CN-121998698-A - Intelligent recommendation method and recommendation system for suppliers facing purchasing of automobile parts

CN121998698ACN 121998698 ACN121998698 ACN 121998698ACN-121998698-A

Abstract

The application belongs to the technical field of provider recommendation, and particularly relates to an intelligent provider recommendation method and system for purchasing automobile parts. The recommendation method comprises the steps of constructing a knowledge graph formed by nodes and edges according to historical orders of a purchasing company, determining accessory model nodes or adaptive vehicle model nodes as demand nodes according to purchasing demands, screening a plurality of qualified provider nodes of the demand nodes in the knowledge graph based on a Monte Carlo Dropout mechanism by a graph neural network model, screening a plurality of potential provider nodes from all the qualified provider nodes based on a self-adaptive multi-objective optimization function, calculating edge note meaning scores of the potential provider nodes, and outputting the top k providers with the edge note meaning scores in descending order as recommendation degree descending order. The method can combine purchasing requirements and preferences on the basis of integrating multi-source heterogeneous data to carry out scientific and reasonable supplier recommendation.

Inventors

LIN CHUANWEN
WANG DINGMIN
LIU RUI
BI CHENXI
CHEN FANGFANG
CUI HAIYING
LU SHENGGAN
XING JING

Assignees

合肥中安数据科技有限公司

Dates

Publication Date: 20260508
Application Date: 20251218

Claims (10)

1. The intelligent recommendation method for the suppliers facing the purchasing of the automobile parts is characterized by comprising the following steps of: S1, constructing a knowledge graph G composed of nodes and edges according to historical orders of a buyer, wherein the nodes comprise supplier nodes, accessory model nodes, adaptation vehicle model nodes and buyer nodes; S2, determining a fitting model node or an adaptive vehicle model node as a demand node according to purchasing demands, and screening a plurality of qualified provider nodes of the demand node in a knowledge graph G on the basis of a Monte Carlo Dropout mechanism by a graph neural network model GNN; S3, screening a plurality of potential provider nodes from all qualified provider nodes based on the self-adaptive multi-objective optimization function; And S4, calculating side note meaning scores of all potential provider nodes, and outputting providers corresponding to the first k potential provider nodes in descending order of the side note meaning scores as a recommendation degree descending order.
2. The intelligent recommendation method for suppliers for purchasing automobile parts according to claim 1, wherein in S1, the method further comprises the following steps: S11, tracking historical orders by a buyer to acquire related data of the automobile parts, wherein the related data comprise order numbers, part models, adaptive vehicle types, suppliers, buyers, part prices, lead time and supplier reputation; S12, using the model of the accessory, the adaptive model, the supplier and the purchasing company as nodes, and connecting two nodes appearing in the same historical order by edges to form a basic map BS, wherein each edge comprises a relationship type and corresponding relationship content of the nodes at two ends in the same historical order; s13, carrying out hierarchical aggregation node embedding on the basic map BS based on the graph neural network model GNN to obtain a knowledge map G, and carrying out hierarchical aggregation node embedding by using the following formula: ; Wherein, the Indicating that node v is on the first The embedded vector of the layer is used to determine, =1,...; Representing a nonlinear activation function; Representing the embedded vector of the neighbor node u at the first layer; Is a relation weight matrix; representing a set of neighbor nodes adjacent to node v under relationship type r; Representing an aggregation function under a relationship type r; all relation types in the basic map BS are represented, wherein the relation types comprise price relation type, adaptation relation type, reputation relation type and lead time relation type, and each node is obtained after the embedded vector of the first layer is initialized and assigned for the graph neural network model GNN according to node information.
3. The intelligent recommendation method for suppliers for purchasing automobile parts according to claim 1, wherein in S2, the method further comprises the following steps: S21, enabling the Monte Carlo Dropout value of the graph neural network model GNN to be located in a set opening value range; s22, determining a fitting model node or an adaptive vehicle model node as a demand node according to the purchasing demand, and performing forward propagation for N times in a knowledge graph G on the basis of the demand node by a graph neural network model GNN to obtain N preliminary qualified provider nodes; S23, using the mean vector of the N preliminary qualified provider nodes Sum of variances As a filtering standard, performing confidence filtering on the N preliminary qualified provider nodes, and then removing the weight to finally obtain M qualified provider nodes serving as demand nodes. ; ; Wherein, the Indicating that the z-th preliminary qualified provider node is in the knowledge graph G Embedding vectors of the layers; Representing the square of the modulus length.
4. The intelligent recommendation method for suppliers for purchasing automobile parts according to claim 3, wherein the training pattern neural network model GNN obtains preliminary qualified supplier nodes through forward propagation, and further comprises the following contents: Step 1a, a batch of historical orders are obtained, and a training knowledge graph PG is constructed; Step 2a, using one of a fitting model, an adaptive vehicle model and a supplier node as a demand node, marking the pairing relation of the fitting model, the adaptive vehicle model and the supplier which contain the demand node and are actually existing in the same historical order as a positive sample, and forming sub-graph training data centering on the current demand node after marking the relations among other nodes and nodes as negative samples; And 3a, sending the subgraph training data taking a certain demand node as a center into a graph neural network model GNN to be trained, and learning potential relations and differences between positive and negative samples in the batch training data of the demand node after the graph neural network model GNN normalizes and vectorizes the relations between each node and each node in the batch training data.
5. The intelligent recommendation method for suppliers to purchasing automobile parts according to claim 4, further comprising the following after step 3 a: Step 4a, after a knowledge graph VG for verification is constructed, a plurality of demand nodes are selected from the knowledge graph VG, the demand nodes are taken as samples of a current verification batch, and accessory models, adaptation vehicle types and supplier pairing relations which are actually existing in the demand nodes are taken as real labels of corresponding samples of the current verification batch; Step 5a, verifying and iteratively optimizing the training effect of the graph neural network model GNN by using the verification knowledge graph VG, the current verification batch sample and the verification batch real label, wherein after calculating the joint loss function L of the current verification batch, the model parameters of the graph neural network model GNN are iteratively updated towards the gradient decreasing direction of the joint loss function of the current verification batch: ; Wherein, the And The local loss weight and the global loss weight of the training data of the current verification batch b are represented respectively, ; And Representing the local and global loss functions of the current validation lot, respectively.
6. The intelligent recommendation method for suppliers to purchasing automobile parts according to claim 5, wherein in step 5a, the method further comprises the following steps: The graph neural network model GNN outputs E prediction results according to A samples of the current verification batch and the verification knowledge graph VG, wherein each prediction result comprises a group of accessory model, an adaptive vehicle model and a supplier pairing relation; The expressions for the local and global penalty functions for the current validation lot are as follows: ; ; Wherein, the An embedded vector representing the model node of the fitting in the x-th prediction result, Representing an embedded vector of an adaptive vehicle model node in the x-th prediction result; The authenticity value of the xth predicted result is represented, if the xth predicted result exists in the real label of the current verification batch, the xth predicted result is represented by the authenticity value of the xth predicted result Otherwise ; Representation of A function; representing the real neighbor node distribution of the g-th sample in the current verification batch; representing the predicted neighbor node distribution of the g-th sample in the current verification batch; Representation of Relative to KL divergence of (c).
7. A method for intelligent recommendation of suppliers for purchasing automobile parts according to claim 3, wherein in S3, the following are included: s31, constructing a self-adaptive multi-objective optimization function F based on a plurality of sub-objective functions of the provider node: ; Wherein, the Representing a total cost sub-objective function corresponding to the fact that a certain supplier meets purchasing requirements; Representing a delay risk sub-objective function corresponding to the fact that a certain provider meets purchasing requirements; representing a reputation sub-objective function corresponding to a certain provider meeting purchasing requirements; representing a probability that a fitting provided by a certain supplier meeting purchasing requirements meets industry standards; Is the standard deviation of the two-dimensional image, = ; Representing business constraints; 、、、 The first, second, third and fourth preference weights respectively representing the current recommended round t, And is also provided with 、、、 The recommendation round refers to the period from the time of acquiring the current purchasing demand to the time of outputting the corresponding recommended supplier; s32, acquiring a pareto optimal solution set when the self-adaptive multi-objective optimization function F is maximum in M qualified provider nodes; S33, calculating self-adaptive multi-objective optimization function values corresponding to all pareto solutions in the pareto optimal solution set, and taking provider nodes corresponding to the first d pareto solutions in descending order of the self-adaptive multi-objective optimization function values as potential provider nodes.
8. The intelligent recommendation method for suppliers oriented to purchasing automobile parts according to claim 5 or 7, wherein: updating the preference weights for the next recommendation round also includes: Step 1b, the purchasing personnel determines the first, second, third and fourth preference configurations of the next recommended round (t+1) 、、、 Wherein, the method comprises the steps of, 、、、 Are all positive numbers ; Step 2b, calculating the preference weight of the next recommendation round according to the preference weight of the current recommendation round and the preference configuration of the next recommendation round: ; ; ; ; Wherein, the A second coefficient of fusion is represented by the first coefficient, ; Updating the global penalty weight for the next validation lot further includes: Step 1c, the technician determines the global loss preference configuration for the next verification lot (b+1) Wherein, the method comprises the steps of, ; Step 2c, calculating the global loss weight of the next verification batch according to the global loss weight of the current verification batch and the global loss preference configuration of the next verification batch: ; Wherein, the A first coefficient of fusion is represented and, 。
9. The intelligent recommendation method for suppliers for purchasing automobile parts according to claim 2, wherein in S4, the method further comprises the following steps: ; ; Wherein, the Representing a buyer node B and a c-th potential provider node Side note force scores in between; Representing node importance weight coefficients; Representing from the buyer node B to the c-th potential provider node Is a side note force weight; Representation of A function; representing a multi-layer perceptron function; representing the embedded vector of the buyer node B; Representing potential provider nodes Embedding vectors of the nodes; representing a buyer node B and a c-th potential provider node The relationship type of the connecting edges; representing a buyer node B and a c-th potential provider node A code vector of a relationship type of the connecting edges; representing a buyer node B and a c-th potential provider node The total edge number contained in all paths between the two paths; the path attenuation coefficient is represented as such, 。
10. The intelligent supplier recommendation system for the purchasing of the automobile parts is characterized by comprising a knowledge graph construction module, a first screening module, a second screening module and a third screening module, wherein the knowledge graph construction module is used for constructing a knowledge graph according to historical orders of the purchasing agents and then sending the knowledge graph to the first screening module, the first screening module is used for screening out a plurality of qualified supplier nodes according to purchasing requirements and then sending the qualified supplier nodes to the second screening module, the second screening module is used for screening out a plurality of potential supplier nodes from all the qualified supplier nodes based on a self-adaptive multi-objective optimization function and then sending the qualified supplier nodes to the third screening module, the third screening module is used for calculating side meaning scores of the potential supplier nodes, and outputting the previous k potential supplier nodes in descending order according to the side meaning scores as suppliers arranged in descending order of recommendation degrees, and each module is programmed or configured to execute the steps of the intelligent supplier recommendation method for the purchasing of the automobile parts.

Description

Intelligent recommendation method and recommendation system for suppliers facing purchasing of automobile parts Technical Field The application belongs to the technical field of provider recommendation, and particularly relates to an intelligent provider recommendation method and system for purchasing automobile parts. Background Along with the rapid popularization of automobiles, the purchasing demands of the whole automobile manufacturing enterprises and the maintenance service market for accessories are continuously increased. In the traditional purchasing mode, the purchasing party relies on manual experience, a fixed supplier directory, or a single price comparison to select suppliers. Not only is this time and effort consuming, but the personal experience and preferences of the purchasing party substantially dominate the selection results, which are highly subjective, and the recommended results are biased to a single dimension, so that the selected provider may have an irreparable short deck in some way. Moreover, the data of the suppliers are scattered in a plurality of sources such as an order system, a quality inspection system and a financial system, and the data of each aspect of the same supplier has the information island condition, so that the purchasing party is difficult to establish effective connection among multidimensional data, and the comprehensive capacity of the suppliers cannot be comprehensively estimated. Therefore, how to combine purchasing requirements and preferences to make scientific and reasonable supplier recommendation based on the integration of multi-source heterogeneous data becomes a challenge to be solved in the field. Disclosure of Invention The application aims to overcome the defects of the prior art, and provides an intelligent supplier recommendation method for purchasing automobile parts, which can combine purchasing requirements and preferences to carry out scientific and reasonable supplier recommendation on the basis of integrating multi-source heterogeneous data. In order to achieve the above purpose, the present application adopts the following technical scheme: an intelligent recommendation method for suppliers facing to automobile accessory purchasing comprises the following steps: S1, constructing a knowledge graph G composed of nodes and edges according to historical orders of a buyer, wherein the nodes comprise supplier nodes, accessory model nodes, adaptation vehicle model nodes and buyer nodes; S2, determining a fitting model node or an adaptive vehicle model node as a demand node according to purchasing demands, and screening a plurality of qualified provider nodes of the demand node in a knowledge graph G on the basis of a Monte Carlo Dropout mechanism by a graph neural network model GNN; S3, screening a plurality of potential provider nodes from all qualified provider nodes based on the self-adaptive multi-objective optimization function; And S4, calculating side note meaning scores of all potential provider nodes, and outputting providers corresponding to the first k potential provider nodes in descending order of the side note meaning scores as a recommendation degree descending order. Preferably, in S1, the following is also included S11, tracking historical orders by a buyer to acquire related data of the automobile parts, wherein the related data comprise order numbers, part models, adaptive vehicle types, suppliers, buyers, part prices, lead time and supplier reputation; S12, using the model of the accessory, the adaptive model, the supplier and the purchasing company as nodes, and connecting two nodes appearing in the same historical order by edges to form a basic map BS, wherein each edge comprises a relationship type and corresponding relationship content of the nodes at two ends in the same historical order; s13, carrying out hierarchical aggregation node embedding on the basic map BS based on the graph neural network model GNN to obtain a knowledge map G, and carrying out hierarchical aggregation node embedding by using the following formula: ; Wherein, the Indicating that node v is on the firstThe embedded vector of the layer is used to determine,=1,...;Representing a nonlinear activation function; Representing the embedded vector of the neighbor node u at the first layer; Is a relation weight matrix; representing a set of neighbor nodes adjacent to node v under relationship type r; Representing an aggregation function under a relationship type r; all relation types in the basic map BS are represented, wherein the relation types comprise price relation type, adaptation relation type, reputation relation type and lead time relation type, and each node is obtained after the embedded vector of the first layer is initialized and assigned for the graph neural network model GNN according to node information. Preferably, in S2, the following is also included S21, enabling the Monte Carlo Dropout value of the graph neural network model G