Artificial intelligence-based procurement project matching method, apparatus, and system

Abstract

The application provides an artificial intelligence-based procurement project matching method, device and system, which comprises the following steps: obtaining a plurality of demand data of a purchaser and corresponding quantization into vector feature data; generating a weight value of each vector feature data in real time; calculating a supply-demand matching degree of the purchaser and a supplier based on the vector feature data and the corresponding weight value by using a Jaccard coefficient algorithm with a regional attenuation factor; and outputting a matching result of the procurement project according to the supply-demand matching degree. In the application, the problems of low efficiency, insufficient precision, poor dynamic adaptability and missing regional influence quantization in the current procurement project matching method are solved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a procurement project matching method, apparatus and system based on artificial intelligence. Background Technology

[0002] In the field of enterprise procurement management, accurate matching of procurement projects with suppliers is a core element in improving procurement efficiency, controlling procurement costs, and ensuring supply chain stability. With the development of the market economy, procurement demands are becoming increasingly diversified and personalized, and supplier resources are becoming increasingly abundant. Traditional procurement project matching methods are no longer sufficient to meet practical application needs, gradually revealing numerous technical shortcomings: First, traditional procurement matching relies heavily on manual screening or simple keyword matching, which is inefficient and highly subjective. In manual screening, procurement personnel must compare supplier qualifications, product parameters, and quotations one by one. Faced with a vast pool of supplier resources, this not only consumes significant manpower and time but is also prone to biases due to individual differences in experience. Simple keyword matching, on the other hand, can only identify explicit overlaps between procurement needs and supplier information, failing to capture implicit needs (such as a preference for potential substitutes or preferred cooperation models) and implicit supplier advantages (such as the ability to fulfill obligations in specific scenarios). This results in insufficient matching accuracy, allowing many suppliers that do not meet actual needs to enter the candidate pool, increasing subsequent screening costs.

[0003] Secondly, traditional methods lack a dynamic adjustment mechanism for feature importance. The importance of different features in procurement needs (such as the type of procurement target, compliance and qualification requirements, and cost sensitivity coefficient) changes with the procurement scenario. However, existing matching methods mostly adopt fixed weight allocation strategies or can only adjust weights manually. They cannot dynamically optimize feature weights based on the industry attributes, urgency, and procurement scale of the procurement project. This results in key features not being given sufficient attention, while non-key features excessively influence the matching results, further reducing matching accuracy.

[0004] Furthermore, the impact of geographical factors on procurement matching has not been scientifically quantified. In actual procurement scenarios, the geographical distance between suppliers and buyers directly affects logistics costs, delivery timeliness, and after-sales response efficiency. This is especially true for procurement projects with high logistics dependence, such as fresh produce, cold chain logistics, and large equipment, where geographical suitability is a core consideration. Traditional methods either completely ignore geographical factors or fail to quantify the geographical attenuation effect based on the logistics cost characteristics of different industries (such as the difference in geographical impact between industries with high and low logistics costs). This leads to the priority recommendation of suppliers with poor geographical suitability but matching characteristics in other aspects, increasing procurement risks and costs.

[0005] In summary, current procurement project matching methods suffer from low efficiency, insufficient accuracy, poor dynamic adaptability, and a lack of quantitative analysis of regional influences. Summary of the Invention

[0006] The main objective of this invention is to provide a procurement project matching method, apparatus, and system based on artificial intelligence, aiming to overcome the problems of low efficiency, insufficient accuracy, poor dynamic adaptability, and lack of quantitative analysis of regional influence in current procurement project matching methods.

[0007] To achieve the above objectives, the present invention provides a procurement project matching method based on artificial intelligence, comprising the following steps: Acquire multiple requirements from the purchasing party and quantify them into vector feature data accordingly; Real-time generation of weight values ​​for each vector feature data; Based on the vector feature data and its corresponding weight values, the Jaccard coefficient algorithm with regional attenuation factor is used to calculate the supply and demand matching degree between the purchaser and the supplier. Based on the supply and demand matching degree, output the matching results of the procurement project.

[0008] Furthermore, the purchaser's demand data includes the type of procurement target, compliance and qualification requirements, cost sensitivity coefficient, and supplier data.

[0009] Furthermore, the supplier data includes parameters of the supply target, credit rating score, and regional suitability index.

[0010] Furthermore, weight values ​​for each vector feature data are generated in real time, including: Obtain historical transaction data of the industry to which the procurement project belongs, and calculate the initial weight values ​​of each vector feature data; Obtain matching results feedback data for similar procurement projects and dynamically adjust the initial weight values; the feedback data includes the transaction rate and performance satisfaction scores of both the supply and demand sides; Based on the urgency and scale of the current procurement project, a scenario influence factor matrix is ​​established. The dynamically corrected weight values ​​are then adjusted a second time to generate the final real-time weight values ​​for each vector feature data.

[0011] Furthermore, based on the supply-demand matching degree, the matching results of the procurement project are output, including: The suppliers are sorted from highest to lowest according to their supply-demand matching scores, and then a second sorting calibration is performed based on the suppliers' real-time response capability parameters to obtain the final sorting result. In-depth feature matching analysis was performed on the top 20% of suppliers in the ranking results to generate a heat map of supply and demand differences, and the difference analysis results with specific differences in key indicators were obtained. Based on the variance analysis results, matching results for the procurement project are generated.

[0012] Furthermore, based on the vector feature data and its corresponding weight values, the Jaccard coefficient algorithm, which incorporates regional attenuation factors, is used to calculate the supply-demand matching degree between the purchaser and the supplier, including: Construct feature sets for both the purchaser and the supplier, and perform feature matching; both sets contain vector feature data of the same dimension. Calculate the number of matching features between the buyer and the supplier, and divide it by the total number of all features of both parties to obtain the basic matching coefficient. The geographical attenuation factor is determined based on the straight-line distance between the purchaser and the supplier; the greater the distance, the smaller the geographical attenuation factor. For each successfully matched feature, the weighted matching degree is calculated by summing its corresponding weight value. Multiplying the basic matching coefficient, weighted matching degree, and regional attenuation factor together yields the supply-demand matching degree between the purchaser and the supplier.

[0013] Furthermore, multiple demand data from the purchasing party are obtained and quantified into vector feature data, including: Obtain the procurement party's demand data and classify it into structured and unstructured data. The structured data includes the type, quantity, and budget range of the procurement target, while the unstructured data includes procurement demand description text, historical cooperation preference documents, and policy compliance statements. Semantic extraction is performed on unstructured data using a bidirectional Transformer model to extract implicit demand features, including potential substitute targets, cooperation mode preferences, and risk tolerance thresholds. Based on a pre-defined feature quantization rule library, structured data and implicit requirement features are uniformly quantified to obtain quantified feature data. Calculate the correlation between the quantitative feature data and the historical demand features of similar procurement projects, and correct the deviation of the quantitative feature data to obtain vector feature data; For time-sensitive demand parameters in vector feature data, an attenuation coefficient is set, and their proportion in the vector feature data is dynamically adjusted over time.

[0014] This invention also provides an artificial intelligence-based procurement project matching system, comprising: The acquisition module is used to acquire multiple demand data from the purchaser and quantify them into vector feature data. The generation module is used to generate weight values ​​for each vector feature data in real time. The calculation module is used to calculate the supply and demand matching degree between the purchaser and the supplier based on the vector feature data and its corresponding weight values, using the Jaccard coefficient algorithm that incorporates regional attenuation factors. The output module is used to output the matching results of procurement projects based on the supply and demand matching degree.

[0015] The present invention also provides an artificial intelligence-based procurement project matching device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0016] This invention provides an AI-based procurement project matching method, apparatus, and system, comprising: acquiring multiple demand data from the purchaser and quantifying them into vector feature data; generating weight values ​​for each vector feature data in real time; calculating the supply-demand matching degree between the purchaser and the supplier using a Jaccard coefficient algorithm incorporating a regional attenuation factor based on the vector feature data and their corresponding weight values; and outputting the matching result of the procurement project based on the supply-demand matching degree. In this invention, by acquiring multiple demand data from the purchaser, quantifying them into vector feature data, and generating weight values ​​for each vector feature data in real time, accurate quantification of demand features and dynamic adjustment of weights are achieved. By using a Jaccard coefficient algorithm incorporating a regional attenuation factor to calculate the supply-demand matching degree between the purchaser and the supplier, regional factors are scientifically integrated, solving the problems of low efficiency, insufficient accuracy, poor dynamic adaptability, and lack of quantitative quantification of regional influence in current procurement project matching methods. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the steps of a procurement project matching method based on artificial intelligence in one embodiment of the present invention; Figure 2 This is a block diagram of a procurement project matching system based on artificial intelligence in one embodiment of the present invention; Figure 3 This is a schematic block diagram of a procurement project matching device based on artificial intelligence according to an embodiment of the present invention.

[0018] The implementation, functional features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] Reference Figure 1 One embodiment of the present invention provides a procurement project matching method based on artificial intelligence, including the following steps: Step S1: Obtain multiple demand data from the purchasing party and quantify them into vector feature data accordingly; Step S2: Generate the weight values ​​of each vector feature data in real time; Step S3: Based on the vector feature data and its corresponding weight values, the Jaccard coefficient algorithm with regional attenuation factor is used to calculate the supply and demand matching degree between the purchaser and the supplier. Step S4: Output the matching results of the procurement project based on the supply and demand matching degree.

[0021] It is particularly important to note that all technical steps, algorithm applications, and parameter settings in the technical solution of this application have clear technical objectives and application value. They do not utilize complex steps and algorithmic formulas to achieve simple functions. To provide detailed explanations of each step and avoid ambiguity, some conventional algorithms are used for illustration. However, this does not mean that the algorithms and technical features listed herein are the only way to implement the technical solution of this application, nor is it intended to limit the scope of protection of this application. This application is not a combination or stacking of the listed algorithms and technical features; its essence is to exemplify the implementation methods of this application to fully explain it. It does not pursue formal complexity by adding meaningless technical steps, nor does it involve the accumulation of technologies divorced from practical needs; it conforms to the conventional logic of technical improvement and design.

[0022] In this embodiment, as described in step S1 above, the core is to transform the dispersed and non-standardized demand information of the purchasing party into structured vector data that can be computed by a computer, providing a data foundation for subsequent matching calculations. First, the demand data of the purchasing party is comprehensively collected, including explicit structured information such as the type of procurement target, budget, and clear compliance qualification requirements, as well as implicit demands (such as potential expectations for delivery cycle and preferences for cooperation mode) mined from demand description texts and historical cooperation records.

[0023] Next, standardized quantification rules were established for different types of demand data: for information with clear judgment standards, such as compliance qualifications, binary encoding was used (1 for compliance, 0 for non-compliance); for demands with varying degrees of sensitivity, such as cost sensitivity, different tiers were defined based on the actual procurement scenario and assigned corresponding values; for regional adaptability requirements, values ​​were mapped to fixed ranges based on factors such as logistical convenience. Finally, all quantified demand data was integrated into vector feature data with unified dimensions to ensure that the data format met the requirements of subsequent calculations and was consistent with the actual needs of the procurement party.

[0024] As described in step S2 above, the aim is to dynamically adjust the importance of each requirement feature based on the procurement scenario, avoiding matching bias caused by fixed weights. First, a weight evaluation logic centered on historical data, real-time feedback, and scenario parameters is constructed: based on historical transaction data of the industry to which the procurement project belongs, the correlation between different requirement features and transaction results is analyzed to determine the initial weight of each feature, ensuring that core requirements (such as the type of procurement target) occupy fundamental importance in the matching process.

[0025] Subsequently, the initial weights are dynamically adjusted based on real-time feedback data from similar procurement projects (such as success rate and performance satisfaction). If a certain feature has a higher success rate, it indicates a greater impact on the procurement decision, and its weight is appropriately increased; conversely, it is decreased. Finally, the adjusted weights are further optimized by considering the characteristics of the current procurement project (such as urgency and procurement scale). Ultimately, normalization is used to ensure that the sum of all feature weights is 1, generating real-time weights that conform to the current scenario.

[0026] As described in step S3 above, it is necessary to comprehensively consider demand feature matching, weight differences, and regional influences to achieve accurate quantification of supply and demand matching. First, a supplier feature set consistent with the buyer's feature dimensions is constructed, covering parameters of the supply target, qualification status, price suitability, regional attributes, etc., and feature matching between the two parties is completed dimension by dimension. For binary-coded qualification features, consistency is directly compared; for continuous features such as cost sensitivity, the degree of matching is determined by calculating the overlap ratio of numerical intervals.

[0027] Next, the basic Jaccard matching coefficient is calculated: the number of features that are successfully matched between the two parties is divided by the total number of all features of both parties (excluding duplicate matching features) to obtain a basic index reflecting the overall feature overlap. At the same time, a geographical attenuation factor is introduced, and the factor value is dynamically adjusted according to the geographical distance between the buyer and the supplier (the closer the distance, the closer the factor is to 1, and the farther the distance, the smaller the factor is), and the factor attenuation magnitude is in line with the characteristics of industry logistics costs.

[0028] Finally, the weights of each matching feature are summed to obtain a weighted total score, which is then combined with the basic Jaccard coefficient and the regional attenuation factor to calculate the result. The result is standardized to the 0-100% range, and finally a supply and demand matching degree that can comprehensively reflect feature matching, core demand fit and regional adaptability is generated.

[0029] As described in step S4 above, the quantified matching degree is transformed into a structured result that the purchaser can directly use to assist in procurement decisions. First, based on the supply and demand matching degree as the core criterion, all suppliers are initially ranked from high to low according to their matching degree. Then, a second calibration is performed by combining the suppliers' real-time response capabilities (such as response time and online status) to ensure that the ranking considers both matching accuracy and the convenience of actual cooperation.

[0030] Subsequently, in-depth feature matching analysis is conducted on high-priority suppliers (such as those ranked highly). The matching advantages and differences between the two parties are presented through visualization (such as a difference heatmap), and an assessment is made as to whether the differences can be compensated for by other supplier strengths, providing the purchaser with a basis for differentiated judgment. Finally, the matching results are output in a tiered manner: quick recommendation cards containing core information (matching degree, core advantages, contact information) are provided for quick filtering by the purchaser; comparative data covering detailed feature matching and performance evaluation is also provided to support in-depth comparison; and for suppliers with low matching degrees, the reasons for non-compliance and directions for improvement are clearly marked. Furthermore, a dynamic update mechanism is established to refresh the matching results in real time based on changes in supplier information, ensuring the timeliness of the output results.

[0031] In this embodiment, multiple demand data from the purchaser are acquired, quantified into vector feature data, and weight values ​​for each vector feature data are generated in real time, achieving accurate quantification of demand features and dynamic adjustment of weights. The supply-demand matching degree between the purchaser and the supplier is calculated by integrating the Jaccard coefficient algorithm with the regional attenuation factor, scientifically integrating regional factors and solving the problems of low efficiency, insufficient accuracy, poor dynamic adaptability, and lack of quantitative analysis of regional influence in current procurement project matching methods.

[0032] In one embodiment, the purchaser's demand data includes the type of the purchase target, compliance and qualification requirements, cost sensitivity coefficient, and supplier data.

[0033] In one embodiment, the supplier data includes parameters of the supply target, credit rating score, and regional suitability index.

[0034] In one embodiment, the weight values ​​of each vector feature data are generated in real time, including: Obtain historical transaction data of the industry to which the procurement project belongs, and calculate the initial weight values ​​of each vector feature data; Obtain matching results feedback data for similar procurement projects and dynamically adjust the initial weight values; the feedback data includes the transaction rate and performance satisfaction scores of both the supply and demand sides; Based on the urgency and scale of the current procurement project, a scenario influence factor matrix is ​​established. The dynamically corrected weight values ​​are then adjusted a second time to generate the final real-time weight values ​​for each vector feature data.

[0035] In this embodiment, firstly, the scope and standards of data collection are defined: focusing on the industry to which the procurement project belongs, historical transaction data from the past 2-3 years are collected. The data must cover complete characteristics of procurement needs (such as the type of procurement target, compliance qualification requirements, cost sensitivity coefficient, etc.) and corresponding transaction results (such as the final selected supplier, transaction amount, cooperation period, etc.). At the same time, incomplete data (such as missing key features, unrecorded transaction results) and abnormal transactions (such as emergency procurement, single-source procurement, and other atypical scenario transactions) are removed to ensure the integrity and representativeness of historical data.

[0036] Next, a feature-transaction correlation analysis model is established: purchasing demand features in historical transaction data are correlated and matched with transaction results, and the frequency of contribution of each vector feature in facilitating the transaction is statistically analyzed. For example, in 100 historical transactions, 92 transactions have a perfect match in the type of the purchased goods, 88 transactions meet all compliance and qualification requirements, and 75 transactions have a suitable cost sensitivity coefficient. This statistical analysis clarifies the degree of correlation between different features and transaction results. Subsequently, feature importance quantification algorithms (such as information gain algorithm and chi-square test algorithm) are used to quantify the degree of correlation: taking the information gain algorithm as an example, the reduction in uncertainty of the transaction result when a certain feature is present or absent (i.e., the information gain value) is calculated. The higher the information gain value, the greater the impact of the feature on the transaction result, and the higher the corresponding initial weight value. Finally, the quantification results of all features are normalized to ensure that the sum of the initial weight values ​​of all vector feature data is 1, generating an initial weight allocation scheme.

[0037] Next, obtain the matching results feedback data of similar procurement projects and dynamically adjust the initial weight values.

[0038] First, define the criteria for similar procurement projects and the feedback data collection cycle: similar procurement projects must meet three core conditions: consistent industry attributes, similar types of procurement targets, and the same procurement scale range; the feedback data collection cycle is set to the past 30-60 days to ensure that the data can reflect recent market dynamics and changes in procurement preferences, and to avoid the feedback data becoming invalid due to an excessively long cycle.

[0039] Then, a quantitative evaluation system for feedback data is constructed: For the transaction rate feedback data, the difference in the probability of success for a specific vector feature when it matches versus when it doesn't is statistically analyzed in similar procurement projects. For example, in similar projects, the transaction rate for suppliers matching compliance requirements is 65%, while the transaction rate for suppliers not matching compliance requirements is only 12%. By calculating the ratio of (matching transaction rate - non-matching transaction rate) / non-matching transaction rate, the actual impact of this feature on the transaction result is quantified. The greater the impact, the more important this feature needs to be at the current stage. Through this type of correlation analysis, the correlation between feature matching degree and performance quality is determined; the higher the correlation, the greater the need to adjust the weight of this feature accordingly.

[0040] Subsequently, a dynamic weighting adjustment rule was established: based on the quantitative results of the feedback data, adjustment coefficients were set. For features with a transaction rate impact exceeding 50% and a performance satisfaction correlation exceeding 70%, their initial weight values ​​were increased by 8%-12%; for features with a transaction rate impact between 20%-50% and a performance satisfaction correlation between 40%-70%, their initial weight values ​​were increased by 3%-7%; and for features with a transaction rate impact below 20% and a performance satisfaction correlation below 40%, their initial weight values ​​were decreased by 1%-3%, ensuring that the adjusted weights accurately reflect the matching effects of similar recent projects. Finally, the adjusted weight values ​​were verified again to ensure that the sum of the weights of all features remained 1, forming the dynamically adjusted weighting scheme.

[0041] Finally, based on the urgency and scale of the current procurement project, a scenario influence factor matrix is ​​established. The dynamically corrected weight values ​​are then adjusted a second time to generate the final real-time weight values ​​for each vector feature data. The key is to ensure that the weight values ​​accurately match the specific scenario of the current procurement project, solving the problem that general weights cannot meet the needs of personalized scenarios, and ensuring that the final weights are scenario-specific. First, the urgency and scale of the current procurement project are categorized and quantified: For urgency, it is divided into three levels based on the project's delivery cycle requirements: Urgent (delivery cycle ≤ 7 days), Urgent (delivery cycle 8-30 days), and Routine (delivery cycle > 30 days), and assigned scenario influence factors of 1.8, 1.4, and 1.0 respectively; for procurement scale, it is divided into four levels based on the amount of a single purchase. The categories of ultra-large (purchase amount > 10 million yuan), large (purchase amount 1 million to 10 million yuan), medium (purchase amount 100,000 to 1 million yuan), and small (purchase amount < 100,000 yuan) are assigned scenario impact factors of 1.6, 1.3, 1.1, and 1.0, respectively. The grading standards and impact factors can be fine-tuned according to industry characteristics. For example, the impact factor of large projects in the construction industry can be appropriately increased, while the impact factor of small projects in the office supplies procurement industry can be kept at a lower level.

[0042] Next, a scenario influence factor matrix is ​​established: a two-dimensional matrix is ​​constructed with the type of vector feature data as rows and the urgency-procurement scale combination scenario as columns. Then, a secondary adjustment calculation is performed: the dynamically corrected weight values ​​of each vector feature are multiplied by the corresponding scenario influence factor under the current procurement project urgency-procurement scale combination scenario to obtain the secondary adjusted weight values.

[0043] Finally, weight constraints and normalization are performed: weight constraints are set to ensure that the final weight value of a single vector feature data does not exceed 45% (to avoid a certain feature from overly dominating the matching result) and is not less than 3% (to avoid secondary features being ignored due to low weight). If the weight value of a certain feature after secondary adjustment exceeds the constraint range, it is truncated according to the upper or lower limit of the constraint. Then, the weight values ​​of all features are normalized again to ensure that the sum of the weight values ​​of all vector feature data is 1, generating the final real-time weight value that meets the needs of the current procurement project scenario.

[0044] In one embodiment, the matching result of the procurement project is output based on the supply and demand matching degree, including: The suppliers are sorted from highest to lowest according to their supply-demand matching scores, and then a second sorting calibration is performed based on the suppliers' real-time response capability parameters to obtain the final sorting result. In-depth feature matching analysis was performed on the top 20% of suppliers in the ranking results to generate a heat map of supply and demand differences, and the difference analysis results with specific differences in key indicators were obtained. Based on the variance analysis results, matching results for the procurement project are generated.

[0045] In this embodiment, firstly, a basic ranking is performed, sorting all participating suppliers in descending order based on their supply-demand matching degree (a quantitative value in the range of 0-100%). Next, a secondary calibration is performed using a supplier real-time response capability parameter. This parameter is designed to reflect the critical impact of response efficiency on cooperation progress in actual procurement scenarios, specifically covering three core dimensions: First, average response time, which is the average time from receiving a request to the first response from the supplier for similar procurement needs historically (e.g., "respond within 1 hour," "respond within 4 hours," "respond within 24 hours"). Shorter times indicate more timely responses. Second, current online status, which involves real-time monitoring of supplier account activity (e.g., "online," "offline," "busy"). Suppliers in an online state can handle request communication more quickly, reducing waiting costs. Third, historical response rate, which calculates the proportion of suppliers responding to procurement needs in the past 30 days (e.g., "received 100 requests, responded to 95," resulting in a 95% response rate). A higher response rate indicates a stronger willingness and commitment from the supplier to cooperate with the procurement needs.

[0046] Subsequently, a secondary calibration rule was established: for suppliers in the initial ranking who are adjacent and have a matching difference of ≤5% (these suppliers have similar core compatibility and need to be further differentiated through responsiveness), the order was readjusted based on their comprehensive real-time response capability score. The comprehensive score was calculated using a weighted average, with suppliers having higher comprehensive scores ranked higher.

[0047] Next, a deep feature matching analysis is performed on the top 20% of suppliers in the ranking results, generating a heatmap of supply and demand differences and obtaining the difference analysis results with specific differences in key indicators. By deeply breaking down the differences in supply and demand characteristics, the purchasing party can clearly identify the advantages and potential risks of suppliers, avoiding the problem of superficial high matching but hidden mismatch. First, the analysis object and key indicator scope are determined: from the ranking results obtained in the first step, the top 20% of suppliers are selected as the analysis sample (these suppliers are the core candidates and need to be evaluated in detail), and the key analysis indicators are defined. The indicators need to cover the core dimensions of procurement needs, including the matching rate of procurement target type, compliance qualification compliance rate, cost sensitivity fit, regional adaptability index, delivery capability fit, etc., to ensure that the analysis dimensions are comprehensive and closely aligned with the core procurement needs.

[0048] Next, a deep feature matching comparison is performed: the feature data of the top 20% of suppliers are broken down and compared with the characteristics of the buyer's needs item by item, and the specific difference value of each key indicator is calculated. Then, a supply and demand difference heatmap is generated: using a visual heatmap tool, a two-dimensional heatmap matrix is ​​constructed with the key indicators as the horizontal axis and the top 20% of suppliers as the vertical axis, and different color gradients are used to indicate the magnitude of the difference values. Through the heatmap, the buyer can intuitively identify the difference distribution of each high-priority supplier, and finally form a difference analysis result with specific difference values ​​of key indicators.

[0049] Finally, based on the variance analysis results, matching results for the procurement project are generated. The core is to transform the variance analysis results into structured matching outcomes that the procuring entity can directly decide upon, ensuring that the results contain core information and support the procuring entity's targeted decision-making. First, a hierarchical logic for the results is constructed: based on the degree and type of variance in the variance analysis results, the top 20% of suppliers are further divided into three levels: Priority Recommendation, Conditional Recommendation, and Observational. Priority Recommendation suppliers have a perfect match to the core requirements, and cooperation can be initiated with only simple communication; Conditional Recommendation suppliers require the procuring entity to communicate with the supplier to confirm the variance solution before proceeding; Observational suppliers are not recommended at this time, but the reasons for the variance must be recorded as a reference for subsequent requirement adjustments or supplier selection.

[0050] Next, tiered matching results are generated: For priority recommended suppliers, a quick decision card is generated, covering the supplier's name, supply-demand matching degree, real-time response capability score, core advantages, contact information, and historical performance evaluation, facilitating quick communication by the purchaser; for conditionally recommended suppliers, a negotiation proposal is generated, clearly indicating differences, negotiable space for the supplier, and the purchaser's negotiation suggestions, providing direction for negotiations between both parties; for suppliers under observation, a difference explanation report is generated, detailing significant differences, their impact on procurement, and supplier improvement suggestions, providing a basis for subsequent follow-up.

[0051] Finally, the results of the hierarchical integration are used to form the final matching result: the quick decision card, the negotiation proposal, and the difference explanation report are integrated according to priority, and a complete version of the supply and demand difference heat map is attached for the convenience of the purchaser to review the whole.

[0052] In one embodiment, based on the vector feature data and its corresponding weight values, the Jaccard coefficient algorithm with fused regional attenuation factors is used to calculate the supply-demand matching degree between the purchaser and the supplier, including: Construct feature sets for both the purchaser and the supplier, and perform feature matching; both sets contain vector feature data of the same dimension. Calculate the number of matching features between the buyer and the supplier, and divide it by the total number of all features of both parties to obtain the basic matching coefficient. The geographical attenuation factor is determined based on the straight-line distance between the purchaser and the supplier; the greater the distance, the smaller the geographical attenuation factor. For each successfully matched feature, the weighted matching degree is calculated by summing its corresponding weight values. Multiplying the basic matching coefficient, weighted matching degree, and regional attenuation factor together yields the supply-demand matching degree between the purchaser and the supplier.

[0053] In this embodiment, firstly, the construction criteria for the feature set are defined: based on the quantified vector feature data of the purchaser, the range of feature dimensions is determined. This covers core dimensions such as the type of procurement target, compliance and qualification requirements, cost sensitivity coefficient, regional adaptability requirements, and delivery capability requirements, ensuring that each dimension corresponds to the purchaser's key needs. Next, supplier feature data collection and quantification are performed: for each supplier, information corresponding to the purchaser's feature dimensions is collected and quantified, with the quantification rules consistent with the purchaser's vector features. Through unified quantification rules, the supplier's unstructured information is transformed into vector feature data in the same format as the purchaser, forming a supplier feature set. Finally, feature matching is performed dimension-by-dimensionally: in a dimension-to-dimensional manner, it is determined whether the purchaser and supplier match on each feature dimension. For binary quantification features (such as whether the target type is consistent, 1 for consistency and 0 for inconsistency), directly compare whether the value is 1; for range or continuous quantification features (such as cost sensitivity and regional suitability index), if the supplier's quantification value falls within the buyer's demand range (or reaches the threshold set by the buyer), it is judged as a match (recorded as 1), otherwise it is judged as a mismatch (recorded as 0).

[0054] Next, the number of matching features between the buyer and supplier is calculated and divided by the total number of features from both sides to obtain the basic matching coefficient. This coefficient aims to quantify the overall compatibility between the supply and demand sides through feature overlap rate, and is the core application of the Jaccard coefficient algorithm. First, the core data is statistically analyzed: one is the number of matching features, and the other is the total number of features from both sides. Since the feature sets of the buyer and supplier have completely identical dimensions, and the same dimension is not repeatedly calculated during feature matching, the total number equals the number of dimensions in the feature set. Therefore, there is no need to additionally remove duplicates, simplifying the calculation logic while ensuring data accuracy.

[0055] Next, the basic matching coefficient is calculated: according to the core formula of the Jaccard coefficient (number of overlapping features / total number of features), the number of matching features is divided by the total number of features to obtain the basic matching coefficient. The coefficient ranges from 0 to 1 (or can be converted to 0%-100%), which directly reflects the overall overlap ratio of features between the supply and demand sides.

[0056] Then, the regional attenuation factor is determined based on the straight-line distance between the buyer and the supplier; the greater the distance, the smaller the regional attenuation factor. Its core function is to quantify the impact of regional factors on procurement cooperation, addressing the shortcomings of traditional matching methods that ignore regional correlations such as logistics costs and delivery time. First, the distance calculation standard is determined: the precise geographical coordinates (such as latitude and longitude) of the buyer and supplier are obtained through a Geographic Information System (GIS), and the straight-line distance between them (in kilometers) is calculated. This distance serves as the core input parameter for the regional attenuation factor, ensuring the objectivity and accuracy of the data and avoiding errors caused by fuzzy classifications such as "same city" or "different location."

[0057] Next, the dynamic calculation rules for the regional attenuation factor are set: the factor value ranges from 0 to 1, and the core logic is that distance is negatively correlated with the factor. The closer the distance, the lower the logistics costs, the faster the delivery time, and the more timely the after-sales response; the smaller the negative impact of location on cooperation, and the closer the factor is to 1. The farther the distance, the weaker the above advantages, the greater the negative impact of location, and the closer the factor is to 0. At the same time, to adapt to the differences in regional sensitivity in different industries, industry calibration parameters can be set to ensure that the factor can truly reflect the impact of location on procurement in different scenarios through differentiated attenuation amplitudes.

[0058] Then, factor calculation and verification are performed: the straight-line distance is substituted into the preset attenuation rule to calculate the regional attenuation factor, and at the same time, the factor is verified to ensure that it falls within the 0-1 range, so as to ensure the rationality of the factor value and provide an effective basis for the quantitative analysis of regional impact for subsequent fusion calculation.

[0059] Then, for each successfully matched feature, the weighted matching degree is calculated by summing the corresponding weight values. The key is to highlight the importance of core features and address the limitation of treating all features equally in the basic matching coefficient, making the matching degree calculation more aligned with the actual needs and priorities of the purchaser. First, key data is extracted: one is the list of successfully matched features, i.e., the feature dimensions determined to be matched in the first step; the other is the real-time weight value corresponding to each feature. This weight value comes from the previous real-time weight generation step and has been optimized by combining historical industry data, feedback data, and scenario parameters. For example, the target type has a weight of 35%, compliance qualifications 30%, cost sensitivity 20%, regional adaptability 10%, and delivery cycle 5%.

[0060] Next, a weighted cumulative calculation is performed: for each successfully matched feature, its corresponding real-time weight value is directly extracted and accumulated to obtain the weighted matching degree. This value ranges from 0 to 1 (or 0% to 100%, which must be consistent with the normalized format of the weight values). It is important to note that the weighted matching degree only calculates the sum of the weights of successfully matched features; the weights of unmatched features are not included in the calculation. Its core significance is that even if the overall feature overlap rate (basic matching coefficient) of the supply and demand sides is the same, if the matched features are mostly high-weight core needs, the weighted matching degree is higher, indicating that the two sides have a better fit in key needs; conversely, the fit in key needs is insufficient.

[0061] Finally, the basic matching coefficient, weighted matching degree, and regional attenuation factor are multiplied to obtain the supply-demand matching degree between the purchaser and the supplier. The core reason for multiplying these three factors is that each parameter represents a key dimension influencing the matching degree, and these dimensions are independent of each other but need to work together. The basic matching coefficient ensures overall fit, the weighted matching degree ensures the priority of core needs, and the regional attenuation factor ensures regional adaptability. Only when all three are at a high level will the final matching degree be high. If there is a weakness in one dimension (such as a regional attenuation factor that is too low), even if other dimensions are excellent, the final matching degree will be lowered, which meets the need for multi-dimensional comprehensive evaluation in actual procurement scenarios. The final matching degree is ensured to fall within the 0%-100% range. If the result exceeds the range due to parameter anomalies (such as a weighted matching degree > 1 due to incorrect weight calculation), a parameter backtracking verification mechanism is triggered to correct the previous data and recalculate, ensuring the accuracy and usability of the final result.

[0062] In one embodiment, multiple demand data from the purchasing party are acquired and quantified into corresponding vector feature data, including: Obtain the procurement party's demand data and classify it into structured and unstructured data. The structured data includes the type, quantity, and budget range of the procurement target, while the unstructured data includes procurement demand description text, historical cooperation preference documents, and policy compliance statements. Semantic extraction is performed on unstructured data using a bidirectional Transformer model to extract implicit demand features, including potential substitute targets, cooperation mode preferences, and risk tolerance thresholds. Based on a pre-defined feature quantization rule library, structured data and implicit requirement features are uniformly quantified to obtain quantified feature data. Calculate the correlation between the quantitative feature data and the historical demand features of similar procurement projects, and correct the deviation of the quantitative feature data to obtain vector feature data; For time-sensitive demand parameters in vector feature data, an attenuation coefficient is set, and their proportion in the vector feature data is dynamically adjusted over time.

[0063] In this embodiment, firstly, the procurement needs information is collected comprehensively through multiple channels, including the procurement management system, demand forms, and offline records, covering both explicit requirements and implicit preferences to avoid missing key information. Then, the data is categorized by format: structured data contains information with clearly defined fields that can be directly quantified, including the type of procurement item (e.g., industrial-grade PLC controllers), quantity (e.g., 50 units), and budget range (e.g., 100,000-150,000 RMB), which can be used for subsequent quantification without additional parsing; unstructured data is text-based information without a fixed format, including demand description text (e.g., requiring support for remote fault diagnosis), historical cooperation preference documents (e.g., prioritizing 60-day payment terms), and policy compliance statements (e.g., complying with the "Regulations for the Implementation of the Government Procurement Law"). This type of data contains implicit requirements and requires professional model parsing. This categorization achieves hierarchical data management, laying the foundation for subsequent processing.

[0064] Next, a bidirectional Transformer model (such as BERT) is employed, leveraging its bidirectional contextual semantic encoding capabilities to accurately capture the deep intent of the text, outperforming traditional keyword matching. Features are extracted from three types of unstructured data: expressions indicating substitutability are identified in the requirement description text to extract potential alternatives (e.g., a DCS control system can be chosen when out of stock); delivery and settlement preferences are analyzed from historical cooperation documents to extract cooperation model features (e.g., phased delivery + payment upon acceptance); and risk tolerance thresholds are quantified by combining policy compliance statements (e.g., FDA certification is mandatory, delays ≤3 days are acceptable). Finally, the extracted implicit features are organized into standardized fields, forming a complete requirement system together with the structured data.

[0065] Then, based on the rule base, structured and implicit features are uniformly quantified, with the goal of eliminating data format differences and transforming them into computer-calculateable numerical values. First, a pre-defined feature quantification rule base is constructed, combining industry characteristics and historical experience to formulate quantification rules covering all dimensions and dynamically updating them. Next, two types of data are quantified separately: structured data is transformed according to rules, target types are coded using industry standards, quantities are converted to actual values ​​or units, and budgets are taken as the median of an interval; implicit features are quantified by dimension, substitute targets are prioritized, cooperation models are based on compatibility, and risk thresholds are based on tolerance. Finally, all data is transformed into 0-1 or specific numerical values, achieving format unification and information quantification.

[0066] Furthermore, historical data verification ensures that the quantitative results align with the procurement intent. First, the database is used to screen similar historical projects in the same industry, target, and scale, extracting verified quantitative demand data to construct a historical feature library as a reference benchmark. Then, algorithms such as Pearson correlation coefficient and cosine similarity are used to calculate the correlation between the current quantitative data and historical data: a correlation of ≥80% indicates no significant deviation; <80% requires analysis of the reasons for the deviation. For example, if a budget of 50,000 yuan is significantly lower than 100,000-150,000 yuan for similar projects, it's necessary to investigate whether it was entered incorrectly or represents a special requirement. During corrections, ambiguous descriptions prompt the procurement party to confirm recalculation; special requirements are noted with reasons and retained values; and outdated rules are updated in the rule library. Finally, the corrected quantitative data is integrated and formed into vector feature data according to preset dimensions.

[0067] Finally, define time-sensitive parameters, such as urgent procurement (7-day delivery), temporary qualification requirements (test reports within the last month), and short-term cooperation preferences (partial delivery this time). The importance of these parameters decreases over time. Then, formulate decay rules based on the strength of timeliness: urgent procurement decays according to the remaining time, temporary qualifications decay according to their validity period, and short-term preferences decay according to project stage. The system uses scheduled tasks (e.g., every 24 hours) to check parameter status, automatically update decay coefficients and adjust proportions (original quantified value × coefficient), and record adjustment logs. When the proportion falls below a threshold (e.g., 0.2), the procurement party is reminded to update their requirements, ensuring the data accurately reflects current needs.

[0068] In one embodiment, the method further includes: The required data is classified into core sensitive data, general sensitive data, and non-sensitive data according to its sensitivity level. Obtain the dimensional features of the required data, and dynamically update the preset encryption algorithm based on the dimensional features; extract the key parameters of the updated encryption algorithm, split them into character fragments, and construct a first binary tree based on the character fragments; Extract the feature values ​​of core sensitive data, general sensitive data, and non-sensitive data, and construct a second binary tree based on each feature value; The first binary tree and the second binary tree are hierarchically aligned and superimposed. The filtering rules are determined by the character characteristics of the root nodes of the two binary trees. The target node is then filtered from other nodes based on the filtering rules. By concatenating the characters on the target node in sequence, a desensitization factor is obtained. Based on the desensitization factor, the required data of different sensitivity levels are desensitized in a differentiated manner and then stored in the database.

[0069] In this embodiment, firstly, a sensitivity level standard for the procurement industry is established: core sensitive data (budget range, core compliance qualifications, emergency procurement identifiers) will cause commercial losses or compliance risks if leaked, and requires the highest level of encryption; general sensitive data (quantity of procurement items, delivery cycle) will only affect efficiency if leaked, and is anonymized using moderately reversible methods; non-sensitive data (project name, common technical parameters) poses no risk when disclosed, and only requires basic hashing. After classification, the requirement data is tagged with sensitive information, and its source and update time are recorded to provide a basis for subsequent dynamic adjustments.

[0070] Next, to address the challenge of traditional fixed algorithms struggling to adapt to data changes, we first extract the dimensional features of the required data (data type, update frequency, association strength), pre-set a basic encryption algorithm such as the national standard SM4, and trigger an update to the encryption algorithm based on these dimensional features. For example, we update parameters such as the encryption factor, number of encryption rounds, and key length. We then extract the key parameters of the updated algorithm (key length, number of rounds, etc.), splitting them into parameter type-parameter value character segments. Using the algorithm type as the root node, parameter type as the left child node, and parameter value as the right child node, we construct the first binary tree.

[0071] Then, feature values ​​are extracted from core sensitive data, general sensitive data, and non-sensitive data: core sensitive data uses the median budget and the number of qualification compliance items; general sensitive data uses the purchase quantity and delivery cycle days; and non-sensitive data uses the hash value of the project name keywords. A second binary tree is constructed with the sensitivity level as the root node (divided into core, general, and non-sensitive), the data dimension as the left child node (e.g., budget and qualification under core sensitive data), and the feature value as the right child node.

[0072] Next, the first and second binary trees are aligned and superimposed according to the root node-left child-right child node, forming a one-to-one association between algorithm parameters and data features. Then, the filtering rules are determined: analyzing the character characteristics of the root nodes of the two trees, such as the first binary tree containing both English and numeric characters, and the second binary tree containing only numeric characters, the filtering rule is to select nodes containing numeric characters in both binary trees as target nodes. The trees are traversed according to the rules to filter out target nodes, eliminating nodes that are not related or have insufficient matching.

[0073] Finally, the character fragments of the target node are concatenated sequentially, or a fixed-length factor can be generated using SHA-256 hashing, combining algorithmic and data feature correlation. Differential desensitization is then performed: core sensitive data is encrypted using a desensitization factor + SM4 irreversible encryption; generally sensitive data is encrypted using a desensitization factor + AES reversible permutation; and non-sensitive data is encrypted using a desensitization factor + basic hash. Desensitized data is partitioned and stored in an encrypted database according to sensitivity level. A binary tree and factor association relationship are encrypted and backed up off-site. Access requires verification of permissions and factor matching; only authorized users can access data according to their sensitivity level.

[0074] In one embodiment, the method further includes: Obtain the multi-dimensional features of the required data, map the multi-dimensional features into character information, and add the character information sequentially to a preset array to construct a first array; Sensitive data is extracted from the required data and added to each node in a preset undirected graph to construct a first undirected graph; Obtain a preset desensitized character array; wherein the desensitized character array includes multiple different characters; Based on the attributes of the first array and the desensitized character array, character replacement rules are determined. Based on the character replacement rules, characters in the desensitized character array are replaced based on the characters in the first array to obtain a replacement desensitized array. Based on the replacement desensitization array, the characters on each node of the first undirected graph are desensitized to obtain a desensitized undirected graph; The characters of each node in the de-identified undirected graph are combined sequentially to obtain de-identified data for storage.

[0075] In this embodiment, firstly, the multi-dimensional features of the demand data are fully extracted, covering data type (e.g., budget is numerical, qualification requirements are text), update frequency (e.g., urgent procurement needs are updated hourly, project names are fixed), correlation strength (e.g., strong correlation between procurement quantity and transportation method, weak correlation between supplier address and delivery cycle), data length (e.g., 18-digit ID number, 6-digit budget code), etc.

[0076] Next, feature-character mapping rules are established to convert each dimension feature into a specific character. For example, "N" represents numeric type, "T" represents text type, "S" represents strong correlation, and "W" represents weak correlation. Numbers such as "5" and "10" represent different data length ranges. Subsequently, the mapped characters are sequentially filled into an array of preset length to form the first array.

[0077] Next, sensitive data is precisely identified and extracted from the demand data, including core sensitive data (such as budget ranges, core technical parameters, and confidentiality requirements) and general sensitive data (such as the quantity of procurement items, delivery cycles, and supplier selection criteria), ensuring that no information that may involve privacy or trade secrets is overlooked. Each piece of sensitive data is treated as an independent node and added to a pre-defined undirected graph framework. The nodes are named using the key identifiers of the original sensitive data, and the specific sensitive values ​​are stored in the node attributes. The resulting first undirected graph clearly displays all the sensitive data.

[0078] The preset desensitized character array is a pre-configured set of characters containing various character types, typically including uppercase letters (AZ), lowercase letters (az), numbers (0-9), and special symbols (such as #, $, %). The number of characters is consistent with the first array; for example, if the first array has 10 characters, the desensitized character array will also have 10 characters. The characters in the array are randomly generated and updated periodically (e.g., automatically refreshed every 24 hours) to prevent the desensitization pattern from being broken due to fixed character patterns. Furthermore, the desensitized character array is specifically configured based on the characteristics of the required data scenario. For example, when processing financial procurement data, the proportion of numeric characters in the array is increased; when processing text-based qualification data, the proportion of alphabetic characters is increased.

[0079] During the acquisition process, the validity of the de-identified character array is automatically verified, including character uniqueness (to avoid too many duplicate characters affecting the randomness of de-identification) and type diversity (to ensure it can match multiple features of the first array). If the verification fails, it is regenerated. The existence of this array makes the de-identification operation independent of the characters in the original sensitive data, reducing the risk of information leakage from the source.

[0080] Next, by establishing association rules between the first array and the desensitized character array, feature-driven character replacement is achieved, providing a precise replacement basis for subsequent node desensitization. First, the core attributes of the two arrays are analyzed: the attribute of the first array is the feature characters (such as "N", "H", "S"), reflecting the essential characteristics of the required data; the attributes of the desensitized character array are the character type (such as numbers, uppercase letters, special symbols) and position index (such as the 1st position, the 5th position). Based on these attributes, character replacement rules are automatically generated: if a character at a certain position in the first array is N (numerical), the corresponding character in the desensitized array is replaced with a random number; if it is T (text), it is replaced with a random letter; if it is H (high-frequency update), it is replaced with an uppercase character; if it is S (strong association), it is replaced with a combination of characters containing special symbols. After the rules are generated, they need to undergo logical verification to ensure that the same feature corresponds to a unique replacement type, avoiding conflicts.

[0081] Subsequently, the desensitized character array is replaced character by character according to the rules to generate a replacement desensitized array. For example, if the 5th character of the first array is N (numerical), and the original 5th character of the desensitized array was B, then it is replaced with 7. The replaced array retains the randomness of the desensitized characters, and the feature association ensures that the replacement logic matches the characteristics of the required data, thus providing a guarantee for accurate desensitization.

[0082] Then, the character replacement of sensitive nodes is achieved by replacing the desensitized array, while preserving the complete association structure of the undirected graph. First, the correspondence between the replacement desensitized array and the nodes of the first undirected graph is established: according to the sensitive data type of the node, the characters matching the features in the array are assigned to the corresponding nodes. For example, the budget upper limit is a numerical sensitive data, corresponding to the numeric characters after feature N replacement in the array.

[0083] Next, the node characters in the first undirected graph are desensitized and replaced according to the corresponding relationships: for numerical sensitive data, the specific values ​​are replaced with numeric characters in the array; for text sensitive data, the key text is replaced with alphanumeric characters; for composite sensitive data containing multiple types, the characters are replaced according to the characteristics of different parts. The final desensitized undirected graph hides the true values ​​of sensitive data while fully preserving the business logic structure of the procurement requirements.

[0084] The de-identified characters are extracted from each node of the de-identified undirected graph in sequence and concatenated to form a complete de-identified data string. After concatenation, the de-identified data undergoes integrity verification to ensure that all nodes are included and there are no duplicates. Once verification passes, the de-identified data is stored in a designated partition of an encrypted database (core sensitive data is stored separately with encrypted access permissions). Simultaneously, the topology information (node ​​relationships) of the de-identified undirected graph and replacement rule logs are stored to facilitate subsequent data retrieval. This ensures two things: first, it allows for the restoration of the business logic related to procurement requirements, supporting normal procurement matching analysis; second, it allows for traceability of the de-identification process, ensuring compliance. The storage process employs an off-site backup strategy to further enhance data security.

[0085] In the above embodiments, this application incorporates some existing algorithms and technical features for explanation and description to make the specification more detailed, clear, and complete, thus complying with the provisions of the Patent Law. However, this is not achieved by using a series of complex steps and algorithmic formulas, nor by complicating the technical solution, nor by combining or stacking conventional or simple features. The existing algorithms and technical features listed are for the purpose of disclosing the specific implementation methods of each step of this application (not to limit this application) and to avoid situations where this application cannot be implemented.

[0086] Reference Figure 2 In another embodiment of the present invention, an artificial intelligence-based procurement project matching system is also provided, comprising: The acquisition module is used to acquire multiple demand data from the purchaser and quantify them into vector feature data. The generation module is used to generate weight values ​​for each vector feature data in real time. The calculation module is used to calculate the supply and demand matching degree between the purchaser and the supplier based on the vector feature data and its corresponding weight values, using the Jaccard coefficient algorithm that incorporates regional attenuation factors. The output module is used to output the matching results of procurement projects based on the supply and demand matching degree.

[0087] The present invention also provides an artificial intelligence-based procurement project matching device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0088] In this embodiment, the specific implementation of each module in the above system embodiment is described in the above method embodiment, and will not be repeated here.

[0089] Reference Figure 3 This invention also provides an artificial intelligence-based procurement project matching device, which can be a server, and its internal structure can be as follows: Figure 3 As shown, the AI-based procurement project matching device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor, designed as a computer, provides computing and control capabilities. The memory of the AI-based procurement project matching device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the AI-based procurement project matching device stores the data corresponding to this embodiment. The network interface of the AI-based procurement project matching device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0090] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the AI-based procurement project matching device to which the present invention is applied.

[0091] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0092] In summary, the AI-based procurement project matching method, apparatus, and system provided in this embodiment of the invention include: acquiring multiple demand data from the purchaser and quantifying them into vector feature data; generating weight values ​​for each vector feature data in real time; calculating the supply-demand matching degree between the purchaser and the supplier using a Jaccard coefficient algorithm that incorporates a regional attenuation factor based on the vector feature data and their corresponding weight values; and outputting the matching result of the procurement project based on the supply-demand matching degree. In this invention, by acquiring multiple demand data from the purchaser, quantifying them into vector feature data, and generating weight values ​​for each vector feature data in real time, accurate quantification of demand features and dynamic adjustment of weights are achieved. By calculating the supply-demand matching degree between the purchaser and the supplier using a Jaccard coefficient algorithm that incorporates a regional attenuation factor, regional factors are scientifically integrated, solving the problems of low efficiency, insufficient accuracy, poor dynamic adaptability, and lack of quantitative quantification of regional influence in current procurement project matching methods.

[0093] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the present invention and embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0094] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0095] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A procurement project matching method based on artificial intelligence, characterized in that, Includes the following steps: Acquire multiple requirements from the purchasing party and quantify them into vector feature data accordingly; Real-time generation of weight values ​​for each vector feature data; Based on the vector feature data and its corresponding weight values, the Jaccard coefficient algorithm with regional attenuation factor is used to calculate the supply and demand matching degree between the purchaser and the supplier. Based on the supply and demand matching degree, output the matching results of the procurement project.

2. The procurement project matching method based on artificial intelligence according to claim 1, characterized in that, The procurement demand data includes the type of procurement target, compliance and qualification requirements, cost sensitivity coefficient, and supplier data.

3. The procurement project matching method based on artificial intelligence according to claim 2, characterized in that, The supplier data includes supplier parameters, credit rating scores, and regional suitability index.

4. The procurement project matching method based on artificial intelligence according to claim 1, characterized in that, Real-time generation of weight values ​​for each vector feature data, including: Obtain historical transaction data of the industry to which the procurement project belongs, and calculate the initial weight values ​​of each vector feature data; Obtain matching results feedback data for similar procurement projects and dynamically adjust the initial weight values; the feedback data includes the transaction rate and performance satisfaction scores of both the supply and demand sides; Based on the urgency and scale of the current procurement project, a scenario influence factor matrix is ​​established. The dynamically corrected weight values ​​are then adjusted a second time to generate the final real-time weight values ​​for each vector feature data.

5. The procurement project matching method based on artificial intelligence according to claim 1, characterized in that, Based on the supply and demand matching degree, the matching results of the procurement project are output, including: The suppliers are sorted from highest to lowest according to their supply-demand matching scores, and then a second sorting calibration is performed based on the suppliers' real-time response capability parameters to obtain the final sorting result. In-depth feature matching analysis was performed on the top 20% of suppliers in the ranking results to generate a heat map of supply and demand differences, and the difference analysis results with specific differences in key indicators were obtained. Based on the variance analysis results, matching results for the procurement project are generated.

6. The procurement project matching method based on artificial intelligence according to claim 1, characterized in that, Based on the vector feature data and its corresponding weight values, the supply-demand matching degree between the purchaser and the supplier is calculated using the Jaccard coefficient algorithm that incorporates regional attenuation factors, including: Construct feature sets for both the purchaser and the supplier, and perform feature matching; both sets contain vector feature data of the same dimension. Calculate the number of matching features between the buyer and the supplier, and divide it by the total number of all features of both parties to obtain the basic matching coefficient. The geographical attenuation factor is determined based on the straight-line distance between the purchaser and the supplier; the greater the distance, the smaller the geographical attenuation factor. For each successfully matched feature, the weighted matching degree is calculated by summing its corresponding weight value. Multiplying the basic matching coefficient, weighted matching degree, and regional attenuation factor together yields the supply-demand matching degree between the purchaser and the supplier.

7. The procurement project matching method based on artificial intelligence according to claim 1, characterized in that, Obtain multiple requirements from the purchasing party and quantify them into vector feature data, including: Obtain the procurement party's demand data and classify it into structured and unstructured data. The structured data includes the type, quantity, and budget range of the procurement target, while the unstructured data includes procurement demand description text, historical cooperation preference documents, and policy compliance statements. Semantic extraction is performed on unstructured data using a bidirectional Transformer model to extract implicit demand features, including potential substitute targets, cooperation mode preferences, and risk tolerance thresholds. Based on a pre-defined feature quantization rule library, structured data and implicit requirement features are uniformly quantified to obtain quantified feature data. Calculate the correlation between the quantitative feature data and the historical demand features of similar procurement projects, and correct the deviation of the quantitative feature data to obtain vector feature data; For time-sensitive demand parameters in vector feature data, an attenuation coefficient is set, and their proportion in the vector feature data is dynamically adjusted over time.

8. A procurement project matching system based on artificial intelligence, characterized in that, include: The acquisition module is used to acquire multiple demand data from the purchaser and quantify them into vector feature data. The generation module is used to generate weight values ​​for each vector feature data in real time. The calculation module is used to calculate the supply and demand matching degree between the purchaser and the supplier based on the vector feature data and its corresponding weight values, using the Jaccard coefficient algorithm that incorporates regional attenuation factors. The output module is used to output the matching results of procurement projects based on the supply and demand matching degree.

9. A procurement project matching device based on artificial intelligence, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

Images