User recommendation method and system based on enterprise product portrait analysis

Abstract

The present application relates to the technical field of portrait construction, and particularly relates to a user recommendation method and system based on enterprise product portrait analysis.The method comprises the following steps: obtaining enterprise product information data; performing supply chain tracing based on the enterprise product information data to generate enterprise product supply chain information data; performing associated enterprise screening on the enterprise product supply chain information data to generate product associated enterprise information data; performing product component production proportion analysis on the enterprise product supply chain information data according to the product associated enterprise information data to generate product component production proportion values; and performing product component input label quantification on the enterprise product information data based on the product component production proportion values to generate product label data.The present application improves the accuracy and comprehensiveness of user recommendation by fusing product supply chain tracing, component proportion quantification, user deep behavior analysis and scene perception recommendation.

Description

Technical Field

[0001] This invention relates to the field of profile building technology, and in particular to a user recommendation method and system based on enterprise product profile analysis. Background Technology

[0002] Traditional user recommendation methods primarily rely on technologies such as collaborative filtering and content recommendation. While these methods achieved some initial success, they are prone to cold start problems and low recommendation accuracy in scenarios with incomplete user profiles and low product information structuring. The widespread application of artificial intelligence, especially deep learning, has injected new vitality into recommendation systems. Enterprises are beginning to build sophisticated product profile models, structurally modeling products from multiple dimensions such as product attributes, functional features, market positioning, and user reviews to form enterprise product profiles. Based on this, multi-dimensional matching analysis is performed by combining user behavior data and preference information to achieve more personalized and accurate user recommendations. However, current traditional recommendation systems are mostly based on user behavior, neglecting a deeper understanding of the structure and origin of the enterprise's products, resulting in low matching accuracy and difficulty in real-time perception of user scenarios, affecting personalization and usability. Summary of the Invention

[0003] Therefore, it is necessary to provide a user recommendation method and system based on enterprise product profile analysis to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a user recommendation method based on enterprise product profile analysis is provided, the method comprising the following steps:

[0005] Step S1: Obtain enterprise product information data; perform supply chain traceability based on enterprise product information data to generate enterprise product supply chain information data; filter related enterprises in the enterprise product supply chain information data to generate product related enterprise information data;

[0006] Step S2: Analyze the production ratio of product components based on the product-related enterprise information data and the enterprise product supply chain information data to generate product component production ratio values; quantify the product component input tags based on the product component production ratio values ​​and the enterprise product information data to generate product tag data.

[0007] Step S3: Construct enterprise product profiles based on product tag data and enterprise product information data; collect purchase user information data using the constructed enterprise product profiles; analyze user interaction behavior of purchase user information data to generate purchase user interaction behavior data, including time-series offset browsing scan and collaborative path reverse behavior tracking.

[0008] Step S4: Purchase user interaction behavior data to locate user scenarios based on low-power Bluetooth beacons, and dynamically generate personalized user recommendation lists based on edge servers to perform user recommendation tasks based on enterprise product profile analysis.

[0009] This invention, by acquiring enterprise product information data and conducting supply chain traceability analysis, can comprehensively grasp the source of product components, production processes, and key nodes, improving supply chain transparency and controllability, and facilitating accountability and optimization. Utilizing the production proportion analysis of product components, it can quantitatively describe the input ratio of each component in the product, aiding in product cost optimization, key component identification, and refined management. Enterprise product profiles constructed based on tagged quantitative data can display product attributes, market positioning, and technical characteristics from multiple dimensions, forming core data support for user behavior analysis and recommendation. Collecting and analyzing user information and their interactive behaviors such as time-series browsing offsets and collaborative path reverse tracing enables deep insights into user intent, providing behavioral basis for personalized services. Leveraging low-power Bluetooth beacon technology to achieve user scenario awareness, combined with dynamic recommendation calculations on edge servers, effectively improves the real-time performance, accuracy, and personalization of user recommendations. Deeply integrating enterprise product profiles with user behavior data achieves closed-loop data flow and intelligent linkage between the enterprise and user ends, enhancing the intelligence level of product marketing strategies. Through dynamic personalized recommendations and interactive behavior feedback mechanisms, enterprises can adjust product strategies in a timely manner to better meet users' personalized needs, enhancing user stickiness and brand loyalty. Therefore, this invention improves the accuracy and comprehensiveness of user recommendations by integrating product supply chain traceability, component proportion quantification, in-depth user behavior analysis, and scenario-aware recommendation.

[0010] Preferably, step S1 includes the following steps:

[0011] Step S11: Obtain enterprise product information data;

[0012] Step S12: Standardize the fields of the acquired enterprise product information data to generate standardized product information data; based on the standardized product information data, track the source of raw materials, processing nodes and circulation records of each batch of products by calling the enterprise resource planning system and product barcode database, and integrate the supply chain data links to generate the initial path data of the enterprise product supply chain.

[0013] Step S13: Classify the initial path data of the enterprise's product supply chain into nodes, identify the key path components of raw material suppliers, processing and manufacturing enterprises, and warehousing and logistics nodes, and generate supply chain sub-node information data; organize the link topology of the supply chain sub-node information data, construct a multi-level supply chain link map, eliminate redundant or broken links, and generate enterprise product supply chain information data.

[0014] Step S14: Calculate the supply intensity value of each node based on the enterprise product supply chain information data, and use the supply intensity value to screen related enterprises and generate a candidate dataset of product related enterprises; perform risk source cross-comparison on the candidate dataset of product related enterprises, screen out upstream and downstream enterprises with high stability, and finally generate product related enterprise information data.

[0015] This invention effectively solves problems such as inconsistent formats and field ambiguity across different data sources through field standardization, laying a solid foundation for subsequent data processing and supply chain analysis. Leveraging Enterprise Resource Planning (ERP) systems and product barcode databases, it tracks raw material sources, processing nodes, and distribution paths, constructing a multi-dimensional initial supply chain path for enterprise products, improving data coverage and traceability depth. By classifying nodes, it identifies core roles such as raw material suppliers, processing manufacturers, and warehousing and logistics, clarifying the supply chain division of labor and providing data support for risk control and node optimization. Using link topology optimization technology, it eliminates redundant and broken nodes, forming a visualized enterprise product supply chain information map, improving the transparency and management efficiency of the supply chain structure. By calculating supply intensity values ​​(e.g., delivery frequency, fulfillment rate, stability) for each level of node, it achieves quantitative assessment of supply chain nodes, improving the scientific rigor and objectivity of related enterprise selection. Combined with cross-comparison of risk sources, it further filters out potentially high-risk enterprises, retaining only stable and reliable upstream and downstream partners, significantly enhancing the resilience and risk resistance of the supply chain. The final generated product-related enterprise information data is of high quality and high credibility, laying a solid data foundation for subsequent component investment analysis, profile construction and personalized recommendations, and realizing systematic optimization from product to enterprise ecosystem.

[0016] Preferably, step S14, which calculates the supply intensity value of each node based on the enterprise's product supply chain information data and uses the supply intensity value to screen related enterprises, includes:

[0017] The enterprise product supply chain information data is structured and layered to identify supply nodes at all levels, including raw material supply nodes, production and processing nodes, assembly and integration nodes, and distribution nodes, generating multi-level supply chain node structure data;

[0018] For each node in the multi-level supply chain node structure data, a supply record statistical matrix is ​​constructed to generate node supply statistics. The supply record statistical matrix includes historical supply quantity, annual fulfillment frequency, and cooperation cycle duration.

[0019] The statistical data on supply at each node are quantified and standardized, and the supply frequency, supply stability and delivery ratio of each node are calculated. A multi-dimensional supply capacity parameter table is constructed to generate supply capacity characteristic data.

[0020] The node supply strength formula is defined based on supply capacity characteristic data, as shown below:

[0021] ;

[0022] In the formula, For the first The supply strength value of each node, For the first The supply frequency of each node, For the first The reliability of each node's performance. For the first Delivery percentage of each node, , , Custom weighting coefficients;

[0023] The supply strength value of each node is calculated using the node supply strength formula to generate supply strength index data. The supply strength index data is then classified into levels, supply strength boundary lines are set, and strong, medium, and weak categories are defined. Related nodes with supply strength at medium to high levels are selected to generate strong supply node screening data.

[0024] The data of strong supply nodes are screened and merged for enterprise uniqueness and supply chain role identification. Redundant or overlapping enterprises are eliminated, and finally a candidate dataset of product-related enterprises is generated.

[0025] This invention, by structurally layering enterprise product supply chain information data, accurately identifies multiple nodes such as raw materials, production and processing, assembly and integration, and distribution, effectively improving the understanding and control of the supply chain hierarchy. It constructs a supply record statistical matrix and extracts indicators such as historical supply quantity, fulfillment frequency, and cooperation duration, systematically describing the actual supply performance of each node and laying a data foundation for scientifically evaluating suppliers. Supply behavior is quantified through three dimensions: supply frequency, fulfillment reliability, and delivery percentage, forming a complete supply capacity parameter table, making supply capacity data more comparable and operable. A node supply intensity formula is used. The system allows for flexible adjustment of indicator weights based on specific business needs, resulting in a node scoring model that better aligns with the company's actual situation. Based on the calculated supply intensity value, it categorizes nodes into strong, medium, and weak levels, accurately identifying key nodes at the medium to high levels, effectively improving the stability and quality assurance of core supply chain links. Through enterprise uniqueness merging and role identification, redundant or duplicate enterprise information is eliminated, ensuring the accuracy and business relevance of the final screening results, laying a clear foundation for subsequent profiling and user recommendations. By constructing a standardized, structured, and quantifiable supply intensity assessment mechanism, enterprises can dynamically monitor and adjust cooperative nodes, reducing supply risks and enhancing the supply chain's ability to cope with unforeseen events.

[0026] Preferably, step S2 includes the following steps:

[0027] Step S21: Map the supply chain path of the product-related enterprise information data, identify the upstream supply source and actual production division ratio of each product component, and generate product component source path data;

[0028] Step S22: Accumulate and statistically analyze the production records of various components in the enterprise's product supply chain information data, and combine them with the product component source path data to extract the actual output ratio of different enterprises at the component granularity, and generate component production allocation matrix data.

[0029] Step S23: Normalize and integrate the component production allocation matrix data to generate standardized product component production percentage data; based on the product component production percentage data, construct a component tag input function:

[0030] ;

[0031] In the formula, For product components The tag weight value, For the first Individual companies for components The production share, For the first Individual enterprise credit rating; using the component label input function to calculate the label weight of the product component production ratio data, and generate component label quantitative data;

[0032] Step S24: Embed tags item by item in the product component list in the enterprise product information data, associate and bind the quantitative data of component tags with specific components, and generate structured and tagged product list data; perform unified product ID labeling, tag weight aggregation and version timestamp synchronization processing on the structured and tagged product list data, and finally generate product tag data.

[0033] This invention, by mapping the information of companies associated with a product, can identify the upstream source and division of labor among companies for each component, achieving supply chain visualization at the component level and providing data support for precise supply management. Combining product supply chain data with component source paths, it performs cumulative statistics and proportion extraction on enterprise output behavior, forming a realistic and quantitative component production allocation matrix that effectively reflects the degree of enterprise participation in component manufacturing. Component tag input function:

[0034] Corporate credit rating By integrating production share calculations, labels are effectively avoided by allocating them solely based on quantity or proportion, thereby guiding higher-quality enterprises to obtain higher label weights and enhancing the credibility and incentive effect of the labeling mechanism. Utilizing component-level and reputation-corrected label quantification data, labels reflect not only "whether they participated," but also "with what quality and contribution," significantly enhancing the label's discriminative power and credibility. Embedding quantified label data into the component structure of enterprise product information creates a structured, tagged product list, providing standardized foundational data for intelligent retrieval, intelligent matching, and visualization. Through unified product ID labeling, weight aggregation, and timestamp synchronization, synchronized updates of label data and product information, along with traceable historical version management, enhance system maintainability and data integrity. As the core input source for subsequent product profiling, the structured, standardized, and dynamically updated characteristics of product label data significantly improve profiling accuracy, label interpretability, and user matching.

[0035] Preferably, step S3, which involves constructing an enterprise product profile based on product tag data, includes:

[0036] The product label data is categorized by label type to generate aggregated label category data.

[0037] Calculate the label density of the tag category aggregated data;

[0038] Weight normalization is performed on the label density data to generate label weight distribution data;

[0039] Product feature vectors are constructed based on label weight distribution data, and principal component fusion is performed on the product feature vector data to generate principal component fused feature data.

[0040] The principal component fusion feature data is processed to construct a label map, generating label map structure data.

[0041] Embedding and mapping of tag graph structure data to generate enterprise product profile vector data;

[0042] Spatial dimensionality reduction is performed on the enterprise product profile vector data to generate enterprise product profile data.

[0043] This invention manages tag information in a structured way by categorizing and aggregating tag data, reducing data redundancy and improving the efficiency and accuracy of subsequent analysis and processing. It calculates the density of tag categories, objectively reflecting the distribution of each category of tags within the product, providing a data foundation for accurately capturing product features. Weight normalization is applied to the tag density data to eliminate the influence of dimensions, rationally allocating the contribution of each category of tags in profile construction, and improving the balance of profile representation. Based on the tag weight distribution, product feature vectors are generated, integrating multi-dimensional information to comprehensively reflect the multi-level and multi-dimensional characteristics of the product. Principal component fusion removes feature redundancy, highlights key components, enhances the discriminative power and stability of the profile, and improves the efficiency and accuracy of subsequent models. The construction of the tag graph reveals the inherent connections and hierarchical relationships between tags, enriching the semantic information of product features and providing multi-dimensional structural support for profiles. Embedding technology is used to transform the tag graph into enterprise product profile vectors, balancing expressive power and computational efficiency, facilitating the storage, retrieval, and analysis of profiles. By simplifying the spatial dimensionality reduction process, the profile vector data is made easier to respond quickly and perform efficient calculations in various application scenarios, while maintaining the core information of the profile. The generated enterprise product profile data has a high degree of information fusion and structure, providing accurate input for subsequent collection of user information and analysis of interaction behavior, and improving the intelligence level of the recommendation system.

[0044] Preferably, the user interaction behavior analyzed in step S3 for purchasing user information data includes:

[0045] Extract user access records and page interaction logs from the purchase user information data;

[0046] Reconstruct user behavior trajectories based on user access records to generate behavior path data;

[0047] Calculate the time-series features of user behavior path data to obtain the user's temporal offset features;

[0048] Based on time-series offset features, the user's browsing and scanning patterns within a specific time window are identified, and browsing and scanning behavior data is generated.

[0049] Based on behavioral path data, construct a collaborative path graph among users, identify existing collaborative path behaviors, and generate collaborative path behavior data.

[0050] Based on the collaborative path behavior data, reverse behavior path tracing is performed on the collaborative path graph to identify path convergence nodes and behavior-based reverse paths, generating collaborative path reverse tracing data.

[0051] By fusing browsing and scanning behavior data and collaborative path reverse tracing data, user interaction behavior pattern characteristics are identified and user interaction behavior pattern data is generated.

[0052] Based on user interaction behavior pattern data, construct structured purchase user interaction behavior data.

[0053] This invention reconstructs user access records and page interaction logs to fully restore user behavior paths, ensuring the spatiotemporal continuity and integrity of user behavior data, laying a solid foundation for accurate analysis. It calculates time-series features of user behavior paths, capturing temporal shifts and dynamic fluctuations in user behavior, improving the understanding of changes in user interests and behavioral rhythms. Utilizing temporal shift features, it analyzes user browsing and scanning behavior within specific time windows, capturing user focus and interest switching on the page, assisting in optimizing content display and recommendation strategies. By establishing a user collaboration path graph through behavior path data, it discovers implicit collaborative behavior patterns among users, helping to uncover group behavior characteristics and social influencing factors. Reverse behavior path tracing identifies convergence nodes and behavioral backward paths in collaborative paths, facilitating accurate analysis of key touchpoints and causal relationships in the user behavior chain, enhancing the explanatory power of behavior patterns. By fusing browsing and scanning behavior data with reverse tracing data of collaborative paths, it integrates multi-dimensional information to extract more representative user interaction behavior patterns, improving the accuracy and richness of the user behavior model. It generates structured and high-dimensional user interaction behavior data, supporting various intelligent application scenarios such as personalized recommendations, user profile improvement, and behavior prediction, enhancing the matching degree between enterprise products and users and the efficiency of interaction. It provides a deep understanding of user interaction behavior and its spatiotemporal evolution, offering enterprises data-driven user insights, optimizing product display and recommendation processes, and significantly improving user satisfaction and conversion rates.

[0054] Preferably, the collaborative path graph is traced backwards based on the collaborative path behavior data to identify path convergence nodes and behavior-based reverse paths, including:

[0055] Construct a directed graph model of the collaborative path graph, where nodes represent user behavior events and edges represent behavior transformation relationships, and generate collaborative path graph model data;

[0056] Based on collaborative path behavior data, target behavior event nodes are selected as the backtracking starting point, tracking depth and path weight thresholds are set, and reverse tracking parameters are initialized.

[0057] Based on the reverse tracing parameters, reverse path search is performed on the collaborative path graph model to identify the upstream behavior path that can be traced back from the target behavior node and generate a set of candidate reverse paths;

[0058] Perform behavioral pattern similarity analysis and path convergence calculation on the candidate reverse path set, identify path nodes that are common to multiple users in the behavioral path, and generate path convergence node data.

[0059] By combining path convergence nodes and reverse path behavior sequences, user behavior paths are clustered and summarized, and behavior reverse path structure patterns are extracted to generate behavior reverse path data.

[0060] By integrating path convergence node data with behavior-based reverse path data, collaborative path reverse tracing data is generated.

[0061] This invention constructs a directed graph model of collaborative path graphs by nodeing user behavior events and edge-directing behavior transitions. This enables a systematic and visual representation of complex user behavior sequences, facilitating in-depth analysis of behavior evolution. Using the target behavior event as the starting point for backtracking, and combining tracking depth and path weight thresholds, the scope and accuracy of reverse path search can be flexibly adjusted to meet different analytical needs, improving the targeting and effectiveness of tracking. Through the reverse path search algorithm, it is possible to quickly backtrack from the target behavior node to multiple upstream behavior paths, uncovering hidden user behavior precursor events and enriching the understanding of user decision-making processes. By comparing the similarity of behavior patterns and calculating path convergence of candidate reverse path sets, key path nodes where multiple users intersect are effectively identified, reflecting the commonalities and key touchpoints of user behavior and promoting the discovery of behavioral patterns. Combining path convergence nodes and reverse path behavior sequences, cluster analysis is used to summarize behavior paths and abstract typical backward path structure patterns, helping to reveal the internal logic and evolution mechanism of user behavior. By organically integrating path convergence node data and behavior-based reverse path data, a complete collaborative path reverse tracking data system is formed, providing a solid foundation for subsequent applications such as user behavior prediction, anomaly detection, and personalized recommendations. Reverse tracking technology delves into the user behavior chain, accurately locating key nodes and paths, enhancing enterprises' insight into the evolution of user behavior, and providing data support for product optimization and user experience improvement. By identifying path convergence and behavior-based reverse structures among users, it helps reveal collaborative behavior characteristics and group influence mechanisms, supporting more effective marketing strategies and user relationship management.

[0062] Preferably, step S4 includes the following steps:

[0063] Step S41: Based on purchased user interaction behavior data, perform low-power Bluetooth beacon sensing processing to generate user scene sensing data; extract the spatial location information of the user scene sensing data to generate user positioning coordinate data;

[0064] Step S42: Perform scene semantic matching analysis on user location coordinate data and enterprise product profile data to generate user behavior profile data; extract behavioral interest factors from user behavior profile data to generate user preference factor data;

[0065] Step S43: Perform correlation measurement analysis on user preference factor data and enterprise product profile to generate profile correlation data; perform edge server task assignment processing on profile correlation data to generate edge-level recommendation request data;

[0066] Step S44: Perform time window adaptive rearrangement on the edge-level recommendation request data to generate dynamic priority ranking data; perform high correlation matching between the dynamic priority ranking data and product feature vector data to generate personalized user recommendation list data;

[0067] Step S45: Perform client-side visualization processing on the personalized user recommendation list data to generate user recommendation interface view data.

[0068] This invention uses low-power Bluetooth beacons to perceive and process user interaction behavior, efficiently acquiring real-time spatial location information of users. This supports precise perception of the user's specific scenario, improving the timeliness and accuracy of recommendations. By combining spatial location information with enterprise product profiles, it completes scene semantic matching analysis, accurately capturing user behavioral characteristics and interests, generating rich user behavior profiles and preference factor data, providing a precise basis for personalized recommendations. Based on the correlation measurement between user preference factors and enterprise product profiles, it achieves deep matching of user needs and product features, ensuring that recommended content highly matches user interests and purchasing behavior, enhancing user satisfaction and conversion rates. Utilizing edge servers for task allocation and adaptive time window reordering effectively optimizes the processing order of recommendation requests, improving the response speed and real-time performance of the recommendation system, reducing network latency, and enhancing system stability and concurrent processing capabilities. Through dynamic priority sorting and high correlation matching of product feature vectors, it can adjust recommendation strategies in real time, flexibly responding to changes in user behavior, and achieving dynamic updates and precise delivery of personalized recommendation lists. Personalized recommendation results are visualized on the client side, generating a clear, intuitive, and user-friendly recommendation interface view to improve user interaction and increase user acceptance and engagement with recommended content. Edge computing and low-power beacon technology are employed to achieve distributed collaboration between computing and sensing, reducing the load on the central server, supporting real-time recommendations for a large number of users, and improving the overall resource utilization efficiency and scalability of the system.

[0069] Preferably, step S44 involves adaptively rearranging the edge-level recommendation request data using a time window to generate dynamic priority ranking data, including:

[0070] Extract timestamps, request frequencies, user identifiers, and content types from edge-level recommendation request data to construct a set of request event sequences; based on the set of request event sequences, set initial sliding time window parameters, including time window size, sliding step size, and trigger threshold, to generate a basic time window structure;

[0071] Analyze the changes in request density and the distribution of request intervals within a time window, and calculate the dynamic activity index for each time window segment;

[0072] The time window size and sliding step size are adaptively adjusted based on the activity index to form an adaptive time window sequence with local optimal response performance.

[0073] Within each adaptive time window, the importance weight of the request is calculated, where the weighting factors for the importance weight include: the level of the request source node, the popularity of the content, the request frequency, and the response latency requirement;

[0074] Based on the importance weight of each request, the recommended requests within the time window are ranked locally by priority, and the cross-window is optimized by combining the position of the global sliding window to generate dynamic priority ranking data.

[0075] This invention dynamically adjusts the time window size and sliding step, enabling the system to respond more flexibly to changes in request density. This avoids delays or resource waste caused by fixed time windows, ensuring more timely processing of recommendation requests. The adaptive time window design automatically adjusts the processing rhythm to accommodate fluctuations in user activity over different time periods, allowing the system to respond quickly during peak hours and conserve computing resources during low-activity periods, achieving load balancing. By combining multi-dimensional weighting factors such as request source node level, content popularity, request frequency, and response latency, the importance of requests can be accurately assessed, ensuring that critical requests are prioritized and improving recommendation relevance and user satisfaction. The combination of local priority ranking and global sliding window cross-tuning makes request sorting more reasonable, effectively reducing resource conflicts and bottlenecks, and improving the concurrent processing capacity of edge servers and system throughput. Dynamic priority ranking data allows the system to self-adjust according to real-time environmental changes, ensuring that recommendation tasks maintain efficient operation in complex network environments and adapting to diverse user needs and scenario changes. The priority ranking mechanism can be flexibly adjusted according to different user and content characteristics, helping to achieve more accurate personalized recommendations, improving user experience and product conversion rates.

[0076] This specification provides a user recommendation system based on enterprise product profile analysis, used to execute the aforementioned user recommendation method based on enterprise product profile analysis. The user recommendation system based on enterprise product profile analysis includes:

[0077] The enterprise screening module is used to obtain enterprise product information data; perform supply chain traceability based on enterprise product information data to generate enterprise product supply chain information data; and screen related enterprises for enterprise product supply chain information data to generate product-related enterprise information data.

[0078] The product tagging module is used to analyze the production ratio of product components based on the information data of the enterprise's product supply chain information data associated with the product, and generate the production ratio value of product components; based on the production ratio value of product components, the module quantifies the product component input tags of the enterprise's product information data, and generates product tag data.

[0079] The user interaction analysis module is used to construct enterprise product profiles based on product tag data; collect purchasing user information data using the constructed enterprise product profiles; analyze the user interaction behavior of purchasing user information data to generate purchasing user interaction behavior data, which includes time-series offset browsing scanning and collaborative path reverse behavior tracking.

[0080] The personalized recommendation module is used to locate user scenarios based on low-power Bluetooth beacons by purchasing user interaction behavior data, and dynamically generate personalized user recommendation lists based on edge servers to perform user recommendation tasks based on enterprise product profile analysis.

[0081] The beneficial effects of this invention lie in the fact that, through the enterprise screening module, the system not only acquires enterprise product information data but also conducts in-depth supply chain node analysis based on supply chain traceability. Combined with a supply intensity assessment model, it identifies key related enterprises, ensuring transparency and traceability of upstream and downstream nodes. The product tagging module calculates the component production ratio based on enterprise capacity allocation data and enterprise reputation factors, thereby achieving quantitative input analysis of tags. This gives product tags traceability and quantifiability, laying the foundation for subsequent personalized recommendations. The user interaction analysis module integrates temporal behavior offset recognition, browsing scanning analysis, and collaborative path reverse tracing technologies to extract users' true interaction intentions and potential interests from access data, generating structured, high-quality user behavior profiles. The personalized recommendation module combines low-power Bluetooth beacons to achieve precise user positioning. Based on semantic matching between the user's actual physical scene and product profile, it performs interest inference. Edge computing enables dynamic sorting and rapid response of recommendation requests, improving the real-time performance and relevance of recommendations. The collaborative design of the modules allows for efficient coupling of product tags and user profiles. Combined with the task allocation capabilities of edge servers, it significantly reduces the central computing burden, improves response speed and system throughput, and adapts to concurrent recommendation needs across multiple scenarios. The functional boundaries between modules are clearly defined, and the data interfaces are designed with standardization, facilitating rapid integration and expansion across different industry chains, product structures, and user behavior models, thus exhibiting excellent system portability and adaptability. Therefore, this invention improves the accuracy and comprehensiveness of user recommendations by integrating product supply chain traceability, component proportion quantification, in-depth user behavior analysis, and scenario-aware recommendation. Attached Figure Description

[0082] Figure 1A flowchart illustrating the steps of a user recommendation method based on enterprise product profile analysis;

[0083] Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S1.

[0084] Figure 3 for Figure 1 A detailed flowchart illustrating the implementation steps of step S2.

[0085] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0086] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0087] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0088] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0089] To achieve the above objectives, please refer to Figures 1 to 3 A user recommendation method based on enterprise product profile analysis, the method comprising the following steps:

[0090] Step S1: Obtain enterprise product information data; perform supply chain traceability based on enterprise product information data to generate enterprise product supply chain information data; filter related enterprises in the enterprise product supply chain information data to generate product related enterprise information data;

[0091] Step S2: Analyze the production ratio of product components based on the product-related enterprise information data and the enterprise product supply chain information data to generate product component production ratio values; quantify the product component input tags based on the product component production ratio values ​​and the enterprise product information data to generate product tag data.

[0092] Step S3: Construct enterprise product profiles based on product tag data and enterprise product information data; collect purchase user information data using the constructed enterprise product profiles; analyze user interaction behavior of purchase user information data to generate purchase user interaction behavior data, including time-series offset browsing scan and collaborative path reverse behavior tracking.

[0093] Step S4: Purchase user interaction behavior data to locate user scenarios based on low-power Bluetooth beacons, and dynamically generate personalized user recommendation lists based on edge servers to perform user recommendation tasks based on enterprise product profile analysis.

[0094] This invention, by acquiring enterprise product information data and conducting supply chain traceability analysis, can comprehensively grasp the source of product components, production processes, and key nodes, improving supply chain transparency and controllability, and facilitating accountability and optimization. Utilizing the production proportion analysis of product components, it can quantitatively describe the input ratio of each component in the product, aiding in product cost optimization, key component identification, and refined management. Enterprise product profiles constructed based on tagged quantitative data can display product attributes, market positioning, and technical characteristics from multiple dimensions, forming core data support for user behavior analysis and recommendation. Collecting and analyzing user information and their interactive behaviors such as time-series browsing offsets and collaborative path reverse tracing enables deep insights into user intent, providing behavioral basis for personalized services. Leveraging low-power Bluetooth beacon technology to achieve user scenario awareness, combined with dynamic recommendation calculations on edge servers, effectively improves the real-time performance, accuracy, and personalization of user recommendations. Deeply integrating enterprise product profiles with user behavior data achieves closed-loop data flow and intelligent linkage between the enterprise and user ends, enhancing the intelligence level of product marketing strategies. Through dynamic personalized recommendations and interactive behavior feedback mechanisms, enterprises can adjust product strategies in a timely manner to better meet users' personalized needs, enhancing user stickiness and brand loyalty. Therefore, this invention improves the accuracy and comprehensiveness of user recommendations by integrating product supply chain traceability, component proportion quantification, in-depth user behavior analysis, and scenario-aware recommendation.

[0095] In this embodiment of the invention, reference is made to Figure 1 The diagram shown illustrates the steps of a user recommendation method based on enterprise product profile analysis according to the present invention. In this example, the user recommendation method based on enterprise product profile analysis includes the following steps:

[0096] Step S1: Obtain enterprise product information data; perform supply chain traceability based on enterprise product information data to generate enterprise product supply chain information data; filter related enterprises in the enterprise product supply chain information data to generate product related enterprise information data;

[0097] Step S2: Analyze the production ratio of product components based on the product-related enterprise information data and the enterprise product supply chain information data to generate product component production ratio values; quantify the product component input tags based on the product component production ratio values ​​and the enterprise product information data to generate product tag data.

[0098] Step S3: Construct enterprise product profiles based on product tag data and enterprise product information data; collect purchase user information data using the constructed enterprise product profiles; analyze user interaction behavior of purchase user information data to generate purchase user interaction behavior data, including time-series offset browsing scan and collaborative path reverse behavior tracking.

[0099] Step S4: Purchase user interaction behavior data to locate user scenarios based on low-power Bluetooth beacons, and dynamically generate personalized user recommendation lists based on edge servers to perform user recommendation tasks based on enterprise product profile analysis.

[0100] In this embodiment of the invention, enterprise product information data is acquired, including product number, category, component information, production batch, raw material details, etc. Based on this information, a blockchain or supply chain management system interface is invoked to perform supply chain traceability operations, tracking the entire process of the product from raw material procurement, component manufacturing, assembly to final delivery, generating enterprise product supply chain information data. Subsequently, all upstream and downstream enterprises in the supply chain data are analyzed. By setting screening conditions such as cooperation frequency, product component importance, and transaction amount, related enterprise screening is performed to extract the structure of enterprises with key contribution relationships, generating product-related enterprise information data. Based on the product-related enterprise information data, the functional component responsibility of each enterprise in the entire product supply chain is identified, and the production proportion of each enterprise's responsible product component is weighted and calculated in conjunction with the bill of materials (BOM) and production records to form a product component production proportion value (e.g., weighted by the number or value of key components). Subsequently, based on this production proportion value, product component input labels are quantified for each component in the enterprise product information data, that is, a weighted label is attached to each component to identify its importance in the whole product, supply risk level, etc., ultimately generating structured product label data. Using product tag data as the core, the system encodes the features of all a company's products to construct a hierarchical product profile. This profile can include dimensions such as product composition distribution, supply chain risk level, component stability indicators, and pricing elasticity index. Based on the constructed product profile, the system collects purchase user information data through methods such as QR code activation, terminal login, and after-sales registration. Further analysis of user behavior paths within this data identifies user interaction behaviors, including but not limited to time-series offset browsing scans (i.e., delayed attention to specific products or components) and collaborative path reverse behavior tracking (i.e., alternating comparisons between similar products). This ultimately generates purchase user interaction behavior data with behavioral pattern characteristics. Based on this user interaction behavior data, when users enter offline stores, showrooms, or industrial display scenarios, the system uses Bluetooth Low Energy (BLE) beacons for real-time location tracking. Combined with the user's historical behavior data, current product display location, and interaction hotspots, a personalized user recommendation list is dynamically generated via an edge computing server. The recommendation list will fully integrate enterprise product profile dimensions (such as high-component weight products, high-stability products, and key enterprise-supplied products) with users' personalized preferences and interaction paths, thereby accurately executing user recommendation operations based on enterprise product profile analysis and improving product guidance efficiency and conversion rate.

[0101] As an example of the present invention, reference is made to... Figure 2 As shown, in this example, step S1 includes:

[0102] Step S11: Obtain enterprise product information data;

[0103] Step S12: Standardize the fields of the acquired enterprise product information data to generate standardized product information data; based on the standardized product information data, track the source of raw materials, processing nodes and circulation records of each batch of products by calling the enterprise resource planning system and product barcode database, and integrate the supply chain data links to generate the initial path data of the enterprise product supply chain.

[0104] Step S13: Classify the initial path data of the enterprise's product supply chain into nodes, identify the key path components of raw material suppliers, processing and manufacturing enterprises, and warehousing and logistics nodes, and generate supply chain sub-node information data; organize the link topology of the supply chain sub-node information data, construct a multi-level supply chain link map, eliminate redundant or broken links, and generate enterprise product supply chain information data.

[0105] Step S14: Calculate the supply intensity value of each node based on the enterprise product supply chain information data, and use the supply intensity value to screen related enterprises and generate a candidate dataset of product related enterprises; perform risk source cross-comparison on the candidate dataset of product related enterprises, screen out upstream and downstream enterprises with high stability, and finally generate product related enterprise information data.

[0106] In this embodiment of the invention, core product-related information data is obtained by connecting to the enterprise's Product Information Management System (PIMS), Enterprise Resource Planning System (ERP), and historical order database. This data includes, but is not limited to, fields such as product code, product name, specifications, production batch, packaging barcode, manufacturing date, and business unit. The data format is a structured table, typically presented in JSON, CSV, or database table format. The data collection process supports API calls or direct database connections to ensure real-time data synchronization. Update records are captured daily or hourly in batch processing to generate raw enterprise product information data, providing basic input for subsequent processing. The raw enterprise product information data undergoes field standardization processing, unifying the naming rules, data types, and encoding formats of each field. For example, synonymous fields such as "productID," "product number," and "product barcode" are uniformly mapped to "ProductCode," and the date format is unified to ISO 8601 (YYYY-MM-DD). Missing fields are filled with null values ​​or filtered to generate standardized product information data. After standardization, the system automatically calls the raw material record interface integrated with the ERP system and the traceability module associated with the product barcode database to perform traceability analysis on each standardized product. Based on the product's batch number and barcode information, the system tracks key supply chain events such as the source of raw materials, production and processing nodes, quality inspection and packaging stages, and warehousing and logistics records. This process supports graph structure storage, constructing a chain-like path structure for each product from the starting point (raw materials), through transit (processing and manufacturing, quality inspection and packaging), to the end point (finished product shipment), ultimately generating initial supply chain path data for the enterprise. Based on this initial supply chain path data, structural parsing and semantic recognition algorithms are used to classify various nodes involved in the path. Through node attribute matching and role label recognition, key links such as raw material supplier nodes, processing and manufacturing enterprise nodes, and warehousing and logistics nodes are distinguished, and corresponding node role data tables are established, generating supply chain sub-node information data. Subsequently, based on this data, graph modeling algorithms (such as NetworkX or Neo4j) are used to construct a multi-level supply chain link graph, hierarchically sorting and visually modeling the dependencies between links, supporting hierarchical retrieval and interactive queries. During the graph construction process, the system automatically identifies isolated nodes, broken links, and duplicate paths. Redundant nodes are pruned and cleaned up by setting weights and connection rules, ultimately generating streamlined and optimized enterprise product supply chain information data. Based on this data, a supply strength index calculation model is used to quantitatively analyze the stability and correlation of nodes at each level. The supply strength index calculation model comprehensively considers indicators such as node supply frequency, historical delivery cycle, supply continuity, and probability of abnormal interruptions. The calculation formula is as follows: In the formula, For the first The supply strength value of each node, For the first The supply frequency of each node, For the first The reliability of each node's performance. For the first Delivery percentage of each node, , , For custom weighting coefficients, the default values ​​are 0.4, 0.3, and 0.3. Based on the calculated intensity values, the companies represented by each node are filtered for correlation, forming a candidate dataset of product-related companies. Next, by calling an enterprise risk monitoring platform, the financial risk, credit rating, legal proceedings records, industry compliance, and other multi-dimensional risk source data of each company in the candidate dataset are cross-referenced to identify upstream and downstream companies with lower risk levels, stable historical cooperation, and strong supply reliability, ultimately generating product-related company information data.

[0107] Preferably, step S14, which calculates the supply intensity value of each node based on the enterprise's product supply chain information data and uses the supply intensity value to screen related enterprises, includes:

[0108] The enterprise product supply chain information data is structured and layered to identify supply nodes at all levels, including raw material supply nodes, production and processing nodes, assembly and integration nodes, and distribution nodes, generating multi-level supply chain node structure data;

[0109] For each node in the multi-level supply chain node structure data, a supply record statistical matrix is ​​constructed to generate node supply statistics. The supply record statistical matrix includes historical supply quantity, annual fulfillment frequency, and cooperation cycle duration.

[0110] The statistical data on supply at each node are quantified and standardized, and the supply frequency, supply stability and delivery ratio of each node are calculated. A multi-dimensional supply capacity parameter table is constructed to generate supply capacity characteristic data.

[0111] The node supply strength formula is defined based on supply capacity characteristic data, as shown below:

[0112] ;

[0113] In the formula, For the first The supply strength value of each node, For the first The supply frequency of each node, For the first The reliability of each node's performance. For the first Delivery percentage of each node, , , Custom weighting coefficients;

[0114] The supply strength value of each node is calculated using the node supply strength formula to generate supply strength index data. The supply strength index data is then classified into levels, supply strength boundary lines are set, and strong, medium, and weak categories are defined. Related nodes with supply strength at medium to high levels are selected to generate strong supply node screening data.

[0115] The data of strong supply nodes are screened and merged for enterprise uniqueness and supply chain role identification. Redundant or overlapping enterprises are eliminated, and finally a candidate dataset of product-related enterprises is generated.

[0116] In this embodiment of the invention, by performing structural layering processing on enterprise product supply chain information data and combining path dependencies and node role labels in the supply chain graph, different levels of supply node types are identified, including raw material supply nodes, production and processing nodes, assembly and integration nodes, and product distribution nodes. By classifying and organizing node attribute fields (such as supplier code, processing plant name, warehouse location, channel provider identifier, etc.), a multi-level chain-structured data model is constructed, generating multi-level supply chain node structure data, facilitating subsequent vertical and horizontal feature calculations between nodes. Next, based on the multi-level supply chain node structure data, a set of supply record statistical matrices is constructed for each node. This matrix comprehensively collects key quantitative data from the node's past supply behavior, including the total historical supply quantity, the number of annual fulfillments in the past three or five years, and the cumulative period of cooperation with target enterprises (in months or quarters). This data is automatically aggregated and statistically analyzed through the database and populated into the node attribute table, generating node supply statistics data as the original input dataset for supply behavior stability. Then, the node supply statistics data undergo unified quantitative indicator standardization processing. The Z-score or Min-Max normalization method is used to standardize core indicators such as supply frequency (number of deliveries / cycle duration), supply stability (fulfillment success rate), and delivery ratio (supply volume at this node / total product usage), constructing a structured multi-dimensional supply capacity parameter table. Based on this, supply capacity characteristic data for each node is generated, ensuring that all dimensions can be comprehensively evaluated within the same order of magnitude. On this basis, a quantitative calculation formula for node supply intensity is defined to measure the supply capacity and reliability of each node in the supply chain. The node supply intensity formula is shown below: In the formula, For the first The supply strength value of each node, For the first The supply frequency of each node, For the first The reliability of each node's performance. For the first Delivery percentage of each node, , , The system allows for custom weighting coefficients; weights can be flexibly configured based on the enterprise's preferred dimensions (e.g., 0.4:0.4:0.2 or other ratios). According to this formula, the system automatically performs batch calculations for each node, generating standardized supply intensity index data. Subsequently, the supply intensity index data is graded, setting supply intensity boundaries based on historical distribution ranges, enterprise experience thresholds, or quantile methods (e.g., upper and lower quartiles), classifying nodes into high, medium, and low intensity levels. The system prioritizes nodes in the medium-to-high level (e.g., supply intensity values ​​in the top 60%) as reliable supplier candidates, forming the strong supply node screening data. Finally, the strong supply node screening data undergoes enterprise uniqueness merging processing. Duplicate records of the same enterprise across multiple nodes are identified using unified enterprise identification codes (e.g., unified social credit code, organization code, etc.), and the merging operation is performed along with supply chain role identification to ensure that an enterprise is not repeatedly selected due to role diversity. The system further eliminates logically redundant or path-repeating enterprises, ultimately outputting a deduplicated product-related enterprise candidate dataset for subsequent risk assessment, cooperation recommendation, or strategic supplier management modules.

[0117] As an example of the present invention, reference is made to... Figure 3 As shown, in this example, step S2 includes:

[0118] Step S21: Map the supply chain path of the product-related enterprise information data, identify the upstream supply source and actual production division ratio of each product component, and generate product component source path data;

[0119] Step S22: Accumulate and statistically analyze the production records of various components in the enterprise's product supply chain information data, and combine them with the product component source path data to extract the actual output ratio of different enterprises at the component granularity, and generate component production allocation matrix data.

[0120] Step S23: Normalize and integrate the component production allocation matrix data to generate standardized product component production percentage data; based on the product component production percentage data, construct a component tag input function:

[0121] ;

[0122] In the formula, For product components The tag weight value, For the first Individual companies for components The production share, For the first Individual enterprise credit rating; using the component label input function to calculate the label weight of the product component production ratio data, and generate component label quantitative data;

[0123] Step S24: Embed tags item by item in the product component list in the enterprise product information data, associate and bind the quantitative data of component tags with specific components, and generate structured and tagged product list data; perform unified product ID labeling, tag weight aggregation and version timestamp synchronization processing on the structured and tagged product list data, and finally generate product tag data.

[0124] In this embodiment of the invention, supply chain path mapping is performed on product-related enterprise information data. The system, by calling a pre-constructed multi-level supply chain map and combining it with the product's Bill of Materials (BOM), traces the actual source path of each component in the supply chain network, identifying the upstream suppliers, transmission nodes, and processing units for each component layer by layer. Combining historical procurement records and order coordination data, the system determines the production responsibility of the actual participating enterprises in the component source path, clarifying the actual supply ratio and production role of each enterprise for that component, thereby generating product component source path data as the basis for subsequent weight allocation. Based on historical production and delivery records in the product supply chain information data, the system focuses on the production responsibility attribution of various components, performing cumulative statistics by time period (e.g., year, quarter) to obtain enterprise-level output data for each component. Combining the product component source path data generated in step S21, the system matches enterprise output records item by item at the component level, statistically analyzes the output quantity and frequency of each enterprise for specific components, and standardizes this into a percentage format (e.g., a component is produced 60% by enterprise A and 40% by enterprise B), forming a two-dimensional mapping relationship. Finally, a component production allocation matrix is ​​constructed, where rows represent component numbers, columns represent participating enterprises, and percentage values ​​are filled into matrix elements. The component production allocation matrix data is normalized, removing abnormal missing items or duplicate data, and standardizing each percentage value to a uniform scale within the 0-1 range. Subsequently, based on the normalized data, the component tag input function is constructed as follows:

[0125] In the formula, For product components The tag weight value, For the first Individual companies for components The production share, For the first The system calculates the credit rating of each enterprise; it also uses a component tagging input function to calculate the tag weights on the production proportion data of product components, generating quantitative component tag data; this function multiplies and sums each enterprise's participation ratio in a component with its credit rating to form a tag weight index reflecting the overall supply quality of components. The system iterates through and calculates the credit ratings of all components. The system outputs corresponding quantitative data for component labels, used for subsequent product lifecycle identification. It structures the component list in the enterprise's product information data and embeds corresponding label weight values ​​for each item, using the component number as the primary key. The system binds label values ​​to components, generating structured product component list data with quantitative label characteristics. Further processing is performed on this data, including unified product ID labeling (e.g., associating with product serial numbers or SKUs), label weight aggregation (e.g., average label score or weighted score for the entire product), and version timestamp processing (recording label generation time, version number, etc.). Finally, the system outputs standardized product label data that can be used for product traceability, quality evaluation, and risk assessment, for use in regulatory, supply chain assessment, or consumer visualization scenarios.

[0126] Preferably, step S3, which involves constructing an enterprise product profile based on product tag data, includes:

[0127] The product label data is categorized by label type to generate aggregated label category data.

[0128] Calculate the label density of the tag category aggregated data;

[0129] Weight normalization is performed on the label density data to generate label weight distribution data;

[0130] Product feature vectors are constructed based on label weight distribution data, and principal component fusion is performed on the product feature vector data to generate principal component fused feature data.

[0131] The principal component fusion feature data is processed to construct a label map, generating label map structure data.

[0132] Embedding and mapping of tag graph structure data to generate enterprise product profile vector data;

[0133] Spatial dimensionality reduction is performed on the enterprise product profile vector data to generate enterprise product profile data.

[0134] In this embodiment of the invention, product label data is categorized by label type. Based on the label's source attributes, performance dimensions, and evaluation objectives, the system divides product labels into multiple categories, such as quality labels (e.g., supply strength, raw material grade), supply labels (e.g., fulfillment timeliness, delivery stability), structural labels (e.g., key component coverage, structural complexity), and safety labels (e.g., traceability credibility, quality inspection compliance). Through clustering and classification operations, a clearly structured aggregated label category data is formed to support refined modeling for subsequent label analysis. Label density is calculated for each label category in the aggregated data. Label density reflects the coverage and refinement of that label category across all products; calculation methods may include label quantity density, label dimension distribution frequency, etc. Label density values ​​measure the representativeness and expressive power of a particular label category towards the overall product information structure, thus providing a data foundation for modeling. After obtaining the label density, the system performs weight normalization processing, unifying the influence of different label categories within the overall label system to the same dimensional range, generating label weight distribution data. This weighted data can serve as a weighting factor when combining features, assigning higher weights to important labels and reducing the interference of secondary labels. Based on the label weight distribution data, the system constructs multi-dimensional product feature vectors for each product, combining all label values ​​according to the label category structure to form a vector representation. For high-dimensional product feature vectors, the system further employs methods such as Principal Component Analysis (PCA) or Independent Component Analysis (ICA) for principal component fusion processing, extracting key components with the largest explained variance from multiple label dimensions to generate principal component fusion feature data. Subsequently, the system performs label graph construction processing based on the principal component fusion feature data. The label graph models the relationships, similarities, or dependencies between product labels as a graph structure representation, where nodes represent various label elements, edges represent the weight relationships between labels, and edge weights can be calculated using methods such as co-occurrence frequency, correlation coefficient, or mutual information, ultimately forming label graph structure data. Next, the system performs embedding mapping on the tag graph structure data. It uses graph neural networks (GNNs), node embedding algorithms (such as Node2Vec and DeepWalk), or graph autoencoder methods (such as Graph AutoEncoder) to transform the graph structure into a dense low-dimensional vector representation, generating enterprise product profile vector data. This vector comprehensively reflects the product's tag features, semantic relationships between tags, and the overall tag distribution pattern. Finally, to facilitate visualization and multi-dimensional analysis, the system performs spatial dimensionality reduction on the enterprise product profile vector data. It uses non-linear dimensionality reduction algorithms such as t-SNE, UMAP, or LLE to compress the high-dimensional vectors into two-dimensional or three-dimensional space, ultimately outputting enterprise product profile data that can be used for evaluation, retrieval, clustering, or trend analysis.

[0135] Preferably, the user interaction behavior analyzed in step S3 for purchasing user information data includes:

[0136] Extract user access records and page interaction logs from the purchase user information data;

[0137] Reconstruct user behavior trajectories based on user access records to generate behavior path data;

[0138] Calculate the time-series features of user behavior path data to obtain the user's temporal offset features;

[0139] Based on time-series offset features, the user's browsing and scanning patterns within a specific time window are identified, and browsing and scanning behavior data is generated.

[0140] Based on behavioral path data, construct a collaborative path graph among users, identify existing collaborative path behaviors, and generate collaborative path behavior data.

[0141] Based on the collaborative path behavior data, reverse behavior path tracing is performed on the collaborative path graph to identify path convergence nodes and behavior-based reverse paths, generating collaborative path reverse tracing data.

[0142] By fusing browsing and scanning behavior data and collaborative path reverse tracing data, user interaction behavior pattern characteristics are identified and user interaction behavior pattern data is generated.

[0143] Based on user interaction behavior pattern data, construct structured purchase user interaction behavior data.

[0144] In this embodiment of the invention, raw user access records and page interaction logs are extracted from purchased user information data. This type of data includes user access timestamps, page click paths, page dwell time, page scroll depth, and interactive control click events. By performing structured parsing on this raw data, the system can obtain the complete interaction trajectory of the user on the product platform or service page. Next, based on the user access records, the system reconstructs the user behavior trajectory. Through time sorting and event recognition methods, the continuous access behaviors of the same user within a certain time window are serialized and organized to form an identifiable user behavior path chain, generating behavior path data. Behavior path data is used to represent the user's specific browsing order, jump logic, and dependencies between pages of interest on the platform. Subsequently, the system performs time series feature calculations on the behavior path data, including user click rhythm, page dwell time distribution, and behavior periodicity, further generating the user's temporal offset features. These features reveal the user's behavioral tendencies and activity patterns at different time periods. After obtaining the temporal offset features, the system identifies the user's browsing scanning pattern within a specific time window based on these features. Through clustering analysis or pattern matching, the system identifies user behaviors such as linear browsing, cross-jumping, and backtracking, ultimately generating browsing and scanning behavior data to analyze how users focus on specific products or information sections. Simultaneously, the system constructs a collaborative path graph among users using behavioral path data. By comparing the overlap and similarity of behavioral paths among different users, it identifies collaborative behavior patterns such as shared information reading and shared product searching, generating collaborative path behavior data. This data can be used to reveal the commonalities and potential correlations in behavior among different user groups. Subsequently, the system performs reverse behavioral path tracing on the collaborative path graph based on the collaborative path behavior data. By reverse analyzing convergence points (i.e., nodes where multiple user behaviors converge) and their upstream behavioral paths, the system can identify key influencing factors in the user's decision-making path, generating collaborative path reverse tracing data for analyzing group behavior attribution and identifying guiding nodes. The browsing and scanning behavior data and the collaborative path reverse tracing data are fused together to extract comprehensive pattern features of individual user behavior and group collaborative behavior, identifying the behavioral decision-making logic, interest shift points, and key behavioral triggering nodes of different users, ultimately generating user interaction behavior pattern data. The user interaction behavior pattern data is structured and organized according to dimensions such as user ID, behavior type, behavior time sequence characteristics, collaboration path characteristics, and key node tags to construct structured purchase user interaction behavior data.

[0145] Of particular importance, based on temporal offset features, identifying a user's browsing and scanning patterns within a specific time window also includes:

[0146] Based on temporal offset features, the browsing and scanning patterns of users within a specific time window are identified, and browsing and scanning behavior data is generated; the time window slices of the interaction behavior time series data are then generated to generate multi-segment time window behavior fragment data.

[0147] Offset difference calculation and rhythm density analysis are performed on multi-segment time window behavior fragment data to generate time series offset feature data; user dwell period identification and scroll jump pattern analysis are performed on time series offset feature data to generate browsing rhythm pattern data.

[0148] Focused area identification and page hotspot trajectory mapping are performed on browsing rhythm pattern data to generate page scanning path marker data; clustering extraction and feature map construction are performed on page scanning path marker data to generate user scanning pattern feature map data;

[0149] Behavioral labels are extracted and user behavior is categorized from user scanning pattern feature map data to ultimately generate browsing and scanning behavior data.

[0150] In this embodiment of the invention, core parameters such as behavior timestamps, dwell time, scrolling speed, and jump frequency are extracted from user page browsing behavior interaction data on digital terminal devices to generate interactive behavior time-series data. Subsequently, the interactive behavior time-series data is sliced ​​into time windows according to a set time interval, constructing multi-segment time-window behavior fragment data with a uniform time length or user activity cycle length. Each behavior fragment represents a continuous browsing activity sequence of the user within a time window. Then, offset difference calculation is performed on each time-window behavior fragment data to identify the time difference sequence between adjacent behavior events and calculate the rhythm density index (i.e., the number of behavior events per unit time), generating time-series offset feature data. Based on this feature data, user dwell cycle characteristics (e.g., fast browsing, slow reading, short dwell time) and scrolling jump patterns (e.g., continuous scrolling, intermittent jumping, area dwell time) are identified, and the output is browsing rhythm pattern data. Further, based on the browsing rhythm pattern data, combined with page structure and area identification information, the view position coordinates in the user behavior data are analyzed to identify the user's focused area on the page. The user's browsing path is reconstructed through hotspot trajectory mapping technology, forming page scanning path marking data. Page scanning path marker data includes hotspot area numbers, scanning order, and dwell intensity. Next, multi-user clustering is performed on the page scanning path marker data. Clustering algorithms (such as K-Means or DBSCAN) are used to identify common patterns in user scanning behavior, thereby constructing a multi-dimensional feature map and generating user scanning pattern feature map data. This map reflects the structured patterns of users' browsing methods, attention distribution, and behavioral rhythm. Finally, behavioral labels are extracted from the user scanning pattern feature map data, including labels such as "quick-skipping reader," "deeply focused reader," and "regionally focused interest reader." Based on this label system, users are categorized into corresponding behavioral characteristic groups, ultimately generating browsing and scanning behavior data. This data can be used for applications such as user intent prediction, content optimization and recommendation, or intelligent page layout adjustment.

[0151] Preferably, the collaborative path graph is traced backwards based on the collaborative path behavior data to identify path convergence nodes and behavior-based reverse paths, including:

[0152] Construct a directed graph model of the collaborative path graph, where nodes represent user behavior events and edges represent behavior transformation relationships, and generate collaborative path graph model data;

[0153] Based on collaborative path behavior data, target behavior event nodes are selected as the backtracking starting point, tracking depth and path weight thresholds are set, and reverse tracking parameters are initialized.

[0154] Based on the reverse tracing parameters, reverse path search is performed on the collaborative path graph model to identify the upstream behavior path that can be traced back from the target behavior node and generate a set of candidate reverse paths;

[0155] Perform behavioral pattern similarity analysis and path convergence calculation on the candidate reverse path set, identify path nodes that are common to multiple users in the behavioral path, and generate path convergence node data.

[0156] By combining path convergence nodes and reverse path behavior sequences, user behavior paths are clustered and summarized, and behavior reverse path structure patterns are extracted to generate behavior reverse path data.

[0157] By integrating path convergence node data with behavior-based reverse path data, collaborative path reverse tracing data is generated.

[0158] In this embodiment of the invention, a directed graph model of the collaborative path graph is constructed. User behavior events are used as nodes in the graph, and behavior transformation relationships are used as directed edges. Combined with attributes such as the timestamp of the behavior occurrence and the transformation frequency, a complete collaborative path graph model data is generated, providing a structured foundation for subsequent path tracing. Secondly, based on the collaborative path behavior data, the target behavior event node is selected as the starting point for reverse tracing. The tracing depth (i.e., the maximum number of reverse backtracking levels) and path weight threshold (used to filter low-weight or low-frequency paths) are set according to business requirements, and the parameters required for the reverse tracing algorithm are initialized. Then, the reverse path search algorithm is executed on the collaborative path graph model using the above parameters. Through a depth-first or breadth-first strategy, all paths tracing back from the target behavior node to the upstream behavior node are identified and collected, forming a candidate reverse path set, covering the user behavior starting point and intermediate key nodes. Next, behavior pattern similarity analysis is performed on the candidate reverse path set. Combining indicators such as node frequency, path length, and path weight, the convergence degree of the path is calculated, and key nodes where multiple user behavior paths intersect and converge, i.e., path convergence nodes, are identified, forming a path convergence node dataset. Subsequently, by combining path convergence nodes and reverse path behavior sequences, clustering algorithms (such as spectral clustering, hierarchical clustering, or density-based clustering methods) are applied to summarize and refine user behavior paths, extracting typical behavior-based reverse path structure patterns and generating behavior-based reverse path data. Finally, the path convergence node data and behavior-based reverse path data are merged and integrated to form a systematic collaborative path reverse tracking data.

[0159] Of particular importance is the analysis of behavioral patterns and calculation of path convergence for the candidate reverse path set, which identifies path nodes that converge in common among multiple users within the behavioral path, and also includes:

[0160] The candidate reverse path set data is standardized by user behavior sequence to generate path behavior standard sequence data; the path behavior standard sequence data is time-aligned and behavior action encoded to generate unified time-action matrix data.

[0161] Multi-path behavior similarity clustering analysis is performed on unified time-action matrix data to generate behavior pattern similarity clustering results data; path intersection points are extracted from the behavior pattern similarity clustering results data, and path convergence frequency weighted calculation and node transfer stability evaluation are performed on the path intersection points to generate path convergence score data;

[0162] Threshold filtering and high-frequency node selection are performed on the path convergence score data to generate common high-frequency convergence node data for the path.

[0163] The path convergence node data is generated by verifying the user source distribution and checking the consistency of behavior orientation on the common high-frequency convergence node data.

[0164] In this embodiment of the invention, a unified coding system for behavioral events is constructed by standardizing the behavioral sequences of different users in the candidate reverse path set data. User behavioral events are uniformly represented as behavioral action sequences in chronological order, and the event granularity and labeling method are standardized to generate standard sequence data of path behaviors. Subsequently, the standard sequence data of path behaviors is time-aligned. Dynamic Time Warping (DTW) or sliding window matching methods are used to ensure consistent sequence lengths for behavioral paths of different durations. A unified time-action matrix is ​​constructed based on the behavioral action codes, where rows represent different users, columns represent time slices, and cells contain the behavioral action codes for the corresponding time slices. Next, multi-path behavior similarity clustering analysis is performed on the unified time-action matrix data. Hierarchical clustering or density clustering (such as DBSCAN) methods are used to cluster based on the edit distance or Jaccard similarity between behavioral action sequences, resulting in behavioral pattern similarity clustering results. Based on the clustering results, frequently occurring path intersections (i.e., nodes that appear repeatedly in multiple user paths) are extracted from each cluster. The path convergence frequency (the percentage of times a node appears within a cluster) and node transition stability (the inverse of the variance of the probability of behavior transition before and after a node) are calculated for these intersections, forming path convergence score data. Then, threshold filtering is applied to the path convergence score data, setting minimum convergence frequency and minimum stability score thresholds to filter out path nodes with significant behavioral commonalities, generating common high-frequency convergence node data. Further, the common high-frequency convergence node data undergoes user source distribution verification (determining whether the convergence node involves multiple user groups) and behavior orientation consistency testing (i.e., whether the node's subsequent behavior tends to be consistent across multiple user paths) to exclude isolated high-frequency nodes and path noise points. Finally, the verified nodes are marked as path convergence nodes, and the output is path convergence node data, which can serve as an important basis for user path backward modeling and collaborative behavior decision-making.

[0165] Preferably, step S4 includes the following steps:

[0166] Step S41: Based on purchased user interaction behavior data, perform low-power Bluetooth beacon sensing processing to generate user scene sensing data; extract the spatial location information of the user scene sensing data to generate user positioning coordinate data;

[0167] Step S42: Perform scene semantic matching analysis on user location coordinate data and enterprise product profile data to generate user behavior profile data; extract behavioral interest factors from user behavior profile data to generate user preference factor data;

[0168] Step S43: Perform correlation measurement analysis on user preference factor data and enterprise product profile to generate profile correlation data; perform edge server task assignment processing on profile correlation data to generate edge-level recommendation request data;

[0169] Step S44: Perform time window adaptive rearrangement on the edge-level recommendation request data to generate dynamic priority ranking data; perform high correlation matching between the dynamic priority ranking data and product feature vector data to generate personalized user recommendation list data;

[0170] Step S45: Perform client-side visualization processing on the personalized user recommendation list data to generate user recommendation interface view data.

[0171] In this embodiment of the invention, BLE (Bluetooth Low Energy) beacon devices (such as iBeacon or Eddystone) are deployed in shopping malls or exhibition halls. Each beacon has a broadcast power of -59dBm and a broadcast interval of 200ms. The user terminal integrates a BLE scanning module with a scanning radius of 20m and a scanning cycle of 500ms. The obtained raw RSSI values ​​are smoothed using a multipath attenuation model (path loss exponent n=2.2) and a Kalman filter algorithm, converting them into distance estimates. Combining distance data from at least three different beacons, a trilateration algorithm is used to calculate the two-dimensional plane coordinates (X,Y), with the error controlled within ±1.5m, generating user scene perception data and corresponding user positioning coordinate data. The system pre-maintains a product profile library in the background, containing attribute tags (such as "technology," "home," and "sports") and scene tags (such as "entrance," "rest area," and "display area") for each product or service. The system maps user location coordinates to a scene map (based on an indoor GIS platform) to identify the user's current scene. Simultaneously, it extracts behavioral events from interaction logs such as terminal touch clicks and browsing duration, mapping them to product profile tags. A TF-IDF weighted model is used to calculate the scene-tag matching degree, generating user behavior profile data. Next, the weights of each tag in the behavior profile are normalized, outputting behavioral interest factors (0–1 range), forming user preference factor data. Cosine similarity is calculated between the user preference factors and the feature vectors in the enterprise product profile (128-dimensional vectors pre-trained using Word2Vec or Doc2Vec models) to obtain a list of profile relevance scores for each product and user. The top 20% of products by relevance and their scores are packaged into edge-level recommendation requests. The request message body includes the user ID, preference tag, and corresponding score threshold (e.g., ≥0.75). These requests are pushed to an edge computing server deployed within the same local area network via MQTT or HTTP / 2, generating edge-level recommendation request data. Upon receiving a recommendation request, the edge server adaptively rearranges the request list based on the request timestamp and a user activity time model (a pre-trained temporal LSTM model predicts the time window when users are most likely to receive push notifications), prioritizing products with high predicted response rates by 10-20%. Subsequently, the rearranged list is matched again with a product feature vector library using high precision (using Annoy or FAISS high-performance vector retrieval libraries), retaining the top 10 most relevant results to generate the final personalized user recommendation list. In the user terminal application, a front-end framework (such as React Native or Flutter) is used to render the recommendation list, displaying product thumbnails, names, similarity scores, and scene matching progress bars using a card-style layout.The recommended card component supports user interaction via swiping, clicking, and saving. The component style is dynamically adjusted via CSS to adapt to different screen sizes and night mode. Meanwhile, the recommendation interface maintains a real-time connection with the backend via WebSocket, automatically refreshing the display when new preference factors are generated or their positions change, forming a smooth user recommendation interface view data.

[0172] Preferably, step S44 involves adaptively rearranging the edge-level recommendation request data using a time window to generate dynamic priority ranking data, including:

[0173] Extract timestamps, request frequencies, user identifiers, and content types from edge-level recommendation request data to construct a set of request event sequences; based on the set of request event sequences, set initial sliding time window parameters, including time window size, sliding step size, and trigger threshold, to generate a basic time window structure;

[0174] Analyze the changes in request density and the distribution of request intervals within a time window, and calculate the dynamic activity index for each time window segment;

[0175] The time window size and sliding step size are adaptively adjusted based on the activity index to form an adaptive time window sequence with local optimal response performance.

[0176] Within each adaptive time window, the importance weight of the request is calculated, where the weighting factors for the importance weight include: the level of the request source node, the popularity of the content, the request frequency, and the response latency requirement;

[0177] Based on the importance weight of each request, the recommended requests within the time window are ranked locally by priority, and the cross-window is optimized by combining the position of the global sliding window to generate dynamic priority ranking data.

[0178] In this embodiment of the invention, the timestamp (accurate to milliseconds), request frequency (the number of times the same content is repeated by the same user within a specified period), user identifier, and content type of each request are extracted from the edge-level recommendation request data. All requests are then constructed into a request event sequence set in ascending order of time. Based on this event sequence, the sliding time window parameters are initialized: the default time window size is set to 300 seconds, the sliding step is 60 seconds, and the trigger threshold is activated when the number of requests within the window exceeds 10 or the same user makes 3 requests. This forms the basic time window structure. Next, the request density (number of requests / window length) and request interval distribution (time difference sequence between adjacent requests) within each basic time window are statistically analyzed, and a dynamic activity index is calculated. Where ρ is the request density of the current window, ρ0 is the global average density, σ_Δt is the standard deviation of the request interval, α=0.6, and β=0.4. If the activity A of a certain time window exceeds a preset high threshold (e.g., 1.2 times the average activity of all windows), the window is reduced to 200 seconds and the sliding step size is adjusted to 40 seconds to improve the response to peak periods; if A is lower than a low threshold (e.g., 0.8 times the average value), the window is expanded to 400 seconds and the step size is increased to 80 seconds to reduce window fragmentation, thereby automatically forming a series of adaptive time window sequences. Then, within each adaptive time window, the importance weight W = ω1·L + ω2·H + ω3·F + ω4·(1 / D) is calculated for the recommended requests within the window, where L is the request source node level (the regional priority of nodes in the edge server cluster, which can be divided into 1–5 levels), H is the content popularity (based on the normalized score of the number of times it has been recommended in the past 24 hours), F is the normalized score of the user request frequency, and D is the response latency requirement (in seconds, the smaller the value, the higher the weight); the default weight coefficients ω1…ω4 are 0.25, 0.35, 0.25, and 0.15, respectively. The request list within each window is sorted locally by W value from largest to smallest. Then, combined with the global sliding window position index, the ranking between adjacent windows is cross-corrected: when the same request appears in multiple windows simultaneously, its final priority is taken from the highest local ranking position and moved down no more than 5 places to ensure cross-window smoothness. Finally, the system merges the local ranking results within all adaptive windows and outputs complete dynamic priority ranking data arranged continuously along the time axis with dynamically changing priorities, which is used for push scheduling and resource allocation of the edge servers.

[0179] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0180] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A user recommendation method based on enterprise product profile analysis, characterized in that, Includes the following steps: Step S1: Obtain enterprise product information data; Based on enterprise product information data, supply chain traceability is performed to generate enterprise product supply chain information data. The system filters related companies from enterprise product supply chain information data and generates product-related company information data. Step S2: Analyze the production ratio of product components based on the product-related enterprise information data and the enterprise product supply chain information data to generate product component production ratio values; Based on the production ratio of product components, the enterprise's product information data is quantified by product component input tags to generate product tag data; Step S3: Construct enterprise product profiles based on product tag data; collect purchasing user information data using the constructed enterprise product profiles; extract user access records and page interaction logs from the purchasing user information data. Reconstruct user behavior trajectories based on user access records to generate behavior path data; Calculate the time-series features of user behavior path data to obtain the user's temporal offset features; Based on time-series offset features, the user's browsing and scanning patterns within a specific time window are identified, and browsing and scanning behavior data is generated. Based on behavioral path data, construct a collaborative path graph among users, identify existing collaborative path behaviors, and generate collaborative path behavior data. Construct a directed graph model of the collaborative path graph, where nodes represent user behavior events and edges represent behavior transformation relationships, and generate collaborative path graph model data; Based on collaborative path behavior data, target behavior event nodes are selected as the backtracking starting point, tracking depth and path weight thresholds are set, and reverse tracking parameters are initialized. Based on the reverse tracing parameters, reverse path search is performed on the collaborative path graph model to identify the upstream behavior path that can be traced back from the target behavior node and generate a set of candidate reverse paths; Perform behavioral pattern similarity analysis and path convergence calculation on the candidate reverse path set, identify path nodes that are common to multiple users in the behavioral path, and generate path convergence node data. By combining path convergence nodes and reverse path behavior sequences, user behavior paths are clustered and summarized, and behavior reverse path structure patterns are extracted to generate behavior reverse path data. By integrating path convergence node data with behavior-based reverse path data, collaborative path reverse tracing data is generated. By fusing browsing and scanning behavior data and collaborative path reverse tracing data, user interaction behavior pattern characteristics are identified and user interaction behavior pattern data is generated. Based on user interaction behavior pattern data, construct structured purchase user interaction behavior data; Step S4: Purchase user interaction behavior data to locate user scenarios based on low-power Bluetooth beacons, and dynamically generate personalized user recommendation lists based on edge servers to perform user recommendation tasks based on enterprise product profile analysis.

2. The user recommendation method based on enterprise product profile analysis according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Obtain enterprise product information data; Step S12: Standardize the fields of the acquired enterprise product information data to generate standardized product information data; based on the standardized product information data, track the source of raw materials, processing nodes and circulation records of each batch of products by calling the enterprise resource planning system and product barcode database, and integrate the supply chain data links to generate the initial path data of the enterprise product supply chain. Step S13: Classify the initial path data of the enterprise's product supply chain into nodes, identify the key path components of raw material suppliers, processing and manufacturing enterprises, and warehousing and logistics nodes, and generate supply chain sub-node information data; organize the link topology of the supply chain sub-node information data, construct a multi-level supply chain link map, eliminate redundant or broken links, and generate enterprise product supply chain information data. Step S14: Calculate the supply intensity value of each node based on the enterprise product supply chain information data, and use the supply intensity value to screen related enterprises and generate a candidate dataset of product related enterprises; perform risk source cross-comparison on the candidate dataset of product related enterprises, screen out upstream and downstream enterprises with high stability, and finally generate product related enterprise information data.

3. The user recommendation method based on enterprise product profile analysis according to claim 2, characterized in that, Step S14 involves calculating the supply intensity value of each node based on the enterprise's product supply chain information data, and using the supply intensity value to screen related enterprises, including: The enterprise product supply chain information data is structured and layered to identify supply nodes at all levels, including raw material supply nodes, production and processing nodes, assembly and integration nodes, and distribution nodes, generating multi-level supply chain node structure data; For each node in the multi-level supply chain node structure data, a supply record statistical matrix is ​​constructed to generate node supply statistics. The supply record statistical matrix includes historical supply quantity, annual fulfillment frequency, and cooperation cycle duration. The statistical data on supply at each node are quantified and standardized, and the supply frequency, supply stability and delivery ratio of each node are calculated. A multi-dimensional supply capacity parameter table is constructed to generate supply capacity characteristic data. The node supply strength formula is defined based on supply capacity characteristic data, as shown below: In the formula, For the first The supply strength value of each node, For the first The supply frequency of each node, For the first The reliability of each node's performance. For the first Delivery percentage of each node, , , Custom weighting coefficients; The supply strength value of each node is calculated using the node supply strength formula to generate supply strength index data. The supply strength index data is then classified into levels, supply strength boundary lines are set, and strong, medium, and weak categories are defined. Related nodes with supply strength at medium to high levels are selected to generate strong supply node screening data. The data of strong supply nodes are screened and merged for enterprise uniqueness and supply chain role identification. Redundant or overlapping enterprises are eliminated, and finally a candidate dataset of product-related enterprises is generated.

4. The user recommendation method based on enterprise product profile analysis according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Map the supply chain path of the product-related enterprise information data, identify the upstream supply source and actual production division ratio of each product component, and generate product component source path data; Step S22: Accumulate and statistically analyze the production records of various components in the enterprise's product supply chain information data, and combine them with the product component source path data to extract the actual output ratio of different enterprises at the component granularity, and generate component production allocation matrix data. Step S23: Normalize and integrate the component production allocation matrix data to generate standardized product component production percentage data; based on the product component production percentage data, construct a component tag input function: In the formula, For product components The tag weight value, For the first Individual companies for components The production share, For the first Individual enterprise credit rating; using the component label input function to calculate the label weight of the product component production ratio data, and generate component label quantitative data; Step S24: Embed tags item by item in the product component list in the enterprise product information data, associate and bind the quantitative data of component tags with specific components, and generate structured and tagged product list data; perform unified product ID labeling, tag weight aggregation and version timestamp synchronization processing on the structured and tagged product list data, and finally generate product tag data.

5. The user recommendation method based on enterprise product profile analysis according to claim 1, characterized in that, Step S3, which involves constructing a corporate product profile based on product tag data, includes: The product label data is categorized by label type to generate aggregated label category data. Calculate the label density of the tag category aggregated data; Weight normalization is performed on the label density data to generate label weight distribution data; Product feature vectors are constructed based on label weight distribution data, and principal component fusion is performed on the product feature vector data to generate principal component fused feature data. The principal component fusion feature data is processed to construct a label map, generating label map structure data. Embedding and mapping of tag graph structure data to generate enterprise product profile vector data; Spatial dimensionality reduction is performed on the enterprise product profile vector data to generate enterprise product profile data.

6. The user recommendation method based on enterprise product profile analysis according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Based on purchased user interaction behavior data, perform low-power Bluetooth beacon sensing processing to generate user scene sensing data; extract the spatial location information of the user scene sensing data to generate user positioning coordinate data; Step S42: Perform scene semantic matching analysis on user location coordinate data and enterprise product profile data to generate user behavior profile data; extract behavioral interest factors from user behavior profile data to generate user preference factor data; Step S43: Perform correlation measurement analysis on user preference factor data and enterprise product profile to generate profile correlation data; perform edge server task assignment processing on profile correlation data to generate edge-level recommendation request data; Step S44: Perform time window adaptive rearrangement on the edge-level recommendation request data to generate dynamic priority ranking data; perform high correlation matching between the dynamic priority ranking data and product feature vector data to generate personalized user recommendation list data; Step S45: Perform client-side visualization processing on the personalized user recommendation list data to generate user recommendation interface view data.

7. The user recommendation method based on enterprise product profile analysis according to claim 6, characterized in that, Step S44 involves adaptively rearranging the edge-level recommendation request data using a time window to generate dynamic priority ranking data, including: Extract timestamps, request frequencies, user identifiers, and content types from edge-level recommendation request data to construct a set of request event sequences; based on the set of request event sequences, set initial sliding time window parameters, including time window size, sliding step size, and trigger threshold, to generate a basic time window structure; Analyze the changes in request density and the distribution of request intervals within a time window, and calculate the dynamic activity index for each time window segment; The time window size and sliding step size are adaptively adjusted based on the activity index to form an adaptive time window sequence with local optimal response performance. Within each adaptive time window, the importance weight of the request is calculated, where the weighting factors for the importance weight include: the level of the request source node, the popularity of the content, the request frequency, and the response latency requirement; Based on the importance weight of each request, the recommended requests within the time window are ranked locally by priority, and the cross-window is optimized by combining the position of the global sliding window to generate dynamic priority ranking data.

8. A user recommendation system based on enterprise product profile analysis, characterized in that, For performing the user recommendation method based on enterprise product profile analysis as described in claim 1, the user recommendation system based on enterprise product profile analysis includes: The enterprise screening module is used to obtain enterprise product information data; perform supply chain traceability based on enterprise product information data to generate enterprise product supply chain information data; and screen related enterprises for enterprise product supply chain information data to generate product-related enterprise information data. The product tagging module is used to analyze the production ratio of product components based on the information data of the enterprise's product supply chain information data associated with the product, and generate the production ratio value of product components; based on the production ratio value of product components, the module quantifies the product component input tags of the enterprise's product information data, and generates product tag data. The user interaction analysis module is used to construct enterprise product profiles based on product tag data; collect purchasing user information data using the constructed enterprise product profiles; analyze the user interaction behavior of purchasing user information data to generate purchasing user interaction behavior data, which includes time-series offset browsing scanning and collaborative path reverse behavior tracking. The personalized recommendation module is used to locate user scenarios based on low-power Bluetooth beacons by purchasing user interaction behavior data, and dynamically generate personalized user recommendation lists based on edge servers to perform user recommendation tasks based on enterprise product profile analysis.

Images