Artificial Intelligence-Based Internet Big Data Analysis Method and System

Abstract

The present invention provides an Internet big data analysis method and system based on artificial intelligence. The method includes: collecting heterogeneous data from different data sources in the medical device industry and performing standardization processing; extracting key features from the standardized data and fusing them into a unified comprehensive feature vector; generating a dynamically updated user profile according to the comprehensive feature vector; constructing a dynamically updated knowledge graph by combining the medical device industry with the dynamically updated user profile to keep the market changes up-to-date; using the knowledge graph for market analysis to provide product demand prediction and user recommendation; and providing a personalized marketing plan according to the comprehensive market strategy suggestions and optimizing it in real time. Through the fusion processing of multi-dimensional data and the construction of an industry knowledge graph, the present invention realizes the comprehensive capture and dynamic update of user needs, and provides intelligent market decision support based on the knowledge graph, overcoming the limitations of the prior art.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence, and particularly relates to an Internet big data analysis method and system based on artificial intelligence. Background Art

[0002] With the rapid development of the medical device industry, the production, sales, and circulation of medical devices have become increasingly complex. The medical device industry covers multiple entities such as hospitals, clinics, pharmacies, and suppliers. The needs of these entities vary, and with the intensification of market competition and policy changes, how to effectively conduct market insight, user management, and precision marketing has become a key challenge for major medical device companies. Against this background, intelligent systems based on Internet and big data technologies have gradually become an effective means to optimize processes and improve efficiency. However, existing Internet big data analysis systems still have many deficiencies when facing the unique needs of the medical device industry.

[0003] First of all, a major problem with the existing technology lies in the limitation of data sources. Most big data analysis systems can only process single-dimensional or structured data, such as order data, user basic information, etc., and it is difficult to accurately judge the all-round needs of users. However, the user needs in the medical device industry are complex and diverse, including not only structured sales and order data, but also a large amount of unstructured data, such as user consultation records, social media interactions, government tender information, etc. If these diverse data cannot be comprehensively integrated and analyzed, it is difficult for enterprises to accurately grasp market needs, leading to a single marketing strategy and poor product promotion effect. In addition, the generation of user portraits in existing systems mostly relies on static data, and it is difficult to dynamically capture changes in user needs, resulting in user portraits lagging behind the actual market situation.

[0004] Secondly, the fragmentation of market information also restricts the decision-making efficiency of enterprises. The medical device industry involves multiple links and a complex supply chain network. From product production, tender procurement to sales and logistics, the data sources are extensive and there is a strong correlation between them. Existing systems cannot effectively integrate these cross-domain market data and lack in-depth insight into the industry panorama. Most current analysis systems rely on traditional data mining and simple statistical models, and it is difficult to discover potential relationships and market opportunities hidden behind the data. Especially in complex scenarios such as partner selection, policy impact analysis, and market demand prediction, the existing technology shows obvious limitations.

[0005] In addition, the existing big data analysis systems have a low level of intelligence in providing market decision-making support. Most systems can only provide simple data reports and trend charts, lacking real intelligent decision-making support and unable to automatically provide the best decision-making suggestions for enterprises based on industry knowledge and market dynamics. Such decision-making support systems cannot respond to the rapidly changing market environment, making it difficult for enterprises to make quick and accurate responses in the face of market competition. Summary of the Invention

[0006] The object of the present invention is to design an Internet big data analysis method and system based on artificial intelligence, which realizes the comprehensive capture and dynamic update of user needs through the fusion processing of multi-dimensional data and the construction of an industry knowledge graph, and provides intelligent market decision-making support based on the knowledge graph, overcoming the limitations of the prior art.

[0007] To achieve the above object, in the first aspect of the present invention, there is provided an Internet big data analysis method based on artificial intelligence, the method comprising:

[0008] Step 1, collect heterogeneous data from different data sources in the medical device industry, perform standardization processing, extract key features from the standardized data, and fuse them into a unified comprehensive feature vector z f ;

[0009] Step 2, generate a dynamically updated user portrait according to the comprehensive feature vector z f Specifically including:

[0010] Step 2.1, use a linear layer plus an activation function to perform demand prediction according to the comprehensive feature vector z f At the same time, add a time-varying weight term λt after the linear layer to adjust the weight of demand prediction according to the change of time, ensuring that the weight of old data is the smallest while the weight of new data is the largest;

[0011] Step 2.2, establish an interest preference analysis model according to the comprehensive feature vector z f At the same time, design a policy influence coefficient η p And combine them to calculate the interest distribution of users for different categories of products;

[0012] Step 2.3, design an emotion analysis network according to the comprehensive feature vector z f And user feedback to generate an emotion score and evaluate the user's satisfaction with the product;

[0013] Step 2.4, fuse the demand prediction, interest distribution and emotion score to generate a comprehensive portrait of the user, and at the same time design a dynamic update mechanism based on time decay. Whenever new data enters, the portrait will be updated according to the time weight;

[0014] Step 3: Construct a dynamically updated knowledge graph by integrating the medical device industry with the dynamically updated user profile to keep the market changes up-to-date, specifically including:

[0015] Step 3.1: Collect market information, where the market information includes the characteristic data P = {p1, p2, …, p n} of medical device products, the policy and market information G = {g1, g2, …, g m}, and the supply chain partner information S = {s1, s2, …, s k}, and construct a dynamic knowledge graph based on the market information and the comprehensive user profile;

[0016] Step 3.2: Use a graph neural network model to learn the complex relationships between the nodes of the dynamic knowledge graph, update the node embeddings, and capture the structural information in the graph;

[0017] Step 3.3: Design a dynamic update mechanism based on time decay. Whenever new data enters, the dynamic knowledge graph will automatically update according to the new user behaviors and external environment;

[0018] Step 4: Use the knowledge graph for market analysis, providing product demand prediction and user recommendations, specifically including:

[0019] Based on the embedding representation of the user nodes, combined with the information of the policy nodes and product nodes, predict the future needs of users and market trends. According to the prediction results, recommend the most suitable products and the most suitable supply chain partners for users, and generate the future needs of users, market trends, the most suitable products, and the most suitable supply chain partners into comprehensive market strategy suggestions;

[0020] Step 5: Provide a personalized marketing plan according to the comprehensive market strategy suggestions and optimize it in real time.

[0021] Furthermore, the heterogeneous data includes structured data, unstructured data, and external data; the standardization process specifically includes the following steps:

[0022] For structured data, use a custom interpolation algorithm for missing value filling and outlier detection based on the historical trends in the time series and the sales data of similar products; for unstructured data, perform text cleaning through NLP techniques, and then use a custom variant of the pre-trained language model BERT to embed the text data into the vector space; for external data, extract features through NLP techniques and combine them with the structured data to obtain the combined structured data;

[0023] Then, extract features from the combined structured data and unstructured data and map them to a unified feature space to obtain a unified feature vector representation z, which is expressed as follows:

[0024] z = W s ·x s +W u ·h i

[0025] where x s represents a vector in the structured data, h i represents a vector in the unstructured data, W s represents the processed structured data, W u represents the unstructured data, W s ∈R p×m and W u ∈R p×d where R p×m and R p×d are weight matrices, projecting the structured data and the unstructured data into the feature space of the same dimension p, and z ∈ R p is the fused feature vector.

[0026] Furthermore, the key features of the data after normalization processing are extracted and fused into a unified fused feature vector, specifically including:

[0027] For the structured data x s , according to its specific values and time series characteristics, a linear transformation with a regularization term is used for mapping, expressed as follows:

[0028]

[0029] where z s represents the mapping vector of the structured data, W s ∈R p×m , R p×m is the linear mapping matrix, mapping the m-dimensional structured data into the p-dimensional unified space, and λ is the regularization term coefficient, used to control the model complexity and prevent overfitting;

[0030] For the unstructured data h i , the transformation matrix W u is used to project it into the p-dimensional space, expressed as follows:

[0031] z u =W u ·h i +γ(h i )

[0032] where z u represents the mapping vector of the unstructured data, γ(h i) is the noise suppression term, which is calculated through the attention mechanism or the context analysis model and filters out the irrelevant information in the feedback; W u ∈R p×d ,R p×d is the projection matrix that maps the unstructured text embedding into a p-dimensional unified feature space;

[0033] For the external data z p , a hysteresis effect control term δ(t) is designed. The hysteresis effect control term δ(t) is related to the policy release time and reflects the gradual impact of the policy on the market, which is expressed as follows:

[0034] z p+1 =W p ·z p ·δ(t)

[0035] where, z p+1 represents the updated external data, W p ∈R p×q ,R p×q is the projection matrix;

[0036] The mapping vector z of the structured data s , the mapping vector z of the unstructured data u and the updated external data z p+1 are fused into a comprehensive feature vector z f , which is expressed as follows:

[0037] z f =α s z s +α u z u +α p z p

[0038] where, α s , α u , α p are the adaptive weights of each modal feature, satisfying α s +α u +α p =1;

[0039] According to the comprehensive feature vector z f a dynamic update mechanism is designed to ensure that the feature vector can be adaptively updated over time, which is expressed as follows:

[0040]

[0041] where, β is the time decay coefficient, is the current feature vector, is the feature vector generated from the latest data;

[0042] During the dynamic update process, a regularization term θ is added to control the smoothness of the feature vector, which is expressed as follows:

[0043]

[0044] where θ is the regularization coefficient, is the Euclidean distance between feature vectors, which is used to control the amplitude of feature vector update to ensure smooth transition.

[0045] Furthermore, in the step 2, the demand prediction is performed based on a dynamic demand prediction model, and the dynamic demand prediction model is constructed as follows:

[0046] Based on the input multi-dimensional feature vector z f , a linear layer plus an activation function is used for demand prediction, and the formula is as follows:

[0047] y d =σ(W d ·z f +b d )

[0048] where W d is the demand mapping vector, W d ∈R 1×p , R 1×p is the mapping matrix that maps the p-dimensional feature vector to the demand prediction dimension, b d is the bias term, σ is the activation function, and y d represents the predicted value of the user's future demand, reflecting the possibility of the user purchasing the product in the future;

[0049] A regularization term Ω(z f ) is added after the linear layer to balance the influence of historical data and new data, which is expressed as follows:

[0050]

[0051] where α represents the regularization term coefficient, W d represents the mapping matrix, and z f represents the comprehensive feature vector;

[0052] The interest preference analysis model is constructed as follows:

[0053] A softmax function is used to map the user's preferences for different product categories into a probability distribution, which is expressed as follows:

[0054] y i =softmax(W i ·z f +b i)

[0055] Among them, W i represents the interest weight, and W i ∈R c×p , where R c×p is the weight matrix that maps the multi-modal feature z f to the interest distribution of c product categories, and b i is the bias term;

[0056] The sentiment analysis network is constructed as follows:

[0057] Using a tanh activation function, the output sentiment score ranges between [-1, 1], and the specific formula is as follows:

[0058] y f = tanh(W f ·z f + b f )

[0059] Among them, W f represents the sentiment weight, and W f ∈R 1×p , where R 1×p is the weight matrix, b f is the bias term, and y f is the sentiment score, indicating the user's satisfaction with the product; among them, y f > 0 indicates that the user's feedback on the product is positive, and y f < 0 indicates negative feedback;

[0060] The comprehensive user profile y profile is expressed as follows:

[0061]

[0062] Among them, y d represents the demand prediction of the user, is the interest preference distribution, is the sentiment analysis result.

[0063] Furthermore, in the dynamic demand prediction model, a time-varying weight term λ(t) is also introduced to adjust the weight of demand prediction according to the change of time, which is expressed as follows:

[0064]

[0065] Among them, λ(t) is the time decay function, represents the updated demand prediction, represents the old demand prediction;

[0066] In the interest preference analysis model, a time decay regularization term ρ is also introduced to balance historical interests and future potential interests, which is expressed as follows:

[0067]

[0068] Among them, is the latest interest distribution, is the historical interest distribution; meanwhile, a policy influence coefficient η is also introduced in the interest preference analysis model p , which is used to enhance the weight influence of policy changes on interest preferences, and is expressed as follows:

[0069]

[0070] Among them, η p reflects the influence of policy changes on interests in different product categories;

[0071] In the sentiment analysis network, a sentiment amplification factor u is also introduced to amplify sentiment fluctuations, which is expressed as follows:

[0072]

[0073] Among them, μ is the amplification factor. When the user feedback contains extreme sentiment, μ will automatically amplify the sentiment score to ensure that the sentiment fluctuations of the user can be fully captured by the system.

[0074] Furthermore, the structure of the dynamic knowledge graph is as follows:

[0075] The nodes include: user node C, product node P, policy node G, and policy node G;

[0076] The user node C uses each user portrait γ profile as the feature information of the user node, which includes demand prediction yd, interest preference y i and sentiment feedback y f ; through the user portrait y profile as the feature input, it is expressed as h C = f(y profile ), where f is the feature embedding function;

[0077] In the product node P, each product node represents a medical device product, which is expressed as p i , and the embedding h P representing the product features is = f(p i );

[0078] The policy node G is an industry policy and market information node, which includes the time of policy release, content influence scope, etc., and is expressed as g i , and the embedding h G= f(g i );

[0079] The supply chain node S represents the node of the supply chain participant, denoted as s i , representing the embedding h of the supply chain participant S = f(s i );

[0080] The relationships include: user-product relationship R cp , user-policy relationship R cg and user-supply chain relationship R cs ;

[0081] The user-product relationship R cp is the relationship between the user and the products they are interested in or have purchased, and the edge weight is determined by the user interest distribution and purchase history;

[0082] The user-policy relationship R cg represents the impact of policies on user needs;

[0083] The user-supply chain relationship R cs represents the interaction relationship between the user and supply chain partners, including order delivery history, feedback, and cooperation intensity;

[0084] A dynamic weight mechanism is introduced to construct the adjacency matrix for embedding nodes and relationships, which is expressed as follows:

[0085] The user-product adjacency matrix A cp (i, j) is expressed as follows:

[0086] A cp (i, j) = α1·sim(y i , p j ) + α2·purchase_history(i, j)

[0087] where sim(y i , p j ) represents the similarity between user interests and product features, purchase_history(i, j) represents the historical purchase record between the user and the product, and α1 and α2 are balancing parameters;

[0088] The user-policy adjacency matrix A cg (i, j) is expressed as follows:

[0089] A cg (i, j) = β1·impact(g j , y i ) + β2·policy_relevance(i, j)

[0090] Among them, impact(g j , y i ) represents the potential impact of the policy on user needs, policy_relevance(i, j) represents the relevance between the policy and user needs, and β1 and β2 are used to regulate the weights of different factors;

[0091] User - supply chain adjacency matrix A cs (i, j), is expressed as follows:

[0092] A cs (i, j) = γ1·order_frequency(i, j) + γ2·logistics_efficiency(i, j)

[0093] Among them, order_frequency(i, j) represents the order frequency between the user and the supplier, and logistics_efficiency(i, j) represents the supply chain.

[0094] Furthermore, using the graph neural network model to learn the complex relationships between the nodes of the dynamic knowledge graph, update the node embeddings, and capture the structural information in the graph, specifically including:

[0095] In each layer of the graph neural network, the features of the nodes are propagated through the adjacency matrix and interact with the neighbor nodes. The features of each node are updated through the following formula:

[0096]

[0097] Among them, represents the feature of node i in the l - th layer, N(i) is the set of neighbor nodes of node i, A ij is the relationship weight between nodes i and j in the adjacency matrix, W (1) is the weight matrix of the l - th layer, b (1) is the bias term, and σ is the non - linear activation function;

[0098] At the same time, introduce a relational regularization term to control the complexity of the relationships between nodes, which is expressed as follows:

[0099]

[0100] Among them, A R is the adjacency matrix under different relationship types, α is the regularization coefficient, ||·|| F is the Frobenius norm;

[0101] The dynamic update mechanism based on time decay is expressed as follows:

[0102]

[0103] Among them, is the feature of the current time node i, is the feature update brought by the new data, and β is the time decay coefficient, which controls the weight distribution between the historical feature and the new feature.

[0104] Furthermore, the prediction of the user's future needs and market trends is performed by combining the information of the policy node and the product node with the embedding representation of the user node as follows:

[0105] Based on the message passing mechanism of the graph neural network, the demand prediction of the user node can be calculated by the following formula:

[0106]

[0107] Among them, is the prediction result of the user demand; W d is the weight matrix that converts the user embedding into the demand prediction; α g is the policy influence weight, impact(g) is the potential influence of the policy on the user demand, and h G is the policy embedding; β p is the product influence weight, and h P is the product embedding;

[0108] The most suitable products are recommended for the user according to the prediction result, and the most suitable supply chain partners are recommended as follows:

[0109] Through the similarity calculation of the user node embedding h C and the product node embedding h P and the historical purchase relationship between the user and the product, a recommendation score is generated, which is expressed as follows:

[0110]

[0111] Among them, y rec is the score of the product recommendation; W rec is the weight matrix of the product recommendation; is the inner product of the user embedding and the product embedding, which is used to calculate the similarity; γ·purchase_history(C, P) is the influence of the historical purchase record, where γ is the historical influence coefficient;

[0112] The supply chain partner recommendation is calculated through the embedding similarity between the user node and the supply chain node, as well as the order frequency and logistics efficiency, and is expressed as follows:

[0113]

[0114] Among them, y sc is the recommended score of supply chain partners; W sc is the weight matrix recommended by the supply chain; is the inner product of the user and the supply chain node embedding, used to calculate similarity; δ·order_frequency(C, S) represents the order frequency of the user and the supply chain partner, where δ is the order frequency weight; θ·logistics_efficiency(S) is the impact of logistics efficiency on partner recommendation, where θ is the logistics efficiency weight.

[0115] Furthermore, step 5 specifically includes:

[0116] Step 5.1: Select appropriate marketing channels for different user groups and reasonably allocate marketing resources according to demand forecasting and product recommendation results;

[0117] Step 5.2: Construct an adaptive user clustering model to cluster users according to user portraits and market feedback data, and combine the marketing resources with the results of user clustering to generate personalized marketing content; among them, the adaptive user clustering model clusters the user embedding vector h C to generate multiple user groups;

[0118] Step 5.3: Design a real-time feedback mechanism. By comparing the actual feedback data of personalized marketing content with the expected effect, adjust the current marketing strategy and channel resource allocation. At the same time, design a step size control mechanism to ensure that the strategy optimization process can converge and avoid strategy fluctuations caused by excessive adjustment.

[0119] In the second aspect of the present invention, an Internet big data analysis system based on artificial intelligence is provided. The system includes:

[0120] A heterogeneous data collection module, used to collect heterogeneous data from different data sources in the medical device industry, perform standardized processing, extract key features from the standardized data, and fuse them into a unified comprehensive feature vector z f ;

[0121] A portrait construction module, used to generate a dynamically updated user portrait according to the comprehensive feature vector z f Specifically, it includes:

[0122] Step 2.1: Use a linear layer plus an activation function to perform demand forecasting according to the comprehensive feature vector z f At the same time, add a time-varying weight term λ(t) after the linear layer to adjust the weight of demand forecasting according to the change of time, ensuring that the weight of old data is the smallest while the weight of new data is the largest;

[0123] Step 2.2: According to the comprehensive feature vector zf Build an interest preference analysis model and design a policy influence coefficient η p Combine them to calculate the interest distribution of users for different categories of products;

[0124] Step 2.3. According to the comprehensive feature vector z f And user feedback, design a sentiment analysis network to generate sentiment scores and evaluate user satisfaction with products;

[0125] Step 2.4. Integrate demand prediction, interest distribution, and sentiment scores to generate a comprehensive user profile. At the same time, design a dynamic update mechanism based on time decay. Whenever new data enters, the profile will be updated according to the time weight;

[0126] The dynamic knowledge graph construction module is used to construct a dynamically updated dynamic knowledge graph by combining the medical device industry with the dynamically updated user profile, so that the market changes are kept up-to-date. Specifically, it includes:

[0127] Step 3.1. Collect market information. The market information includes the characteristic data P = {p1, p2,..., p n} of medical device products, the policy and market information G = {g1, g2,..., g m}, and the supply chain partner information S = {s1, s2,..., s k}. Construct a dynamic knowledge graph based on the market information and the comprehensive user profile;

[0128] Step 3.2. Use a graph neural network model to learn the complex relationships between the nodes of the dynamic knowledge graph, update the node embeddings, and capture the structural information in the graph;

[0129] Step 3.3. Design a dynamic update mechanism based on time decay. Whenever new data enters, the dynamic knowledge graph will be automatically updated according to the new user behaviors and external environment;

[0130] The dynamic knowledge graph analysis module is used to perform market analysis using the knowledge graph, provide product demand prediction and user recommendations. Specifically, it includes:

[0131] Through the embedding representation of the user node, combine the information of the policy node and the product node to predict the future needs of users and market trends. According to the prediction results, recommend the most suitable products and the most suitable supply chain partners for users, and generate the future needs of users, market trends, the most suitable products, and the most suitable supply chain partners into comprehensive market strategy suggestions;

[0132] The real-time optimization module is used to provide a personalized marketing plan according to the comprehensive market strategy suggestions and optimize it in real time.

[0133] The beneficial technical effects of the present invention are at least as follows:

[0134] First of all, through the multi-dimensional data fusion technology, the present invention breaks through the limitations of the existing big data analysis system in terms of data processing dimensions. The system can integrate structured and unstructured data in the medical device industry, including users' purchase behaviors, online consultation records, government tender information, industry policies, etc., and uses deep learning and natural language processing technologies to process and fuse multi-modal data. Compared with the deficiencies of traditional technologies that only rely on a single structured data source, the present invention can generate a more accurate and comprehensive user portrait. More importantly, this user portrait is dynamically updated, capable of real-time tracking of users' behavior changes, ensuring that the portrait reflects users' latest needs, and significantly improving the accuracy of marketing strategies.

[0135] Secondly, by constructing a knowledge graph of the medical device industry, the present invention solves the problem that existing systems cannot integrate cross-domain market information. Based on different entities (such as products, users, suppliers) and their relationships in the industry, the knowledge graph forms a networked association between data, which can help the system discover market opportunities and industry rules hidden behind the data. By analyzing these relationships, the system can identify potential changes in market demand, the impact of policies on the market, and discover suitable business partners. Compared with traditional systems that can only provide data reports and trend charts, the knowledge graph of the present invention can provide a deeper level of industry analysis, especially suitable for complex decision-making scenarios such as market trend prediction and partner recommendation.

[0136] Finally, through in-depth analysis of the knowledge graph, the present invention provides an intelligent market decision-making support function, overcoming the problem of insufficient intelligence level of existing systems. The system can not only generate conventional market analysis reports, but also automatically identify potential opportunities in the market, predict product demand, and even provide personalized marketing strategy suggestions based on the knowledge graph. Enterprises can intelligently optimize marketing plans, select the best partners, and conduct market layout through this system, significantly improving the market response speed and the accuracy of decision-making. Brief Description of the Drawings

[0137] The present invention is further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation to the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the following drawings.

[0138] Figure 1 It is a flowchart of the Internet big data analysis method based on artificial intelligence of the present invention. Detailed Embodiments

[0139] Embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation of the present invention.

[0140] In one or more embodiments, as Figure 1 shown, an Internet big data analysis method based on artificial intelligence of the present invention is disclosed, and the method includes steps S1 - S5:

[0141] Step 1: Collect heterogeneous data from different data sources in the medical device industry, perform standardization processing, extract key features from the standardized data, and fuse them into a unified comprehensive feature vector z f .

[0142] Specifically, first, the present invention obtains multi-dimensional data in the medical device industry from multiple sources, and these data include but are not limited to the following categories:

[0143] D s : Structured data, such as sales records, order information, purchase behaviors of users, etc. Each record of D s can be represented as a vector x s = [x s1 , x s2 ,..., x sm , where m is the dimension of the structured data.

[0144] D u : Unstructured data, mainly text data from user consultation records, feedback information, and social media interactions. The present invention represents the unstructured data as a set of texts D u = {t1, t2,..., t n}, where t i is a single text.

[0145] D e : External data, such as government tender information, industry policy changes, etc. These data are publicly available and usually exist in the form of reports and announcements, and can be represented by text data.

[0146] All data sources D = D s ∪D u ∪D e all need to be standardized to ensure data consistency and operability. The input data has multimodality, which is a major difficulty in this patent.

[0147] Furthermore, for the structured data D s, missing value imputation and outlier detection are required. The present invention uses a custom interpolation algorithm for missing value imputation. This method performs dynamic interpolation based on the historical trends in the time series and the sales data of similar products. The specific calculation formula is as follows:

[0148]

[0149] Where is the data point after imputation, is the historical data of similar product i on feature j, and w i is the weight based on historical similarity. This interpolation method can dynamically adjust the weight to ensure adaptability to the similarity between different products, thereby improving the accuracy of handling missing data.

[0150] For unstructured data D u , the present invention performs text cleaning through NLP techniques, including removing stop words, punctuation marks, and redundant information. Then, a custom variant of the pre-trained language model BERT is used to embed the text data into a vector space. Specifically, each piece of text t i after being transformed by the BERT model, obtains a vectorized representation h i , that is:

[0151] h i = BERT(t i )

[0152] Where h i ∈R d is the feature representation of the text, and d is the dimension of the embedding space. The embedded text data will be used as part of the subsequent data fusion.

[0153] For external data D e , the present invention extracts the features of key policy changes or tender information through NLP techniques and combines them with structured data. For example, keywords related to market impact are extracted from government policy texts and converted into actionable metric information.

[0154] Furthermore, after data cleaning and standardization, next, feature extraction is performed on different types of data and mapped into a unified feature space. To integrate structured and unstructured data, the present invention designs a self-supervised learning model that can perform feature extraction and fusion on multi-modal data.

[0155] First, for structured data D s , the present invention directly uses the feature x s as the input. For unstructured text data, the embedded representation h i has been obtained through the BERT model. Next, the present invention defines a fusion function f(xs , h i ), combine the data features of these two different modalities into a unified representation. The specific fusion calculation is as follows:

[0156] z = W s ·x s + W u ·h i

[0157] Among them, W s ∈ R p×m and W u ∈ R p×d are weight matrices, project the structured data and unstructured data into the feature space of the same dimension p, and z ∈ R p is the fused feature vector. This fusion process is designed for the special needs in the field of the present invention, because in the medical device industry, the needs of users depend not only on purchase behavior, but also on unstructured feedback and policy information.

[0158] Finally, after all types of data are processed as above, a unified feature vector representation z is obtained. This feature vector not only includes the purchase behavior of users, but also combines external policy information and user feedback. Each feature vector z is an important input for subsequent generation of user portraits and knowledge graphs.

[0159] Since the behavior and needs of each user are time-sensitive, the feature vector z also needs to be dynamically updated. For example, when new sales data or feedback information enters the system, the feature vector will be recalculated in real time and sent to the subsequent module. This dynamic update mechanism is one of the keys designed by the present system for the medical device field and can ensure that the user portrait always maintains accuracy.

[0160] Specifically, the present invention designs a multi-modal feature mapping network to map data of different modalities to a unified feature space. For the unique properties of each type of data in the medical device industry, the present invention designs different mapping rules for structured data, unstructured data, and policy data respectively.

[0161] First, for the structured data x s , since they are specific numerical values and time series features, the present invention uses a linear transformation with a regularization term for mapping. Particularly considering the overfitting phenomenon that appears in historical data, a penalty term λ is added during the transformation process to control the overly complex mapping:

[0162]

[0163] Among them, W s ∈ R p×mis a linear mapping matrix that maps m-dimensional structured data to a p-dimensional unified space. λ is the regularization term coefficient, which is used to control the model complexity and prevent overfitting. This is particularly important for the medical device industry because order and sales data often have large fluctuations and must be controlled by the regularization term.

[0164] For the unstructured text embedding h i , the present invention uses another transformation matrix W u to project it into a p-dimensional space. At the same time, text data often contains potential noise (such as irrelevant user feedback or sentiment), so the present invention designs an adaptive noise reduction term γ(h i ) to suppress noise according to the context importance weight of the text:

[0165] z u = W u ·h i + γ(h i )

[0166] where γ(h i ) is the noise suppression term, which is calculated by the attention mechanism or the context analysis model to filter out the irrelevant information in the feedback. W u ∈R p×d is the projection matrix that maps the unstructured text embedding to a p-dimensional unified feature space.

[0167] For the external policy data z p , since the policy has a lag and indirectness on the market demand, the present invention designs a lag effect control term δ(t), which is related to the policy release time and reflects the gradual impact of the policy on the market:

[0168] z p = W p ·z p ·δ(t)

[0169] where W p ∈R p×q is the projection matrix, and δ(t) is the time lag function. As time t progresses, the lag effect gradually weakens. This formula fully considers the timeliness of the policy data to ensure that the impact of the policy on user needs and purchase behavior is fully captured.

[0170] Furthermore, in order to unify and fuse the three different modal data into a comprehensive feature vector, the present invention adopts an adaptive weighted fusion strategy. The present invention designs an adaptive weight generation network that dynamically assigns weights according to the data characteristics of each modality to ensure that more important modalities obtain higher weights. The fusion formula is as follows:

[0171] z f = αs z s +α u z u +α p z p

[0172] where α s ,α u ,α p are the adaptive weights of each modal feature, satisfying α s +α u +α p = 1, and they are adaptively generated by the network. These weights are not only related to the modality, but also related to factors such as data timeliness and data quality. The result of this step is the fused unified feature vector z f ∈R p . The specially designed adaptive weights ensure that sudden changes in the data of the medical device industry (such as new policies, major feedback, etc.) can be captured in a timely manner, avoiding the lag problem caused by fixed weights in traditional weighting methods.

[0173] Furthermore, after the feature vector z f is generated, since user behavior and market dynamics will constantly change, the present invention must design a dynamic update mechanism to ensure that the feature vector can be adaptively updated over time. The present invention introduces a time-weighted update algorithm here:

[0174]

[0175] where β is the time decay coefficient, is the current feature vector, is the feature vector generated from the latest data. This weighted update mechanism ensures the timely impact of new data on the user profile, while retaining the long-term memory effect of part of the historical data.

[0176] In addition, in this dynamic update mechanism, the present invention also adds a regularization term θ to control the smoothness of the feature vector, avoiding large fluctuations in the feature vector caused by excessive noise:

[0177]

[0178] where θ is the regularization coefficient, is the Euclidean distance between feature vectors, used to control the amplitude of feature vector update and ensure smooth transition.

[0179] In summary, the process of feature extraction and multi-modal fusion is divided into four steps: 1) Initialization of input data; 2) Cross-modal feature transformation; 3) Feature fusion; 4) Dynamic update of feature vectors. Through this systematic feature extraction scheme, the problems of complex, variable, and dynamically updated multi-modal data in the medical device industry are solved, and highly accurate input is provided for subsequent user profiling and intelligent market decision-making.

[0180] Step 2. Generate a dynamically updated user profile based on the comprehensive feature vector z f

[0181] Specifically, demand prediction is one of the core tasks in user profile generation. Especially in the medical device industry, user demands are often affected by policy changes, seasonal fluctuations, and the user's own historical behavior. To capture these complex influencing factors, the present invention designs an innovative dynamic demand prediction model to predict the user's future demands based on z f Based on the input multi-dimensional feature vector z f , the present invention uses a linear layer plus an activation function for demand prediction, and the formula is as follows:

[0182] y d = σ(W d ·z f + b d )

[0183] where W d ∈R 1×p is a mapping matrix that maps the p-dimensional feature vector to the demand prediction dimension, b d is a bias term, and σ is an activation function. Usually, ReLU or Sigmoid is used to limit the range of the output. y d represents the predicted value of the user's future demand, reflecting the possibility of the user purchasing the product in the future.

[0184] To consider the impact of external policies or market fluctuations on demand, the present invention introduces a time-varying weight term λ(t), whose role is to adjust the weight of demand prediction according to the change of time, ensuring that the weight of old data gradually decreases while the weight of new data gradually increases:

[0185]

[0186] where λ(t) is a time decay function, which can be adjusted exponentially or dynamically based on historical data. This time-varying term enables the system to adapt to market or policy changes and timely adjust the results of demand prediction.

[0187] ​Furthermore, during the demand forecasting process, the user demands in the medical device industry are usually affected by some unforeseen fluctuations. To prevent the model from relying too much on historical data and ignoring potential new demand changes, the present invention designs an innovative regularization term Ω(z f ), which is used to balance the impacts of historical data and new data:

[0188]

[0189] This term controls the overfitting problem of the model parameters in demand forecasting, ensuring that the feature mapping has a certain smoothness and robustness, so that the model can adapt to the fluctuations in demand.

[0190] Furthermore, in addition to demand forecasting, the user's interest preferences for different products are also an important part of the accurate user portrait. The present invention uses the feature vector z f after multi-modal fusion to establish an interest preference analysis model to calculate the interest distribution of the user for different categories of products. The present invention uses a softmax function to map the user's preferences for different product categories into a probability distribution. The specific formula is as follows:

[0191] y i = softmax(W i ·z f + b i )

[0192] where W i ∈ R c×p is the weight matrix that maps the multi-modal feature z f to the interest distribution of c product categories, and b i is the bias term. Through the softmax function, the model outputs a normalized interest distribution, and y i represents the degree of interest of the user in each product category.

[0193] To prevent the interest preference analysis from relying too much on the user's historical purchase behavior and resulting in a single recommendation result, the present invention introduces a time decay regularization term ρ to balance historical interest and future potential interest:

[0194]

[0195] where is the latest interest distribution, is the historical interest distribution. Through this regularization term, the present invention can dynamically adjust in the recommendation, prevent the user portrait from relying too much on historical data, and give higher recommendation weights to new products.

[0196] Furthermore, due to the relatively slow product update and iteration speed in the medical device industry, but the relatively fast policy and demand changes, a policy impact coefficient η is also added to the model p , which is specifically used to enhance the weight impact of policy changes on interest preferences:

[0197]

[0198] where η p reflects the impact of policy changes on the interest in different product categories, ensuring that after the introduction of certain policies, the relevant product categories can obtain higher recommendation weights.

[0199] Furthermore, in the medical device industry, the feedback and emotional tendencies of users will have an important impact on future purchase behaviors and demands. The present invention designs an emotion analysis network, which combines text feedback data and multi-modal fusion feature z f to generate an emotion score y f , which is used to evaluate the satisfaction of users with products. The emotion score model uses a tanh activation function, and the output emotion score ranges between [-1, 1]. The specific formula is as follows:

[0200] y f = tanh(W f ·z f + b f )

[0201] where is the weight matrix, b f is the bias term, and y f is the emotion score, representing the satisfaction of users with products. y f > 0 indicates that the user's feedback on the product is relatively positive, and y f < 0 indicates negative feedback.

[0202] Furthermore, in view of the large emotional changes of users (for example, strong dissatisfaction or extreme satisfaction with certain products), the present invention introduces an emotion amplification factor μ to amplify the emotional fluctuations:

[0203]

[0204] where μ is the amplification factor. When the user's feedback contains extreme emotions, μ will automatically amplify the emotion score to ensure that the emotional fluctuations of the user can be fully captured by the system.

[0205] Furthermore, through demand prediction, interest preference analysis, and emotion analysis, the present invention finally generates a comprehensive user portrait y profile . This portrait can comprehensively reflect the potential needs, interest distribution, and emotional tendencies of users. The specific formula is as follows:

[0206]

[0207] Among them, y d represents the user's demand prediction, is the interest preference distribution, and is the result of sentiment analysis. This portrait will be used to formulate precise marketing strategies and personalized product recommendations in subsequent steps.

[0208] Furthermore, in order to ensure the timeliness of the user portrait, the present invention designs a dynamic update mechanism based on time decay. Whenever new data enters the system, the portrait will be updated according to the time weight:

[0209]

[0210] Among them, β is the time decay coefficient, which controls the weight distribution between the old portrait and the new portrait. This can ensure that the portrait captures new demands while retaining historical information, generating a dynamic user portrait with timeliness.

[0211] Step 3: Construct a dynamically updated knowledge graph by combining the medical device industry with the dynamically updated user portrait, so that the market changes are kept up-to-date.

[0212] Specifically, by generating a user portrait This portrait includes the results of user demand prediction, interest preference distribution, and sentiment analysis. The present invention has also obtained the characteristic data P = {p1, p2,..., p n} of medical device products, policy and market information G = {g1, g2,..., g m}, and supply chain partner information S = {s1, s2,..., s k}. These data will be used to construct a dynamic knowledge graph.

[0213] The purpose of this step is to integrate these structured and unstructured data from different sources into a knowledge graph KG that can describe different entities and their relationships such as users, products, policies, and supply chains in the medical device industry, and to discover potential market opportunities, optimize supply chain cooperation relationships, and better serve user needs through this graph. The knowledge graph will play a key role in subsequent intelligent decision-making processes.

[0214] Furthermore, the construction of the knowledge graph first requires defining each type of entity in the data as a node of the graph, and the relationships between different entities as the edges of the graph.

[0215] Node definition:

[0216] User node C: The portrait y of each user profileAs the feature information of the user node, it includes demand prediction y d , interest preference y i and emotional feedback y f .

[0217] Product node P: Each product node represents a medical device product, and the features of the node include price, performance, type, etc., denoted as p i .

[0218] Policy node G: Industry policy and market information node, including the time of policy release, content influence scope, etc., denoted as g i .

[0219] Supply chain node S: Node representing supply chain participants such as producers, distributors, and logistics companies, denoted as s i .

[0220] Relationship definition:

[0221] The relationships between different nodes are represented by edges:

[0222] User-product relationship R cp : Defines the relationship between a user and the products they are interested in or have purchased. The edge weight is determined by the user's interest distribution and purchase history.

[0223] User-policy relationship R cg : Represents the impact of policies on user needs. Policies may affect user needs through price fluctuations, product compliance, and supply chain changes.

[0224] User-supply chain relationship R cs : Represents the interaction relationship between a user and supply chain partners (such as producers or logistics providers), including order delivery history, feedback, and cooperation intensity.

[0225] Furthermore, in order to effectively model nodes and their relationships, especially the interaction between users and the external environment (such as policies, supply chains), the present invention introduces an adaptive graph embedding model to map different entities and their relationships to a common feature space.

[0226] Node embedding:

[0227] The initial features of each node come from the data generated previously, denoted as:

[0228] User node embedding h C : Through the user profile y profile as the feature input, denoted as h C = f(y profile ), where f is the feature embedding function.

[0229] Product node embedding hP : Embedded h representing product features P = f(p i ).

[0230] Policy node embedding h G : Embedded h representing policy features G = f(g i ).

[0231] Supply chain node embedding h S : Embedded h representing supply chain participants S = f(s i ).

[0232] Adjacency matrix construction: The relationships between different entities are characterized by an adjacency matrix. The present invention introduces a dynamic weight mechanism to construct the adjacency matrix, and these weights change with time or data updates. It is defined as follows:

[0233] User-product adjacency matrix A cp :

[0234] A cp (i, j) = α1 · sim(y i , p j ) + α2 · purchase_history(i, j)

[0235] Where sim(y i , p j ) represents the similarity between user interests and product features, and purchase_history(i, j) represents the historical purchase records of the user and the product. α1 and α2 are balance parameters.

[0236] User-policy adjacency matrix A cg :

[0237] A cg (i, j) = β1 · impact(g j , y i ) + β2 · policy_relevance(i, j)

[0238] Where impact(g j , y i ) represents the potential impact of the policy on user needs, and policy_relevance(i, j) represents the relevance between the policy and user needs. β1 and β2 are used to regulate the weights of different factors.

[0239] User-supply chain adjacency matrix A cs :

[0240] A cs(i, j) = γ1·order_frequency(i, j) + γ2·logistics_efficiency(i, j)

[0241] Among them, order_frequency(i, j) represents the order frequency between the user and the supplier, logistics_efficiency(i, j) represents the logistics efficiency of the supply chain, and γ1 and γ2 control the weights of each factor.

[0242] Furthermore, in order to learn the complex relationships between the nodes in the graph and provide support for subsequent intelligent decision-making, the present invention uses a graph neural network to update the node embeddings and capture the structural information in the graph.

[0243] Furthermore, in each layer of the graph neural network, the features of the nodes are propagated through the adjacency matrix and interact with the neighbor nodes. The features of each node are updated through the following formula:

[0244]

[0245] Among them, represents the feature of node i in the l-th layer, is the set of neighbor nodes of node i, A ij is the relationship weight between nodes i and j in the adjacency matrix, W (l) is the weight matrix of the l-th layer, b (l) is the bias term, and σ is a non-linear activation function (such as ReLU). This formula describes how the user node obtains information from the product, policy, and supply chain nodes and updates the feature representation through multi-layer propagation. This process can effectively capture the potential associations between nodes, such as the indirect impact of a certain policy on user demand or the impact of the logistics efficiency of the supply chain on user satisfaction.

[0246] Furthermore, in order to ensure that the embedding representation of the graph neural network is not overly complex while retaining the important relationships in the graph structure, the present invention introduces a relational regularization term to control the complexity of the relationships between nodes:

[0247]

[0248] Among them, A R is the adjacency matrix under different relationship types, α is the regularization coefficient, and ||·|| F is the Frobenius norm. Through this regularization term, the present invention can prevent the graph structure from being overly complex, thereby maintaining the relationship sparsity in the knowledge graph and making it easier to interpret and apply.

[0249] Furthermore, in the medical device industry, factors such as user needs, market policies, and supply chain efficiency are constantly changing. Therefore, the update mechanism of the knowledge graph must have dynamic adaptability.

[0250] Furthermore, the present invention designs a dynamic update mechanism based on time decay. Whenever new data enters the system, the knowledge graph will be automatically updated according to new user behaviors and the external environment:

[0251]

[0252] where, is the feature of node i at the current moment, is the feature update brought by the new data, and β is the time decay coefficient, which controls the weight distribution between historical features and new features. This mechanism ensures that the knowledge graph can change with market dynamics, maintain the latest state, and provide real-time support for intelligent decision-making.

[0253] Step 4: Use the knowledge graph for market analysis to provide product demand forecasting and user recommendations.

[0254] Specifically, the present invention first combines the embedded representation h C of the user node with the information of the policy node h G and the product node h P to predict the future needs of users and market trends.

[0255] Based on the message passing mechanism of the graph neural network, the demand prediction of the user node can be calculated by the following formula:

[0256]

[0257] where, is the prediction result of the user demand; W d is the weight matrix that converts the user embedding into demand prediction; α g is the policy influence weight, impact(g) is the potential impact of the policy on the user demand, and h G is the policy embedding; β p is the product influence weight, and h P is the product embedding. This formula combines the influences of user embedding, policy, and product to predict the future demand changes of users. To capture the dynamic changes in the market, the influence weights α g and β p of the policy node and the product node can be automatically learned through the training process of the model.

[0258] Furthermore, based on demand forecasting, the system further recommends the most suitable products to the user. The present invention constructs an adaptive product recommendation model according to the interest distribution of the user profile, the impact of policies on the market, and the user's historical purchase records.

[0259] Through the similarity calculation of the user node embedding h C and the product node embedding h P , and the historical purchase relationship between the user and the product, a recommendation score is generated:

[0260]

[0261] where y rec is the score of product recommendation; W rec is the weight matrix of product recommendation; is the inner product of the user embedding and the product embedding, used to calculate the similarity; γ·purchase_history(C, P) is the impact of historical purchase records, where γ is the historical impact coefficient. The product recommendation score realizes precise personalized recommendation by combining the similarity between the user and the product, historical behavior, and policy impact. The softmax function ensures that the recommended products have a normalized score, and the final product list provided to the user will be sorted according to the score.

[0262] Furthermore, in addition to predicting demand and product recommendation, the system also needs to recommend the most suitable supply chain partners for the enterprise to optimize supply chain efficiency, reduce costs, and improve order fulfillment speed. Here, the present invention uses the embedding representation of the supply chain node h S to make recommendations by analyzing the user's demand, product information, and supply chain historical data. The supply chain partner recommendation is calculated through the embedding similarity between the user node and the supply chain node, as well as the order frequency and logistics efficiency:

[0263]

[0264] where y sc is the supply chain partner recommendation score; W sc is the weight matrix of supply chain recommendation; is the inner product of the user and the supply chain node embedding, used to calculate the similarity; δ·order_frequency(C, S) represents the order frequency between the user and the supply chain partner, where δ is the order frequency weight; θ·logistics_efficiency(S) is the impact of logistics efficiency on partner recommendation, where θ is the logistics efficiency weight.

[0265] Through this formula, the system can comprehensively consider the user's demand, product characteristics, order history, and supply chain efficiency to recommend the most suitable partners for the enterprise and help it optimize operations.

[0266] Further, finally, based on the results of demand forecasting, product recommendation, and supply chain partner recommendation, the system generates comprehensive market strategy suggestions. The strategies include product promotion plans, target user group selection, partner optimization, etc. Through the dynamic update of the knowledge graph and the adaptive adjustment of the prediction model, all strategies can respond in a timely manner to changes in market and user demands.

[0267] The system also provides visual decision-making outputs for enterprises, including market trend charts, changes in user demand forecasts, supply chain cooperation suggestions, etc. This information can be used for business analysis by decision-makers to help them better plan resources and market promotion.

[0268] Step 5: Provide personalized marketing solutions based on the comprehensive market strategy suggestions and optimize them in real time.

[0269] Specifically, it is first necessary to select appropriate marketing channels for different user groups and reasonably allocate marketing resources. The selection and allocation of marketing channels should be based on the intensity of user needs, the priority of product recommendations, and historical marketing effects. The present invention designs an innovative channel weight allocation model, which is based on user demand forecasting and recommendation scores and combines the historical performance of channels to dynamically adjust the priority of each channel. The specific weight allocation formula is as follows:

[0270]

[0271] where w c represents the weight of marketing channel c, controlling the resource allocation ratio of each channel; represents the predicted value of user demand, used to measure the future potential demand of users; y rec is the product recommendation score, representing the priority of product recommendations for users; h conv (c) is the historical conversion rate of marketing channel c, representing the performance of this channel in past marketing activities; λ1, λ2, and λ3 are adjustment coefficients used to balance the impacts of demand, recommendation, and historical performance.

[0272] This formula innovatively combines user demand forecasting, product recommendation priority, and historical conversion rate, dynamically optimizing channel weights while ensuring reasonable resource allocation, and achieving more efficient resource utilization.

[0273] Furthermore, in the medical device industry, the needs and preferences of users vary. Therefore, it is crucial to achieve precise marketing through user segmentation. Based on the user portraits and market feedback data in Step 4, the present invention segments users through an adaptive user segmentation model. Each segment can obtain different marketing strategies and personalized content recommendations to ensure that the marketing content matches the interests of users.

[0274] Furthermore, user clustering is performed on the user embedding vector h C to generate multiple user groups. The personalized marketing content for each group is generated by weighting dimensions such as user interest distribution and demand prediction. The formula is as follows:

[0275]

[0276] where y content (g) represents the personalized marketing content of user group g; W content is the weight matrix for content generation; aggregate(h C , g) represents the feature aggregation operation of users within user group g; α is the weight of the recommendation priority, y rec (i) is the product recommendation score; β is the weight of demand prediction, is the demand prediction result for this user group. This formula aggregates the demands, product recommendations, and user feature aggregation after user clustering to generate personalized marketing content, ensuring a high degree of match between the content and user demands, thereby improving the marketing conversion rate.

[0277] Furthermore, the feedback data (such as click-through rate, conversion rate, etc.) in the marketing activity provides an important basis for strategy optimization. The present invention designs a real-time feedback mechanism that continuously adjusts the current marketing strategy and channel resource allocation by comparing the actual feedback data with the expected effect.

[0278] Furthermore, the channel weights are adjusted through real-time feedback. If the actual performance of a certain channel is better than expected, more resources are invested; otherwise, the investment is reduced. The specific formula is as follows:

[0279]

[0280] where, is the updated channel weight; η is the feedback adjustment coefficient that controls the weight adjustment amplitude; actual_rate(c) is the actual conversion rate of channel c; expected_rate(c) is the expected conversion rate of channel c. This feedback adjustment formula can adjust the resource allocation in a timely manner according to market feedback, enabling the marketing strategy to adapt to the dynamic market environment and remain efficient.

[0281] Furthermore, in the medical device industry, factors such as market policies and user demands are dynamically changing. Therefore, the optimization of the marketing strategy requires adaptive adjustment. The present invention designs a step size control mechanism to ensure that the strategy optimization process can gradually converge and avoid strategy fluctuations caused by excessive adjustment.

[0282] Furthermore, after each marketing cycle, the step size η gradually decays to ensure the gradual convergence of the strategy. The formula is:

[0283] η new = η old ·γ

[0284] Where η new is the new step size; η old is the current step size; γ is the step size decay coefficient, usually taking values between (0, 1). Through step size control, the system can dynamically optimize the marketing strategy, ensuring that the strategy converges gradually while responding to market changes and avoiding large fluctuations.

[0285] Finally, after multiple strategy adjustments and optimizations, the system will generate the final marketing strategy output, including:

[0286] Channel resource allocation suggestions: The resource allocation ratios for different channels;

[0287] User segmentation and personalized content suggestions: Personalized marketing strategies for different user groups;

[0288] Long-term strategy optimization path: Based on historical data and market feedback, give suggestions for optimizing future marketing strategies.

[0289] By continuously obtaining new market data and performing cyclic feedback optimization, the system ensures that the marketing strategy can be continuously improved in the long term.

[0290] In this step, through integrating market decision data, intelligent marketing strategy optimization realizes a complete optimization process of marketing channel selection and resource allocation, user segmentation and personalized content generation, real-time feedback and dynamic adjustment. Through an innovative channel weight allocation model, personalized content generation formula, and adaptive step size control mechanism, the system can dynamically respond to market changes, ensure continuous optimization and gradual convergence of the marketing strategy, thereby improving the overall marketing efficiency.

[0291] In the second embodiment of the present invention, an Internet big data analysis system based on artificial intelligence is provided. The system includes:

[0292] A heterogeneous data collection module 101, which is used to collect heterogeneous data from different data sources in the medical device industry, perform standardization processing, extract key features from the standardized data, and fuse them into a unified comprehensive feature vector z f ;

[0293] A portrait construction module 102, which is used to generate a dynamically updated user portrait according to the comprehensive feature vector z f Specifically including:

[0294] Step 2.1. According to the comprehensive feature vector z fUse a linear layer plus an activation function for demand prediction. At the same time, add a time-varying weight term λ(t) after the linear layer to adjust the weight of demand prediction according to the change of time, ensuring that the weight of old data is minimized while the weight of new data is maximized;

[0295] Step 2.2: According to the comprehensive feature vector z f Build an interest preference analysis model and design a policy influence coefficient η p Combine them to calculate the interest distribution of users for different categories of products;

[0296] Step 2.3: According to the comprehensive feature vector z f And user feedback, design a sentiment analysis network to generate sentiment scores and evaluate user satisfaction with products;

[0297] Step 2.4: Integrate demand prediction, interest distribution, and sentiment scores to generate a comprehensive user profile. At the same time, design a dynamic update mechanism based on time decay. Whenever new data enters, the profile will be updated according to the time weight;

[0298] The dynamic knowledge graph construction module 103 is used to construct a dynamically updated dynamic knowledge graph by combining the medical device industry with the dynamically updated user profile, so that the market changes are kept up-to-date, specifically including:

[0299] Step 3.1: Collect market information. The market information includes the characteristic data P = {p1, p2,..., p n} of medical device products, the policy and market information G = {g1, g2,..., g m}, and the supply chain partner information S = {s1, s2,..., s k}. Construct a dynamic knowledge graph according to the market information and the comprehensive user profile;

[0300] Step 3.2: Use a graph neural network model to learn the complex relationships between the nodes of the dynamic knowledge graph, update the node embeddings, and capture the structural information in the graph;

[0301] Step 3.3: Design a dynamic update mechanism based on time decay. Whenever new data enters, the dynamic knowledge graph will be automatically updated according to the new user behavior and external environment;

[0302] The dynamic knowledge graph analysis module 104 is used to conduct market analysis using the knowledge graph and provide product demand prediction and user recommendations, specifically including:

[0303] Based on the embedded representation of user nodes, combined with the information of policy nodes and product nodes, predict the future needs of users and market trends, recommend the most suitable products and the most appropriate supply chain partners for users according to the prediction results, and generate the prediction of the future needs of users and market trends, the most suitable products and the most appropriate supply chain partners into comprehensive market strategy recommendations;

[0304] The real-time optimization module 105 is used to provide a personalized marketing plan and optimize it in real time according to the comprehensive market strategy recommendations.

[0305] In summary, the present invention proposes an Internet big data analysis method and system based on artificial intelligence. Through the fusion processing of multi-dimensional data and the construction of an industry knowledge graph, it realizes the comprehensive capture and dynamic update of user needs, and provides intelligent market decision support based on the knowledge graph, overcoming the limitations of the prior art.

[0306] The above-disclosed are only some preferred embodiments of the present invention. Of course, the scope of the rights of the present invention cannot be limited thereby. Those of ordinary skill in the art can understand all or part of the processes of implementing the above embodiments, and the equivalent changes made according to the claims of the present invention still fall within the scope covered by the invention.

Claims

1. An Internet big data analysis method based on artificial intelligence, characterized in that: The method comprises: Step 1: Collect heterogeneous data from different data sources in the medical device industry, standardize them, extract key features from the standardized data, and fuse them into a unified comprehensive feature vector ; Step 2: Based on the comprehensive feature vector Generate dynamically updated user profiles, including: Step 2.1: Based on the comprehensive feature vector Use a linear layer plus an activation function to predict demand, and add a time-varying weight term after the linear layer. Adjust the weight of demand forecast according to the change of time, so that the weight of old data is the smallest and the weight of new data is the largest; Step 2.2: Based on the comprehensive feature vector Establish interest preference analysis model and design policy impact coefficient Combine and calculate the user's interest distribution in different categories of products; Step 2.3: Based on the comprehensive feature vector Design sentiment analysis network based on user feedback, generate sentiment scores, and evaluate user satisfaction with the product; Step 2.4: Integrate demand prediction, interest distribution, and sentiment score to generate a comprehensive user portrait. At the same time, design a dynamic update mechanism based on time decay. Whenever new data enters, the portrait will be updated according to the time weight. Step 3: Combine the medical device industry with dynamically updated user portraits to build a dynamically updated knowledge graph to keep the market changes up to date, including: Step 3.1: Collect market information, including characteristic data of medical device products , policy and market information , and supply chain partner information , build a dynamic knowledge graph based on market information and comprehensive user portraits; Step 3.2: Use the graph neural network model to learn the complex relationships between nodes in the dynamic knowledge graph, embed and update the nodes, and capture the structural information in the graph; Step 3.3: Design a dynamic update mechanism based on time decay. Whenever new data comes in, the dynamic knowledge graph will automatically update according to the new user behavior and external environment. Step 4: Use the knowledge graph to conduct market analysis and provide product demand forecasts and user recommendations, including: Through the embedded representation of user nodes, combined with the information of policy nodes and product nodes, the user's future needs and market trends are predicted, and the most suitable products and supply chain partners are recommended to the user based on the predicted results. The user's future needs and market trends are predicted, the most suitable products and the most suitable supply chain partners are generated into comprehensive market strategy recommendations; Step 5: Provide personalized marketing plans based on comprehensive market strategy recommendations and optimize in real time.

2. The method for analyzing Internet big data based on artificial intelligence according to claim 1, characterized in that: The heterogeneous data includes structured data, unstructured data and external data; the standardization process specifically includes the following steps: For structured data, we use a custom interpolation algorithm to fill missing values ​​and detect outliers based on historical trends in time series and sales data of similar products. For unstructured data, we use NLP technology to clean text, and then use a custom variant of the pre-trained language model BERT to embed text data into the vector space. For external data, we use NLP technology to extract features and combine them with structured data to obtain the combined structured data. Then, the combined structured data and unstructured data are subjected to feature extraction and mapped to a unified feature space to obtain a unified feature vector representation. , which is expressed as follows: ； in, Represents vectors in structured data, Represents vectors in unstructured data, Represents the mapping matrix of the processed structured data, A mapping matrix representing unstructured data, and ,in and is the weight matrix, projecting structured data and unstructured data into the same dimension In the feature space, , is the fused feature vector.

3. The method for analyzing Internet big data based on artificial intelligence according to claim 2, characterized in that: The key features of the standardized data are extracted and fused into a unified fusion feature vector, specifically including: For structured data , according to its specific numerical value and time series characteristics, a linear transformation with a regular term is used for mapping, which is expressed as follows: ; in, A mapping vector representing structured data, , is the mapping matrix, The structured data of the dimension is mapped to Dimension unified space, is the regular term coefficient, which is used to control the complexity of the model and prevent overfitting; For unstructured data , using the mapping matrix Project it onto dimensional space, expressed as follows: ; in, A mapping vector representing unstructured data, It is a noise suppression term, which is calculated through the attention mechanism or context analysis model and filters irrelevant information in the feedback; , is the mapping matrix of unstructured data, which maps the unstructured text embedding to dimensional unified feature space; For external data , a hysteresis effect control term is designed , the hysteresis effect control term Related to the time of policy release, reflecting the gradual impact of the policy on the market, as shown below: ; in, Represents updated external data, , is the projection matrix; Mapping structured data to vector , mapping vectors for unstructured data and updated external data Fusion into a comprehensive feature vector , which is expressed as follows: ; in, is the adaptive weight of each modal feature, satisfying ; According to the comprehensive feature vector Design a dynamic update mechanism to ensure that the feature vector can be adaptively updated over time, as shown below: ; in, is the time attenuation coefficient, is the current feature vector, The feature vector generated for the most recent data; At the same time, in the dynamic update process, add regular terms To control the smoothness of the eigenvector, it is expressed as follows: ; in, is the regularization coefficient, is the Euclidean distance between feature vectors, which is used to control the amplitude of feature vector update to ensure smooth transition.

4. The method for analyzing Internet big data based on artificial intelligence according to claim 1, characterized in that: In step 2, the demand forecast is based on a dynamic demand forecast model, and the dynamic demand forecast model is constructed as follows: Multidimensional feature vector based on input , use a linear layer plus activation function to perform demand forecasting, the formula is as follows: ； in, , is the mapping matrix, The dimension feature vector is mapped to the demand forecast dimension, is the bias term, is the activation function, It indicates the predicted value of the user's future demand, reflecting the possibility of the user purchasing the product in the future; Add a regularization term after the linear layer Balancing the impact of historical data and new data is expressed as follows: ； in, represents the regularization coefficient, represents the mapping matrix, represents the comprehensive feature vector; The interest preference analysis model is constructed as follows: A softmax function is used to map users’ preferences for different product categories into probability distributions, as shown below: ； in, represents the interest weight, , is the weight matrix, which transforms the multimodal features Map to Interest distribution of product categories, is the bias term; The sentiment analysis network is constructed as follows: Using a tanh activation function, the output sentiment score range is between [-1, 1]. The specific formula is as follows: ; in, represents the sentiment weight, , is the weight matrix, is the bias term, is the sentiment score, which indicates the user's satisfaction with the product; Indicates that users' feedback on the product is positive. Indicates negative feedback; The user portrait , which is expressed as follows: ; in, represents the user demand forecast, is the interest preference distribution, It is the result of sentiment analysis.

5. The method for analyzing Internet big data based on artificial intelligence according to claim 4, characterized in that: In the dynamic demand forecasting model, a time-varying weight term is also introduced , the weight of demand forecast is adjusted according to the change of time, which is expressed as follows: ; in, is the time decay function, represents the updated demand forecast, represents the old demand forecast; In the interest preference analysis model, a time decay regularization term is also introduced , balancing historical interest and future potential interest, expressed as follows: ; in, For the latest interest distribution, is the historical interest distribution; at the same time, the policy influence coefficient is also introduced into the interest preference analysis model , which is used to enhance the weighted impact of policy changes on interest preferences, is expressed as follows: ; in, reflects the impact of policy changes on interest in different product categories; In the sentiment analysis network, the sentiment amplification factor is also introduced , amplifying the emotional fluctuations, expressed as follows: ; in, is the amplification factor. When user feedback contains extreme emotions, The sentiment score will be automatically amplified to ensure that the user's emotional fluctuations can be fully captured by the system.

6. The method for analyzing Internet big data based on artificial intelligence according to claim 1, characterized in that: The structure of the dynamic knowledge graph is as follows: Nodes include: User nodes , Product Node , Policy Node and supply chain nodes ; The user node Each user portrait As the characteristic information of user nodes, including demand forecast , interest preferences and emotional feedback ; Through user portraits As feature input, it is expressed as ,in is the feature embedding function; The user node In , each product node represents a medical device product, expressed as , represents the embedding of product features ; The policy node is the industry policy and market information node, expressed as , represents the embedding of policy features ; The supply chain node Nodes representing supply chain participants are represented by , represents the embedding of supply chain participants ; Relationships include: User-Product Relationship , User-Policy Relationship and user-supply chain relationships ; The user-product relationship The relationship between users and the products they are interested in or have purchased. The edge weight is determined by the user's interest distribution and purchase history. The user-policy relationship Indicates the impact of policies on user needs; The user-supply chain relationship Represents the interactive relationship between users and supply chain partners, including order delivery history, feedback, and cooperation intensity; A dynamic weight mechanism is introduced to construct the adjacency matrix to embed nodes and relationships, which is expressed as follows: The user-product adjacency matrix , which is expressed as follows: ; in, Indicates the similarity between user interests and product features, Represents the historical purchase records of users and products. and is the balance parameter; User-Policy Adjacency Matrix , which is expressed as follows: ; in, represents the potential impact of the policy on user demand, Indicates the relevance of policies to user needs, and Used to adjust the weights of different factors; User-supply chain adjacency matrix , which is expressed as follows: ; in, represents the order frequency between users and suppliers, Represents the supply chain.

7. The method for analyzing Internet big data based on artificial intelligence according to claim 6, characterized in that: The graph neural network model is used to learn the complex relationships between nodes in the dynamic knowledge graph, embed and update the nodes, and capture the structural information in the graph, including: In each layer of the graph neural network, the features of the nodes are propagated through the adjacency matrix and interact with neighboring nodes. The features of each node are updated using the following formula: ; in, Indicates Layer Node Features, Is a node The set of neighbor nodes of is the node in the adjacency matrix and The relationship weight between It is The weight matrix of the layer, is the bias term, is a nonlinear activation function; At the same time, the relationship regularization term is introduced to control the complexity of the relationship between nodes, which is expressed as follows: ; in, is the adjacency matrix under different relationship types, is the regularization coefficient, is the Frobenius norm; The dynamic update mechanism based on time decay is expressed as follows: ； in, The current time node Features, Update features for new data, is the time decay coefficient, which controls the weight distribution of historical features and new features.

8. The method for analyzing Internet big data based on artificial intelligence according to claim 1, characterized in that: The embedding representation of the user node is combined with the information of the policy node and the product node to predict the user's future needs and market trends, which is performed as follows: Based on the message passing mechanism of graph neural network, the demand forecast of user nodes is calculated by the following formula: ; in, It is the predicted result of user demand; is a weight matrix that converts user embeddings into demand predictions; is the policy impact weight, is the potential impact of policies on user needs, for policy embedding; is the product impact weight, Embed for products; The method of recommending the most suitable product and the most suitable supply chain partner to the user based on the prediction results is performed as follows: Embedded by user node and product node embedding The similarity calculation and the historical purchase relationship between the user and the product are used to generate the recommendation score, which is expressed as follows: ; in, is the score of the product recommendation; is the weight matrix of product recommendation; It is the inner product of user embedding and product embedding, which is used to calculate similarity; The impact of historical purchase records, including is the historical impact coefficient; Supply chain partner recommendation is calculated by the embedding similarity of user nodes and supply chain nodes, as well as order frequency and logistics efficiency, as shown below: ; in, is the supply chain partner recommendation score; is the weight matrix of supply chain recommendations; is the inner product of the embedding of the user and the supply chain node, which is used to calculate the similarity; Indicates the order frequency between users and supply chain partners, is the order frequency weight; The impact of logistics efficiency on partner recommendations, is the logistics efficiency weight.

9. The method for analyzing Internet big data based on artificial intelligence according to claim 1, characterized in that: The step 5 specifically includes: Step 5.1: Based on demand forecast and product recommendation results, select appropriate marketing channels for different user groups and allocate marketing resources reasonably; Step 5.2: Based on user portraits and market feedback data, an adaptive user clustering model is constructed to cluster users, and marketing resources and user clustering results are combined to generate personalized marketing content; wherein, the adaptive user clustering model embeds user vectors Perform clustering to generate multiple user groups; Step 5.3: Design a real-time feedback mechanism to adjust the current marketing strategy and channel resource allocation by comparing the actual feedback data of personalized marketing content with the expected results. At the same time, design a step control mechanism to ensure that the strategy optimization process can converge and avoid strategy fluctuations caused by excessive adjustments.

10. The Internet big data analysis system based on artificial intelligence is characterized by: The system comprises: Heterogeneous data collection module is used to collect heterogeneous data from different data sources in the medical device industry, standardize it, extract key features from the standardized data, and fuse them into a unified comprehensive feature vector ; The portrait construction module is used to Generate dynamically updated user profiles, including: Step 2.1: Based on the comprehensive feature vector Use a linear layer plus an activation function to predict demand, and add a time-varying weight term after the linear layer. Adjust the weight of demand forecast according to the change of time, so that the weight of old data is the smallest and the weight of new data is the largest; Step 2.2: Based on the comprehensive feature vector Establish interest preference analysis model and design policy impact coefficient Combine and calculate the user's interest distribution in different categories of products; Step 2.3: Based on the comprehensive feature vector Design sentiment analysis network based on user feedback, generate sentiment scores, and evaluate user satisfaction with the product; Step 2.4: Integrate demand prediction, interest distribution, and sentiment score to generate a comprehensive user portrait. At the same time, design a dynamic update mechanism based on time decay. Whenever new data enters, the portrait will be updated according to the time weight. The dynamic knowledge graph construction module is used to build a dynamically updated dynamic knowledge graph based on the medical device industry and dynamically updated user portraits, so that market changes are kept up to date. Specifically, it includes: Step 3.1: Collect market information, including characteristic data of medical device products , policy and market information , and supply chain partner information , build a dynamic knowledge graph based on market information and comprehensive user portraits; Step 3.2: Use the graph neural network model to learn the complex relationships between nodes in the dynamic knowledge graph, embed and update the nodes, and capture the structural information in the graph; Step 3.3: Design a dynamic update mechanism based on time decay. Whenever new data comes in, the dynamic knowledge graph will automatically update according to the new user behavior and external environment. The dynamic knowledge graph analysis module is used to use the knowledge graph to conduct market analysis, provide product demand forecasts and user recommendations, including: Through the embedded representation of user nodes, combined with the information of policy nodes and product nodes, the user's future needs and market trends are predicted, and the most suitable products and supply chain partners are recommended to the user based on the predicted results. The user's future needs and market trends are predicted, the most suitable products and the most suitable supply chain partners are generated into comprehensive market strategy recommendations; The real-time optimization module is used to provide personalized marketing plans and optimize in real time based on comprehensive market strategy recommendations.

Images