Intelligent medicine shopping guide and recommendation system for drugstore

Abstract

The invention relates to the field of intelligent medical services, and discloses an intelligent medicine shopping guide and recommendation system for a drugstore, which comprises a user information collection module, a medicine database module, a symptom matching module, a recommendation optimization module and a result output module, the user information collection module is used for receiving health data and personalized requirements input by a user and converting the health data and the personalized requirements into structured user feature data; the drug database module stores components, indications, prices and drug interaction information of drugs, and constructs a drug interaction knowledge graph for representing an interaction relationship between drugs and a risk thereof. By introducing a drug interaction knowledge graph, the interaction risk between drugs is dynamically evaluated, and the risk information is fused into a recommendation decision. Compared with a recommendation method which only depends on static rule matching in the prior art, the problem that potential drug conflicts in recommendation results are difficult to find is effectively solved, and the drug use safety is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent medical services, and in particular to an intelligent medicine shopping guide and recommendation system for pharmacies. Background Art

[0002] At present, with the development of informatization in the pharmaceutical industry, drug recommendation technology has been initially applied in the intelligent services of pharmacies. Some existing systems use symptoms or keywords input by users to screen out suitable drugs from the drug database based on rule matching or simple search algorithms. This type of technology can help users quickly obtain relevant drug information to a certain extent and provide pharmacies with an efficient service means. For example, existing technologies can recommend commonly used drugs related to symptoms by matching the relationship between user input and drug indications; at the same time, some drug recommendation systems can also combine drug prices and inventory information to provide users with simple selection references. However, due to the wide variety of drugs and complex user needs, the existing systems still have certain limitations in functional design and recommendation effects.

[0003] In practical applications, existing technologies still face some problems and challenges. First, the accuracy of symptom matching relies on the recognition of keywords or simple rules, lacks the understanding of semantic relationships in natural language input, and is difficult to handle complex or ambiguous user descriptions. Secondly, the interaction risks between drugs have not been dynamically evaluated, and the recommended results may include drugs that are risky to use at the same time, which cannot fully guarantee the safety of medication. In addition, the existing system has limited consideration of user personalized needs (such as budget, brand preference, and health records), and the recommended results lack pertinence. Finally, user input requirements may change dynamically, but existing technologies usually only generate static recommendation results, which cannot meet the user's real-time adjustment needs during the drug purchase process. Therefore, there is an urgent need for a drug shopping guide and recommendation system that can provide a more intelligent, personalized and secure system to improve the accuracy of drug recommendations and user experience. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides an intelligent drug shopping guide and recommendation system for pharmacies, which solves the problems of inaccurate symptom matching, unassessed drug interaction risks, and difficulty in meeting personalized needs in the drug recommendation process in pharmacies.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: an intelligent drug shopping guide and recommendation system for drug stores, comprising a user information collection module, a drug database module, a symptom matching module, a recommendation optimization module and a result output module; The user information collection module is used to receive health data and personalized needs input by the user and convert them into structured user feature data; The drug database module stores the ingredients, indications, prices and drug interaction information of drugs, and constructs a drug interaction knowledge graph to represent the interaction relationship between drugs and their risks; The symptom matching module screens out a set of candidate drugs based on the symptom characteristics input by the user and the drug indication information; The recommendation optimization module adopts a variational optimization model, comprehensively considers symptom matching, drug interaction risk, and user preferences, and generates a drug recommendation list by optimizing a multi-objective loss function; The result output module evaluates the safety of the drug recommendation list based on the drug interaction knowledge graph and outputs the final recommendation result.

[0006] Preferably, a user information collection module, a drug database module, a symptom matching module, a recommendation optimization module and a result output module; The user information collection module is used to receive health data and personalized needs input by the user and convert them into structured user feature data; The drug database module stores the ingredients, indications, prices and drug interaction information of drugs, and constructs a drug interaction knowledge graph to represent the interaction relationship between drugs and their risks; The symptom matching module screens out a set of candidate drugs based on the symptom characteristics input by the user and the drug indication information; The recommendation optimization module adopts a variational optimization model, comprehensively considers symptom matching, drug interaction risk, and user preferences, and generates a drug recommendation list by optimizing a multi-objective loss function; The result output module evaluates the safety of the drug recommendation list based on the drug interaction knowledge graph and outputs the final recommendation result.

[0007] Preferably, the drug interaction knowledge graph constructed by the drug database module includes interaction nodes and interaction relationship edges between drugs, and the weights of the relationship edges are used to characterize the risk level of drug interactions.

[0008] Preferably, the symptom matching module screens the candidate drug set by calculating the similarity between the user's symptom characteristics and the drug indication characteristics, wherein the similarity is calculated by the cosine similarity between the vector representation of the user's input symptoms and the drug indication.

[0009] Preferably, the recommendation optimization module uses a variational autoencoder model to embed user feature data and a set of candidate drugs, and optimizes the following multi-objective loss function based on the drug recommendation distribution: Symptom matching objective, which is used to maximize the fit between the recommended drugs and the user-input symptoms; Drug interaction risk objectives, used to minimize the risk of interactions between recommended drugs; User preference objective, which is used to minimize the deviation between recommended drugs and user budget and brand preference.

[0010] Preferably, the decoder of the variational autoencoder model generates a distribution of recommended drugs through a Softmax function, which is used to select an optimal recommended drug list from a set of candidate drugs.

[0011] Preferably, in the recommendation optimization module, the drug interaction risk is calculated by using the relationship edge weights in the drug interaction knowledge graph and is used to evaluate the overall risk score of the recommended drug list.

[0012] Preferably, the result output module includes a risk assessment unit for replacing drugs in the recommended drug list whose drug interaction risk exceeds a safety threshold and regenerating the final recommendation result.

[0013] Preferably, the user characteristic data includes the user's symptom description, age, gender, allergy history, budget and brand preference, and the user characteristic data is embedded in a high-dimensional vector space as a model input.

[0014] Preferably, the final recommendation result includes drug indication information, drug price information and interaction risk information between drugs.

[0015] The present invention provides an intelligent drug shopping guide and recommendation system for drug stores, which has the following beneficial effects: 1. The present invention introduces a drug interaction knowledge graph to dynamically evaluate the interaction risk between drugs and incorporate risk information into recommendation decisions. Compared with the recommendation method in the prior art that only relies on static rule matching, it effectively solves the problem of difficulty in discovering potential drug conflicts in recommendation results and improves drug safety.

[0016] 2. The present invention uses variational optimization technology to comprehensively consider symptom matching, drug interaction risk and user preferences, optimizes the multi-objective loss function to generate a recommendation list. Compared with the recommendation method of single-objective optimization or fixed rules in the prior art, it significantly improves the personalization and accuracy of the recommendation results and meets the diverse needs of users.

[0017] 3. The present invention uses natural language processing technology to analyze the user's symptom description, converts complex natural language into structured data, and combines the symptom matching module to achieve intelligent recommendation. Compared with the traditional method of relying on manual input of keywords in the prior art, it solves the problems of high user input threshold and poor system responsiveness, and significantly improves the system's ease of use and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the system module of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] Please refer to the attached Figure 1 , an embodiment of the present invention provides an intelligent drug shopping guide and recommendation system for a drug store, including a user information collection module, a drug database module, a symptom matching module, a recommendation optimization module and a result output module; The user information collection module is used to receive the health data and personalized needs input by the user and convert them into structured user feature data; The drug database module stores the ingredients, indications, prices and drug interaction information of drugs, and builds a drug interaction knowledge graph to represent the interaction relationship and risks between drugs; The symptom matching module selects a set of candidate drugs based on the symptom characteristics input by the user and the drug indication information; The recommendation optimization module adopts a variational optimization model, comprehensively considers symptom matching, drug interaction risk, and user preferences, and generates a drug recommendation list by optimizing a multi-objective loss function; The result output module evaluates the safety of the drug recommendation list based on the drug interaction knowledge graph and outputs the final recommendation result; The user information collection module includes a natural language processing module for parsing the user's natural language input into symptom features and extracting the user's personalized demand information; The drug interaction knowledge graph constructed by the drug database module contains interaction nodes and interaction relationship edges between drugs. The weight of the relationship edge is used to characterize the risk level of drug interaction. The symptom matching module screens the candidate drug set by calculating the similarity between the user's symptom characteristics and the drug indication characteristics. The similarity is calculated by the cosine similarity between the vector representation of the user's input symptoms and the drug indication; The recommendation optimization module uses a variational autoencoder model to embed user feature data and candidate drug sets, and optimizes the following multi-objective loss function based on the drug recommendation distribution: Symptom matching objective, which is used to maximize the fit between the recommended drugs and the user-input symptoms; Drug interaction risk objectives, used to minimize the risk of interactions between recommended drugs; User preference objective, used to minimize the deviation between recommended drugs and user budget and brand preference; The decoder of the variational autoencoder model generates the distribution of recommended drugs through the Softmax function, which is used to select the best recommended drug list from the candidate drug set; In the recommendation optimization module, the drug interaction risk is calculated by the relationship edge weights in the drug interaction knowledge graph and used to evaluate the overall risk score of the recommended drug list; The result output module includes a risk assessment unit, which is used to replace the drugs in the recommended drug list whose drug interaction risk exceeds the safety threshold and regenerate the final recommendation result; User feature data includes the user's symptom description, age, gender, allergy history, budget, and brand preference. User feature data is embedded in a high-dimensional vector space as model input; The final recommendation results include drug indication information, drug price information and drug interaction risk information.

[0021] Specifically, the user information collection module includes a natural language processing module and a user feature vector generation unit. The natural language processing module is used to parse the symptom description input by the user and convert it into structured symptom features; the user feature vector generation unit embeds all user data into a unified high-dimensional vector space to form a user feature vector.

[0022] Specifically, the natural language processing module first receives the symptom description input by the user in natural language, such as "I have had frequent headaches and a sore throat recently". In one possible implementation, the module uses semantic analysis technology based on a pre-trained language model (such as BERT or BioGPT) to parse the user input into a set of semantic embedding vectors. In general, the semantic embedding vector can capture important symptom keywords (such as "headache", "throat inflammation") and their contextual semantic information in the input text. Through the feature representation of the embedding vector, the system can efficiently extract the user's health status information.

[0023] As an option, the natural language processing module can also use the symptom classification model to further parse the user input. For example, the symptoms described by the user can be classified into headache, pharyngitis, or other symptom categories. In this implementation, the classification results can be combined with the embedding vector to enhance the accuracy of feature expression.

[0024] In some embodiments, the user feature vector generation unit also combines other health data and personalized needs of the user, for example: Health data includes the user's age, gender, and medical history (such as allergies, chronic diseases, etc.). These data can be entered directly into the system or collected through question-and-answer format.

[0025] Personalized requirements include budget and drug brand preference. For example, users may set "budget limit is less than 50 yuan" or "prefer brand A drugs". These requirements are quantified into numerical features and embedded into high-dimensional vectors together with other features.

[0026] The feature vector generation unit embeds all data using the following formula:

[0027] in: The symptom embedding vector representing the user input is parsed and generated by the natural language processing module; Represents the user's health profile characteristics, such as age, gender, and allergies; Represents the user's budget and brand preference characteristics, usually normalized to a unified numerical range.

[0028] In a possible implementation, the user feature vector generation unit adopts the following embedding function:

[0029] in: Represents the semantic embedding function generated by the natural language processing module; is a function that normalizes health data; is a mapping function that transforms user preferences into embedding vectors.

[0030] Generated user feature vector It is a unified high-dimensional vector that can represent the user's symptom information, health status and personalized needs. Generally, this vector will be passed to the symptom matching module for comparison with the indication characteristics in the drug database.

[0031] In this embodiment, the combination of the natural language processing module and the user feature vector generation unit can ensure that all information input by the user is efficiently parsed and quantified. Whether it is a complex natural language description or a specific numerical requirement, the system can convert it into a standardized feature representation for use by subsequent modules. For example, if the user enters "I need a painkiller that costs less than 50 yuan, and I am allergic to certain antibiotics", the system will parse out "painkillers" as a symptom feature, "less than 50 yuan" as a budget feature, and "antibiotic allergy" as a health feature.

[0032] The drug database module is used to store relevant information about drugs and provide data support for the symptom matching module and the recommendation optimization module. In general, this module contains basic information about drugs (such as drug ingredients, indications, prices, etc.) and interaction information between drugs (such as drug interactions and their risks). By constructing a drug interaction knowledge graph, this module can not only provide basic drug properties, but also represent the interaction between drugs, providing support for drug safety assessment. The design of the drug database module enables the system to quickly extract the required data from diverse drug information and complex interaction relationships, thereby improving the efficiency and accuracy of recommendations.

[0033] In this embodiment, the drug database module stores structured drug information, mainly including the following types of data: Drug ingredients: Describe the active ingredients of the drug and their content.

[0034] Drug indications: including applicable diseases, symptoms or special usage scenarios.

[0035] Drug prices: record market prices or drug store prices.

[0036] Drug interaction information: includes the risk of potential adverse reactions or interactions between drugs.

[0037] Specifically, each drug in the drug database is represented by a multidimensional feature vector:

[0038] in: is the drug component vector, indicating the chemical composition of the drug; is the indication vector, which indicates the scope of application of the drug; is a price feature, usually normalized to a certain range; It is an interactive feature, which is used to represent the interactive relationship between a drug and other drugs.

[0039] In one possible implementation, this module can also update user features in real time. For example, when a user adds more symptom information or modifies their budget, the system can dynamically adjust the user feature vector to ensure the accuracy and real-time nature of the recommendation results. This dynamic adjustment mechanism is implemented through the module's built-in real-time feature update unit.

[0040] In one possible implementation, the drug database module uses a relational database (such as MySQL) to store basic drug data, and combines it with a graph database (such as Neo4j) to build a drug interaction knowledge graph. Generally speaking, a relational database is suitable for storing and querying structured data, while a graph database is more suitable for representing complex drug interaction relationships.

[0041] In some embodiments, the drug interaction knowledge graph is composed of the following elements: Node: Each node represents a drug, and node attributes include drug name, ingredients, and indications.

[0042] Edge: Each edge represents the interaction between two drugs, and the edge weight represents the degree of interaction risk.

[0043] In a possible implementation, the edge weight of the drug interaction knowledge graph can be calculated by the following formula:

[0044] in: For medicine and The severity of the interaction; The likelihood of an interaction occurring; The credibility of the data source.

[0045] The construction of the knowledge graph can be achieved in many ways. As an option, the system can extract drug interaction information from public drug databases (such as DrugBank or the FDA drug database) and expand it with drug inventory data from local pharmacies. Specifically, the system establishes interactive relationships between drugs based on their ingredients and indications. For example, drug A and drug B may have potential risks due to the interaction of certain ingredients, and these relationships are recorded in the knowledge graph in the form of edges.

[0046] In another possible implementation, the drug database module also supports real-time data updates. For example, when new drugs are added or the interaction risk of existing drugs changes, the system can automatically update the database and knowledge graph. This dynamic update function can be achieved by regularly crawling the public drug database or by connecting with the drugstore's inventory management system.

[0047] In this embodiment, the drug database module also provides an interface for the symptom matching module and the recommendation optimization module. For example, the symptom matching module can obtain the indication characteristics of the drug by querying the drug database. , and compare it with the symptom characteristics entered by the user. The recommendation optimization module can use the interactive risk data in the knowledge graph Evaluate the safety of a list of recommended medicines.

[0048] The symptom matching module is one of the core functional modules in the intelligent drug shopping guide and recommendation system of the present invention, and is used to screen out a set of candidate drugs that may match from the drug database module based on the symptom features input by the user. Generally, the symptoms input by the user are in the form of natural language descriptions, while the indication data in the drug database are structured feature representations. The main task of the symptom matching module is to use the user features generated by the user information collection module to perform semantic comparison with the indication information in the drug database, calculate the degree of matching, and screen the drugs by similarity scores. This module provides a highly correlated set of candidate drugs for the recommendation optimization module, thereby improving the accuracy and efficiency of the recommendation results.

[0049] In this embodiment, the symptom matching module matches the user's symptoms with the drug indications through the following steps: First, the system receives the user's symptom characteristics from the user information collection module ,This feature is extracted by natural language processing technology and contains a high-dimensional embedding representation of the symptom information described by the user.

[0050] At the same time, the symptom matching module obtains the indication characteristics of the drug from the drug database module , where each drug The indication feature is a structured high-dimensional vector that describes the range of symptoms for which the drug is applicable.

[0051] Specifically, the symptom matching module uses the following formula to calculate the similarity between the user's symptom characteristics and the drug indication characteristics:

[0052] in: Indicates the semantic similarity between symptoms and indications; is the user symptom feature vector; It's medicine Indications vector.

[0053] In one possible implementation, the symptom matching module can also be combined with a weighted feature similarity calculation method to enhance the matching accuracy. Specifically, different symptoms may have different importance to drug indications. For example, "persistent fever" may be more critical to drug selection than "mild headache". Therefore, the system assigns weights to each user's symptom feature component. , and its calculation formula is as follows:

[0054] in, The weights of the feature components are input by the user; and The first A quantity.

[0055] As an option, the symptom matching module can also assist in processing user input through a classification model. For example, a multi-label classifier based on deep learning is used to classify user symptom input into multiple possible disease categories. Each disease category is compared with the indication category of the drug to narrow the matching range and reduce the interference of irrelevant drugs.

[0056] In another possible implementation, the symptom matching module supports matching of multimodal inputs. For example, a user can input symptom descriptions and physical examination report data at the same time. The system achieves more accurate symptom matching by jointly embedding natural language symptom descriptions and structured physical examination data.

[0057] For example, a user may enter "dizziness accompanied by high blood pressure" and upload a medical report at the same time. The system converts the blood pressure data in the medical report into numerical features, which together with the natural language description form a joint feature. , and then perform similarity calculation with the drug indication vector.

[0058] This embodiment also supports real-time updating of symptom matching results. When the user supplements or modifies the input information, such as adding the symptom of "mild cough", the system will automatically update the user's symptom characteristics. This dynamic update capability enables the system to quickly respond to user needs and provide real-time drug matching results.

[0059] The recommendation optimization module is a key part of the present invention. Its main function is to generate a drug recommendation list through a multi-objective optimization model based on the candidate drug set output by the symptom matching module, combined with user feature data and drug interaction knowledge graph. In general, the goal of this module is to comprehensively consider the symptom matching degree of the drug, the risk of drug interaction and the personalized needs of the user, to ensure that the recommendation results are both accurate and in line with drug safety and user preferences. The recommendation optimization module adopts a variational optimization model with a multi-objective loss function as the core to optimize and screen candidate drugs and provide a sorted list of recommended drugs.

[0060] In this embodiment, the recommendation optimization module is based on the candidate drug set candidate, user feature vector A multi-objective optimization model was built based on the drug interaction knowledge graph, and the following three main objective functions were defined: This goal is used to maximize the matching degree between the user's symptom characteristics and the recommended drug indication characteristics. The specific calculation method is:

[0061] in: Matching losses for symptoms; The symptom feature vector is input by the user; Candidate drug The indication feature vector of is the vector similarity function, and cosine similarity is used.

[0062] This goal is used to minimize the interaction risk between drugs in the recommended drug list. Specifically, based on the weights of the edges in the drug interaction knowledge graph, the calculation formula for drug interaction risk is defined as:

[0063] in: drug interaction risk score for interaction risk; For medicine and Interactive risk weights in interactive knowledge graphs.

[0064] As an option, the interaction risk target can also set a safety threshold based on actual conditions. When the drug interaction risk is higher than the threshold, the candidate drug will be excluded.

[0065] This goal is used to minimize the deviation between the recommended drugs and the user's budget and brand preference. The specific formula is:

[0066] in: is the total loss function; , , are the weights of the three objective functions, used to control the priorities of different objectives; is the model parameter.

[0067] In some embodiments, the recommendation optimization module supports online learning, that is, the recommendation model is updated in real time based on user feedback. For example, when a user selects a drug recommended by the system but the feedback is not good, the system can adjust the weight parameters to , , Or retrain the model to improve the accuracy of subsequent recommendations.

[0068] In another possible implementation, the recommendation optimization module can also support a hierarchical optimization strategy. Specifically, the system can first preliminarily screen the candidate drug set, for example, sort them by symptom matching, select the top-ranked drugs, and then perform interactive risk assessment and user preference adjustment on these drugs. This can reduce the amount of calculation and improve optimization efficiency.

[0069] The result output module is the last functional module of the intelligent drug shopping guide and recommendation system in the present invention. It is used to present the recommendation results based on the drug recommendation list generated by the recommendation optimization module, and provide users with intuitive medication information reference. Generally, this module includes a risk assessment function and a recommendation result display function. By further evaluating the safety of the drug recommendation list and outputting the optimized results, it provides users with reliable and intuitive drug selection suggestions. The result output module also supports interactive operations with users, such as filtering or adjusting the recommended drug list to better meet the personalized needs of users. This module ensures that the final output results of the recommendation system meet user expectations and are scientific and operational.

[0070] In this embodiment, the design of the result output module includes the following main functions: First, the system receives the drug recommendation list generated by the recommendation optimization module ,Each drug in the list contains indication information, price information and ,drug interaction risk information.,In order to improve the safety and credibility of the ,recommendation results, this module further conducts interaction risk assessment on ,the drugs in the recommendation list.

[0071] Specifically, this module evaluates the interaction risk between every two drugs in the recommended drug list based on the drug interaction knowledge graph. The calculation formula for interaction risk is:

[0072] in: Represents the total interaction risk score of the entire recommendation list; For medicine and The interaction risk is represented by the edge weights in the interaction knowledge graph.

[0073] Generally, when the total interaction risk score exceeds a preset threshold, the system triggers a risk adjustment mechanism. As an option, the system can automatically replace drugs with higher interaction risks. For example, if and Interaction risk weight If the risk is higher than the threshold, the system can select alternative drugs with lower risks from the candidate drug set. replace or .

[0074] In this embodiment, the result output module also supports the sorting and screening functions of recommended drugs. Specifically, the system sorts the drugs in the recommendation list according to the comprehensive scores generated by the multi-objective loss function of the recommendation optimization module. The sorting is based on the following formula:

[0075] in: For medicine The overall score of represents the symptom matching score; Indicates medicine Interaction risk score; for drug prices; Budget for users.

[0076] In a possible implementation, the result output module supports the visualization of the recommendation results. Specifically, the module presents the recommended drug list in the form of a table or card. The main information of each drug includes: Name of the drug; Description of indications; Drug prices; Drug interaction risk warning.

[0077] For example, the recommendation results can be presented in the following format: Users can use the interface filtering function to adjust the recommended list according to price range, interaction risk threshold or brand preference. For example, users can set the price cap to 50 yuan, and the system will automatically filter out drugs above 50 yuan and re-sort them.

[0078] Drug Name Indications price Interactive risk warning Drug A Headache, fever 35 Yuan No interaction risk Drug B Sore throat 50 NTD There is a low risk Drug C cold 45 Yuan No interaction risk As an option, the interaction risk prompt can be presented in a color-coded manner, such as green for no interaction risk, yellow for low risk, and red for high risk.

[0079] In another possible implementation, the result output module supports multiple rounds of user interaction. Users can further propose requirements based on the initial output results of the recommendation list, such as adjusting the budget range, adding or deleting specific drugs, etc. This module will dynamically update the recommendation list based on user feedback. For example, if the user adds the condition that "the drug needs to have anti-inflammatory effect", the system will re-screen drugs with anti-inflammatory effect from the candidate drug set and update the recommendation list.

[0080] In this embodiment, the result output module also provides a medication guidance function. For example, when the user selects a recommended drug, the system will provide instructions for the drug, including usage and dosage, possible side effects and contraindications. In addition, the system can give relevant medication reminders based on the user's health data, such as "This drug is not suitable for patients with liver dysfunction."

[0081] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent drug shopping guide and recommendation system for pharmacies, characterized in that: It includes user information collection module, drug database module, symptom matching module, recommendation optimization module and result output module; The user information collection module is used to receive health data and personalized needs input by the user and convert them into structured user feature data; The drug database module stores the ingredients, indications, prices and drug interaction information of drugs, and constructs a drug interaction knowledge graph to represent the interaction relationship between drugs and their risks; The symptom matching module screens out a set of candidate drugs based on the symptom characteristics input by the user and the drug indication information; The recommendation optimization module adopts a variational optimization model, comprehensively considers symptom matching, drug interaction risk, and user preferences, and generates a drug recommendation list by optimizing a multi-objective loss function; The result output module evaluates the safety of the drug recommendation list based on the drug interaction knowledge graph and outputs the final recommendation result.

2. The intelligent drug shopping guide and recommendation system for pharmacies according to claim 1, characterized in that: The user information collection module includes a natural language processing module, which is used to parse the user's natural language input into symptom features and extract the user's personalized demand information.

3. The intelligent drug shopping guide and recommendation system for pharmacies according to claim 1, characterized in that: The drug interaction knowledge graph constructed by the drug database module includes interaction nodes and interaction relationship edges between drugs, and the weights of the relationship edges are used to characterize the risk level of drug interaction.

4. The intelligent drug shopping guide and recommendation system for pharmacies according to claim 1, characterized in that: The symptom matching module screens the candidate drug set by calculating the similarity between the user's symptom characteristics and the drug indication characteristics, where the similarity is calculated by the cosine similarity between the vector representation of the user's input symptoms and the drug indication.

5. The intelligent drug shopping guide and recommendation system for pharmacies according to claim 1, characterized in that: The recommendation optimization module uses a variational autoencoder model to embed user feature data and candidate drug sets, and optimizes the following multi-objective loss function based on drug recommendation distribution: Symptom matching objective, which is used to maximize the fit between the recommended drugs and the user-input symptoms; Drug interaction risk objectives, used to minimize the risk of interactions between recommended drugs; User preference objective, which is used to minimize the deviation between recommended drugs and user budget and brand preference.

6. The intelligent drug shopping guide and recommendation system for pharmacies according to claim 1, characterized in that: The decoder of the variational autoencoder model generates a distribution of recommended drugs through a Softmax function, which is used to select an optimal recommended drug list from a set of candidate drugs.

7. The intelligent drug shopping guide and recommendation system for pharmacies according to claim 1, characterized in that: In the recommendation optimization module, the drug interaction risk is calculated by the relationship edge weights in the drug interaction knowledge graph and is used to evaluate the overall risk score of the recommended drug list.

8. The intelligent drug shopping guide and recommendation system for pharmacies according to claim 1, characterized in that: The result output module includes a risk assessment unit, which is used to replace drugs in the recommended drug list whose drug interaction risks exceed the safety threshold and regenerate the final recommendation result.

9. The intelligent drug shopping guide and recommendation system for pharmacies according to claim 1, characterized in that: The user feature data includes the user's symptom description, age, gender, allergy history, budget and brand preference, and the user feature data is embedded in a high-dimensional vector space as a model input.

10. The intelligent drug shopping guide and recommendation system for pharmacies according to claim 1, characterized in that: The final recommendation results include drug indication information, drug price information, and interaction risk information between drugs.

Images