A data stream processing method for dynamically updating knowledge graphs

Abstract

The present invention discloses a data stream processing method for dynamically updating a knowledge graph, and relates to the technical field of data stream. In the invention of the method, a data stream for data update is obtained through multiple data channels to form a real-time data stream, and text parsing, data verification and processing are performed on the real-time data stream to form drug feature information, and the drug information in the drug feature information group is converted, and a bag-of-words model or a word embedding method is used to form a vector form data group, and multiple directed edge models are established by using a graph database or a graph algorithm for the vector form data group to simulate the multi-faceted interaction relationship between drugs and drug features, and a directed edge model set is established, and the directed edge model set is used to match the directed edge models already existing in the system to form a directed edge query result, and the directed edge model set is dynamically updated according to the query result to achieve real-time updated drug relationships in terms of drug interactions and side effects.

Description

Technical Field

[0001] The present invention relates to the field of data stream technology, and specifically to a data stream processing method for dynamically updating a knowledge graph. Background Art

[0002] In the field of knowledge management, knowledge graphs, as a key information representation and relationship modeling method, are widely used in multiple fields, including medicine and bioinformatics. With the rapid development of the medical field, researchers have accumulated a lot of knowledge about drug interactions, side effects and indications. However, with the continuous evolution of medical knowledge, traditional static knowledge graphs have the problem of difficulty in reflecting these new knowledge in real time. Therefore, a more flexible and real-time knowledge graph update method is needed to better adapt to the dynamic evolution of knowledge in the medical field.

[0003] In traditional knowledge graph update methods, due to the use of static storage and query modes, the rapid response to new knowledge is subject to some limitations. The timeliness and diversity of medical knowledge make it difficult for static update methods to keep up with the emergence of new data. In addition, traditional methods may have problems of information lag and untimely updates when dealing with dynamically changing drug relationships, which makes it difficult for knowledge graphs to accurately reflect the latest research results in the medical field and adapt to the dynamic evolution of medical knowledge.

[0004] Among them, in the study of drug interactions and side effects, affected by individual differences and diversity, traditional methods face challenges in dealing with the personalized needs of a large number of patients. The updating process may fail to fully consider the differences between different populations, resulting in some cases where the treatment plan fails to fully consider the patient's specific physiological and pathological characteristics, thereby reducing the accuracy of the treatment. At the same time, medical research continues to innovate, and the update cycle of traditional methods is long, resulting in information lag in the knowledge graph, which may make doctors fail to fully consider the latest relevant information when formulating treatment plans, thus affecting the treatment effect of patients. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] In view of the shortcomings of the prior art, the present invention provides a data stream processing method for dynamically updating knowledge graphs, which solves the problems mentioned in the background technology.

[0007] (II) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: A data stream processing method for dynamically updating a knowledge graph comprises the following steps:

[0009] Update collection: monitor multiple data channels of medical literature and clinical trial data, identify and capture data changes in the data channels, including new literature releases and experimental data updates, and then intercept them in real time in the form of data streams to form real-time data streams;

[0010] Feature extraction: By performing text parsing on real-time data streams, natural language processing (NLP) technology is used to extract drug information related to drug interactions, side effects, and indications, and data verification and processing are performed to form a drug feature information group;

[0011] Vector construction: convert the drug information in the drug feature information group into numerically quantifiable data, and use the bag-of-words model or word embedding method to form a vector data group;

[0012] Data modeling: Use graph databases or graph algorithms to build multiple directed edge models for vector data sets to simulate the multi-faceted interactions between drugs and drug characteristics, and build a set of directed edge models including interactions, side effects, and adaptive symptoms;

[0013] Relationship search: Match the directed edge models existing in the system through the directed edge model set to obtain the directed edge models related to the start node and the end node in the directed edge model set to form a directed edge query result;

[0014] Dynamic update: Dynamically update the directed edge model set based on the directed edge query results, including using incremental learning methods to expand, overlap, downwardly compatible, upwardly compatible and add new operations to the directed edge model set of the existing model.

[0015] Preferably, the multiple data channels include: medical literature, clinical trial data, medical journals and clinical trial registration centers, and the data changes that occur are intercepted in real time through subscription mechanisms, crawler technology and API calls, while capturing the changed information in the form of data streams.

[0016] Preferably, identifying data changes is determined by timestamp comparison, content comparison, keyword matching, machine algorithms, and manual review;

[0017] Among them, timestamp comparison is achieved through the following steps:

[0018] Timestamp Step 1: Add a timestamp to the information to record the time when the information was published or updated;

[0019] Timestamp Step 2: Compare the timestamp of the newly collected information with the timestamp of the existing information;

[0020] Timestamp step three: Determine whether it is an update. If the timestamp of the new information is later than that of the existing information, it is determined to be an update.

[0021] Preferably, text parsing: processing text in real-time data streams and structuring them into processable text forms, including word segmentation and noise removal preprocessing steps;

[0022] Data verification and processing: Verify and process the text form of data stream conversion, including recording noise, HTML tags, special characters, special symbols, uppercase and lowercase, word segmentation, stop words and punctuation in the text, and perform deletion, replacement and supplement operations.

[0023] Preferably, the conversion includes word segmentation, word embedding and text vectorization, and the conversion of drug information is completed by the following steps to form a vector form data group;

[0024] Word segmentation step: split the text into multiple individual words or phrases to form a vocabulary;

[0025] Word embedding step: Each word needs to be converted into a vector that the computer can understand, where each word is mapped to a vector of a fixed dimension;

[0026] Text vectorization step: combine the vectors of all words into a vector representing the entire text.

[0027] Preferably, the edge-oriented model includes: relationship type, directionality, timestamp, mechanism description, clinical significance, interaction, side effects and adaptive symptoms.

[0028] Preferably, drug interaction: directed edges represent the interaction between two drugs, including synergistic, antagonistic and synergistic effects between drugs;

[0029] Drug side effects: describe the various side effects that the drug may cause, including digestive system side effects, nervous system side effects, immune system side effects, cardiovascular system side effects, metabolic and endocrine system side effects, blood system side effects, urinary system side effects, and skin side effects;

[0030] Drug indication relationship: represents the association between a drug and a certain disease or symptom. The association will identify the name of the specific disease or symptom that the drug is used to treat, including: indication name, ICD code, symptom description, treatment direction and medication recommendations.

[0031] Preferably, the matching includes node matching, relationship matching and obtaining matching results;

[0032] The node matching includes the following steps:

[0033] Node step 1: extract the start node and end node identifiers in the user query;

[0034] Node Step 2: Match the start and end nodes in the graph that have the same identifier as in the user query;

[0035] Node step 3: if there is no query result during the query process of the extraction step and the matching step, the matching process is stopped and the query result is fed back or fed back;

[0036] Relationship matching includes the following steps:

[0037] Relationship step 1: extract relationship type, direction and weight information in the directed edge model;

[0038] Relationship step 2: Check the start and end nodes in the model to match the nodes in the user query information;

[0039] Relationship step three: When the nodes are matched, continue to check whether the relationship type, direction and weight match the data in the user query;

[0040] Relationship step 4: When the relationship matches, the model is added to the matching result set, otherwise the relationship mismatch information is fed back.

[0041] Preferably, the matching result includes the following steps:

[0042] Result step 1: Return the matching results, which contain the directed edge model of the data in the user query;

[0043] Result Step 2: When the matching result set is empty, the system returns an empty prompt, indicating that no matching directed edge model is found.

[0044] Preferably, the dynamic update includes the following steps to perform the update:

[0045] Dynamic update step 1: determine the directed edge model set that needs to be updated based on the directed edge query result;

[0046] Dynamic update step 2: For the query results of the directed edge model that needs to be updated, select the execution operation, including expansion, overlap, downward compatibility, upward compatibility and addition;

[0047] Dynamic update step three: Use expansion, overlap, downward compatibility, upward compatibility and addition to perform specific operations on the directed edge model.

[0048] (III) Beneficial effects

[0049] The present invention provides a data stream processing method for dynamically updating a knowledge graph, which has the following beneficial effects:

[0050] (1) In the embodiment of the method, a data stream of data update is obtained through multiple data channels to form a real-time data stream, the real-time data stream is parsed, and data verification and processing are performed to form drug feature information, the drug information in the drug feature information group is converted, and a bag of words model or a word embedding method is used to form a vector form data group, and a plurality of directed edge models are established by using a graph database or a graph algorithm for the vector form data group to simulate the multi-faceted interaction relationship between drugs and drug features, and a directed edge model set is established, and the directed edge model set is used to match the directed edge models already existing in the system to obtain directed edge models that match the directed edge models. The directed edge model related to the starting node and the end node in the model set is formed to form a directed edge query result. According to the directed edge query result, the directed edge model set is dynamically updated, including using the incremental learning method to expand, overlap, downward compatible, upward compatible and add new operations on the directed edge model set of the existing model, so as to achieve the dynamic update method in terms of drug interactions and side effects. New relationships can be discovered in time, providing doctors with more comprehensive and reliable drug selection and decision-making support, thereby improving the treatment effect of patients. The advantage of this processing flow is that it can quickly adapt to changes in the medical field and better serve the purpose of medical decision-making and research needs.

[0051] (2) By using natural language processing (NLP) technology for feature extraction, the system can extract key information on drug interactions, side effects, and indications from text data. Compared with traditional manual extraction methods, the application of NLP technology makes feature extraction more automated and efficient, improves the processing efficiency of large-scale text data, and also reduces the workload of researchers.

[0052] (3) The introduction of node matching and relationship matching has greatly improved query efficiency. By extracting the node identifiers in the user query, the system can quickly match the start and end nodes with the same identifiers in the graph, thereby locating the information queried by the user. The automation and efficiency of this process effectively reduce the workload of researchers. Compared with the traditional manual query method, the query speed is significantly improved. At the same time, the dynamic update mechanism has brought significant improvements to the timeliness and flexibility of the knowledge graph. By determining the set of directed edge models that need to be updated based on the directed edge query results and selecting specific update operations, including expansion, overlap, downward compatibility, upward compatibility and addition, the system can maintain the integrity of the graph while timely adapting to the continuous changes in medical knowledge. This incremental update method is more efficient than the traditional full update method, effectively reducing the time and cost of system maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flowchart diagram of a data stream processing method for dynamically updating a knowledge graph according to the present invention; DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0055] In the field of knowledge management, knowledge graphs, as a key information representation and relationship modeling method, are widely used in multiple fields, including medicine and bioinformatics. With the rapid development of the medical field, researchers have accumulated a lot of knowledge about drug interactions, side effects and indications. However, with the continuous evolution of medical knowledge, traditional static knowledge graphs have the problem of difficulty in reflecting these new knowledge in real time. Therefore, a more flexible and real-time knowledge graph update method is needed to better adapt to the dynamic evolution of knowledge in the medical field.

[0056] In traditional knowledge graph update methods, due to the use of static storage and query modes, the rapid response to new knowledge is subject to some limitations. The timeliness and diversity of medical knowledge make it difficult for static update methods to keep up with the emergence of new data. In addition, traditional methods may have problems of information lag and untimely updates when dealing with dynamically changing drug relationships, which makes it difficult for knowledge graphs to accurately reflect the latest research results in the medical field and adapt to the dynamic evolution of medical knowledge.

[0057] Among them, in the study of drug interactions and side effects, affected by individual differences and diversity, traditional methods face challenges in dealing with the personalized needs of a large number of patients. The updating process may fail to fully consider the differences between different populations, resulting in some cases where the treatment plan fails to fully consider the patient's specific physiological and pathological characteristics, thereby reducing the accuracy of the treatment. At the same time, medical research continues to innovate, and the update cycle of traditional methods is long, resulting in information lag in the knowledge graph, which may make doctors fail to fully consider the latest relevant information when formulating treatment plans, thus affecting the treatment effect of patients.

[0058] Example 1

[0059] The present invention provides a data stream processing method for dynamically updating knowledge graphs. Figure 1 , including the following steps:

[0060] Update collection: monitor multiple data channels of medical literature and clinical trial data, identify and capture data changes in the data channels, including new literature releases and experimental data updates, and then intercept them in real time in the form of data streams to form real-time data streams;

[0061] Feature extraction: By performing text parsing on real-time data streams, natural language processing (NLP) technology is used to extract drug information related to drug interactions, side effects, and indications, and data verification and processing are performed to form a drug feature information group;

[0062] Vector construction: convert the drug information in the drug feature information group into numerically quantifiable data, and use the bag-of-words model or word embedding method to form a vector data group;

[0063] Data modeling: Use graph databases or graph algorithms to build multiple directed edge models for vector data sets to simulate the multi-faceted interactions between drugs and drug characteristics, and build a set of directed edge models including interactions, side effects, and adaptive symptoms;

[0064] Relationship search: Match the directed edge models existing in the system through the directed edge model set to obtain the directed edge models related to the start node and the end node in the directed edge model set to form a directed edge query result;

[0065] Dynamic update: According to the directed edge query results, the directed edge model set is dynamically updated, including using incremental learning methods to expand, overlap, downwardly compatible, upwardly compatible and add new operations to the directed edge model set of the existing model.

[0066] In the embodiment of the method, a data stream of data update is obtained through multiple data channels to form a real-time data stream, the real-time data stream is subjected to text parsing, and data verification and processing are performed to form drug characteristic information, the drug information in the drug characteristic information group is converted, and a bag of words model or a word embedding method is used to form a vector form data group, and a plurality of directed edge models are established by using a graph database or a graph algorithm for the vector form data group to simulate the multi-faceted interaction relationship between drugs and drug characteristics, and a directed edge model set is established, and the directed edge model set is matched with the directed edge model already existing in the system to obtain the directed edge model. The directed edge models related to the start node and the end node are concentrated to form a directed edge query result. According to the directed edge query result, the directed edge model set is dynamically updated, including using incremental learning methods to expand, overlap, downwardly compatible, upwardly compatible and add new operations to the directed edge model set of the existing model, so as to achieve the dynamic update method in terms of drug interactions and side effects. New relationships can be discovered in a timely manner, providing doctors with more comprehensive and reliable drug selection and decision-making support, thereby improving the treatment effect of patients. The advantage of this processing flow is that it can quickly adapt to changes in the medical field and better serve the purpose of medical decision-making and research needs.

[0067] Example 2

[0068] This embodiment is explained in Example 1, please refer to Figure 1 , specifically:

[0069] Multiple data channels include: medical literature, clinical trial data, medical journals and clinical trial registration centers. Through subscription mechanisms, crawler technology and API calls, data changes are intercepted in real time, and the changing information is captured in the form of data streams.

[0070] Identifying data changes is determined through timestamp comparison, content comparison, keyword matching, machine algorithms, and manual review;

[0071] Among them, timestamp comparison is achieved through the following steps:

[0072] Timestamp Step 1: Add a timestamp to the information to record the time when the information was published or updated;

[0073] Timestamp Step 2: Compare the timestamp of the newly collected information with the timestamp of the existing information;

[0074] Timestamp step three: Determine whether it is an update. If the timestamp of the new information is later than that of the existing information, it is determined to be an update.

[0075] Specific example:

[0076] Suppose there are two pieces of information:

[0077] Existing information: Interaction study of drug A, including key information: A interacts with B, timestamp is 2023-01-01 12:00:00;

[0078] Newly collected information: the latest drug A interaction study results, updated interaction information: A and B interaction, C side effects, timestamp 2023-02-01 14:30:00;

[0079] The steps for timestamp comparison are as follows:

[0080] Timestamp Step 1: Add timestamp to information to record the time when the information is published or updated:

[0081] The timestamp of existing information is: 2023-01-01 12:00:00 and the timestamp of newly collected information is: 2023-02-01 14:30:00;

[0082] Timestamp Step 2: Compare the timestamp of the newly collected information with the timestamp of the existing information:

[0083] Compare the timestamps of existing and new information;

[0084] Time step three: Determine whether it is an update. If the timestamp of the new information is later than that of the existing information, it is determined to be an update:

[0085] The timestamp of the new information (2023-02-01 14:30:00) is later than the timestamp of the existing information (2023-01-01 12:00:00), so it is determined to be updated.

[0086] The steps for content comparison are as follows:

[0087] Content Step 1: Compare the content of existing information and new information:

[0088] Existing information content: A and B interact with each other;

[0089] New information collected: interaction between A and B, side effects of C;

[0090] Content Step 2: Determine whether it is an update:

[0091] Determine whether the newly collected information is different from the existing information;

[0092] Content Step 3: Update information:

[0093] Merge the newly collected information into the existing information to form updated information.

[0094] Text parsing: Process the text in the real-time data stream and structure it into a processable text form, including word segmentation and noise removal preprocessing steps;

[0095] Data verification and processing: Verify and process the text form of data stream conversion, including recording noise, HTML tags, special characters, special symbols, uppercase and lowercase, word segmentation, stop words and punctuation in the text, and perform deletion, replacement and supplement operations.

[0096] The conversion includes word segmentation, word embedding and text vectorization. Specifically, the drug information is converted through the following steps to form a vector data set;

[0097] Word segmentation step: split the text into multiple individual words or phrases to form a vocabulary;

[0098] Word embedding step: Each word needs to be converted into a vector that the computer can understand, where each word is mapped to a vector of a fixed dimension;

[0099] Text vectorization step: combine the vectors of all words into a vector representing the entire text.

[0100] In this implementation, by using natural language processing (NLP) technology for feature extraction, the system can extract key information on drug interactions, side effects and indications from text data. Compared with traditional manual extraction methods, the application of NLP technology makes feature extraction more automated and efficient, improves the processing efficiency of large-scale text data, and also reduces the workload of researchers.

[0101] Example 3

[0102] This embodiment is explained in Example 1, please refer to Figure 1 , specifically:

[0103] The edge-directed model includes: relationship type, directionality, timestamp, mechanism description, clinical significance, interaction, side effects and adaptive symptoms.

[0104] Directionality: Directed edges have directionality, which means there is a starting point and an end point between the two connected nodes. This directionality is used to describe drug relationships;

[0105] Timestamp: records the time when the directed edge is established, which is used to track the evolution of the relationship in the knowledge graph and mark the changes in drug interactions or the establishment of new relationships;

[0106] Mechanistic description: describes the biological or chemical mechanism of the relationship represented by the directed edge;

[0107] Clinical significance: Directed edges may contain information about the potential impact of this relationship on patient treatment, including evaluation information on treatment effectiveness and safety.

[0108] Drug interactions: Directed edges represent the interactions between two drugs, including synergistic, antagonistic, and synergistic effects between drugs;

[0109] Drug side effects: describe the various side effects that the drug may cause, including digestive system side effects, nervous system side effects, immune system side effects, cardiovascular system side effects, metabolic and endocrine system side effects, blood system side effects, urinary system side effects, and skin side effects;

[0110] Drug indication relationship: represents the association between a drug and a certain disease or symptom. The association will identify the name of the specific disease or symptom that the drug is used to treat, including: indication name, ICD code, symptom description, treatment direction and medication recommendations.

[0111] Example 4

[0112] This embodiment is explained in Example 1, please refer to Figure 1 , specifically:

[0113] The matching includes node matching, relationship matching and obtaining matching results;

[0114] The node matching includes the following steps:

[0115] Node step 1: extract the start node and end node identifiers in the user query;

[0116] Node Step 2: Match the start and end nodes in the graph that have the same identifier as in the user query;

[0117] Node step 3: if there is no query result during the query process of the extraction step and the matching step, the matching process is stopped and the query result is fed back or fed back;

[0118] Relationship matching includes the following steps:

[0119] Relationship step 1: extract relationship type, direction and weight information in the directed edge model;

[0120] Relationship step 2: Check the start and end nodes in the model to match the nodes in the user query information;

[0121] Relationship step three: When the nodes are matched, continue to check whether the relationship type, direction and weight match the data in the user query;

[0122] Relationship step 4: When the relationship matches, the model is added to the matching result set, otherwise the relationship mismatch information is fed back.

[0123] Matching results includes the following steps:

[0124] Result step 1: Return the matching results, which contain the directed edge model of the data in the user query;

[0125] Result Step 2: When the matching result set is empty, the system returns an empty prompt, indicating that no matching directed edge model is found.

[0126] Dynamic update includes the following steps:

[0127] Dynamic update step 1: determine the directed edge model set that needs to be updated based on the directed edge query result;

[0128] Dynamic update step 2: For the query results of the directed edge model that needs to be updated, select the execution operation, including expansion, overlap, downward compatibility, upward compatibility and addition;

[0129] Dynamic update step three: Use expansion, overlap, downward compatibility, upward compatibility and addition to perform specific operations on the directed edge model.

[0130] In this embodiment, the query efficiency is greatly improved by introducing node matching and relationship matching. By extracting the node identifiers in the user query, the system can quickly match the start and end nodes with the same identifiers in the graph, thereby locating the information queried by the user. The automation and efficiency of this process effectively reduce the workload of researchers. Compared with the traditional manual query method, the query speed is significantly improved. At the same time, the dynamic update mechanism has brought significant improvements to the timeliness and flexibility of the knowledge graph. By determining the set of directed edge models that need to be updated based on the directed edge query results, and selecting specific update operations, including expansion, overlap, downward compatibility, upward compatibility and addition, the system can maintain the integrity of the graph while timely adapting to the continuous changes in medical knowledge. This incremental update method is more efficient than the traditional full update method, effectively reducing the time and cost of system maintenance.

[0131] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A data stream processing method for dynamically updating a knowledge graph, characterized in that: The following steps are involved: Update collection: monitor multiple data channels of medical literature and clinical trial data, identify and capture data changes in the data channels, including new literature releases and experimental data updates, and then intercept them in real time in the form of data streams to form real-time data streams; Feature extraction: By performing text parsing on real-time data streams, natural language processing (NLP) technology is used to extract drug information related to drug interactions, side effects, and indications, and data verification and processing are performed to form a drug feature information group; Vector construction: convert the drug information in the drug feature information group into numerically quantifiable data, and use the bag-of-words model or word embedding method to form a vector data group; Data modeling: Use graph databases or graph algorithms to build multiple directed edge models for vector data sets to simulate the multi-faceted interactions between drugs and drug characteristics, and build a set of directed edge models including interactions, side effects, and adaptive symptoms; Relationship search: Match the directed edge models existing in the system through the directed edge model set to obtain the directed edge models related to the start node and the end node in the directed edge model set to form a directed edge query result; The matching includes node matching, relationship matching and obtaining matching results; The node matching includes the following steps: Node step 1: extract the start node and end node identifiers in the user query; Node Step 2: Match the start and end nodes in the graph that have the same identifier as in the user query; Node step 3: if there is no query result during the query process of the extraction step and the matching step, the matching process is stopped and the query result is fed back or fed back; Relationship matching includes the following steps: Relationship step 1: extract relationship type, direction and weight information in the directed edge model; Relationship step 2: Check the start and end nodes in the model to match the nodes in the user query information; Relationship step three: When the nodes are matched, continue to check whether the relationship type, direction and weight match the data in the user query; Relationship step 4: When the relationship matches, the model is added to the matching result set, otherwise the relationship mismatch information is fed back; Matching results includes the following steps: Result step 1: Return the matching results, which contain the directed edge model of the data in the user query; Result Step 2: When the matching result set is empty, the system returns an empty prompt, indicating that no matching directed edge model is found; Dynamic update: Dynamically update the directed edge model set based on the directed edge query results, including using incremental learning methods to expand, overlap, downwardly compatible, upwardly compatible, and add new operations to the directed edge model set of the existing model; Dynamic update includes the following steps: Dynamic update step 1: determine the directed edge model set that needs to be updated based on the directed edge query result; Dynamic update step 2: For the query results of the directed edge model that needs to be updated, select the execution operation, including expansion, overlap, downward compatibility, upward compatibility and addition; Dynamic update step three: Use expansion, overlap, downward compatibility, upward compatibility and addition to perform specific operations on the directed edge model.

2. A data stream processing method for dynamically updating a knowledge graph according to claim 1, characterized in that: Multiple data channels include: medical literature, clinical trial data, medical journals and clinical trial registration centers. Through subscription mechanisms, crawler technology and API calls, data changes are intercepted in real time, and the changing information is captured in the form of data streams.

3. According to claim 1, a data stream processing method for dynamically updating a knowledge graph is characterized in that: Identifying data changes is determined through timestamp comparison, content comparison, keyword matching, machine algorithms, and manual review; Among them, timestamp comparison is achieved through the following steps: Timestamp Step 1: Add a timestamp to the information to record the time when the information was published or updated; Timestamp Step 2: Compare the timestamp of the newly collected information with the timestamp of the existing information; Timestamp step three: Determine whether it is an update. If the timestamp of the new information is later than that of the existing information, it is determined to be an update.

4. A data stream processing method for dynamically updating a knowledge graph according to claim 1, characterized in that: Text parsing: Process the text in the real-time data stream and structure it into a processable text form, including word segmentation and noise removal preprocessing steps; Data verification and processing: Verify and process the text form of data stream conversion, including recording noise, HTML tags, special characters, special symbols, uppercase and lowercase, word segmentation, stop words and punctuation in the text, and perform deletion, replacement and supplement operations.

5. According to claim 1, a data stream processing method for dynamically updating a knowledge graph is characterized in that: The conversion includes word segmentation, word embedding and text vectorization. Specifically, the drug information is converted through the following steps to form a vector data set; Word segmentation step: split the text into multiple individual words or phrases to form a vocabulary; Word embedding step: Each word needs to be converted into a vector that the computer can understand, where each word is mapped to a vector of a fixed dimension; Text vectorization step: combine the vectors of all words into a vector representing the entire text.

6. A data stream processing method for dynamically updating a knowledge graph according to claim 1, characterized in that: The edge-directed model includes: relationship type, directionality, timestamp, mechanism description, clinical significance, interaction, side effects and adaptive symptoms.

7. A data stream processing method for dynamically updating a knowledge graph according to claim 1, characterized in that: Drug interactions: Directed edges represent the interactions between two drugs, including synergistic, antagonistic, and synergistic effects between drugs; Drug side effects: describe the various side effects that the drug may cause, including digestive system side effects, nervous system side effects, immune system side effects, cardiovascular system side effects, metabolic and endocrine system side effects, blood system side effects, urinary system side effects, and skin side effects; Drug indication relationship: represents the association between a drug and a certain disease or symptom. The association will identify the name of the specific disease or symptom that the drug is used to treat, including: indication name, ICD code, symptom description, treatment direction and medication recommendations.

Images