Point exchange system and method for point shopping mall

Abstract

The invention relates to the technical field of point shopping malls, and discloses a point exchange system and method for a point shopping mall, and the system comprises the steps: obtaining user data and public medical case data, and carrying out the preprocessing; constructing a historical time sequence data set and a historical node graph based on the preprocessed data; the training module is used for constructing a body state prediction model and training the body state prediction model by adopting a historical time sequence data set and a historical node graph; integral commodity exchange information is generated based on the final physical state prediction result and is sent to the user terminal; according to the invention, based on the previous purchase data, behavior data, feedback data and health data of the user and the previous case data of the corresponding user, the hidden associated information can be effectively obtained, so that the future physical state of the user is predicted, and a time period is set before a disease occurs; and adaptively changing and pushing the commodity exchange information of the point shopping mall of the corresponding user.

Description

Technical Field

[0001] The present invention relates to the technical field of point malls, and more specifically, to a point redemption system and method for point malls. Background Art

[0002] As e-commerce continues to evolve, mall points redemption systems are playing an increasingly important role in motivating user behavior and enhancing brand loyalty.

[0003] However, the existing point system has many shortcomings, such as the opacity of the point value, the single redemption method and the low user participation. In addition, when the existing point mall recommends products that can be redeemed with points to users, it mostly makes recommendations based on the mall's accumulated goods information, user search, browsing information and user's past purchase information. Although this method can effectively obtain information data on the goods that the current user needs or wants to purchase, if the user has not searched, browsed or added to the shopping cart in the recent period, or the user's required purchase of the product does not match the user's search, browsing or shopping cart information data, the point mall will lose accurate recommendation parameters, resulting in the point redemption products pushed to the corresponding user not containing the products that the current user needs or really wants to buy. That is, the existing point mall cannot predict the user's demand parameters and hidden needs at the current or future corresponding time points based on past data, which will cause the point redemption system to lose the corresponding order volume, reduce user satisfaction, and thus lose customers. Summary of the invention

[0004] The purpose of the present invention is to provide a points redemption system and method for a points mall in order to solve the above-mentioned problems.

[0005] The present invention provides a points redemption method for a points mall, comprising the following steps:

[0006] Step 100: Obtain the user's historical purchase data, behavior data, feedback data, and health data, and obtain the corresponding user's public medical case data from the public medical database, and pre-process the user's historical purchase data, behavior data, feedback data, health data, and public medical case data;

[0007] Step 200: construct a historical time series data set based on the pre-processed historical purchase data, behavior data, and feedback data of the user;

[0008] Step 300: construct a historical node graph based on the pre-processed historical health data and public medical case data of the user;

[0009] Step 400: construct a physical state prediction model, and use the historical time series data set and the historical node graph to train the physical state prediction model to obtain a trained physical state prediction model;

[0010] The body state prediction model includes LSTM branch, GNN branch and output layer;

[0011] Step 500: Obtain the user's real-time purchase data, behavior data, feedback data, and health data, process the user's real-time purchase data, behavior data, feedback data, and health data to generate a real-time time series data set and a real-time node graph, input the real-time time series data set and the real-time node graph into the trained body state prediction model to generate the final body state prediction result, generate point product redemption information based on the final body state prediction result and send it to the user terminal.

[0012] As a further optimization solution of the present invention, the historical purchase history data includes transaction records, commodity categories, and purchase frequency;

[0013] Historical behavior data includes browsing behavior, search history, and click behavior;

[0014] Historical feedback data includes product ratings, product feedback, and service reviews;

[0015] Historical health data includes basic user information and lifestyle information;

[0016] User basic information includes name, ID number, age, gender, height, weight, health status, allergy history, and family medical history;

[0017] Lifestyle information includes exercise status, daily routine, drinking habits, stress level, smoking and drinking;

[0018] Public medical case data includes basic patient information, medical history information, symptoms and signs, diagnosis results, treatment plans, and efficacy evaluation.

[0019] As a further optimization scheme of the present invention, the preprocessing of the user's historical purchase data, behavior data, feedback data, health data and public medical case data specifically includes the following steps:

[0020] 1) Data cleaning:

[0021] Deduplication: Eliminate duplicate records;

[0022] Missing value processing: fill in or delete missing information;

[0023] Format standardization: unified data format;

[0024] Outlier detection: identifying and processing abnormal data;

[0025] Data conversion: convert different data types into numerical types that the model can understand;

[0026] 2) Feature Engineering:

[0027] Purchase frequency: calculate the number of purchases per unit time;

[0028] Average transaction amount: calculate the average amount of each transaction;

[0029] Page View Duration: Calculates the average time a user spends on a specific page;

[0030] Search keyword frequency: record the search frequency of keywords;

[0031] Click-through rate: calculate the click-through rate of users on recommended products;

[0032] Feedback sentiment analysis: extract the sentiment score of the feedback text;

[0033] Feedback rating stability: Calculate the consistency or standard deviation of user ratings;

[0034] BMI: calculated by height and weight;

[0035] Health risk indicators: construct health risk scores based on family medical history and past medical history;

[0036] Healthy Lifestyle Score: Assess your health level based on diet and exercise information.

[0037] As a further optimization scheme of the present invention, the historical time series data set is constructed based on the pre-processed historical purchase data, behavior data, and feedback data of the user, specifically:

[0038] Align all data by timestamp to ensure that data at the same time point corresponds to the same time series;

[0039] Different types of data are concatenated into a vector containing multi-dimensional features to form a complete time series X = {x1, x2, ..., x t}, where x t represents the feature vector at time t.

[0040] As a further optimization solution of the present invention, the historical node graph is constructed based on the pre-processed historical health data and public medical case data of the user, specifically:

[0041] Construct a node graph G = (V, E), where V represents a node set, each node represents a user, E represents an edge set, each node has an edge, and an edge represents the similarity between two nodes;

[0042] Constructing a node feature vector for each node based on the pre-processed user's historical health data and public medical case data;

[0043] The weight of the edge between each node is calculated based on the node feature vector.

[0044] As a further optimization solution of the present invention, the LSTM branch includes a first hidden layer and a first fully connected layer; the first hidden layer is used to capture sequence features, and the first fully connected layer is used to map the output of the first hidden layer to a first body state prediction result;

[0045] The calculation formula for the first hidden layer is:

[0046] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0047] i t =σ(W i ·[h t-1 ,x t ]+b i );

[0048]

[0049] o t =σ(W o ·[h t-1 ,x t ]+b o );

[0050] h t =o t ·tanh(c t );

[0051] Among them, f t represents the forget gate vector, σ represents the Sigmoid activation function, W f represents the weight matrix of the forget gate, h t-1 represents the hidden state of the previous time step, x t represents the input sequence feature vector of the current time step, b f Represents the bias vector of the forget gate, i t represents the input gate vector, represents the new candidate memory, tanh represents the hyperbolic tangent activation function, Wi , W c represents the weight matrix of the input gate and candidate memory, b i 、b c represents the bias vector of the input gate and candidate memory, c t represents the cell state at the current time step, c t-1 represents the cell state at the previous time step, ⊙ element-wise multiplication, o t represents the output gate vector, h t represents the hidden state of the current time step, W o represents the weight matrix of the output gate, b o represents the bias vector of the output gate;

[0052] The calculation formula of the first fully connected layer is:

[0053] Q t =W q h t +b q ;

[0054] Among them, Q t represents the output of the first fully connected layer, W q represents the weight matrix of the first fully connected layer, b q Represents the bias vector of the first fully connected layer.

[0055] As a further optimization solution of the present invention, the GNN branch includes a second hidden layer and a second fully connected layer; the second hidden layer is used to capture the relationship in the graph structure and the relationship aggregation information between nodes, and the second fully connected layer is used to map the output of the second hidden layer to a second body state prediction result;

[0056] The calculation formula for the second hidden layer is:

[0057]

[0058] Among them, S i represents the vertex update feature of the i-th vertex, φ i represents the aggregation coefficient of the i-th vertex, K i represents the set of vertices that have edges with the i-th vertex, σ represents the activation function, W2 represents the second weight, E j Represents the vertex feature of the jth vertex;

[0059] The calculation formula of the aggregation coefficient of the i-th vertex is:

[0060]

[0061] Z i =W1E i ;

[0062] Z j =W1E j ;

[0063] Among them, E i and E j Represent the vertex features of the i-th and j-th vertices respectively, W1 represents the first weight, a represents the third weight, T represents transpose, exp represents the natural exponential function, LeakyRelu represents the LeakyRelu activation function, and K i represents the set of vertices that have edges with the i-th vertex;

[0064] The calculation formula of the second fully connected layer is:

[0065] Y i =W y S i +b y ;

[0066] Among them, Y represents the output of the second fully connected layer, W y represents the weight matrix of the second fully connected layer, b y Represents the bias vector of the second fully connected layer.

[0067] As a further optimization scheme of the present invention, the calculation formula of the output layer is:

[0068] R=α·Q t +(1-α)Y i ;

[0069] Among them, R represents the final body state prediction result, and α represents the fusion coefficient.

[0070] A points redemption system for a points mall, used to execute a points redemption method for a points mall as described above, comprising:

[0071] The data acquisition module acquires the user's historical purchase data, behavior data, feedback data, and health data, and acquires the corresponding user's public medical case data from the public medical database, and pre-processes the user's historical purchase data, behavior data, feedback data, health data, and public medical case data;

[0072] The data processing module constructs a historical time series data set based on the pre-processed user's historical purchase data, behavior data, and feedback data;

[0073] Construct a historical node graph based on the pre-processed historical health data and public medical case data of the user;

[0074] A body state prediction module is used to construct a body state prediction model, and train the body state prediction model using a historical time series data set and a historical node graph to obtain a trained body state prediction model;

[0075] The body state prediction model includes LSTM branch, GNN branch and output layer;

[0076] The LSTM branch includes a first hidden layer and a first fully connected layer; the first hidden layer is used to capture sequence features, and the first fully connected layer is used to map the output of the first hidden layer to a first body state prediction result;

[0077] The GNN branch includes a second hidden layer and a second fully connected layer; the second hidden layer is used to capture the relationship in the graph structure and the relationship aggregation information between nodes, and the second fully connected layer is used to map the output of the second hidden layer to a second body state prediction result;

[0078] The points redemption push module is used to obtain the user's real-time purchase data, behavior data, feedback data, and health data, and generate a real-time time series data set and a real-time node graph after processing the user's real-time purchase data, behavior data, feedback data, and health data. The real-time time series data set and the real-time node graph are input into the trained physical state prediction model to generate the final physical state prediction result, and the points product redemption information is generated based on the final physical state prediction result and sent to the user terminal.

[0079] A storage medium stores non-transitory computer-readable instructions. When the non-transitory computer-readable instructions are executed by a computer, the steps in the points redemption system and method of the points mall as described above can be executed.

[0080] The beneficial effects of the present invention are as follows: based on the user's past purchase data, behavior data, feedback data, health data and the corresponding user's past case data, the present invention can effectively obtain the hidden related information therein, thereby predicting the user's future physical state, and setting a time period before the disease occurs, and adaptively changing and pushing the corresponding user's point mall product redemption information, thereby effectively improving the point utilization rate of the point mall and the user's trust and recognition of the point mall. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 It is a schematic diagram of a flow chart of a points redemption method of a points mall of the present invention;

[0082] Figure 2 It is a module block diagram of a points redemption system of a points mall of the present invention. DETAILED DESCRIPTION

[0083] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only intended to enable those skilled in the art to better understand and implement the subject matter described herein. In addition, the features described relative to some examples may also be combined in other examples.

[0084] like Figure 1 As shown, a point redemption method of a point mall includes the following steps:

[0085] Step 100: Obtain the user's historical purchase data, behavior data, feedback data, and health data, and obtain the corresponding user's public medical case data from the public medical database, and pre-process the user's historical purchase data, behavior data, feedback data, health data, and public medical case data;

[0086] Among them, historical purchase history data includes transaction records, product categories, and purchase frequency;

[0087] Transaction records: Record the detailed information of each transaction.

[0088] Example: The user purchased 2 units of "Vitamin C" on September 15, 2024, for a transaction amount of RMB 100.

[0089] Product category: Categorize and record the type of products purchased.

[0090] Examples: health supplements, medicines, personal care products, etc.

[0091] Purchase frequency: The number of times a user makes purchases within a specific time period.

[0092] Example: The user purchased health products five times in the past three months.

[0093] Among them, historical behavior data includes browsing behavior, search history, and click behavior;

[0094] Browsing behavior: user's visit records in the mall.

[0095] Example: A user browsed the page for "immune enhancers" on September 10, 2024, and stayed on the page for 5 minutes.

[0096] Search history: Keywords entered by users in the search box.

[0097] Example: Users recently searched for keywords such as "drugs to boost immunity" and "constipation remedies".

[0098] Click behavior: product links and recommendations clicked by users.

[0099] Example: User clicked on "Probiotics", viewed its details, and then returned to the home page.

[0100] Among them, historical feedback data includes product ratings, product feedback, and service evaluations;

[0101] Product ratings: User reviews and ratings of purchased products.

[0102] Example: A user gave a 4-star rating to “Calcium Tablets” and left a comment “The effect is good, but the packaging is not so good”.

[0103] Product feedback: user experience and effects of the product.

[0104] Example: A user submitted feedback about an “anti-fatigue drug” and said that he felt significantly better after using it.

[0105] Service evaluation: User satisfaction evaluation of mall services.

[0106] Example: A user commented in customer service feedback that “delivery was fast and customer service was great.”

[0107] Among them, historical health data includes basic user information and lifestyle information;

[0108] User basic information includes name, ID number, age, gender, height, weight, health status, allergy history, and family medical history;

[0109] Name: User's name;

[0110] Example: Li Hua;

[0111] ID card number; not used as an example here;

[0112] Age: the user’s birth year or age;

[0113] Example: 28 years old;

[0114] Gender: user's gender;

[0115] Example: Female;

[0116] Height: User's height (in centimeters);

[0117] Example: 162 cm;

[0118] Weight: user's weight (in kilograms);

[0119] Example: 56 kg;

[0120] Health status: the user’s current health status;

[0121] Example: no history of major illness, occasional cold;

[0122] Allergy history: whether the user has allergic reactions and allergens;

[0123] Example: Allergy to penicillin;

[0124] Family medical history: whether there are genetic diseases in the family;

[0125] Example: Father has high blood pressure and mother has diabetes;

[0126] Among them, lifestyle information includes exercise status, work and rest routine, drinking habits, stress level, smoking and drinking;

[0127] Eating habits: users’ daily eating preferences;

[0128] Examples: prefer vegetarian diet, eat fruits and vegetables frequently, and prefer a low-sugar diet;

[0129] Exercise status: the user's exercise frequency and type;

[0130] Example: Work out 4 times a week, including running and yoga;

[0131] Work and rest habits: the user's work and rest time and sleep quality;

[0132] Example: Go to bed at 11pm every night, get up at 7am, and have good sleep quality;

[0133] Drinking habits: the user’s daily water intake;

[0134] Example: Drink about 2 liters of water per day;

[0135] Stress level: the user’s sources of stress and how to manage them;

[0136] Example: Work stress is high, take meditation and weekend trips to relieve stress;

[0137] Smoking and drinking: users’ tobacco and alcohol consumption;

[0138] Example: No smoker, occasional drinker, drinks once a month at a party.

[0139] Among them, public medical case data includes basic patient information, medical history information, symptoms and signs, diagnosis results, treatment plans, and efficacy evaluation;

[0140] Basic information of the patient:

[0141] Examples: name, ID number, gender, age, height, weight.

[0142] Sample data: male, 32 years old, 172 cm, weight 75 kg.

[0143] Medical history information:

[0144] Examples: past medical history, family medical history.

[0145] Sample data: No history of chronic disease, parents have hypertension.

[0146] Symptoms and Signs:

[0147] Example: Description of symptoms during initial visit.

[0148] Example Data: A patient reports fatigue, headaches, and occasional diarrhea on initial visit.

[0149] Diagnosis:

[0150] Example: Type of confirmed disease and related test results.

[0151] Example data: A patient is diagnosed with allergic rhinitis and test results show elevated IgE levels.

[0152] Treatment options:

[0153] Examples: Medications and therapies taken during treatment.

[0154] Example data: Allergy immunotherapy with antihistamines.

[0155] Efficacy evaluation:

[0156] Example: Feedback on symptom improvement after treatment.

[0157] Sample data: After three months of treatment, the symptoms improved significantly and the headache disappeared.

[0158] In one embodiment of the present invention, public medical databases include the Centers for Disease Control, the World Health Organization, the Health Information Administration, etc., and open data provided by the corresponding databases can be obtained, and some case data can also be obtained from medical research databases (such as PubMed, ClinicalTrials.gov) and clinical case reports.

[0159] In one embodiment of the present invention, the user's historical purchase data, behavior data, feedback data, health data, and public medical case data are preprocessed, specifically including the following steps:

[0160] 1) Data cleaning:

[0161] Deduplication: Eliminate duplicate records.

[0162] Example: If a user purchases the same item multiple times on the same date, keep one record.

[0163] Missing value handling: filling or removing missing information.

[0164] Example: If a transaction is missing a product name, delete the record or fill it in based on other records;

[0165] If the user does not fill in "Height", you can decide to fill in the mean or delete the record;

[0166] If a case lacked important diagnostic information, it was marked or deleted.

[0167] Format standardization: unified data format;

[0168] like:

[0169] Timestamp processing: ensure the consistency of time information, that is, convert timestamps in different formats into a standard time format.

[0170] Example: Change the purchase date format to "YYYY-MM-DD", such as changing "2024 / 09 / 15" to "2024-09-15".

[0171] Standardized field names: Unify field names for health information.

[0172] Example: "Weight" and "Weight" are unified into "Weight (kg)";

[0173] Unify "hypertension" into "hypertension";

[0174] Convert non-standard disease names into ICD codes, such as "diabetes" is marked as "E10-E14";

[0175] Unify the two labels of "care products" and "skin care products" into one;

[0176] Unify ratings to one standard, such as standardizing 5-star reviews to scores from 1 to 5;

[0177] Encode the purchased commodity categories into a unified code, such as "Vitamin C" is coded as 1, "Protein powder" is coded as 2, etc.

[0178] Outlier detection: Identify and handle abnormal data.

[0179] Example: A user browsed 200 products in one hour. This may be a data collection error and should be investigated.

[0180] The user's height is marked as 250 cm. Verification should be performed and a decision should be made to fill in the mean or delete the record.

[0181] Data conversion: Convert different data types into numerical types that the model can understand.

[0182] Example: Encode categorical data as numeric values; for example, one-hot encode gender;

[0183] The user name is one-hot encoded to form a unique identification value for each user.

[0184] 2) Feature Engineering:

[0185] Purchase frequency: Calculate the number of purchases per unit time.

[0186] Example: In the past three months, the user made 10 purchases.

[0187] Average transaction amount: Calculate the average amount of each transaction.

[0188] Example: The user spent a total of 1,000 yuan, the number of transactions was 10, and the average transaction amount was 100 yuan.

[0189] Page View Duration: Calculates the average time a user spends on a specific page.

[0190] Example: The average time a user spends on the “Immunity Booster” page is 5 minutes.

[0191] Search keyword frequency: record the search frequency of keywords.

[0192] Example: "Boost immunity" was searched 3 times, and "constipation remedy" was searched 2 times.

[0193] Click-through rate: Calculates the click rate of users on recommended products.

[0194] Example: 8 products are recommended and the user clicks on 4 of them, with a click-through rate of 50%.

[0195] Feedback Sentiment Analysis: Extract the sentiment score (positive, negative) of the feedback text.

[0196] Example: A user’s review of “probiotics” is “works great” and the sentiment score is 0.8 (positive).

[0197] Feedback rating stability: Calculate the consistency or standard deviation of user ratings.

[0198] Example: Users rated 5 products 4, 5, 4, 5, 4, with a standard deviation of 0.45 (low).

[0199] BMI (Body Mass Index): Calculated by height and weight.

[0200] Example: The user is 170cm tall and weighs 65kg, so BMI = 65 / (1.7^2)≈22.49.

[0201] Health risk indicators: Build a health risk score based on family medical history, past medical history, etc.

[0202] Example: Having a family history of hypertension is scored +1, for a total score of 2 (low risk).

[0203] Healthy Lifestyle Score: Assess your health level based on information such as diet, exercise, etc.

[0204] Example: Exercising 3 times a week is worth 2 points, and eating a balanced diet is worth 3 points, for a total score of 5 points (out of 10).

[0205] It should be noted that after preprocessing, the raw data can be converted into high-quality features, providing strong support for subsequent analysis and model training.

[0206] Step 200: Based on the pre-processed historical purchase data, behavior data, and feedback data of the user, a historical time series data set is constructed, specifically:

[0207] Align all data by timestamp to ensure that data at the same time point corresponds to the same time series;

[0208] For example:

[0209] Timestamp category amount stay_time clicks rating sentiment_score 2024-01-01 1 0.25 0.1 0.3 0.8 0.8 2024-01-07 2 0.6 0.2 0.5 1.0 0.9 2024-01-15 1 0.3 0.14 0.4 0.6 0.6 ... ... ... ... ... ... ...

[0210] Different types of data are concatenated into a vector containing multi-dimensional features to form a complete time series X = {x1, x2, ..., x t}, where x t represents the feature vector at time t;

[0211] In the above example, x1 = {1, 0.25, 0.1, 0.3, 0.8, 0.8}, x2 = {2, 0.6, 0.2, 0.5, 1.0, 0.9}. It should be noted that this is only an example and does not represent the overall time series characteristics.

[0212] Step 300: construct a historical node graph based on the pre-processed historical health data and public medical case data of the user, specifically:

[0213] Construct a node graph G = (V, E), where V represents a node set, each node represents a user, E represents an edge set, each node has an edge, and an edge represents the similarity between two nodes;

[0214] Constructing a node feature vector for each node based on the pre-processed user's historical health data and public medical case data;

[0215] For example, let the node feature vector of user 1 be y1, and the node feature vector of user 2 be y2;

[0216] y1={170,70,22.49,1,3,1,1,0,1,0};

[0217] y2={165,68,21.57,3,4,1,0,1,1,0};

[0218] Among them, the first five dimensions represent height, weight, BMI, health risk index, and healthy lifestyle score, and the last five dimensions represent the binary presence or absence of certain diseases or drugs, such as diabetes, hypertension, heart disease, drug A, and drug B;

[0219] The weight of the edge between each node is calculated based on the node feature vector. The edge weight calculation formula is the Euclidean distance calculation formula.

[0220] Take the above calculation of the node feature vector y1 of user 1 and the node feature vector y2 of user 2 as an example:

[0221]

[0222] Step 400: construct a physical state prediction model, and use the historical time series data set and the historical node graph to train the physical state prediction model to obtain a trained physical state prediction model;

[0223] The body state prediction model includes an LSTM branch, a GNN branch and an output layer; the historical time series data set in step 200 is used as an input sequence of the LSTM branch, and the historical node graph in step 300 is used as an input of the GNN branch;

[0224] The LSTM branch includes a first hidden layer and a first fully connected layer; the first hidden layer is used to capture sequence features, and the first fully connected layer is used to map the output of the first hidden layer to a first body state prediction result;

[0225] The calculation formula for the first hidden layer is:

[0226] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0227] i t =σ(W i ·[h t-1 ,x t ]+b i );

[0228]

[0229] o t =σ(W o ·[ht-1 ,x t ]+b o );

[0230] h t =o t ·tanh(c t );

[0231] Among them, f t represents the forget gate vector, which determines how much previous memory is discarded; σ represents the Sigmoid activation function, which scales the output to [0,1]; W f represents the weight matrix of the forget gate; h t-1 represents the hidden state of the previous time step; x t Represents the input sequence feature vector of the current time step; b f Represents the bias vector of the forget gate; i t Represents the input gate vector, which determines how much new information is updated at this moment; W represents a new candidate memory, which is used to update the state; tanh represents the hyperbolic tangent activation function, which scales the output to [-1,1]. i , W c represents the weight matrix of the input gate and candidate memory; b i 、b c represents the bias vector of the input gate and candidate memory; c t Represents the cell state at the current time step; c t-1 Represents the cell state at the previous time step; ⊙ element-wise multiplication; o t Represents the output gate vector, which determines how much current memory state is output; h t represents the hidden state of the current time step; W o represents the weight matrix of the output gate; b o Represents the bias vector of the output gate.

[0232] The calculation formula of the first fully connected layer is:

[0233] Q t =W q h t +b q

[0234] Among them, Q t represents the output of the first fully connected layer, W q represents the weight matrix of the first fully connected layer, b q Represents the bias vector of the first fully connected layer.

[0235] The GNN branch includes a second hidden layer and a second fully connected layer; the second hidden layer is used to capture the relationship in the graph structure and the relationship aggregation information between nodes, and the second fully connected layer is used to map the output of the second hidden layer to a second body state prediction result;

[0236] The calculation formula for the second hidden layer is:

[0237]

[0238] Among them, S i represents the vertex update feature of the i-th vertex, represents the aggregation coefficient of the i-th vertex, K i represents the set of vertices that have edges with the i-th vertex, σ represents the activation function, W2 represents the second weight, E j Represents the vertex feature of the jth vertex;

[0239] The calculation formula of the aggregation coefficient of the i-th vertex is:

[0240]

[0241] Z i =W1E i

[0242] Z j =W1E j

[0243] Among them, E i and E j Represent the vertex features of the i-th and j-th vertices respectively, W1 represents the first weight, represents the third weight, T represents transpose, exp represents the natural exponential function, LeakyRelu represents the LeakyRelu activation function, K i represents the set of vertices that have edges with the i-th vertex;

[0244] The calculation formula of the second fully connected layer is:

[0245] Y i =W y S i +b y

[0246] Among them, Y represents the output of the second fully connected layer, W y represents the weight matrix of the second fully connected layer, b y Represents the bias vector of the second fully connected layer.

[0247] The calculation formula of the output layer is:

[0248] R=α·Q t+(1-α)Y i

[0249] Among them, R represents the final body state prediction result, α represents the fusion coefficient, and the fusion coefficient is determined by verification of training data.

[0250] It should be noted that the final physical condition prediction results include the types of diseases that may occur in the body, suitable drug products, and the corresponding time of disease occurrence.

[0251] In one embodiment of the present invention, the training process of the body state prediction model is as follows:

[0252] Model training process

[0253] 1). Initialization:

[0254] Initialize model parameters such as weights and biases.

[0255] 2). Forward propagation:

[0256] LSTM Feedforward:

[0257] Input time series data and obtain the time-dependent feature representation of each patient.

[0258] GNN feedforward:

[0259] Input the initial graph structure and perform multi-layer message passing to obtain the graph feature representation of each node.

[0260] Feature Fusion:

[0261] Concatenate the LSTM and GNN outputs together to form a comprehensive feature vector.

[0262] Output layer:

[0263] The comprehensive features are mapped into prediction results (disease type, applicable drug product information, and disease occurrence time) through the fully connected layer.

[0264] 3). Loss calculation

[0265] Using a loss function appropriate for the task (such as cross entropy loss), calculate the error between the predicted value and the true label.

[0266] 4) Back Propagation

[0267] Compute the gradient and backpropagate the error via the chain rule.

[0268] 5) Parameter update

[0269] Update the model parameters using an optimization algorithm such as Adam or SGD to minimize the loss.

[0270] 6) Iterative training

[0271] Multiple rounds of iterations are performed on the entire dataset until the loss converges or a set stopping condition is reached.

[0272] 7) Model Validation

[0273] Evaluate model performance on a validation set to prevent overfitting.

[0274] 8) Hyperparameter tuning

[0275] Search for optimal hyperparameters (such as learning rate, batch size, number of network layers) to improve model performance.

[0276] Cross-validation was used to ensure the generalization ability of the model.

[0277] 9) Model Evaluation

[0278] Perform final evaluation on the test set, using metrics such as accuracy, AUC, F1 score, etc.

[0279] It should be noted that the physical condition prediction model can effectively obtain hidden related information based on the user's past purchase data, behavior data, feedback data, health data and the corresponding user's past case data, thereby predicting the user's future physical condition, and setting a time period before the disease occurs, and adaptively changing and pushing the product redemption information of the corresponding user's points mall. After the user obtains the information or develops the corresponding disease within the corresponding time period, he will choose to use points to redeem the corresponding drug products to reduce the actual capital consumption, thereby effectively improving the points utilization rate of the points mall and the user's trust and recognition of the points mall.

[0280] Step 500: Obtain the user's real-time purchase data, behavior data, feedback data, and health data, process the user's real-time purchase data, behavior data, feedback data, and health data to generate a real-time time series data set and a real-time node graph, input the real-time time series data set and the real-time node graph into the trained body state prediction model to generate the final body state prediction result, generate point product redemption information based on the final body state prediction result and send it to the user terminal.

[0281] In one embodiment of the present invention, the user terminal includes a mobile terminal, a PC terminal, a TV terminal, etc.

[0282] like Figure 2 As shown, a points redemption system of a points mall is used to execute a points redemption method of a points mall as described above, including:

[0283] The data acquisition module acquires the user's historical purchase data, behavior data, feedback data, and health data, and acquires the corresponding user's public medical case data from the public medical database, and pre-processes the user's historical purchase data, behavior data, feedback data, health data, and public medical case data;

[0284] The data processing module constructs a historical time series data set based on the pre-processed user's historical purchase data, behavior data, and feedback data;

[0285] Construct a historical node graph based on the pre-processed historical health data and public medical case data of the user;

[0286] A body state prediction module is used to construct a body state prediction model, and train the body state prediction model using a historical time series data set and a historical node graph to obtain a trained body state prediction model;

[0287] The body state prediction model includes LSTM branch, GNN branch and output layer;

[0288] The LSTM branch includes a first hidden layer and a first fully connected layer; the first hidden layer is used to capture sequence features, and the first fully connected layer is used to map the output of the first hidden layer to a first body state prediction result;

[0289] The GNN branch includes a second hidden layer and a second fully connected layer; the second hidden layer is used to capture the relationship in the graph structure and the relationship aggregation information between nodes, and the second fully connected layer is used to map the output of the second hidden layer to a second body state prediction result;

[0290] The points redemption push module is used to obtain the user's real-time purchase data, behavior data, feedback data, and health data, and generate a real-time time series data set and a real-time node graph after processing the user's real-time purchase data, behavior data, feedback data, and health data. The real-time time series data set and the real-time node graph are input into the trained physical state prediction model to generate the final physical state prediction result, and the points product redemption information is generated based on the final physical state prediction result and sent to the user terminal.

[0291] A storage medium stores non-transitory computer-readable instructions. When the non-transitory computer-readable instructions are executed by a computer, the steps in the points redemption system and method of the points mall as described above can be executed.

[0292] The present embodiment is described above, but the present embodiment is not limited to the above-mentioned specific implementation manner. The above-mentioned specific implementation manner is merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, which are all within the protection of the present embodiment.

Claims

1. A points redemption method for a points mall, characterized in that: The following steps are involved: Step 100: Obtain the user's historical purchase data, behavior data, feedback data, and health data, and obtain the corresponding user's public medical case data from the public medical database, and pre-process the user's historical purchase data, behavior data, feedback data, health data, and public medical case data; Step 200: construct a historical time series data set based on the pre-processed historical purchase data, behavior data, and feedback data of the user; Step 300: construct a historical node graph based on the pre-processed historical health data and public medical case data of the user; Step 400: construct a physical state prediction model, and use the historical time series data set and the historical node graph to train the physical state prediction model to obtain a trained physical state prediction model; The body state prediction model includes LSTM branch, GNN branch and output layer; Step 500: Obtain the user's real-time purchase data, behavior data, feedback data, and health data, process the user's real-time purchase data, behavior data, feedback data, and health data to generate a real-time time series data set and a real-time node graph, input the real-time time series data set and the real-time node graph into the trained body state prediction model to generate the final body state prediction result, generate point product redemption information based on the final body state prediction result and send it to the user terminal.

2. A points redemption method for a points mall according to claim 1, characterized in that: The historical purchase history data includes transaction records, product categories, and purchase frequency; Historical behavior data includes browsing behavior, search history, and click behavior; Historical feedback data includes product ratings, product feedback, and service reviews; Historical health data includes basic user information and lifestyle information; User basic information includes name, ID number, age, gender, height, weight, health status, allergy history, and family medical history; Lifestyle information includes exercise status, daily routine, drinking habits, stress level, smoking and drinking; Public medical case data includes basic patient information, medical history information, symptoms and signs, diagnosis results, treatment plans, and efficacy evaluation.

3. A points redemption method for a points mall according to claim 2, characterized in that: The preprocessing of the user's historical purchase data, behavior data, feedback data, health data, and public medical case data specifically includes the following steps: 1) Data cleaning: Deduplication: Eliminate duplicate records; Missing value processing: fill in or delete missing information; Format standardization: unified data format; Outlier detection: identifying and processing abnormal data; Data conversion: convert different data types into numerical types that the model can understand; 2) Feature Engineering: Purchase frequency: calculate the number of purchases per unit time; Average transaction amount: calculate the average amount of each transaction; Page View Duration: Calculates the average time a user spends on a specific page; Search keyword frequency: record the search frequency of keywords; Click-through rate: calculate the click-through rate of users on recommended products; Feedback sentiment analysis: extract the sentiment score of the feedback text; Feedback rating stability: Calculate the consistency or standard deviation of user ratings; BMI: calculated by height and weight; Health risk indicators: construct health risk scores based on family medical history and past medical history; Healthy Lifestyle Score: Assess your health level based on diet and exercise information.

4. A points redemption method for a points mall according to claim 3, characterized in that: The historical time series data set is constructed based on the pre-processed historical purchase data, behavior data, and feedback data of the user, specifically: Align all data by timestamp to ensure that data at the same time point corresponds to the same time series; Different types of data are concatenated into a vector containing multi-dimensional features to form a complete time series X = {x1, x2, ..., x t }, where x t represents the feature vector at time t.

5. A points redemption method for a points mall according to claim 4, characterized in that: The historical node graph is constructed based on the pre-processed historical health data and public medical case data of the user, specifically: Construct a node graph G = (V, E), where V represents a node set, each node represents a user, E represents an edge set, each node has an edge, and an edge represents the similarity between two nodes; Constructing a node feature vector for each node based on the pre-processed user's historical health data and public medical case data; The weight of the edge between each node is calculated based on the node feature vector.

6. A points redemption method for a points mall according to claim 5, characterized in that: The LSTM branch includes a first hidden layer and a first fully connected layer; the first hidden layer is used to capture sequence features, and the first fully connected layer is used to map the output of the first hidden layer to a first body state prediction result; The calculation formula for the first hidden layer is: f t =σ(W f ·[h t-1 ,x t ]+b f ); i t =σ(W i ·[h t-1 ,x t ]+b i ); the t =σ(W o ·[h t-1 ,x t ]+b o ); h t =o t ·tanh(c t ); Among them, f t represents the forget gate vector, σ represents the Sigmoid activation function, W f represents the weight matrix of the forget gate, h t-1 represents the hidden state of the previous time step, x t represents the input sequence feature vector of the current time step, b f Represents the bias vector of the forget gate, i t represents the input gate vector, represents the new candidate memory, tanh represents the hyperbolic tangent activation function, W i , W c represents the weight matrix of the input gate and candidate memory, b i 、b c represents the bias vector of the input gate and candidate memory, c t represents the cell state at the current time step, c t-1 represents the cell state at the previous time step, ⊙ element-wise multiplication, o t represents the output gate vector, h t represents the hidden state of the current time step, W o represents the weight matrix of the output gate, b o represents the bias vector of the output gate; The calculation formula of the first fully connected layer is: Q t =W q h t +b q ; Among them, Q t represents the output of the first fully connected layer, W q represents the weight matrix of the first fully connected layer, b q Represents the bias vector of the first fully connected layer.

7. A points redemption method for a points mall according to claim 6, characterized in that: The GNN branch includes a second hidden layer and a second fully connected layer; the second hidden layer is used to capture the relationship in the graph structure and the relationship aggregation information between nodes, and the second fully connected layer is used to map the output of the second hidden layer to a second body state prediction result; The calculation formula for the second hidden layer is: S i =σ[∑ j∈Ki f i W2E j ]; Among them, S i represents the vertex update feature of the i-th vertex, φ i represents the aggregation coefficient of the i-th vertex, K i represents the set of vertices that have edges with the i-th vertex, σ represents the activation function, W2 represents the second weight, E j Represents the vertex feature of the jth vertex; The calculation formula of the aggregation coefficient of the i-th vertex is: WITH i =W1E i ; WITH j =W1E j ; Among them, E i and E j Represent the vertex features of the i-th and j-th vertices respectively, W1 represents the first weight, a represents the third weight, T represents transpose, exp represents the natural exponential function, LeakyRelu represents the LeakyRelu activation function, and K i represents the set of vertices that have edges with the i-th vertex; The calculation formula of the second fully connected layer is: Y i =W y S i +b y ; Among them, Y represents the output of the second fully connected layer, W y represents the weight matrix of the second fully connected layer, b y Represents the bias vector of the second fully connected layer.

8. A points redemption method for a points mall according to claim 7, characterized in that: The calculation formula of the output layer is: R=α·Q t +(1-a)Y i ; Among them, R represents the final body state prediction result, and α represents the fusion coefficient.

9. A points redemption system for a points mall, characterized in that: It is used to execute a points redemption method of a points mall as described in any one of claims 1 to 8, comprising: The data acquisition module acquires the user's historical purchase data, behavior data, feedback data, and health data, and acquires the corresponding user's public medical case data from the public medical database, and pre-processes the user's historical purchase data, behavior data, feedback data, health data, and public medical case data; The data processing module constructs a historical time series data set based on the pre-processed user's historical purchase data, behavior data, and feedback data; Construct a historical node graph based on the pre-processed historical health data and public medical case data of the user; A body state prediction module is used to construct a body state prediction model, and train the body state prediction model using a historical time series data set and a historical node graph to obtain a trained body state prediction model; The body state prediction model includes LSTM branch, GNN branch and output layer; The LSTM branch includes a first hidden layer and a first fully connected layer; the first hidden layer is used to capture sequence features, and the first fully connected layer is used to map the output of the first hidden layer to a first body state prediction result; The GNN branch includes a second hidden layer and a second fully connected layer; the second hidden layer is used to capture the relationship in the graph structure and the relationship aggregation information between nodes, and the second fully connected layer is used to map the output of the second hidden layer to a second body state prediction result; The points redemption push module is used to obtain the user's real-time purchase data, behavior data, feedback data, and health data, and generate a real-time time series data set and a real-time node graph after processing the user's real-time purchase data, behavior data, feedback data, and health data. The real-time time series data set and the real-time node graph are input into the trained physical state prediction model to generate the final physical state prediction result, and the points product redemption information is generated based on the final physical state prediction result and sent to the user terminal.

10. A storage medium, characterized in that: It stores non-transitory computer-readable instructions, and when the non-transitory computer-readable instructions are executed by a computer, it can execute the steps in a points redemption method of a points mall as described in any one of claims 1-8.

Images