Drug data recommendation method and system based on neural network learning, and medium

Abstract

The invention provides a medicine data recommendation method and system based on neural network learning and a medium, and the method comprises the steps: S1, when a medicine database is established, guaranteeing the consistency and accuracy of terms by using a standardized medical term set, and carrying out the importance statistics of a medicine keyword set through an IDF (Inverse Document Function); s2, accurately estimating an association probability between the main node and the sub-nodes based on expert recommended drug data or expert knowledge; the correlation degree is calculated through importance statistics; and S3, setting a recommendation weight according to the keyword importance statistics and the correlation degree, and performing drug data recommendation operation according to the set recommendation weight.

Description

Technical Field

[0001] The present invention relates to the field of big data analysis, and in particular to a drug data recommendation method, system and medium based on neural network learning. Background Art

[0002] The prior art is to obtain a large amount of matching data information in the database, especially in the field of medical data. Since the massive data cannot automatically match the user's needs, the vast user group cannot match the drugs they want from the massive drug SKUs through multi-dimensional drug attribute search, and cannot complete the search for highly subjective combination conditions such as product name, manufacturer, specification, indication description, etc. If these data can be converted into valuable information and knowledge, the accurate matching of diseases and drug indications to target drugs will greatly improve drug search. In CN110085292A, the target function is obtained according to the implicit feature information and operation behavior information, and the recommended drugs are obtained through the processed neural network. However, if the required data is queried in the database, the recommendation content becomes slower and slower due to the mismatch of the data calculation method, and the user cannot retrieve the data he wants to query, which increases the user's waiting time and reduces the work efficiency. It is a technical problem that technicians in this field urgently need to solve. Summary of the invention

[0003] The present invention aims to at least solve the technical problems existing in the prior art, and in particular innovatively proposes a drug data recommendation method, system and medium based on neural network learning.

[0004] In order to achieve the above-mentioned object of the present invention, the present invention provides a drug data recommendation method based on neural network learning, comprising:

[0005] S1, when establishing a drug database, use a standardized medical terminology set to ensure the consistency and accuracy of the terminology, and use IDF to perform importance statistics on the drug keyword set;

[0006] S2, accurately estimates the association probability between the main node and the sub-nodes based on expert recommended drug data or expert knowledge; calculates the association degree through importance statistics;

[0007] S3, setting recommendation weights according to keyword importance statistics and relevance, and performing drug data recommendation operations according to the set recommendation weights.

[0008] Preferably, the above technical solution is that S1 comprises:

[0009] Using IDF calculation of retrieval and text mining, the IDF is the inverse document frequency, and the drug data set of the drug database is determined. According to the importance statistics of drug keywords, the IDF formula is:

[0010]

[0011] t is a specific drug word; D is a drug word set; N is the total number of words in the drug word set;

[0012] In |{d∈D:t∈d}|, d is the number of words containing the specific drug word t, and k is the smoothing factor, which helps to enhance the robustness of the algorithm.

[0013] Set the minimum drug word frequency threshold C min , calculated in C min The IDF value of the keywords that appear in the drug word set is adjusted to filter out overly sparse keywords, thereby improving the performance of the recommendation system.

[0014] If the set of drugs in which a specific drug word t appears is less than C min , then its IDF value is set to 0, indicating that this drug word is not considered in the recommendation system.

[0015] Preferably, the above technical solution, S1 further comprises:

[0016] Set to:

[0017]

[0018] In the above technical solution, preferably, S2 comprises:

[0019] The calculation of association probability is based on conditional probability, which needs to take into account the preference information of the user's input keywords:

[0020] Association probability P b (A|B): The association probability between the user input keyword A and the related drug B calculated based on prior knowledge.

[0021] In order to simultaneously consider the impact of user preferences, drug categories, and expert knowledge on operational behavior events, the probability function is calculated The calculated probability that the operation behavior event of expert knowledge i for expert recommendation of drug j belongs to the specific event set E, η ij is the operational behavior event of expert knowledge i recommending drug j to the expert, K m is the feature matrix of expert knowledge, L n is the feature matrix of expert recommended drugs, ω mi is the behavioral weight of user preference m to expert knowledge i; c nj K is the weight of the expert's recommendation of the category n of drug j on the expert's knowledge, si is the feature vector of expert knowledge i, L sjis the feature vector of the expert’s recommendation for drug j, f(*) is the correlation function used to calculate the interaction between them, and Πf(*) is the multiplication used to calculate the combined impact of all relevant factors.

[0022] Preferably, the above technical solution, S2 further includes:

[0023] In order to Incorporate it into the association probability between the user input keyword A and the related drug B,

[0024] Use the weighted average method to fuse this information. The specific steps are as follows:

[0025] Calculate the weighted average:

[0026] For each expert knowledge i, we use the probability of its expert knowledge on the operation behavior event of the expert recommending the drug as the weight and the basic association probability P b Multiply (A|B) and then add all the results together to get the weighted average probability;

[0027] Weighted Average=∑ i,j p(η ij ∈E|K m ,L n ,ω mi ,c nj )·P b (A|B),

[0028] Normalize and multiply the probability of each expert knowledge by P b (A|B),

[0029]

[0030] Thus, the association probability P between the fused disease and the drug is obtained. f (A|B).

[0031] In the above technical solution, S3 preferably includes:

[0032] The calculation of the comprehensive recommendation weight is in the form of a weighted sum, where the IDF and conditional probability are multiplied by different weight coefficients to reflect their importance in the recommendation. The specific formula is as follows:

[0033] Q=α·IDF(t,D)+β·P final (A|B),

[0034] Among them, α is the weight of the importance statistic, and β is the weight of the association probability.

[0035] The present invention also discloses a drug data recommendation system based on neural network learning, comprising:

[0036] Importance statistics module, which is used to use standardized medical terminology to ensure the consistency and accuracy of terms when establishing a drug database, and to perform importance statistics of drug keyword sets through IDF;

[0037] The association module is used to accurately estimate the association probability between the main node and the sub-nodes based on expert recommended drug data or expert knowledge; and calculate the association degree through importance statistics;

[0038] The comprehensive recommendation module is used to set recommendation weights based on keyword importance statistics and relevance, and perform drug data recommendation operations based on the set recommendation weights.

[0039] The present invention also discloses a computer-readable storage medium, wherein the computer-readable storage medium stores a processing program including the method described in claim 1.

[0040] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0041] The present invention improves the consistency and accuracy of recommended data, and uses a standardized medical terminology set to establish a drug database, which helps to reduce misunderstandings and confusion caused by inconsistent or incorrect terms, thereby improving the reliability of data processing and analysis. The importance statistics of drug keyword sets are performed using IDF (inverse document frequency), which can more accurately reflect the relative importance of each keyword in a specific context. By accurately estimating the association probability between the main node and the child node through expert recommendation drug data or expert knowledge, and calculating the association degree in combination with importance statistics, this method can more deeply explore the intrinsic connections and laws between drug data. This helps to improve the accuracy and effectiveness of the recommendation system. According to the keyword importance statistics and association degree, the recommendation weight is set, and the drug data recommendation operation is performed according to these weights. This method can meet the needs of users in a more personalized and accurate manner. This helps to improve user experience and satisfaction, and also helps to improve the efficiency of drug matching data.

[0042] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0044] Figure 1 It is a schematic diagram of the working method of the present invention;

[0045] Figures 2 to 6 It is a working state diagram of drug data retrieval recommendation of the present invention;

[0046] Figure 7 It is an implementation effect diagram of the present invention. DETAILED DESCRIPTION

[0047] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0048] like Figures 1 to 7 As shown, the present invention discloses a drug data recommendation method based on neural network learning, comprising the following steps:

[0049] As the degree of standardization of medicines becomes increasingly higher, professional terms will be stored in the database of the products produced. During the user's use, browsing and collection process, the medicine database will record the corresponding user traces and perform medicine data recommendation operations based on user needs.

[0050] S1. When establishing a drug database, use standardized medical terminology sets (such as ICD, SNOMED CT, etc.) to ensure the consistency and accuracy of terminology, and use IDF to perform importance statistics on drug keyword sets;

[0051] S2, accurately estimates the association probability between the main node and the sub-nodes based on expert recommended drug data or expert knowledge; calculates the association degree through importance statistics;

[0052] S3, setting recommendation weights according to keyword importance statistics and relevance, and performing drug data recommendation operations according to the set recommendation weights.

[0053] The S1 includes:

[0054] Using IDF calculation of retrieval and text mining, the IDF is the inverse document frequency, and the drug data set of the drug database is determined. According to the importance statistics of drug keywords, the IDF formula is:

[0055]

[0056] t is a specific drug word; D is a drug word set; N is the total number of words in the drug word set;

[0057] In |{d∈D:t∈d}|, d is the number of words containing the specific drug word t, and k is the smoothing factor, which helps to enhance the robustness of the algorithm.

[0058] Set the minimum drug word frequency threshold C min , calculated in C minThe IDF value of the keywords that appear in the drug word set is adjusted to filter out overly sparse keywords, thereby improving the performance of the recommendation system.

[0059] If the set of drugs in which a specific drug word t appears is less than C min , then its IDF value is set to 0, indicating that this drug word is not considered in the recommendation system.

[0060] Set to:

[0061]

[0062] The search capability is configured based on the scenario of the user input keywords and the expected recommended drug data. The basic scenario of the user is met and the keyword matching search is performed based on the main function and drug name of the drug data. The contents that need to be considered include: the first letter of the product pinyin, the manufacturer's name, the product specification information, the product type (medical insurance, etc.), the indication capability, the search capability, etc.

[0063] The S2 includes:

[0064] Through actual drug data or expert knowledge data, the association probability between the main node (such as expert knowledge) and the child node (such as expert recommended drugs) is accurately estimated.

[0065] The calculation of association probability is based on conditional probability, which needs to take into account the preference information of the user's input keywords:

[0066] Association probability P b (A|B): The association probability between the user input keyword A (including disease, drug name, chemical name, common name) and the related drug B calculated based on prior knowledge.

[0067] In order to simultaneously consider the impact of user preferences, drug categories, and expert knowledge on operational behavior events, the probability function is calculated The calculated probability that the operation behavior event of expert knowledge i for expert recommendation of drug j belongs to the specific event set E, η ij is the operational behavior event of expert knowledge i recommending drug j to the expert, K m is the feature matrix of expert knowledge, L n is the feature matrix of expert recommended drugs, ω mi is the behavioral weight of user preference m to expert knowledge i; c nj is the weight of the expert's recommended drug category n on the expert's knowledge, that is, the weight of the expert's recommended drug category, K si is the feature vector of expert knowledge i, L sjis the feature vector of the expert's recommendation for drug j, f(*) is the correlation function, which is used to calculate the interaction between them, and Πf(*) is the multiplication, which is used to calculate the comprehensive impact of all relevant factors.

[0068] In order to Integrating it into the association probability between the user input keyword A and the related drug B requires expert knowledge to regard the expert's operational behavior of recommending drugs as a kind of evidence, which supports or does not support the association between the input keyword A and the related drug B.

[0069] The probability of the expert knowledge-expert recommended drug operation behavior event is regarded as evidence to support or not support the association between input keyword A and related drug B. If the operation behavior shows that the expert knowledge tends to associate the drug with the disease, then the probability value is high; otherwise, it is low.

[0070] Use the weighted average method to fuse this information. The specific steps are as follows:

[0071] Calculate the weighted average:

[0072] For each expert knowledge i, we use the probability of its expert knowledge on the operation behavior event of the expert recommending the drug as the weight and the basic association probability P b Multiply (A|B) and then add all the results together to get the weighted average probability;

[0073] Weighted Average=Σ i,j p(η ij ∈E|K m ,L n ,ω mi ,c nj )·P b (A|B),

[0074] Normalize and multiply the probability of each expert knowledge by P b (A|B), the result needs to be normalized to ensure that the final probability value is between 0 and 1.

[0075]

[0076] Thus, the association probability P between the fused disease and the drug is obtained. f (A|B).

[0077] According to P final The value of (A|B) can explain the strength of association between the input keyword A and the related drug B. f (A|B) is close to 1, indicating P final (A|B) has a strong correlation; if it is close to 0, it means the correlation is weak.

[0078] By calculating P final (A|B), the system can evaluate the matching degree between the recommended drug D and the user's needs when the user inputs a specific keyword E. This probability-based evaluation method helps to filter out irrelevant or low-relevance drugs, thereby improving the accuracy of recommendations. As the user input keywords change, the probability value of the recommended drug is calculated and updated in real time, thereby dynamically adjusting the recommendation list to ensure that the recommendation results are always highly consistent with the user's current needs.

[0079] Combined with the user's historical search behavior data (such as browsing history, purchase history, etc.), the system can further refine P final The calculation of (A|B) takes into account the user's personal preferences and medication habits. This personalized probability calculation helps improve the user experience of the recommendation system and makes users feel more intimate and professional service.

[0080] Traditional recommendation methods may not be able to effectively cover the demand for drugs in some unpopular or specialized fields. final (A|B), the system can capture these long-tail demands more accurately and provide users with corresponding recommendations to meet more diverse medication needs.

[0081] Through the recommendation system, users can final The recommended drug list calculated by (A|B) and accurate and personalized drug recommendations can enhance users’ trust and satisfaction with the recommendation system, thereby enhancing user stickiness and promoting users to continue using the recommendation system and share it with more people.

[0082] The S3 includes:

[0083] Combine IDF and conditional probability to enhance the performance of the drug recommendation system, use IDF to evaluate the importance of drug keywords, and use conditional probability to evaluate the association between drugs and diseases. Then, we can combine these two indicators to define a comprehensive recommendation weight for drug recommendation.

[0084] The calculation of the comprehensive recommendation weight is in the form of a weighted sum, where the IDF and conditional probability are multiplied by different weight coefficients to reflect their importance in the recommendation. The specific formula is as follows:

[0085] Q=α·IDF(t,D)+β·P final (A|B),

[0086] Among them, α is the weight of importance statistics, β is the weight of association probability,

[0087] Segment the search keywords in the user input information according to the dictionary to obtain a set of segmented words; for example, for "Amoxicillin and Clavulanate Potassium Tablets", the set of segmented words after segmentation is "阿莫", "西林", "钾片", "克拉维酸", "阿莫西林", "阿莫西林克拉维酸钾片", "片", etc.;

[0088] Query the synonym dictionary and add the synonyms to the set of segmented words; as an example, "respiratory tract infection" is an associated attribute of the indication of "Amoxicillin", and the indication "*** infection" is added to the set of segmented words.

[0089] Remove the stop words, mainly to remove the modal particles "哪", "吧", "的", "了", "呢", which are not subject to subsequent operations such as word segmentation and matching. The user obtains the keywords, such as "Amoxicillin", and synchronizes the synonyms related to the indication in this category, such as "infection", "pharyngitis", "anti-inflammatory", etc., and corresponding related words are added to the synonym dictionary in the system.

[0090] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the claims and their equivalents.

Claims

1. A drug data recommendation method based on neural network learning, characterized in that: include: S1, when establishing a drug database, use a standardized medical terminology set to ensure the consistency and accuracy of the terminology, and use IDF to perform importance statistics on the drug keyword set; S2, accurately estimates the association probability between the main node and the sub-nodes based on expert recommended drug data or expert knowledge; calculates the association degree through importance statistics; S3, setting recommendation weights according to keyword importance statistics and relevance, and performing drug data recommendation operations according to the set recommendation weights.

2. The drug data recommendation method based on neural network learning according to claim 1, characterized in that: The S1 includes: Using IDF calculation of retrieval and text mining, the IDF is the inverse document frequency, and the drug data set of the drug database is determined. According to the importance statistics of drug keywords, the IDF formula is: t is a specific drug word; D is a drug word set; N is the total number of words in the drug word set; In |{d∈D:t∈d}|, d is the number of words containing the specific drug word t, and k is the smoothing factor, which helps to enhance the robustness of the algorithm. Set the minimum drug word frequency threshold C min , calculated in C min The IDF value of the keywords that appear in the drug word set is adjusted to filter out overly sparse keywords, thereby improving the performance of the recommendation system. If the set of drugs in which a specific drug word t appears is less than C min , then its IDF value is set to 0, indicating that this drug word is not considered in the recommendation system.

3. The drug data recommendation method based on neural network learning according to claim 2 is characterized in that: The S1 further comprises: Set to:

4. The drug data recommendation method based on neural network learning according to claim 1, characterized in that: The S2 includes: The calculation of association probability is based on conditional probability, which needs to take into account the preference information of the user's input keywords: Association probability P b (A|B): The association probability between the user input keyword A and the related drug B calculated based on prior knowledge. In order to simultaneously consider the impact of user preferences, drug categories, and expert knowledge on operational behavior events, the probability function is calculated The calculated probability that the operation behavior event of expert knowledge i for expert recommendation of drug j belongs to the specific event set E, η ij is the operational behavior event of expert knowledge i recommending drug j to the expert, K m is the feature matrix of expert knowledge, L n is the feature matrix of expert recommended drugs, ω mi is the behavioral weight of user preference m to expert knowledge i; c nj K is the weight of the expert's recommendation of the category n of drug j on the expert's knowledge, si is the feature vector of expert knowledge i, L sj is the feature vector of the expert’s recommendation for drug j, f(*) is the correlation function, which is used to calculate the interaction between them, and ∏f(*) is the multiplication, which is used to calculate the combined impact of all relevant factors.

5. The drug data recommendation method based on neural network learning according to claim 4 is characterized in that: The S2 further includes: In order to Incorporate it into the association probability between the user input keyword A and the related drug B, Use the weighted average method to fuse this information. The specific steps are as follows: Calculate the weighted average: For each expert knowledge i, we use the probability of the expert knowledge on the operation behavior event of the expert recommending the drug as the weight and the basic association probability P b Multiply (A|B) and then add all the results together to get the weighted average probability; Weighted Average=∑ i,j p(η ij ∈E|K m ,L n ,ω mi ,c nj )·P b (A|B), Normalize and multiply the probability of each expert knowledge by P b (A|B), Thus, the association probability P between the fused disease and the drug is obtained. f (A|B).

6. The drug data recommendation method based on neural network learning according to claim 1, characterized in that: The S3 includes: The calculation of the comprehensive recommendation weight is in the form of a weighted sum, where the IDF and conditional probability are multiplied by different weight coefficients to reflect their importance in the recommendation. The specific formula is as follows: Q=α·IDF(t,D)+β·P final (A|B), Among them, α is the weight of the importance statistic, and β is the weight of the association probability.

7. A drug data recommendation system based on neural network learning, characterized in that: include: Importance statistics module, which is used to use standardized medical terminology to ensure the consistency and accuracy of terms when establishing a drug database, and to perform importance statistics of drug keyword sets through IDF; An association module, which is used to accurately estimate the association probability between the main node and the child nodes based on expert recommended drug data or expert knowledge; Calculate associations through importance statistics; The comprehensive recommendation module is used to set recommendation weights based on keyword importance statistics and relevance, and perform drug data recommendation operations based on the set recommendation weights.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a processing program including the method of claim 1.

Images