Supplier matching method and system based on multi-source heterogeneous data fusion

Abstract

This application discloses a supplier matching method and system based on multi-source heterogeneous data fusion, relating to the field of construction engineering procurement analysis. The method includes: performing feature parsing on the procurement demand text to generate a structured procurement demand vector; dynamically adjusting the weights of key indicators in the supplier profile based on multi-source data fusion and historical time series prediction; selecting feature mapping relationships based on project type to extract the Top-K profile indicators most relevant to the demand, and weighting and aggregating key profile indicators within the same supplier to generate its unique representation vector; and finally achieving high-precision candidate supplier screening by calculating the similarity between this vector and the procurement demand vector.

Description

Technical Field

[0001] This application relates to the field of procurement analysis technology in construction engineering, and in particular to a supplier matching method and system based on multi-source heterogeneous data fusion. Background Technology

[0002] In the construction engineering field, supplier selection and management span the entire process of project bidding, procurement, and contract performance, directly impacting project quality, schedule, and cost control. With the widespread application of digital and big data technologies, the industry has seen the emergence of diverse and heterogeneous data resources, including contract management systems, quality monitoring platforms, internal corporate financial data, and external data such as third-party credit ratings, legal proceedings, and public opinion monitoring. However, existing procurement systems lack the capacity to effectively integrate this data, hindering the full realization of the value of multi-source information in supplier management.

[0003] Currently, most systems still suffer from data silos and insufficient timeliness: on the one hand, there is a lack of effective data integration and sharing mechanisms between internal systems and external platforms; on the other hand, updates to external credit and risk information often lag significantly. When purchasing parties make decisions, the data they rely on is often a static snapshot of "yesterday" or "last month," which is insufficient to reflect the latest performance status and potential changes of suppliers.

[0004] Furthermore, supplier profile construction relies heavily on static information such as historical average fulfillment rates, qualification levels, and financial indicators, while key elements such as dynamic feedback and on-site evaluations are not updated in a timely manner. Correspondingly, procurement needs are diverse and time-sensitive—different types of projects have different focuses in terms of quality, delivery timeliness, and technical qualifications, but existing matching models often use fixed weight configurations, ignoring the actual matching preferences brought about by differences in needs, resulting in generalized and inaccurate recommendation results, requiring a large amount of manual secondary screening. Summary of the Invention

[0005] This application provides a supplier matching method, system, storage medium, computer program product, and electronic device based on multi-source heterogeneous data fusion, which aims to at least solve the problems of data silos, insufficient response time, and rigid matching modes in current related technologies.

[0006] Firstly, embodiments of this application provide a supplier matching method based on multi-source heterogeneous data fusion. The method includes: performing feature parsing on engineering procurement demand text according to a preset set of procurement feature dimensions to generate a corresponding procurement demand vector; the procurement feature dimensions include any one of the following: quality focus, performance timeliness, risk tolerance, and price sensitivity; determining the real-time monitoring values ​​of supplier profile indicators based on multiple data sources, and updating the supplier profile by adjusting the profile weights of the supplier profile indicators when the deviation between the real-time monitoring values ​​of the supplier profile indicators and the corresponding predicted values ​​exceeds a preset threshold; the predicted values ​​of the indicators are real-time predicted values ​​calculated based on historical indicator data of the supplier profile indicators, and the supplier profile indicators include any one of the following: performance delay rate, quality pass rate, financial health index, and judicial public opinion compliance; selecting feature mapping relationships according to the procurement project type, extracting the Top-K feature subset with the highest relevance to the procurement demand vector from the updated supplier profile, and generating a supplier-specific representation vector by weighting; calculating the similarity distance between the supplier-specific representation vector and the procurement demand vector to screen candidate suppliers.

[0007] Secondly, embodiments of this application provide a supplier matching system based on multi-source heterogeneous data fusion. The system includes: a procurement demand analysis unit, used to perform feature parsing on engineering procurement demand text according to a preset set of procurement feature dimensions to generate a corresponding procurement demand vector; the procurement feature dimensions include any one of the following: quality focus, performance timeliness, risk tolerance, and price sensitivity; and a supplier profile update unit, used to determine the real-time monitoring values ​​of supplier profile indicators based on multiple data sources, and when the deviation between the real-time monitoring values ​​of supplier profile indicators and the predicted values ​​of the corresponding indicators exceeds a preset threshold, adjust the supplier profile indicators... The weights are used to update the supplier profile; the predicted value of the indicator is a real-time predicted value calculated based on the historical indicator data of the supplier profile indicator, which includes any one of the following: performance delay rate, quality pass rate, financial health index, and judicial public opinion compliance; the Top-K vector generation unit is used to select feature mapping relationship according to the procurement project type, extract the Top-K feature subset with the highest relevance to the procurement demand vector from the updated supplier profile, and generate a weighted supplier-specific representation vector; the supplier screening unit is used to calculate the similarity distance between the supplier-specific representation vector and the procurement demand vector to screen candidate suppliers.

[0008] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the supplier matching method based on multi-source heterogeneous data fusion according to any embodiment of the present application.

[0009] Fourthly, embodiments of this application provide a storage medium storing a computer program thereon, characterized in that, when the program is executed by a processor, it implements the steps of the supplier matching method based on multi-source heterogeneous data fusion according to any embodiment of this application.

[0010] Fifthly, embodiments of this application provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the supplier matching method based on multi-source heterogeneous data fusion according to any embodiment of this application.

[0011] The supplier matching method and system based on multi-source heterogeneous data fusion provided in this application can produce at least the following technical effects:

[0012] (1) By setting multi-dimensional procurement feature dimensions (such as quality focus, performance timeliness, risk tolerance, and price sensitivity), the engineering procurement requirements expressed in natural language are structured and analyzed. A vectorized modeling mechanism with the semantic features of procurement requirements as the core is constructed, which transforms the expression of requirements from unstructured subjective descriptions into quantifiable, multi-dimensional semantic vectors, significantly enhancing the system's ability to identify the differences between different engineering types and procurement objectives.

[0013] (2) By collecting real-time indicator values ​​(such as performance delay rate, quality pass rate, financial health index, and judicial public opinion compliance) from multiple heterogeneous platforms, and comparing them with the predicted values, a real-time monitoring and deviation control model is introduced into the dynamic update mechanism of the supplier profile. This model can trigger a weight adjustment strategy when there is a significant deviation between the performance of the profile indicators and the predicted values ​​based on historical indicator data, thereby achieving adaptive updates of the supplier profile.

[0014] (3) By setting feature mapping relationships for different project types, the Top-K feature subset most relevant to the current needs is extracted from the updated supplier profiles, and a unique representation vector for each supplier is generated accordingly. This constructs a feature mapping and selection mechanism between the demand side and the supply side, effectively extracting the correlation features between profile dimensions and procurement objectives. By introducing a similarity distance algorithm, accurate supplier matching and ranking decisions based on vector calculation are achieved. This improves the focus of feature matching, ensures priority response to key capability dimensions during candidate screening, and avoids interference from weakly correlated or irrelevant features in the matching calculation.

[0015] This technical solution integrates dynamic analysis of procurement demand feature vectors with multi-source heterogeneous data, enabling real-time updates and adaptive weight adjustments to supplier profiles. This accurately reflects the latest status of suppliers across multiple dimensions, including performance timeliness, quality pass rate, financial health, and compliance risks. Furthermore, based on the matching of procurement project types and demand-side characteristics, it intelligently filters key supplier features highly relevant to actual needs, generating targeted supplier-specific representation vectors and accurately recommending candidate suppliers based on similarity distance. This significantly improves the fit and timeliness of supplier matching results, effectively reduces manual intervention, and enhances the intelligence of procurement decisions and risk management capabilities. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating an example of a supplier matching method based on multi-source heterogeneous data fusion according to an embodiment of this application is shown;

[0018] Figure 2 A flowchart illustrating an example of dynamic updating of a supplier profile according to an embodiment of this application is shown.

[0019] Figure 3 A flowchart illustrating an example of generating a supplier-specific representation vector based on an updated supplier profile, according to an embodiment of this application, is provided.

[0020] Figure 4 The simulation diagram shows a comparison of the Top-M hit rates of different methods;

[0021] Figure 5A structural block diagram of an example supplier matching system based on multi-source heterogeneous data fusion according to an embodiment of this application is shown. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] It should be noted that in recent years, with the continuous expansion of the scale of construction projects and the increasing complexity of procurement management, supplier matching has become a core part of the mathematical analysis of bidding. Traditional supplier matching methods for construction projects mainly rely on manual experience and static indicator scoring. The common practice is for procurement personnel to manually screen or score suppliers based on sub-indicators such as qualification certificates, historical cooperation records, quotations, and contract performance. Although this method has a certain degree of operability, it has obvious subjectivity and is difficult to fully reflect the supplier's true capabilities and risk status.

[0024] To improve matching efficiency and objectivity, some companies have introduced automated scoring systems based on databases and rule engines. These systems use preset rules to structure data on suppliers' company size, financial status, performance capabilities, price competitiveness, and other factors. Meanwhile, some research has proposed using multi-indicator decision-making methods such as AHP (Analytic Hierarchy Process) and TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) for supplier evaluation and selection. These systems can standardize multiple indicators and allocate weights, thus improving the scientific nature of the decision-making process.

[0025] Furthermore, with the popularization of information technology and big data technology, supplier matching is gradually incorporating machine learning and data mining methods. Related research attempts to build risk assessment and recommendation models based on historical procurement data, public opinion information, and third-party credit data. For example, methods such as decision trees, support vector machines, and cluster analysis are used to classify and predict suppliers, aiming to reduce labor costs and improve matching accuracy.

[0026] Despite continuous technological advancements, the following technical shortcomings persist in practical applications within the construction engineering field: First, significant data silos exist for supplier information, hindering effective integration of data across projects, departments, and platforms. This results in incomplete supplier profiles and an inability to dynamically reflect changes in supplier capabilities. Second, most existing models employ static scoring or single rules, failing to adapt to the dynamic matching needs across multiple dimensions and scenarios, including project type, project stage, and risk appetite. Third, current data-driven methods often overlook the complex relationship and contextual adaptability between procurement needs and supplier capabilities, leading to insufficient model generalization and interpretability, and a disconnect between matching results and actual business needs. Finally, some AI-based models suffer from weak cold-start and online self-learning capabilities, making it difficult to adapt promptly to new suppliers, new business models, or market changes.

[0027] Therefore, how to achieve the integration of multi-source heterogeneous data, dynamically optimize supplier profiles, and accurately adapt to procurement needs in various engineering scenarios remains a pressing challenge in current construction engineering supplier matching technology.

[0028] It should be understood that the above description of the relevant technologies is intended only to help the public better understand the inventive spirit and motivation of this application, and is not intended to limit this application. Furthermore, the technical solutions described in the above-mentioned relevant technologies are not prior art, and may also be undisclosed technical solutions, such as those under research or in the laboratory stage.

[0029] Figure 1 A flowchart illustrating an example of a supplier matching method based on multi-source heterogeneous data fusion according to an embodiment of this application is shown, which constructs a highly timely intelligent matching technology system for construction engineering suppliers.

[0030] Regarding the execution subject of the method in the embodiments of this application, it can be any controller or processor with computing or processing capabilities, which improves the intelligence and accuracy of supplier matching, realizes the dynamic response and differentiated adaptation of procurement decisions to diverse data, and promotes efficient collaboration and risk prevention and control in the construction engineering supply chain management.

[0031] In some examples, it can be integrated into an electronic device or terminal through software, hardware, or a combination of both, and the type of terminal or electronic device can be diverse, such as mobile phones, tablets, or desktop computers, etc.

[0032] like Figure 1 As shown, in step S110, the engineering procurement requirement text is parsed according to the preset procurement feature dimension set to generate the corresponding procurement requirement vector.

[0033] It should be noted that procurement requirements documents for engineering projects often come from a wide range of sources, including design briefs, tender documents, and user requirements specifications. They are expressed in various ways and have different focuses. Therefore, it is necessary to map the original requirements documents into a structured feature expression.

[0034] Here, the pre-defined set of procurement characteristic dimensions provides a standardized framework for requirements analysis, covering the core elements that procurement engineering focuses on. For example, procurement characteristic dimensions include any one of the following: quality focus, delivery timeliness, risk tolerance, and price sensitivity.

[0035] Quality focus reflects the procuring entity's emphasis on product or service quality, such as whether it stresses high-standard construction and quality control at key stages. Delivery timeliness measures the procuring entity's sensitivity to schedule or delivery deadlines, such as whether it requires shortened deadlines or phased delivery. Risk tolerance reflects the procuring entity's ability to accept potential risks related to supplier creditworthiness, compliance, and litigation history. Price sensitivity reflects the tightness of the project budget and the priority given to price.

[0036] In some implementations, deep semantic parsing techniques (such as the BERT language model) can be used to extract semantic elements from the text and map them to a predefined set of procurement feature dimensions, forming a multi-dimensional procurement demand vector. Each dimension value represents the weight of that factor in the procurement text. For example, if a project emphasizes quality and timeliness but is not price-sensitive, the values ​​of the "quality focus" and "performance timeliness" dimensions will be significantly higher than those of "price sensitivity." This achieves a standardized, structured, and quantitative expression of procurement needs, generating fine-grained and hierarchical feature expressions for procurement needs of different project types or with different priorities.

[0037] It should be understood that the above description of the procurement characteristics is for illustrative purposes only, so that the public can have a clearer understanding of the technical concept and motivation of this application, and should not be regarded as a limitation on the scope of implementation of this application.

[0038] Furthermore, the set of procurement characteristic dimensions can be customized for the corresponding project type; that is, the set of procurement characteristic dimensions can be different for different project types. For example, for rail transit projects, the set of procurement characteristic dimensions could be performance timeliness, compliance risk control, and safety standard compliance rate, while for residential building projects, it could be quality focus, price sensitivity, and after-sales maintenance service capabilities.

[0039] In step S120, the real-time monitoring value of the supplier profile indicator is determined based on multiple data sources. When the deviation between the real-time monitoring value of the supplier profile indicator and the predicted value of the corresponding indicator exceeds a preset amplitude threshold, the supplier profile is updated by adjusting the profile weight of the supplier profile indicator.

[0040] It should be noted that supplier profiling, as a core input for supply-demand matching, directly impacts the reliability of recommendation results due to its accuracy and timeliness. In engineering projects, a supplier's performance status can change significantly over time and with project circumstances. Traditional methods rely solely on historical contract performance, qualification levels, and partial financial statements, and these information is often outdated, failing to reflect the supplier's latest developments.

[0041] In this embodiment, through multi-source data fusion and prediction deviation mechanism, the supplier profile indicators are dynamically updated and the weights are adaptively adjusted, so that the profile can reflect the current and trend status of the supplier in a timely and accurate manner.

[0042] Here, the types of data sources can also be diverse, to obtain real-time profile indicators (real-time monitoring values) of suppliers from a variety of channels, such as the supplier's internal management system (such as enterprise contract performance system, quality supervision platform and project progress monitoring system, etc.), external data sources (such as data from third-party credit rating agencies, judicial litigation disclosure platforms, public opinion monitoring tools, etc.), and financial data sources (the supplier's financial statements, debt ratio, accounts receivable turnover cycle, etc.).

[0043] In addition, the supplier profile metrics include any one of the following: performance delay rate, quality pass rate, financial health index, and legal and public opinion compliance.

[0044] The performance delay rate reflects the proportion of tasks not completed on time within the stipulated delivery period. The quality pass rate is calculated by statistically analyzing the proportion of project nodes that have passed acceptance out of the total number of acceptances. The financial health index is constructed by combining liquidity indicators and debt levels. Judicial public opinion compliance can be comprehensively evaluated by combining administrative penalties, court judgments, and negative online public opinion.

[0045] It should be understood that the above description of supplier profile indicators is only used as an example to help the public better understand the technical concept and motivation of this application, and should not be regarded as a limitation on the scope of implementation of this application.

[0046] After obtaining the aforementioned real-time values, they are compared with the predicted values ​​of the indicators stored in the supplier profile. The predicted values ​​are real-time predictions calculated from historical indicator data of the supplier profile indicators, such as those predicted by a time series model based on historical indicator data. When the deviation between the real-time value and the predicted value exceeds a preset threshold (e.g., 15%), a weight adjustment mechanism is automatically triggered. For example, if a supplier's fulfillment delay rate suddenly increases significantly, the weight of that indicator in the overall profile will increase, the overall profile will lean towards a "high-risk" type, and additional profile indicator early warning functions can be implemented; conversely, the opposite is also true.

[0047] This enables a continuous monitoring and dynamic feedback mechanism for supplier behavior, allowing for real-time monitoring of changes in supplier performance capabilities. This significantly enhances the consistency between the profile and the actual situation, avoiding risk misjudgments caused by static and outdated profiles.

[0048] In step S130, a feature mapping relationship is selected according to the procurement project type, and the Top-K feature subset with the highest relevance to the procurement demand vector is extracted from the updated supplier profile, and a weighted representation vector for the supplier is generated.

[0049] It should be noted that while the indicators for supplier profiling are diverse, not every indicator is required for the procurement and bidding projects. Furthermore, different types of projects (such as municipal infrastructure, residential construction, and industrial plants) have different focuses in their requirements for supplier capabilities. Therefore, by introducing a procurement type-driven feature mapping mechanism, key profile features of the procurement project type can be selected from the supplier profile as needed for matching, thereby constructing a unique supplier representation vector.

[0050] For example, a predefined feature mapping template is invoked based on the procurement project type (which can come from structured project type tags or be derived from parsed procurement text). This template can be constructed based on historical project experience and expert knowledge to analyze the sensitivity and weight priority of different project types to various supplier profile features. After the feature mapping relationship is determined, a Top-K subset of features with the highest relevance (such as Pearson correlation coefficient or dot product similarity) to the current procurement demand vector is selected from the supplier profile, ensuring that the extracted features have significant numerical coupling with the procurement demand. Simultaneously, the weight of each feature in the supplier-specific representation vector can be adjusted according to the corresponding dimension in the procurement demand vector, achieving multi-dimensional alignment and semantic mapping.

[0051] In this embodiment, compared with the fixed full feature matching method, dynamic mapping with the semantic content of the procurement task can effectively eliminate the interference of irrelevant or redundant features, and significantly improve the adaptability and focus of the supplier capability expression in the procurement task, thus realizing customized supplier capability modeling.

[0052] In step S140, the similarity distance between the supplier-specific representation vector and the procurement demand vector is calculated to screen candidate suppliers.

[0053] Here, the similarity distance can be measured in various ways, such as cosine similarity, Euclidean distance, Mahalanobis distance, etc. By calculating the similarity score between the procurement demand vector and the unique representation vector of each supplier, personalized ranking of supplier matching results is achieved.

[0054] Through the embodiments of this application, from the analysis of customized procurement needs for engineering types to the construction of exclusive coupled features for dynamic supplier profiles, and the matching and ranking based on vector similarity calculation results, it is possible to more accurately identify the high degree of fit between suppliers and project needs, and also to keenly capture and respond to potential risks and changes in demand, significantly improving the scientific nature and security of decision-making in construction engineering supply chain management.

[0055] Regarding the implementation details of step S110, in some examples of embodiments of this application, the engineering procurement requirement text is extracted from multiple tender requirement documents issued by the procuring entity. Each tender requirement document indicates the same type of procurement project and has a unique corresponding tender release time. Specifically, for each procurement feature dimension in a preset set of procurement feature dimensions, the feature time series value corresponding to the procurement feature dimension is extracted from the engineering procurement requirement text based on consecutive tender release times. Furthermore, a procurement requirement vector is generated based on the feature time series values ​​corresponding to each procurement feature dimension.

[0056] Here, by performing structured semantic analysis on the tender documents of the same type of procurement project, the time-series feature values ​​corresponding to the preset feature dimensions are extracted, and then these time-series values ​​are combined to generate a more trend-oriented dynamic procurement demand vector, which can reflect the changes in the procurement preferences or priorities of the procurement party for the project.

[0057] It should be noted that construction projects are typically characterized by long cycles, multiple phases, and repetitive types. The same type of procurement project often issues tender documents multiple times at different time points. Although each document revolves around the same project objectives, it may present different expressions of requirements due to changes in the external environment, policy guidance, or project phase.

[0058] Therefore, by using a dynamic time-series generation method to generate procurement demand vectors and modeling based on the characteristic time-series values ​​extracted from continuous bidding documents, it is possible to more objectively and systematically reflect the long-term focus and emphasis of the procuring entity on this type of project, realize the modeling of the procuring entity's long-term behavioral preferences, and at the same time improve the model's responsiveness to the evolution of procurement intentions.

[0059] For example, taking the procurement project type of "urban rail transit tunnel construction" as an example, a certain unit issued 6 tender documents in the past three years, spanning from Q1 of 2021 to Q2 of 2024. After system processing, the following trends were obtained: the quality attention score remained above 0.85 for a long time, indicating that the procuring unit consistently adhered to high construction standards; the contract performance timeliness improved significantly in 2023 due to the overall compression of the subway line cycle; price sensitivity briefly increased in the early stage and then fell back; risk tolerance continued to decline, reflecting stricter supervision. Therefore, the dynamic changes in the weights of different procurement characteristic dimensions can be expressed, giving the procurement demand vector a richer and more precise semantic expression.

[0060] It should be noted that in actual bidding and procurement scenarios for construction projects, the level of detail in the bidding requirements text varies greatly. This may result in the inability to fully extract the information required for the preset set of procurement feature dimensions (such as quality focus, performance timeliness, risk tolerance, and price sensitivity) from certain texts during structured analysis.

[0061] Therefore, in some examples of embodiments of this application, it is detected whether the engineering procurement requirement text covers each procurement feature dimension in the procurement feature dimension set. If it covers, a procurement requirement vector can be directly generated; if it does not cover, it proves that there are missing dimensions. In this case, a historical engineering procurement data set matching the procurement engineering type can be obtained, the feature average value of the missing procurement feature dimension corresponding to the historical engineering procurement data set can be extracted, and the missing procurement feature dimension can be improved based on the feature average value. Then, a procurement requirement vector is generated based on the improved procurement feature values.

[0062] In this embodiment, dimensional coverage detection is performed on the current procurement requirement text. If missing dimension items exist, a complete procurement requirement vector cannot be directly constructed. To address this, historical project procurement data sets matching the current project type can be accessed, and the average feature values ​​of these historical data on the missing dimensions can be extracted as an approximate expression for the corresponding dimension in the current requirement vector. The detected dimension values ​​are then fused with the completed missing dimension values ​​to construct a structurally complete procurement requirement vector.

[0063] For example, if "risk tolerance" is missing in the current text, the system will extract feature values ​​for that dimension from historical similar projects, such as [0.42, 0.48, 0.44, 0.46], and take the average of 0.45 as the complete estimate for the missing item "risk tolerance". This ensures that even when the information in the engineering procurement requirements text is incomplete, it can still maintain stable vector representation capabilities, avoiding the disruption of the entire supplier matching process due to missing dimension values.

[0064] Regarding the details of filtering candidate suppliers by calculating similarity distance, in some examples of embodiments of this application, it may involve mapping the supplier-specific representation vector to the procurement demand vector in a dimensionality subset space. Here, the procurement demand vector... Supplier-specific representation vectors Both are based on differential feature subsets Constructed within the same dimensional space. Differential feature subsets ( The vectors are selected based on the feature relevance score of the procurement needs and correspond to the profile dimensions of the suppliers. This ensures that when calculating similarity, the two vectors are aligned in the same feature dimension space, without the risk of dimension misalignment.

[0065] Based on the normalized weights of each feature within the subset of differentiated features, the weighted Euclidean distance and weighted cosine similarity between the supplier-specific representation vector and the procurement demand vector are calculated respectively.

[0066] Here, normalized weights are used to weight each feature, calculating the weighted Euclidean distance and weighted cosine similarity. The weighted Euclidean distance measures the degree of fit between supplier capabilities and procurement needs along the same feature dimension by measuring the absolute deviation between them. Furthermore, the weighted cosine similarity measures the "directional" similarity between supplier capabilities and procurement needs in the feature space, making it particularly suitable for structural matching between supplier capabilities and needs.

[0067] In the calculation of weighted Euclidean distance, a profile vector for each supplier is generated. and procurement demand vector According to normalized weights For each feature dimension Calculate the squares of the weighted differences and sum them:

[0068] Equation (1)

[0069] In the formula, Features Normalized weights, For suppliers In features Standardized values ​​on For the procurement demand vector in features The demand value.

[0070] In the calculation of weighted cosine similarity, the supplier profile vector is... and procurement demand vector Weighted cosine similarity is calculated based on feature-normalized weights:

[0071] Equation (2)

[0072] Therefore, weighted Euclidean distance measures the absolute fit between suppliers and demand, especially when strict control is required over certain features (such as cost and delivery time). Weighted cosine similarity focuses more on the matching of feature structures and can identify the degree of alignment between suppliers and demand in terms of direction. It is particularly suitable for scenarios with diverse demands and a focus on long-term cooperation.

[0073] The weighted Euclidean distance of all suppliers is normalized to obtain a distance matching score, which is then weighted and fused with the weighted cosine similarity according to a preset fusion weight to generate a comprehensive supplier similarity score.

[0074] To ensure a fair comparison between different measurement methods, the weighted Euclidean distance and weighted cosine similarity need to be normalized to ensure consistent value ranges and facilitate fusion calculation. The normalized weighted Euclidean distance reflects the "absolute difference" between the supplier and demand; the normalized weighted cosine similarity reflects the "directional alignment" between the supplier and demand. Then, the two are weighted and fused according to a preset fusion weight (e.g., 0.5, or dynamically adjusted based on demand) to obtain a comprehensive similarity score. By normalizing and weighted fusion, the dimensional differences between different similarity measurement methods were eliminated.

[0075] Finally, based on the supplier's overall similarity score and preset screening rules, all suppliers are ranked to select candidate suppliers.

[0076] Here, all suppliers are ranked based on their overall similarity score. A higher score indicates a stronger fit for the current procurement needs. Therefore, a minimum matching score threshold or a Top-N filtering rule can be set to ensure that recommended suppliers meet the buyer's core requirements.

[0077] It should be noted that matching suppliers with procurement needs relies not only on a single feature fit, but also on a comprehensive evaluation across multiple dimensions. For example, the differences between supplier capabilities and procurement needs (such as delivery timeliness and quality control) are crucial; therefore, weighted Euclidean distance can quantify these absolute numerical differences. Simultaneously, procurement needs often emphasize structural fit, such as whether supplier capabilities match demand features in the same direction, which requires weighted cosine similarity for measurement. Therefore, by fusing weighted Euclidean distance and weighted cosine similarity, both absolute fit and directional matching can be comprehensively considered, ensuring that the recommendation system can more comprehensively evaluate the fit between suppliers and procurement needs.

[0078] It should be noted that in the analysis of bidding and procurement for construction projects, the actual performance of suppliers is highly dynamic and time-sensitive. Key profile indicators such as performance quality, delivery capability, and compliance level may fluctuate over time due to factors such as environmental changes, resource pressure, and policy supervision. Static profile models often cannot reflect these changes in real time, which may lead to matching bias and risk exposure.

[0079] In view of this, Figure 2 The flowchart illustrates an example of dynamic updating of supplier profiles according to an embodiment of this application. It proposes a supplier profile updating method based on historical prediction-real-time observation comparison and weight adaptive adjustment mechanism to improve the dynamic perception capability and expression reliability of the profile.

[0080] like Figure 2 As shown, in step S210, for each supplier profile indicator, the historical time series values ​​of the supplier profile indicators are processed based on the exponential weighted moving average method to obtain the corresponding indicator prediction values.

[0081] Here, supplier profile metrics, as important quantitative data for measuring their performance capabilities (such as performance delay rate, quality pass rate, financial health index, and legal compliance), typically exhibit certain time-series characteristics. To determine whether current supplier behavior is "deviation from normal," a reasonable historical trend baseline must first be established. Specifically, the historical time-series values ​​of each profile metric are processed using an EWMA (Exponentially Weighted Moving Average) model to obtain its predicted value for the current real-time moment, which is then used as a reference baseline for the "normal state."

[0082] Specifically, for each supplier profile indicator, the system maintains a time series consisting of observation data at continuous time points, denoted as... , This represents the historical value of the indicator at time point t.

[0083] Predicted values ​​are generated using the EWMA algorithm:

[0084] Equation (3)

[0085] In the formula, This represents the predicted value of the indicator at time point t. This represents the actual historical observation value at time point t-1; This is a smoothing factor; a larger value indicates greater sensitivity to recent data. The predicted value is for time point t-2. The initial value can be the historical average or the first observation.

[0086] To accommodate the dynamic nature of different indicators, the system allows for different settings to be applied to different indicators. Value. For example, for an indicator like "delay rate," which is highly volatile and fluctuates significantly in the short term, a larger value can be selected. (e.g., 0.6~0.8); for indicators like the "Financial Health Index," which change slowly, a smaller value can be selected. (e.g., 0.2~0.4). As a result, the generated predicted values ​​have the characteristics of both trend smoothness and short-term sensitivity, effectively avoiding misjudgments caused by short-term fluctuations and enhancing the system's ability to grasp the evolution trend of indicators.

[0087] In step S220, the raw observation values ​​of the supplier profile indicators are obtained from multiple data sources, and the real-time monitoring values ​​of the supplier profile indicators are obtained through weighted fusion.

[0088] To obtain accurate performance data for current supplier profile metrics, raw observations need to be collected from multiple business platforms and data sources. These sources include internal management systems (such as project management platforms and quality inspection systems) and external platforms (such as third-party credit rating agencies, public opinion monitoring systems, and court announcement platforms). Due to the heterogeneity and varying reliability of data sources, a weighted fusion mechanism must be introduced to integrate the observations and form more stable and reliable real-time monitoring values.

[0089] For example, suppose a certain portrait indicator starts from... The observations collected from different data sources are Weights are assigned to each observation. It satisfies the normalization condition: Then the real-time monitoring value of this portrait indicator The calculation formula is:

[0090] Equation (4)

[0091] In the formula, Indicates the first Observations from one data source Indicates the first The fusion weight of each data source can be set based on factors such as data credibility and update frequency.

[0092] Therefore, by integrating multi-source observation data, the accuracy and real-time performance of profile indicator monitoring can be significantly improved, avoiding delays or false alarms that may result from relying on a single source, and enhancing the system's ability to perceive the actual status of suppliers.

[0093] In step S230, when the deviation between the real-time monitoring value and the predicted value of the supplier profile indicator exceeds a preset threshold, the weight of the corresponding supplier profile indicator is adjusted according to the deviation.

[0094] On the other hand, when the preset threshold is not exceeded, there is no need to adjust the weight of the supplier profile indicator.

[0095] Here, if there is a significant deviation between the real-time monitoring value and the historical predicted value, it indicates that the performance of the profile indicator is abnormal in the current period. Its weight should be adjusted to strengthen its influence on the overall profile, thereby improving the profile's sensitivity to actual risks. Conversely, if the deviation is acceptable, no adjustment is needed.

[0096] Specifically, let the predicted value of a certain indicator be... The real-time monitoring value is Calculate its relative deviation magnitude :

[0097] Equation (5)

[0098] The system sets a preset deviation threshold. The judgment condition is:

[0099] like If the indicator is significantly abnormal, its weight should be adjusted; if If the fluctuation is normal, no adjustment will be made.

[0100] For indicators that require weight adjustments, their new weights Calculate as follows:

[0101] Equation (6)

[0102] In the formula, This is the old weight of the indicator; The adjustment coefficient controls the sensitivity to deviation amplification (setting it to 1 indicates a linear response).

[0103] In addition, to prevent the weights from expanding too rapidly, the system can set a maximum weight limit. Cutting:

[0104] Equation (7)

[0105] Unlike traditional fixed-weight profiling models, the embodiments of this application can dynamically respond to abnormal fluctuations in supplier profiling indicators and enhance their weight in the profiling, so that the profiling results can quickly reflect current risk points or potential performance issues, thereby enhancing the adaptability and risk warning capabilities of the profiling.

[0106] In step S240, the profile weights of all supplier profile metrics are normalized to update the supplier profiles.

[0107] Here, after the profile weight adjustment as described in step S230, the sum of the weights of each profile indicator may change, no longer meeting the weighted consistency requirement. To maintain profile comparability and structural integrity, all indicator weights need to be normalized, and the supplier profile updated accordingly. This ensures that the sum of the dimensional weights of the supplier profile is 1, maintaining structural consistency and avoiding imbalances caused by local weight adjustments.

[0108] Figure 3 A flowchart illustrating an example of generating a supplier-specific representation vector based on an updated supplier profile, according to an embodiment of this application, is shown.

[0109] like Figure 3 As shown, in step S310, based on the procurement project type identifier... Load the pre-trained engineering type-feature mapping matrix ,in The number of associated procurement feature dimensions is equal to the total number of dimensions in the procurement feature dimension set. The total number of pre-trained and calibrated related supplier profile metrics.

[0110] It should be noted that the rows of the feature mapping matrix (i.e. The feature set (dimensions) is consistent with the procurement feature dimension set, which means that there will be no dimension misalignment or information loss during feature extraction and feature mapping, which helps to make procurement requirements "traceable" and "explainable". In addition, improving feature relevance and business matching ensures a tight coupling between the feature mapping matrix and procurement requirements. The selected differentiated feature subset (i.e., Top-K features) must be what procurement actually focuses on, thus improving the accuracy of the results.

[0111] In some implementations, the system stores a set of pre-trained project type-feature mapping matrices, which record the feature mapping matrices corresponding to different project types. The system loads a matching pre-trained model based on the type of project procurement needs (e.g., construction, equipment procurement, road construction, etc.). Each pre-trained model contains a feature mapping matrix related to that project type, used to map various features of the supplier profile into a high-dimensional space. This ensures that the model training and feature mapping match the needs of a specific project type, avoiding mismatches caused by generic feature mapping.

[0112] For example, suppose there is Various characteristic dimensions related to procurement needs (such as quality, delivery time, risk, etc.), and The system will utilize project type as a supplier characteristic dimension. The corresponding feature mapping matrix Feature mapping is performed to map the features of each supplier to a feature space related to the project type, ensuring maximum relevance to procurement requirements.

[0113] In step S320, based on the procurement demand vector With project type - feature mapping matrix Calculate the feature relevance score vector.

[0114] Equation (8)

[0115] In the formula, express The transpose of , This represents the feature relevance score vector.

[0116] Specifically, equation (8) reflects a linear feature-weighted projection. Procurement demand vector This represents the level of attention and quantitative requirements for different characteristics (such as price, delivery time, quality, etc.) in this procurement, while the project type-feature mapping matrix... This depicts the structured mapping relationship between each procurement feature and all supplier profile indicators. Multiplying the two is equivalent to using the importance of each feature of the procurement requirement to correspondingly weight the correlation strength between each feature under the project type and all supplier profile features, thus realizing a "weighted correlation assessment" of each supplier profile indicator.

[0117] In other words, the result of matrix multiplication Each component in the model represents "the contribution or relevance of each profile feature to meeting the requirements under the current project type, given the weight of the procurement needs." This allows the system to automatically quantify which supplier capabilities are most critical in the current context, enabling targeted selection and recommendation of supplier indicators. The linear mapping method quantifies the correlation between current project requirements and supplier profile features while preserving business interpretability.

[0118] In step S330, select The largest value The index of each supplier profile metric constitutes a subset of differentiated features. .

[0119] Here, after obtaining the matching scores between suppliers and procurement needs, the system will sort these scores and select the one with the highest score. We use a supplier profile metrics to highlight the most relevant metrics while avoiding interference from low-relevance or noisy dimensions.

[0120] Specifically, the system scores based on feature relevance vectors. Select the first from among them A collection of indexes for supplier profile metrics This indicates the highest degree of matching with the current needs. Each supplier profile indicator. Furthermore, adjustments can be made... The number of candidate supplier profile indicators can be flexibly controlled to meet the actual needs of different business scenarios or system resources.

[0121] In step S340, supplier-specific representation vectors are extracted from the updated supplier profile based on a subset of differentiated features.

[0122] In some implementations, after selecting the optimal After identifying the supplier profile metrics, the system further extracts feature vectors from the values ​​under these metrics to reflect the supplier's adaptability to current procurement needs.

[0123] The system updates the supplier-specific representation vector using the following formula:

[0124] Equation (9)

[0125] Equation (10)

[0126] In the formula, Indicates supplier The supplier-specific representation vector, For suppliers Supplier profile metrics Standardized index values, Supplier profiling metrics The normalized weights represent the first... The influence of each feature within a subset of differentiated features; This is an adjustment coefficient used to adjust the degree of concentration or dispersion of the weight distribution; For normalized traversal variables, it represents the index of any feature within the subset of differentiated features; Represents the first in the subset of differentiated features The relevance score of each feature.

[0127] This application's embodiments introduce a feature mapping matrix based on procurement project type. Through operations with the procurement demand vector, it identifies the most relevant subset of profile indicators for the current task and performs weighted aggregation on these indicators within each supplier. Furthermore, the weighting is adaptively adjusted based on the relevance score to the actual procurement needs, allowing the profile-specific vector to focus more on the capability dimensions relevant to this task, thus improving the accuracy of the matching and ranking results.

[0128] Regarding the training of the set of project type-feature mapping matrices, in some examples of embodiments of this application, it can be achieved through the following operations.

[0129] First, from project data of multiple historical project types, we collect the procurement demand vector set, supplier profile indicator set, and supplier matching results of historical projects to initialize the mapping matrix meta-parameters shared by all project types. .

[0130] It should be understood that the effectiveness of the project type-feature mapping matrix highly depends on the comprehensiveness and accuracy of three types of historical data: procurement needs, supplier profiles, and actual matching results. For example, from project data of multiple historical project types (such as roads, bridges, and residential buildings), the procurement needs vector set (such as required materials and technical parameters), the supplier profile indicator set (such as performance capability, cost, and reputation), and the actual matching results of the project with different suppliers (such as winning bids, scoring, and contract performance) are collected and organized in batches for each project.

[0131] By structuring this data, procurement needs and supplier capabilities can be uniformly mapped to a standardized feature space, providing data support for training multiple types of matrices. Based on this, a global meta-parameter is initialized using the historical sample averaging method. This serves as the initial parameter for the mapping matrix of all project types. Therefore, by uniformly initializing the meta-parameters, the efficiency of knowledge transfer and parameter sharing among different project types can be effectively improved.

[0132] Then, for each project type, the meta-parameters are adjusted using the procurement demand vector, supplier profile indicators, and historical project matching results for that project type. Perform gradient updates a predetermined number of times to generate the initial project type-feature mapping matrix. :

[0133] Equation (11)

[0134] In the formula, For learning rate, For different types of projects The loss function is defined by combining procurement needs, supplier profiles, and historical matching results.

[0135] It should be noted that the requirements and supplier capabilities vary for each project type, necessitating an analysis of global meta-parameters. Only by making fine-tuning adjustments to the type can we obtain a unique mapping matrix that reflects the characteristics of this project type.

[0136] Specifically, for each specific project type We selected procurement demand vectors, supplier profile indicators, and historical matching result data related to this type as the training set.

[0137] Loss functions for constructing feature mapping matrix models for various project types The project type-feature mapping matrix model can employ linear mapping models or neural networks to optimize the feature mapping matrix for the corresponding project type. This ensures that the matching effect between procurement needs and supplier profiles, after matrix mapping, can predict the actual matching effect as accurately as possible. Therefore, this loss function comprehensively reflects the deviation between the matching effect of procurement needs and supplier profiles and historical observations. It can employ mean squared error, cross-entropy, or weighted loss functions tailored to various project types. For example, the mean squared error loss function can be used to measure the mean squared error between the model-predicted supplier matching score and the actual matching result for all historical samples.

[0138] by Based on this, the gradient descent method is used to iterate the loss function one or more times, updating the parameters to minimize the matching bias, thus obtaining the feature mapping matrix specific to this project type. .

[0139] This enables feature mapping matrices for different project types to accurately express the correlation between procurement characteristics and supplier profiles for that type, improving the accuracy and targeting of intelligent supplier matching. Simultaneously, it supports parameter sharing and knowledge transfer between types, enhancing the robustness and generalization ability of training new type matrices in small-sample scenarios.

[0140] Furthermore, by jointly optimizing the finely tuned sum of losses across all project types, the following objective function is minimized to obtain the meta-parameters with optimal generalization ability. :

[0141] Equation (12)

[0142] In the formula, This indicates the total number of historical project types. Indicates the meta-parameter In terms of project type Lower loss function The gradient is used to guide the... Adjustments to reduce project types The matching loss.

[0143] Here, to further improve the model's generalization ability across different project types, a multi-task learning approach is adopted. The sum of losses after fine-tuning for all project types is jointly minimized to obtain the globally optimal meta-parameters. Therefore, meta-learning and automatic differentiation are employed. Multi-task training is conducted to make it the optimal starting point for fine-tuning the matrix specific to each engineering type.

[0144] If updated project data is detected, the updated meta-parameters are fine-tuned using the updated project data. And update the project type-feature mapping matrix set.

[0145] Here, an online adaptive update mechanism is adopted. When new project data (such as new procurement requirements, supplier profiles, and the latest matching / performance performance) arrives in real time, these updated project data are used as increments, and the same gradient update or fine-tuning strategy as historical data is applied to the global meta-parameters. Or a dedicated matrix for the corresponding project type Local optimization is performed. As a result, the mapping matrix can continuously track and reflect dynamic changes in the market, reducing the risk of decreased adaptability or failure caused by long-term reliance on static models.

[0146] Through the embodiments of this application, based on historical procurement needs, supplier profiles, and actual matching results, a unique feature mapping matrix closely corresponding to historical performance is trained for each project type. Furthermore, efficient knowledge transfer between different model types is achieved through multi-task optimization of global meta-parameters, reducing the cold-start difficulty for new or data-scarce types. In addition, through multi-task joint optimization and continuous online updates, the system can learn and adaptively reflect the actual business patterns in various project scenarios.

[0147] To further verify the superiority of the method in the embodiments of this application, the details of the experimental part of the method based on the project type-feature mapping matrix model in the embodiments of this application will be elaborated below.

[0148] Specifically, to verify the effectiveness and superiority of the proposed "supplier matching method based on project type-feature mapping matrix" in real-world engineering procurement scenarios, this paper conducts experiments based on procurement datasets from multiple real-world engineering projects in the construction industry. The experimental data covers procurement demand characteristics, supplier profile indicators, and historical matching scores, including multiple project types such as bridges, residential buildings, and roads, and covers different project scales, complexities, and supplier types.

[0149] In terms of data preprocessing, all features were standardized, and the labels were uniformly set as historical matching scores (or performance scores) between procurement needs and supplier profiles. The dataset was randomly divided into training and testing sets to ensure the objectivity of the experimental results.

[0150] Regarding evaluation metrics, the following mainstream evaluation metrics were used to assess the matching effectiveness of each method:

[0151] 1) Mean Squared Error (MSE): Reflects the error between the model's predicted score and the actual score.

[0152] 2) Mean Absolute Error (MAE): Measures the average deviation between the predicted value and the actual value.

[0153] 3) Top-M Hit Rate: In the procurement requirements of the test set, whether the actual preferred supplier is listed in the model's Top-M recommendation results (measures ranking and recommendation accuracy).

[0154] 3) AUC: Reflects the consistency and discriminative power of model ranking.

[0155] To comprehensively compare the performance of the proposed method, two existing models were selected for comparative experiments. The first was a traditional static weighted model, which sets weights solely based on human experience or fixed rules, and performs linear weighted matching according to each feature without distinguishing between project types. The second was a standard multi-index linear regression model, which directly concatenates procurement requirements and supplier profile features as input and outputs a matching score without using a mapping matrix structure.

[0156] The two comparative models mentioned above are compared with the matching method based on the project type-feature mapping matrix model proposed in this paper. Specifically, the project type-feature mapping matrix model introduces project type differentiation and uses the feature mapping matrix training and fine-tuning mechanism proposed in this paper to perform correlation projection and intelligent matching between procurement needs and supplier profiles.

[0157] Table 1 shows the main evaluation metrics for different methods on the test set.

[0158]

[0159] As shown in Table 1, the experimental results demonstrate that the proposed method outperforms traditional methods and ordinary linear regression models across all evaluation metrics, particularly in key metrics such as Top-5 hit rate and AUC. Therefore, by introducing an engineering type-feature mapping matrix mechanism, the complex correspondence between procurement needs and supplier profiles under different engineering types can be effectively modeled, achieving personalized, intelligent, and accurate supplier matching.

[0160] Figure 4 A simulation diagram comparing the Top-M hit rates of different methods is shown. Specifically, Figure 4 The Top-M hit rates of the project type-feature mapping matrix (the method presented in this paper), the multi-index linear regression model, and the traditional static weighted model on the same test set are compared through simulation. The horizontal axis represents the length M of the recommendation list (i.e., Top-M), ranging from 1 to 10; the vertical axis represents the corresponding Top-M hit rate (%), that is, the proportion of historical preferred suppliers that appear in the top M positions of the model's recommendations.

[0161] like Figure 4 The results shown are for the static weighted model (dotted line): each feature is linearly weighted using fixed empirical weights, and the hit rate increases slowly as M increases;

[0162] Multi-indicator linear regression (block line): directly splices procurement demand with supplier profile features, and uses linear regression to predict matching scores. Its Top-M hit rate is higher than that of static methods.

[0163] The proposed method (triangle line) is an intelligent matching scheme based on the project type-feature mapping matrix. It significantly outperforms the comparison methods under all M values, especially when M=5, the hit rate reaches over 82%, which verifies the efficient modeling ability of this method to model the complex correspondence between procurement needs and supplier profiles under different project types.

[0164] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of combined actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application. In the above embodiments, the descriptions of each embodiment have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0165] Figure 5 A structural block diagram of an example supplier matching system based on multi-source heterogeneous data fusion according to an embodiment of this application is shown.

[0166] like Figure 5 As shown, the supplier matching system 500 based on multi-source heterogeneous data fusion includes a procurement demand analysis unit 510, a supplier profile update unit 520, a Top-K vector generation unit 530, and a supplier screening unit 540.

[0167] The procurement demand analysis unit 510 is used to perform feature parsing on the engineering procurement demand text according to the preset procurement feature dimension set to generate the corresponding procurement demand vector; the procurement feature dimensions include any one of the following: quality focus, performance timeliness, risk tolerance and price sensitivity.

[0168] The supplier profile update unit 520 is used to determine the real-time monitoring values ​​of supplier profile indicators based on multiple data sources, and when the deviation between the real-time monitoring value of the supplier profile indicator and the corresponding predicted value exceeds a preset threshold, the supplier profile is updated by adjusting the profile weight of the supplier profile indicator; the predicted value of the indicator is a real-time predicted value calculated based on the historical indicator data of the supplier profile indicator, and the supplier profile indicator includes any one of the following: performance delay rate, quality pass rate, financial health index and judicial public opinion compliance.

[0169] The Top-K vector generation unit 530 is used to select feature mapping relationships according to the procurement project type, extract the Top-K feature subset with the highest relevance to the procurement demand vector from the updated supplier profile, and generate a supplier-specific representation vector by weighting.

[0170] The supplier screening unit 540 is used to calculate the similarity distance between the supplier-specific representation vector and the procurement demand vector in order to screen candidate suppliers.

[0171] In some embodiments, this application provides a non-volatile computer-readable storage medium storing one or more programs including execution instructions. These execution instructions can be read and executed by electronic devices (including but not limited to computers, servers, or network devices) to perform the steps of any of the supplier matching methods based on multi-source heterogeneous data fusion described above.

[0172] In some embodiments, this application also provides a computer program product, the computer program product including a computer program stored on a non-volatile computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the steps of any of the above-described supplier matching methods based on multi-source heterogeneous data fusion.

[0173] In some embodiments, this application also provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform steps of a supplier matching method based on multi-source heterogeneous data fusion.

[0174] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0175] The electronic devices in this application embodiments exist in various forms, including but not limited to:

[0176] (1) Mobile communication devices: These devices are characterized by their mobile communication capabilities and primarily aim to provide voice and data communication. These terminals include smartphones, multimedia phones, feature phones, and low-end phones.

[0177] (2) Ultra-mobile personal computer devices: These devices fall under the category of personal computers, possessing computing and processing capabilities, and generally also have mobile internet access features. These terminals include: PDAs, MIDs, and UMPCs, etc.

[0178] (3) Portable entertainment devices: These devices can display and play multimedia content. This category includes audio and video players, handheld game consoles, e-book readers, as well as smart toys and portable car navigation devices.

[0179] (4) Other airborne electronic devices with data interaction capabilities, such as vehicle-mounted systems installed on vehicles.

[0180] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0181] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0182] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A supplier matching method based on multi-source heterogeneous data fusion, characterized in that, The method includes: The engineering procurement requirement text is analyzed based on a preset set of procurement feature dimensions to generate a corresponding procurement requirement vector. The procurement feature dimensions include any one of the following: quality focus, performance timeliness, risk tolerance, and price sensitivity. The engineering procurement requirement text is extracted from multiple tender requirement documents issued by the procuring party. Each tender requirement document indicates the same type of procurement project and has a unique corresponding tender release time. The real-time monitoring values ​​of supplier profile indicators are determined based on multiple data sources. When the deviation between the real-time monitoring value of the supplier profile indicator and the predicted value of the corresponding indicator exceeds a preset threshold, the supplier profile is updated by adjusting the profile weight of the supplier profile indicator. The predicted value of the indicator is a real-time predicted value calculated based on the historical indicator data of the supplier profile indicator. The supplier profile indicator includes any one of the following: performance delay rate, quality pass rate, financial health index, and judicial public opinion compliance. Select feature mapping relationships based on the type of procurement project, extract the Top-K feature subsets with the highest relevance to the procurement demand vector from the updated supplier profile, and generate a weighted supplier-specific representation vector; Calculate the similarity distance between the supplier-specific representation vector and the procurement demand vector to filter candidate suppliers; The step of parsing the engineering procurement requirement text based on a preset set of procurement feature dimensions to generate a corresponding procurement requirement vector includes: For each procurement feature dimension in the preset procurement feature dimension set, the feature time sequence value corresponding to the procurement feature dimension is extracted from the engineering procurement requirement text according to the continuous bidding release time. A procurement demand vector is generated based on the time series values ​​of the features corresponding to each of the procurement feature dimensions.

2. The method according to claim 1, characterized in that, The step of parsing the engineering procurement requirement text based on a preset set of procurement feature dimensions to generate a corresponding procurement requirement vector includes: Detect whether the engineering procurement requirement text covers each procurement feature dimension in the procurement feature dimension set; When not covered, obtain the historical project procurement data group that matches the procurement project type, extract the feature average value of the missing procurement feature dimension corresponding to the historical project procurement data group, and improve the missing procurement feature dimension based on the feature average value. Based on the refined procurement feature values, a procurement demand vector is generated.

3. The method according to claim 1, characterized in that, The process involves determining real-time monitoring values ​​of supplier profile metrics based on multiple data sources, and updating the supplier profile by adjusting the profile weights of the supplier profile metrics when the deviation between the real-time monitoring values ​​of the supplier profile metrics and the predicted values ​​of the corresponding metrics exceeds a preset threshold. This includes: For each supplier profile indicator, the historical time series values ​​of the supplier profile indicators are processed based on the exponential weighted moving average method to obtain the corresponding indicator prediction values; The raw observations of supplier profile metrics are obtained from multiple data sources, and the real-time monitoring values ​​of supplier profile metrics are obtained through weighted fusion. When the deviation between the real-time monitoring value and the predicted value of a supplier profile indicator exceeds a preset threshold, the weight of the corresponding supplier profile indicator is adjusted according to the deviation. The profile weights of all supplier profile metrics are normalized to update the supplier profiles.

4. The method according to claim 3, characterized in that, The step of selecting feature mapping relationships based on the procurement project type, extracting the Top-K feature subsets with the highest relevance to the procurement demand vector from the updated supplier profile, and generating a weighted supplier-specific representation vector includes: According to the type of procurement project Load the pre-trained engineering type-feature mapping matrix ,in The number of associated procurement feature dimensions is equal to the total number of dimensions in the procurement feature dimension set. The total number of pre-trained and calibrated related supplier profile metrics; Based on procurement demand vector With project type-feature mapping matrix Calculate the feature relevance score vector: ， In the formula, express The transpose of , Represents the feature relevance score vector; Select The largest value The index of each supplier profile metric constitutes a subset of differentiated features. ; Extract supplier-specific representation vectors from the updated supplier profile based on a subset of differentiated features: ， ， In the formula, Indicates supplier The supplier-specific representation vector, For suppliers Supplier profile metrics Standardized index values, Supplier profiling metrics The normalized weights represent the first... The influence of each feature within a subset of differentiated features; This is an adjustment coefficient used to adjust the degree of concentration or dispersion of the weight distribution; For normalized traversal variables, it represents the index of any feature within the subset of differentiated features; Represents the first in the subset of differentiated features The relevance score of each feature.

5. The method according to claim 4, characterized in that, Training for the set of project type-feature mapping matrices includes: From project data of multiple historical engineering types, we collect sets of procurement demand vectors, sets of supplier profile indicators, and supplier matching results from historical projects to initialize the common mapping matrix parameters for all engineering types. ; For each project type, the meta-parameters are adjusted using the procurement demand vector, supplier profile indicators, and historical project matching results for that project type. Perform gradient updates a predetermined number of times to generate the initial project type-feature mapping matrix. : ， In the formula, For learning rate, For different types of projects The loss function is defined by combining procurement needs, supplier profiles, and historical matching results. Indicates the meta-parameter In terms of project type Lower loss function The gradient is used to guide the... Adjustments to reduce project types Matching loss; By jointly optimizing the finely tuned sum of losses across all project types, the following objective function is minimized to obtain the meta-parameters with optimal generalization ability. : ， In the formula, This indicates the total number of historical project types; If updated project data is detected, the updated meta-parameters are fine-tuned using the updated project data. And update the project type-feature mapping matrix set.

6. The method according to claim 4, characterized in that, The step of calculating the similarity distance between the supplier-specific representation vector and the procurement demand vector to filter candidate suppliers includes: The supplier-specific representation vector and the procurement demand vector are mapped dimensionally within the differential feature subset space; Based on the normalized weights of each feature within the subset of differentiated features, the weighted Euclidean distance and weighted cosine similarity between the supplier-specific representation vector and the procurement demand vector are calculated respectively. The weighted Euclidean distance of all suppliers is normalized to obtain a distance matching score, which is then weighted and fused with the weighted cosine similarity according to a preset fusion weight to generate a comprehensive supplier similarity score. Based on the supplier's comprehensive similarity score and preset screening rules, all suppliers are sorted to select candidate suppliers.

7. A supplier matching system based on multi-source heterogeneous data fusion, characterized in that, The system includes: The procurement demand analysis unit is used to perform feature parsing on the engineering procurement demand text according to a preset set of procurement feature dimensions to generate a corresponding procurement demand vector. The procurement feature dimensions include any one of the following: quality focus, performance timeliness, risk tolerance, and price sensitivity. The engineering procurement demand text is extracted from multiple tender demand documents issued by the procuring party. Each tender demand document indicates the same type of procurement project and has a unique corresponding tender release time. The supplier profile update unit is used to determine the real-time monitoring values ​​of supplier profile indicators based on multiple data sources, and when the deviation between the real-time monitoring value of the supplier profile indicator and the corresponding predicted value exceeds a preset threshold, the supplier profile is updated by adjusting the profile weight of the supplier profile indicator; the predicted value of the indicator is a real-time predicted value calculated based on the historical indicator data of the supplier profile indicator, and the supplier profile indicator includes any one of the following: performance delay rate, quality pass rate, financial health index and judicial public opinion compliance. The Top-K vector generation unit is used to select feature mapping relationships according to the procurement project type, extract the Top-K feature subset with the highest relevance to the procurement demand vector from the updated supplier profile, and generate a supplier-specific representation vector by weighting. The supplier screening unit is used to calculate the similarity distance between the supplier-specific representation vector and the procurement demand vector in order to screen candidate suppliers; The step of parsing the engineering procurement requirement text based on a preset set of procurement feature dimensions to generate a corresponding procurement requirement vector includes: For each procurement feature dimension in the preset procurement feature dimension set, the feature time sequence value corresponding to the procurement feature dimension is extracted from the engineering procurement requirement text according to the continuous bidding release time. A procurement demand vector is generated based on the time series values ​​of the features corresponding to each of the procurement feature dimensions.

Images