragment is found in the immediately preceding clause as the subject. The above method ensures that each extracted target information has an accurate and appropriate subject, thereby improving the accuracy of information extraction and matching.

[0048] S7, output the target information that meets the screening rules obtained in step S5 and the subject matched in step S6.

[0049] The present invention effectively extracts information about diseases, symptoms and their causes from Chinese medical texts, and ensures the comprehensiveness, accuracy and efficiency of information extraction. The present invention is of great significance for the automated analysis of medical texts and can provide technical support for medical research, auxiliary diagnosis and other fields.

[0050] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, material or apparatus including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, material or apparatus.

[0051] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for extracting diseases, their symptoms and causes from medical texts, characterized in that: The following steps are involved: S1. Obtain and store medical texts containing diseases and their related symptoms and causes; S2, extracting symptom group information and cause chain information from the medical text obtained in step S1 and outputting them; S3, according to the preset sentence rules, perform a first round of sentence rule matching on all the sentences in the medical text obtained in step S1, segment the sentences that meet the preset sentence rules, and obtain clause groups containing valid information and clause groups that do not contain valid information; S4, discarding the clause group that does not contain valid information, performing a second round of sentence rule matching on the clauses in the clause group that contains valid information according to the preset sentence rules, extracting information from each matched clause, and extracting target information on symptom manifestations and inducements; S5, selecting target information that meets the screening rules from the extracted target information according to the screening rules; S6, matching a subject for each clause in the target information obtained in step S5 and meeting the screening rule according to the marking rule; S7, output the target information that meets the screening rules obtained in step S5 and the subject matched in step S6.

2. The method for extracting diseases, their symptoms and causes from medical texts according to claim 1, characterized in that: The method for extracting symptom group information and predisposing chain information is as follows: based on symptom group keywords and predisposing chain keywords, a relationship analysis is performed on sentences in a medical text, the logical associations between sentences are identified, and symptom group and predisposing chain information is obtained and extracted.

3. The method for extracting diseases, their symptoms and causes from medical texts according to claim 2, characterized in that: In the process of extracting symptom group information and predisposing chain information, if multiple consecutive sentences contain marker words of symptom groups or predisposing chain, the multiple sentences are regarded as a whole and marked as symptom groups or predisposing chains of the corresponding diseases.

4. The method for extracting diseases, their symptoms and causes from medical texts according to claim 1 or 3, characterized in that: The method for segmenting sentences that meet the preset sentence rules is: using the punctuation mark closest to the sentence rule marker and the punctuation mark closest to the sentence rule marker as the segmentation boundary, and splitting the selected sentences that meet the preset sentence rules.

5. The method for extracting diseases, their symptoms and causes from medical texts according to claim 4, characterized in that: The method for extracting information from each matched clause is as follows: when extracting information from each matched clause, a corresponding operation strategy is adopted according to the sentence pattern rule matched by the clause, which is specifically as follows: If a clause matches only one sentence rule, information extraction is performed directly according to the sentence rule; If a clause matches multiple sentence rules at the same time, first, check whether there are the same sentence markers but different numbers of sentence markers in these matching rules. If so, discard the sentence rule with fewer sentence markers and select the sentence rule with more sentence markers to take effect and extract the target information; Secondly, if the sentence marker is an independent single sentence rule marker, multiple sentence rules are allowed to take effect at the same time; finally, when a sentence rule contains two or more sentence markers, it is determined whether the extraction information corresponding to the sentence marker contains an inclusion relationship, and the sentence rules whose corresponding extraction information does not have an inclusion relationship are discarded, and the sentence rules whose corresponding extraction information has an inclusion relationship are selected to take effect and extract the target information.

6. The method for extracting diseases, symptoms and causes from medical texts according to claim 5, characterized in that: The extracted symptom manifestations and causes are simple entities or composite entities. The composite entity includes a simple entity and additional information. The additional information includes a part word, a quantifier, a degree adverb, and a directional adverb.

7. The method for extracting diseases, symptoms and causes from medical texts according to claim 6, characterized in that: The symptom manifestations include clinical symptoms at the onset of the disease and other information related to the disease manifestations, and the other information related to the disease manifestations include one or more combinations of the disease-affecting population, the onset time of the disease, the severity of the disease, the onset site of the disease, and complications.

8. The method for extracting diseases, their symptoms and causes from medical texts according to claim 7, characterized in that: The subject is a literal word with a clear semantic expression function, and the subject is a disease or symptom manifestation.

9. The method for extracting diseases, symptoms and causes from medical texts according to claim 8, characterized in that: The method for matching the subject of each clause in the target information that meets the screening rules obtained in step S5 according to the marking rules is as follows: first, the positions of all disease entities in the medical text obtained in step S1 are marked to provide a reference for subject identification. Then, for each piece of screened target information, the text fragment before the sentence rule marker word is checked to determine whether the text fragment meets the condition of being the subject of the target information; if it meets the condition, the text fragment is determined to be the subject of the target information; if it does not meet the condition, the disease entity closest to the current sentence rule marker word is searched among all previously marked disease entities according to specific rules as the subject of the target information. If the sentence rule requires that the subject be a symptom manifestation or a inducement, and a qualified subject cannot be found in the current clause, a qualified text fragment is searched in the immediately preceding clause as the subject.

10. The method for extracting diseases, their symptoms and causes from medical texts according to claim 9, characterized in that: The medical texts are derived from professional medical textbooks and / or professional medical websites. One method for obtaining the medical texts is: scanning and extracting professional medical textbooks using optical character recognition technology; and collecting medical texts from professional medical websites using web crawler technology.

Images