A marketing management system based on AI algorithm

Abstract

The present invention relates to the field of marketing management technology, and in particular to a marketing management system based on an AI algorithm. It includes a commodity monitoring module, a sales analysis module, an inventory management module, a sales strategy optimization module, and a feedback adjustment module. The commodity monitoring module collects sales data through image recognition and sensors. The sales analysis module uses a deep learning algorithm to analyze data and predict sales trends. The inventory management module uses a stochastic conjugate gradient descent algorithm to optimize inventory scheduling based on the analysis report. The sales strategy optimization module generates pricing, promotion, and recommendation strategies through an AI algorithm. The feedback adjustment module extracts key information based on a sparse representation algorithm and adjusts inventory and sales strategies. The present invention can realize intelligent and automated marketing management, improve the operating efficiency of the enterprise, reduce labor costs, and enhance market competitiveness.

Description

Technical Field

[0001] The present invention relates to the technical field of marketing management, and in particular to a marketing management system based on AI algorithm. Background Art

[0002] With the rapid development of the Internet, e-commerce and online marketing have gradually become the main ways for companies to promote their products and services. However, with the intensification of market competition, companies need more precise marketing strategies to increase their market share and customer loyalty. At present, there are still the following problems: the existing technology relies on manual entry or traditional sensor collection, and there are problems such as untimely data updates and low accuracy, making it difficult to obtain accurate commodity sales data in a timely manner; the existing technology is usually based on simple historical data analysis, lacking the ability to deeply mine complex market trends and multi-dimensional data, making it difficult to provide accurate sales trend forecasts and unable to provide effective support for marketing strategies; the existing system is usually based on simple historical data analysis, lacking the ability to deeply mine complex market trends and multi-dimensional data, making it difficult to provide accurate sales trend forecasts and unable to provide effective support for marketing strategies. Summary of the invention

[0003] To solve the above problems, the present invention provides a marketing management system based on AI algorithm, which solves the problems of inaccurate sales data collection, incomplete sales analysis, lack of intelligence in inventory management, lagging sales strategy optimization and untimely feedback processing response, and effectively improves the intelligence level of marketing management and the market response speed.

[0004] To achieve the above object, the technical solution adopted by the present invention is:

[0005] A marketing management system based on AI algorithm, comprising a commodity monitoring module, a sales analysis module, an inventory management module, a sales strategy optimization module and a feedback adjustment module connected in sequence;

[0006] The commodity monitoring module is used to collect sales data of commodities based on image recognition and sensor network;

[0007] The sales analysis module is used to comprehensively analyze the sales data obtained by the commodity monitoring module through a deep learning algorithm, predict the best sales trend and regional distribution of commodities, and generate a sales analysis report;

[0008] The inventory management module is used to optimize inventory scheduling based on the sales analysis report using a stochastic conjugate gradient descent algorithm to achieve intelligent inventory replenishment, commodity allocation, and supply chain automation management;

[0009] The sales strategy optimization module is used to generate an optimal sales strategy using an AI algorithm based on the sales analysis report and inventory level; the sales strategy includes product pricing adjustment, promotion plan and product recommendation;

[0010] The feedback adjustment module is used to extract key information from feedback data using a sparse representation algorithm to quickly adjust inventory management and sales strategies.

[0011] Furthermore, the sales data includes product sales, customer purchase data, inventory levels and product display status.

[0012] Furthermore, the operation process of the sales analysis module includes the following steps:

[0013] Receive sales data collected by the commodity monitoring module and perform pre-processing operations;

[0014] Based on the sales data, a graph neural network algorithm with an attention mechanism is used to construct an interaction graph of products, customers, and supply chains, capture the complex dependencies between nodes, and predict product sales trends;

[0015] Based on product sales trends, use the Geographically Weighted Regression algorithm to analyze regional sales data and identify high demand trends in different regions;

[0016] Based on the above analysis results, a sales analysis report is generated; the sales analysis report includes sales trend forecasts, customer behavior analysis and regional recommendations.

[0017] Furthermore, the formula of the geographically weighted regression algorithm is as follows:

[0018]

[0019] Among them, y i represents the sales volume of goods in the i-th geographical area; x ia represents the ath influencing factor of the i-th geographical region; β0(u i , v i ) represents the intercept term of the i-th geographical area; β a (u i , v i ) represents the regression coefficient of the ath influencing factor in the ith geographical region; that is, the degree of influence of this factor on the sales volume of goods in this region; (u i , v i ) represents the longitude and latitude coordinates of the i-th geographical area; ∈ i represents the random error of the ith region; p represents the total number of variables that affect the sales volume of the product.

[0020] Furthermore, the operation process of the inventory management module includes the following steps:

[0021] Based on the sales trend forecast in the sales analysis report, combined with historical sales data and real-time inventory levels, a recurrent neural network is used to predict future inventory demand;

[0022] According to the demand forecast results, the stochastic conjugate gradient descent algorithm is used to calculate the replenishment quantity, optimize the inventory replenishment frequency and quantity, and combine the supply chain response speed to achieve intelligent allocation and inventory scheduling;

[0023] Automatically trigger supply chain operations based on inventory scheduling results, and execute replenishment requests or commodity transfer operations in real time through automated interfaces with suppliers and warehouses;

[0024] Receive key information returned by the feedback adjustment module and dynamically adjust the inventory management strategy.

[0025] Furthermore, the formula of the stochastic conjugate gradient descent algorithm is as follows:

[0026]

[0027] Among them, Q k+1 Indicates the updated inventory replenishment or transfer quantity after iteration; Q k Indicates the replenishment quantity, transfer quantity or supply chain operation quantity under the current inventory level; α k It represents the step size coefficient of the kth iteration, which controls the magnitude of inventory adjustment in the current iteration; represents the gradient of the objective function at the kth iteration; d k-1 represents the search direction of the k-1th iteration, which is used to determine the current inventory optimization direction; f(Q k ) represents the inventory management objective; Represents the conjugate coefficient, which determines the relationship between the current search direction and the direction of the previous iteration.

[0028] Furthermore, the operation process of the sales strategy optimization module includes the following steps:

[0029] Based on the sales trend forecast in the sales analysis report, use the reinforcement learning algorithm to analyze the effects of different sales strategies and establish a sales strategy optimization model;

[0030] According to the sales strategy optimization model, combined with the collaborative filtering algorithm, product recommendations are made, the recommendation effects of different product combinations are analyzed, the recommendation strategy is dynamically adjusted, and the optimal sales strategy solution is generated;

[0031] Generate a personalized product recommendation list based on customer behavior data through rule-based algorithms;

[0032] Continuously optimize sales strategies based on key information provided by the feedback adjustment module.

[0033] Furthermore, the process of constructing the sales strategy optimization model includes the following steps:

[0034] Setting initial strategies for product pricing, promotional plans, and product recommendations, and dynamically adjusting the initial strategies based on the sales data and inventory levels;

[0035] The gradient boosted decision tree algorithm is used to optimize the impact of different pricing and promotional schemes on sales volume and optimize product sales strategies.

[0036] Furthermore, the sparse representation algorithm combines sparse coding and decoding mechanisms to perform deep compression processing on the feedback data while retaining the core information of the feedback data, and realizes effective transmission and storage of the feedback data by decoding and reconstructing the compressed data.

[0037] The beneficial effects of the present invention are as follows: the commodity monitoring module of the present invention uses image recognition and sensor networks to automatically collect sales data of commodities, realizes real-time and accurate data collection, and reduces the cost and error of manual monitoring. The sales analysis module comprehensively analyzes sales data through deep learning algorithms, can predict the best sales trend and regional distribution of commodities, provides scientific sales analysis reports, and provides a strong basis for subsequent strategy optimization. The inventory management module optimizes inventory scheduling through the random conjugate gradient descent algorithm, realizes intelligent inventory replenishment, commodity allocation and supply chain automation management, avoids excess or shortage of inventory, and improves the efficiency and flexibility of inventory management. Based on the sales analysis report and inventory level, the sales strategy optimization module uses AI algorithms to generate the best sales strategy, including commodity pricing adjustment, promotion plan and personalized commodity recommendation, thereby improving the accuracy and conversion rate of sales. The feedback adjustment module uses sparse representation algorithms to extract key information from feedback data, can quickly adjust inventory management and sales strategies, and forms a data-driven closed-loop management, so that the system has higher flexibility and response speed, and continuously optimizes marketing effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a module schematic diagram of a marketing management system based on AI algorithm of the present invention.

[0039] Figure 2 It is a flowchart of the operation process of the sales analysis module provided by an embodiment of the present invention.

[0040] Figure 3 It is a flowchart of the operation process of the inventory management module provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0041] See also Figure 1-Figure 3 As shown, the present invention relates to a marketing management system based on AI algorithm.

[0042] Example

[0043] A marketing management system based on AI algorithm, comprising a commodity monitoring module, a sales analysis module, an inventory management module, a sales strategy optimization module and a feedback adjustment module connected in sequence;

[0044] The commodity monitoring module is used to collect commodity sales data based on image recognition and sensor networks; the sales data includes commodity sales, customer purchase data, inventory levels and commodity display status;

[0045] It should be noted that the product monitoring module is responsible for comprehensively and accurately capturing the various states of the product in the sales process through sensor networks, image recognition technology and data transmission networks. It requires real-time collection of multi-dimensional data on the product, including sales quantity, customer behavior, inventory changes, product display conditions, etc.

[0046] The hardware components of the commodity monitoring module include:

[0047] High-definition cameras: used to capture the display status of goods and monitor the status of goods on shelves, booths, and even self-service checkout counters.

[0048] Sensor network: Multiple sensors deployed on shelves, warehouses or display areas, such as weight sensors, infrared sensors, RFID (radio frequency identification) tags, etc., to collect data on the physical changes of goods.

[0049] Image recognition technology identifies different types of goods and sales scenarios. The details are as follows:

[0050] Product recognition: The camera captures the image of each product on the shelf in real time and identifies the specific product type, specification and brand. The applications of image recognition technology include the following:

[0051] Identify the number of products: By continuously collecting image streams, the system can automatically detect changes in the number of products on the shelf. For example, after a specific product is taken away by a customer, the system can know the reduction of the product on the shelf in real time through image comparison.

[0052] Product placement status analysis: The system can detect whether the products are placed correctly, whether they are placed randomly, upside down or blocked. By analyzing these situations, merchants can optimize product display in a timely manner to ensure better display effects.

[0053] Display optimization monitoring: The system can monitor the visual appeal of product placement and, combined with historical sales data, evaluate the impact of different product display methods on sales. Through feedback, the display strategy is optimized to increase sales.

[0054] Shopping route analysis: By using multiple cameras to track the movement paths of customers in the store, the system can understand the main shopping routes of customers and help optimize the placement of goods and advertising areas.

[0055] The commodity monitoring module needs to rely on high-speed data processing and transmission networks. The specific implementation includes:

[0056] Edge computing: Since cameras and sensors generate a large amount of data, the commodity monitoring module can use edge computing devices to perform preliminary processing locally to reduce transmission delays and data volume. Edge devices can pre-process images locally, such as detecting changes and extracting key features, and only necessary information will be uploaded to the cloud or central server.

[0057] Wireless sensor network (WSN): Sensors form a network through wireless communication technologies (such as ZigBee, LoRa, Wi-Fi, etc.) and transmit sensor data to the central server in real time. Sensor data includes the specific status of each product on the shelf, whether a customer has taken the product, and the specific timestamp.

[0058] Real-time data transmission protocol: Use low-latency transmission protocols (such as MQTT or CoAP) to ensure that all types of data collected from cameras and sensors can be quickly transmitted to the backend system for analysis and processing. This data transmission method is suitable for scenarios that require frequent updates and real-time responses.

[0059] The product monitoring module is not only responsible for data collection, but also provides intuitive visualization tools to help merchants monitor product sales in real time. Specific implementation includes:

[0060] Real-time monitoring interface: The system displays real-time product sales data on the front-end interface, including sales quantity, inventory status, customer interaction, etc. Merchants can check sales dynamics at any time through the monitoring screen or mobile device application.

[0061] Data alarm: When there is insufficient inventory or abnormal sales of goods, the system will automatically issue an alarm to remind managers to take corresponding measures, such as replenishing stocks, re-displaying goods or analyzing the causes of abnormal situations.

[0062] The sales analysis module is used to comprehensively analyze the sales data obtained by the commodity monitoring module through a deep learning algorithm, predict the best sales trend and regional distribution of commodities, and generate a sales analysis report;

[0063] The operation process of the sales analysis module includes the following steps:

[0064] Receive sales data collected by the commodity monitoring module and perform pre-processing operations;

[0065] Based on the sales data, a graph neural network algorithm with an attention mechanism is used to construct an interaction graph of products, customers, and supply chains, capture the complex dependencies between nodes, and predict product sales trends;

[0066] Specifically, the system uses graph neural network (GNN) technology to build an interaction graph reflecting the relationship between these entities based on the sales data of goods, customers, and supply chains.

[0067] Graph node definition: Each product, customer, and supply chain node is represented as a node in the graph. The attributes of the node include product sales data (such as sales volume, inventory), customer behavior data (such as purchase history, preferences), and supply chain data (such as inventory changes, shipping time).

[0068] Graph edge definition: The edges of a graph represent the interaction between different nodes.

[0069] For example, the edge between a product and a customer represents the customer's purchasing behavior, and the weight of the edge can be determined by the frequency and quantity of the customer's purchases. The edge between a product and a supply chain represents the impact of the supply chain on the product, such as the frequency of inventory replenishment and the time of transportation.

[0070] The interaction graph captures the complex dependencies between products, customers, and supply chains. Through this graph structure, the system can understand which customer groups prefer which products, which products have a high degree of sales dependence on the supply chain, and the mutual influence between the sales of different products (such as the relationship between complementary products or substitutes).

[0071] After constructing the interaction graph, the sales analysis module uses a graph neural network (GNN) algorithm based on the attention mechanism to more accurately capture the complex dependencies between nodes and predict the sales trends of products.

[0072] GNN propagates information between nodes through an iterative message passing mechanism. For example, a product node receives information from adjacent customer nodes and supply chain nodes to update its feature representation. Through multiple iterations, the features of the node will reflect the global information of the entire graph. This approach enables the system to understand the mutual influence of products, customers, and supply chains in the overall network, and how these influences drive changes in sales trends. On the basis of traditional GNN, an attention mechanism is introduced, which can dynamically assign different weights to each node to determine which neighboring nodes are most important to the prediction target. For example, when predicting the sales of a certain product, the attention mechanism can identify which customer nodes and supply chain nodes have a greater impact on the sales of this product. Through the attention mechanism, the model can flexibly capture the complex relationships between different nodes, especially when dealing with large-scale interaction graphs, it can highlight key dependencies and avoid redundant information propagation. After updating the features of each product node through the GNN model, the system will predict the future sales trend of the product. The model can identify peaks and troughs in the sales cycle and predict the sales trend of a certain product in the future. This prediction result will be used in key decisions such as inventory management and promotion strategy formulation.

[0073] Based on product sales trends, use the Geographically Weighted Regression algorithm to analyze regional sales data and identify high demand trends in different regions;

[0074] Specifically, in order to more accurately analyze the sales situation in different regions, the system uses the Geographically Weighted Regression (GWR) algorithm to identify high demand trends in different regions by performing weighted regression analysis on sales data in different regions. The specific steps include:

[0075] The sales data of various sales areas (such as different cities, stores, etc.) are obtained from the product monitoring module, and combined with geographic information (such as the economic level of the region, population density, consumer behavior characteristics, etc.).

[0076] The GWR algorithm builds a regional weighted regression model based on the geographic location and sales data of each region. The model assigns different weights to different regions to capture the differences between regions. For example, some commodities may have higher sales demand in high-density cities, but relatively less in rural areas.

[0077] By analyzing the results of the regression model, the system is able to identify high-demand trends in various regions and determine where inventory or promotion efforts should be increased, thereby improving sales efficiency.

[0078] Based on the above analysis results, a sales analysis report is generated; the sales analysis report includes sales trend forecasts, customer behavior analysis and regional recommendations.

[0079] Furthermore, the formula of the geographically weighted regression algorithm is as follows:

[0080]

[0081] Among them, y i represents the sales volume of goods in the i-th geographical area; x ia represents the ath influencing factor in the i-th geographical area, and the influencing factors include price factors, promotion factors, commodity inventory level and commodity display status; β0(u i , v i ) represents the intercept term of the ith geographic region, and its value varies according to the geographical location of the region (u i , v i ) varies with the region, reflecting the sales level of basic commodities in the region and is not affected by other factors; a (u i , v i ) represents the regression coefficient of an influencing factor a in the ith geographical region; that is, the degree of influence of the factor on the sales volume of goods in the region; (u i , v i ) represents the longitude and latitude coordinates of the ith geographic region, which is used to calculate the location weight in the geographically weighted regression model; ∈ i represents the random error of the ith region; p represents the total number of variables that affect the sales volume of the product.

[0082] The inventory management module is used to optimize inventory scheduling based on the sales analysis report using a stochastic conjugate gradient descent algorithm to achieve intelligent inventory replenishment, commodity allocation, and supply chain automation management;

[0083] The operation process of the inventory management module includes the following steps:

[0084] Based on the sales trend forecast in the sales analysis report, combined with historical sales data and real-time inventory levels, a recurrent neural network is used to predict future inventory demand;

[0085] Specifically, the system uses a recurrent neural network model to effectively capture the time dependency and historical patterns of product sales.

[0086] The input data of the recurrent neural network model includes historical sales data, trend forecasts from sales analysis reports, and real-time inventory status. The model is trained with these multi-dimensional time series data. The input data is processed sequentially through time steps, and its hidden layer state is gradually updated, so that it can capture long-term and short-term sales patterns from historical data. The output of the model is a forecast of inventory demand in the future several time steps, that is, the amount of inventory the system expects to be required for each commodity in a certain period of time in the future. The forecast results can also be adjusted during the training process, especially in combination with seasonal forecasts in sales analysis reports (such as sales surges during holidays or promotions) to improve the accuracy of the forecast.

[0087] According to the demand forecast results, the stochastic conjugate gradient descent algorithm is used to calculate the replenishment quantity, optimize the inventory replenishment frequency and quantity, and combine the supply chain response speed to achieve intelligent allocation and inventory scheduling;

[0088] Specifically, the system uses the stochastic conjugate gradient descent (SCGD) algorithm, which finds the optimal solution through multiple iterations, optimizes the replenishment frequency and replenishment quantity, and reduces inventory costs.

[0089] The optimization goal of inventory scheduling is to find a balance point so that the inventory of goods can meet the predicted demand while avoiding overstocking or out-of-stock. The optimization problem can be expressed as:

[0090] Minimize inventory costs: including inventory holding costs, losses due to expired goods, sales losses due to out-of-stock, etc.

[0091] Minimize replenishment frequency: Reduce transportation costs through appropriate replenishment frequency and avoid frequent small-batch replenishment.

[0092] Maximize service levels: Ensure that customers have adequate inventory supply during peak demand to avoid sales losses due to stockouts.

[0093] The SCGD algorithm process includes: starting from the initial replenishment quantity and replenishment time, these parameters are gradually optimized through iteration. The initial replenishment quantity can be set based on the historical average demand, and the replenishment time can be estimated based on the supply chain response speed. In each iteration, the SCGD algorithm calculates the impact of the current inventory scheduling plan on the cost function, and updates the inventory replenishment plan through the conjugate direction, gradually approaching the optimal solution. Each update combines the future inventory demand predicted by the RNN, the current inventory status, and the supply chain response time. When the iterative results of the SC GD algorithm no longer change significantly within a certain range, the algorithm is considered to have converged and outputs the optimal inventory replenishment plan.

[0094] Automatically trigger supply chain operations based on inventory scheduling results, and execute replenishment requests or commodity transfer operations in real time through automated interfaces with suppliers and warehouses;

[0095] It should be noted that once the replenishment quantity and time output by the SCGD algorithm are determined, the system will issue a replenishment request in real time through the automated interface with suppliers and warehouses. For example, the system sends a replenishment order to the supplier, indicating the type, quantity and delivery time of the goods to be replenished. It issues a delivery instruction to the warehouse to deliver the goods to the designated store or warehouse.

[0096] If the system decides to transfer inventory, it will automatically trigger the corresponding logistics operation. The system will send a transfer request to the third-party or internal warehouse management system responsible for logistics, including the type and quantity of goods to be transferred, the starting warehouse, the target warehouse or store, and the optimal transportation route.

[0097] Supply chain response speed is an important factor in determining replenishment and allocation plans. The system needs to dynamically adjust inventory management strategies based on the supplier's production and delivery capabilities, logistics transportation efficiency, etc. For example, if a supplier has a long delivery cycle, the system will issue a replenishment request in advance to avoid out-of-stock situations.

[0098] Receive key information returned by the feedback adjustment module and dynamically adjust the inventory management strategy.

[0099] Furthermore, the formula of the stochastic conjugate gradient descent algorithm is as follows:

[0100]

[0101] Among them, Q k+1 Indicates the updated inventory replenishment or transfer quantity after iteration; Q k Indicates the replenishment quantity, transfer quantity or supply chain operation quantity under the current inventory level; α k The step coefficient of the kth iteration controls the magnitude of inventory adjustment in the current iteration. In a specific scheme, it determines the scale and speed of each replenishment or transfer. For example, when the step size is large, the system may replenish inventory quickly; when the step size is small, the system will adjust the inventory more cautiously. represents the gradient of the objective function at the kth iteration; d k-1 represents the search direction of the k-1th iteration, which is used to determine the current inventory optimization direction, that is, the inventory replenishment or transfer direction selected in the k-1th iteration; f(Q k ) represents the inventory management goal, which can be to minimize inventory costs, improve supply chain responsiveness, or maximize inventory replenishment efficiency; Represents the conjugate coefficient, which determines the relationship between the current search direction and the direction of the previous iteration.

[0102] The sales strategy optimization module is used to generate an optimal sales strategy using an AI algorithm based on the sales analysis report and inventory level; the sales strategy includes product pricing adjustment, promotion plan and product recommendation;

[0103] The operation process of the sales strategy optimization module includes the following steps:

[0104] Based on the sales trend forecast in the sales analysis report, use the reinforcement learning algorithm to analyze the effects of different sales strategies and establish a sales strategy optimization model;

[0105] In one embodiment, the specific steps of the reinforcement learning algorithm are as follows:

[0106] Definition of state and action: The state refers to the current sales environment, including the current sales situation of the product, inventory level, market demand, customer preferences, etc. The system updates these states in real time through data in the sales analysis report. Actions are different options for sales strategies, such as adjusting prices, launching promotions, modifying recommended products, etc. The reinforcement learning algorithm evaluates the impact of each action on sales results.

[0107] Reward function design: The reward function defines the effect evaluation criteria for each strategy action. In this embodiment, the reward function is usually associated with product sales, profit, customer satisfaction, and inventory consumption. For example, increasing sales can get positive rewards. Reducing inventory pressure will get positive rewards, while inventory backlogs will bring negative rewards. If the sales strategy improves the customer's purchasing experience, such as successful personalized recommendations or reasonable pricing, rewards will also be obtained.

[0108] The Q-learning algorithm in reinforcement learning is used to evaluate the long-term effects of each action under different states. The system continuously explores and records the reward values ​​of taking different sales strategies under different states through interaction with the environment, and finally forms an optimal sales strategy. In a more complex sales environment, the deep Q-learning (DQN) algorithm can be used in combination with a deep neural network to estimate the Q value. DQN can more accurately find the optimal strategy in a high-dimensional sales state space.

[0109] According to the sales strategy optimization model, combined with the collaborative filtering algorithm, product recommendations are made, the recommendation effects of different product combinations are analyzed, the recommendation strategy is dynamically adjusted, and the optimal sales strategy solution is generated;

[0110] In one embodiment, the system uses a user-based collaborative filtering algorithm, which specifically includes: analyzing the purchasing behavior and preferences of different customers to find similar user groups. For example, user A purchased products X and Y, and user B also purchased product X, then the system can recommend product Y to user B. The system uses purchase history, browsing history, product ratings, etc. from customer behavior data to generate a customer similarity matrix. The similarity can be calculated based on cosine similarity or Pearson correlation coefficient. Based on the feedback from the reinforcement learning model, the system can dynamically adjust the combination of recommended products. For example, if the system finds that some combinations are not effective (that is, the customer did not buy after the recommendation), the optimal recommendation can be regenerated by adjusting the weights of the recommendation list.

[0111] In one embodiment, the system uses an item-based collaborative filtering algorithm, and the system can recommend items that are often purchased together. For example, customers who buy mobile phones often also buy mobile phone cases. The algorithm generates a similarity matrix between items by analyzing the co-occurrence rate between items. The system will continuously monitor the recommendation effect and optimize the recommended product combination in combination with the reward feedback mechanism of reinforcement learning. In some situations, the system may increase the opportunity for cross-selling (for example, recommending items that are less related but complementary).

[0112] Generate a personalized product recommendation list based on customer behavior data through rule-based algorithms;

[0113] Continuously optimize sales strategies based on key information provided by the feedback adjustment module.

[0114] The construction process of the sales strategy optimization model includes the following steps:

[0115] Setting initial strategies for product pricing, promotional plans, and product recommendations, and dynamically adjusting the initial strategies based on the sales data and inventory levels;

[0116] The gradient boosted decision tree algorithm is used to optimize the impact of different pricing and promotional schemes on sales volume and optimize product sales strategies.

[0117] It should be noted that the system introduces the Gradient Boosting Decision Tree (GBDT) algorithm. The GBDT model analyzes the optimal pricing strategy based on the relationship between historical pricing data and sales volume. The specific steps are as follows:

[0118] Input variables include historical pricing, sales volume, inventory levels, customer purchasing behavior, etc. GB DT mines the complex nonlinear relationship between pricing and sales volume through the integration of multiple decision trees.

[0119] Through the prediction of the GBDT model, the system can determine whether a certain price range has a positive impact on sales, thereby dynamically adjusting the price of the product. The system will also combine promotional strategies to further refine the pricing model.

[0120] At the same time, the effectiveness of promotional strategies can also be optimized through the GBDT model, including promotion type (discounts, gifts, etc.), promotion duration, number of participating users, inventory consumption, etc. GBDT uses multiple rounds of learning to identify which promotional plans can bring the highest sales growth. Based on the results of the model, the system can adjust the promotion intensity, duration, and frequency of activities.

[0121] The feedback adjustment module is used to extract key information from feedback data using a sparse representation algorithm to quickly adjust inventory management and sales strategies. The sparse representation algorithm combines sparse coding and decoding mechanisms to perform deep compression processing on feedback data while retaining the core information of the feedback data, and by decoding and reconstructing the compressed data, the feedback data can be effectively transmitted and stored.

[0122] Specifically, the sparse coding process includes: converting a large amount of redundant feedback data into a sparse representation. Through natural language processing technology, the system extracts high-frequency words and key sentences from a large amount of feedback data to remove redundant or irrelevant information. For text comments, the system first performs pre-processing operations such as word segmentation and part-of-speech tagging on the comment data, and then assigns a weight value to each comment through natural language processing to extract important keywords and phrases. The system will give priority to comments with low scores or comments directly related to product performance, price, and service, retaining these valuable content for further processing. For time series data in inventory and sales data, the system will extract key trends such as changes in product sales and inventory fluctuations through deep learning models such as autoencoders to reduce the storage of redundant information.

[0123] The sparse decoding process includes: when the system needs to adjust the feedback data, the sparsely represented feedback data is decoded and reconstructed through the decoder. The decoding process reconstructs the key features of the original data through a deep learning model, and the system can quickly make corresponding adjustments based on this reconstructed information. The decoding model is usually based on an autoencoder, which can retain important change information in the time series during reconstruction. The system reconstructs the sparse text feedback into complete sentences with semantic information, retaining the core opinions related to goods and services. For example, feedback such as "poor quality, low cost performance" is extracted as a simplified opinion and then passed to the operations team. The system can decode compressed numerical feedback data into approximate original data, restore the changing trends of key features such as sales and inventory, and help judge the effectiveness of past strategies.

[0124] To sum up, this application uses the product monitoring module, sensor networks and image recognition technology, and the system can collect real-time sales data of products, including sales volume, inventory, display status and customer behavior, to ensure the comprehensiveness and accuracy of the data, and provide a solid foundation for subsequent sales analysis and optimization.

[0125] The sales analysis module in this application uses graph neural networks and attention mechanisms. By building an interaction graph between products, customers, and supply chains, it can deeply capture complex dependencies and accurately predict the sales trends and regional distribution of products, thereby helping merchants to develop more targeted sales strategies. Through the geographically weighted regression algorithm, the system can identify the trends of high-demand products in different regions, and combine the specific factors of each region to generate refined sales analysis reports, helping merchants optimize inventory configuration and promotion strategies in different regions and improve sales efficiency.

[0126] This application uses a recursive neural network to predict future inventory demand, and combines the random conjugate gradient descent algorithm to optimize the frequency and quantity of inventory replenishment. The system can realize intelligent inventory scheduling, avoid overstocking or out-of-stock, reduce inventory costs, and improve the automation level of supply chain management. Through reinforcement learning and collaborative filtering algorithms, the system can generate the best sales strategy based on sales data and inventory levels, covering product pricing, promotion plans and product recommendations, and continuously optimize sales performance. In addition, the system will dynamically adjust these strategies to ensure their optimal effect in different scenarios.

[0127] This application uses sparse representation algorithms to extract core information from feedback data, combined with compression decoding technology, to quickly adjust inventory management and sales strategies, ensuring that merchants can dynamically adjust operating strategies based on the latest feedback, thereby achieving the ability to quickly respond to market changes. Through edge computing, wireless sensor networks and low-latency transmission protocols, the system can efficiently process and transmit large-scale real-time data, ensuring that merchants can grasp the sales dynamics and inventory status of goods in real time, and improve decision-making efficiency through data visualization tools.

[0128] The above implementation modes are merely descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering and technical personnel in the field shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A marketing management system based on AI algorithm, characterized in that: It includes a commodity monitoring module, a sales analysis module, an inventory management module, a sales strategy optimization module and a feedback adjustment module which are connected in sequence; The commodity monitoring module is used to collect sales data of commodities based on image recognition and sensor network; The sales analysis module is used to comprehensively analyze the sales data obtained by the commodity monitoring module through a deep learning algorithm, predict the best sales trend and regional distribution of commodities, and generate a sales analysis report; The inventory management module is used to optimize inventory scheduling based on the sales analysis report using a stochastic conjugate gradient descent algorithm to achieve intelligent inventory replenishment, commodity allocation, and supply chain automation management; The sales strategy optimization module is used to generate an optimal sales strategy using an AI algorithm based on the sales analysis report and inventory level; the sales strategy includes product pricing adjustment, promotion plan and product recommendation; The feedback adjustment module is used to extract key information from feedback data using a sparse representation algorithm to quickly adjust inventory management and sales strategies; The operation process of the sales analysis module includes the following steps: Receive sales data collected by the commodity monitoring module and perform pre-processing operations; Based on the sales data, a graph neural network algorithm with an attention mechanism is used to construct an interaction graph of products, customers, and supply chains, capture the complex dependencies between nodes, and predict product sales trends; Based on product sales trends, use the Geographically Weighted Regression algorithm to analyze regional sales data and identify high demand trends in different regions; Based on the above analysis results, a sales analysis report is generated; the sales analysis report includes sales trend forecast, customer behavior analysis and regional recommendations; The formula for the geographically weighted regression algorithm is as follows: Among them, y i represents the sales volume of goods in the i-th geographical area; x ia represents the ath influencing factor of the i-th geographical region; β0(u i , v i ) represents the intercept term of the i-th geographical area; β a (u i , v i ) represents the regression coefficient of the ath influencing factor in the ith geographical region; that is, the degree of influence of this factor on the sales volume of goods in this region; (u i , v i ) represents the longitude and latitude coordinates of the i-th geographical area; ∈ i represents the random error of the ith region; p represents the total number of variables that affect the sales volume of the product; The operation process of the inventory management module includes the following steps: Based on the sales trend forecast in the sales analysis report, combined with historical sales data and real-time inventory levels, a recurrent neural network is used to predict future inventory demand; According to the demand forecast results, the stochastic conjugate gradient descent algorithm is used to calculate the replenishment quantity, optimize the inventory replenishment frequency and quantity, and combine the supply chain response speed to achieve intelligent allocation and inventory scheduling; Automatically trigger supply chain operations based on inventory scheduling results, and execute replenishment requests or commodity transfer operations in real time through automated interfaces with suppliers and warehouses; Receive key information returned by the feedback adjustment module and dynamically adjust the inventory management strategy; The formula of the stochastic conjugate gradient descent algorithm is as follows: Among them, Q k+1 Indicates the updated inventory replenishment or transfer quantity after iteration; Q k Indicates the replenishment quantity, transfer quantity or supply chain operation quantity under the current inventory level; α k It represents the step size coefficient of the kth iteration, which controls the magnitude of inventory adjustment in the current iteration; represents the gradient of the objective function at the kth iteration; d k-1 represents the search direction of the k-1th iteration, which is used to determine the current inventory optimization direction; f(Q k ) represents the inventory management objective; Represents the conjugate coefficient, which determines the relationship between the current search direction and the direction of the previous iteration.

2. A marketing management system based on AI algorithm according to claim 1, characterized in that: The sales data includes product sales, customer purchase data, inventory levels and product display status.

3. The marketing management system based on AI algorithm according to claim 1, characterized in that: The operation process of the sales strategy optimization module includes the following steps: Based on the sales trend forecast in the sales analysis report, use the reinforcement learning algorithm to analyze the effects of different sales strategies and establish a sales strategy optimization model; According to the sales strategy optimization model, combined with the collaborative filtering algorithm, product recommendations are made, the recommendation effects of different product combinations are analyzed, the recommendation strategy is dynamically adjusted, and the optimal sales strategy solution is generated; Generate a personalized product recommendation list based on customer behavior data through rule-based algorithms; Continuously optimize sales strategies based on key information provided by the feedback adjustment module.

4. A marketing management system based on AI algorithm according to claim 3, characterized in that: The construction process of the sales strategy optimization model includes the following steps: Setting initial strategies for product pricing, promotional plans, and product recommendations, and dynamically adjusting the initial strategies based on the sales data and inventory levels; The gradient boosted decision tree algorithm is used to optimize the impact of different pricing and promotional schemes on sales volume and optimize product sales strategies.

5. The marketing management system based on AI algorithm according to claim 1, characterized in that: The sparse representation algorithm combines sparse coding and decoding mechanisms to perform deep compression processing on the feedback data while retaining the core information of the feedback data, and realizes effective transmission and storage of the feedback data by decoding and reconstructing the compressed data.

Images