WO2021159640A1 - Drug recommendation method based on artificial intelligence, and related device - Google Patents

Abstract

Provided are a drug recommendation method and system based on artificial intelligence, a computer device, and a computer-readable storage medium. The method comprises the following steps: collecting historical medical data to construct an original profile (S10); pre-processing the original profile and outputting a feature data set (S20); by taking the pre-processed feature data set as training data, respectively performing pre-training to obtain an XGBOOST recommendation model and a deep neural network recommendation model (S30); obtaining a first recommendation strength probability value and a second recommendation strength probability value by means of the XGBOOST recommendation model and the deep neural network recommendation model; linearly adding the first recommendation strength probability value and the second recommendation strength probability value of each target drug to obtain each target recommendation strength probability value (S70); and according to target recommendation strength probability values, screening out a preset number of pieces of top-ranked target drug information, and pushing the target drug information to a user. According to the method, model results of the two models are linearly added, so that the drug recommendation accuracy can be effectively improved.

Description

Artificial intelligence-based drug recommendation method and related equipment

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on February 13, 2020, the application number is CN2020100906810, and the invention title is "artificial intelligence-based drug recommendation method and related equipment", the entire content of which is incorporated by reference In this application.

Technical field

This application relates to the field of smart medicine, and in particular to an artificial intelligence-based drug recommendation method, system, computer equipment, and computer-readable storage medium.

Background technique

When existing patients buy medicines, they usually choose to go to physical medicine retail stores such as chain pharmacies. Specifically, the patients describe their symptoms to the pharmacist, and the pharmacist chooses the corresponding medicines for the patient. However, this method has the following shortcomings: The level of licensed pharmacists in pharmacies may be uneven, resulting in that the recommended drugs are not necessarily suitable for patients. In addition, physical stores need to be equipped with multiple pharmacists, which causes labor costs to rise rapidly. Therefore, there are many drugs that can be recommended to users. Most of the existing drug recommendation systems are basically based on the drug's own indications to match the drug indications. It is theoretically feasible, but it ignores a large amount of specialized drug history data.

technical problem

The inventor realizes that the completeness and validity of the medical history data will determine the accuracy of drug recommendation. Therefore, how to improve the accuracy of drug recommendation is a problem that needs to be solved at present.

Technical solutions

The purpose of this application is to provide an artificial intelligence-based drug recommendation method, system, computer equipment, and computer-readable storage medium, which can effectively improve the accuracy of drug recommendation.

In order to achieve the objective, this application provides an artificial intelligence-based drug recommendation method, which includes the following steps:

Collect historical medical data to construct original portraits;

Preprocess the original image and output the feature data set;

The pre-processed feature data set is used as the training data, and the XGBOOST recommendation model and the deep neural network recommendation model are pre-trained respectively;

Obtain the user's drug request information and input it into the XGBOOST recommendation model and the deep neural network recommendation model;

The XGBOOST recommendation model calculates and obtains the first recommendation intensity probability value of the target drug corresponding to the drug request information according to the drug request information;

The deep neural network recommendation model calculates and obtains the second recommendation intensity probability value of the target drug corresponding to the drug request information according to the drug request information;

Linearly adding the first recommended intensity probability value and the second recommended intensity probability value of each of the target drugs to obtain the target recommended intensity probability value of each of the target drugs;

The target drug information of each target drug is sorted according to the corresponding target recommendation intensity probability value from high to low, the top preset number of target drug information is screened out, and the target drug information is pushed to the user.

This application also provides an artificial intelligence-based drug recommendation system, which includes:

Collection module, which is used to collect historical medical data to construct original portraits;

A preprocessing module, which is used to preprocess the original image and output a feature data set;

Model training module, which is used to pre-train the pre-processed feature data set as training data to obtain the XGBOOST recommendation model and the deep neural network recommendation model;

Input module, which is used to obtain the user's drug request information and input it into the XGBOOST recommendation model and the deep neural network recommendation model;

The processing module is used to calculate and obtain the first recommendation strength probability value of the target drug corresponding to the drug request information through the XGBOOST recommendation model, and obtain the second recommendation of the target drug corresponding to the drug request information through the deep neural network recommendation model Intensity probability value; and linearly adding the first recommended intensity probability value and the second recommended intensity probability value of each of the target drugs to obtain the target recommended intensity probability value of each of the target drugs;

The output module is used to sort the target drug information of each target drug according to the corresponding target recommendation intensity probability value from high to low, filter out the preset number of target drug information in the top ranking, and push the target drug information To the user.

The present application also provides a computer device that includes a memory, a processor, and computer-readable instructions that are stored in the memory and can run on the processor. The processor implements the computer-readable instructions when the processor executes the computer-readable instructions. Steps of artificial intelligence-based drug recommendation method:

Collect historical medical data to construct original portraits;

Preprocess the original image and output the feature data set;

The pre-processed feature data set is used as the training data, and the XGBOOST recommendation model and the deep neural network recommendation model are pre-trained respectively;

Obtain the user's drug request information and input it into the XGBOOST recommendation model and the deep neural network recommendation model;

The XGBOOST recommendation model calculates and obtains the first recommendation intensity probability value of the target drug corresponding to the drug request information according to the drug request information;

The deep neural network recommendation model calculates and obtains the second recommendation intensity probability value of the target drug corresponding to the drug request information according to the drug request information;

Linearly adding the first recommended intensity probability value and the second recommended intensity probability value of each of the target drugs to obtain the target recommended intensity probability value of each of the target drugs;

The target drug information of each target drug is sorted according to the corresponding target recommendation intensity probability value from high to low, the top preset number of target drug information is screened out, and the target drug information is pushed to the user.

This application also provides a computer-readable storage medium on which computer-readable instructions are stored, and when the computer-readable instructions are executed by a processor, the steps of the method for recommending drugs based on artificial intelligence are implemented:

Collect historical medical data to construct original portraits;

Preprocess the original image and output the feature data set;

The pre-processed feature data set is used as the training data, and the XGBOOST recommendation model and the deep neural network recommendation model are pre-trained respectively;

Obtain the user's drug request information and input it into the XGBOOST recommendation model and the deep neural network recommendation model;

The XGBOOST recommendation model calculates and obtains the first recommendation intensity probability value of the target drug corresponding to the drug request information according to the drug request information;

The deep neural network recommendation model calculates and obtains the second recommendation intensity probability value of the target drug corresponding to the drug request information according to the drug request information;

Linearly adding the first recommended intensity probability value and the second recommended intensity probability value of each of the target drugs to obtain the target recommended intensity probability value of each of the target drugs;

The target drug information of each target drug is sorted according to the corresponding target recommendation intensity probability value from high to low, and the preset number of target drug information in the top ranking is screened out, and the target drug information is pushed to the user.

Beneficial effect

This application combines the XGBOOST recommendation model and the deep neural network recommendation model, and linearly adds the results of the two models. The accuracy of the recommendation is 200% higher than that of the non-model, and the recommendation rate of the top three correct drugs Increased from 75% to 95%. This application extracts more hidden information from the historical medical data in the original portrait through feature expansion to ensure the integrity of feature values, and uses saturation exploration to filter and delete feature values that are not meaningful for training, thereby forming complete and effective feature data set. In addition, this application combines the XGBOOST recommendation model and the deep neural network recommendation model, which has a high degree of automation, and can be adaptively trained and predicted, which improves the efficiency of drug recommendation.

Description of the drawings

Figure 1 is a flow chart of the application of the artificial intelligence-based drug recommendation method;

FIG. 2 is a flowchart of the classification processing when the historical medical data in step S10 in FIG. 1 involves non-standard drug names;

Figure 3 is a block diagram of the artificial intelligence-based drug recommendation system of this application;

Fig. 4 is a schematic diagram of the hardware structure of the computer equipment of the artificial intelligence-based drug recommendation method of the present application.

Reference signs:

1. A drug recommendation system based on artificial intelligence 10. Acquisition module 20. Pre-processing module

30. Model training module 40. Input module 50. Processing module

60. Output module 2. Computer equipment 21. Memory 22. Processor

Embodiments of the present invention

In order to make the purpose, technical solutions, and advantages of this application clearer and clearer, the following further describes the application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the application, and are not used to limit the application. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of this application.

As shown in Figure 1, the artificial intelligence-based drug recommendation method of this application includes the process of pre-training to form a drug recommendation model and the process of performing drug recommendation based on the drug recommendation model, which includes the following steps:

Step S10: Collect historical medical data to construct an original portrait;

Step S20: preprocess the original image and output a feature data set;

Step S30: The pre-processed feature data set is used as training data, and the XGBOOST recommendation model and the deep neural network recommendation model are obtained by pre-training respectively;

Step S40: Obtain the user's drug request information and input it into the XGBOOST recommendation model and the deep neural network recommendation model;

Step S50: The XGBOOST recommendation model calculates and obtains the first recommendation intensity probability value of the target drug corresponding to the drug request information according to the drug request information;

Step S60: The deep neural network recommendation model calculates and obtains the second recommendation strength probability value of the target drug corresponding to the drug request information according to the drug request information;

Step S70: linearly add the first recommended strength probability value and the second recommended strength probability value of each target drug to obtain the target recommended strength probability value of each target drug;

Step S80: Sort the target drug information of each target drug according to the corresponding target recommendation intensity probability value from high to low, filter out the preset number of target drug information in the top ranking, and push the target drug information to the user.

Please refer to Figure 2. Figure 2 is a flow chart of the classification process when the historical medical data in step S10 in Figure 1 involves non-standard drug names. If the drug name in the historical medical data is not strictly a national standard ICD10 code, Unable to accurately locate and classify medicines. This application uses the text extraction and matching method of the readable medium to locate and match the non-standard drug names in the historical medical data, so as to facilitate the accurate positioning and classification of the non-standard drug names.

The classification of non-standard drug names includes the following steps:

Step S110: Separate the names of non-standard drugs and standard drugs.

Step S120: Based on the Word2Vec shallow neural network, and based on the skip_gram algorithm, the words in the non-standard drug name and the standard drug name are respectively converted into word vectors; for example, the word "Sense" is transformed into word2vec through the network training of Word2vec The vector [1.90334,2.9874,...,0.988] has a dimension of 5×1.

Step S130: Use the pooling rule to splice the vectors vertically to obtain the word vector splicing of non-standard drug names and standard drug names; for example, cold medicine can be transformed into a vector [1.0334,10.51184,...,1.8115], dimension 15 ×1.

Step S140: Calculate the vector distance between the non-standard drug name and the standard drug name.

Step S150: Find the standard drug name with the closest vector distance to the non-standard drug name.

Step S160: Classify the non-standard drug names according to the classification criteria of the standard drug names with the closest vector distance.

In this embodiment, the original portrait in step S10 includes basic information and medical history information. The data source of the original portrait includes multiple different data sources, and the multi-dimensional features included in the multiple different data sources are not consistent.

The basic information includes the gender, age, blood type, marriage, mobile phone number, etc. of the drug purchaser or medical staff;

The medical record information includes a summary of the medical record such as current symptoms, duration of symptoms, severe symptoms, history of surgery, medical expenses, drugs, drug formulations, and hospitalization days, hospital orders, etc. Medical history information, corresponding time of onset, medications, etc. of acute diseases such as influenza, chickenpox, hand, foot and mouth; follow-up information of chronic diseases such as hypertension, diabetes, and chronic obstructive pulmonary disease, including duration of illness, medications, etc.

Different data sources contain inconsistent multi-dimensional features. For example, sentinel hospitals will have richer and sufficient multi-dimensional features, usually including gender, age, blood type, marriage, diastolic blood pressure, systolic blood pressure, various physical examinations, sampling indicators, etc. Small and medium-sized outpatient clinics, community hospitals, etc. may only have basic information; there are even fewer brand-name pharmacies, which may only have user phone numbers, drugs, and drug formulations.

While collecting historical medical data in step S10, establish a table correspondence between the data in the historical medical data and the data in the associated database through the association extraction method, and collect the associated data in the associated database to construct the original portrait, where the associated database It is a database other than the medical field, and can be a database in multiple fields such as insurance, finance, and technology. The related database can be an internal database or an external free database (such as MySQL database). The information is expanded and filled in multiple dimensions to construct a more complete original portrait. Wherein, the primary key is an ID, such as a social security account number/ID number of a person after data desensitization.

Taking linked data as the data in the financial field as an example, by analyzing and mining the data in the financial field linked database, the user's consumption capacity characteristic dimension is established, thereby expanding the consumption capacity characteristic dimension of the personal portrait. If the financial domain related database data is Including those who did not participate in financial derivative products, and did not purchase insurance, then establish the user's consumption power characteristic dimension-low consumption power; if the data in the financial sector related database includes participating in the financial management of many financial derivatives, and purchase a variety of insurance , The user’s spending power feature dimension—high spending power is established, and the user’s spending power feature dimension is used as training data for training to obtain the XGBOOST recommendation model and the deep neural network recommendation model. The medicines recommended by the XGBOOST recommendation model and the deep neural network recommendation model trained on the training data with low spending power correspond to their spending power, that is, recommend cheap but slower drugs; XGBOOST trained on the training data with high spending power The drugs recommended by the recommendation model and the deep neural network recommendation model correspond to their spending power, that is, expensive and effective drugs are recommended.

While collecting historical medical data in step S10, other related data is also collected, and the information in the original portrait is expanded and filled in multiple dimensions through the multi-table primary key association extraction method.

In this embodiment, the preprocessing in step S20 includes feature expansion, saturation detection and screening, missing value filling, data set heterogeneity and expansion, feedback selection, and correlation screening.

The specific implementation of feature expansion is as follows. The tsfresh algorithm is used to expand, and many intermediate statistical variables such as variance, standard deviation, extreme value, various mean values and so on are calculated. For example, for every two columns of features f1 and f2, statistical indicators such as variance, standard deviation, extreme value, and various mean values are calculated. After expansion, there will be missing data in some columns, or features that are not clearly distributed, such as the entire column is 0 or -1. Such feature columns are indistinguishable, and it is meaningless to enter the drug recommendation model. Such data columns will be deleted. .

,H_m }，再进行粒子扫描，所述粒子扫描包括单维度粒子扫描与多维度粒子扫描。In this embodiment, the original portrait includes the features of an n×m dimensional matrix. When preprocessing the original portrait, the tsfresh algorithm is used to simplify the features of the n×m dimensional matrix into a horizontal splicing of several n×1 dimensional vectors. For example {H_1,H_2,

,H_m }, and then perform particle scanning, which includes single-dimensional particle scanning and multi-dimensional particle scanning.

In this embodiment, when the original image is preprocessed, the single-dimensional particle scan uses a sliding scan to derive a single-dimensional expanded feature set L1, and the multi-dimensional particle scan uses a sliding scan to derive a multi-dimensional expanded feature Set L2, and then combine the single-dimensional extended feature set, the multi-dimensional extended feature set and the original image to obtain a feature data set containing more information.

Taking a single-dimensional particle scan as an example, each of the above H is a vector of n×1 dimensions.

×1进行函数计算，该函数可以为各类统计指标如统计指标如方差、标准差、极值、均值等函数，也可以是tanh、relu等非线性函数。计算得到粒子扫描值。然后，进行窗口滑动，再进行扫描，然后再进行窗口滑动扫描循环下去直至整个单列扫描完毕，则每一个单列特征可以衍生出多列方差、标准差、极值等等多维度的特征。将衍生出来的特征concat成一个单维度扩充特征集L1。原始的特征画像中的特征结合衍生出来的单维度扩充特征集形成预处理后的特征数据集。First, define an a×1 scanning particle with a window_size of k for scanning. When scanning in a single dimension, a particle a×1 with a window_size of k is used for sliding scanning. When each sliding to the corresponding single dimension, the scanning data obtained by the particle in the single dimension a

×1 for function calculation, the function can be various statistical indicators such as statistical indicators such as variance, standard deviation, extreme value, mean value and other functions, or non-linear functions such as tanh and relu. Calculate the particle scan value. Then, window sliding is performed, scanning is performed, and then the window sliding scanning cycle is performed until the entire single column is scanned. Then, each single column feature can derive multi-dimensional features such as multi-column variance, standard deviation, extreme value, and so on. Concat the derived features into a single-dimensional extended feature set L1. The features in the original feature portrait are combined with the derived single-dimensional extended feature set to form a preprocessed feature data set.

Taking multi-dimensional particle scanning as an example, C can be taken as a list as an alternative dimension. Then take out c1, c2, etc. from C in a loop to operate. Assuming that c2 is taken out this time, the feature of column c2 will be scanned with a particle shape of d×c2 in a loop, and a new window_size_2 will also be defined to slide and scan on the c2 dimension vector. The data obtained by scanning is also calculated by function.

In the end, each set of c2 features can derive a set of derivative features. But it is possible to perform c2 scanning on all features in a loop. In addition, you can take different small cs in C, such as c1, c3,... for scanning. Finally, concat the image obtained by multi-particle scanning into a multi-dimensional extended feature set L2. Finally, L1, L2 and the original portrait are spliced together to obtain a portrait W that contains more information. The features in the original feature portrait are combined with the derived multi-dimensional extended feature set to form a preprocessed feature data set.

The specific implementation of the saturation detection and screening is as follows. A certain dimensional feature is searched and filtered to delete the dimensional feature with low saturation from the original image. Take the age search and filter as an example. Assuming that there are a total of 100 people, 70 people have age data feature records, and 30 people do not, the saturation of the age column data is 70%, and the features with low saturation will be deleted as appropriate, because The missing is too serious to retain valid information.

The specific implementation of the missing value filling is as follows. The missing value filling is performed based on the nonlinear interpolation method and the rpart method, and the missing value filling scheme is selected through the result backtracking.

The specific implementation of data set heterogeneity and expansion is as follows. Take the data set heterogeneity and expansion based on text extraction from readable media as an example. It is assumed that there are diseases such as diabetes, hypertension, etc. in the history of 1 million people. , The accumulation of so much text information in this column of features cannot allow the model to directly obtain an effective text medium. Therefore, the extraction medium with diabetes in this list of features is made into a new list of features. If there are diabetes in the 1 million people, the new list of features will be marked as 1, and no diabetes will be marked as 0 in this list of features; and so on. Types of diseases and other conditions.

In this embodiment, the specific implementation of the XGBOOST recommendation model and the deep neural network recommendation model obtained by pre-training in step S30: first train according to the feature data set of step S20 to obtain the initial deep neural network recommendation model and the initial XGBOOST recommendation model, after testing After data set testing and optimization, the optimized deep neural network recommendation model and XGBOOST recommendation model are obtained.

The output data of the XGBOOST recommendation model and the deep neural network recommendation model are the drugs to be recommended and the corresponding probability values.

The optimization means that after the preprocessing of step S20 is completed, the XGBOOST recommendation model and the deep neural network recommendation model are built, the parameters and hyperparameters of the two models are adjusted according to computing resources and time, and thresholds are set for the parameters and hyperparameters to limit The upper and lower limits prevent entering an endless loop.

The feature data set of step S20 is used as training data, and an initial deep neural network recommendation model is obtained through training. Among them, the initial deep neural network recommendation model is a 5-layer neural network model, which includes a fully connected layer. The fully connected layer contains 256 neurons, 128 neurons, and 64 neurons, and Dropout is added to the fully connected layer. Floor. The initial deep neural network recommendation model uses the back propagation algorithm to find the partial derivative, which follows the chain rule; and through the TensorFlow deep learning framework for continuous network training and parameter adjustment, the optimized deep neural network recommendation model is finally obtained. Use the feature data set of step S20 as the training data to train to obtain the initial XGBOOST recommendation model. The initial XGBOOST recommendation model uses its built-in feature importance to further rank the importance of feature factors, and filter out the factors with lower weights, and finally obtain the optimized Recommended model of XGBOOST.

In this embodiment, the user's drug request information in step S40 can be obtained in a variety of ways, the voice input is converted into text information or codes in a standard format, and the medical records are scanned and recognized and converted into text information or codes in a standard format, or After manually inputting key field information, it is converted into standard format text information or code, and the converted standard format text information or code is used as drug request information. The drug request information entered by the user through the system includes age, symptoms, medical history, etc. The drug request information entered by different users is not completely the same. For example, the drug request information of the first user includes age 3 years old, the first symptom is a fever of 39 degrees for four days, the second symptom is a dry cough for two days, and the third symptom is a runny nose for two days. The medicine request information of the second user includes age 65, dizziness, medical history of hypertension for ten years, and medication for ten years.

In this embodiment, different drug request information in step S50 obtains different probability values of the first recommendation intensity through the calculation of the XGBOOST recommendation model. For example, the XGBOOST recommendation model calculates and obtains the first recommended strength probability value of each drug in the drug library according to the user's drug request information, which are 0.19 for the first drug; 0.55 for the second drug; 0.55 for the third drug, 0.97;... ….

In this embodiment, different medicine request information in step S60 is calculated by the deep neural network recommendation model to obtain different second recommendation strength probability values. For example, the deep neural network recommendation model calculates the second recommended strength probability value of each drug in the drug library according to the user's drug request information, which are 0.89 for the first drug; 0.15 for the second drug; and 0.07 for the third drug. ;.......

In this embodiment, the first recommended strength probability value and the second recommended strength probability value in step S70 are linearly added according to the weight ratio, where the weight ratio is adjusted through linear regression, and finally obtained The target recommended strength probability value of the combined model is calculated. Suppose that the weight ratio of the XGBOOST recommendation model is the first weight ratio, the weight ratio of the deep neural network recommendation model is the second weight ratio, and the first weight ratio + the second weight ratio=1.

The probability value of the target recommended strength of the first drug = the probability value of the first recommended strength of the first drug x the proportion of the first weight + the probability value of the second recommended strength of the first drug x the proportion of the second weight.

The probability value of the target recommended strength of the second drug=the probability value of the first recommended strength of the second drug×the proportion of the first weight+the probability value of the second recommended strength of the second drug×the proportion of the second weight.

By analogy, the probability value of the target recommended strength of all drugs in the drug library is calculated.

In this embodiment, the specific screening method in step S80 may be: selecting the top three drugs and saving them after sorting the target recommendation intensity probability values in reverse order. The information stored in the drug includes the ranking information of the drug after screening and the name of the drug after screening, and the probability value of the target recommendation intensity after the screening can also be selectively stored. The drug ranking information, drug name and the target recommendation intensity probability value are in one-to-one correspondence. For example, the drug recommendation results of the first user are respectively the first ranking-the third drug-0.97; the second ranking-the seventh drug-0.92; the third ranking-the fiftieth drug-0.89. The drug recommendation results of the second user are respectively the first ranking-the first drug-0.89; the second ranking-the second drug-0.83; the third ranking-the third medicine-0.81.

This application combines the XGBOOST recommendation model and the deep neural network recommendation model, and linearly adds the results of the two models. The accuracy of the recommendation is 200% higher than the accuracy of the non-learning model recommendation, and it ranks the top three correct drugs The recommendation rate has increased from 75% to 95%. This application extracts more hidden information from the historical medical data in the original portrait through feature expansion to ensure the integrity of feature values, and uses saturation exploration to filter and delete feature values that are not meaningful for training, thereby forming complete and effective feature data set. In addition, this application can be used in smart medical treatment by combining the XGBOOST recommendation model and the deep neural network recommendation model, with a high degree of automation, and adaptive training and prediction, which improves the efficiency of drug recommendation.

Please refer to Figure 3. This application also provides an artificial intelligence-based drug recommendation system, which includes:

The collection module 10 is used to collect historical medical data to construct an original portrait;

The preprocessing module 20 is used to preprocess the original image and output a feature data set;

The model training module 30 is used for pre-training the pre-processed feature data set as training data to obtain the XGBOOST recommendation model and the deep neural network recommendation model;

The input module 40 is used to obtain the user's drug request information and input it into the XGBOOST recommendation model and the deep neural network recommendation model;

The processing module 50 is configured to calculate and obtain the first recommended strength probability value of the target drug corresponding to the drug request information through the XGBOOST recommendation model, and obtain the second recommended strength probability value of the target drug corresponding to the drug request information through the deep neural network recommendation model. Recommended intensity probability value; and linearly add the first recommended intensity probability value and the second recommended intensity probability value of each target drug to obtain the target recommended intensity probability value of each target drug;

The output module 60 is used to sort the target drug information of each target drug according to the corresponding target recommendation intensity probability value from high to low, filter out the top preset number of target drug information, and compare the target drug information Push to the user.

Please refer to FIG. 4, this application also provides a computer device 2, and the computer device 2 includes:

The memory 21 is used to store executable program codes; and

The processor 22 is configured to call the executable program code in the memory 21, and the execution steps include the above-mentioned artificial intelligence-based drug recommendation method.

In FIG. 4, a processor 22 is taken as an example.

The memory 21, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs and modules, such as the artificial intelligence-based drug recommendation method in the embodiment of the present application Corresponding program instructions/modules. The processor 22 executes various functional applications and data processing of the computer device 2 by running non-volatile software programs, instructions, and modules stored in the memory 21, that is, realizing the artificial intelligence-based medicine in any of the above-mentioned method embodiments Recommended method.

The memory 21 may include a storage program area and a storage data area. The storage program area may store an operating system and an application program required by at least one function; the storage data area may store historical medical data of the user in the computer device 2. In addition, the memory 21 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 21 may optionally include a memory 21 remotely provided with respect to the processor 22, and these remote memories 21 may be connected to the artificial intelligence-based medicine recommendation system 1 via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

The one or more modules are stored in the memory 21, and when executed by the one or more processors 22, the artificial intelligence-based drug recommendation method in any of the foregoing method embodiments is executed, for example, the foregoing description is executed Figure 1-Figure 2 of the program.

The above-mentioned products can execute the methods provided in the embodiments of the present application, and have functional modules and beneficial effects corresponding to the execution methods. For technical details not described in detail in this embodiment, please refer to the method provided in the embodiment of this application.

The computer equipment 2 of the embodiment of the present application exists in various forms, including but not limited to:

(1) Mobile communication equipment: This type of equipment is characterized by mobile communication functions, and its main goal is to provide voice and data communications. Such terminals include: smart phones (such as iPhone), multimedia phones, functional phones, and low-end phones.

(2) Ultra-mobile personal computer equipment: This type of equipment belongs to the category of personal computers, has calculation and processing functions, and generally also has mobile Internet features. Such terminals include: PDA, MID, and UMPC devices, such as iPad.

(3) Portable entertainment equipment: This type of equipment can display and play multimedia content. Such devices include: audio, video players (such as iPod), handheld game consoles, e-books, as well as smart toys and portable car navigation devices.

(4) Server: A device that provides computing services. The composition of a server includes a processor, hard disk, memory, system bus, etc. The server is similar to a general computer architecture, but due to the need to provide highly reliable services, it is in terms of processing power and stability. , Reliability, security, scalability, and manageability.

(5) Other electronic devices with data interaction function.

Another embodiment of the present application further provides a non-volatile computer-readable storage medium, the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are executed by one or more processors, such as One processor 22 in FIG. 4 can enable the one or more processors 22 to execute the artificial intelligence-based drug recommendation method in any of the foregoing method embodiments, for example, to execute the programs in FIGS. 1 to 2 described above. The computer-readable storage medium may be non-volatile or volatile.

The device embodiments described above are merely illustrative, where the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place. , Or it can be distributed to at least two network units. Some or all of the modules may be selected according to actual needs to achieve the objectives of the solutions of the embodiments of the present application. Those of ordinary skill in the art can understand and implement it without creative work.

Through the description of the above implementation manners, those of ordinary skill in the art can clearly understand that each implementation manner can be implemented by means of software plus a general hardware platform, and of course, it can also be implemented by hardware. Those of ordinary skill in the art can understand that all or part of the processes in the methods of the foregoing embodiments can be implemented by computer-readable instructions to instruct relevant hardware. The programs can be stored in a computer-readable storage medium. When the program is executed, it may include the processes of the above-mentioned method embodiments. Wherein, the storage medium may be a magnetic disk, an optical disc, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM), etc.

The serial numbers of the embodiments of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments.

Through the description of the above embodiments, those skilled in the art can clearly understand that the method of the embodiment can be implemented by means of software plus the necessary general hardware platform, of course, it can also be implemented by hardware, but in many cases the former is more The best way to implement it.

The above are only the preferred embodiments of the application, and do not limit the scope of the patent for this application. Any equivalent structure or equivalent process transformation made using the content of the description and drawings of the application, or directly or indirectly applied to other related technical fields , The same reason is included in the scope of patent protection of this application.

Claims (20)

An artificial intelligence-based drug recommendation method, which includes the following steps: Collect historical medical data to construct original portraits; Preprocess the original image and output the feature data set; The pre-processed feature data set is used as the training data, and the XGBOOST recommendation model and the deep neural network recommendation model are pre-trained respectively; Obtain the user's drug request information and input it into the XGBOOST recommendation model and the deep neural network recommendation model; The XGBOOST recommendation model calculates and obtains the first recommendation intensity probability value of the target drug corresponding to the drug request information according to the drug request information; The deep neural network recommendation model calculates and obtains the second recommendation intensity probability value of the target drug corresponding to the drug request information according to the drug request information; Linearly adding the first recommended intensity probability value and the second recommended intensity probability value of each of the target drugs to obtain the target recommended intensity probability value of each of the target drugs; The target drug information of each target drug is sorted according to the corresponding target recommendation intensity probability value from high to low, and the preset number of target drug information in the top ranking is screened out, and the target drug information is pushed to the user. The artificial intelligence-based drug recommendation method according to claim 1, wherein: the original image includes basic information and medical record information, and the data source of the original image includes multiple different data sources, and the multiple different data sources include The multi-dimensional characteristics are inconsistent. The artificial intelligence-based drug recommendation method according to claim 1, wherein: while collecting historical medical data, the data in the historical medical data and the data in the associated database are established through the association extraction method to establish a table correspondence relationship, and the association is obtained by collecting Link data in the database to construct the original portrait. The artificial intelligence-based drug recommendation method according to claim 1, wherein: the preprocessing includes feature expansion, saturation exploration and screening, missing value filling, data set heterogeneity and expansion, feedback selection, and correlation screening. The artificial intelligence-based medicine recommendation method according to claim 1, wherein: the original portrait includes features of an n×m dimensional matrix, and the preprocessing of the original portrait includes: Using the tsfresh algorithm, the feature of the n×m dimension matrix is simplified into the horizontal splicing of several n×1 dimension vectors, and then particle scanning is performed. The particle scanning includes single-dimensional particle scanning and multi-dimensional particle scanning. The artificial intelligence-based medicine recommendation method according to claim 5, wherein: the preprocessing of the original image comprises: the single-dimensional particle scan is used to derive a single-dimensional extended feature set L1 through a sliding scan, and the multi-dimensional Particle scanning uses sliding scanning to derive a multi-dimensional extended feature set L2, and then a single-dimensional extended feature set, a multi-dimensional extended feature set and the original image are spliced together to obtain a feature data set containing more information. The artificial intelligence-based drug recommendation method according to claim 1, wherein: when the historical medical data is collected, the classification of non-standard drug names includes the following steps: Separate the names of non-standard drugs and standard drugs; Based on the Word2Vec shallow neural network, and based on the skip_gram algorithm, the characters in the non-standard drug name and the standard drug name are respectively converted into word vectors; Use the pooling rule to splice the vectors vertically to obtain the word vector splicing of non-standard drug names and standard drug names; Calculate the vector distance between the non-standard drug name and the standard drug name; Find the standard drug name with the closest vector distance to the non-standard drug name; The non-standard drug names are classified according to the classification criteria of the standard drug names with the closest vector distance. A drug recommendation system based on artificial intelligence, including: Collection module, which is used to collect historical medical data to construct original portraits; A preprocessing module, which is used to preprocess the original image and output a feature data set; Model training module, which is used to pre-train the pre-processed feature data set as training data to obtain the XGBOOST recommendation model and the deep neural network recommendation model; Input module, which is used to obtain the user's drug request information and input it into the XGBOOST recommendation model and the deep neural network recommendation model; The processing module is used to calculate and obtain the first recommendation strength probability value of the target drug corresponding to the drug request information through the XGBOOST recommendation model, and obtain the second recommendation of the target drug corresponding to the drug request information through the deep neural network recommendation model Intensity probability value; and linearly adding the first recommended intensity probability value and the second recommended intensity probability value of each of the target drugs to obtain the target recommended intensity probability value of each of the target drugs; The output module is used to sort the target drug information of each target drug according to the corresponding target recommendation intensity probability value from high to low, filter out the preset number of target drug information in the top ranking, and push the target drug information To the user. The artificial intelligence-based drug recommendation system according to claim 8, wherein: the original portrait includes basic information and medical record information, and the data source of the original portrait includes multiple different data sources, and the multiple different data sources include The multi-dimensional characteristics are inconsistent. The artificial intelligence-based drug recommendation system according to claim 8, wherein: the collection module is used to collect historical medical data and at the same time establish a table correspondence between the data in the historical medical data and the data in the associated database through the correlation extraction method Relations, and collect the associated data in the associated database to construct the original portrait. The artificial intelligence-based drug recommendation system according to claim 8, wherein: the preprocessing includes feature expansion, saturation exploration and screening, missing value filling, data set heterogeneity and expansion, feedback selection, and correlation screening. The artificial intelligence-based medicine recommendation system according to claim 8, wherein: the original portrait includes features of an n×m dimensional matrix; The preprocessing module is used to simplify the feature of the n×m dimensional matrix into a horizontal splicing of several n×1 dimensional vectors by using the tsfresh algorithm, and then perform particle scanning. The particle scanning includes single-dimensional particle scanning and multi-dimensional particle scanning. The artificial intelligence-based drug recommendation system according to claim 12, wherein: the particle scanning performed by the pre-processing module comprises: the single-dimensional particle scanning uses sliding scanning to derive a single-dimensional extended feature set L1, and the multi-dimensional Particle scanning uses sliding scanning to derive a multi-dimensional extended feature set L2, and then a single-dimensional extended feature set, a multi-dimensional extended feature set and the original image are spliced together to obtain a feature data set containing more information. A computer device comprising a memory, a processor, and computer readable instructions stored in the memory and capable of running on the processor, wherein: the processor executes the computer readable instructions to implement a manual operation The following steps of the smart drug recommendation method: Collect historical medical data to construct original portraits; Preprocess the original image and output the feature data set; The pre-processed feature data set is used as the training data, and the XGBOOST recommendation model and the deep neural network recommendation model are pre-trained respectively; Obtain the user's drug request information and input it into the XGBOOST recommendation model and the deep neural network recommendation model; The XGBOOST recommendation model calculates and obtains the first recommendation intensity probability value of the target drug corresponding to the drug request information according to the drug request information; The deep neural network recommendation model calculates and obtains the second recommendation intensity probability value of the target drug corresponding to the drug request information according to the drug request information; Linearly adding the first recommended intensity probability value and the second recommended intensity probability value of each of the target drugs to obtain the target recommended intensity probability value of each of the target drugs; The target drug information of each target drug is sorted according to the corresponding target recommendation intensity probability value from high to low, and the preset number of target drug information in the top ranking is screened out, and the target drug information is pushed to the user. The computer device according to claim 14, wherein: the original portrait includes basic information and medical record information, the data source of the original portrait includes multiple different data sources, and the multi-dimensional features included in the multiple different data sources are inconsistent . The computer device according to claim 14, wherein: said collecting historical medical data to construct an original portrait comprises: While collecting historical medical data, the data in the historical medical data and the data in the associated database are established by the association extraction method to establish a table correspondence relationship, and the associated data in the associated database is collected to construct the original portrait. The computer device according to claim 14, wherein: the preprocessing includes feature expansion, saturation exploration and screening, missing value filling, data set heterogeneity and expansion, feedback selection, and correlation screening. A computer-readable storage medium having computer-readable instructions stored thereon, wherein: when the computer-readable instructions are executed by a processor, the following steps of an artificial intelligence-based drug recommendation method are implemented: Collect historical medical data to construct original portraits; Preprocess the original image and output the feature data set; The pre-processed feature data set is used as the training data, and the XGBOOST recommendation model and the deep neural network recommendation model are pre-trained respectively; Obtain the user's drug request information and input it into the XGBOOST recommendation model and the deep neural network recommendation model; The XGBOOST recommendation model calculates and obtains the first recommendation intensity probability value of the target drug corresponding to the drug request information according to the drug request information; The deep neural network recommendation model calculates and obtains the second recommendation intensity probability value of the target drug corresponding to the drug request information according to the drug request information; Linearly adding the first recommended intensity probability value and the second recommended intensity probability value of each of the target drugs to obtain the target recommended intensity probability value of each of the target drugs; The target drug information of each target drug is sorted according to the corresponding target recommendation intensity probability value from high to low, and the preset number of target drug information in the top ranking is screened out, and the target drug information is pushed to the user. The computer-readable storage medium according to claim 18, wherein: the original portrait includes basic information and medical record information, the data source of the original portrait includes a plurality of different data sources, and the plurality of different data sources includes a plurality of The dimensional characteristics are inconsistent. 18. The computer-readable storage medium according to claim 18, wherein: the preprocessing includes feature expansion, saturation exploration and screening, missing value filling, data set heterogeneity and expansion, feedback selection, and correlation screening.