Industrial chain resource data reconstruction and integration system based on digital transformation enterprise

Abstract

The invention discloses an industrial chain resource data reconstruction and integration system based on a digital transformation enterprise, and relates to the field of artificial intelligence and enterprise service platforms. The system comprises a user behavior module, a user portrait module, a resource knowledge base module, an intelligent display module, a collaborative application module and a self-learning optimization module. The user behavior module captures user behaviors and constructs a behavior feature set after judgment of a preset rule; the user portrait module processes and analyzes the behavior feature set, and generates and dynamically updates a user interest portrait; the intelligent display module is used for realizing the matching and display of personalized contents by calculating the fusion similarity between a user interest vector and a resource knowledge base content vector; the collaborative application module automatically processes user applications and pushes the user applications to related departments; the self-learning optimization module optimizes various parameters and algorithms of the system. According to the invention, integration and automation of data integration, intelligent recommendation and business collaboration are realized.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence and enterprise service platforms, and belongs to a system for reconstructing and integrating supply chain resource data of enterprises undergoing digital transformation. Background Technology

[0002] As digital transformation deepens, enterprises face severe challenges when integrating upstream and downstream resources across the industry chain, including large data volumes, dispersed sources, and heterogeneous structures. Traditional enterprise data management systems or content recommendation platforms often rely on static rules or simple keyword matching technologies, lacking the ability to dynamically perceive and analyze users' real-time, in-depth behavioral intentions. This makes it difficult for the system to accurately distinguish between users' casual browsing behavior and genuine interests, resulting in delayed and inaccurate updates to the generated user profiles.

[0003] Currently, incubator websites or platforms typically showcase company resources in a centralized manner, but the display format is mostly static categorization, lacking personalized recommendation mechanisms. Different types of users (entrepreneurs, investors, partner institutions) have significantly different needs during their visits, and the system cannot dynamically adjust content based on behavior, resulting in low resource matching, fragmented application processes, and a poor user experience.

[0004] Furthermore, in terms of resource recommendation, traditional systems often fail to effectively integrate users' long-term interests with their short-term intentions, resulting in content recommendations that are not highly relevant and lack accuracy, leading to low utilization of valuable industry chain resources. In business collaboration, user-submitted cooperation or service requests rely heavily on manual identification, allocation, and follow-up, resulting in slow response times and low process efficiency, failing to meet the core requirements of digital transformation for real-time business response and intelligent processes. Summary of the Invention

[0005] The purpose of this invention is to provide a system for reconstructing and integrating supply chain resource data for enterprises undergoing digital transformation, in order to solve the problems mentioned in the background art.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a system for reconstructing and integrating supply chain resource data for digitally transformed enterprises, comprising a user behavior module, a resource knowledge base module, a user profile module, and an intelligent display module; The user behavior module is used by the user behavior module to obtain user behavior data and establish a set of behavior features. The user profiling module is used to process all data based on behavioral feature sets and analyze the user's interest weights. The resource knowledge base module is used to store, manage, and label all recommendable content resources in the system; it is a structured content database. The intelligent display module matches web page content that is highly similar to the user's interests in real time based on user profiles and resource knowledge base; The user behavior module includes an initial interest analysis unit, a click count unit, a stay duration timing unit, a scroll line count recording unit, a keyword extraction unit, and an application monitoring module; The initial interest analysis unit includes that when a user first enters the website, the initial interest analysis unit sends a pop-up window instruction, and an interest analysis questionnaire pops up on the user side. After the user fills out the interest analysis questionnaire, the initial interest analysis unit screens out the content with a high similarity to the user's interest in the resource knowledge base module and displays it to the user as the initial web page; The user behavior module initializes a preset pre-interest pool and a non-interest pool, incorporates the data determined to be of pre-interest into the pre-interest pool, and incorporates the data determined to be non-interest into the non-interest pool; When the user performs a click operation, the click count unit obtains user data C and counts the number of clicks; When C >= Cmin times, it is determined that the user is interested in the content, marked as pre-interest, and at the same time obtain d; when C < Cmin times, it is determined that the user is not interested in the content, marked as non-interest; C is the number of user clicks, and Cmin is the preset value of the number of clicks; When the user stays on the web page, the stay duration timing unit obtains user data T and determines the interest concentration based on the user's stay duration on the page; T is the user's stay duration, and Tmin is the preset value of the stay duration Tmin; Among them, the web page is divided into a main page and sub-pages, and the determination is further refined by placing the user's stay duration T in different page scenarios; When T > Tmin seconds, it is determined that the user is interested in the content, but the specific content cannot be determined, so only obtain d; When T > Tmin seconds, it is determined that the user is interested in the content, marked as pre-interest, and at the same time obtain d; when T < Tmin seconds, it is determined that the user is not interested in the content, marked as non-interest; Calculate T total = S1211 + S1221 and fuse all d data; T total is the total of the user's stay duration; When the user uses the scroll wheel to roll back the page, the scroll line count recording unit reads the data S and records the number of lines; S is the number of user scroll lines, and Smin is the preset value of the number of scroll lines; When S > Smin lines and there is no rollback, it is determined that the user is not interested in the content, marked as non-interest; When S > Smin lines, but there is a scroll, determine whether the stay duration meets the pre-interest preset value through the stay duration timing unit; When S < Smin lines and T > Tmin, it is determined that the user is interested in the content, marked as pre-interest, and at the same time obtain d; When a user searches for content through a website, the keyword extraction unit obtains user data t for keyword extraction; t is the keyword.

[0007] According to the above technical solution, the user behavior module further includes: When T > Tmin and C > Cmin and S > Smin, it is determined as pre - attention, and d is obtained simultaneously; When T > Tmin and C < Cmin and S > Smin, it is determined as pre - attention, and d is obtained simultaneously; When T > Tmin and C < Cmin and S < Smin, it is determined as pre - attention, and d is obtained simultaneously; When T < Tmin and C > Cmin and S > Smin, it is determined as non - attention; When T < Tmin and C < Cmin and S < Smin, it is determined as non - attention; When the user clicks the web page application button, the application monitoring module reads the user's click behavior and directly determines this behavior of the user as pre - attention; Based on the pre - attention pool and non - attention pool, a behavior feature set is established, and a feature value is added to each data.

[0008] According to the above technical solution, the user portrait module includes a numerical classification unit, a numerical analysis unit, and a type analysis unit; The user portrait module extracts all feature values from the behavior feature set, performs comparison processing on each feature value, and converts the feature values into a unified numerical range; The numerical classification unit divides the data analysis in the behavior feature set while classifying, and classifies the click count, stay duration, and scroll lines into numerical features; classifies keywords and application behaviors into categorical features; When processing numerical features, the numerical analysis unit scales the numerical features to the range of 0 to 1; When processing categorical features, the type analysis unit converts the categorical features into a numerical vector that can reflect their importance.

[0009] According to the above technical solution, the user portrait module calculates the initial weight of each interest category based on the processed numerical features and categorical features; First, a weight coefficient is assigned to each feature, and all coefficients form a weight matrix, which is dynamically provided by the self - learning optimization module; Second, calculate the score of each interest category. The score is the sum of the products of all user feature values and their corresponding weights, plus an adjustment term; Then, dynamically update the interest weights according to the user's historical behavior data; During this period, the interest weights are updated regularly. The new weights are jointly determined by the old weights at the previous moment and the current new scores, with the old weights accounting for a larger proportion to achieve the smooth evolution of the weights. Finally, the user portrait is output, which includes the user's interest weight vector and a list of the main interest categories.

[0010] According to the above technical solution, the intelligent display and recommendation module includes a real-time calculation unit and a real-time monitoring unit. The intelligent display module obtains the user interest weight vector U from the user portrait module and simultaneously obtains the feature vectors of all web page contents from the resource knowledge base module. ; The resource knowledge base module converts unstructured content resources into feature vectors that can be understood and calculated by machines. ; The intelligent display module obtains the current and recent associated data d from the user portrait module in real time and constructs a dynamic interest context vector D. The real-time calculation unit analyzes the data d and extracts the mouse hover hotspots, the in-page search term t, and the rolled-back paragraphs in the current user session.

[0011] According to the above technical solution, when the real-time calculation unit analyzes the mouse hover hotspots in the data d: First, the web page interface currently accessed by the user is divided into n logical blocks, and a unique area identifier is assigned to each block. At the same time, the system records the core content elements included in each block, and these core elements include text paragraphs, pictures, video players, data tables, and hyperlinks. When the user hovers the mouse over a certain block, the hover duration is recorded and the content of the page of that block is obtained, and the recording stops when leaving the block. The intelligent display and recommendation module continuously monitors the hover duration of the user's mouse in each block through the real-time monitoring unit; when the user's mouse enters a block, the hover duration is recorded; the recording stops when leaving the block; the original hover duration is recorded as Traw, and at the same time, an effective hover determination is performed to filter out invalid hover behaviors: Set the minimum effective hover preset value Tfast = 0.3 seconds; if Traw < Tfast, it is determined that the mouse moves too fast, and the hover duration of this time is not included in the cumulative data. Set the maximum effective hover preset value Tslow = 30 seconds; if Traw > Tslow, it is determined that the mouse moves too slowly, and the system automatically divides the hover duration, and only records the first Tslow duration as the effective value, that is, Tvalid = Tslow, and the excess part is regarded as invalid. Convert the hover duration into a high-frequency timestamp, continuously collect the area identifier where the mouse cursor is located and its continuous停留时间,建立区块-时长-内容的关联; It should be noted that there is an unclear expression "持续停留时间,建立区块-时长-内容的关联" in the original text. I have translated it as best as possible based on the context, but this part may need further clarification for a more accurate translation. Subsequently, the hover durations of multiple hovers in the same region of each block are accumulated to obtain the Thover of that block; a normalization algorithm is used to map the Thover to a block Wheat in the range [0,1]. The longer the hover duration of a block, the higher its Wheat; at the same time, the hot zone is associated with the specific web page content carried by the block. Thover is the total effective hover duration, Tmin is the minimum hover duration of all blocks, Tmax is the maximum hover duration of all blocks, and Wheat is the heat weight value; when a block contains two tags at the same time, each of the two tags receives a portion of Wheat.

[0012] According to the above technical solution, the real-time monitoring unit also includes continuously recording and analyzing the mouse's movement speed, trajectory patterns, and interaction patterns within the webpage to assess the user's proficiency in computer operation. Mouse movement speed is determined by acquiring position coordinates in real time, calculating the average displacement distance per unit time, and combining it with the number of clicks C, hover duration, and hotspot distribution. When the system detects that a user's mouse movement speed is consistently low, the number of clicks is infrequent, and the mouse hotspot distribution is uneven, the user is determined to be an inexperienced operator. When the mouse moves quickly, clicks frequently, and the hot zone is evenly distributed, the system determines that the user is a skilled operator. The heat weight values ​​of each block and their associated content are integrated into the dynamic interest context vector D; The real-time calculation unit parses data d and extracts the current user's rollback behavior. First, the system defines and captures the user's rollback behavior, defining a complete reading session as starting from page loading and ending when the user jumps away or closes the page; In the current session page, the system continuously monitors the scroll bar position. When it detects that the scroll direction changes from downward to upward, it marks the start of a rollback behavior and records the starting position of the scroll bar when the rollback starts and the target position of the scroll bar when the rollback stops. The system calculates the scroll speed based on the scroll bar's starting and target positions to determine whether the user's scroll is invalid or valid. For valid scrolls, the system obtains the content of the scrolled paragraph, extracts the text features of that paragraph, and integrates them into the dynamic interest context vector D.

[0013] According to the above technical solution, the real-time calculation unit quantizes the parsed d data and generates a feature vector D to construct a dynamic interest context vector D; The real-time calculation unit performs a two-layer similarity fusion calculation to obtain a similarity score, using the following formula: sim(U,Ci)=(U*Ci) / (||U||*||Ci||) calculates long-term interest similarity; sim(U,Ci) is the similarity score between the user interest vector U and the content feature vector Ci; U*Ci represents the dot product of vectors U and Ci; ||U|| is the Euclidean norm of vector U; ||Ci|| is the Euclidean norm of vector Ci; The real-time calculation unit uses the following formula: sim(D,Ci)=(D*Ci) / (||D||*||Ci||) calculates the real-time context similarity; sim(D,Ci) represents the real-time similarity between the dynamic interest context vector D and the content feature vector Ci; D*Ci is the dot product of vectors D and Ci; ||D|| is the Euclidean norm of vector D; ||Ci|| is the Euclidean norm of vector Ci; Finally, the fusion similarity score is calculated: FinalSim(i)=α*sim(U,Ci)+β*sim(D,Ci); FinalSim(i) is the final similarity score, α is the first fusion weight coefficient, β is the second fusion weight coefficient, and α+β=1; The value of coefficient β is dynamically provided by the self-learning optimization module based on the signal strength of D; The intelligent display module sorts webpage content based on similarity scores and selects recommended content with a similarity score greater than θ; θ is a preset similarity value. The intelligent display module dynamically determines the insertion position of recommended content and replaces the content displayed on the user's webpage based on information from the webpage visited by the user and data d. The real-time monitoring unit fine-tunes recommended content based on the user's real-time behavior.

[0014] According to the above technical solution, it also includes a collaborative application module, which is used by users to submit the required application through a webpage, and the system automatically pushes the user information to relevant departments or administrators to automatically promote internal processes.

[0015] According to the above technical solution, it also includes a self-learning optimization module, which is an AI model used to store historical data and feedback information, and automatically adjust and optimize various parameters and algorithms within the system.

[0016] Compared with the prior art, the beneficial effects achieved by the present invention are: This invention accurately constructs a user interest model through multi-dimensional user behavior analysis and high-quality data filtering: The system integrates multi-dimensional data such as click count, dwell time, scroll count, search keywords, and application behavior, and effectively distinguishes the intensity of user interests and filters invalid behaviors through preset thresholds and refined scenario judgment rules, providing a high-quality and high-purity set of behavioral features for subsequent analysis, thus ensuring the accuracy of user profiles from the source. This invention significantly improves content matching accuracy by dynamically updating user profiles and fusing recommendations with a two-layer similarity system: the user profile module employs normalization processing and a dynamic weight matrix, combining historical behavior to smoothly evolve interest weights. The intelligent display module integrates long-term interest profiles with real-time context vectors, achieving accurate content matching that balances consistent user preferences with instantaneous intent. This makes recommended content more aligned with users' real-time needs, significantly improving click-through rates and user satisfaction. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a system schematic diagram of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 The present invention provides a technical solution: a system for reconstructing and integrating supply chain resource data of enterprises undergoing digital transformation, including a user behavior module, a user profile module, a resource knowledge base module, an intelligent display module, a collaborative application module, and a self-learning optimization module; The user behavior module is used to obtain user behavior data and establish a set of behavior features.

[0020] The user behavior module includes an initial interest analysis unit, a click counting unit, a dwell time timing unit, a scrolling line count recording unit, a keyword extraction unit, and an application monitoring module.

[0021] Where C is the number of user clicks, Cmin is the preset number of clicks; T is the user dwell time, Tmin is the preset dwell time Tmin; S is the number of rows scrolled, Smin is the preset number of rows scrolled; and t is the keyword.

[0022] When a user first enters the website, the initial interest analysis unit sends a pop-up window instruction, and an interest analysis questionnaire pops up on the user side. After the user fills out the interest analysis questionnaire, the initial interest analysis unit filters out content with a high similarity to the user's interests in the resource knowledge base module and displays it as the initial web page to the user.

[0023] The user behavior module initializes a preset pre-interest pool and a non-interest pool, incorporates the data determined to be of pre-interest into the pre-interest pool, and incorporates the data determined to be of non-interest into the non-interest pool. The data in the non-interest pool is not included in the reference data range of the behavior feature set.

[0024] When the user performs a click operation, the click counting unit obtains user data C for click count; When C >= Cmin times, it is determined that the user is interested in the content and marked as pre-interest, and at the same time, d is obtained; when C < Cmin times, it is determined that the user is not interested in the content and marked as non-interest; Using the click count C as the judgment basis can effectively filter out the user's interest concentration; however, there are a large number of misclicks and short-term browsing situations in the user's behavior data. By setting the preset value Cmin, it is possible to filter out misclicks and short-term browsing situations, and only incorporate the behavior user data of repeated clicks, showing an interest concentration, into the pre-interest, providing a high-quality data basis for the subsequent establishment of the behavior feature set.

[0025] When the user stays on the web page, the stay duration timing unit obtains user data T and judges the interest concentration according to the user's stay duration on the page; The web page is divided into a main page and sub-pages, and the determination is further refined by placing the user's stay duration T in different page scenarios, while ensuring that it is judged in combination with the reading depth, immersion level, and the user's stay duration T.

[0026] The content of the main page is extensive. When the user stays on the main page, it can be judged through T whether the user is browsing aimlessly or is interested in a specific content, thereby improving the accuracy of the judgment.

[0027] When T > Tmin seconds, it is determined that the user is interested in the content, but the specific content cannot be determined, so only d is obtained; by obtaining d, the system can obtain the user's potential interest points, providing a data basis for the subsequent intelligent display and recommendation module to recommend content directionally.

[0028] The content of the sub-pages is more in-depth. When the user stays on a sub-page, the T of each sub-page is obtained, and the T of the sub-page can more directly reflect the interest concentration, thereby accurately positioning the specific content items that the user is interested in.

[0029] When T > Tmin seconds, it is determined that the user is interested in the content, marked as pre - attention, and d is obtained at the same time; when T < Tmin seconds, it is determined that the user is not interested in the content, marked as not interested. The sub - page with a long stay is determined as pre - attention, which represents a signal of strong interest concentration of the user and can accurately locate the specific content items that the user is interested in.

[0030] The sub - page with a short stay is determined as not interested, which can effectively filter out invalid behaviors such as mis - clicks or quick jumps, prevent them from polluting the data of the behavior feature set, and save the analysis resources of the system.

[0031] Calculate T total = S1211 + S1221 and fuse all d data. Where T total is the total sum of the user's stay duration; T total records the summary index of the user's single - session, correlates the user's behavior on the main page and the behavior on the sub - page to form a complete stay duration, and its data is used for subsequent establishment of the behavior feature set.

[0032] The fused data d forms a data set including the interest concentration tendency and interest preference. The system can automatically assign different weights to data from different sources. The interest concentration weights generated by homepage browsing and in - depth reading of sub - pages are higher than those generated by the number of clicks, which can greatly improve the accuracy of the user portrait module constructed through the behavior feature set.

[0033] When the user uses the scroll wheel to roll back the page, the data S is read by the scroll line recording unit for line recording.

[0034] Taking the number of user scroll lines S as the judgment basis is because the interest concentration can be calculated according to the user's roll - back behavior, providing a high - quality data basis for subsequent establishment of the behavior feature set.

[0035] When S > Smin lines and there is no roll - back, it usually means that they are quickly browsing to find specific information and may leave directly after finding that the content is not relevant. This is a non - reading behavior; it is determined that the user is not interested in the content, marked as not interested. Determining this behavior as not interested can effectively prevent the system from wrongly recording the user's invalid browsing or negative feedback as interest concentration, thus maintaining the purity of the subsequent user portrait module.

[0036] When S > Smin lines, but there is scrolling, the stay duration timing unit is used to determine whether the stay duration meets the pre - attention preset value; usually, roll - back indicates that the user's reading speed cannot keep up with the scrolling speed and needs to look back to understand the content or is particularly interested in a certain part of the content and is savoring it repeatedly. When S < Smin and T > Tmin, it indicates that the information required by the user is on the page, and the content is highly attractive so that the user can obtain it without scrolling. It is determined that the user is interested in the content and marked as pre - concerned. Meanwhile, d is obtained; the data d helps to analyze which content on the page is required by the user, providing data support for the content optimization of the subsequent intelligent display and recommendation module.

[0037] When the user searches for content on the website, the user data t is obtained by the keyword extraction unit for keyword extraction.

[0038] When T > Tmin, C > Cmin and S > Smin, it is determined as pre - concerned, and d is obtained simultaneously; this indicates that the user not only stays for a long time, but also has active interactive click behavior and browses most of the content; the system can obtain a strong and stable interest concentration signal from it. This determination rule can accurately identify the user's full - range and in - depth participation in the content, assign the highest weight to this interest category, and provide the best training samples for the self - learning optimization module.

[0039] When T > Tmin, C < Cmin and S > Smin, it is determined as pre - concerned, and d is obtained simultaneously; This indicates that the user stays for a long time and reads deeply, but clicks less; the system can avoid misjudging users with an interest concentration in in - depth reading due to insufficient click times according to this determination rule, and it helps the system discover high - quality resources that can make users immerse and have complete content.

[0040] When T > Tmin, C < Cmin and S < Smin, it is determined as pre - concerned, and d is obtained simultaneously; This indicates that the user stays in a small area of the page for a long time, neither scrolling deeply nor clicking frequently; this enables the system to accurately capture the user's interest concentration in the page or specific segment content. It prevents the system from ignoring the user's attention due to few user interaction behaviors, and at the same time obtains the data d for use in the subsequent intelligent display and recommendation module.

[0041] When T < Tmin, C > Cmin and S > Smin, it is determined as not concerned; This indicates that the user scrolls quickly and clicks frequently in a short time, but the total stay time is very short; this enables the system to accurately identify the user's state, usually indicating a search failure or poor content. Classifying this behavior as not concerned can effectively filter out negative feedback behaviors and prevent the system from misjudging.

[0042] When T < Tmin, C < Cmin and S < Smin, it is determined as not concerned; This indicates that the user leaves almost without any effective interaction; the system can filter it most efficiently, directly saving computing resources, and providing the clearest negative samples for content quality and recommendation relevance.

[0043] When a user clicks the application button on the webpage, the application monitoring module reads the user's click behavior and directly determines the user's behavior as a pre-follower.

[0044] A behavioral feature set is built based on the pre-attention pool and the non-attention pool, and a feature value is added to each data point.

[0045] When a user submits an application request on the supply chain solutions page of the enterprise digital transformation platform, the system directly determines the behavior as pre-following based on preset conditions, and includes all relevant data of this interaction, including the application type (trial application), the source page (supply chain solutions), the timestamp, and the user's basic information, into the pre-following pool. At the same time, the system determines the data of industry news pages that the user briefly browses but does not interact with in the same session as not following and includes them in the unfollowing pool.

[0046] Based on these two data pools, the system builds a behavioral feature set, adding high-weight feature values ​​such as application intent strength (0.95) and business priority (0.9) to the trial application behavior in the pre-follow pool, thereby constructing a behavioral feature set with significant differentiation, providing a high-quality data foundation for subsequent user profile modeling and accurate recommendations.

[0047] The user profiling module is used to process all data based on behavioral feature sets and analyze the user's interest weights.

[0048] The user profiling module includes numerical classification, numerical analysis, and type analysis units. The user profiling module extracts all feature values ​​from the behavioral feature set, compares each feature value, and converts the feature values ​​into a uniform numerical range. The numerical classification unit separates and divides the data from the behavioral feature set, classifying click count, dwell time, and number of scrolls into numerical features, and keywords and application behavior into categorical features.

[0049] When processing numerical features, the numerical analysis unit scales the numerical features to a range of 0 to 1.

[0050] For example, in one session, a user clicked the Smart Factory Solution page C=3 times, stayed for T=45 seconds, and scrolled S=12 rows. Based on historical data or preset judgments, the system sets normalization benchmarks for each feature, and statistically obtains the historical maximum number of clicks Cmax=10 times, the historical maximum stay Tmax=300 seconds, and the historical maximum number of scrolled rows Smax=50 rows. According to the normalization rule S211, the system performs the following calculations: the normalized value of the number of clicks is 3 / 10=0.3; the normalized value of the stay time is 45 / 300=0.15; and the normalized value of the number of scrolled rows is 12 / 50=0.24. After this processing, the original multi-dimensional data of user A's behavior is successfully transformed into a standardized feature vector [0.3,0.15,0.24] that can be used for direct comparison and weighted calculation, providing standard and reliable data input for the accurate calculation of interest weights in the subsequent user profiling module.

[0051] When processing categorical features, the categorical analysis unit converts the categorical features into numerical vectors that reflect their importance.

[0052] For example, a user performs two key categorical actions in a single session: first, they search for the keyword "Industrial Internet," and then they submit a request for expert consultation. The system first converts the keyword "Industrial Internet" into a 128-dimensional dense vector, such as [0.12, -0.45, 0.78, ..., 0.33], using a pre-trained word embedding model. This vector is close to the "Intelligent Manufacturing" vector in the semantic space, capturing its technical attributes. Simultaneously, the system assigns a high-weight numerical identifier vector [0, 0, 1, 0.95] to the action of requesting expert consultation, based on preset business rules. The last dimension, 0.95, represents a very high importance score for this request. Finally, the system weighted and concatenates these two numerical vectors, fusing them into a composite categorical feature vector that comprehensively reflects the user's current intention and interest level, used to construct an accurate user profile.

[0053] Based on the processed numerical and categorical features, calculate the initial weight for each interest category; First, a weight coefficient is assigned to each feature, and all coefficients form a weight matrix, which is dynamically provided by the system's self-learning optimization module. Secondly, calculate the score for each interest category. The score is the sum of the products of all user feature values ​​and their corresponding weights, plus an adjustment term. Then, the interest weights are dynamically updated based on the user's historical behavior data; During this period, the interest weights are updated regularly. The new weights are determined by the old weights from the previous moment and the current new score, with the old weights accounting for a larger proportion, in order to achieve a smooth evolution of the weights. Finally, output the user profile, which includes the user's interest weight vector and a list of major interest categories; The resource knowledge base module is used to store, manage, and label all recommended content resources in the system; it is a structured content database.

[0054] The intelligent display module matches web page content that is highly similar to the user's interests in real time based on user profiles and resource knowledge base; The intelligent display and recommendation module includes a real-time calculation unit and a real-time monitoring unit. The intelligent display module obtains the user interest weight vector U from the user profiling module, and simultaneously obtains the feature vectors of all web page content from the resource knowledge base module. ; The resource knowledge base module transforms unstructured content resources into feature vectors that can be understood and computed by machines. This ensures that content can be quickly matched and retrieved, guarantees that recommended content aligns with users' consistent preferences, and ensures a consistent and predictable user experience.

[0055] The intelligent display module obtains current and recent related data (d) from the user profile module in real time and constructs a dynamic interest context vector (D). By parsing real-time signals within the page, the intelligent display module can capture the user's instantaneous interest intent in the current session, ensuring that the recommended content is timely.

[0056] The real-time processing unit parses the data d and extracts information such as the mouse hover hotspot, search terms t within the page, scrollback paragraphs, and specific content blocks where the user is hovering in the current user session.

[0057] Specifically, when parsing the mouse hover hotspot in data d: First, the webpage currently accessed by the user is divided into n logical blocks, and a unique area identifier is assigned to each block. At the same time, the system records the core content elements contained in each block, which include, but are not limited to, text paragraphs, images, video players, data tables, and hyperlinks.

[0058] When a user hovers their mouse over a certain area, the system starts recording the hover duration and retrieves the content of that area's page; recording stops when the user leaves the area. By binding hovering behavior to specific page areas and content, the system can more accurately understand the user's focus.

[0059] The intelligent display and recommendation module continuously monitors the duration of the user's mouse hovering within each block through a real-time monitoring unit. Recording begins when the user's mouse enters a block and stops when the mouse leaves. The original hover duration is recorded as "Traw," and valid hovering behavior is filtered out. Set the minimum effective hover preset value Tfast = 0.3 seconds; if Traw < Tfast, it is determined that the mouse moves too fast, and the hover duration of this time is not included in the cumulative data. This is used to avoid misjudgment caused by rapid mouse sliding and ensure that the heat zone data only reflects the areas that users are intentionally interested in.

[0060] Set the maximum effective hover preset value Tslow = 30 seconds; if Traw > Tslow, it is determined that the mouse moves too slowly, and the system automatically divides the hover duration. Only the first Tslow duration is recorded as the valid value, that is, Tvalid = Tslow, and the exceeded part is regarded as invalid. This is used to prevent the overestimation of attention caused by the user leaving the seat or being distracted for a long time and improve the authenticity of the heat weight.

[0061] Convert the hover duration into high-frequency timestamps and continuously collect the area identifier where the mouse cursor is located and its continuous stay time, establishing the association of block-duration-content, so that the system can comprehensively judge the interest concentration according to the block content and hover duration.

[0062] Subsequently, accumulate the multiple hover durations of the same area identifier in each block to obtain Thover of the block; use the normalization algorithm to map Thover to a block Wheat in the range of [0,1]. The longer the hover duration of the block, the higher the Wheat of the block; at the same time, associate the heat zone with the specific web page content carried by the block.

[0063] Among them, Thover is the total effective hover duration, Tmin is the minimum hover duration of all blocks, Tmax is the maximum hover duration of all blocks, and Wheat is the heat weight value. If a block contains two labels at the same time, each of these two labels gets a part of Wheat.

[0064] Exemplarily, in the product specification block of a web page product details page, the system monitors that the total hover duration of the user is 5 seconds. Through normalization calculation, the heat weight value Wheat of this block is 0.286; since this block contains two content labels, technical parameters and product pictures, the system distributes the heat value as 0.172 and 0.114 according to the preset weight ratio and integrates it into the dynamic interest context vector D, so that subsequent intelligent recommendations are more inclined to display content related to technical parameters and product appearance, realizing precise delivery and real-time response.

[0065] Among them, the real-time monitoring unit also includes continuously recording and analyzing the moving speed, trajectory pattern and interaction mode of the mouse within the web page, which is used to evaluate the user's proficiency in computer operation.

[0066] The mouse moving speed is judged by obtaining the position coordinates in real time, calculating the average displacement distance per unit time, and combining the click count C, hover duration and heat zone distribution.

[0067] When the system detects that the user's mouse movement speed remains low, the number of clicks is small (C < Cmin), and the distribution of the mouse heat zone is highly uneven (the heat weight value of a certain block is significantly higher than that of other blocks), the user is determined to be an unskilled operator. This usually indicates that the user is hesitant in interface navigation, has low exploration efficiency, or is unfamiliar with interactive functions, and may require more guidance and simplified operation paths.

[0068] Conversely, when the mouse movement speed is fast, the click behavior is frequent (C ≥ Cmin), and the heat zone distribution is relatively uniform (the difference in heat weight values of each block is not large), the system determines that the user is a skilled operator. This indicates that the user can often quickly locate the target content, efficiently complete interactive tasks, and demonstrates strong information retrieval and interface adaptation abilities.

[0069] Based on the determination result of the user's operation proficiency, the system can further customize the interface layout, simplify or deepen the information hierarchy, and dynamically insert operation guides or basic content recommendations for unskilled users, and provide more efficient and intensive information display and shortcut functions for skilled users, thereby comprehensively improving the usability of the platform and the adaptability of the user experience.

[0070] Finally, based on the heat zone analysis results, the system dynamically inserts the most relevant recommended content near the high-heat blocks that the user is most concerned about, achieving precise placement of what is seen is what is recommended. Continuously collect heat zone data, generate a visual heat map, provide data-driven decision support for optimizing web page content, so as to place high-concern content in a position where users are more likely to notice, and improve the overall resource conversion efficiency. Integrate the heat weight values of each block and their associated content into the dynamic interest context vector D. The content of the block with a high heat weight value will obtain a higher feature weight in the dynamic interest context vector D.

[0071] The real-time calculation unit analyzes the data d and extracts the current user's rollback behavior; First, the system defines and captures the user's rollback behavior. A complete reading session of the user is defined as starting from page loading until the user jumps away or closes the page.

[0072] On the current session page, the system continuously monitors the position of the scroll bar. When it detects that the scroll direction changes from down to up, it is marked as the start of a rollback behavior, and the starting position Pend of the scroll bar at the start of the rollback and the target position Pstart of the scroll bar at the end of the rollback are recorded; Set a preset minimum effective rollback amplitude value Pmin. If ΔP < Pmin, it is determined to be an operation jitter or an accidental fine-tuning, and this rollback is not counted as valid data.

[0073] Record the duration of rollback events (Troll) and set a maximum effective rollback speed preset value (Vmax). Calculate the rollback speed Vroll=ΔP / Troll. If Vroll>Vmax, it is determined that the user is quickly skimming over to find other content rather than carefully rereading it, and this rollback will not be counted as valid data.

[0074] A rollback is considered valid if ΔP >= Pmin and Vroll <= Vmax. For a valid rollback, the system obtains the content of the rolled-back paragraph, extracts the text features of that paragraph, and integrates them into the dynamic interest context vector D.

[0075] The real-time processing unit quantizes the parsed d data to generate a feature vector D that reflects the user's real-time, micro-level intent, thereby constructing a dynamic interest context vector D.

[0076] For example, if a user hovers their mouse over the price area of ​​solution A on a webpage, the weight of the price feature of solution A in vector D is increased. Simultaneously, effective hover detection is performed to avoid misjudgments caused by rapid mouse scrolling or inflated attention levels due to user leaving their seat or prolonged distraction.

[0077] The real-time calculation unit performs a two-layer similarity fusion calculation to obtain a similarity score. The long-term interest similarity is calculated using the formula: sim(U,Ci)=(U*Ci) / (||U||*||Ci||. This ensures that the system will not deviate from the user's core interests due to a temporary or accidental user behavior, thus maintaining the robustness of the recommendation system.

[0078] Where sim(U,Ci) represents the similarity score between the user interest vector U and the content feature vector Ci. Its value range is usually between [0,1], and the closer the value is to 1, the more similar it is. U*Ci: Represents the dot product of vectors U and Ci, which calculates the sum of the products of corresponding components of the two vectors; ||U||: denotes the Euclidean norm of vector U; ||Ci||: Denotes the Euclidean norm of vector Ci.

[0079] For example, the user interest vector U and content feature vector Ci are obtained as follows: U=[0.8,0.2,0.5]; Ci=[0.6,0.3,0.1]; First, calculate the dot product: U*Ci=(0.8×0.6)+(0.2×0.3)+(0.5×0.1)=0.48+0.06+0.05=0.59; Recalculate the norm:

[0080]

[0081] Finally, the similarity was calculated:

[0082] The similarity score was approximately 0.903, indicating that the user had a high degree of interest in the content.

[0083] The real-time calculation unit calculates the real-time context similarity using the formula: sim(D,Ci)=(D*Ci) / (||D||*||Ci||). This enables the system to improve the accuracy of recommendations, respond to the content with the highest user interest at that moment, and intelligently recommend content that is highly consistent with the current browsing context, thereby significantly improving click-through rate and user satisfaction.

[0084] Wherein, sim(D,Ci) represents the real-time similarity between the dynamic interest context vector D and the content feature vector Ci, which is used to measure the degree of matching between the user's current interest and the content i.

[0085] D*Ci represents the dot product of vectors D and Ci; ||D|| denotes the Euclidean norm of vector D; ||Ci|| denotes the Euclidean norm of vector Ci; Finally, the fusion similarity score is calculated: FinalSim(i)=α*sim(U,Ci)+β*sim(D,Ci); Where FinalSim(i) is the final similarity score, α is the first fusion weight coefficient, β is the second fusion weight coefficient, and α+β=1; The value of coefficient β is dynamically provided by the self-learning optimization module based on the signal strength of the dynamic context D. When the signal represented by the D vector is strong and clear, the value of β increases, and the real-time intent has a higher weight in the recommendation.

[0086] To ensure the optimal balance between stability and sensitivity, the system can grasp the basic content that users are interested in and flexibly respond to users' real-time needs; the α and β coefficients are dynamically adjusted to adapt to behavioral signals of different intensities.

[0087] The intelligent display module sorts web page content based on similarity scores and selects recommended content with a similarity score greater than θ. Here, θ is the preset similarity value, used to determine whether the similarity meets the standard. By setting the preset value θ, the system ensures that the recommended content has high relevance, effectively filtering out low-quality or weakly relevant content, thereby improving the overall quality of the recommendation list and avoiding information overload for users.

[0088] The intelligent display module dynamically determines the insertion position of recommended content and replaces the content displayed on the user's webpage based on information from the webpage visited by the user and data d. For example, if it is detected that a user has recently scrolled back to the technical documentation area at the bottom of the page, relevant content recommendations will be dynamically inserted nearby the next time the user scrolls back to that area.

[0089] Recommended content is seamlessly embedded into the user's browsing experience. This significantly reduces the user's cognitive load and operational costs, making recommended content an aid to browsing and improving recommendation acceptance and conversion efficiency.

[0090] The real-time monitoring unit fine-tunes recommended content based on real-time user behavior; it continuously monitors user behavior on the page, and once new data with strong interest signals is generated, the recommended content is refreshed in real time through the intelligent display module.

[0091] This transforms the system's real-time interactive recommendations from a one-time calculation service to one that continuously interacts with the user throughout the entire page session, dynamically changing according to the user's interest level and providing an immersive, personalized experience.

[0092] The user profile module records user feedback on recommended content and uses it to update the behavioral feature set. It is also transmitted to the self-learning optimization module to accumulate historical data.

[0093] The collaborative application module allows users to submit their applications via a webpage, and the system automatically pushes the user information to relevant departments or administrators to automate internal processes. The collaborative application module determines the relevant department or administrator based on the application type; if multiple departments are involved, the primary department is selected based on priority or load balancing; the collaborative application module pushes information to the relevant department or administrator's system via the message queue API interface; records the push status and notifies the user that the application has been submitted; The self-learning optimization module is an AI model used to store historical data and feedback information, and automatically adjust and optimize various parameters and algorithms within the system.

[0094] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0095] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A system for reconstructing and integrating supply chain resource data for enterprises undergoing digital transformation, characterized in that: It includes a user behavior module, a resource knowledge base module, a user portrait module, and an intelligent display module; The user behavior module is used to obtain user behavior data and establish a behavior feature set; The user portrait module is used to process all data based on the behavior feature set and analyze the user's interest weights; The resource knowledge base module is used to store, manage, and label all recommended content resources of the system, and is a structured content database; The intelligent display module matches web page content highly similar to the user's interest concentration in real time according to the user portrait and the resource knowledge base; The user behavior module includes an initial interest analysis unit, a click counting unit, a stay duration timing unit, a scroll line number recording unit, a keyword extraction unit, and an application monitoring module; The initial interest analysis unit includes that when the user first enters the website, the initial interest analysis unit sends a pop-up window instruction, and an interest analysis questionnaire pops up to the user side. After the user fills in the interest analysis questionnaire, the initial interest analysis unit screens out content with a high similarity to the user's interest in the resource knowledge base module and displays it to the user as the initial web page; The user behavior module initializes a preset pre-concern pool and a non-concern pool, incorporates the data determined to be pre-concerned into the pre-concern pool, and incorporates the data determined to be non-concerned into the non-concern pool; When the user performs a click operation, the click counting unit obtains user data C and counts the number of clicks; When C >= Cmin times, it is determined that the user is interested in the content and marked as pre-concerned, and at the same time, d is obtained; when C < Cmin times, it is determined that the user is not interested in the content and marked as non-concerned; The C is the number of user clicks, and the Cmin is the preset value of the number of clicks; When the user stays on the web page, the stay duration timing unit obtains user data T and judges the user's interest concentration based on the stay duration of the user on the page; The T is the user's stay duration, and the Tmin is the preset value of the stay duration Tmin; Among them, the web page is divided into a main page and a sub-page, and the determination is further refined by placing the user's stay duration T in different page scenarios; When T > Tmin seconds, it is determined that the user is interested in the content, but the specific content cannot be determined, so only d is obtained; When T > Tmin seconds, it is determined that the user is interested in the content and marked as pre-concerned, and at the same time, d is obtained; when T < Tmin seconds, it is determined that the user is not interested in the content and marked as non-concerned; Calculate T total = S1211 + S1221 and fuse all d data; the T total is the total of the user's stay duration; When the user uses the scroll wheel to roll back the page, the scroll line number recording unit reads the data S and records the number of lines; The S is the number of user scroll lines, and the Smin is the preset value of the number of scroll lines; When S > Smin lines and there is no rollback, it is determined that the user is not interested in the content and marked as non-concerned; When S > Smin lines but there is a scroll, the stay duration timing unit is used to judge whether the stay duration meets the pre-concern preset value; [[ID= When a user searches for content through a website, the keyword extraction unit obtains user data t for keyword extraction; the t is a keyword.

2. The supply chain resource data reconstruction and integration system for digitally transformed enterprises as described in claim 1, characterized in that, The user behavior module further includes: When T > Tmin and C > Cmin and S > Smin, it is determined as pre - attention, and d is obtained simultaneously; When T > Tmin and C < Cmin and S > Smin, it is determined as pre - attention, and d is obtained simultaneously; When T > Tmin and C < Cmin and S < Smin, it is determined as pre - attention, and d is obtained simultaneously; When T < Tmin and C > Cmin and S > Smin, it is determined as not interested; When T < Tmin and C < Cmin and S < Smin, it is determined as not interested; When the user clicks the web page application button, the application monitoring module reads the user's click behavior and directly determines this behavior of the user as pre - attention; Based on the pre - attention pool and the not - interested pool, a behavior feature set is established, and a feature value is added to each data.

3. The supply chain resource data reconstruction and integration system for digitally transformed enterprises according to claim 2, characterized in that, The user portrait module includes a numerical classification unit, a numerical analysis unit, and a type analysis unit; The user portrait module extracts all feature values from the behavior feature set, performs comparison processing on each feature value, and converts the feature values into a unified numerical range; The numerical classification unit divides and analyzes the data in the behavior feature set while stripping. It classifies the click count, stay duration, and scroll lines into numerical features; and classifies keywords and application behaviors into categorical features; When processing numerical features, the numerical analysis unit scales the numerical features to the range of 0 to 1; When processing categorical features, the type analysis unit converts the categorical features into a numerical vector that can reflect their importance.

4. The supply chain resource data reconstruction and integration system for digitally transformed enterprises according to claim 3, characterized in that, The user portrait module calculates the initial weight of each interest category based on the processed numerical features and categorical features; First, a weight coefficient is assigned to each feature, and all coefficients form a weight matrix, which is dynamically provided by the self - learning optimization module; Secondly, the score of each interest category is calculated. The score is the sum of the products of all user feature values and their corresponding weights, plus an adjustment term; Then, the interest weights are dynamically updated according to the user's historical behavior data; During this period, the interest weights are updated regularly. The new weight is jointly determined by the old weight at the previous moment and the current new score, where the old weight accounts for a larger proportion to achieve a smooth evolution of the weight; Finally, a user portrait is output, which includes the user's interest weight vector and a list of main interest categories.

5. The supply chain resource data reconstruction and integration system for digitally transformed enterprises according to claim 4, characterized in that, The intelligent display recommendation module includes a real - time calculation unit and a real - time monitoring unit; The intelligent display module obtains the user interest weight vector U from the user profiling module, and simultaneously obtains the feature vectors of all web page content from the resource knowledge base module. ; The resource knowledge base module transforms unstructured content resources into feature vectors that can be understood and computed by machines. ; The intelligent display module obtains the current and recent associated data d from the user portrait module in real - time and constructs a dynamic interest context vector D; The real - time calculation unit analyzes the data d and extracts the mouse - hover hot zone, the in - page search term t, and the rolled - back paragraphs in the current user session.

6. The supply chain resource data reconstruction and integration system based on digitally transformed enterprises according to claim 1, characterized in that, When the real - time calculation unit analyzes the mouse - hover hot zone in the data d: First, divide the web page interface currently accessed by the user into n logical blocks, and assign a unique area identifier to each block. At the same time, the system records the core content elements contained in each block, and these core elements include text paragraphs, pictures, video players, data tables, and hyperlinks. When the user hovers the mouse over a certain block, start recording the hovering duration and obtain the content of the page of this block, and stop recording when leaving this block. The intelligent display recommendation module continuously monitors the hovering duration of the user's mouse in each block through the real-time monitoring unit; when the user's mouse enters a block, start recording the hovering duration. Stop recording when leaving the block; record the original hovering duration as Traw, and at the same time perform an effective hovering determination to filter out invalid hovering behaviors: Set the minimum effective hovering preset value Tfast = 0.3 seconds. If Traw < Tfast, it is determined that the mouse moves too fast, and the hovering duration of this time is not included in the cumulative data. Set the maximum effective hovering preset value Tslow = 30 seconds; if Traw > Tslow, it is determined that the mouse moves too slowly, and the system automatically divides the hovering duration, and only records the first Tslow duration as the effective value, that is, Tvalid = Tslow, and the excess part is regarded as invalid. Convert the hovering duration into a high-frequency timestamp, continuously collect the area identifier where the mouse cursor is located and its continuous stay time, and establish the association of block-duration-content. Subsequently, accumulate the hovering durations of multiple times with the same area identifier for each block to obtain the Thover of this block; use the normalization algorithm to map Thover to a block Wheat in the interval [0, 1]. The longer the hovering duration of a block, the higher the Wheat of this block; at the same time, the hot area is associated with the specific web page content carried by this block. The Thover is the total effective hovering duration, the Tmin is the minimum hovering duration of all blocks, the Tmax is the maximum hovering duration of all blocks, and the Wheat is the heat weight value; when a block contains two labels at the same time, each of these two labels obtains a part of Wheat.

7. The supply chain resource data reconstruction and integration system for digitally transformed enterprises according to claim 6, characterized in that, The real-time monitoring unit also includes continuously recording and analyzing the moving speed, trajectory pattern and interaction mode of the mouse in the web page, and evaluating the user's proficiency in computer operation. The mouse moving speed is judged by obtaining the position coordinates in real time, calculating the average displacement distance per unit time, and combining the click count C, hovering duration and hot area distribution. ​ ​ ​ ​ ​ In the current session page, the system continuously monitors the scroll bar position. When it detects that the scroll direction changes from downward to upward, it marks the start of a rollback behavior and records the starting position of the scroll bar when the rollback starts and the target position of the scroll bar when the rollback stops. The system calculates the scroll speed based on the scroll bar's starting and target positions to determine whether the user's scroll is invalid or valid. For valid scrolls, the system obtains the content of the scrolled paragraph, extracts the text features of that paragraph, and integrates them into the dynamic interest context vector D.

8. The supply chain resource data reconstruction and integration system for digitally transformed enterprises according to claim 1, characterized in that, The real-time solution unit quantizes the parsed d data to generate a feature vector D, thereby constructing a dynamic interest context vector D. The real-time calculation unit performs a two-layer similarity fusion calculation to obtain a similarity score, using the following formula: sim(U,Ci)=(U*Ci) / (||U||*||Ci||) calculates long-term interest similarity; The sim(U,Ci) is the similarity score between the user interest vector U and the content feature vector Ci. U*Ci represents the dot product of vectors U and Ci; The ||U|| is the Euclidean norm of vector U; The ||Ci|| is the Euclidean norm of vector Ci; The real-time calculation unit uses the following calculation formula: sim(D,Ci)=(D*Ci) / (||D||*||Ci||) calculates the real-time context similarity; The sim(D,Ci) represents the real-time similarity between the dynamic interest context vector D and the content feature vector Ci. D*Ci is the dot product of vectors D and Ci; The ||D|| is the Euclidean norm of vector D; The ||Ci|| is the Euclidean norm of vector Ci; Finally, the fusion similarity score is calculated: FinalSim(i)=α*sim(U,Ci)+β*sim(D,Ci); FinalSim(i) is the final similarity score, α is the first fusion weight coefficient, β is the second fusion weight coefficient, and α+β=1; The value of coefficient β is dynamically provided by the self-learning optimization module based on the signal strength of D; The intelligent display module sorts webpage content based on similarity scores and selects recommended content with a similarity score greater than θ; where θ is a preset similarity value. The intelligent display module dynamically determines the insertion position of recommended content and replaces the content displayed on the user's webpage based on the information in the webpage visited by the user and in the data d. The real-time monitoring unit fine-tunes the recommended content based on the user's real-time behavior.

9. The supply chain resource data reconstruction and integration system for digitally transformed enterprises according to claim 1, characterized in that, It also includes a collaborative application module, which allows users to submit their applications via a webpage, and the system automatically pushes the user information to relevant departments or administrators to automate internal processes.

10. The supply chain resource data reconstruction and integration system for digitally transformed enterprises according to claims 1 to 9, characterized in that, It also includes a self-learning optimization module, which is an AI model used to store historical data and feedback information, and automatically adjust and optimize various parameters and algorithms within the system.

Images