Intelligent procurement recommendation system and method based on AI technology

Abstract

The present application provides an intelligent procurement recommendation system and method based on AI technology, which relates to the field of intelligent recommendation. It uses AI-based data information analysis and processing to extract text information from the bidding project files uploaded by the demander and the bidding files provided by the first supplier and perform semantic coding. At the same time, it performs semantic coding and kernel dynamic aggregation on the evaluation information captured on the Internet, so as to intelligently determine whether to recommend the first supplier based on the joint features between the bidding semantic features and the evaluation aggregation features and the fitness measure between the bidding semantic features. In this way, automated file review can be achieved, and deep meanings that cannot be touched by keyword matching can be captured, which is conducive to more accurate selection of suppliers, thereby achieving more intelligent procurement recommendations.

Description

Technical Field

[0001] The present application relates to the field of intelligent recommendation, and more specifically, to an intelligent procurement recommendation system and method based on AI technology. Background Art

[0002] Procurement is a key link in business operations, aiming to enhance competitiveness through efficient resource acquisition. For different users such as enterprises, universities and scientific research institutions, procurement is a key link to ensure efficient operations and meet specific needs. Enterprises need to constantly find suitable suppliers and products to meet production and operation needs; universities and scientific research institutions need special experimental equipment, scientific research materials and technical services. Users usually purchase high-value, high-service, and high-quality products through bidding. An effective procurement strategy can establish a stable supplier relationship and enhance the resilience and flexibility of the supply chain to better respond to market changes.

[0003] However, the traditional procurement process usually involves a lot of manual judgment and simple keyword matching. Specifically, the traditional process of handling bidding documents and tender documents is highly dependent on manual review. This process is not only time-consuming and labor-intensive, but also difficult to ensure the consistency and objectivity of manual review when faced with large-scale documents, resulting in inconsistent standards. In addition, the traditional method of matching information between user needs and suppliers often uses keyword matching. Although this method can quickly locate certain specific information, for complex business terms, technical specifications and other content, simple keyword searches often fail to capture the deep meaning of the document and cannot accurately express its true intentions. Especially for unpopular products, users have limited knowledge of suppliers in the market, and it is difficult to ensure that they choose suppliers and goods with reasonable prices and reliable quality.

[0004] Therefore, an intelligent procurement recommendation solution based on AI technology is expected. Summary of the invention

[0005] In order to solve the above technical problems, this application is proposed. The embodiments of this application provide an intelligent purchase recommendation system and method based on AI technology.

[0006] According to one aspect of the present application, an AI-based intelligent procurement recommendation system is provided, comprising: a bidding project file acquisition module, used to acquire a bidding project file uploaded by a demander; a bidding text encoding module, used to extract text information of the bidding project from the bidding project file, and semantically encode the text information of the bidding project to obtain bidding project information semantic features; a bidding document acquisition module, used to acquire a bidding document provided by a first supplier; a bidding text encoding module, used to extract text information of the bidding project from the bidding document, and semantically encode the text information of the bidding project to obtain bidding project information semantic features; an evaluation information acquisition module, used to capture a set of evaluation information for the first supplier through the Internet; an evaluation kernel semantic aggregation module, used to semantically encode the set of evaluation information of the first supplier and dynamically semantically aggregate kernel features to obtain evaluation kernel semantic aggregation features; a bidding-evaluation semantic combination module, used to combine the bidding project information semantic features and the evaluation kernel semantic aggregation features to obtain the first supplier bidding-evaluation semantic combination features; a supplier recommendation module, used to calculate a fitness metric between the first supplier bidding-evaluation semantic combination features and the bidding project information semantic features, and determine whether to recommend the first supplier based on the fitness metric.

[0007] In the above-mentioned intelligent procurement recommendation system based on AI technology, the bidding text encoding module is used to: extract the text information of the bidding project from the bidding project file; perform word segmentation on the text information of the bidding project and input it into a bidding project semantic encoder containing a word embedding layer to obtain a bidding project information semantic encoding vector as the bidding project information semantic feature.

[0008] In the above-mentioned intelligent procurement recommendation system based on AI technology, the bidding text encoding module is used to: extract the text information of the bidding project from the bidding document; perform word segmentation on the text information of the bidding project and input it into a bidding project semantic encoder including a word embedding layer to obtain a bidding project information semantic encoding vector as the bidding project information semantic feature.

[0009] In the above-mentioned intelligent procurement recommendation system based on AI technology, the evaluation kernel semantic aggregation module includes: an evaluation information semantic encoding unit, which is used to semantically encode each evaluation information in the set of evaluation information of the first supplier to obtain a set of evaluation information semantic encoding vectors; an evaluation information semantic dynamic aggregation unit, which is used to perform kernel feature dynamic semantic aggregation on the set of evaluation information semantic encoding vectors to obtain the evaluation kernel semantic aggregation feature.

[0010] In the above-mentioned intelligent procurement recommendation system based on AI technology, the evaluation information semantic dynamic aggregation unit includes: an evaluation information initial aggregation representation calculation subunit, which is used to calculate the positional mean vector of the set of evaluation information semantic coding vectors to obtain the evaluation information initial aggregation representation vector; an evaluation information static energy potential factor calculation subunit, which is used to calculate the static energy potential factor of each evaluation information semantic coding vector in the set of evaluation information semantic coding vectors to obtain a set of evaluation information static energy potential factors; an evaluation information dynamic energy potential factor calculation subunit, which is used to calculate the dynamic energy potential factor of each evaluation information semantic coding vector in the set of evaluation information semantic coding vectors relative to the evaluation information initial aggregation representation vector to obtain a set of evaluation information dynamic energy potential factors; an evaluation information superposition state energy potential factor calculation subunit A unit is used to determine the superposition state energy potential factor of each evaluation information semantic coding vector in the set of evaluation information semantic coding vectors based on the static energy potential factor and the dynamic energy potential factor of each evaluation information semantic coding vector in the set of evaluation information semantic coding vectors to obtain a set of evaluation information superposition state energy potential factors; an evaluation information superposition state energy potential factor screening subunit is used to input the set of evaluation information superposition state energy potential factors into the energy potential factor gated weighting network to obtain a set of evaluation information semantic feature superposition energy potential weight coefficients; an evaluation kernel semantic aggregation feature generation subunit is used to calculate the position-weighted sum of the set of evaluation information semantic coding vectors based on the set of evaluation information semantic feature superposition energy potential weight coefficients to obtain an evaluation kernel semantic aggregation feature vector as the evaluation kernel semantic aggregation feature.

[0011] In the above-mentioned intelligent procurement recommendation system based on AI technology, the evaluation information static energy potential factor calculation subunit is used to: calculate the mean and standard deviation of the evaluation information semantic coding vector to obtain the evaluation information semantic mean and the evaluation information semantic standard deviation; calculate the positional subtraction between the evaluation information semantic coding vector and the evaluation information semantic mean, divide the evaluation information semantic offset vector after the subtraction by the evaluation information semantic standard deviation, and calculate the fourth power of the eigenvalue of each position in the vector after the division to obtain the evaluation information semantic static energy potential vector; calculate the positional mean of the evaluation information semantic static energy potential vector to obtain the evaluation information static energy potential factor.

[0012] In the above-mentioned intelligent procurement recommendation system based on AI technology, the evaluation information dynamic potential factor calculation subunit is used to: matrix multiply the evaluation information aggregation weight matrix with the evaluation information initial aggregation representation vector to obtain the evaluation information initial aggregation weight representation vector; matrix multiply the evaluation information weight matrix with the evaluation information semantic coding vector to obtain the evaluation information semantic coding weight vector; after positionally dot-multiplying the evaluation information initial aggregation weight representation vector with the evaluation information semantic coding weight vector, positionally add the obtained evaluation information semantic coding fusion weight vector and the evaluation information bias vector to obtain the evaluation information semantic dynamic potential vector; calculate the square of the Frobenius norm of the evaluation information semantic dynamic potential vector to obtain the evaluation information dynamic potential factor.

[0013] In the above-mentioned intelligent procurement recommendation system based on AI technology, the bid-evaluation semantic joint module is used to: cascade the bid project information semantic encoding vector and the evaluation kernel semantic aggregation feature vector to obtain the first supplier bid-evaluation semantic joint encoding vector as the first supplier bid-evaluation semantic joint feature.

[0014] In the above-mentioned intelligent procurement recommendation system based on AI technology, the supplier recommendation module is used to: input the first supplier's bid-evaluation semantic joint encoding vector and the bidding project information semantic encoding vector into the procurement matching measurement network to obtain the bidding project-first supplier fitness measurement value as the fitness measurement value; based on the comparison between the bidding project-first supplier fitness measurement value and a preset threshold, determine whether to recommend the first supplier.

[0015] According to another aspect of the present application, an intelligent procurement recommendation method based on AI technology is provided, which includes: obtaining a bidding project file uploaded by a demander; extracting text information of the bidding project from the bidding project file, and semantically encoding the text information of the bidding project to obtain semantic features of the bidding project information; obtaining a bidding document provided by a first supplier; extracting text information of the bidding project from the bidding document, and semantically encoding the text information of the bidding project to obtain semantic features of the bidding project information; capturing a set of evaluation information for the first supplier through the Internet; semantically encoding and dynamically semantically aggregating kernel features of the set of evaluation information of the first supplier to obtain evaluation kernel semantic aggregation features; combining the bidding project information semantic features and the evaluation kernel semantic aggregation features to obtain a first supplier bid-evaluation semantic joint feature; calculating a fitness measure between the first supplier bid-evaluation semantic joint feature and the bidding project information semantic feature, and determining whether to recommend the first supplier based on the fitness measure.

[0016] Compared with the prior art, the AI-based intelligent procurement recommendation system and method provided by this application uses AI-based data information analysis and processing to extract text information from the bidding project files uploaded by the demander and the bidding files provided by the first supplier and perform semantic coding, and at the same time, semantic coding and kernel dynamic aggregation of the evaluation information captured on the Internet, so as to intelligently determine whether to recommend the first supplier based on the fit between the joint features of the bidding semantic features and the evaluation aggregation features and the bidding semantic features. In this way, automated file review can be achieved, and deep meanings that cannot be touched by keyword matching can be captured, which is conducive to more accurate selection of suppliers, thereby achieving more intelligent procurement recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] In the attached picture: Figure 1 This is a system block diagram of an intelligent procurement recommendation system based on AI technology according to an embodiment of the present application.

[0019] Figure 2 This is a data flow diagram of an intelligent procurement recommendation system based on AI technology according to an embodiment of the present application.

[0020] Figure 3 This is a block diagram of an evaluation kernel semantic aggregation module in an AI-based intelligent procurement recommendation system according to an embodiment of the present application.

[0021] Figure 4 This is a block diagram of a semantic dynamic aggregation unit of evaluation information in an intelligent procurement recommendation system based on AI technology according to an embodiment of the present application.

[0022] Figure 5 The present invention is a flowchart of an intelligent purchasing recommendation method based on AI technology according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0024] It should be noted that in this application, all actions to obtain signals, information or data are carried out in compliance with the relevant data protection laws and policies of the country where the device is located and with the authorization of the owner of the corresponding device.

[0025] Procurement activities are a key component of corporate operations, and their purpose is to enhance the company's market competitiveness by optimizing resource allocation. Whether it is a company, university or scientific research organization, procurement is a key link to ensure its efficient operation and meet specific needs. Companies need to continuously look for suitable suppliers and products to meet their production and operation needs; while universities and scientific research institutions need to purchase specific experimental equipment, scientific research materials and technical services. When purchasing high-value, high-service, and high-quality products, bidding is usually used. An effective procurement strategy can establish a solid supplier relationship, enhance the flexibility and adaptability of the supply chain, and better respond to market changes.

[0026] However, traditional procurement processes often involve a lot of manual operations and judgments, as well as simple keyword matching. Specifically, when processing bidding documents, traditional methods rely heavily on manual review, which is not only time-consuming and labor-intensive, but also difficult to ensure the consistency and objectivity of manual review when processing a large number of documents, resulting in inconsistent evaluation criteria. In addition, traditional methods often use keyword matching technology to match user needs with supplier information. Although this method can quickly locate specific information, for complex business terms and technical specifications, simple keyword searches often fail to capture the deep meaning of the document and cannot accurately reflect its true intentions. Especially for rare commodities on the market, users have limited knowledge of suppliers and it is difficult to ensure that they can choose suppliers and goods with reasonable prices and reliable quality.

[0027] In response to the above technical problems, the technical concept of this application is to obtain the bidding project documents uploaded by the demander and the bidding documents provided by the first supplier, and to capture a collection of evaluation information for the first supplier, and to use AI-based data information analysis and processing methods to extract text information from the bidding project and bidding documents respectively and perform semantic encoding, and at the same time perform semantic encoding and kernel dynamic aggregation on the evaluation information, so as to intelligently determine whether to recommend the first supplier based on the joint features between the bidding semantic features and the evaluation aggregation features and the fitness measure between the bidding semantic features. In this way, it is possible to capture the deep meaning that cannot be touched by keyword matching, reduce the reliance on manual review, and by collecting and analyzing the evaluation information on the market, even if the user has limited knowledge of the supplier, it is possible to analyze potential high-quality suppliers based on a large amount of data to help the user make a better choice, thereby providing the user with a more efficient and accurate procurement recommendation solution.

[0028] Figure 1 This is a system block diagram of an intelligent procurement recommendation system based on AI technology according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the intelligent purchase recommendation system based on AI technology according to an embodiment of the present application. Figure 1 and Figure 2 As shown, in the intelligent procurement recommendation system 100 based on AI technology, it includes: a bidding project file acquisition module 110, which is used to acquire the bidding project file uploaded by the demander; a bidding text encoding module 120, which is used to extract the text information of the bidding project from the bidding project file, and semantically encode the text information of the bidding project to obtain the semantic features of the bidding project information; a bidding document acquisition module 130, which is used to acquire the bidding document provided by the first supplier; a bidding text encoding module 140, which is used to extract the text information of the bidding project from the bidding document, and semantically encode the text information of the bidding project to obtain the semantic features of the bidding project information; an evaluation information acquisition module 150, which is used to A collection of evaluation information for the first supplier is captured through the Internet; an evaluation kernel semantic aggregation module 160 is used to perform semantic encoding and kernel feature dynamic semantic aggregation on the collection of evaluation information of the first supplier to obtain evaluation kernel semantic aggregation features; a bid-evaluation semantic combination module 170 is used to combine the bidding project information semantic features and the evaluation kernel semantic aggregation features to obtain the first supplier's bid-evaluation semantic combination features; a supplier recommendation module 180 is used to calculate a fitness measure between the first supplier's bid-evaluation semantic combination features and the bidding project information semantic features, and determine whether to recommend the first supplier based on the fitness measure.

[0029] In the embodiment of the present application, the bidding project file acquisition module 110 is used to acquire the bidding project file uploaded by the demander. It should be understood that the bidding project file uploaded by the demander generally contains an overview of the project, detailed description of the required products or services, delivery requirements, and supplier qualification requirements. By collecting the bidding project file and analyzing the information in the bidding project file, the bidding requirements of the project can be clarified, which is conducive to more accurately evaluating the suitability of the supplier, thereby improving the reliability of the procurement decision.

[0030] In an embodiment of the present application, the bidding text encoding module 120 is used to extract the text information of the bidding project from the bidding project file, and semantically encode the text information of the bidding project to obtain the semantic features of the bidding project information. Specifically, in an embodiment of the present application, the bidding text encoding module is used to: extract the text information of the bidding project from the bidding project file; perform word segmentation on the text information of the bidding project and input it into a bidding project semantic encoder containing a word embedding layer to obtain a bidding project information semantic encoding vector as the bidding project information semantic feature. It should be understood that the bidding project file contains a large amount of project information, such as the background, objectives, budget constraints, time schedule, etc. of the project, and in order to focus on the key text information in the file, in the technical solution of the present application, the text information of the bidding project is extracted from the bidding project file. Then, in order to be able to more deeply understand the semantic information of the specific needs of the demander for the project, further, the text information of the bidding project is semantically encoded to obtain the bidding project information semantic encoding vector. In particular, in a specific embodiment of the present application, the text information of the bidding project is segmented and then input into a bidding project semantic encoder containing a word embedding layer to obtain a bidding project information semantic encoding vector. In this way, the context information contained in each word can be understood and analyzed more clearly and finely, so as to ensure that the subsequent recommended suppliers can truly meet the actual needs of the demand side.

[0031] In the embodiment of the present application, the bidding document acquisition module 130 is used to acquire the bidding document provided by the first supplier. It should be understood that the bidding document provided by the first supplier generally includes the basic profile information of the supplier, the specific technical solution and description information for the bidding project, the qualification certification information of the supplier, etc. By collecting the bidding documents provided by the first supplier and analyzing the contents of the documents, it is possible to determine whether the supplier meets the requirements of the bidding project, and then determine whether to recommend the supplier.

[0032] In an embodiment of the present application, the bid text encoding module 140 is used to extract the text information of the bid project from the bid document, and semantically encode the text information of the bid project to obtain the semantic features of the bid project information. Specifically, in an embodiment of the present application, the bid text encoding module is used to: extract the text information of the bid project from the bid document; perform word segmentation on the text information of the bid project and input it into a bid project semantic encoder containing a word embedding layer to obtain a bid project information semantic encoding vector as the semantic features of the bid project information. It should be understood that the bid document contains the supplier's detailed response to the bidding project, including key information such as technical solutions and qualification certificates. This information is an important basis for evaluating whether the supplier meets the bidding requirements. In order to provide an accurate and complete data source for subsequent data analysis, it is first necessary to extract the text information of the bid project from the bid document. Accordingly, considering that the text information of the bidding project is generally composed of basic units-vocabulary, the efficiency of data processing and analysis can be effectively improved by performing word segmentation on the text information of the bidding project. In order to better capture and understand the key information in the bidding project, the segmented text information can be input into the bidding project semantic encoder containing the word embedding layer to obtain the bidding project information semantic features that can well reflect the bidding project text context information.

[0033] In the embodiment of the present application, the evaluation information acquisition module 150 is used to capture a collection of evaluation information for the first supplier through the Internet. It should be understood that the evaluation information usually includes feedback on the supplier's performance in the market, service level and other aspects, and this information is an important window to understand the overall situation of the supplier. By obtaining the evaluation information, it is possible to make up for the lack of information or one-sidedness that may exist in the bidding documents, which helps to have a deeper and more comprehensive understanding of the supplier, and then help make a clearer recommendation decision.

[0034] In the embodiment of the present application, the evaluation kernel semantic aggregation module 160 is used to perform semantic encoding and kernel feature dynamic semantic aggregation on the set of evaluation information of the first supplier to obtain evaluation kernel semantic aggregation features. Specifically, Figure 3 FIG. 1 is a block diagram of an evaluation kernel semantic aggregation module in an AI-based intelligent purchasing recommendation system according to an embodiment of the present application. Figure 3 As shown, the evaluation kernel semantic aggregation module 160 includes: an evaluation information semantic encoding unit 161, used to semantically encode each evaluation information in the set of evaluation information of the first supplier to obtain a set of evaluation information semantic encoding vectors; an evaluation information semantic dynamic aggregation unit 162, used to perform kernel feature dynamic semantic aggregation on the set of evaluation information semantic encoding vectors to obtain the evaluation kernel semantic aggregation feature.

[0035] It should be understandable that the set of evaluation information of the first supplier contains information on multiple aspects, such as service quality, delivery capability, after-sales service, etc. Therefore, in order to capture and mine the key and implicit deep semantic information in the evaluation information, in the technical solution of the present application, each evaluation information in the set of evaluation information of the first supplier is semantically encoded to obtain a set of evaluation information semantic encoding vectors. In this way, the deep meaning in the evaluation information can be more accurately captured, which helps to more accurately understand the intentions and feelings of the evaluator. In particular, in a specific embodiment of the present application, a converter-based BERT model can be used to semantically encode each evaluation information in the set of evaluation information of the first supplier, thereby obtaining a set of evaluation information semantic encoding vectors.

[0036] Furthermore, considering that there are complex relationships between different evaluation information semantic features in the set of evaluation information semantic coding vectors, such as the mutual influence and dependency between different evaluation information, and each feature contains key semantic information and essential core features of different aspects, in order to integrate the key core features of the semantic coding vectors of multiple evaluation information to form a comprehensive evaluation result to cover multiple aspects of the supplier, such as service quality, delivery capability, price competitiveness and other semantic information, in the technical solution of the present application, the kernel features of the set of evaluation information semantic coding vectors are dynamically semantically aggregated to obtain an evaluation kernel semantic aggregation feature vector as an evaluation kernel semantic aggregation feature.

[0037] In detail, first, in order to find an overall feature that can represent all evaluation information, that is, to identify a representative center point or standard vector to help eliminate interference factors in the data and make the data distribution clearer, the positional mean vector of the set of semantic encoding vectors of the evaluation information is calculated to obtain the initial aggregate representation vector of the evaluation information. Next, the static potential factor of each evaluation information semantic encoding vector is calculated to quantify the importance of each evaluation information and its unique characteristic attributes, so as to highlight those evaluation information that are crucial to the supplier evaluation, and obtain a set of static potential factors of the evaluation information. Then, in order to consider the semantic association interaction between each individual vector and the initial aggregation (center point), the dynamic potential factor of each evaluation information semantic encoding vector relative to the evaluation information initial aggregation representation vector is calculated to obtain a set of dynamic potential factors of the evaluation information, so that the weight of each subsequent evaluation information can be dynamically adjusted according to the interaction between each evaluation information and the initial aggregation representation vector. After that, the importance of each evaluation information is comprehensively evaluated from multiple dimensions to reflect its comprehensive influence in the whole process. The static potential factors and dynamic potential factors of the semantic encoding vectors of each evaluation information are superimposed to obtain a set of superimposed potential factors of the evaluation information, so as to more accurately evaluate the overall performance of the supplier. Next, in order to adaptively adjust the weight of each evaluation information according to the superimposed potential factors of the evaluation information, the set of superimposed potential factors of the evaluation information is input into the potential factor gated weighting network to obtain a set of superimposed potential weight coefficients of the semantic features of the evaluation information. That is, through the gated weighting network, the network can dynamically optimize the weights and flexibly adjust them according to the specific evaluation information to more effectively identify the key and important features for the final result and reduce the influence of noise evaluation information. Finally, the set of evaluation information semantic encoding vectors is weighted and summed based on the set of weight coefficients to obtain the final evaluation kernel semantic aggregation feature vector containing the key kernel evaluation semantic information of each feature.

[0038] Specifically, Figure 4 FIG. 1 is a block diagram of a semantic dynamic aggregation unit of evaluation information in an intelligent purchasing recommendation system based on AI technology according to an embodiment of the present application. Figure 4As shown, the evaluation information semantic dynamic aggregation unit 162 includes: an evaluation information initial aggregation representation calculation subunit 1621, which is used to calculate the positional mean vector of the set of evaluation information semantic coding vectors to obtain the evaluation information initial aggregation representation vector; an evaluation information static energy potential factor calculation subunit 1622, which is used to calculate the static energy potential factor of each evaluation information semantic coding vector in the set of evaluation information semantic coding vectors to obtain a set of evaluation information static energy potential factors; an evaluation information dynamic energy potential factor calculation subunit 1623, which is used to calculate the dynamic energy potential factor of each evaluation information semantic coding vector in the set of evaluation information semantic coding vectors relative to the evaluation information initial aggregation representation vector to obtain a set of evaluation information dynamic energy potential factors; an evaluation information superposition state energy potential factor calculation subunit 1624, which is used to calculate the dynamic energy potential factor of each evaluation information semantic coding vector in the set of evaluation information semantic coding vectors relative to the evaluation information initial aggregation representation vector to obtain a set of evaluation information dynamic energy potential factors. Based on the static potential factors and dynamic potential factors of each evaluation information semantic coding vector in the set of evaluation information semantic coding vectors, the superposition state potential factors of each evaluation information semantic coding vector in the set of evaluation information semantic coding vectors are determined to obtain a set of evaluation information superposition state potential factors; an evaluation information superposition state potential factor screening subunit 1625 is used to input the set of evaluation information superposition state potential factors into a potential factor gated weighting network to obtain a set of evaluation information semantic feature superposition potential weight coefficients; an evaluation kernel semantic aggregation feature generation subunit 1626 is used to calculate the position-weighted sum of the set of evaluation information semantic coding vectors based on the set of evaluation information semantic feature superposition potential weight coefficients to obtain an evaluation kernel semantic aggregation feature vector as the evaluation kernel semantic aggregation feature.

[0039] More specifically, in an embodiment of the present application, the evaluation information static energy potential factor calculation subunit is used to: calculate the mean and standard deviation of the evaluation information semantic coding vector to obtain the evaluation information semantic mean and the evaluation information semantic standard deviation; calculate the positional subtraction of the evaluation information semantic coding vector and the evaluation information semantic mean, divide the evaluation information semantic offset vector after the subtraction by the evaluation information semantic standard deviation, and calculate the fourth power of the eigenvalue of each position in the vector after the division to obtain the evaluation information semantic static energy potential vector; calculate the positional mean of the evaluation information semantic static energy potential vector to obtain the evaluation information static energy potential factor.

[0040] More specifically, in an embodiment of the present application, the evaluation information dynamic energy potential factor calculation subunit is used to: matrix multiply the evaluation information aggregation weight matrix with the evaluation information initial aggregation representation vector to obtain the evaluation information initial aggregation weight representation vector; matrix multiply the evaluation information weight matrix with the evaluation information semantic coding vector to obtain the evaluation information semantic coding weight vector; perform positional dot multiplication of the evaluation information initial aggregation weight representation vector and the evaluation information semantic coding weight vector, and then perform positional addition of the obtained evaluation information semantic coding fusion weight vector and the evaluation information bias vector to obtain the evaluation information semantic dynamic energy potential vector; calculate the square of the Frobenius norm of the evaluation information semantic dynamic energy potential vector to obtain the evaluation information dynamic energy potential factor.

[0041] More specifically, in an embodiment of the present application, the evaluation information superposition state energy potential factor calculation subunit is used to: calculate an exponential function with the natural constant e as the base and the static energy potential factor of the evaluation information semantic coding vector as the exponent to obtain the evaluation information semantic bias static energy potential factor; calculate an exponential function with the natural constant e as the base and the square of the dynamic energy potential factor of the evaluation information semantic coding vector as the exponent to obtain the evaluation information semantic bias dynamic energy potential factor; add the evaluation information semantic bias static energy potential factor and the evaluation information semantic bias dynamic energy potential factor to obtain the superposition state energy potential factor of the evaluation information semantic coding vector.

[0042] More specifically, in an embodiment of the present application, the evaluation information superposition state energy potential factor screening subunit is used to: normalize the set of the evaluation information superposition state energy potential factors to obtain a set of normalized evaluation information superposition state energy potential factors; input the set of normalized evaluation information superposition state energy potential factors into a mask function for gated mask screening to obtain a set of evaluation information semantic feature superposition energy potential weight coefficients, wherein, in response to each normalized evaluation information superposition state energy potential factor in the set of normalized evaluation information superposition state energy potential factors being greater than a preset threshold, the normalized evaluation information superposition state energy potential factor is set to the original value, and the rest are set to zero, so as to obtain the set of evaluation information semantic feature superposition energy potential weight coefficients.

[0043] More specifically, in an embodiment of the present application, the set of evaluation information superposition state energy potential factors is normalized to obtain a set of normalized evaluation information superposition state energy potential factors, including: calculating an exponential function value with the natural constant e as the base and the negative of each evaluation information superposition state energy potential factor in the set of evaluation information superposition state energy potential factors as the exponent to obtain a set of evaluation information superposition state energy potential exponential factors; adding the set of evaluation information superposition state energy potential exponential factors to the constant one by position, and then calculating the reciprocal of each evaluation information superposition state energy potential exponential modulation factor in the obtained set of evaluation information superposition state energy potential exponential modulation factors to obtain the set of normalized evaluation information superposition state energy potential factors.

[0044] In the embodiment of the present application, the evaluation kernel semantic aggregation module is used to: perform kernel feature dynamic semantic aggregation on the set of evaluation information semantic encoding vectors according to the following formula; wherein the formula is: .

[0045] in, is the set of semantic encoding vectors of the evaluation information, , , and are the first, second, and third evaluation information semantic encoding vectors in the set. and A semantic encoding vector of evaluation information, is the number of vectors in the set of evaluation information semantic encoding vectors, is the initial aggregate representation vector of the evaluation information, yes The eigenvalues ​​at each position in , and They are The mean and standard deviation of yes The number of eigenvalues ​​in , yes The corresponding evaluation information static energy potential factor, is the evaluation information aggregation weight matrix, is the evaluation information weight matrix, It is the point multiplication by position. is the evaluation information bias vector, is the square of the Frobenius norm of the calculated vector, yes The corresponding evaluation information dynamic energy potential factor, yes The corresponding evaluation information superposition state energy potential factor is, yes The corresponding normalized evaluation information superposition state energy potential factor is, For gated mask operation, is the preset threshold, yes The corresponding evaluation information semantic feature superposition potential weight coefficient, is the evaluation kernel semantic aggregation feature vector.

[0046] In an embodiment of the present application, the bidding-evaluation semantic union module 170 is used to combine the bidding project information semantic features and the evaluation kernel semantic aggregation features to obtain the first supplier bidding-evaluation semantic union features. Specifically, in an embodiment of the present application, the bidding-evaluation semantic union module is used to: cascade the bidding project information semantic encoding vector and the evaluation kernel semantic aggregation feature vector to obtain the first supplier bidding-evaluation semantic union encoding vector as the first supplier bidding-evaluation semantic union feature. It should be understood that in order to combine the first supplier bidding information and evaluation information to form a comprehensive evaluation result, so as to more accurately and comprehensively characterize the information of the first supplier, in the technical solution of the present application, the bidding project information semantic encoding vector and the evaluation kernel semantic aggregation feature vector are cascaded to obtain the first supplier bidding-evaluation semantic union encoding vector. That is to say, the first supplier bidding-evaluation semantic union encoding vector can more accurately reflect the overall performance of the supplier, including its performance in the bidding document and its performance in the historical evaluation, so as to improve the accuracy and reliability of subsequent recommendations.

[0047] In an embodiment of the present application, the supplier recommendation module 180 is used to calculate the fitness metric between the first supplier's bid-evaluation semantic joint feature and the bidding project information semantic feature, and determine whether to recommend the first supplier based on the fitness metric. Specifically, in an embodiment of the present application, the supplier recommendation module is used to: input the first supplier's bid-evaluation semantic joint encoding vector and the bidding project information semantic encoding vector into the procurement matching metric network to obtain the bidding project-first supplier fitness metric as the fitness metric; based on the comparison between the bidding project-first supplier fitness metric and a preset threshold, determine whether to recommend the first supplier.

[0048] It should be understood that the first supplier bid-evaluation semantic joint coding vector mainly expresses the semantic integration information about the supplier's bid and comments. The bidding project information semantic coding vector clearly expresses the semantic information of the demand side's needs and requirements for the bidding project. Therefore, in order to comprehensively evaluate the compatibility of the first supplier and the bidding project to ensure the accuracy of the recommendation results, in the technical solution of the present application, the first supplier bid-evaluation semantic joint coding vector and the bidding project information semantic coding vector are input into the procurement matching degree measurement network to obtain the bidding project-first supplier fitness measurement value. In this way, the matching degree between the supplier and the bidding project can be more accurately reflected, thereby helping the demand side to better understand and make decisions. In particular, in a specific embodiment of the present application, the cosine similarity between the first supplier bid-evaluation semantic joint coding vector and the bidding project information semantic coding vector can be calculated as the bidding project-first supplier fitness measurement value.

[0049] Then, based on the comparison between the bidding project-first supplier fitness metric and the preset threshold, determine whether to recommend the first supplier. That is, the fitness metric obtained by using the first supplier's bid-evaluation semantic joint coding vector and the bidding project information semantic coding vector to measure fitness is compared with the preset threshold to intelligently determine whether to recommend the first supplier. In this way, it is possible to capture the deep meaning that cannot be touched by keyword matching, reduce the reliance on manual review, and by collecting and analyzing the evaluation information on the market, even if the user has limited knowledge of the supplier, it can analyze potential high-quality suppliers based on a large amount of data to help the user make a better choice, thereby providing the user with a more efficient and accurate procurement recommendation solution.

[0050] In particular, when the bidding project information semantic encoding vector, the evaluation kernel semantic aggregation feature vector and the tender project information semantic encoding vector respectively represent the encoded semantic features of the supplier's bidding documents, the set of supplier's evaluation information and the text information of the tender project, when the matching measurement network is input to calculate the fitness measurement value, the first supplier bidding-evaluation semantic joint encoding vector and the tender project information semantic encoding vector will also have a remote global semantic matching space domain regression understanding representation imbalance relative to the feature distribution field selective state space of the local semantic space domain, thereby deviating from the semantic matching space domain mapping understanding of the source encoding semantic feature distribution representation, affecting the calculation accuracy of the tender project-first supplier fitness measurement value.

[0051] Preferably, inputting the first supplier bid-evaluation semantic joint coding vector and the tender project information semantic coding vector into a procurement matching degree measurement network to obtain a tender project-first supplier adaptation degree measurement value includes: concatenating the first supplier bid-evaluation semantic joint coding vector and the tender project information semantic coding vector into a supplier bid-tender project information semantic joint coding feature vector; determining the number of zero eigenvalues ​​in the supplier bid-tender project information semantic joint coding feature vector, and respectively calculating the reciprocal of the logarithm of the number of zero eigenvalues ​​with base two and the exponential value of the reciprocal of the number of zero eigenvalues ​​with base e to obtain a first supplier bid-tender project information semantic joint coding representation value and a second supplier bid-tender project information semantic joint coding representation value; calculating the power function of each eigenvalue of the supplier bid-tender project information semantic joint coding feature vector with the reciprocal of the number of zero eigenvalues ​​as the exponent to obtain a supplier bid-tender project information semantic joint coding structure state vector; comparing the supplier bid-tender project information semantic joint coding structure state vector with the first supplier The supplier bidding-tendering project information semantic joint coding representation value is point multiplied to obtain the supplier bidding-tendering project information semantic joint coding information representation vector; the autocorrelation matrix of the supplier bidding-tendering project information semantic joint coding structure state vector is point multiplied with the second supplier bidding-tendering project information semantic joint coding representation value and the weight hyperparameter respectively to obtain the supplier bidding-tendering project information semantic joint coding regression understanding matrix; the supplier bidding-tendering project information semantic joint coding information representation vector is matrix multiplied with the supplier bidding-tendering project information semantic joint coding regression understanding matrix to obtain the optimized supplier bidding-tendering project information semantic joint coding feature vector; the optimized supplier bidding-tendering project information semantic joint coding feature vector is split into an optimized first supplier bidding-evaluation semantic joint coding vector and an optimized tendering project information semantic coding vector; the optimized first supplier bidding-evaluation semantic joint coding vector and the optimized tendering project information semantic coding vector are calculated and input into the procurement matching degree measurement network to obtain the tendering project-first supplier fitness degree measurement value.

[0052] Here, the semantic joint coding representation value of the first supplier bidding-tendering project information and the semantic joint coding representation value of the second supplier bidding-tendering project information are expressed as: ;in, is the number of zero eigenvalues ​​in the semantic joint encoding feature vector of the supplier bidding-tendering project information, represents the logarithmic function with base 2, represents the natural exponential function, The semantic joint coding representation value of the bidding-tendering project information of the first supplier is A semantic joint encoding representation value of the bidding-tendering project information is provided for the second supplier.

[0053] And the optimized supplier bidding-tendering project information semantic joint encoding feature vector is expressed as: ;in, is the semantic joint encoding feature vector of the supplier bidding-tendering project information, and is a row vector, is the number of zero eigenvalues ​​in the semantic joint encoding feature vector of the supplier bidding-tendering project information, It means point multiplication by position. The semantic joint coding representation value of the bidding-tendering project information of the first supplier is represents a power function that calculates each eigenvalue of the semantic joint encoding eigenvector of the supplier bidding-tendering project information with the inverse of the number of zero eigenvalues ​​as the exponent, The information representation vector of the supplier bidding-tendering project information semantic joint encoding is obtained. is the weight hyperparameter, The semantic joint encoding representation value of the bidding-tendering project information of the second supplier is for The transposed vector of represents matrix multiplication, The semantic joint encoding regression understanding matrix of the supplier bidding-tendering project information is obtained. A semantic joint encoding feature vector for the optimized supplier bidding-tendering project information is provided.

[0054] That is, the present application realizes effective long-range modeling of the first supplier bid-evaluation semantic joint encoding vector and the bidding project information semantic encoding vector under field linear complexity by modeling the vector field zero dimension of the supplier bid-bidding project information semantic joint encoding feature vector obtained by cascading the first supplier bid-evaluation semantic joint encoding vector and the bidding project information semantic encoding vector as a global field redundant dependency. In this way, in the feature mapping regression process of the supplier bid-bidding project information semantic joint encoding feature vector, structured correlation self-distillation between information representation and regression understanding is performed in a selective state space based on feature distribution, thereby maintaining the overall information complexity of the feature distribution of the supplier bid-bidding project information semantic joint encoding feature vector while promoting the long-range dynamic regression understanding balance of its feature convergence, so as to improve the understanding consistency between the matching metric mapping regression and the feature extraction process, and improve the calculation accuracy of the bidding project-first supplier fitness metric. In this way, it is possible to capture deeper meanings that cannot be reached by keyword matching, reduce reliance on manual review, and by collecting and analyzing evaluation information on the market, even if users have limited knowledge of suppliers, they can analyze potential high-quality suppliers based on a large amount of data to help users make better choices, thereby providing users with a more efficient and accurate procurement recommendation solution.

[0055] In summary, the AI-based intelligent procurement recommendation system 100 according to the embodiment of the present application is explained, which uses AI-based data information analysis and processing to extract text information from the bidding project files uploaded by the demander and the bidding files provided by the first supplier and perform semantic coding, and at the same time, semantic coding and kernel dynamic aggregation are performed on the evaluation information captured on the Internet, so as to intelligently determine whether to recommend the first supplier based on the joint features between the bidding semantic features and the evaluation aggregation features and the fitness measure between the bidding semantic features. In this way, automated file review can be achieved, and deep meanings that cannot be touched by keyword matching can be captured, which is conducive to more accurate selection of suppliers, thereby achieving more intelligent procurement recommendations.

[0056] As described above, the intelligent procurement recommendation system 100 based on AI technology according to the embodiment of the present application can be implemented in various terminal devices, such as a server for intelligent procurement recommendation based on AI technology. In one example, the intelligent procurement recommendation system 100 based on AI technology according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the intelligent procurement recommendation system 100 based on AI technology can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the intelligent procurement recommendation system 100 based on AI technology can also be one of the many hardware modules of the terminal device.

[0057] Alternatively, in another example, the AI ​​technology-based intelligent procurement recommendation system 100 and the terminal device may also be separate devices, and the AI ​​technology-based intelligent procurement recommendation system 100 may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0058] Figure 5 Flow chart of the intelligent purchasing recommendation method based on AI technology according to an embodiment of the present application. Figure 5 As shown, in the intelligent procurement recommendation method based on AI technology, it includes: S110, obtaining a bidding project file uploaded by the demander; S120, extracting the text information of the bidding project from the bidding project file, and semantically encoding the text information of the bidding project to obtain the semantic features of the bidding project information; S130, obtaining a bidding document provided by a first supplier; S140, extracting the text information of the bidding project from the bidding document, and semantically encoding the text information of the bidding project to obtain the semantic features of the bidding project information; S150, capturing a set of evaluation information for the first supplier through the Internet; S160, semantically encoding the set of evaluation information of the first supplier and dynamically semantically aggregating the kernel features to obtain the evaluation kernel semantic aggregation features; S170, combining the bidding project information semantic features and the evaluation kernel semantic aggregation features to obtain the first supplier bidding-evaluation semantic joint features; S180, calculating the fitness metric between the first supplier bidding-evaluation semantic joint features and the bidding project information semantic features, and determining whether to recommend the first supplier based on the fitness metric.

[0059] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned intelligent procurement recommendation method based on AI technology have been referred to above. Figures 1 to 4 The description of the intelligent procurement recommendation system based on AI technology has been introduced in detail, and therefore, its repeated description will be omitted.

[0060] In summary, the intelligent procurement recommendation method based on AI technology based on the embodiment of the present application is explained, which uses AI-based data information analysis and processing to extract text information from the bidding project files uploaded by the demander and the bidding files provided by the first supplier and perform semantic coding, and at the same time, semantic coding and kernel dynamic aggregation are performed on the evaluation information captured on the Internet, so as to intelligently determine whether to recommend the first supplier based on the joint features between the bidding semantic features and the evaluation aggregation features and the fitness measure between the bidding semantic features. In this way, automated file review can be achieved, and deep meanings that cannot be touched by keyword matching can be captured, which is conducive to more accurate selection of suppliers, thereby achieving more intelligent procurement recommendations.

[0061] In other embodiments of the present application, a personalized recommendation method for bidding products based on AI is also provided, and the technical ideas for its implementation include the following steps: a. First, widely obtain data from multiple relevant websites, and then perform strict data cleaning operations to remove noise and outliers in the original data. Then perform feature construction, extract key features from the data and encode them reasonably, and then store the processed high-quality data in the database to provide a solid data foundation for subsequent model training. b. Use the data in the database to carefully construct training, test and verification data sets, and constantly adjust parameters to fine-tune the embedding model, rerank model and chat model. Here, the embedding model is used to map high-dimensional data to low-dimensional vector space for subsequent calculation and analysis; the rerank model is used to re-rank the search results to improve the accuracy of the recommendation; and the chat model provides users with interactive functions to better understand user needs. c. Obtain the user's relevant data A in real time through the online system, and use the embedding model to vectorize it to obtain vectorized data B. d. Use data A and data B to perform keyword and vectorized hybrid retrieval in elasticsearch. It should be noted that elasticsearch has efficient indexing and searching functions, and can quickly return a certain amount of sorted result data C. e. Use the rerank model to calculate the relevance between data A and data C, and accurately sort data C according to the calculated relevance value. The sorted result is represented by D. f. Modify data D according to business rules. For example, insert new data E into data D to meet the needs of real-time updates; or delete some data in data D that does not meet the requirements; or modify the sorting of data D to better adapt to user preferences. g. Recommend the result data obtained after the above processing to the user to provide the user with a personalized recommendation experience.

Claims

1. An intelligent purchasing recommendation system based on AI technology, characterized in that: include: The bidding project file acquisition module is used to acquire the bidding project files uploaded by the demander; A bidding text encoding module, used for extracting text information of the bidding project from the bidding project file, and performing semantic encoding on the text information of the bidding project to obtain semantic features of the bidding project information; A bidding document acquisition module, used to acquire the bidding document provided by the first supplier; A bidding text encoding module, used to extract the text information of the bidding project from the bidding document, and semantically encode the text information of the bidding project to obtain the semantic features of the bidding project information; an evaluation information acquisition module, used to capture the set of evaluation information for the first supplier through the Internet; an evaluation kernel semantic aggregation module, used to semantically encode the set of evaluation information of the first supplier and dynamically semantically aggregate kernel features to obtain evaluation kernel semantic aggregation features; A bidding-evaluation semantic union module is used to combine the bidding project information semantic features and the evaluation kernel semantic aggregation features to obtain the first supplier bidding-evaluation semantic union features; a supplier recommendation module is used to calculate the fitness measure between the first supplier bidding-evaluation semantic union features and the bidding project information semantic features, and determine whether to recommend the first supplier based on the fitness measure.

2. The intelligent purchasing recommendation system based on AI technology according to claim 1 is characterized in that: The bidding text encoding module is used to: extract the text information of the bidding project from the bidding project file; perform word segmentation on the text information of the bidding project and input it into a bidding project semantic encoder including a word embedding layer to obtain a bidding project information semantic encoding vector as the bidding project information semantic feature.

3. The intelligent purchasing recommendation system based on AI technology according to claim 2 is characterized in that: The bidding text encoding module is used to: extract the text information of the bidding project from the bidding document; perform word segmentation on the text information of the bidding project and input it into a bidding project semantic encoder including a word embedding layer to obtain a bidding project information semantic encoding vector as the bidding project information semantic feature.

4. The intelligent purchasing recommendation system based on AI technology according to claim 3 is characterized in that: The evaluation kernel semantic aggregation module includes: an evaluation information semantic encoding unit, which is used to semantically encode each evaluation information in the set of evaluation information of the first supplier to obtain a set of evaluation information semantic encoding vectors; an evaluation information semantic dynamic aggregation unit, which is used to perform kernel feature dynamic semantic aggregation on the set of evaluation information semantic encoding vectors to obtain the evaluation kernel semantic aggregation feature.

5. The intelligent purchasing recommendation system based on AI technology according to claim 4 is characterized in that: The evaluation information semantic dynamic aggregation unit includes: an evaluation information initial aggregation representation calculation subunit, which is used to calculate the positional mean vector of the set of evaluation information semantic coding vectors to obtain the evaluation information initial aggregation representation vector; an evaluation information static energy potential factor calculation subunit, which is used to calculate the static energy potential factor of each evaluation information semantic coding vector in the set of evaluation information semantic coding vectors to obtain a set of evaluation information static energy potential factors; an evaluation information dynamic energy potential factor calculation subunit, which is used to calculate the dynamic energy potential factor of each evaluation information semantic coding vector in the set of evaluation information semantic coding vectors relative to the evaluation information initial aggregation representation vector to obtain a set of evaluation information dynamic energy potential factors; an evaluation information superposition state energy potential factor calculation subunit, which is used to calculate the dynamic energy potential factor of each evaluation information semantic coding vector in the set of evaluation information semantic coding vectors relative to the evaluation information initial aggregation representation vector to obtain a set of evaluation information dynamic energy potential factors. The static potential factor and dynamic potential factor of each evaluation information semantic coding vector in the set of information semantic coding vectors are used to determine the superposition state potential factor of each evaluation information semantic coding vector in the set of evaluation information semantic coding vectors to obtain a set of evaluation information superposition state potential factors; an evaluation information superposition state potential factor screening subunit is used to input the set of evaluation information superposition state potential factors into the potential factor gated weighting network to obtain a set of evaluation information semantic feature superposition potential weight coefficients; an evaluation kernel semantic aggregation feature generation subunit is used to calculate the position-weighted sum of the set of evaluation information semantic coding vectors based on the set of evaluation information semantic feature superposition potential weight coefficients to obtain an evaluation kernel semantic aggregation feature vector as the evaluation kernel semantic aggregation feature.

6. The intelligent purchasing recommendation system based on AI technology according to claim 5 is characterized in that: The evaluation information static energy potential factor calculation subunit is used to: calculate the mean and standard deviation of the evaluation information semantic coding vector to obtain the evaluation information semantic mean and the evaluation information semantic standard deviation; calculate the positional subtraction between the evaluation information semantic coding vector and the evaluation information semantic mean, divide the evaluation information semantic offset vector after the subtraction by the evaluation information semantic standard deviation, and calculate the fourth power of the eigenvalue of each position in the vector after the division to obtain the evaluation information semantic static energy potential vector; The positional mean of the evaluation information semantic static energy potential vector is calculated to obtain the evaluation information static energy potential factor.

7. The intelligent purchasing recommendation system based on AI technology according to claim 6 is characterized in that: The evaluation information dynamic energy potential factor calculation subunit is used to: matrix multiply the evaluation information aggregation weight matrix with the evaluation information initial aggregation representation vector to obtain the evaluation information initial aggregation weight representation vector; matrix multiply the evaluation information weight matrix with the evaluation information semantic coding vector to obtain the evaluation information semantic coding weight vector; perform positional dot multiplication of the evaluation information initial aggregation weight representation vector and the evaluation information semantic coding weight vector, and then add the obtained evaluation information semantic coding fusion weight vector and the evaluation information bias vector by position to obtain the evaluation information semantic dynamic energy potential vector; The square of the Frobenius norm of the semantic dynamic potential vector of the evaluation information is calculated to obtain the dynamic potential factor of the evaluation information.

8. The intelligent purchasing recommendation system based on AI technology according to claim 7 is characterized in that: The bidding-evaluation semantic joint module is used to: cascade the bidding project information semantic coding vector and the evaluation kernel semantic aggregation feature vector to obtain the first supplier bidding-evaluation semantic joint coding vector as the first supplier bidding-evaluation semantic joint feature.

9. The intelligent purchasing recommendation system based on AI technology according to claim 8 is characterized in that: The supplier recommendation module is used to: input the first supplier bid-evaluation semantic joint encoding vector and the bidding project information semantic encoding vector into the procurement matching measurement network to obtain the bidding project-first supplier fitness measurement value as the fitness measurement value; based on the comparison between the bidding project-first supplier fitness measurement value and a preset threshold, determine whether to recommend the first supplier.

10. An intelligent purchasing recommendation method based on AI technology, characterized in that: include: Acquire a bidding project file uploaded by the demander; extract text information of the bidding project from the bidding project file, and semantically encode the text information of the bidding project to obtain semantic features of the bidding project information; acquire a bidding document provided by the first supplier; Extracting text information of the bidding project from the bidding document, and semantically encoding the text information of the bidding project to obtain semantic features of the bidding project information; Capturing a collection of evaluation information for the first supplier through the Internet; semantically encoding the collection of evaluation information of the first supplier and dynamically semantically aggregating kernel features to obtain evaluation kernel semantic aggregation features; combining the bidding project information semantic features and the evaluation kernel semantic aggregation features to obtain the first supplier bidding-evaluation semantic joint features; calculating a fitness measure between the first supplier bidding-evaluation semantic joint features and the bidding project information semantic features, and determining whether to recommend the first supplier based on the fitness measure.

Images