CN-122022966-A - Bidding purchasing provider optimization method, system and equipment based on AI technology

CN122022966ACN 122022966 ACN122022966 ACN 122022966ACN-122022966-A

Abstract

The application provides a bid-recruitment purchasing provider optimization method, system and equipment based on an AI technology, and relates to the technical field of artificial intelligence. The method comprises the steps of responding to a bid-inviting purchasing request from a purchasing terminal, generating corresponding purchasing intention vectors, screening a first provider sequence from a provider database based on the purchasing intention vectors and a pre-trained multipath screening matching model, and determining a preference conflict strength value between the purchasing intention vectors and preset dynamic user preference vectors based on a panoramic feature matching matrix corresponding to the first provider sequence. And under the condition that the preference conflict intensity value is larger than a first preset threshold value, generating dynamic correction weights based on the purchasing intention vector and the dynamic user preference vector to carry out weighted correction on the panoramic feature matching matrix according to the dynamic correction weights, screening a second provider sequence from the first provider sequence based on the corrected panoramic feature matching matrix, and sending the second provider sequence to the purchasing terminal.

Inventors

GUO JIAN
REN XUESHEN
ZHANG GUOWEI
ZHAO GUANG
LI HOUMING

Assignees

山东土地数字科技集团有限公司

Dates

Publication Date: 20260512
Application Date: 20260123

Claims (10)

1. An AI technology-based bidding purchasing provider preference method, the method comprising: responding to a bid-up purchase request from a purchase terminal, and generating a corresponding purchase intention vector; Screening a first supplier sequence from a supplier database based on the purchasing intention vector and a pre-trained multipath screening matching model; Determining a preference conflict strength value between the purchasing intention vector and a preset dynamic user preference vector based on a panoramic feature matching matrix corresponding to the first provider sequence, wherein the dynamic user preference vector is obtained based on dynamic analysis of historical bid-winning records, purchasing evaluation data and purchasing browsing records associated with the purchasing terminal; generating a dynamic correction weight based on the purchasing intention vector and the dynamic user preference vector under the condition that the preference conflict intensity value is larger than a first preset threshold value, so as to carry out weighted correction on the panoramic feature matching matrix according to the dynamic correction weight; And screening a second provider sequence from the first provider sequence based on the corrected panoramic feature matching matrix, and sending the second provider sequence to the purchasing terminal.
2. The method of claim 1, wherein generating the corresponding purchase intention vector comprises: carrying out semantic coding on the purchasing demand text corresponding to the bidding purchasing request to obtain a first depth semantic vector; entity extraction is carried out on the purchasing demand text to obtain corresponding key entities, wherein the key entities at least comprise a technical parameter entity, a delivery demand entity and a budget constraint entity; Matching each extracted key entity with a preset industrial knowledge graph to obtain an implicit demand entity set corresponding to the purchasing demand text by reasoning based on entity relation in the preset industrial knowledge graph, and determining a corresponding implicit demand feature vector; Feature fusion is carried out on the first depth semantic vector and the implicit demand feature vector, and a feature fusion result is input into a preset multi-head self-attention module for processing, so that a second depth semantic vector is generated, wherein the second depth semantic vector at least comprises demand intensity values corresponding to all demand dimensions respectively; And determining confidence scores corresponding to all the demand dimensions of the second depth semantic vector respectively based on the demand satisfaction quantitative information in the historical purchasing item, so as to construct the purchasing intention vector according to all the demand intensity values, all the confidence scores and all the demand dimensions.
3. The method of claim 2, wherein the screening from the vendor database for the first vendor sequence based on the purchase intent vector and a pre-trained multipath screening matching model, comprises: According to a preset weight list and the confidence scores in the purchasing intention vectors, matching first weight values corresponding to the demand dimensions respectively, and carrying out weighting and normalization processing on the demand intensity values based on the first weight values to obtain second weight vectors; inputting the second weight vector into the multipath screening matching model to determine a demand matching score corresponding to each candidate provider based on the feature vector of each candidate provider in the provider database and the second weight vector; Matching corresponding similar purchase item information in the historical purchase item by taking the demand dimension and the demand intensity value in the purchase intention vector as user characteristics through the multipath screening matching model, and determining collaborative filtering recommendation scores of candidate suppliers in the supplier database according to a winning supplier set and the purchase item similarity in the similar purchase item information; Performing expansion traversal on entity nodes corresponding to the required dimensions, of which the confidence scores are larger than a preset scoring threshold, in the purchasing intention vector as seed nodes in the preset industrial knowledge graph to determine that provider entity nodes with connection relations with the seed nodes are candidate suppliers, and determining knowledge graph association scores of the candidate suppliers according to reachable paths between the seed nodes and the provider entity nodes; And determining corresponding comprehensive recommendation scores based on a screening score multi-group corresponding to the same candidate provider, so that the candidate providers are added to the first provider sequence in sequence according to the order of the comprehensive recommendation scores from large to small, wherein the screening score multi-group comprises at least two of the demand matching scores, the collaborative filtering recommendation scores and the knowledge graph association scores.
4. The method of claim 1, wherein determining a preference collision strength value between the purchase intention vector and a preset dynamic user preference vector based on a panoramic feature matching matrix corresponding to the first vendor sequence, specifically comprises: The method comprises the steps of taking a demand intensity value of each demand dimension in the purchasing intention vector as a third weight vector, and carrying out weighted calculation on the third weight vector and the panoramic feature matching matrix to obtain the intention comprehensive score of each supplier in the first supplier sequence, wherein the behavior of each supplier of the panoramic feature matching matrix is listed as basic matching degree of the supplier corresponding to each demand dimension; Taking each dimension preference intensity value in the preset dynamic user preference vector as a fourth weight vector, and carrying out weighted calculation with the panoramic feature matching matrix to obtain preference comprehensive scores of each provider in the first provider sequence, wherein the dynamic user preference vector at least comprises the dimension preference intensity value of each required dimension and the corresponding preference confidence coefficient; respectively carrying out descending order sorting on the first provider sequence according to the intention comprehensive score and the preference comprehensive score to generate an intention sorting list and a preference sorting list, and calculating sorting difference degree of the intention sorting list and the preference sorting list; determining the first suppliers in the intention sorting list and the preference sorting list as a first intention supplier and a first preference supplier respectively, and determining corresponding first supplier characteristic difference values according to matching degree vectors respectively corresponding to the first intention supplier and the first preference supplier in the panoramic characteristic matching matrix; screening the first supplier sequence according to a preset purchasing decision standard score based on the intention comprehensive score and the preference comprehensive score to obtain an intention qualified supplier set and a preference qualified supplier set so as to calculate a corresponding decision result difference value; And based on a preset preference conflict weight set, carrying out weighted summation on the sorting difference degree, the first provider characteristic difference value and the decision result difference value, and determining the preference conflict strength value.
5. The method of claim 1, wherein generating dynamic correction weights based on the purchase intent vector and the dynamic user preference vector, comprises: According to the purchasing intention vector and the dynamic user preference vector, calculating conflict index values corresponding to each demand dimension respectively; when the conflict index value is larger than a second preset threshold value, judging that the corresponding demand dimension is a conflict dimension, and calculating a trend coefficient corresponding to the conflict dimension based on the purchasing intention vector and the dynamic user preference vector; and matching the tendency coefficient with each preset correction condition interval to calculate the dynamic correction weight according to the correction conditions obtained by matching, wherein different preset correction condition intervals are applied to different correction conditions for calculating the dynamic correction weight.
6. The method according to claim 1, wherein the weighting correction is performed on the panoramic feature matching matrix according to the dynamic correction weight, specifically comprising: expanding the dynamic correction weight vector corresponding to the dynamic correction weight into a diagonal matrix; And correcting the panoramic feature matching matrix through matrix multiplication according to the diagonal matrix.
7. The method according to claim 1, wherein screening a second vendor sequence from the first vendor sequence based on the modified panoramic feature matching matrix, in particular comprises: Taking the demand intensity value of each demand dimension in the purchasing intention vector as a third weight vector; Calculating dynamic comprehensive matching scores of all suppliers in the first supplier sequence according to the third weight vector and the corrected panoramic feature matching matrix; And sorting the first supplier sequences in a descending order according to the dynamic comprehensive matching scores, and screening the second supplier sequences from the first supplier sequences subjected to descending order according to a preset dynamic comprehensive matching score threshold.
8. A method according to claim 3, wherein determining knowledge-graph association scores for each candidate provider based on reachable paths between the seed node and the provider entity node, comprises: Determining each reachable path from the seed node to the provider entity node, and determining a path association strength value corresponding to the reachable path based on the path length and the relationship type on the path edge, wherein the path association strength value and the path length are in a negative correlation relationship; summing the path association intensity values corresponding to all reachable paths from the same seed node to the provider entity node to obtain single seed association intensity values, taking the provider entity node as a candidate provider, and calculating the sum value of the single seed association intensity values corresponding to each provider entity node to serve as the knowledge graph association score.
9. An AI technology-based bidding purchase provider preference system, the system comprising: The first generation module is used for responding to the bid-up purchase request from the purchase terminal and generating a corresponding purchase intention vector; The screening module is used for screening a first supplier sequence from a supplier database based on the purchasing intention vector and a pre-trained multipath screening matching model; The determining module is used for determining a preference conflict strength value between the purchasing intention vector and a preset dynamic user preference vector based on a panoramic feature matching matrix corresponding to the first provider sequence, wherein the dynamic user preference vector is obtained by dynamically analyzing historical bid-winning records, purchasing evaluation data and purchasing browsing records associated with the purchasing terminal; The second generation module is used for generating dynamic correction weights based on the purchasing intention vector and the dynamic user preference vector under the condition that the preference conflict intensity value is larger than a first preset threshold value so as to carry out weighted correction on the panoramic feature matching matrix according to the dynamic correction weights; and the screening and transmitting module is used for screening a second provider sequence from the first provider sequence based on the corrected panoramic feature matching matrix and transmitting the second provider sequence to the purchasing terminal.
10. An AI technology-based bidding purchase provider preference apparatus, the apparatus comprising: at least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores instructions executable by the at least one processor to enable the at least one processor to perform an AI-technology-based bidding purchasing provider preference method of any of the preceding claims 1-8.

Description

Bidding purchasing provider optimization method, system and equipment based on AI technology Technical Field The application relates to the technical field of artificial intelligence, in particular to a bid-recruitment purchasing provider optimization method, system and equipment based on an artificial intelligence (ARTIFICIAL INTELLIGENCE, AI) technology. Background With the penetration of enterprise digital transformation, an electronic bidding purchasing platform has become a mainstream purchasing mode. In order to efficiently and accurately screen out optimal candidates from mass suppliers, intelligent recommendation of the suppliers is realized by utilizing an artificial intelligence technology and becomes an industry development trend. Currently, the mainstream intelligent recommendation scheme usually analyzes the bidding documents through natural language processing technology, extracts the explicit requirements of the purchasing party, and then outputs a comprehensive recommendation list which is more in line with the requirements. The recommendation method is used for statically balancing the relation between the demand and the provider, the bidding preference of the purchasing party is not flexibly integrated into the recommendation method, and the reliability of the recommendation result is poor. After the recommendation result of the supplier is obtained, manual operation is still relied on, and purchasing personnel perform subjective judgment and screening according to historical cooperation experience, basic information of the supplier, qualification files and the like, so that human resource cost is consumed. Disclosure of Invention The embodiment of the application provides a bidding and purchasing supplier optimizing method, system and equipment based on an AI technology, which are used for solving the technical problems that the prior art lacks a bidding and purchasing supplier optimizing scheme which can integrate user bidding preference, dynamically screen suppliers related to purchasing demands, reduce labor cost investment and improve supplier recommending efficiency. In a first aspect, an embodiment of the present application provides a preferred method for bidding and purchasing suppliers based on AI technology, the method including: responding to a bid-up purchase request from a purchase terminal, and generating a corresponding purchase intention vector; Screening a first supplier sequence from a supplier database based on the purchasing intention vector and a pre-trained multipath screening matching model; Determining a preference conflict strength value between the purchasing intention vector and a preset dynamic user preference vector based on a panoramic feature matching matrix corresponding to the first provider sequence, wherein the dynamic user preference vector is obtained based on dynamic analysis of historical bid-winning records, purchasing evaluation data and purchasing browsing records associated with the purchasing terminal; generating a dynamic correction weight based on the purchasing intention vector and the dynamic user preference vector under the condition that the preference conflict intensity value is larger than a first preset threshold value, so as to carry out weighted correction on the panoramic feature matching matrix according to the dynamic correction weight; And screening a second provider sequence from the first provider sequence based on the corrected panoramic feature matching matrix, and sending the second provider sequence to the purchasing terminal. In one implementation of the present application, generating a corresponding purchase intention vector specifically includes: carrying out semantic coding on the purchasing demand text corresponding to the bidding purchasing request to obtain a first depth semantic vector; entity extraction is carried out on the purchasing demand text to obtain corresponding key entities, wherein the key entities at least comprise a technical parameter entity, a delivery demand entity and a budget constraint entity; Matching each extracted key entity with a preset industrial knowledge graph to obtain an implicit demand entity set corresponding to the purchasing demand text by reasoning based on entity relation in the preset industrial knowledge graph, and determining a corresponding implicit demand feature vector; Feature fusion is carried out on the first depth semantic vector and the implicit demand feature vector, and a feature fusion result is input into a preset multi-head self-attention module for processing, so that a second depth semantic vector is generated, wherein the second depth semantic vector at least comprises demand intensity values corresponding to all demand dimensions respectively; And determining confidence scores corresponding to all the demand dimensions of the second depth semantic vector respectively based on the demand satisfaction quantitative information in the historical purchasing item, so