Purchase management decision support system driven by AI large model

Abstract

The invention discloses a purchase management decision support system driven by an AI large model, and particularly relates to the technical field of intelligent purchase management. A data acquisition module obtains historical and current purchase, market price, inventory, sales and customer feedback data of commodities; the clustering analysis module divides purchasing groups based on inventory states and market prices; the self-portrait construction module generates a comprehensive self-portrait through client feedback data; the market analysis module analyzes sales data and predicts future demand changes; the optimization adjustment module optimizes the purchase quantity in combination with demand prediction, customer preferences and purchase group data, so that the accuracy and efficiency of purchase decision making are improved; through intelligent data processing and model prediction, the limitation of dependence on experience and static rules in traditional purchase management is avoided, fluctuation of market demands can be responded in real time, the purchase quantity is ensured to be matched with the market demands, inventory overstock and shortage phenomena are avoided, and customer satisfaction and market competitiveness are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent procurement management technology, and more specifically, to a procurement management decision support system driven by an AI big model. Background Art

[0002] With the rapid development of global economic integration and e-commerce, modern enterprises are facing increasingly complex market demands and competitive pressures. Traditional procurement management methods usually rely on historical experience and static rules, and cannot respond to market changes in real time, which can easily lead to inventory backlogs, capital waste, or insufficient supply, which in turn affects the company's operational efficiency and customer satisfaction. Therefore, traditional procurement decision-making methods can no longer meet the modern market's requirements for procurement accuracy, timeliness, and flexibility.

[0003] In recent years, the rapid development of artificial intelligence has provided new solutions for procurement management. Through intelligent data analysis and prediction, it can help companies more accurately grasp market demand, optimize inventory management, and flexibly adjust procurement strategies according to changes in customer demand. However, most existing intelligent procurement systems are limited to a single data source or static analysis, and it is difficult to fully consider multi-dimensional data such as historical sales, customer feedback, market changes, and procurement group behavior.

[0004] Therefore, there is an urgent need for a system that integrates multi-source data, has predictive capabilities, and can dynamically optimize procurement decisions based on real-time feedback to achieve intelligent, precise, and efficient procurement processes. The procurement management decision support system proposed in this invention, which is driven by an AI big model, aims to improve the accuracy and response speed of procurement decisions by integrating multiple data sources and applying big data analysis and machine learning algorithms, so as to provide all-round intelligent support for the procurement management of enterprises. Summary of the invention

[0005] To achieve the above object, the present invention provides the following technical solutions: AI big model-driven procurement management decision support system, including data collection module, cluster analysis module, self-portrait building module, market analysis module, and optimization and adjustment module; The data collection module is used to collect historical and current purchase data, market price data, inventory status data, sales data, and customer feedback data of various preset commodities; The cluster analysis module is used to divide the purchasing groups based on the historical inventory status data and market price data of each category of goods through a preset cluster analysis algorithm to obtain the purchasing groups that are most similar to the current inventory status and market conditions; The self-portrait construction module is used to construct a customer feedback self-portrait of each category of goods based on the historical customer feedback data of each category of goods, and generate a comprehensive self-portrait based on customer preferences through a preset comprehensive model; The market analysis module is used to analyze market trends based on historical sales data and predict future changes in commodity demand; The optimization and adjustment module is used to optimize the purchase volume of each category of goods during the latest purchase based on the predicted results of future changes in commodity demand and the comprehensive self-portrait of customer preferences, combined with the purchase data of the most similar purchasing groups.

[0006] In a preferred embodiment, the clustering analysis module includes a data conversion unit and a clustering unit. The data conversion unit is used to convert the current inventory status data and market price data of the product into a first feature vector, and convert the historical inventory status data and market price data of the product into a second feature vector, and then transmit the first feature vector and all the second feature vectors to the clustering unit for clustering analysis.

[0007] In a preferred embodiment, the clustering unit uses a K-means clustering algorithm when performing cluster analysis to identify the purchasing group that is most similar to the current inventory situation and market situation.

[0008] In a preferred embodiment, the customer feedback self-portrait includes a sentiment classification portrait, a preset keyword frequency portrait, and a satisfaction rating portrait.

[0009] In a preferred embodiment, the sentiment classification portrait is obtained by performing sentiment classification on historical customer feedback data using a support vector machine, the preset keyword frequency portrait is extracted from the historical customer feedback data using a preset text feature extraction algorithm, the preset keywords are preset and their occurrence frequencies are counted, and then the portrait is obtained, and the satisfaction score portrait is obtained by performing satisfaction score analysis on the historical customer feedback data using a questionnaire survey combined with an assignment method.

[0010] In a preferred embodiment, generating a comprehensive self-portrait based on customer preferences through a preset comprehensive model refers to: The sentiment classification portrait, preset keyword frequency portrait, and satisfaction score portrait results of each category of goods are obtained respectively, and then input into the preset convolutional neural network model together, and the customer preference score corresponding to each category of goods is output. Then, the product types are sorted in descending order according to the customer preference scores of all products to obtain a comprehensive self-portrait.

[0011] In a preferred embodiment, the market analysis module is used to analyze market trends based on historical sales data and predict future changes in demand for commodities, which means: Obtain m influencing factors to construct a multivariate regression model. When constructing the multivariate regression model, based on the dependency of the interaction terms in the multivariate regression model, calculate the mutual information values ​​of all interaction terms, and count the mean, standard deviation and maximum value. Perform weighted summation on the mean, standard deviation and maximum value to obtain a regularization index. Compare the regularization index with the preset threshold one and threshold two. If the threshold one is less than the threshold two, if the regularization index is greater than the threshold two, select L1 regularization. If the regularization index is less than the threshold one, select L2 regularization. If the regularization index is within the range of the threshold one and the threshold two, select elastic net regularization. Finally, the trained multivariate regression model is used to predict future changes in demand for commodities.

[0012] In a preferred embodiment, the logic of influencing factors is: using the Pearson correlation coefficient to calculate the correlation coefficient between each preset influencing factor and the sales volume, and then taking the influencing factors corresponding to the first m correlation coefficients to construct a multivariate regression model.

[0013] In a preferred embodiment, the logic of optimization adjustment is: Get the prediction results of the same category of products separately , the average purchase data of the most similar purchase groups , Customer preference values ​​in comprehensive self-portraits , and then substitute it into the following formula: ; and They are the maximum and minimum purchase quantity limits corresponding to the commodity type. To optimize the adjusted purchase volume, , and All are preset non-zero scale factors.

[0014] Technical effects and advantages of the present invention: This system can accurately predict future changes in demand for goods through in-depth analysis of historical data, market trends, customer feedback, and multi-dimensional data of purchasing group behavior, thereby optimizing purchasing decisions. Through intelligent data processing and model prediction, the system avoids the limitations of traditional purchasing management that relies on experience and static rules, and can respond to market demand fluctuations in real time, ensuring that the purchase volume matches market demand and avoiding inventory backlogs and shortages.

[0015] The present invention uses a self-portrait construction module, combined with sentiment classification, keyword frequency extraction and satisfaction scoring, to comprehensively analyze customer feedback data, thereby accurately understanding customer preferences. The comprehensively generated customer preference self-portrait helps companies better understand customer needs and expectations, and can better meet customer needs when formulating procurement strategies, thereby improving customer satisfaction and market competitiveness.

[0016] The present invention can divide the most similar purchasing groups according to the historical inventory status and market price data of commodities through the cluster analysis module. This function can effectively draw on the purchasing behavior and inventory management experience of similar groups to help enterprises optimize inventory levels and reduce inventory backlogs and capital waste. At the same time, combined with the purchase quantity optimization function of the optimization adjustment module, the efficiency of inventory management and the rationality of procurement plans can be further improved.

[0017] The present invention uses a market analysis module to track market changes and trends in real time and predict future demand fluctuations for commodities. Combined with multiple regression analysis and regularization methods, the system can flexibly adjust procurement strategies based on real-time data and changing trends, ensuring that enterprises can respond to market changes in a timely manner and improving the flexibility and responsiveness of the supply chain.

[0018] The present invention helps enterprises reduce over-purchasing and inventory backlogs through intelligent forecasting and optimization adjustments, and avoids losses caused by out-of-stock or excessive inventory. At the same time, accurate purchase volume calculation can improve supply chain efficiency, reduce resource waste, reduce overall procurement costs, and provide enterprises with higher operational benefits.

[0019] The present invention can not only provide decision support based on static data, but also has the ability to make dynamic adjustments. Based on real-time market data and sales feedback, the system can automatically update the procurement decision model, continuously optimize procurement strategies, reduce manual intervention, and improve decision-making efficiency. Through the deep learning capabilities of the AI ​​big model, the present invention can extract valuable insights from massive amounts of data, provide decision makers with scientific decision support based on data, reduce subjective judgments and human errors, improve the company's intelligent decision-making level, and promote the company's digital transformation. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings; Figure 1 This is a schematic diagram of the procurement management decision support system driven by the AI ​​big model in the present invention.

[0021] Figure 2 It is a schematic diagram of the cluster analysis module in the present invention.

[0022] Figure 3 It is a schematic diagram of the self-portrait building module in the present invention. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] Reference Figure 1-3 The following embodiments are obtained: Embodiment 1:

[0025] In the modern business environment, enterprises are faced with ever-changing market demands, complex supply chain management, and diverse customer preferences. Traditional procurement management decision-making methods often rely on historical experience and static rules, which can easily lead to inventory backlogs or shortages, thereby affecting the operational efficiency and cost control of enterprises. Therefore, how to effectively integrate historical sales data, customer feedback, market changes and forecast results, and scientifically adjust procurement volume has become the key to improving the supply chain management and operational efficiency of enterprises.

[0026] With the rapid development of technologies such as big data, artificial intelligence (AI) and machine learning, more and more companies are beginning to seek data-driven intelligent procurement management decision-making systems. In particular, the application of AI big models can mine deep rules and associations in massive data, optimize procurement decisions, and improve the responsiveness and flexibility of the supply chain.

[0027] The AI ​​big model-driven procurement management decision support system proposed in this invention builds a dynamic adjustment mechanism based on the AI ​​big model by comprehensively utilizing multi-dimensional information such as historical sales data, customer feedback, market forecasts, and procurement group analysis, thereby providing enterprises with more accurate and intelligent procurement decision support. By using the AI ​​big model and machine learning algorithms, a large amount of historical data, market trends, and customer feedback information can be automatically analyzed to help decision makers make scientific procurement decisions and reduce the interference of human subjective factors.

[0028] Based on historical sales data and market analysis, the system predicts future changes in commodity demand, and combines the analysis results of customer preferences and purchasing groups to optimize the purchase volume in real time to avoid inventory backlogs or out-of-stock situations. Through the self-portrait construction module, combined with factors such as customer sentiment analysis, keyword frequency extraction and satisfaction rating, a comprehensive self-portrait based on customer preferences is generated to help companies understand and meet customer needs more accurately, thereby improving customer satisfaction. The system can obtain dynamic information such as market prices, inventory status, sales data, etc. in real time, and flexibly adjust procurement strategies based on the historical data of the most similar purchasing groups to improve the response speed and efficiency of the supply chain.

[0029] Through the calculation of regularization index and the selection mechanism of regularization methods (such as L1, L2, elastic net regularization), the regression model is dynamically adjusted according to the correlation of feature interaction terms to optimize the accuracy of demand forecasting, avoid overfitting, and improve the stability of the model. Through accurate demand forecasting and optimized purchase volume adjustment, it helps companies reduce procurement costs, avoid capital waste, and improve overall operational efficiency while maintaining a reasonable inventory level.

[0030] The present invention proposes a procurement management decision support system driven by an AI big model, comprising a data acquisition module, a cluster analysis module, a self-portrait construction module, a market analysis module, and an optimization and adjustment module; the data acquisition module, the cluster analysis module, the self-portrait construction module, the market analysis module, and the optimization and adjustment module are communicatively connected with each other.

[0031] The data collection module is used to collect the historical and current purchase data, market price data, inventory status data, sales data, and customer feedback data of various commodities; the data collection module is the basis of this system. It provides the necessary raw data for subsequent analysis and decision-making by collecting the historical and current purchase data, market price data, inventory status data, sales data and customer feedback data of various commodities. The data collection module not only provides data reflecting key elements such as commodity sales, market prices and inventory status, but also integrates customer feedback information, thus providing comprehensive input for subsequent modules. Collection of historical data: By collecting historical data, it can help the system understand past purchasing behaviors, sales trends and inventory management, and provide support for subsequent trend forecasting and decision optimization. Market price data: Understand the price fluctuations of commodities in the market, help the system adjust the procurement strategy in real time, and optimize price management. Customer feedback data: By obtaining customer feedback, it can reveal the changes in market demand and customer preferences, and provide data support for the subsequent construction and optimization of customer portraits. Whether it is demand forecasting, cluster analysis or customer portrait construction, all modules need to rely on the data provided by this module, so the data collection module is the basis of the system.

[0032] The cluster analysis module is used to divide the purchasing groups based on the historical inventory status data and market price data of each category of goods through the preset cluster analysis algorithm to obtain the purchasing groups that are most similar to the current inventory situation and market situation; the cluster analysis module divides the goods into different purchasing groups based on the preset clustering algorithm (such as K-means or DBSCAN) by analyzing the historical inventory data and market price data of each category of goods, and then identifies the purchasing groups that are most similar to the current inventory situation and market situation. In this way, the system can learn from the purchasing behavior of similar historical groups to make more appropriate decisions. Identify similar groups: Cluster analysis helps to classify goods according to market environment, inventory status and other factors. For each group, the system can learn the best purchasing patterns and trends from historical data. Improve decision-making accuracy: By identifying groups similar to the current situation, the system can learn from the behavioral patterns in similar situations in the past and predict more accurate purchasing needs. Importance: This module is crucial for optimizing procurement plans. By identifying the most similar purchasing groups, the system can more effectively adjust procurement strategies and reduce inventory backlogs or shortages.

[0033] The self-portrait construction module is used to construct a customer feedback self-portrait for each category of goods based on the historical customer feedback data of each category of goods. Through the preset comprehensive model, a comprehensive self-portrait based on customer preferences is generated; it deeply understands the needs and preferences of customers, and provides a specific expression of customer preferences through sentiment analysis, keyword frequency analysis and satisfaction score multi-dimensional data. Customer sentiment analysis: Through sentiment classification, identify the positive or negative feedback of customers on the goods, and help understand the overall attitude of the market towards the goods. Keyword extraction and frequency statistics: Extract keywords from customer feedback and analyze their frequency to further understand customer demand trends. Satisfaction score: Based on customer ratings and feedback data, quantify customer satisfaction and provide a reference for customer satisfaction for subsequent purchasing decisions. By building a customer self-portrait, the system can better understand the changes in customer needs, thereby optimizing the procurement strategy of goods, improving customer satisfaction and purchase conversion rate.

[0034] The market analysis module is used to analyze market trends based on historical sales data and predict future changes in demand for goods. Through in-depth analysis of historical sales data, it helps the system identify and predict the changing patterns of demand for goods so that purchasing adjustments can be made in advance. Utilize historical sales data, combined with statistical methods and machine learning models (such as multivariate regression models), to predict future trends in demand for goods. Not only analyze past sales data, but also consider influencing factors such as market environment, seasonal factors, and economic changes to help the system fully understand market changes. Market trend forecasting is crucial for accurately adjusting purchase volume. Through accurate market analysis, the system can effectively foresee future changes in demand and make adjustments based on forecast results to avoid over-purchasing or insufficient inventory.

[0035] The optimization and adjustment module is used to optimize and adjust the purchase volume of each category of goods during the latest purchase of goods based on the forecast results of future demand changes and the comprehensive self-portrait of customer preferences, combined with the purchase data of the most similar purchasing groups. It can ensure the precise adjustment of the purchase volume to minimize the purchase cost and ensure the efficient operation of the supply chain. Combine multi-dimensional information such as future demand forecasts, customer preferences, and historical purchase data of the most similar purchasing groups to ensure that the purchase decision is based on a global and comprehensive perspective. Through the optimization formula, the purchase volume of each category of goods is dynamically adjusted to ensure that the purchase plan is accurate and forward-looking. Based on real-time data and forecast results, the purchase volume of each category of goods is ensured to be within a reasonable range to avoid excessive or shortage inventory. This module is the core of the entire system. Through intelligent optimization and adjustment, it ensures that enterprises can make reasonable purchase decisions in a dynamically changing market environment, reduce waste and improve operational efficiency.

[0036] The clustering analysis module includes a data conversion unit and a clustering unit. The data conversion unit is used to convert the current inventory status data and market price data of the product into a first feature vector, and convert the historical inventory status data and market price data of the product into a second feature vector. The first feature vector and all the second feature vectors are then sent to the clustering unit for clustering analysis. The clustering unit uses the K-means clustering algorithm to identify the purchasing group that is most similar to the current inventory situation and market situation.

[0037] The main task of the data conversion unit is to convert the inventory status data and market price data of the goods into feature vectors suitable for cluster analysis. This process includes the following steps: Transformation of the current inventory status data of goods: The current inventory status data of goods mainly includes the remaining inventory, inventory consumption frequency, etc. These data can reflect the inventory status of goods at the current time point and the consumption speed of goods in the market. These data are standardized to generate a first feature vector describing the current inventory status.

[0038] Transformation of commodity historical inventory status data: Historical inventory status data can provide information about the inventory changes of commodities in the past period of time. These data also include information such as inventory remaining quantity and inventory consumption frequency. Through similar standardization steps, a second feature vector describing the historical inventory status of commodities is generated.

[0039] Conversion of market price data: Market price data includes the purchase price and sales price of commodities. These two data are the pricing information of commodities in the market and can reflect the market value and sales strategy of commodities. After standardizing these two data, they are converted into feature vectors.

[0040] Through the above steps, the data conversion unit converts the current inventory status data and market price data of the commodity into a first feature vector and a second feature vector. These two vectors contain the inventory and market pricing information of the commodity and are the basic inputs for cluster analysis.

[0041] The task of the clustering unit is to divide the commodities into different purchasing groups. Specifically, the clustering analysis is performed using the K-means clustering algorithm, which is a prior art well known to those skilled in the art. The steps are briefly described as follows in the present invention: Initialize cluster centers: The K-means clustering algorithm first requires specifying the number of clusters (K). During initialization, K data points are randomly selected as initial cluster centers. These centers will represent K different purchasing groups. The number (K) can be determined using the inflection point method.

[0042] Assign each data point to the nearest cluster center: For each product's first eigenvector and second eigenvector, calculate its distance from each cluster center, usually using the Euclidean distance metric. Assign each product to its closest cluster center, so that the products in each cluster have similar inventory status and market price characteristics.

[0043] Update cluster center: Once all products are assigned to a cluster, the cluster center will be updated based on the feature vectors of all products in the cluster. The new cluster center is the average of the feature vectors of all products in the cluster, that is, the weighted average of the inventory status and market price of all products in the cluster.

[0044] Repeat the iteration: Continue the assignment and update steps until the cluster center stabilizes and no longer changes significantly. This indicates that the algorithm has converged and all products have been accurately assigned to the most appropriate cluster.

[0045] Through K-means cluster analysis, the system will identify the purchasing groups that are most similar to the current inventory status and market conditions. The characteristics of these groups are that their inventory status and market price feature vectors are highly similar, which can help companies make more accurate purchasing decisions based on the purchasing patterns of similar historical groups. For example, the system may find that: one type of goods has a higher sales price and purchase price when the inventory surplus is high and the consumption frequency is low, which is suitable for inventory management with less replenishment. Another type of goods has a lower purchase price and sales price when the inventory surplus is low and the consumption frequency is high, which is suitable for inventory management with higher replenishment.

[0046] Customer feedback self-portraits include sentiment classification portraits, preset keyword frequency portraits, and satisfaction rating portraits. Sentiment classification portraits are obtained by using support vector machines to perform sentiment classification on historical customer feedback data and to create portraits. Preset keyword frequency portraits are obtained by extracting historical customer feedback data through a preset text feature extraction algorithm, presetting keywords and counting their frequency of occurrence, and then creating portraits. Satisfaction rating portraits are obtained by combining questionnaire surveys with the assignment method to perform satisfaction ratings on historical customer feedback data and create portraits.

[0047] Sentiment classification portrait is a graphical display based on the sentiment tendency in customer feedback data. Sentiment classification is a sentiment analysis of customer feedback, mainly classifying customer evaluations into different sentiment categories, such as positive, negative or neutral. The sentiment classification process is usually implemented using machine learning algorithms such as support vector machines (SVM). Support vector machines are a supervised learning model that is often used for classification problems. In sentiment analysis, support vector machines learn a large amount of labeled training data to find an optimal hyperplane to separate different categories of sentiment feedback (such as positive and negative). During the training process, support vector machines can distinguish different categories of feedback text by finding the optimal boundary. For example, positive customer feedback and negative customer feedback can be classified by support vector machines, thereby generating sentiment classification portraits for each category of goods. The portrait generates a sentiment tendency distribution map of each product by counting the proportion of positive, negative and neutral feedback for each product, as well as the intensity of the feedback. Sentiment classification portraits not only reflect the overall sentiment of customers towards the product, but also reveal potential changes in customer emotions, providing a basis for subsequent decision-making.

[0048] The preset keyword frequency profile is constructed by extracting specific keywords from customer feedback and calculating their frequency of occurrence. This process is usually carried out through a text feature extraction algorithm, with the goal of identifying the most frequently mentioned words in customer feedback, thereby reflecting the focus and trends of customer attention. Text feature extraction algorithm: Commonly used text feature extraction methods include term frequency-inverse document frequency (TF-IDF), bag-of-words model, etc. These algorithms extract pre-defined keywords (such as quality, price, service, etc.) by segmenting customer feedback, and count the frequency of occurrence of these keywords in all feedback. Keyword frequency calculation: During the analysis process, the system calculates the frequency of occurrence of each preset keyword and measures the importance of the keyword by the frequency. If a keyword appears frequently in most feedback, it means that this factor may be a key issue that customers are concerned about. In this way, a keyword frequency profile of the product can be generated to show the features that customers are most concerned about (such as a certain problem or demand) and emotional preferences. This profile can help merchants identify customer demand hotspots, pain points and expectations.

[0049] The satisfaction score profile is obtained by scoring and analyzing customer feedback data, usually through questionnaires or customer evaluation systems, combined with the assignment method to quantify customer feedback. Each customer's feedback on the product is assigned a score to indicate their satisfaction. By counting and analyzing these scores, a satisfaction score profile for each product can be generated.

[0050] Questionnaire survey and value assignment method: By designing a questionnaire or rating system, customers are asked to rate the products, and the ratings are usually on a scale of 1 to 5. According to the customer's rating, the feedback is quantified. For example, 1 point means extremely dissatisfied, and 5 points means very satisfied. The value assignment method directly maps the customer's rating to a numerical value, which facilitates subsequent statistics and analysis. Rating statistics and portrait generation: Once all customer rating data are collected and assigned, the system can count the average rating of each product, the distribution of ratings (such as the proportion of high or low ratings), and calculate the overall customer satisfaction. This satisfaction rating will be presented in the form of a portrait, such as a bar chart, pie chart, or other visual method to show the proportion of different ratings and their average values. The satisfaction rating portrait can quantify the customer's satisfaction with the product, provide merchants with data support based on customer feedback, and help them determine which products need to be improved and which products can meet customer needs.

[0051] Generating a comprehensive self-portrait based on customer preferences through a preset comprehensive model means: obtaining the sentiment classification portrait, preset keyword frequency portrait, and satisfaction score portrait results of each category of goods respectively, and then inputting them into the preset convolutional neural network model together, outputting the customer preference score corresponding to each category of goods, and then sorting the product types in descending order according to the customer preference scores of all products to obtain a comprehensive self-portrait.

[0052] In order to combine sentiment classification portraits, keyword frequency portraits, and satisfaction rating portraits to generate a comprehensive customer preference score, the system uses a preset convolutional neural network (CNN) model. Convolutional neural networks can process complex input data from multiple perspectives, especially effectively extracting and combining features of multiple information such as images and text.

[0053] Input data: Sentiment classification images, keyword frequency images, and satisfaction rating images are all converted into a format suitable for CNN model input (for example, by converting text information into vectors through encoding or feature embedding).

[0054] Convolutional layers: Convolutional neural networks extract features from input data through multiple convolutional layers. Each layer captures higher-level patterns by locally perceiving the input features, such as extracting sentiment trends, keyword relevance, and other information from customer feedback.

[0055] Pooling layer: The pooling layer reduces the dimension of the features output by the convolutional layer, extracts the most important features, reduces the amount of calculation, and improves the efficiency and generalization ability of the model.

[0056] Fully connected layer: Finally, the pooled features are passed to the fully connected layer, and the network is trained and optimized to obtain a comprehensive customer preference score, which reflects the degree of preference for each product in the customer's mind.

[0057] After the convolutional neural network processes the input portraits, the system outputs the customer preference score for each category of goods. The score represents the popularity of the goods and the customer's preference. Through these preference scores, the system can sort all the goods in descending order to obtain a comprehensive self-portrait of the goods. All goods are sorted according to the customer preference score, with the goods with higher preference scores at the front and the goods with lower preference scores at the back. This sorting can help decision makers identify which goods are currently the most popular and which goods may need further optimization or adjustment. The sorted product list can not only reflect the overall preferences of customers, but also reveal the popularity of various types of goods among different customer groups, providing an important reference for companies in formulating procurement, marketing and product optimization strategies.

[0058] By generating this comprehensive self-portrait, companies can understand which products are most popular in the market and adjust product strategies based on customer preferences, such as increasing sales of popular products, improving the characteristics of products with poor feedback, optimizing inventory management, ensuring that procurement volumes are highly matched with customer demand, and avoiding inventory backlogs or shortages.

[0059] The market analysis module is used to analyze market trends based on historical sales data and predict future changes in commodity demand. Obtain m influencing factors to construct a multivariate regression model. When constructing the multivariate regression model, based on the dependency of the interaction terms in the multivariate regression model, calculate the mutual information values ​​of all interaction terms, and count the mean, standard deviation and maximum value. Perform weighted summation on the mean, standard deviation and maximum value to obtain a regularization index. Compare the regularization index with the preset threshold one and threshold two. If the threshold one is less than the threshold two, if the regularization index is greater than the threshold two, select L1 regularization. If the regularization index is less than the threshold one, select L2 regularization. If the regularization index is within the range of the threshold one and the threshold two, select elastic net regularization. Finally, the trained multivariate regression model is used to predict future changes in demand for commodities.

[0060] By selecting features with high correlation, a multivariate regression model can be constructed to predict the value of the target variable. In the classic linear regression model, the target variable Y is a linear combination of the feature variables. Assume that the feature variables have and However, the traditional linear regression model only considers the linear effect of a single feature on the target variable, but in fact, there may be interactions between features. Adding interaction terms can improve the performance of the model. For example: ; , and are all regression coefficients. By adding interaction terms, the model can capture the mutual influence between different features. is the error term. When fitting data, the least squares method can be used to estimate the regression coefficient to minimize the fitting error. The multivariate regression model may be at risk of overfitting, especially when there are many features. In order to prevent overfitting, L1 regularization (Lasso regression), L2 regularization (Ridge regression) or elastic net regularization can be introduced, which can control the complexity of the model by imposing a penalty term on the regression coefficient: Lasso regression (L1 regularization): Add the absolute value penalty term (L1 norm) of the regression coefficient to the least squares objective function: ; Ridge regression (L2 regularization): Add the square penalty term (L2 norm) of the regression coefficient to the least squares objective function: ; Among them, λ is the penalty coefficient of regularization. Regularization helps to reduce the complexity of the model and improve the generalization ability of the model. are the estimated coefficients of the regression model, i.e. the parameters to be optimized. is the dependent variable of the ith observation, which is sales volume in this invention, is the independent variable vector (feature variable such as market price) of the ith observation, n represents the total number of observations in the data set (i.e. the number of samples), m represents the number of independent variables, i.e. how many features need to be fitted, and j represents the index of the feature or variable in the regression model. j varies from 1 to m, indicating that there are a total of m features (independent variables) involved in the regression; It represents the regression coefficient of the jth feature, reflecting the influence of the jth feature on the dependent variable; Represents the j-th eigenvalue of the i-th observation, that is, the value of the j-th feature in the i-th data point.

[0061] Elastic Net regularization combines L1 and L2 regularization to better handle the correlation between features and is particularly effective when processing high-dimensional data, especially when there is correlation between features. Specifically, Elastic Net regularization balances the two by introducing a mixed regularization term, usually through two hyperparameters to control the weights of Lasso and Ridge regularization.

[0062] By evaluating the nonlinear relationship between interaction items through mutual information and calculating a comprehensive regularization index based on their distribution, we can help identify the nonlinear dependencies between features and dynamically adjust the regularization strategy based on these relationships to select the most suitable regularization method (L1, L2 or elastic net). The following are the detailed implementation steps: Mutual information is an indicator of the dependency between two random variables, which can capture both linear and nonlinear relationships. The higher the mutual information, the more information is shared between the two features (or interaction terms), that is, the stronger the dependency between them. The calculation formula is not described here.

[0063] Calculate all mutual information values ​​between feature pairs to obtain a mutual information matrix. Once the mutual information values ​​of all interaction items are calculated, these values ​​can be statistically analyzed to obtain their distribution information. For example: mean, standard deviation and maximum value. These statistics will help understand the strength of the dependency relationship between features in the data and its distribution. According to the mutual information distribution of all interaction items, a comprehensive regularization index can be calculated to determine which regularization method to use. The mean, standard deviation and maximum value are first multiplied by the preset corresponding non-zero weight coefficients, and then accumulated and summed to obtain the regularization index that quantifies the strength of the nonlinear relationship between features. If the regularization index is small, it means that the dependency between most features is weak, and L2 regularization (Ridge regression) can be used to reduce the overfitting of the model. If the regularization index is large, it means that there is a strong nonlinear dependency between features, and L1 regularization (Lasso regression) can be used for feature selection to remove some redundant features. If the regularization index is at a medium value, elastic net regularization can be used to combine L1 and L2 regularization, which is suitable for situations where there are some related features but other features still want to be retained.

[0064] The logic of the influencing factors is: the Pearson correlation coefficient is used to calculate the correlation coefficient between each preset influencing factor and the sales volume, and then the influencing factors corresponding to the first m correlation coefficients are taken to construct a multiple regression model. When predicting sales volume, it is first necessary to identify and select the key factors that affect the sales of goods. In order to ensure the accuracy and reliability of the prediction model, the selection of influencing factors is crucial. This method uses the Pearson correlation coefficient to evaluate the correlation between each preset influencing factor and the sales volume of the goods, and determines which factors have an important impact on the sales volume based on this correlation, thereby constructing a multiple regression model. The following is a detailed explanation of the principle: The Pearson correlation coefficient is a commonly used statistical method to measure the strength of the linear relationship between two variables. Its value range is from -1 to +1. The closer the value is to +1, the higher the degree of positive correlation between the variables; the closer the value is to -1, the higher the degree of negative correlation between the variables; the closer the value is to 0, the more there is no linear relationship between the two. Compare each preset influencing factor with the sales volume of the goods, and measure the linear relationship between each influencing factor and the sales volume by calculating the Pearson correlation coefficient. Common preset influencing factors include market price, advertising budget, promotional activities, seasonal changes, economic environment, inventory status, etc.

[0065] For each influencing factor, calculate the Pearson correlation coefficient between it and sales volume. The Pearson correlation coefficient is obtained by statistically calculating the influencing factor data and sales volume data, and reflects the strength of the relationship between the influencing factor and sales volume. According to the calculated Pearson correlation coefficient, select the influencing factor with the strongest correlation with sales volume. Usually, in order to improve the effect of the regression model, factors with a high positive or negative correlation with sales volume are selected. According to the value of the Pearson correlation coefficient, select the first m factors with the strongest correlation coefficients from all preset influencing factors. Selecting m factors with high correlation can ensure that these factors have a significant impact on the change in sales volume. In the selection process, the most appropriate m value can be determined based on actual needs and business background. The choice of m value depends on the characteristics of the data and the complexity of the problem.

[0066] The logic of optimization and adjustment is: obtain the prediction results of the same category of products separately , the average purchase data of the most similar purchase groups , Customer preference values ​​in comprehensive self-portraits , and then substitute it into the following formula: ; and They are the maximum and minimum purchase quantity limits corresponding to the commodity type. The maximum purchase quantity limit can prevent the purchase quantity from exceeding the market demand limit or inventory limit, and the minimum purchase quantity limit can prevent the purchase quantity from falling below a certain level to avoid out-of-stock situations. To optimize the adjusted purchase volume, , and are all preset non-zero proportional coefficients, corresponding to the future demand forecast, the average purchase volume of the most similar purchasing groups, and the impact of customer preferences. Their values ​​reflect the degree of influence of these factors relative to the optimization of purchase volume. The optimization adjustment formula is designed to dynamically adjust the purchase volume of goods by integrating the future demand of goods, the historical purchase behavior of the most similar groups, and customer preferences. By adjusting the coefficients and setting reasonable upper and lower limits for purchases, the system can avoid over-purchase and out-of-stock situations while ensuring supply chain efficiency. This method can improve the accuracy and response speed of procurement decisions, thereby helping companies optimize inventory management and procurement plans.

[0067] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0068] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0069] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0070] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0071] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. AI big model driven procurement management decision support system, characterized by: It includes data collection module, cluster analysis module, self-portrait building module, market analysis module, and optimization and adjustment module; The data collection module is used to collect historical and current purchase data, market price data, inventory status data, sales data, and customer feedback data of various preset commodities; The cluster analysis module is used to divide the purchasing groups based on the historical inventory status data and market price data of each category of goods through a preset cluster analysis algorithm to obtain the purchasing groups that are most similar to the current inventory status and market conditions; The self-portrait construction module is used to construct a customer feedback self-portrait of each category of goods based on the historical customer feedback data of each category of goods, and generate a comprehensive self-portrait based on customer preferences through a preset comprehensive model; The market analysis module is used to analyze market trends based on historical sales data and predict future changes in commodity demand; The optimization and adjustment module is used to optimize the purchase volume of each category of goods during the latest purchase based on the predicted results of future changes in commodity demand and the comprehensive self-portrait of customer preferences, combined with the purchase data of the most similar purchasing groups.

2. The AI ​​big model driven procurement management decision support system according to claim 1 is characterized in that: The clustering analysis module includes a data conversion unit and a clustering unit. The data conversion unit is used to convert the current inventory status data and market price data of the product into a first feature vector, and convert the historical inventory status data and market price data of the product into a second feature vector, and then transmit the first feature vector and all the second feature vectors to the clustering unit for clustering analysis.

3. The AI ​​big model driven procurement management decision support system according to claim 2 is characterized in that: The clustering unit uses the K-means clustering algorithm when performing cluster analysis to identify the purchasing group that is most similar to the current inventory situation and market situation.

4. The AI ​​big model driven procurement management decision support system according to claim 3 is characterized in that: The customer feedback self-portrait includes sentiment classification portrait, preset keyword frequency portrait, and satisfaction rating portrait.

5. The AI ​​big model driven procurement management decision support system according to claim 4 is characterized in that: The sentiment classification portrait is obtained by performing sentiment classification on historical customer feedback data through a support vector machine and then performing a portrait. The preset keyword frequency portrait is extracted from historical customer feedback data through a preset text feature extraction algorithm, and the preset keywords are counted and their occurrence frequency is counted before the portrait is obtained. The satisfaction score portrait is obtained by performing a satisfaction score on historical customer feedback data through a questionnaire survey combined with the assignment method and then performing a portrait.

6. The AI ​​big model driven procurement management decision support system according to claim 5 is characterized in that: Generating a comprehensive self-portrait based on customer preferences through a preset comprehensive model means: The sentiment classification portrait, preset keyword frequency portrait, and satisfaction score portrait results of each category of goods are obtained respectively, and then input into the preset convolutional neural network model together, and the customer preference score corresponding to each category of goods is output. Then, the product types are sorted in descending order according to the customer preference scores of all products to obtain a comprehensive self-portrait.

7. The AI ​​big model driven procurement management decision support system according to claim 6 is characterized in that: The market analysis module is used to analyze market trends based on historical sales data and predict future changes in commodity demand. Obtain m influencing factors to construct a multivariate regression model. When constructing the multivariate regression model, based on the dependency of the interaction terms in the multivariate regression model, calculate the mutual information values ​​of all interaction terms, and count the mean, standard deviation and maximum value. Perform weighted summation on the mean, standard deviation and maximum value to obtain a regularization index. Compare the regularization index with the preset threshold one and threshold two. If the threshold one is less than the threshold two, if the regularization index is greater than the threshold two, select L1 regularization. If the regularization index is less than the threshold one, select L2 regularization. If the regularization index is within the range of the threshold one and the threshold two, select elastic net regularization. Finally, the trained multivariate regression model is used to predict future changes in demand for commodities.

8. The AI ​​big model driven procurement management decision support system according to claim 7 is characterized in that: The logic of influencing factors is as follows: the Pearson correlation coefficient is used to calculate the correlation coefficient between each preset influencing factor and sales volume, and then the influencing factors corresponding to the first m correlation coefficients are taken to construct a multiple regression model.

9. The AI ​​big model driven procurement management decision support system according to claim 8, characterized in that: The logic of optimization adjustment is: Get the prediction results of the same category of products separately , the average purchase data of the most similar purchase groups , Customer preference values ​​in comprehensive self-portraits , and then substitute it into the following formula: ； and They are the maximum and minimum purchase quantity limits corresponding to the commodity type. To optimize the adjusted purchase volume, , and All are preset non-zero scale factors.

Images