Multi-mode electronic commerce data integrated management system

Abstract

The invention discloses a multi-mode electronic commerce data integrated management system, and relates to the technical field of electronic commerce data management. The multi-modal electronic commerce data integrated management system comprises a data acquisition module used for acquiring the current sales volume and commodity sales data of each commodity after the current period is ended, a data analysis module used for analyzing the customer flow index and the customer emotion fluctuation index of each commodity based on the commodity sales data, and a data storage module used for storing the customer flow index and the customer emotion fluctuation index of each commodity. The prediction analysis module is used for predicting the sales prediction quantity of each commodity in the next period based on the customer flow index and the customer emotion fluctuation index of each commodity, and the inventory calling module is used for carrying out scheduling supplementary processing on each commodity based on the sales prediction quantity of the next period. According to the method, the customer flow index and the customer emotion fluctuation index are integrated, the purchase intention and market demand change of the customer can be reflected more comprehensively, and the sales volume of each commodity can be accurately predicted through the multi-dimensional data fusion mode.

Description

Technical Field

[0001] The present invention relates to the technical field of e-commerce data management, and in particular to a multi-modal e-commerce data integrated management system. Background Art

[0002] With the rapid growth of the e-commerce market, sales forecasting and inventory management have become key issues facing companies. Traditional sales forecasting methods usually rely on historical sales data and some basic statistical analysis models to predict future sales. However, these methods often face problems such as poor accuracy and slow response. Especially in the case of unstable market demand or diversified customer behavior, traditional methods mostly ignore the individual differences of customers, real-time emotional fluctuations and behavioral patterns, resulting in inaccurate sales forecasts and difficulty in flexibly adjusting inventory management according to actual needs, which may lead to inventory backlogs or out-of-stock phenomena, thereby affecting the profitability and operational efficiency of the company.

[0003] Based on the above, it is found that the limitations of the existing technology include at least the following problems. First, most of the existing e-commerce data management methods rely on single historical sales data and customer behavior data for sales forecasting, but do not conduct comprehensive analysis of multimodal data such as customer mobility index and emotional fluctuation index. In the absence of these dynamic factors, it is difficult for existing technology to accurately reflect the complex behavior of customers in the purchase process, such as the emotional fluctuations of customers when browsing products, their interest in products, and the frequency of customers jumping between different product pages. These factors have a direct impact on the customer's final purchase decision, and it is often difficult to capture them by simply relying on traditional sales data for prediction. Specifically, the lack of a combination of customer sentiment fluctuations and liquidity data will lead to deviations in sales forecast results, making it difficult to accurately predict customer purchase intentions. For example, the purchase intention of certain products may decrease due to customer sentiment fluctuations, but if existing technologies are based only on historical sales data, it is difficult to identify the negative impact of such sentiment changes, and thus higher sales volumes are still predicted, resulting in excess or insufficient inventory. In addition, customer liquidity is not fully considered. The time customers spend on product pages and the frequency of jumps are important indicators for predicting sales trends. If this is not included in the prediction model, it is easy to cause a disconnect between the prediction results and actual demand. Summary of the invention

[0004] In view of the deficiencies of the prior art, the present invention provides a multimodal e-commerce data integrated management system, which solves the problem that the prior art fails to comprehensively consider the dynamic factors of customer mobility and emotional fluctuations, resulting in a lack of accuracy in sales forecasts and difficulty in effectively adapting to changes in market demand.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a multimodal e-commerce data integrated management system, including: a data acquisition module, a data analysis module, a prediction analysis module, and an inventory retrieval module; the data acquisition module is used to obtain the current sales volume and commodity sales data of each commodity after the current cycle ends; the data analysis module is used to analyze the customer flow index and customer emotion fluctuation index of each commodity based on the commodity sales data of each commodity; the prediction analysis module is used to predict the sales forecast of each commodity in the next cycle based on the customer flow index and customer emotion fluctuation index of each commodity; the inventory retrieval module is used to schedule and replenish each commodity based on the sales forecast of the next cycle; wherein the specific formula for calculating the sales forecast of each commodity in the next cycle is as follows: ;in, For the next cycle The sales forecast of the product, For the The current sales volume of the product, For the The customer flow index of the product, is the customer flow impact coefficient stored in the database, For the The customer sentiment fluctuation index of the product, is the customer emotion fluctuation correction coefficient stored in the database, is the customer emotion fluctuation influence coefficient stored in the database, , , The number of commodities.

[0006] Furthermore, the product sales data includes the purchase browsing time of each customer, the return rate, the product page loading time of each customer when purchasing the product, and the text information of each comment, the number of comment forwardings, and the number of comment likes.

[0007] Furthermore, the specific steps for analyzing the customer flow index of each commodity are as follows: read the purchase browsing time of each customer for each commodity and conduct a comprehensive analysis to obtain the purchase browsing index of each commodity; read the product page loading time of each customer when purchasing each commodity and conduct a comprehensive analysis to obtain the product page loading index of each commodity; obtain the browsing path complexity index, interaction intensity index, and historical purchase index of each commodity, and conduct a comprehensive analysis in combination with the purchase browsing index and product page loading index of each commodity to obtain the customer flow index of each commodity.

[0008] Furthermore, the specific formula for calculating the customer flow index of each product is as follows: ;in, For the The customer flow index of the product, For the The purchase browsing index of the product, For the The browsing path complexity index of the product, is the browsing influence coefficient stored in the database, For the The interaction intensity index of the commodity, is the interaction influence coefficient stored in the database, For the The product page loading index of the product, is the load adjustment factor stored in the database, is the loading influence coefficient stored in the database, For the The historical purchase index of the commodity, is the historical purchase impact coefficient stored in the database, , The number of commodities.

[0009] Furthermore, the specific steps for obtaining the browsing path complexity index of each product are as follows: for each product, respectively obtain the page jump times, page dwell time, and click depth values ​​of each customer when purchasing the product, and conduct comprehensive analysis to obtain the page jump index, page dwell index, and click depth index of each product; conduct comprehensive analysis on the page jump index, page dwell index, and click depth index of each product to obtain the browsing path complexity index of each product.

[0010] Furthermore, the specific steps to obtain the interaction intensity index of each product are as follows: for each product, obtain the number of product clicks, product comments, product likes, and product shares of each customer, and conduct comprehensive analysis to obtain the product click index, product comment index, product like index, and product sharing index of each product; conduct comprehensive analysis on the product click index, product comment index, product like index, and product sharing index of each product to obtain the interaction intensity index of each product.

[0011] Furthermore, the specific steps of analyzing the customer emotion fluctuation index of each product are as follows: read the text information of each comment of each product, and perform sentiment analysis to obtain the sentiment score of each comment of each product, and perform comprehensive analysis to obtain the sentiment score index of each product; read the return rate of each customer of each product, and perform comprehensive analysis to obtain the return index of each product; read the number of comment forwardings and the number of comment likes of each comment of each product, and perform comprehensive analysis to obtain the comment interaction index of each product; obtain the comment time difference index of each product, and perform comprehensive analysis based on the sentiment score index, comment interaction index and return index of each product to obtain the customer emotion fluctuation index of each product.

[0012] Furthermore, the specific formula for calculating the customer sentiment fluctuation index of each product is as follows: ;in, For the The customer sentiment fluctuation index of the product, For the The sentiment score index of the product. For the The comment interaction index of the product. For the The time difference index of reviews of products. is the time difference impact coefficient of the comments stored in the database, For the The return index of the product, is the return impact coefficient stored in the database, , The number of commodities.

[0013] Furthermore, the text information of each comment of each product is read, and sentiment analysis is performed to obtain the sentiment score of each comment of each product. The specific steps are as follows: based on natural language processing technology, the text information of each comment of each product is processed separately; each comment of each product after text processing is input into the pre-trained analysis model for sentiment analysis to obtain the sentiment score of each comment of each product.

[0014] Furthermore, based on natural language processing technology, the specific steps of performing text processing on the text information of each comment of each product are as follows: performing noise removal processing on the text information of each comment of each product; after the noise removal processing, performing word segmentation processing on the text information of each comment of each product; after the word segmentation processing, performing word removal processing on the text information of each comment of each product.

[0015] The present invention has the following beneficial effects: (1) This multimodal e-commerce data integrated management system can more comprehensively reflect customers' purchasing intentions and changes in market demand by integrating the customer flow index and the customer emotion fluctuation index. For example, the complexity of customers' browsing paths, interaction intensity, and emotion fluctuations, such as comments, likes, and returns, directly affect their purchasing decisions. Traditional methods only rely on sales data or customer behavior data, ignoring these dynamic factors, which can easily lead to deviations in sales forecasts. Through this multi-dimensional data fusion method, the system can accurately predict the sales volume of each product, thereby providing enterprises with more timely and accurate sales data support and optimizing inventory management.

[0016] (2) The multimodal e-commerce data integration management system can dynamically adjust inventory scheduling based on predicted sales volume. By combining sales forecasts and customer sentiment fluctuation index, the system can adjust the replenishment plan of goods in real time, effectively avoiding inventory backlogs or out-of-stock phenomena caused by inaccurate sales forecasts. Specifically, the system further optimizes sales forecasts by analyzing data such as return rate and review time difference index, making inventory management more flexible and accurate, avoiding unnecessary inventory backlogs, and ensuring the balance between product supply and market demand.

[0017] (3) This multimodal e-commerce data integration management system can not only predict customer needs, but also provide more personalized product recommendations and marketing strategies through in-depth analysis of customer behavior and emotions. Customers' browsing time, comment interaction, likes and sharing data can reflect their real needs and emotional fluctuations for products. The system analyzes these data and provides companies with accurate customer portraits. This precise personalized recommendation and dynamic adjustment of marketing strategies can enhance customers' purchasing experience and increase customer stickiness, thereby improving sales efficiency and conversion rate, and promoting corporate profit growth.

[0018] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a block diagram of a multimodal e-commerce data integrated management system of the present invention.

[0020] Figure 2 The present invention is a flowchart of the specific steps for obtaining the browsing path complexity index of each commodity in a multimodal e-commerce data integrated management system.

[0021] Figure 3 The present invention is a flowchart of the specific steps for obtaining the interaction strength index of each commodity in a multimodal e-commerce data integrated management system. DETAILED DESCRIPTION

[0022] The overall idea of ​​the problem in the embodiment of this application is as follows: Through the collaborative work of multiple modules, accurate sales forecasting and inventory scheduling can be achieved. First, in the data acquisition module, the system collects the product sales data and customer behavior data of the current cycle, such as purchase browsing time, return rate, page loading time, etc. Then, the data analysis module will analyze the customer flow index and customer sentiment fluctuation index of each product based on these data. This analysis relies on the comprehensive calculation of factors such as the complexity of the customer's browsing path, interaction intensity, and comment interaction. Subsequently, the predictive analysis module uses these indexes to predict the sales volume of each product in the next cycle, and schedules and replenishes the products through the inventory retrieval module to ensure that inventory and demand match.

[0023] See also Figure 1 The embodiment of the present invention provides a technical solution: a multimodal e-commerce data integrated management system, comprising: a data acquisition module, a data analysis module, a prediction analysis module, and an inventory retrieval module; the data acquisition module is used to obtain the current sales volume and commodity sales data of each commodity after the current cycle ends; the data analysis module is used to analyze the customer flow index and customer emotion fluctuation index of each commodity based on the commodity sales data of each commodity; the prediction analysis module is used to predict the sales forecast of each commodity in the next cycle based on the customer flow index and customer emotion fluctuation index of each commodity; the inventory retrieval module is used to schedule and replenish each commodity based on the sales forecast of the next cycle; wherein the specific formula for calculating the sales forecast of each commodity in the next cycle is as follows: ;in, For the next cycle The sales forecast of the product, For the The current sales volume of the product, For the The customer flow index of the product, is the customer flow impact coefficient stored in the database, For the The customer sentiment fluctuation index of the product, is the customer emotion fluctuation correction coefficient stored in the database, is the customer emotion fluctuation influence coefficient stored in the database, , , The number of commodities.

[0024] It should be explained that the customer flow impact coefficient stored in the database , Customer emotion fluctuation influence coefficient The specific steps to obtain are: First, customer mobility influence coefficient By analyzing customer interaction data (such as browsing frequency, number of clicks, dwell time, etc.) to measure the customer's interest in the product, this data can be obtained through the visit records or behavior logs of the product page. Next, the customer's emotional fluctuation influence coefficient Sentiment analysis of customer reviews is performed using sentiment analysis techniques (such as BERT, VADER, etc.) to obtain the sentiment score of each review (such as positive and negative sentiment fluctuations). These two data sources are combined with historical sales data and trained using regression analysis or machine learning models to obtain these two coefficients, which reflect the joint impact of customer mobility and sentiment fluctuations on sales forecasts.

[0025] Customer emotion fluctuation correction factor stored in the database The specific steps to obtain are: determine the correction effect on sales forecasts by analyzing the interaction between customer mood fluctuations and customer mobility. This coefficient reflects the impact of the relationship between customer mood fluctuations and their behaviors (such as browsing paths, dwell time) on sales. When customers' moods fluctuate, this fluctuation may interact with the customer's behavioral mobility (for example, whether they frequently browse products, click on other products, etc.), thereby changing their purchasing decisions. In order to obtain this correction coefficient, it is necessary to combine the customer's mood fluctuation data with the customer's behavioral data (such as browsing history, click volume, etc.), and estimate the correction coefficient by considering the interaction between the two through regression analysis or machine learning models to ensure that sales forecasts can more accurately reflect the impact of customer mood changes and behavioral dynamics on sales.

[0026] The product sales data includes the purchase browsing time of each customer, the return rate, the product page loading time of each customer when purchasing the product, and the text information of each comment, the number of comment forwarding times, and the number of comment likes.

[0027] Among them, purchase browsing time refers to the browsing time of customers on a certain product page, that is, the customer's attention to the product. Specifically, it is the total time that customers browse product pages, view product details, compare prices, view reviews, etc. when considering purchasing a certain product.

[0028] The return rate refers to the proportion of customers who return goods after purchase. It is usually an important indicator to measure the stability of customers' purchasing decisions. It can be obtained from order data and calculated based on the ratio of the number of returns of purchased goods to the total number of purchases.

[0029] The product page loading time when a customer purchases a product refers to the time required for the customer to click on the product link until the product page is fully displayed. It reflects the response speed of the web page or application, that is, the time interval between the user requesting the product page and the page being fully rendered and displayed. It can be obtained from server logs or data tracking tools (such as Google Analytics).

[0030] The text information of the comment, indicating the specific content of the comment.

[0031] The specific implementation example of calculating the sales forecast of each commodity in the next cycle is as follows. The following data is available: Including the current sales volume, customer flow index, and customer sentiment fluctuation index of five commodities. The specific data are shown in Table 1: Table 1 Examples of current sales volume, customer turnover index, and customer sentiment fluctuation index data for five commodities

[0032] In addition, the customer turnover impact coefficient stored in the database Approximately: 0.092; Customer emotion fluctuation correction factor stored in the database Approximately: 1.243; Customer emotion fluctuation impact coefficient stored in the database Approximately: 0.048; Substituting the above coefficients and the data in Table 1 into the specific formula for calculating the sales forecast of each commodity in the next cycle, we get: The sales forecast of the first commodity in the next cycle = 524*(1+0.092*0.78+0.048*(0.63) 1.243 )≈524*(1+0.092*0.78+0.048*0.563)≈576.

[0033] The sales forecast of the second commodity in the next cycle = 1132*(1+0.092*1.16+0.048*(0.52) 1.243 )≈1132*(1+0.092*1.16+0.048*0.443)≈1276.

[0034] The sales forecast of the third product in the next cycle = 742*(1+0.092*0.88+0.048*(0.71) 1.243 )≈742*(1+0.092*0.88+0.048*0.653)≈825.

[0035] The sales forecast of the fourth product in the next cycle = 312*(1+0.092*0.65+0.048*(0.76) 1.243 )≈312*(1+0.092*0.65+0.048*0.711)≈341.

[0036] The sales forecast of the fifth commodity in the next cycle = 1584*(1+0.092*1.48+0.048*(0.88) 1.243 )≈1584*(1+0.092*1.48+0.048*0.853)≈1865.

[0037] Specifically, the specific steps for analyzing the customer flow index of each commodity are as follows: read the purchase browsing time of each customer for each commodity and conduct a comprehensive analysis (i.e., mean analysis) to obtain the purchase browsing index of each commodity; read the product page loading time of each customer when purchasing each commodity and conduct a comprehensive analysis (i.e., mean analysis) to obtain the product page loading index of each commodity; obtain the browsing path complexity index, interaction intensity index, and historical purchase index of each commodity, and conduct a comprehensive analysis in combination with the purchase browsing index and product page loading index of each commodity to obtain the customer flow index of each commodity.

[0038] The historical purchase index is obtained by performing mean analysis on the historical purchase quantity of each customer.

[0039] The specific formula for calculating the customer flow index of each commodity is as follows: ;in, For the The customer flow index of the product, For the The purchase browsing index of the product, For the The browsing path complexity index of the product, is the browsing influence coefficient stored in the database, For the The interaction intensity index of the commodity, is the interaction influence coefficient stored in the database, For the The product page loading index of the product, It is a load adjustment factor stored in the database, used to prevent the denominator from being zero. is the loading influence coefficient stored in the database, For the The historical purchase index of the commodity, is the historical purchase impact coefficient stored in the database, , The number of commodities.

[0040] It should be explained that the browsing impact coefficient stored in the database , interaction influence coefficient , Loading influence coefficient , Historical purchase influence coefficient The specific steps to obtain are: First, browse the influence coefficient By analyzing the frequency of customers browsing and the time they stay on the product page, we can evaluate the customer's attention to the product. The data comes from the visit log of the product page, and the interaction impact coefficient The calculation is performed by analyzing customers’ comments, likes, shares and other interactive behaviors to reflect customers’ emotions and purchase intentions towards the product. These data can be obtained through product review systems or social media platforms and loaded with influence coefficients. By measuring the loading time of the product page, the impact of page loading speed on customer purchase decisions is evaluated. This is usually obtained through website performance monitoring tools. Finally, the historical purchase influence coefficient By analyzing the customer's historical purchase records to predict the customer's future purchasing behavior, estimates are made based on data such as the categories and frequency of goods purchased by the customer in the past. All of this data is modeled through regression analysis or machine learning models to derive four influence coefficients for optimizing sales forecasts.

[0041] In this implementation scheme, by performing mean analysis on multiple indicators such as the purchase browsing time, product page loading time, browsing path complexity, interaction intensity and historical purchase index of each product, the system can comprehensively evaluate the customer's behavior and preferences in all aspects of the purchase process. This comprehensive analysis can not only accurately reflect the customer's interest in the product and purchase intention, but also provide the enterprise with an accurate customer flow index to help them predict future sales trends. Specifically, the refined calculation of the browsing influence coefficient, the interaction influence coefficient, the loading influence coefficient and the historical purchase influence coefficient makes the sales forecast no longer rely on a single sales data or customer behavior, but is based on multi-dimensional customer activity data to fully capture customer demand changes and emotional fluctuations. This data-driven model can flexibly respond to fluctuations in the market and customer behavior, thereby improving the accuracy of the prediction results. By comprehensively considering these influencing factors, the system can not only optimize sales forecasts, but also help enterprises to reasonably dispatch inventory and avoid problems such as inventory backlog or out of stock. At the same time, accurate sales forecasts can also help enterprises adjust marketing strategies, optimize commodity supply chain management, improve operational efficiency and reduce costs. This refined analysis method effectively reduces the errors caused by ignoring customer dynamic behavior in traditional sales forecasts, and provides enterprises with more competitive decision support in the fierce market competition.

[0042] Specifically, Figure 2As shown in the figure, the specific steps for obtaining the browsing path complexity index of each product are as follows: for each product, obtain the page jump times, page stay time, and click depth value of each customer when purchasing the product, and perform comprehensive analysis (i.e., mean analysis) to obtain the page jump index, page stay index, and click depth index of each product; perform comprehensive analysis on the page jump index, page stay index, and click depth index of each product to obtain the browsing path complexity index of each product.

[0043] Among them, the page jump count value indicates the number of times a customer jumps from one product page to another. Each jump indicates that a customer leaves one page and enters another page, which may be to browse other products, view related information, or return to the homepage, etc. The number of customer jumps can reflect the complexity of the customer's browsing behavior. If customers jump frequently between different pages, it means that they compare multiple products or information in a short period of time, showing a strong interest or exploration. A large number of jumps usually means that the customer's browsing path is more complicated, which can easily lead to a longer purchase decision time. For example: customers jump from product A page to product B page, and then to product C page. Each jump indicates that they are interested in different products or pages.

[0044] The page dwell time value indicates the time a customer stays on a certain page, that is, the length of time the customer browses, views, and interacts on the page. The dwell time reflects the customer's interest in the page. Generally speaking, the longer the dwell time, the more interested the customer is in the content or product of the page. A shorter dwell time may indicate that the customer is not very interested in the content of the page. For example, if a customer stays on the page of product A for 5 minutes, it may mean that they are looking at the product description, comments, or other related content in detail. On the contrary, if the customer only stays on the page for 30 seconds, it may mean that they just browsed it briefly and did not have a deep understanding of the product.

[0045] The click depth value indicates the depth of the level of clicks made by customers on the product page, that is, how many levels of sub-pages or detailed pages the customer enters when browsing the product page. For example, from the basic information page of the product, click to enter the comments, specifications, accessories and other pages. The more levels there are, the deeper the click depth is. The click depth usually indicates the intensity of the customer's interest in the product. The deeper the level of clicks made by customers, the more invested they are in exploring the product. On the contrary, if the customer only stays on the surface information of the product page and does not further understand other details, the click depth is shallow. For example, the customer enters the detailed information page from the main page of product A, and then clicks to view comments, related products and other sub-pages. At this time, the click depth is 3 levels. If the customer only stays on the main page of product A and does not click to enter other pages, the click depth is 1 level.

[0046] The specific formula for calculating the browsing path complexity index of each product is as follows: ;in, For the The browsing path complexity index of the product, For the The page jump index of the product, is the page jump impact coefficient stored in the database, For the The page dwell index of the product, is the page stay adjustment coefficient stored in the database. is the page stay influence coefficient stored in the database, For the The click depth index of the purchase of the product. is the click depth adjustment coefficient stored in the database, is the click depth influence coefficient stored in the database, , The number of commodities.

[0047] It needs to be explained that the page stay adjustment coefficient stored in the database , click depth adjustment coefficient The specific steps to obtain are: First, the page stay adjustment coefficient It measures the impact of the time customers spend on a product page on sales forecasts. To obtain this coefficient, you need to collect data on the time customers spend on a product page through website behavior analysis tools (such as Google Analytics or heat map analysis tools). Then, analyze the relationship between the customer's stay time and their final purchase behavior. For example, are customers who stay longer more likely to buy the product? By combining the customer's stay time with sales data through regression analysis or machine learning models, you can calculate this coefficient. Click depth adjustment coefficient The coefficient measures the moderating effect of the customer's click depth on the product page (i.e., how many layers of pages the customer clicked and how many product details the customer viewed) on the sales forecast. The steps to obtain this coefficient are similar. First, the customer's click depth on the product page needs to be analyzed through the customer's click stream data. The data can be obtained through a click analysis tool to calculate the page level visited by the customer and the number of clicks on each page. Then, by combining it with sales data, using regression analysis or machine learning models, the moderating effect of click depth on sales can be estimated.

[0048] Page jump impact coefficient stored in the database , Page dwell influence coefficient , Click depth influence coefficient The specific steps to obtain are: page jump influence coefficient Indicates the impact of the frequency of customers jumping between different product pages on sales. To obtain this coefficient, it is necessary to analyze the behavior of customers jumping from one product page to another. Through the website's user behavior log or click stream analysis tools (such as Google Analytics), the number of customers jumping between different product pages, the jump path and the jump order are counted. Through regression analysis or machine learning models, the jump data is combined with historical sales data to calculate the impact of page jumps on sales and the page stay impact coefficient. Measures the impact of customer dwell time on product pages on sales forecasts. The steps to obtain this coefficient are similar to the page dwell adjustment coefficient. The data comes from the customer's dwell time on the product page, which can be obtained through heat map analysis tools or behavior tracking tools. Then, through regression analysis or machine learning modeling, combined with historical sales data, calculate the impact of dwell time on sales and click depth impact coefficient. It reflects the impact of the depth of customer browsing on the product page on sales forecasts. To obtain this coefficient, it is necessary to collect customer clickstream data and analyze the depth of customer clicks on the product page (such as how many levels of detail pages are viewed and which sub-pages are clicked). The data is obtained through click analysis tools and combined with sales data, using regression analysis or machine learning models to estimate the direct impact of click depth on sales.

[0049] In this implementation scheme, by comprehensively analyzing the browsing behavior of customers on product pages, it is possible to more accurately reflect the customer's interest in the product and their purchasing intention. First, the three indicators of page jump count, dwell time, and click depth capture the customer's behavioral characteristics from different dimensions, helping to more comprehensively evaluate the customer's activity and participation in the product browsing process. By performing mean analysis on these behavioral data and combining them with the influence coefficients stored in the database, the browsing path complexity index of each product can be accurately calculated, thereby improving the accuracy of sales forecasts. Specifically, the customer's in-depth browsing, dwell time, and jump frequency on the product page directly reflect the complexity of their purchase decisions. These are dynamically incorporated into the prediction model, which helps to predict demand fluctuations for different products and optimize inventory management. In addition, calculating the page jump impact coefficient, page dwell impact coefficient, and click depth impact coefficient can further help companies understand customer behavior patterns, dynamically adjust sales forecasts and product supply based on these factors, effectively avoid overstocking or out-of-stock, improve operational efficiency, and reduce inventory costs.

[0050] Specifically, Figure 3As shown in , the specific steps for obtaining the interaction intensity index of each product are as follows: for each product, obtain the number of product clicks, product comments, product likes, and product shares of each customer, and conduct comprehensive analysis (i.e., mean analysis) to obtain the product click index, product comment index, product like index, and product sharing index of each product; conduct comprehensive analysis on the product click index, product comment index, product like index, and product sharing index of each product to obtain the interaction intensity index of each product.

[0051] Among them, the number of product clicks refers to the number of times customers click on the product page, reflecting the customer's interest in the product. Each click represents the customer's attention or exploration of the product. A large number of clicks usually means that the customer has a high interest in the product. The number of clicks is a direct way to measure the customer's interaction with the product page. If a customer frequently clicks on different parts of the product page (such as pictures, descriptions, prices, etc.), it means that the customer is gaining a deeper understanding of the product.

[0052] The number of product reviews indicates the number of comments left by customers on the product page, indicating the customer's emotions and involvement in the product. Reviews reflect the customer's deep concern for the product, especially the emotional expression or evaluation of the product. The more reviews, the deeper the customer's interaction with the product, which is usually accompanied by a higher emotional investment. For example, when a customer leaves a comment on a product page, it may be an evaluation of the product quality, a sharing of the purchase experience, or a question about the product features.

[0053] The number of product likes indicates the number of times customers like the product page, indicating the customers' recognition and support for the product. Likes are usually a kind of positive feedback from customers to the product. The more likes, the better the customers' evaluation of the product, reflecting the customers' positive emotions. For example, customers like the content of the product page or the comments of other customers, indicating their liking or recognition of the product.

[0054] The number of times a product is shared indicates the number of times a customer shares the product with others (for example, sharing it on social media, sending it to friends), reflecting the customer's recommendation behavior. The more times it is shared, the higher the customer's interest and trust. For example, if a customer shares a link to a product page on a social network or sends it to a friend via email, it indicates the customer's intention to recommend and promote the product.

[0055] The specific formula for calculating the interaction intensity index of each product is as follows: ;in, For the The interaction intensity index of the commodity, For the The product click index of the product, For the The product review index of the product, is the impact coefficient of the comments stored in the database, For the The product like index of the product, The likes stored in the database refer to the influence coefficient. For the The commodity sharing index of the commodity, is the sharing influence coefficient stored in the database, , , The number of commodities.

[0056] It should be explained that the comment influence coefficient stored in the database , Like refers to the influence coefficient , Shared Impact Coefficient The specific steps to obtain are: comment influence coefficient Measures the impact of customer reviews on product sales forecasts. To obtain this coefficient, you first need to analyze the sentiment score of each review through sentiment analysis (for example, using sentiment analysis tools such as BERT or VADER). Then, count the number of reviews and sentiment scores for each product, combine them with historical sales data, and calculate the impact of reviews on sales through regression analysis or machine learning models. It measures the impact of customers’ likes on product sales forecasts. The steps to obtain this coefficient are to analyze the likes data of customers on the product page, that is, to count the number of likes for each product. This information can be obtained through the likes data on e-commerce platforms or social media. Through regression analysis or machine learning modeling, the likes data is combined with sales data to calculate the impact of likes on sales forecasts and share the impact coefficient. It measures the impact of customers sharing product information on social platforms on sales forecasts. To obtain this coefficient, it is necessary to analyze customers' product sharing behavior, that is, the frequency of customers sharing products on social platforms or through other channels. The data can usually be obtained from the sharing records of social media platforms or e-commerce platforms. Then, combined with sales data, regression analysis or machine learning methods are used to calculate the impact of sharing behavior on sales.

[0057] In this implementation scheme, by comprehensively analyzing customers' clicking behavior, comment interactions, likes behavior and sharing behavior, it is possible to comprehensively evaluate customers' interests and emotional investment, thereby more accurately predicting the sales trend of goods. By average analyzing the number of clicks, comments, likes and shares of each product, the system can quantify customers' multi-dimensional interactions on goods, reflecting the customer's interest intensity and purchasing intention in the goods. Combined with the comment, like and share influence coefficients in the database, the system can evaluate the direct impact of these interactive behaviors on sales volume, and then adjust sales forecasts and inventory management. For example, if customers frequently like or share goods, it means that the product has a high degree of recognition and recommendation intention among customers, which can effectively promote the sales of goods. By accurately calculating the interaction intensity index, the system can not only improve the accuracy of sales forecasts, but also adjust marketing strategies and inventory scheduling according to customer participation, reduce excess inventory or out-of-stock situations, and optimize supply chain management, thereby improving the company's operating efficiency and profitability.

[0058] Specifically, the specific steps for analyzing the customer emotion fluctuation index of each product are as follows: read the text information of each comment of each product, and perform sentiment analysis to obtain the sentiment score of each comment of each product, and perform comprehensive analysis (i.e., mean analysis) to obtain the sentiment score index of each product; read the return rate of each customer of each product, and perform comprehensive analysis (i.e., mean analysis) to obtain the return index of each product; read the number of comment forwardings and comment likes of each comment of each product, and perform comprehensive analysis (first perform weighted analysis on the number of comment forwardings and comment likes, and then perform mean analysis on the weighted analysis results of each comment after the weighted analysis) to obtain the comment interaction index of each product; obtain the comment time difference index of each product, and perform comprehensive analysis on the sentiment score index, comment interaction index, and return index of each product to obtain the customer emotion fluctuation index of each product; wherein, the comment time difference index is specifically obtained by performing mean analysis on the time difference between each comment.

[0059] The specific formula for calculating the customer sentiment fluctuation index of each product is as follows: ;in, For the The customer sentiment fluctuation index of the product, For the The sentiment score index of the product. For the The comment interaction index of the product. For the The time difference index of reviews of products. is the time difference impact coefficient of the comments stored in the database, For the The return index of the product, is the return impact coefficient stored in the database, , The number of commodities.

[0060] It needs to be explained that the time difference impact coefficient of the comments stored in the database The specific steps to obtain are: This represents the impact of the time difference between customer comments on sales forecasts. To obtain this coefficient, we first need to analyze the timestamps of customer comments, that is, the time when each comment is published. Then, calculate the time difference between comments on the same product, and count the interval time between comments. This data can usually be extracted from the comment timestamps in the database. After that, combined with sales data, the comment time difference is associated with sales volume through regression analysis or machine learning models, so as to obtain the impact of the comment time difference on sales forecasts.

[0061] Return impact coefficient stored in the database The specific steps to obtain are: This coefficient reflects the impact of customer return behavior on product sales forecasts. To obtain this coefficient, it is necessary to analyze customer return data, including the return rate of each product, the frequency of returns, and the type of returned products. First, collect the return records of the products and calculate the return rate of each product. Then, analyze the relationship between return behavior and sales volume. Through regression analysis or machine learning modeling, combine the return data with historical sales data to derive the return impact coefficient, which represents the corrective effect of return behavior on sales forecasts.

[0062] In this implementation scheme, by comprehensively analyzing the customer's comment sentiment, comment interaction, return behavior and comment time difference, it is possible to more accurately measure the customer's emotional fluctuations and their impact on sales. By performing sentiment analysis on each comment, combined with the customer's return rate, comment forwarding and number of likes, the system can fully understand the customer's true feelings and involvement in the product. In addition, through the calculation of the comment time difference index, the system can further capture the changing trends of customer emotions, especially how emotional fluctuations affect purchasing decisions between comments in different time periods. These multi-dimensional emotional data help eliminate the errors caused by ignoring emotional fluctuations in traditional sales forecasts, make the forecast results more in line with actual market demand, and improve the accuracy of sales forecasts. Combined with the return impact coefficient in the database, the system can further optimize sales forecasts, reduce inventory problems caused by customer emotional fluctuations and return behavior, avoid product backlogs or out-of-stocks, and help companies achieve more efficient inventory management and more accurate marketing strategies, thereby improving the overall operational efficiency and profitability of the company.

[0063] Specifically, the text information of each comment of each product is read, and sentiment analysis is performed to obtain the sentiment score of each comment of each product. The specific steps are as follows: based on natural language processing technology, text processing is performed on the text information of each comment of each product; each comment of each product after text processing is input into the pre-trained analysis model for sentiment analysis to obtain the sentiment score of each comment of each product.

[0064] The analysis model is specifically the BERT model, which is built based on the Transformer encoder architecture and includes the following components: Input embedding layer: The input of BERT is a tokenized text sequence. In BERT, each word is first encoded through the WordPiece embedding layer to represent each word as a vector.

[0065] BERT’s input consists of three parts: Token Embeddings: Word embeddings, which represent the vector representation of each input word or token.

[0066] Segment Embeddings: Represents two sentences of a text (for question answering tasks, etc.), used to distinguish between different sentences A and B.

[0067] Position Embeddings: In order to handle the order of words in the sequence, BERT introduces position embeddings to indicate the position of each word in the sentence.

[0068] These input embedding vectors are added together to form the final input representation.

[0069] Multi-layer Transformer encoder, which captures the relationship between words through the self-attention mechanism. BERT's Transformer encoder layer stacks multiple layers, and the depth of the model can vary depending on the parameters, for example: BERT-base: 12-layer Transformer encoder.

[0070] BERT-large: 24-layer Transformer encoder.

[0071] Each layer of Transformer encoder consists of the following components: Self-Attention: Through the self-attention mechanism, the model can adjust the representation of each word according to the relationship between words. The self-attention mechanism allows the model to consider contextual information at the same time, rather than just processing text from left to right or right to left.

[0072] Feedforward Neural Network: The output of each layer of Transformer is further processed by a feedforward neural network.

[0073] Layer Normalization: Each layer uses layer normalization to help speed up training and stabilize the training process.

[0074] Residual connections: In each layer, residual connections are used to avoid the gradient vanishing problem, which helps in training deep networks.

[0075] Output layer: For classification tasks such as sentiment analysis, BERT outputs a sentiment score representing the entire sentence or text.

[0076] Among them, the pre-training steps of the BERT model are as follows: The review text of each product is collected and organized into a data set, where each review includes the review content and the corresponding product ID.

[0077] Each review was cleaned and normalized; Converting to an input format acceptable to the BERT model mainly includes: Token Embeddings: Each word / subword is converted into a vector representation.

[0078] Segment Embeddings: used to represent the relationship between two sentences (for example, required in question answering tasks).

[0079] Position Embeddings: Represents the position information of each word.

[0080] Convert to model input format: Convert the text into the input format accepted by BERT through tokenization, including: Input IDs: Map each word or subword after tokenization to an ID in the vocabulary.

[0081] Attention Mask: Indicates which tokens are actual inputs and which are paddings, helping the model identify valid inputs.

[0082] Token Type IDs: distinguish sentence A from sentence B, which can usually be omitted in single-sentence sentiment analysis.

[0083] The pre-training of the BERT model includes two core tasks: masked language model and next sentence prediction. These pre-training steps enable BERT to understand the relationship between words and sentences and accurately handle sentiment analysis tasks.

[0084] Masked Language Model: Task: For the input text, BERT randomly masks 15% of the words in the input sequence and lets the model predict these masked words.

[0085] Goal: In this way, BERT learns the relationship between contexts and understands the meaning of each word.

[0086] Application: In sentiment analysis tasks, BERT learns how to understand sentiment and semantic relationships in context through MLM tasks, especially sentiment changes in cases such as polysemous words and long texts.

[0087] Next sentence prediction: Task: BERT is given two sentences during pre-training, and the task is to determine whether the second sentence is the next sentence of the first sentence.

[0088] Objective: BERT learns semantic relationships between sentences, which helps it understand the context or connections between paragraphs in sentiment analysis tasks.

[0089] Application: For sentiment analysis, BERT can better understand the changes in sentiment across sentences and paragraphs, especially in the reasoning of sentiment.

[0090] Fine-tuning the BERT model: After completing pre-training, BERT needs to be fine-tuned to adapt to specific tasks. Sentiment analysis is a text classification task. During the fine-tuning stage, BERT will be adjusted to specifically solve sentiment classification problems.

[0091] Set up the sentiment analysis task: Goal: Classify the sentiment of each product review, that is, directly generate a sentiment score (e.g., a rating from 0 to 1).

[0092] Fine-tuning process: Data preparation: Classify product reviews into different sentiment categories (such as positive, negative, or neutral) through manual or automatic labeling. For example, you can use a labeled sentiment dataset (such as the Amazon product review dataset) for training.

[0093] Input data: Each comment is passed as input to the BERT model and is tokenized, converted to input ID, etc.

[0094] Fine-tuning training: Use the labeled sentiment tags to train BERT and update the weights of the BERT model. The goal is to minimize the loss function of the sentiment classification task (e.g., cross-entropy loss function).

[0095] Output layer: The output layer of BERT is usually a fully connected layer used to convert the output of BERT into a probability distribution for sentiment classification.

[0096] In this implementation, by applying the BERT model for sentiment analysis, it is possible to accurately analyze the sentiment scores of each product review, and then evaluate the emotional fluctuations of customers towards the product. This sentiment analysis is based on natural language processing technology and can deeply understand the context and sentiment details in the review, especially the sentiment changes in the case of polysemous words and long texts. Compared with traditional sentiment analysis methods, BERT can fully consider the context information through the self-attention mechanism, making the sentiment scores more accurate and reliable. In addition, the pre-training and fine-tuning processes of BERT enable the model to learn the subtle changes in customer sentiment from a large number of product reviews and make accurate predictions in different products and scenarios. Through these sentiment scores, the system can better understand the emotional tendencies of customers, optimize sales forecasts and inventory management, thereby improving customer satisfaction, reducing return rates, and enhancing the market competitiveness of enterprises.

[0097] Specifically, based on natural language processing technology, the specific steps for text processing of the text information of each review of each product are as follows: Remove noise from the text information of each review of each product, that is, delete irrelevant characters (such as special symbols, HTML tags, URLs, etc.) and remove stop words (such as "de", "he", etc.); After removing noise, perform word segmentation on the text information of each review of each product, that is, segment the review text. Usually, the jieba library is used for Chinese, and the nltk or spaCy library is used for English. In Chinese sentiment analysis, word segmentation is very important because Chinese has no obvious word boundaries; After word segmentation, perform word removal on the text information of each review of each product, that is, remove low-frequency words or irrelevant vocabulary according to word frequency statistics or business requirements to improve the accuracy of analysis.

[0098] In this implementation scheme, the accuracy and efficiency of sentiment analysis can be effectively improved through refined text processing steps. First, noise removal ensures the cleanliness and high quality of the comment text, and avoids irrelevant information such as special symbols, HTML tags and URLs interfering with the analysis process. The step of removing stop words is equally important, because stop words often have no practical meaning in sentiment analysis. After removal, we can better focus on effective information. Secondly, word segmentation is a key step in Chinese sentiment analysis. Because Chinese text has no obvious word boundaries, tools such as word segmentation can accurately split the text into meaningful words, thereby extracting the customer's emotional tendencies. Finally, by removing low-frequency words or irrelevant words through word removal, the efficiency of the model can be further improved, noise can be reduced, and the accuracy of the analysis results can be improved. Through the comprehensive application of these steps, emotional information can be better extracted from a large number of comments, helping companies to achieve accurate sentiment analysis and sales forecasts.

[0099] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0100] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A multimodal e-commerce data integrated management system, characterized in that: include: Data acquisition module, data analysis module, forecasting analysis module, inventory retrieval module; The data acquisition module is used to acquire the current sales volume and commodity sales data of each commodity after the current cycle ends; The data analysis module is used to analyze the customer flow index and customer emotion fluctuation index of each commodity based on the commodity sales data of each commodity; The forecast analysis module is used to forecast the sales volume of each commodity in the next cycle based on the customer flow index and customer emotion fluctuation index of each commodity; The inventory retrieval module is used to schedule and replenish each commodity based on the sales forecast for the next cycle; The specific formula for calculating the sales forecast for each commodity in the next cycle is as follows: ； in, For the next cycle The sales forecast of the product, , , The first The current sales volume of the product, customer turnover index, customer sentiment fluctuation index, , , They are the customer flow impact coefficient, customer mood fluctuation impact coefficient, and customer mood fluctuation correction coefficient stored in the database, respectively. , , The number of commodities.

2. The multimodal e-commerce data integrated management system according to claim 1, characterized in that: The product sales data includes the purchase browsing time of each customer, the return rate, the product page loading time of each customer when purchasing the product, and the text information of each comment, the number of comment forwarding times, and the number of comment likes.

3. The multimodal e-commerce data integrated management system according to claim 2, characterized in that: The specific steps to analyze the customer flow index of each product are as follows: Read the purchase browsing time of each customer for each product and conduct a comprehensive analysis to obtain the purchase browsing index of each product; Read the product page loading time of each customer when purchasing each product and conduct a comprehensive analysis to obtain the product page loading index of each product; Obtain the browsing path complexity index, interaction intensity index, and historical purchase index of each product, and conduct a comprehensive analysis based on the purchase browsing index and product page loading index of each product to obtain the customer flow index of each product.

4. The multimodal e-commerce data integrated management system according to claim 3, characterized in that: The specific formula for calculating the customer flow index of each commodity is as follows: ； in, , , , , , The first The customer flow index, purchase browsing index, browsing path complexity index, interaction intensity index, product page loading index, and historical purchase index of the products are , , , They are the browsing influence coefficient, interactive influence coefficient, loading influence coefficient, and historical purchase influence coefficient stored in the database. is the load adjustment factor stored in the database, , The number of commodities.

5. The multimodal e-commerce data integrated management system according to claim 3, characterized in that: The specific steps to obtain the browsing path complexity index of each product are as follows: For each product, obtain the page jump count, page stay time, and click depth value of each customer when purchasing the product, and conduct comprehensive analysis to obtain the page jump index, page stay index, and click depth index of each product; A comprehensive analysis is conducted on the page jump index, page dwell index, and click depth index of each product to obtain the browsing path complexity index of each product.

6. The multimodal e-commerce data integrated management system according to claim 3, characterized in that: The specific steps to obtain the interaction strength index of each product are as follows: For each product, obtain the number of product clicks, product comments, product likes, and product shares of each customer, and conduct comprehensive analysis to obtain the product click index, product comment index, product like index, and product share index of each product; A comprehensive analysis is conducted on the product click index, product comment index, product like index, and product sharing index of each product to obtain the interaction intensity index of each product.

7. The multimodal e-commerce data integrated management system according to claim 2, characterized in that: The specific steps to analyze the customer sentiment fluctuation index of each product are as follows: Read the text information of each comment of each product, perform sentiment analysis, obtain the sentiment score of each comment of each product, and perform comprehensive analysis to obtain the sentiment score index of each product; Read the return rate of each customer for each product and conduct a comprehensive analysis to obtain the return index of each product; Read the number of reposts and likes of each comment for each product, and conduct a comprehensive analysis to obtain the comment interaction index of each product; The review time difference index of each product is obtained, and a comprehensive analysis is performed based on the sentiment score index, review interaction index, and return index of each product to obtain the customer sentiment fluctuation index of each product.

8. The multimodal e-commerce data integrated management system according to claim 7, characterized in that: The specific formula for calculating the customer sentiment fluctuation index of each product is as follows: ； in, , , , , The first The customer sentiment fluctuation index, sentiment score index, review interaction index, review time difference index, and return index of the product. , They are the time difference impact coefficient and return impact coefficient of the comments stored in the database. , The number of commodities.

9. The multimodal e-commerce data integrated management system according to claim 7, characterized in that: The specific steps for reading the text information of each comment of each product and performing sentiment analysis to obtain the sentiment score of each comment of each product are as follows: Based on natural language processing technology, the text information of each review of each product is processed separately; After text processing, each comment of each product is input into the pre-trained analysis model for sentiment analysis to obtain the sentiment score of each comment of each product.

10. The multimodal e-commerce data integrated management system according to claim 9, characterized in that: Based on natural language processing technology, the specific steps for text processing of each comment of each product are as follows: Perform noise removal on the text information of each review of each product; After removing the noise, the text information of each review of each product is segmented separately; After the word segmentation process, the text information of each review of each product is de-worded.

Images