Internet hospital patient propaganda and education method based on recommendation algorithm

Abstract

The invention provides an Internet hospital patient propaganda and education method based on a recommendation algorithm, and the method comprises the steps: collecting the historical browsing behaviors of real-name patients in the hospital Internet, and constructing a relation graph between the patients and propaganda and education articles; mining similarity of propaganda and education articles according to patient behaviors; recommending by utilizing the similarity between the propaganda and education articles; and recommending the first K propaganda and education articles with the highest interest to different patients on the home page of the Internet hospital. Personalized recommendation not only can improve the disease prevention and treatment ability of the patient, but also can guide the public to pay attention to more professional and valuable health propaganda and education knowledge to a certain extent, and has social education significance.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to an Internet hospital patient education method based on a recommendation algorithm. Background Art

[0002] With the development of the Internet, various medical hotspots are happening all the time, but this knowledge is mixed and of varying quality. Patients who lack a medical background find it difficult to identify the most valuable information. Medical institutions bear the social responsibility of patient education. Large tertiary hospitals usually have their publicity departments push a fixed number of education articles with fixed content every week through their official WeChat accounts, or organize the education knowledge by department or disease. Patients are required to actively search through the vast amount of knowledge, and there is a lack of attention to the correlation between the articles and the patients' diseases and interests.

[0003] One feasible solution to the above problem is to use recommendation algorithms as an effective tool for health knowledge education. Based on the mining of patients' historical data, knowledge with high relevance to patients' basic information, medical records, preferences and behaviors can be screened out, and then recommended on the hospital's official Internet hospital mini program to improve the personalization and accuracy of education.

[0004] A common recommendation algorithm based on user similarity is to find a set of people with similar interests to the target user, guess the target user's rating of the item based on the users' ratings of the item to be recommended and their similarity with the target user, and then take the top K items.

[0005] The shortcomings of the recommendation algorithm based on user similarity are: first, in terms of performance, it is only suitable for occasions with fewer users. If there are many users, the cost of calculating the user similarity matrix is ​​very high. However, the number of outpatients in large tertiary hospitals is huge, far greater than the number of health knowledge articles edited by experts in the hospital; second, it is only suitable for areas where user personalization is not obvious, while patients' demand for health knowledge is constantly changing with the change of disease, time, and social hot spots, and has obvious personalized needs; third, when users have new behaviors, it does not necessarily cause immediate changes in the recommendation results. Summary of the invention

[0006] The present invention provides an Internet hospital patient education method based on a recommendation algorithm, comprising:

[0007] Collect the historical browsing behavior of real-name patients on the hospital Internet and build a relationship diagram between patients and education articles;

[0008] Mining similarities in educational articles based on patient behavior;

[0009] Use similarities between missionary articles to make recommendations;

[0010] On the homepage of the Internet hospital, the top K educational articles of greatest interest are recommended to different patients.

[0011] Furthermore, the relationship diagram between patients and educational articles is constructed, including:

[0012] Set the click behavior to 1 point, the full text browsing behavior to 1 point, the like behavior to 1 point, the collection behavior to 1 point, and the forwarding behavior to 1 point. The accumulated points are the patient's interest score for the educational article.

[0013] Furthermore, the relationship diagram between patients and educational articles is constructed, including:

[0014] For patients who have not interacted with any educational articles, we recommend initial educational articles to them based on their most recent diagnosis information. Based on the patient's real-name authentication information in the Internet hospital, we link the patient's basic information and multi-dimensional diagnosis and treatment behaviors in the hospital to obtain the diagnostic ICD code. We invite experts in the hospital to mark the articles with diagnostic marks. Based on the patient's diagnosis, we make the first article recommendation for him.

[0015] Furthermore, similarities of educational articles mined based on patient behavior include:

[0016] Record the patient's interest in the jth educational article he has interacted with as like(patient, a j ), the similarity between article i and article j is sim(a i , a j ); Patient refers to a patient. This function is used to calculate the patient's interest score for a certain article. According to the above set rules, for example, if the patient clicks, browses the entire article, and collects article a1, then like(patient, a1) = 1 + 1 + 1 = 3;

[0017] Create an index of (patients → educational articles) to record the IDs of the educational articles that each patient has recently clicked and interacted with. Based on this index, given any patient ID, find the list of m educational articles that he is interested in recently and the corresponding interest scores:

[0018] patient→((a1, like(patient, a1), (a2, like(patient, a2),…, (a m , like(patient, a m ));

[0019] Create an index of (education article → education article) and calculate the similarity between education articles. Each article indexes the K articles that are most similar to it. Based on this index, given any education article ID, quickly find its K most similar article IDs and similarity scores:

[0020] article→((a1, sim(article, a1)), (a2, sim(article, a2)), ..., (a k ,sim(article,a k )));

[0021] Similarity calculation, the similarity between the candidate article and the jth article is sim(a j , article), the main idea is that the higher the audience overlap of the two educational articles, the stronger the relevance of the two educational articles; the formula for estimating the patient's interest in the candidate educational article is:

[0022]

[0023] Calculate the similarity sim(a) between the candidate education article and the education articles of interest to the patient j , article); let the patients who like article a1 be set P1, and the patients who like article a2 be set P2, and define the set P2 that likes both a1 and a2. The similarity between the two articles is:

[0024]

[0025] Considering the patient's interest in each article, each article is represented as a sparse vector, and each element of the vector corresponds to a patient. The similarity is the cosine of the angle between the two vectors, so cosine similarity is used to represent sim(a1, a2);

[0026]

[0027] Substitute the above sim(a1, a2) similarity formula into the formula for estimating the patient's interest in the candidate education article Given any patient ID and any candidate article, the patient's interest in the candidate article can be calculated.

[0028] Further, recommendations based on similarities between missionary articles include:

[0029] If the patient is interested in education article a1, and education articles a1 and a2 are similar, then the user is also interested in article a2;

[0030] Given a patient ID, find the list of educational articles that the patient is interested in recently through the (user->article) index established in the offline phase L n ;

[0031] For L nFor each article in the collection, use the previous (article->article) index to obtain the K articles most similar to each article, recorded as top-k, and the total number of candidate articles is at most nk;

[0032] For the top-k article set obtained, use the formula to estimate the patient's interest score for each article;

[0033] Sort the missionary articles by scores from high to low, and return the K articles with the highest scores;

[0034] Based on the patient's real-name ID registered in the Internet hospital, these K articles are inserted into the carousel on the homepage as the recommended knowledge for this education.

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

[0036] 1. The similarity between the educational articles in this method is calculated based on the implicit scores of millions of patients and can remain stable over a period of time. Therefore, it has the characteristics of large offline computing volume and small online computing volume. Different scores are assigned to different behaviors of patients to ensure that after the patient has online behavioral data, the system can quickly update the recommended articles.

[0037] 2. Improve patient experience. Personalized recommendations can help patients save time in screening articles and directly obtain health education knowledge content of interest, thereby greatly improving patient experience.

[0038] 3. Enhance the efficiency of original health knowledge dissemination and create a large-scale tertiary hospital brand effect. Through personalized and accurate recommendations, health education knowledge can be quickly delivered to the patient groups most likely to be interested, improving the dissemination efficiency and influence of original health education knowledge.

[0039] 4. Improve patient stickiness. Personalized health education knowledge can serve as an important tool for Internet hospitals to attract and retain patients, and help hospitals build user groups.

[0040] 5. Possessing certain social value, personalized recommendations can not only enhance patients’ ability to prevent and treat diseases, but also, to a certain extent, guide the public to not blindly believe in or follow the crowd, but to pay attention to more professional and valuable health education knowledge, which has social educational significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a flow chart of the four stages of the present invention;

[0042] Figure 2 It is a flow chart of the online recall phase. DETAILED DESCRIPTION

[0043] The present invention aims to solve the technical problem that the health education knowledge recommendation content in the prior art has poor segmentation and cannot display to patients the education knowledge that is highly relevant to their health information and interests.

[0044] The present invention relates to the field of health knowledge education for Internet hospital patients, and in particular to a personalized education method for Internet hospital patients based on an ItemCF recommendation algorithm.

[0045] It includes four stages: the data collection stage, which collects the historical browsing behavior of real-name patients in the hospital's Internet hospital education module and constructs a (patient, education article) relationship diagram; the offline calculation stage, which mines the similarity of education articles based on patient behavior; the online recall stage, which uses the similarity between education articles to make recommendations; the education article display stage, which recommends the top K education articles of greatest interest to different patients in the carousel module on the Internet hospital homepage.

[0046] During the data collection stage, since each patient has interacted with several educational articles, it can be assumed that the educational articles that the patient has clicked, liked, collected, and forwarded are the educational articles that the patient is interested in.

[0047] Quantify the patient's interest in the educational article according to the patient's behavior, assign different scores to different patient behaviors to distinguish the degree of similarity of different educational contents. The patient's interest in the educational article that has not been read can be predicted through the educational articles that have been interacted with. Preferably, according to the actual application scenario and needs, the click behavior can be set to 1 point, the full text browsing behavior can be set to 1 point, the like behavior can be set to 1 point, the collection behavior can be set to 1 point, and the forwarding behavior can be set to 1 point. The accumulated value is the patient's interest score for the educational article.

[0048] Optionally, to solve the cold start problem, for patients who have not interacted with any educational articles, we recommend initial educational articles based on the patient's most recent diagnosis information. Specifically, based on the patient's real-name authentication information in the Internet hospital, we associate the patient's basic information and multi-dimensional diagnosis and treatment behaviors in the hospital to obtain the diagnostic ICD code; we invite experts in the hospital to mark the articles with diagnostic information. Based on the patient's diagnosis, we recommend the patient the first article.

[0049] Offline calculation stage

[0050] (1) Record the patient's interest in the jth educational article that he / she has interacted with as like(patient, a j ), the similarity between article i and article j is sim(a i , a j ).

[0051] (2) Establish an index of (patient→educational article) to record the ID of the education article that each patient has recently clicked on and interacted with. Based on this index, given any patient ID, we can find a list of m education articles that he is interested in recently and the corresponding interest score.

[0052] patient→((a1, like(patient, a1), (a2, like(patient, a2),…, (a m , like(patient, a m ))

[0053] (3) Establish an index of (missionary article → missionary article) and calculate the similarity between missionary articles. Each article indexes the K articles that are most similar to it. Based on this index, given any missionary article ID, its K most similar article IDs and similarity scores can be quickly found.

[0054] article→((a1, sim(article, a1)), (a2, sim(article, a2)), ..., (a k ,sim(article,a k )))

[0055] (4) Similarity calculation module, record the similarity between the candidate propaganda article article and the jth propaganda article as sim(a j , article), the main idea is that the higher the audience overlap of the two educational articles, the stronger the relevance of the two educational articles. The formula for estimating the patient's interest in the candidate educational article is

[0056]

[0057] (5) Calculate the similarity sim(a) between the candidate education article and the education articles of interest to the patient j , article). Let the patients who like article a1 be set P1, and the patients who like article a2 be set P2. Define the set P1 that likes both a1 and a2. The similarity between the two articles is:

[0058]

[0059] Considering the patient's interest in each article, each article is represented as a sparse vector. Each element of the vector corresponds to a patient. The similarity is the cosine of the angle between the two vectors. Therefore, cosine similarity is used to represent sim(a1, a2).

[0060]

[0061] (6) Substitute the above sim(a1, a2) similarity formula into the formula for estimating the patient's interest in the candidate education article. Given any patient ID and any candidate article, the patient's interest in the candidate article can be calculated.

[0062] In the online recall phase, if the patient is interested in educational article a1, and educational articles a1 and a2 are similar, then the user is likely to be interested in article a2 as well.

[0063] (1) Given a patient ID, find the list of educational articles that the patient is interested in recently through the (user->article) index established in the offline phase. n .

[0064] (2) For L n For each article in the collection, we use the previous (article->article) index to get the K articles most similar to each article, recorded as top-k, and the total number of candidate articles is at most nk.

[0065] (3) For the top-k article sets obtained, use the formula to estimate the patient’s interest score for each article.

[0066] (4) Sort the missionary articles by scores from high to low and return the K articles with the highest scores.

[0067] (5) Based on the patient's real-name ID registered in the Internet hospital, insert these K articles into the carousel on the homepage as recommended knowledge for this education campaign.

[0068] The present invention innovatively applies the recommendation algorithm to the patient education scenario of large-scale Internet hospitals. This method considers the patient's behavioral data, assigns scores to different behaviors, and ensures that the system can be quickly updated after the patient has behavioral data online. In addition, the similarity between the education articles of the recommended articles is calculated based on the implicit scores of millions of patients, which can remain stable over a period of time. Therefore, it has the characteristics of large offline calculation amount and small online calculation amount. It is conducive to realizing personalized education and enhancing the efficiency of original health knowledge dissemination.

[0069] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0070] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0071] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A patient education method for an Internet hospital based on a recommendation algorithm, characterized in that: include: Collect the historical browsing behavior of real-name patients on the hospital Internet and build a relationship diagram between patients and education articles; Mining similarities in educational articles based on patient behavior; Use similarities between missionary articles to make recommendations; On the homepage of the Internet hospital, the top K educational articles of greatest interest are recommended to different patients.

2. The Internet hospital patient education method based on recommendation algorithm according to claim 1 is characterized in that: Constructing a relationship diagram between patients and educational articles specifically includes: Set the click behavior to 1 point, the full text browsing behavior to 1 point, the like behavior to 1 point, the collection behavior to 1 point, and the forwarding behavior to 1 point. The accumulated points are the patient's interest score for the educational article.

3. The Internet hospital patient education method based on recommendation algorithm according to claim 1 is characterized in that: Constructing a relationship diagram between patients and educational articles specifically includes: For patients who have not interacted with any educational articles, we recommend initial educational articles to them based on their most recent diagnosis information. Based on the patient's real-name authentication information in the Internet hospital, we link the patient's basic information and multi-dimensional diagnosis and treatment behaviors in the hospital to obtain the diagnostic ICD code. We invite experts in the hospital to mark the articles with diagnostic marks. Based on the patient's diagnosis, we make the first article recommendation for him.

4. The method for patient education in an Internet hospital based on a recommendation algorithm according to claim 3, characterized in that: Mining similarities in educational articles based on patient behavior includes: Record the patient's interest in the jth educational article he interacted with as like(patient, a j ), the similarity between article i and article j is sim(a i , a j ); Create an index of (patients → educational articles) to record the IDs of the educational articles that each patient has recently clicked and interacted with. Based on this index, given any patient ID, find the list of m educational articles that he is interested in recently and the corresponding interest scores: patient→((a1,like(patient,a1),(a2,like(patient,a2),…,(a m ,like(patient,a m )); Create an index of (education article → education article) and calculate the similarity between education articles. Each article indexes the K articles that are most similar to it. Based on this index, given any education article ID, quickly find its K most similar article IDs and similarity scores: article→((a1,sim(article,a1)),(a2,sim(article,a2)),…,(a k ,sim(article,a k ))); Similarity calculation, the similarity between the candidate article and the jth article is sim(a j , article), the main idea is that the higher the audience overlap of the two educational articles, the stronger the relevance of the two educational articles; the formula for estimating the patient's interest in the candidate educational article is: Calculate the similarity sim(a) between the candidate education article and the education articles of interest to the patient j , article); let the patients who like article a1 be set P1, and the patients who like article a2 be set P2, and define the set that likes both a1 and a2 The similarity between the two articles is: Considering the patient's interest in each article, each article is represented as a sparse vector, and each element of the vector corresponds to a patient. The similarity is the cosine of the angle between the two vectors, so cosine similarity is used to represent sim(a1, a2); Substitute the above sim(a1, a2) similarity formula into the formula for estimating the patient's interest in the candidate education article Given any patient ID and any candidate article, the patient's interest in the candidate article can be calculated.

5. The Internet hospital patient education method based on recommendation algorithm according to claim 1 or 4, characterized in that: Recommendations based on similarities between missionary articles include: If the patient is interested in education article a1, and education articles a1 and a2 are similar, then the user is also interested in article a2; Given a patient ID, find the list of educational articles that the patient is interested in recently through the (user->article) index established in the offline phase L n ; For L n For each article in the collection, use the previous (article->article) index to obtain the K articles most similar to each article, recorded as top-k, and the total number of candidate articles is at most nk; For the top-k article set obtained, use the formula to estimate the patient's interest score for each article; Sort the missionary articles by scores from high to low, and return the K articles with the highest scores; Based on the patient's real-name ID registered in the Internet hospital, these K articles are inserted into the carousel on the homepage as the recommended knowledge for this education.

Images