Intelligent information retrieval method and system for mass medical care data
By constructing a health care database containing text and image information and using intelligent information retrieval methods, the problem of inefficient retrieval of massive health care data is solved, and high-precision and high-intelligence search results are achieved.
Patent Information
- Application Number
- CN202410799720.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-20
- Publication Date
- 2024-11-08
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing technology is difficult to effectively retrieve and manage massive amounts of health care data, resulting in inefficient searches and inaccurate results.
By constructing a health care database containing text and image information, intelligent information retrieval methods are adopted, including user input search methods, collecting and matching results, forming mapping subsets, merging search methods with similar semantics, and optimizing searches based on user clicks and browsing behaviors Recommended results.
It improves the accuracy and reliability of search results, reduces results that users do not care about when browsing, and enhances the intelligence of search.
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data retrieval, and in particular to a method and system for intelligent information retrieval of massive medical care data. Background Art
[0002] The description of the background technology in the present invention belongs to the related technology related to the present invention and is only used to illustrate and facilitate the understanding of the invention content of the present invention. It should not be understood that the applicant explicitly believes or infers that the applicant believes that it is the prior art of the present invention on the filing date of the first application.
[0003] The amount of healthcare data is huge. With the development of society, healthcare has received more and more attention from people and a large number of experts and scholars. However, for the development of healthcare, historical data is a good reference standard. Whether it is users designing their own healthcare plans, healthcare experts and doctors solving problems encountered in healthcare, or in healthcare education, the use of historical data is very important. Historical data can be a reference for many projects.
[0004] However, healthcare data is complex, and intelligent information retrieval can greatly improve retrieval efficiency. Currently, there is an urgent need for an intelligent information retrieval method. Summary of the Invention
[0005] The purpose of the embodiments of the present invention is to provide a method and system for intelligent information retrieval of massive medical care data. The method of the present invention has a high degree of intelligent retrieval, and the retrieval results are accurate and reliable.
[0006] A method for intelligent information retrieval of massive medical care data, comprising the following steps:
[0007] S1. Build a healthcare database, wherein the healthcare database includes text information and image information;
[0008] S2. Matching search terms with search results: After the user enters the search term, the search results are displayed to the customer according to preset rules;
[0009] S3. Search formula collection: When a user searches, the user's search formula is collected to obtain a search formula database;
[0010] S4. User click result collection: The user clicks on the search result that the user needs according to the displayed search results, and the search result clicked by the user is determined to be a valid search result. The valid search results are mapped to the search terms entered by the user to form a mapping subset;
[0011] S5. Merge the mapping subsets to obtain a mapping set: Merge the mapping subsets obtained by semantically similar search expressions. When merging the mapping subsets, only one identical search result is retained and the number of repetitions is marked in the search results.
[0012] S6. Mapping set recommendation: When the number of search results clicked by the user that are identical to the results in a certain mapping set reaches a set value, the mapping set is recommended to the user.
[0013] Furthermore, the step S3 collects the user's search formulas and performs the following processing: two search formulas with a search result repetition rate of not less than 80% are defined as each other's interoperable search formulas, and the number of interoperable search formulas of each search formula is marked as a quantity label, and the quantity label is linked to the corresponding interoperable search formula.
[0014] Furthermore, the mapping subset in step S4 is divided into two virtual areas, the first area is the valid search results, and the second area is the potential valid search results; the potential valid search results are determined according to the following steps: all mapping subsets are repeatedly compared, and the mapping subset whose number of valid search results in the first area accounts for a ratio greater than a set ratio of the valid search results in the first area is taken as a potential mapping subset, and the search results other than the valid search results in the potential mapping subset are placed in the second area as potential valid search results.
[0015] Furthermore, in step S4, the retrieval result clicked by the user is determined to be a valid retrieval result based on the time the user spends browsing the retrieval result. When the user's browsing time is greater than the set time, the clicked result is determined to be a valid retrieval result.
[0016] Furthermore, in step S5, the search terms with similar semantics are determined in the following manner:
[0017] (1) Using semantic analysis software, semantically similar search terms are defined as pre-selected search terms;
[0018] (2) performing pairwise comparisons on the search results of the preselected search formulas, determining the repetition rate between the two preselected search formulas, and calculating the mean repetition rate of each preselected search formula;
[0019] (3) Eliminate the preselected search formulas whose mean repetition rate is less than the set mean repetition rate, and define the remaining search formulas as the first search formula set;
[0020] (4) The search results whose ratio of the number of occurrences in the first search formula set to the number of pre-selected search formulas in the first search formula set is greater than 50% are defined as core search results;
[0021] (5) comparing the core search results with the search formula results in the search formula database to find a second search formula set, wherein the search formulas in the second search formula set meet the following conditions: covering more than 60% of the core search results, and the proportion of the core search results in the search results of the search formula is greater than 20%;
[0022] (6) The first search formula set and the second search formula set are merged to obtain search formulas with similar semantics, and the repeated search formulas in the first search formula set and the second search formula set are used as core search formulas.
[0023] Furthermore, the retrieval of image information of the retrieval method includes the following steps:
[0024] Get the original image: The original image is the retrieval target provided by the user;
[0025] Capture key position images: perform a scaled comparison between the original image and the reference image in the image database, and capture the key position images for retrieval;
[0026] Key position image positioning: set a reference point, establish a virtual coordinate system with the reference point as the origin, and determine the position coordinates of the key position image;
[0027] A key position image is input at the search position, and the key position image has a virtual coordinate system; a virtual coordinate system is generated at the same reference point in the medical care database, and an image area with the same coordinates as the key position image is intercepted as a search comparison area. The key position image and the search comparison area are compared for coverage at the same proportion, and an image with a coverage rate that meets the set value is displayed as a search result.
[0028] An intelligent information retrieval system for massive medical care data, comprising:
[0029] A medical care database, wherein the medical care database includes text information and image information;
[0030] Search formula and search result matching module: According to preset rules, after the user enters the search formula, the search results are displayed to the customer;
[0031] Retrieval formula database: When a user searches, the user's search formula is collected to obtain a search formula database;
[0032] User click result collection module: The user clicks on the search result they need based on the displayed search results. The search result clicked by the user is determined to be a valid search result. The valid search results are mapped to the search terms entered by the user to form a mapping subset.
[0033] Merge module: Merges the mapping subsets obtained from semantically similar search expressions. When merging mapping subsets, only one identical search result is retained and the number of repetitions is marked in the search results.
[0034] Mapping set recommendation module: When the number of search results clicked by the user that are identical to the results in a certain mapping set reaches a set value, the mapping set is recommended to the user.
[0035] The embodiments of the present invention have the following beneficial effects:
[0036] The method of the present invention determines the search results that the user actually needs when using a certain search formula based on the original search rules and the user's search and search result screening process, and marks and classifies each search result according to the number of times the user has selected it. During use, the method intelligently recommends search formulas and search results to the user, and continuously optimizes the search results during use, so that the search results are gradually accurate, and the user gradually reduces the number of search results that he or she does not care about. The method has a high degree of intelligence and accurate search results. DETAILED DESCRIPTION
[0037] The present application will be further described below with reference to the embodiments.
[0038] To more clearly illustrate the embodiments of the present invention or technical solutions in the prior art, different "one embodiment" or "embodiment" in the following description do not necessarily refer to the same embodiment. Different embodiments may be replaced or combined. Those skilled in the art can also derive other implementation methods based on these embodiments without inventive effort.
[0039] A method for intelligent information retrieval of massive medical care data, comprising the following steps:
[0040] S1. Constructing a healthcare database, which includes text and image information. The healthcare database contains a large amount of text and image information. For example, how to conservatively treat diabetic patients corresponds to a large number of case studies and conservative treatment plans. For example, when a user (a patient or a doctor) needs to understand a patient's condition and specify a treatment plan, such as a patient with bone hyperplasia X-ray, the X-ray can be directly placed in the search system for retrieval. The specific retrieval method will be illustrated in the preferred embodiment below.
[0041] S2. Matching search terms with search results: According to preset rules, after the user enters the search term, the search results are displayed to the customer. For example, if the customer enters the search term as: male + 45 years old + lung cancer + stage 3 + non-chemotherapy + annual cost less than 200,000 yuan, the search system will display the cases, treatment plans and effects containing the above information as search results to the user. The user can select the appropriate search results according to their needs. Of course, the retrieved cases generally only contain information related to the condition and treatment, and generally should not include the patient's personal information and other information that should not be disclosed or is irrelevant to treatment and health care needs.
[0042] S3. Search formula collection: When a user searches, the user's search formula is collected to obtain a search formula database;
[0043] S4. User click result collection: The user clicks on the search result that the user needs according to the displayed search results, and the search result clicked by the user is determined to be a valid search result. The valid search results are mapped to the search terms entered by the user to form a mapping subset;
[0044] S5. Merge the mapping subsets to obtain a mapping set: Merge the mapping subsets obtained by semantically similar search expressions. When merging the mapping subsets, only one identical search result is retained and the number of repetitions is marked in the search results.
[0045] S6. Mapping set recommendation: When the number of search results clicked by the user that are identical to the results in a certain mapping set reaches a set value, the mapping set is recommended to the user.
[0046] The method of the present invention determines the search results that the user actually needs when using a certain search formula based on the original search rules and the user's search and search result screening process, and marks and classifies each search result according to the number of times the user has selected it. During use, the method intelligently recommends search formulas and search results to the user, and continuously optimizes the search results during use, so that the search results are gradually accurate, and the user gradually reduces the number of search results that he or she does not care about. The method has a high degree of intelligence and accurate search results.
[0047] As a preferred solution, the step S3 collects the user's search formulas and performs the following processing: two search formulas with a search result repetition rate of not less than 80% are defined as mutually interoperable search formulas, the number of interoperable search formulas for each search formula is marked as a quantity label, and the quantity label is linked to the corresponding interoperable search formula. It should be noted here that there are many different search formula methods. Due to reasons such as cultural level or search habits, users may have different search results even when their needs are the same, which requires the search formula to be processed. In some embodiments, if the results retrieved by two search formulas are 80% the same, it is considered that the intentions of the users who entered the two search formulas are the same. Of course, the 80% repetition rate is only a standard in this embodiment, and other ratios can be set according to different levels of accuracy, such as 90%, 95%, or 88%.
[0048] As a preferred solution, the mapping subset in step S4 is divided into two virtual areas, the first area being valid search results and the second area being potential valid search results. Potential valid search results are determined as follows: all mapping subsets are repeatedly compared, and mapping subsets whose number of valid search results in the first area accounts for a greater than a set ratio of the valid search results in the first area are selected as potential mapping subsets, and search results other than the valid search results in the potential mapping subset are placed in the second area as potential valid search results. It is understandable that after a user enters a search formula, a large number of search results appear, such as 1,000, 10,000, or other numbers. In this case, after browsing a portion of the search results, the user finds a search result that he or she believes meets his or her needs and may no longer need to browse other search results. Since users with different similar needs browse different results, for example, if the search results are numbered 1-1000, user A stops clicking after clicking on the 20th numbered search result. When user B searches, although user B has similar needs to user A, the search results he or she clicks on are generally not the same as those clicked by user A. However, there will generally be some repeated clicks. If the search results clicked by B include 80% of the results clicked by A, of course, it can also be other set ratios, and it is believed that A and B can verify each other, then the search results of B's clicks, except for the search results that are repeated with A, will be regarded as A's potential valid search results.
[0049] As a preferred solution, in step S4, the retrieval result clicked by the user is determined to be a valid retrieval result based on the length of time the user browses the retrieval result. When the user's browsing time is greater than the set time, the clicked result is determined to be a valid retrieval result. It can be understood here that the user clicking on the retrieval result does not mean that it is the retrieval result the user needs. It can be determined in a variety of ways. In this embodiment, the length of browsing time is used as the basis, such as browsing for more than 1 minute, more than 5 minutes, etc. Of course, it can also be determined in other ways. For example, in some embodiments, the retrieval results have options for users to choose, such as useful, favorite, irrelevant, etc., and can also be determined based on the customer's choice. There are many ways to determine it. The above are just two preferred embodiments for easy understanding, and cannot be used as a limitation on the scope of protection of this application.
[0050] As a preferred solution, the semantically similar search terms in step S5 are determined in the following manner:
[0051] (1) Using semantic analysis software, semantically similar search terms are defined as pre-selected search terms. There are many ways to determine semantic similarity, and this is a very mature technology, such as Google semantic analysis, which will not be listed or elaborated here.
[0052] (2) Compare the search results of the pre-selected search formulas in pairs, determine the repetition rate between the two pre-selected search formulas, and calculate the average repetition rate of each pre-selected search formula; this is easy to understand, that is, use the semantically similar search formulas obtained by the software semantic analysis as screening targets, and compare them in pairs. For example, after the analysis of search formula A and search formula B by the semantic analysis software, it is preliminarily considered that the two are semantically similar, but there are complex differences between the search formulas that are difficult to analyze through the surface of the text. Therefore, it cannot be judged as similar simply by the approximation given by the semantic analysis software. It also needs to be verified based on the search results. The analysis of search formula A is as follows: search formula B, search formula C Compare the results of search formulas D and E with those of search formula A. The number of search results of search formula B and search results of search formula A accounts for 60% of the search results of search formula A; the number of search results of search formula C and search results of search formula A accounts for 55% of the search results of search formula A; the number of search results of search formula D and search results of search formula A accounts for 70% of the search results of search formula A, and the number of search results of search formula E and search results of search formula A accounts for 95% of the search results of search formula A; then the average repetition rate of search formula A is (60+55+70+95)÷4=70, so it is considered that the average repetition rate of search formula A is 70;
[0053] (3) Eliminate the pre-selected search formulas whose mean repetition rate is less than the set mean repetition rate, and define the remaining search formulas as the first search formula set; because the semantic analysis of commonly used software has a certain degree of universality, but often lacks specificity, after the software analyzes the search formulas with similar semantics and calculates the mean repetition rate, there will be some search formulas that do not meet the standards. For example, if the set mean repetition rate is 60, and the mean repetition rate of search formula B is 45, it is considered that search formula B should be eliminated;
[0054] (4) The search results whose ratio of the number of occurrences in the first search formula set to the number of pre-selected search formulas in the first search formula set is greater than 50% are defined as core search results. This is easy to understand. For example, if there are 1,000 search results in the first search formula set and a search result appears more than 500 times, then this search result is considered a core search result.
[0055] (5) Compare the core search results with the search formula results in the search formula database to find a second search formula set, where the search formulas in the second search formula set meet the following conditions: covering more than 60% of the core search results, and the proportion of the core search results in the search results of the search formula is greater than 20%; because ordinary semantic analysis will not only select some search formulas that do not meet the standards, but also miss many search formulas that meet the standards, but after the above screening, the core search results have been found, and a search result coverage ratio standard is set, such as 60% in this embodiment. If it is higher than this standard, it is considered to be a search formula that meets the standards;
[0056] (6) The first search formula set and the second search formula set are combined to obtain search formulas with similar semantics, and the repeated search formulas in the first and second search formula sets are used as core search formulas. The first search formula set and the second search formula set are established based on two standards. The standard for the first search formula set is: semantic analysis software determination + average repetition rate; and the standard for determining the second search formula set is: core search result coverage rate. The result sets screened out by different standards are necessarily different, but both are considered to be search formula sets that meet the requirements;
[0057] As a preferred solution, the retrieval of image information of the retrieval method includes the following steps:
[0058] Get the original image: The original image is the retrieval target provided by the user; for example, an X-ray taken by a doctor for a patient;
[0059] Capture images of key positions: perform overlay comparison of the original image and the reference image in the image database at the same proportion, and capture images of key positions for retrieval. What needs to be understood here is that there are often multiple different problems in the X-ray film shooting part, and if the entire X-ray film is used for retrieval, there will often be a situation where the image in the database contains not only the problem to be retrieved, but also other problems. If repeated comparisons are performed in this way, the results are often unsatisfactory. Therefore, a part can be captured, such as using a smallest circle that can encircle the image of the part to be retrieved as the standard circle, and a circle that is concentric with the standard circle but has a radius greater than a certain proportion of the standard circle as the key position image, such as a circle with a radius of 1.2 times the standard circle. The reason here is that other problems appear within a certain range around the part to be retrieved, which may affect the diagnosis of the disease and the determination of the plan. This ratio can be set according to different parts and needs, such as 1.5, 1.3, etc.
[0060] Image positioning at key positions: set a reference point, establish a virtual coordinate system with the reference point as the origin, and determine the position coordinates of the key position image; the key positions of the human body can be used as the coordinate origin of the image, such as the knee, ankle, etc. Since each person's size is different, the images in the database and the images used for retrieval are compared at the same scale, but in order to eliminate the influence of this factor, the scaling ratio will be marked on each image, and the doctor can directly restore it when using it.
[0061] A key position image is input at the search position, and the key position image has a virtual coordinate system; a virtual coordinate system is generated at the same reference point in the medical care database, and an image area with the same coordinates as the key position image is intercepted as a search comparison area. The key position image and the search comparison area are compared for coverage at the same proportion, and an image with a coverage rate that meets the set value is displayed as a search result.
[0062] An intelligent information retrieval system for massive medical care data, comprising:
[0063] A medical care database, wherein the medical care database includes text information and image information;
[0064] Search formula and search result matching module: According to preset rules, after the user enters the search formula, the search results are displayed to the customer;
[0065] Retrieval formula database: When a user searches, the user's search formula is collected to obtain a search formula database;
[0066] User click result collection module: The user clicks on the search result they need based on the displayed search results. The search result clicked by the user is determined to be a valid search result. The valid search results are mapped to the search terms entered by the user to form a mapping subset.
[0067] Merge module: Merges the mapping subsets obtained from semantically similar search expressions. When merging mapping subsets, only one identical search result is retained and the number of repetitions is marked in the search results.
[0068] Mapping set recommendation module: When the number of search results clicked by the user that are identical to the results in a certain mapping set reaches a set value, the mapping set is recommended to the user.
[0069] It should be noted that the above embodiments can be freely combined as needed. The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that the present invention is susceptible to various modifications and variations. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.
Claims
1. A method for intelligent information retrieval of massive medical care data, characterized in that: The steps include: S1. constructing a medical care database, wherein the medical care database includes text information and image information; S2. Matching search terms with search results: After the user enters the search terms, the search results are displayed to the customer according to the preset rules; S3. Search formula collection: when the user searches, the user's search formula is collected to obtain a search formula database; S4. Collection of user clicked results: The user clicks on the search result required by the user according to the displayed search results, and the search result clicked by the user is determined as a valid search result, and the valid search result is mapped to the search formula entered by the user to form a mapping subset; S5. Merge the mapping subsets to obtain a mapping set: merge the mapping subsets obtained by semantically similar search expressions. When merging the mapping subsets, only one identical search result is retained and the number of repetitions is marked in the search results. S6. Mapping set recommendation: When the number of search results clicked by the user that are identical to the results in a certain mapping set reaches a set value, the mapping set is recommended to the user.
2. The intelligent information retrieval method for massive medical care data according to claim 1, characterized in that: After collecting the user's search formulas, the step S3 performs the following processing: define two search formulas with a search result repetition rate of not less than 80% as mutually interoperable search formulas, mark the number of interoperable search formulas of each search formula as a quantity label, and link the quantity label to the corresponding interoperable search formula.
3. The method for intelligent information retrieval of massive medical care data according to claim 2, characterized in that: The mapping subset in step S4 is divided into two virtual areas, the first area is the valid search results, and the second area is the potential valid search results; the potential valid search results are determined according to the following steps: all mapping subsets are repeatedly compared, and the mapping subset whose number of valid search results in the first area accounts for a ratio greater than a set ratio of the valid search results in the first area is taken as a potential mapping subset, and the search results other than the valid search results in the potential mapping subset are placed in the second area as potential valid search results.
4. The method for intelligent information retrieval of massive medical care data according to claim 3, characterized in that: In step S4, the search result clicked by the user is determined to be a valid search result based on the time the user spends browsing the search result. When the user browses for more than a set time, the clicked result is determined to be a valid search result.
5. An intelligent information retrieval system for massive medical care data, characterized in that: include: A medical care database, wherein the medical care database includes text information and image information; Search formula and search result matching module: according to the preset rules, after the user enters the search formula, the search results are displayed to the customer; Retrieval formula database: when a user searches, the user's search formula is collected to obtain a search formula database; User click result collection module: The user clicks on the search result the user needs based on the displayed search results, and the search result clicked by the user is determined as a valid search result. The valid search result is matched with the search formula input by the user to form a mapping subset; Merge module: merges the mapping subsets obtained by semantically similar search expressions. When merging the mapping subsets, only one identical search result is retained and the number of repetitions is marked in the search results. Mapping set recommendation module: When the number of search results clicked by the user that are identical to the results in a certain mapping set reaches a set value, the mapping set is recommended to the user.
Citation Information
Patent Citations
Method and device for establishing coupled relation between documents
CN101097574A
Search recommendation term pushing method and device
CN108197244A
Retrieval system for knowledge base in medical industry
CN110389997A
Search term recommendation method and device
CN110532454A
