An intelligent retrieval system for medical devices

Abstract

The present invention discloses an intelligent retrieval system for medical devices, belongs to the field of medical device knowledge retrieval, and solves the problem of how to provide an intelligent retrieval system so that relevant personnel can obtain accurate medical device knowledge information; a knowledge graph construction module constructs a knowledge graph about medical devices according to medical device information; a search term parsing module parses the search terms input by a user terminal, obtains a search triple set or a search keyword set and sends it to an association matching module; the association matching module analyzes the obtained search triple set or search keyword set, obtains a matching index term set and sends it to a knowledge recommendation module; the knowledge recommendation module pushes the knowledge graph related knowledge information stored in the database to the user terminal according to the obtained matching index term set; and provides the user terminal with knowledge graph knowledge information that contains the most search term information and is arranged in the front, thereby improving work efficiency or learning efficiency.

Description

[0001] This application is a divisional application of the application filed on August 29, 2023, with application number 202311094559.0 and invention name “A medical device intelligent retrieval system”. Technical Field

[0002] The present invention belongs to the field of medical device knowledge retrieval, involves knowledge graph technology, and specifically is an intelligent retrieval system for medical devices. Background Art

[0003] Various medical devices are needed in medical treatment or other medical research. Medical staff or researchers can obtain relevant knowledge about the medical devices they need by searching in the database, which helps them make better decisions and conduct in-depth research.

[0004] In the prior art, when searching a database of medical devices, if the search term is too long, accurate knowledge information cannot be retrieved, resulting in relevant personnel being unable to obtain the required knowledge information, thereby delaying work progress or learning progress. To this end, the present invention proposes an intelligent retrieval system for medical devices. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes an intelligent retrieval system for medical devices, which solves the problem of how to provide an intelligent retrieval system so that relevant personnel can obtain accurate medical device knowledge information.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A medical device intelligent retrieval system includes: a knowledge graph construction module for analyzing medical device information from different data sources, thereby constructing a knowledge graph about the medical device and storing it in a database;

[0008] The search term parsing module is used to parse the search term input by the user, identify and extract the entity of the obtained search term according to the natural language processing technology, obtain one or more search triples, and store the obtained search triples into the search triple set; if the search triple cannot be extracted, the keywords of the search term are extracted and stored into the search keyword set; the obtained search triple set or search keyword set is marked with a timestamp and a user ID tag and sent to the association matching module;

[0009] The association matching module is used to analyze the acquired search triple set or search keyword set with timestamp and user ID label, obtain the matching index word set and send it to the knowledge recommendation module; when the acquired search triple set is used, the index word set of the entity or relation word corresponding to each search triple is obtained according to the synonym dictionary for one or more search triples in the search triple set, and the first matching index word, the second matching index word and the third matching index word of the corresponding triple are obtained according to whether the knowledge graph related knowledge linked to each index word contains the subject entity, relation word and object entity of the corresponding triple, and then the matching index word set of the search triple set containing only one search triple is obtained, and at the same matching level, the index words are arranged in descending order of frequency, so as to obtain the matching index word set of the search triple set containing multiple search triples; when the acquired search keyword set is used, the matching index word set of the search keyword set is obtained after analysis; the association matching module adds the acquired matching index word set to the timestamp and user ID label and sends it to the knowledge recommendation module;

[0010] The knowledge recommendation module is used to push the knowledge graph-related knowledge information stored in the database to the user end based on the obtained matching index word set with timestamp and user ID label.

[0011] Furthermore, the construction process of the knowledge graph construction module is as follows:

[0012] Obtain medical device information from different data sources;

[0013] Process the noise, duplicate data, missing data or erroneous data contained in the acquired medical device information, and convert data with different data structures into a unified data structure and data format;

[0014] Through natural language processing technology, entity recognition and extraction are performed on the text data of medical devices with unified data structure and data format, and the subject entity, object entity and relational word are represented as triples;

[0015] After forming multiple triples, the same subject entities or object entities are linked through a synonym dictionary to construct a knowledge graph about medical devices and store it in the database. The database uses a graph algorithm to index the knowledge graph; among them, the synonym dictionary is to merge different words or terms in the early stage so that synonyms can be associated.

[0016] Furthermore, medical device information includes the name, specifications, functions, performance, brand, manufacturer, instructions for use, use, indications and symptom description of the medical device, among which the name, brand and manufacturer of the medical device are the main entity information, and the specifications, functions, performance, instructions for use, use, indications and symptom description of the medical device are the object entity information.

[0017] Furthermore, the process of the search term parsing module acquiring the search triples and storing them in the search triple set is as follows:

[0018] Count the number of search triples; if multiple search triples are obtained through the search term, arrange the extracted triples in the order in which they appear in the search term; and count the frequency of each search triple. If there are search triples with a frequency greater than 1, delete the search triples with a frequency greater than 1 until the frequency is equal to 1; store the processed search triples in the search triple set, and add a timestamp and user ID tag to send it to the associated matching module.

[0019] Furthermore, when a set of retrieval triples with timestamps and user ID tags is obtained, the analysis process of the association matching module is as follows:

[0020] Count the number of retrieval triples contained in the retrieval triple set;

[0021] If the number of search triplets is equal to 1, then:

[0022] Mark the unique search triple in the search triple set as the target triple, and mark the subject entity word, relation word and object subject word contained in it as the first target entity, the target relation word and the second target entity respectively;

[0023] According to the synonym dictionary, the first target entity, the target relationship word and the second target entity are traversed in the index words contained in the database respectively, and the subject entity index word set, the relationship word index word set and the object entity index word set belonging to the first target entity, the target relationship word and the second target entity are obtained respectively;

[0024] Traverse the knowledge graph entities linked to the index terms contained in each index term set, other entities linked to the entity, and the relationship terms between the entity and other entities;

[0025] According to the synonym dictionary, determine whether the entity corresponding to the knowledge graph, other entities linked to the entity, and the relationship terms between the entity and other entities are semantically similar to the first target entity, the second target entity, and the target relationship terms; if all three semantics are similar, mark the index term as the first matching index term; if there are two semantic similarities, mark the index term as the second matching index term; if there is one semantic similarity, mark the index term as the third matching index term;

[0026] Filter out the first matching index word, the second matching index word and the third matching index word from each index word set, delete duplicate index words, and arrange the index words after deleting duplicates in order according to the matching level, so as to obtain the matching index word set of the target triple, and add the timestamp and user ID tag to the obtained matching index word set and send it to the knowledge recommendation module;

[0027] If the number of search triplets is greater than 1, then:

[0028] Process each search triple in the search triple set in the manner that the number of search triples is equal to 1, obtain a matching index word set for each search triple, and merge them into a matching index word set;

[0029] Count the frequency of each index word, arrange the index words contained in each matching level in the merged matching index word set in order of frequency from large to small, delete duplicate index words, and add the processed matching index word set with a timestamp and user ID tag and send it to the knowledge recommendation module.

[0030] Furthermore, when a search keyword set with a timestamp and a user ID tag is obtained, the analysis process of the association matching module is as follows:

[0031] If the number of search keywords is equal to 1, the only search keyword in the search keyword set is marked as the target keyword. According to the synonym dictionary, the target keyword is traversed in the index words contained in the database to obtain the matching index word set belonging to the target keyword, and the obtained matching index word set is marked with a timestamp and a user ID tag and sent to the knowledge recommendation module;

[0032] If the number of search keywords is greater than 1, then the matching index word sets of each search keyword are obtained according to the above method that the number of search keywords is equal to 1, and merged into one matching index word set;

[0033] Count the frequency of each index word, arrange the index words in the merged matching index word set in descending order of frequency, delete the duplicate index words, add timestamp and user ID tag and send to the knowledge recommendation module.

[0034] Furthermore, the knowledge recommendation module extracts each index word contained in the index word set according to the index word set; and sends the knowledge information of the knowledge graph corresponding to the link to the user terminal in sequence according to the order of the index words in the index word set for the user terminal to view.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. In the present invention, the search term parsing module parses the search term input by the user, identifies and extracts the entity of the obtained search term, so as to obtain the search triple. When the search term is long, multiple search triples can be obtained according to the search term. If the search term is short or the search triple cannot be extracted from the search term, the search keyword is directly extracted; the search term parsing module sends the obtained search triple set or search keyword set to the association matching module, and provides the association matching module with the data of the user ID corresponding to the timestamp of the search term, so as to provide the search data in the format and standard so that the subsequent search term can be accurately linked to the knowledge graph related knowledge.

[0037] 2. In the present invention, the acquired search triple set or search keyword set is analyzed by the association matching module, and then the matching index word set for the corresponding timestamp and user ID tag is obtained and sent to the knowledge recommendation module, and the knowledge recommendation module sends the corresponding knowledge in the knowledge graph to the user end according to the order of the index words in the matching index word set; wherein, the acquisition of the matching index word set is divided into two situations, namely, the analysis and acquisition of the search triple set and the analysis and acquisition of the search keyword set; for one or more search triples in the search triple set, the index word set of the entity or relationship word corresponding to each search triple is obtained according to the synonym dictionary, and then the knowledge graph linked to each index word is obtained. Whether the relevant knowledge contains the subject entity, relational term and object entity of the corresponding triple, so as to obtain the first matching index term, the second matching index term and the third matching index term of the corresponding triple, and then obtain the matching index term set of the retrieval triple set containing only one retrieval triple, and at the same matching level, arrange the index terms in order of frequency from large to small, so as to obtain the matching index term set of the retrieval triple set containing multiple retrieval triples; through the association matching module and the knowledge recommendation module, the user terminal can be provided with the knowledge graph knowledge information that contains the most retrieval term information and is arranged in the front, so that the user can obtain more accurate knowledge information, thereby improving work efficiency or learning efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The figure is a block diagram of an intelligent retrieval system for medical devices of the present invention. DETAILED DESCRIPTION

[0039] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than 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.

[0040] like Figure 1 As shown, a medical device intelligent retrieval system includes a knowledge graph construction module, a search term parsing module, an association matching module, a knowledge recommendation module and a database;

[0041] The knowledge graph construction module is used to analyze medical device information from different data sources, and then construct a knowledge graph about medical devices; the process is as follows:

[0042] Obtain medical device information from different data sources, including the name, specifications, functions, performance, brand, manufacturer, instructions for use, use, indications, disease description, etc. of the medical device; the data sources come from medical device manufacturers, medical databases, etc.;

[0043] Process the noise, duplicate data, missing data or erroneous data contained in the acquired medical device information, and convert data with different data structures into a unified data structure and data format;

[0044] Through natural language processing technology, entity recognition and extraction are performed on the text data of medical devices with unified data structure and data format, including: recognition of subject entity information such as the name, brand, and manufacturer of the medical device, recognition of object entity information such as the specifications, functions, performance, instructions for use, uses, indications, and descriptions of symptoms of the medical device, and extraction of relationship word information between subject entities and object entities, such as having, including, applicable to, and treating; the subject entity, object entity, and relationship word information are represented as triples;

[0045] After forming multiple triples, the same subject entities or object entities are linked through a synonym dictionary to construct a knowledge graph about medical devices and store it in the database. The database uses a graph algorithm to index the knowledge graph.

[0046] Among them, the thesaurus is to merge different words or terms in the early stage so that synonyms can be associated;

[0047] The search term parsing module is used to parse the search term input by the user terminal, including:

[0048] Obtain the search term input by the user; perform semantic analysis on the obtained search term, identify the subject entity, the object entity and the relational terms between the two, and establish a search triple; if the search term has a large number of characters, multiple search triples can be extracted; if the search triple cannot be established based on the search term, obtain the entity or relational term as a keyword;

[0049] When the search triples are obtained through the search terms, we have:

[0050] Count the number of search triples; if multiple search triples are obtained through the search term, arrange the extracted triples in the order in which they appear in the search term; and count the frequency of each search triple. If there is a search triple with a frequency greater than 1, delete the search triple with a frequency greater than 1 until the frequency is equal to 1;

[0051] The processed retrieval triples are stored in the retrieval triple set, and are labeled with a timestamp and user ID and sent to the association matching module;

[0052] When no search triples are obtained through the search terms, there are:

[0053] The extracted entities or relation words are used as keywords, and the keywords are stored in the search keyword set, and are labeled with timestamps and user IDs and sent to the association matching module;

[0054] Failure to obtain the retrieval triple means that only one entity or relation word can be extracted, or multiple entities are extracted but cannot form a triple;

[0055] In this embodiment, the search term parsing module parses the search term input by the user, identifies and extracts entities of the obtained search term, thereby obtaining a search triple. When the search term is long, multiple search triples can be obtained according to the search term. If the search term is short or the search triple cannot be extracted from the search term, the search keyword is directly extracted. The search term parsing module sends the obtained search triple set or search keyword set to the association matching module, and provides the association matching module with data of the user ID corresponding to the timestamp of the search term, thereby providing search data that conforms to the format and standard for the subsequent search term to be accurately linked to the knowledge graph related knowledge.

[0056] The association matching module is used to analyze the acquired search triple set or search keyword set with timestamp and user ID tag, obtain the matching index word set and send it to the knowledge recommendation module; it is divided into two situations, and the process is as follows:

[0057] Scenario 1: When the retrieved triples are labeled with timestamp and user ID;

[0058] Count the number of retrieval triples contained in the retrieval triple set;

[0059] If the number of search triplets is equal to 1, then:

[0060] Mark the unique search triple in the search triple set as the target triple, and mark the subject entity word, relation word and object subject word contained in it as the first target entity, the target relation word and the second target entity respectively;

[0061] According to the synonym dictionary, the first target entity, the target relationship word and the second target entity are traversed in the index words contained in the database respectively, and the subject entity index word set, the relationship word index word set and the object entity index word set belonging to the first target entity, the target relationship word and the second target entity are obtained respectively;

[0062] Traverse the knowledge graph entities linked to the index terms contained in each index term set, other entities linked to the entity, and the relationship terms between the entity and other entities;

[0063] According to the synonym dictionary, determine whether the entity corresponding to the knowledge graph, other entities linked to the entity, and the relationship terms between the entity and other entities are semantically similar to the first target entity, the second target entity, and the target relationship terms; if all three semantics are similar, mark the index term as the first matching index term; if there are two semantic similarities, mark the index term as the second matching index term; if there is one semantic similarity, mark the index term as the third matching index term;

[0064] Filter out the first matching index word, the second matching index word and the third matching index word from each index word set, delete duplicate index words, and arrange the index words after deleting duplicates in order according to the matching level, so as to obtain the matching index word set of the target triple, and add the timestamp and user ID tag to the obtained matching index word set and send it to the knowledge recommendation module;

[0065] If the number of search triplets is greater than 1, then:

[0066] Process each search triple in the search triple set in the above manner, obtain a matching index word set for each search triple, and merge them into a matching index word set;

[0067] Count the frequency of each index word, arrange the index words contained in each matching level in the merged matching index word set in descending order of frequency, delete duplicate index words, and add a timestamp and user ID tag to the processed matching index word set and send it to the knowledge recommendation module;

[0068] Scenario 2: When the retrieved keyword set is labeled with a timestamp and user ID;

[0069] If the number of search keywords is equal to 1, the only search keyword in the search keyword set is marked as the target keyword. According to the synonym dictionary, the target keyword is traversed in the index words contained in the database to obtain the matching index word set belonging to the target keyword, and the obtained matching index word set is marked with a timestamp and a user ID tag and sent to the knowledge recommendation module;

[0070] If the number of search keywords is greater than 1, then the matching index word sets of each search keyword are obtained according to the above method and merged into one matching index word set;

[0071] Count the frequency of each index word, arrange the index words in the merged matching index word set in descending order of frequency, delete duplicate index words, add timestamp and user ID tag and send to the knowledge recommendation module;

[0072] The knowledge recommendation module is used to push the knowledge graph-related knowledge information stored in the database to the user end according to the obtained matching index word set with timestamp and user ID tag;

[0073] According to the index word set, extract each index word contained in the index word set;

[0074] According to the order of the index words in the index word set, the knowledge information of the knowledge graph corresponding to the link is sent to the user end in turn for the user end to view;

[0075] In this embodiment, the acquired search triple set or search keyword set is analyzed by the association matching module, and then the matching index word set for the corresponding timestamp and user ID tag is obtained and sent to the knowledge recommendation module, and the knowledge recommendation module sends the corresponding knowledge in the knowledge graph to the user end according to the order of the index words in the matching index word set; wherein, the acquisition of the matching index word set is divided into two situations, namely, the analysis and acquisition of the search triple set and the analysis and acquisition of the search keyword set; for one or more search triples in the search triple set, the index word set of the entity or relationship word corresponding to each search triple is obtained according to the synonym dictionary, and then the knowledge graph linked to each index word is obtained. Whether the relevant knowledge contains the subject entity, relational term and object entity of the corresponding triple, so as to obtain the first matching index term, the second matching index term and the third matching index term of the corresponding triple, and then obtain the matching index term set of the retrieval triple set containing only one retrieval triple, and at the same matching level, arrange the index terms in order of frequency from large to small, so as to obtain the matching index term set of the retrieval triple set containing multiple retrieval triples; through the association matching module and the knowledge recommendation module, the user terminal can be provided with the knowledge graph knowledge information that contains the most retrieval term information and is arranged in the front, so that the user can obtain more accurate knowledge information, thereby improving work efficiency or learning efficiency.

[0076] In the embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation; the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the method of this embodiment.

[0077] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An intelligent retrieval system for medical devices, characterized in that: include: The knowledge graph construction module is used to analyze the medical device information from different data sources, and then construct a knowledge graph about medical devices and store it in the database; The search term parsing module is used to parse the search term input by the user, identify and extract the entity of the obtained search term according to the natural language processing technology, obtain one or more search triples, and store the obtained search triples into the search triple set; if the search triple cannot be extracted, the keywords of the search term are extracted and stored into the search keyword set; the obtained search triple set or search keyword set is marked with a timestamp and a user ID tag and sent to the association matching module; The association matching module is used to analyze the acquired search triple set or search keyword set with timestamp and user ID label, obtain the matching index word set and send it to the knowledge recommendation module; when the acquired search triple set is used, the index word set of the entity or relation word corresponding to each search triple is obtained according to the synonym dictionary for one or more search triples in the search triple set, and the first matching index word, the second matching index word and the third matching index word of the corresponding triple are obtained according to whether the knowledge graph related knowledge linked to each index word contains the subject entity, relation word and object entity of the corresponding triple, and then the matching index word set of the search triple set containing only one search triple is obtained, and at the same matching level, the index words are arranged in descending order of frequency, so as to obtain the matching index word set of the search triple set containing multiple search triples; when the acquired search keyword set is used, the matching index word set of the search keyword set is obtained after analysis; the association matching module adds the acquired matching index word set to the timestamp and user ID label and sends it to the knowledge recommendation module; When a set of retrieval triples with timestamps and user ID tags is obtained, the analysis process of the association matching module is as follows: Count the number of retrieval triples contained in the retrieval triple set; If the number of search triplets is equal to 1, then: Mark the unique search triple in the search triple set as the target triple, and mark the subject entity word, relation word and object subject word contained in it as the first target entity, the target relation word and the second target entity respectively; According to the synonym dictionary, the first target entity, the target relationship word and the second target entity are traversed in the index words contained in the database respectively, and the subject entity index word set, the relationship word index word set and the object entity index word set belonging to the first target entity, the target relationship word and the second target entity are obtained respectively; Traverse the knowledge graph entities linked to the index terms contained in each index term set, other entities linked to the entity, and the relationship terms between the entity and other entities; Determine whether the entity corresponding to the knowledge graph, other entities linked to the entity, and the relationship words between the entity and other entities are semantically similar to the first target entity, the second target entity, and the target relationship words according to the synonym dictionary; If the three terms are similar in meaning, the index term is marked as the first matching index term; If there are two semantically similar terms, the index term is marked as the second matching index term; if there is one semantically similar term, the index term is marked as the third matching index term; Filter out the first matching index word, the second matching index word and the third matching index word from each index word set, delete duplicate index words, and arrange the index words after deleting duplicates in order according to the matching level, so as to obtain the matching index word set of the target triple, and add the timestamp and user ID tag to the obtained matching index word set and send it to the knowledge recommendation module; If the number of search triplets is greater than 1, then: Process each search triple in the search triple set in the manner that the number of search triples is equal to 1, obtain a matching index word set for each search triple, and merge them into a matching index word set; Count the frequency of each index word, arrange the index words contained in each matching level in the merged matching index word set in descending order of frequency, delete duplicate index words, and add a timestamp and user ID tag to the processed matching index word set and send it to the knowledge recommendation module; The knowledge recommendation module is used to push the knowledge graph-related knowledge information stored in the database to the user end based on the obtained matching index word set with timestamp and user ID tag; The construction process of the knowledge graph building module is as follows: Obtain medical device information from different data sources; Process the noise, duplicate data, missing data or erroneous data contained in the acquired medical device information, and convert data with different data structures into a unified data structure and data format; Through natural language processing technology, entity recognition and extraction are performed on the text data of medical devices with unified data structure and data format, and the subject entity, object entity and relational word are represented as triples; After forming multiple triples, the same subject entities or object entities are linked through a synonym dictionary to construct a knowledge graph about medical devices and store it in the database. The database uses a graph algorithm to index the knowledge graph. Among them, the synonym dictionary is to merge different words or terms in the early stage so that synonyms can be associated. When a search keyword set with a timestamp and user ID tag is obtained, the analysis process of the association matching module is as follows: If the number of search keywords is equal to 1, the only search keyword in the search keyword set is marked as the target keyword. According to the synonym dictionary, the target keyword is traversed in the index words contained in the database to obtain the matching index word set belonging to the target keyword, and the obtained matching index word set is marked with a timestamp and a user ID tag and sent to the knowledge recommendation module; If the number of search keywords is greater than 1, then the matching index word sets of each search keyword are obtained according to the above method that the number of search keywords is equal to 1, and merged into one matching index word set; Count the frequency of each index word, arrange the index words in the merged matching index word set in descending order of frequency, delete duplicate index words, add timestamp and user ID tag and send to the knowledge recommendation module; The knowledge recommendation module extracts each index word contained in the index word set according to the index word set; and sends the knowledge information of the knowledge graph corresponding to the link to the user terminal in sequence according to the order of the index words in the index word set for the user terminal to view.

2. The intelligent medical device retrieval system according to claim 1, characterized in that: Medical device information includes the name, specifications, functions, performance, brand, manufacturer, instructions for use, purpose, indications and symptom description of the medical device, among which the name, brand and manufacturer of the medical device are the main entity information, and the specifications, functions, performance, instructions for use, purpose, indications and symptom description of the medical device are the object entity information.

3. The intelligent medical device retrieval system according to claim 1, characterized in that: The process of the search term parsing module acquiring the search triples and storing them in the search triple set is as follows: Count the number of search triples; if multiple search triples are obtained through the search term, arrange the extracted triples in the order in which they appear in the search term; and count the frequency of each search triple. If there are search triples with a frequency greater than 1, delete the search triples with a frequency greater than 1 until the frequency is equal to 1; store the processed search triples in the search triple set, and add a timestamp and user ID tag to send it to the associated matching module.

Images