Commodity searching and sorting method based on purchase data

Abstract

The invention discloses a commodity searching and sorting method based on purchase data, relates to the technical field of commodity searching and sorting, and solves the problems that a traditional sorting method is static and one-sided in value evaluation, is disjointed with a real-time market supply and demand state, and lacks quantitative avoidance of supply chain multi-dimensional risks. According to the method, the supply continuity risk, the quality and performance risk and the price fluctuation risk are brought into the final sorting calibration link, and the right of high-risk commodities is systematically reduced through the risk penalty term, so that the final sorting list pursues the value and opportunity, and meanwhile, suppliers with high commodity shortage risk, high after-sales problems or unstable cost can be automatically avoided, and the efficiency of sorting the commodities is improved. Therefore, the long-term robustness of the purchase decision and the safety of the supply chain are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of product search and ranking technology, specifically a product search and ranking method based on procurement data. Background Technology

[0002] In e-commerce and enterprise procurement platforms, the key technology for improving procurement efficiency and reducing decision-making costs lies in how to intelligently filter, sort, and present products that best meet the needs of the buyer from a massive number of suppliers and products.

[0003] Traditional ranking methods primarily rely on the static attribute matching degree, historical sales volume, or price of products for sorting. These methods have significant limitations: First, they typically only reflect past performance (such as total sales volume) and cannot effectively identify new or emerging products with upward trends and growth potential, potentially missing valuable opportunities. Second, the ranking process is often disconnected from real-time market supply and demand. For example, even if a product has high value, if its current inventory is extremely tight, recommending it to buyers may result in invalid inquiries, reducing procurement efficiency. Furthermore, existing methods lack quantitative consideration of supplier long-term stability and fulfillment risks, potentially prioritizing products with frequent stockouts, quality fluctuations, or drastic price volatility, thus introducing potential risks to procurement decisions. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a product search and ranking method based on procurement data, which solves the problems of traditional ranking methods having static and one-sided value assessment, being out of touch with real-time market supply and demand, and lacking quantitative avoidance of multi-dimensional risks in the supply chain.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a product search and ranking method based on procurement data, comprising the following steps:

[0006] Step 1: Calculate the total popularity score, growth potential score, and fulfillment stability score for each product based on recent sales data;

[0007] Step 2: Based on the query terms and historical successful procurement cases, obtain the corresponding weights of the total popularity value, growth potential value, and performance stability score, calculate the basic comprehensive score of each product, and generate an initial sorted list;

[0008] Step 3: Calculate the competition factor based on the real-time inventory of the products and the short-term demand based on search popularity. For any two adjacent products in the initial ranking list, the product with tighter inventory is penalized according to the competition factor, and a fine-tuned ranking is generated.

[0009] Step 4: Calculate the supply continuity, quality fulfillment, and price fluctuation risks of each commodity to obtain the total risk factor. Combine this with the competition factor to apply a double penalty calibration to the scores of the commodities in the fine-tuned ranking. Output the ranking list in descending order based on the final scores.

[0010] As a further aspect of the present invention: the calculation of the total heat value is specifically as follows:

[0011] Obtain the daily sales sequence of the product for the past N days, assign a weight to the daily sales volume in the sequence that decays over time, and sum the products of the daily sales volume and the corresponding weights to obtain the total popularity H.

[0012] As a further aspect of the present invention: the calculation of the growth potential value is specifically as follows:

[0013] The recent sales series of a product is divided into multiple analysis windows that are consecutive and overlapping in time.

[0014] Calculate the slope β of sales change over time within each window. K ;

[0015] Calculate the trend consistency coefficient C based on the slope of all windows;

[0016] Calculate the recent sales growth rate Gr and the trend increase rate A;

[0017] Preset weights are assigned to the recent sales growth rate Gr, the trend increase rate A, and the trend consistency coefficient C, and then a weighted sum is performed to obtain the growth potential value G of the product.

[0018] As a further aspect of the present invention: the calculation of the performance stability score is specifically as follows:

[0019] Obtain the on-time delivery rate, logistics anomaly rate, and quality refund rate of goods within the statistical period;

[0020] The data of the statistical period is divided into multiple consecutive time windows, and the average value of each indicator is calculated in each window.

[0021] For each indicator, its volatility across the window is calculated to obtain the stability coefficients of on-time performance, anomaly rate, and refund rate.

[0022] Extract the longest consecutive number of days without orders for a product during the observation period, Dj, and calculate the supply continuity penalty coefficient P;

[0023] By combining the stability coefficients of various indicators with the supply continuity penalty coefficient P, the fulfillment stability score Si of the commodity is calculated.

[0024] As a further aspect of the present invention: the method of obtaining the comprehensive matching degree between query terms and products specifically involves:

[0025] Calculate the semantic similarity between the query term and the product title to obtain the semantic matching score ScoreA;

[0026] The relevance between query terms and product categories is calculated based on the product category system to obtain the category matching score ScoreB;

[0027] The attribute matching score (Score C) is obtained by analyzing the key attributes in the query terms and comparing them with the product attributes.

[0028] The overall matching score is calculated using the formula: (ScoreA+ScoreB+ScoreC) / 3.

[0029] As a further aspect of the present invention: the process of obtaining the corresponding weights includes:

[0030] Based on the query terms, relevant successful procurement cases are retrieved from the historical database;

[0031] Obtain the average performance scores μH, μG, and μSi of the total popularity value, growth potential value, and fulfillment stability score of the products in these cases to construct a three-dimensional vector;

[0032] The three-dimensional vector formed by the average performance score is normalized to obtain weights WH, WG, and WSi that are adapted to the current query scenario, and WH+WG+WSi=1.

[0033] As a further aspect of the present invention: the calculation of the competition factor specifically includes:

[0034] Get the real-time available inventory Lj for each product;

[0035] Based on the recent search popularity sequence for each product, the short-term potential demand Uj is calculated using the exponentially weighted moving average method. The specific calculation formula is as follows:

[0036] , where γ is the attenuation factor;

[0037] Calculate the inventory demand tension Tj = Uj / (Lj + ε), where ε is a very small positive number;

[0038] The inventory demand tension Tj is normalized to obtain the competition factor Kj. The larger the Kj value, the more intense the competition.

[0039] As a further aspect of the present invention: the penalty based on the competition factor specifically refers to:

[0040] For product pairs (m, n) with similar basic comprehensive scores, the formula is used:

[0041] Sm=Sbasem×exp(-η×max(0,Km-Kn));

[0042] Sn = Sbasen × exp(-η × max(0, Kn - Km)); Calculate the adjusted scores Sm and Sn, where η is the penalty intensity coefficient, Km and Kn are the competition factors of goods m and n respectively, and Sbasem and Sbasen are the basic comprehensive scores of goods m and n respectively.

[0043] As a further aspect of the present invention: the obtained total risk factor specifically refers to:

[0044] Based on the longest consecutive number of days without orders, Dj, the supply continuity risk value ra is calculated using the formula: ra=Dj / Dcrit;

[0045] Based on the logistics anomaly rate and the quality refund rate, the quality and performance risk value rb is calculated using the formula: rb = (logistics anomaly rate + refund rate) / 2.

[0046] Based on the recent historical purchase price series, the price volatility risk value rc is calculated using the formula: rc=α(Pt) / μ(Pt)+ε, where Pt represents the historical purchase price series of the past N, t=1,2,…N, α(Pt) represents the standard deviation of the series, μ(Pt) represents the arithmetic mean of the series, and ε is a very small positive number.

[0047] The total risk factor Rj is obtained by averaging the supply continuity risk value ra, the quality and performance risk value rb, and the price volatility risk value rc.

[0048] As a further aspect of the present invention: the final score calculation formula is specifically as follows:

[0049] Fj = Sj / [(Kj)] α [×(1+β×Rj)], where Sj is Sm or Sn, α is the competition penalty index, β is the risk penalty coefficient, Kj is its competition factor, and Rj is its total risk factor.

[0050] This invention provides a product search and ranking method based on procurement data. Compared with existing technologies, it has the following advantages:

[0051] (1) This invention not only calculates the historical sales popularity of goods, but also analyzes their growth potential in depth through multi-window trend regression and assesses the stability of their fulfillment supply. This makes the ranking results not only reflect the goods that sold well in the past, but also highlight the goods with future potential and more reliable supply, guiding buyers to focus on the options with better overall value. By analyzing the characteristic distribution of historical successful procurement cases, the weight of different characteristics is dynamically adjusted, so that the ranking strategy can adapt to different procurement intentions and scenarios.

[0052] (2) By integrating real-time inventory with short-term demand based on search popularity, this invention calculates competition factors and performs local sorting adjustments for products with similar value based on inventory tightness. This ensures that when recommending products with similar value, products with relatively sufficient inventory and higher purchase success rate will be displayed first, effectively avoiding user frustration and resource waste caused by recommending seemingly good but unavailable products.

[0053] (3) This invention incorporates supply continuity risk, quality and performance risk, and price fluctuation risk into the final ranking calibration process. Through risk penalty items, high-risk products are systematically downgraded, so that the final ranking list can automatically avoid suppliers with high stockout risk, high after-sales problems, or unstable costs while pursuing value and opportunities, thereby greatly improving the long-term robustness of procurement decisions and supply chain security. Attached Figure Description

[0054] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] Example 1

[0057] Please see Figure 1 This application provides a product search and ranking method based on procurement data, including the following steps:

[0058] Step 1: Calculate the total popularity score, growth potential score, and fulfillment stability score for each product based on recent sales data;

[0059] Step 2: Based on the query terms and historical successful procurement cases, obtain the corresponding weights of the total popularity value, growth potential value, and performance stability score, calculate the basic comprehensive score of each product, and generate an initial sorted list.

[0060] Step 3: Calculate the competition factor based on the real-time inventory of the products and the short-term demand based on search popularity. For any two adjacent products in the initial ranking list, the product with tighter inventory is penalized according to the competition factor, and a fine-tuned ranking is generated.

[0061] Step 4: Calculate the supply continuity, quality fulfillment, and price fluctuation risks of each commodity to obtain the total risk factor. Combine this with the competition factor to apply a double penalty calibration to the scores of the commodities in the fine-tuned ranking. Output the ranking list in descending order based on the final scores.

[0062] Example 2

[0063] This illustrates another embodiment of the present invention. Based on Embodiment 1, the content of Embodiment 1 is described in further detail. This application provides a product search and sorting method based on procurement data, specifically including the following steps:

[0064] Step 1: For the same type of target product, obtain the sales quantity sequence Q[q1, q2, ..., qn] for each product in the most recent N days, where qn is the sales volume before day n;

[0065] Using the formula Wi=α （i-1） Assign a weight [W1, W2, ..., Wn] to the daily sales volume in the sales volume sequence Q, where i is from 1 to n, α is the decay coefficient, and satisfy W1>W2D>...Wn and ∑Wi=1;

[0066] The total popularity H of a product is obtained by summing the products of daily sales volume and corresponding weights in the sales volume data series.

[0067] Secondly, with N days as the total number of days, the recent sales sequence of each product is divided into multiple overlapping analysis windows. Each window covers sales data for L consecutive days, with an offset of one day between adjacent windows, thus forming a series of analysis units that are continuously interleaved in time, ultimately obtaining K windows. The first window contains sales data for L days starting from the most recent day, the second window starts from the second most recent day, and so on, until the entire observation period is covered.

[0068] For each window W K Calculate the slope β of each window. K The specific formula is as follows:

[0069] β K =∑[(ti-tavg)(yi-yavg)] / ∑(ti-tavg)², where =tavg is the arithmetic mean of all time points ti within this window, and yavg is the arithmetic mean of all sales values ​​yi within the current window;

[0070] For the slope β of all windows K The consistency coefficient C is calculated using the following formula:

[0071] C=|μβ| / (|μβ+|αβ), where μβ is all β K The mean of αβ is the standard deviation;

[0072] Simultaneously calculate the recent sales growth rate Gr and the trend increase rate A, using the following formula:

[0073] Gr = (Total sales in the most recent L days / Total sales in the previous L days) - 1;

[0074] A = (β of the nearest window) K -β of the farthest window K (Nearest time window point - farthest time window point);

[0075] Finally, Gr, A, and C are substituted into the weighted combination of indicators to obtain the potential value G of the product. The specific formula is: G = λ1 × Gr + λ2 × A + λ3 × C.

[0076] Three key metrics were extracted from the data of each product over the past N days: On-time delivery rate = On-time order quantity / Total order quantity; Logistics anomaly rate = Number of abnormal logistics orders / Total orders; Refund rate = Number of quality refund orders / Total orders.

[0077] The data from M days is divided into W consecutive windows according to a fixed period. Each window contains data from several days. For each indicator, the average value within each window is calculated to obtain the average values ​​of on-time delivery rate, logistics anomaly rate, and quality refund rate for the Wth window.

[0078] For each key indicator, calculate its fluctuation degree between different windows, and use the formula: mean standard deviation of a single indicator across all windows / mean value of a single indicator across all windows to calculate the stability coefficients of on-time rate, anomaly rate, and refund rate.

[0079] Extract the longest consecutive number of days without orders for this product within N days, Dj, and obtain the penalty coefficient P, specifically:

[0080] P = e^(-Z×Dj), where Z is an adjustment parameter (e.g., Z = 0.1-0.5).

[0081] The final stability score Si is calculated based on the stability coefficient and the penalty coefficient, as follows:

[0082] Si = (On-time rate stability coefficient × Abnormality rate stability coefficient × Refund rate stability coefficient)^(1 / 3) × P × 100.

[0083] Step Two:

[0084] The system obtains the query terms from purchasing users. Based on these terms, it uses a semantic analysis model to vectorize the query terms and product titles, calculates the cosine similarity between the two in the semantic space, and obtains the semantic matching score ScoreA. Simultaneously, based on a pre-defined product category system, it calculates the degree of association between the query terms and their respective category C, and obtains the category matching score ScoreB. Furthermore, it analyzes the key attribute requirements in the query terms and compares them with the detailed attribute set of the product, and calculates the attribute matching score ScoreC.

[0085] The final overall matching score is calculated as M = (ScoreA + ScoreB + ScoreC) / 3;

[0086] Based on the query terms, relevant successful procurement cases are retrieved from the historical database. The performance distribution of products in these cases on three features—total popularity, potential value, and stability score—is obtained. The average score of each feature in the relevant cases is calculated and denoted as μH, μG, and μSi, respectively, forming a three-dimensional vector. This three-dimensional vector is normalized to obtain the objective weights WH, WG, and WSi that best match the current query scenario, satisfying WH+WG+WSi=1.

[0087] For the total popularity H, potential value G, and stability score Si of all products, minimum-maximum normalization is applied to linearly transform each feature value to the interval [0, 1] to obtain standardized feature values ​​H', G', and Si', so as to ensure that different features are comparable;

[0088] Finally, calculate the comprehensive ranking score Sbase for each product, specifically: Sbase = M × F, where F is the feature score, F = WH × H' + WG × G' + WSi × Si'. Sort all products in descending order of Sbase and obtain the top J products to form the initial ideal value ranking list.

[0089] Step 3:

[0090] For the first J items in the initial ideal value ranking list, query in real time the total amount of immediately sellable inventory in all available warehouses, denoted as inventory quantity Lj, where j=1, 2, ...j;

[0091] Based on the search popularity sequence Vj,t of related queries for this product on the platform in the recent period (e.g., the past M days) (t=1, 2, ..., M, t=1 being the most recent day), the potential demand Uj in the short term (the next operating cycle T days) is calculated using the exponentially weighted moving average method. The specific formula is as follows:

[0092] , where γ is the attenuation factor (e.g., γ=0.9);

[0093] Based on the real-time inventory Lj and the potential demand Uj, calculate the inventory demand tension of commodity j Tj=Uj / (Lj+ε), where ε is a very small positive number (e.g. ε=0.001). The higher the tension Tj, the greater the demand pressure faced by each unit of inventory.

[0094] Furthermore, the tension is transformed into a competitive factor Kj that can be used as a penalty term, and then normalized:

[0095] Kj = 1 + (Tj - Tmin) / (Tmax - Tmin) makes Kj fall within the interval [1, 2], and the larger the value, the more intense the competition.

[0096] Iterate through the list of items sorted in descending order of Sbase. For any two adjacent items m and n (Sbasem ≥ Sbasen), calculate the relative difference in their scores:

[0097] δ = (Sbasem - Sbasen) / Sbasem. If δ < θ, then the comprehensive scores of products m and n are considered to be similar, where θ is a preset threshold.

[0098] For each pair of similar goods (m, n), based on the competition factors Km and Kn of goods m and n, if Km > Kn, then a penalty adjustment is applied to goods m; if Kn > Km, then a penalty adjustment is applied to goods n. The specific penalty adjustment is as follows:

[0099] Sm=Sbasem×exp(-η×max(0,Km-Kn));

[0100] Sn=Sbasen×exp(-η×max(0,Kn-Km));

[0101] Where η is the penalty intensity coefficient (e.g., η=0.1), the function max(0, ΔK) ensures that only the party with the most intense competition is penalized, the exponential term exp(-η×ΔK) constitutes the penalty factor, and ΔK is Km-Kn or Kn-Km;

[0102] Finally, based on the adjustment scores of all products (products that are not adjusted retain their original values), they are reordered in descending order to obtain a product ranking list that has been fine-tuned by competition perception. This list, while respecting the original comprehensive value ranking, prioritizes products with relatively ample inventory within a local range of similar values.

[0103] Step Four:

[0104] Based on the longest consecutive number of days without orders (Dj) extracted in step one, the supplier's continuous risk value (ra) is calculated using the following formula:

[0105] ra = Dj / Dcrit, where Dcrit is a preset critical day threshold (e.g., Dcrit = 10). When Dj ≥ Dcrit, the supplier's continuous risk is considered to be at its maximum (value is 1.0).

[0106] Based on the logistics anomaly rate and refund rate from step one, calculate the quality and fulfillment risk value rb using the following formula:

[0107] rb = (Logistics anomaly rate + Refund rate) / 2;

[0108] Obtain the historical purchase price sequence Pt of the commodity over the past N days, and calculate its price volatility risk value rc. The specific formula is as follows:

[0109] rc = α(Pt) / μ(Pt) + ε, where Pt represents the historical purchase unit price sequence of the last N, t = 1, 2, ... N, α(Pt) represents the standard deviation of the sequence, μ(Pt) represents the arithmetic mean of the sequence, and ε is a very small positive number (e.g., ε = 0.01).

[0110] The total risk factor Rj is calculated based on the supplier continuous risk value ra, the quality and performance risk value rb, and the price fluctuation risk value rc. The specific formula is: Rj(ra+rb+rc)×(1 / 3).

[0111] Based on the fine-tuned product scores Sj (Sm or Sn) and Kj obtained in step three, and the risk factor Rj, the final score Fj for each product is calculated using the following formula:

[0112] Fj = Sj / [(Kj)] α [×(1+β×Rj)], where (Kj) α This is the penalty term for the competition factor, where α is the competition penalty index (e.g., α=0.05) and β is the risk penalty coefficient (e.g., β=0.3). The larger Rj is, the larger this term is, and the stronger its weakening effect on the score.

[0113] Finally, obtain the final score Fj for each product in the ideal value ranking list, sort all products in descending order based on the final ranking score Fj, and output this final product ranking list as the response result of this procurement search.

[0114] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0115] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A product search and ranking method based on procurement data, characterized in that, Includes the following steps: Step 1: Calculate the total popularity score, growth potential score, and fulfillment stability score for each product based on recent sales data; Step 2: Based on the query terms and historical successful procurement cases, obtain the corresponding weights of the total popularity value, growth potential value, and performance stability score, calculate the basic comprehensive score of each product, and generate an initial sorted list; Step 3: Calculate the competition factor based on the real-time inventory of the products and the short-term demand based on search popularity. For any two adjacent products in the initial ranking list, the product with tighter inventory is penalized according to the competition factor, and a fine-tuned ranking is generated. Step 4: Calculate the supply continuity, quality fulfillment, and price fluctuation risks of each commodity to obtain the total risk factor. Combine this with the competition factor to apply a double penalty calibration to the scores of the commodities in the fine-tuned ranking. Output the ranking list in descending order based on the final scores.

2. The product search and sorting method based on procurement data according to claim 1, characterized in that, The calculation of the total heat value is as follows: Obtain the daily sales sequence of the product for the past N days, assign a weight to the daily sales volume in the sequence that decays over time, and sum the products of the daily sales volume and the corresponding weights to obtain the total popularity H.

3. The product search and sorting method based on procurement data according to claim 1, characterized in that, The growth potential value is calculated as follows: The recent sales series of a product is divided into multiple analysis windows that are consecutive and overlapping in time. Calculate the slope β of sales change over time within each window. K ; Calculate the trend consistency coefficient C based on the slope of all windows; Calculate the recent sales growth rate Gr and the trend increase rate A; Preset weights are assigned to the recent sales growth rate Gr, the trend increase rate A, and the trend consistency coefficient C, and then a weighted sum is performed to obtain the growth potential value G of the product.

4. The product search and sorting method based on procurement data according to claim 1, characterized in that, The performance stability score is calculated as follows: Obtain the on-time delivery rate, logistics anomaly rate, and quality refund rate of goods within the statistical period; The data of the statistical period is divided into multiple consecutive time windows, and the average value of each indicator is calculated in each window. For each indicator, its volatility across the window is calculated to obtain the stability coefficients of on-time performance, anomaly rate, and refund rate. Extract the longest consecutive number of days without orders for a product during the observation period, Dj, and calculate the supply continuity penalty coefficient P; By combining the stability coefficients of various indicators with the supply continuity penalty coefficient P, the fulfillment stability score Si of the commodity is calculated.

5. The product search and sorting method based on procurement data according to claim 1, characterized in that, The overall matching degree between the query terms and the products is specifically as follows: Calculate the semantic similarity between the query term and the product title to obtain the semantic matching score ScoreA; The relevance between query terms and product categories is calculated based on the product category system to obtain the category matching score ScoreB; The attribute matching score (Score C) is obtained by analyzing the key attributes in the query terms and comparing them with the product attributes. The overall matching score is calculated using the formula: (ScoreA+ScoreB+ScoreC) / 3.

6. The product search and sorting method based on procurement data according to claim 1, characterized in that, The process of obtaining the corresponding weights includes: Based on the query terms, relevant successful procurement cases are retrieved from the historical database; Obtain the average performance scores μH, μG, and μSi of the total popularity value, growth potential value, and fulfillment stability score of the products in these cases to construct a three-dimensional vector; The three-dimensional vector formed by the average performance score is normalized to obtain weights WH, WG, and WSi that are adapted to the current query scenario, and WH+WG+WSi=1.

7. The product search and sorting method based on procurement data according to claim 1, characterized in that, The calculation of the competition factor specifically includes: Get the real-time available inventory Lj for each product; Based on the recent search popularity sequence for each product, the short-term potential demand Uj is calculated using the exponentially weighted moving average method. The specific calculation formula is as follows: , where γ is the attenuation factor; Calculate the inventory demand tension Tj = Uj / (Lj + ε), where ε is a very small positive number; The inventory demand tension Tj is normalized to obtain the competition factor Kj. The larger the Kj value, the more intense the competition.

8. The product search and sorting method based on procurement data according to claim 1, characterized in that, The specific method for penalizing them based on competition factors is as follows: For product pairs (m, n) with similar basic comprehensive scores, the formula is used: Sm=Sbasem×exp(-η×max(0,Km-Kn)); Sn = Sbasen × exp(-η × max(0, Kn - Km)); Calculate the adjusted scores Sm and Sn, where η is the penalty intensity coefficient, Km and Kn are the competition factors of goods m and n respectively, and Sbasem and Sbasen are the basic comprehensive scores of goods m and n respectively.

9. The product search and sorting method based on procurement data according to claim 1, characterized in that, The obtained total risk factor is specifically as follows: Based on the longest consecutive number of days without orders, Dj, the supply continuity risk value ra is calculated using the formula: ra=Dj / Dcrit; Based on the logistics anomaly rate and the quality refund rate, the quality and performance risk value rb is calculated using the formula: rb = (logistics anomaly rate + refund rate) / 2. Based on the recent historical purchase price series, the price volatility risk value rc is calculated using the formula: rc=α(Pt) / μ(Pt)+ε, where Pt represents the historical purchase price series of the past N, t=1,2,…N, α(Pt) represents the standard deviation of the series, μ(Pt) represents the arithmetic mean of the series, and ε is a very small positive number. The total risk factor Rj is obtained by averaging the supply continuity risk value ra, the quality and performance risk value rb, and the price volatility risk value rc.

10. The product search and ranking method based on procurement data according to claim 1, characterized in that, The specific formula for calculating the final score is as follows: Fj = Sj / [(Kj)] α [×(1+β×Rj)], where Sj is Sm or Sn, α is the competition penalty index, β is the risk penalty coefficient, Kj is its competition factor, and Rj is its total risk factor.

Images