Project transaction push method and system based on artificial intelligence

Abstract

The present invention discloses a project transaction push method and system based on artificial intelligence, the method comprising: obtaining project transaction information through project information sources; performing data preprocessing on the project transaction information to obtain preprocessed data; training the initial model according to the preprocessed data to obtain a project transaction push model; obtaining data to be processed, inputting the data to be processed into the project transaction push model, and outputting project transaction push information. By deeply mining and analyzing the user's historical transaction data, browsing behavior and other information, the model can accurately identify the user's needs and preferences and realize accurate personalized push. The model adopts advanced deep learning algorithms and distributed computing technologies, and can efficiently process large-scale project transaction data and extract valuable information from it. The model has iterative update capabilities, and can adjust and optimize the push strategy according to market changes and user needs to achieve accurate push.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based project transaction push method and system. Background Art

[0002] At present, there are the following problems and shortcomings in project transaction push: Low push accuracy: Existing project transaction push methods are often based on simple rule matching or keyword search, and cannot achieve accurate personalized push. This results in many of the push information received by users not meeting their actual needs, reducing the user experience. Limited data processing capabilities: With the continuous growth of project transaction data, existing data processing methods can no longer meet the processing needs of large-scale data. This has led to a lot of valuable information being buried in massive data and cannot be effectively utilized. Difficult model update: Due to the rapid and complex changes in the market, existing push models often require manual intervention to update and adjust. This not only increases labor costs, but may also lead to untimely model updates and inability to adapt to market changes. Summary of the invention

[0003] The present invention aims to solve at least one of the technical problems in the above-mentioned technologies to a certain extent. To this end, the first purpose of the present invention is to propose a project transaction push method based on artificial intelligence. Through big data and artificial intelligence technology, a push algorithm model for public resource transaction projects and services is constructed, and the public service platform completes the information push to market entities, so that various types of markets can understand market trends and seize transaction opportunities. By deeply mining and analyzing users' historical transaction data, browsing behavior and other information, the model can accurately identify users' needs and preferences and achieve accurate personalized push. The model adopts advanced deep learning algorithms and distributed computing technologies, which can efficiently process large-scale project transaction data and extract valuable information from it. The model has iterative update capabilities, and can adjust and optimize push strategies according to market changes and user needs to achieve accurate push.

[0004] The second object of the present invention is to propose a project transaction push system based on artificial intelligence.

[0005] To achieve the above-mentioned purpose, the first embodiment of the present invention proposes a project transaction push method based on artificial intelligence, comprising:

[0006] Obtain project transaction information through project information sources;

[0007] Perform data preprocessing on project transaction information to obtain preprocessed data;

[0008] The initial model is trained based on the preprocessed data to obtain the project transaction push model;

[0009] Obtain the data to be processed, input the data to be processed into the project transaction push model, and output the project transaction push information.

[0010] According to some embodiments of the present invention, sources of project information include government procurement projects, construction projects and open procurement projects; project transaction information includes project name, project type, transaction method, transaction amount and project description.

[0011] According to some embodiments of the present invention, data preprocessing is performed on project transaction information, including:

[0012] Remove duplicate, invalid and irregularly formatted data from project transaction information to obtain clean data;

[0013] Standardize and normalize the cleaned data to obtain the target data;

[0014] Feature extraction is performed on the target data to determine project features; wherein the project features include the manufacturer role, the number of historical completed projects of the supplier, the supplier's project type preference, and the purchaser's project type preference.

[0015] According to some embodiments of the present invention, the initial model is trained according to the preprocessed data to obtain the project transaction push model, including:

[0016] Divide the preprocessed data into training set, validation set and test set;

[0017] The initial model is trained through the training set based on the machine learning algorithm, the validation set is used to adjust the model parameters to prevent overfitting, and the test set is used to evaluate the model performance. During the training process, the initial model will continuously iterate and optimize the parameters to eventually obtain the project transaction push model.

[0018] According to some embodiments of the present invention, it also includes: online learning or updating of the project transaction push model based on the user's tracking feedback data; the tracking feedback data includes click-through rate and conversion rate.

[0019] According to some embodiments of the present invention, training the initial model through a training set based on a machine learning algorithm includes:

[0020] Based on the machine learning algorithm, the initial model is trained through the training set in the application scenario of the purchaser recommending suppliers and the application scenario of the supplier recommending the purchaser to obtain the first model;

[0021] Based on the machine learning algorithm, the initial model is trained through the training set in the supplier recommendation project application scenario and the project recommendation supplier application scenario to obtain the second model;

[0022] A project transaction push model is determined according to the first model and the second model.

[0023] According to some embodiments of the present invention, the initial model is trained based on a machine learning algorithm through a training set in an application scenario where a purchaser recommends a supplier and an application scenario where a supplier recommends a purchaser, to obtain a first model, including:

[0024] In the model training stage, the enterprise entity information and historical project information are determined based on the training set. The bidding unit is recorded as the purchaser role, and the winning bidder is recorded as the supplier role. The purchaser and historical project information, and the supplier and historical project information are output respectively.

[0025] Construct a purchaser portrait table based on the purchaser and historical project information; the purchaser portrait table includes the purchaser's basic information and the purchaser's historical purchase project preferences; train a purchaser clustering model based on the purchaser portrait table; the purchaser clustering model is used to output purchaser categories; cluster the purchaser categories based on the Kmeans clustering method to obtain purchaser clustering results; calculate the distance from the cluster category to the cluster center based on the purchaser clustering results, calculate the purchaser's basic score based on the distance to the cluster center, and obtain a purchaser category table;

[0026] Construct a supplier portrait table based on supplier and historical project information; the supplier portrait table includes supplier basic information, supplier historical winning project preferences, and supplier historical bidding project preferences; train a supplier clustering model based on the supplier portrait table; the supplier clustering model is used to output supplier categories; cluster the supplier categories based on the Kmeans clustering method to obtain supplier clustering results; calculate the distance from the cluster category to the cluster center based on the supplier clustering results, calculate the supplier basic score based on the distance to the cluster center, and obtain a supplier category table;

[0027] According to the purchaser category table, supplier category table and historical transaction project information, the correlation relationship and correlation degree between the categories are analyzed and learned from historical projects, and the first correlation relationship table is output to generate the first model.

[0028] According to some embodiments of the present invention, the initial model is trained in the supplier recommendation project application scenario and the project recommendation supplier application scenario based on the machine learning algorithm through the training set to obtain the second model, including:

[0029] Get the supplier category table;

[0030] Build a project portrait table based on the supplier's corresponding historical project information, which includes basic project information and the industry to which the project belongs; train a project classification model based on bid information, output a three-level industry classification, and output a project category table based on the project portrait table;

[0031] According to the supplier category table and the project category table, the correlation relationship and correlation degree between each subject category are analyzed and learned from the historical transaction projects, and the second correlation relationship table is output to generate the second model.

[0032] According to some embodiments of the present invention, a purchaser portrait table is constructed based on the purchaser and historical project information, including:

[0033] Perform word frequency statistics on the purchaser and historical project information to obtain a purchaser text information vocabulary list; generate a first feature vector about the purchaser with word position information and context association information based on the purchaser text information vocabulary list;

[0034] Perform project relationship analysis on the purchaser and historical project information to obtain multi-level project objects located at the purchaser's project neighboring nodes as the purchaser's second feature vector; the multi-level project objects include project duration, steps, and project communication times;

[0035] The first feature vector and the second feature vector are input into a fully connected layer, and an intermediate vector is output. The fully connected layer is used to learn parameters describing the corresponding relationship between the input first feature vector and the second feature vector and the purchaser label; the intermediate vector is input into n label classifiers in a multi-dimensional label classifier, and the label prediction probability of the purchaser in n dimensions is output, where n is a positive integer; each of the n label classifiers includes different logistic regression function layers, which are used to output purchaser labels of different dimensions;

[0036] Sort the predicted probabilities of labels on n dimensions in descending order; select the top k purchaser labels from the purchaser labels on n dimensions according to the sorting results;

[0037] Construct a purchaser profile table based on the first k purchaser labels.

[0038] To achieve the above-mentioned purpose, the second embodiment of the present invention proposes a project transaction push system based on artificial intelligence, including:

[0039] An acquisition module is used to acquire project transaction information through project information sources;

[0040] A preprocessing module is used to perform data preprocessing on project transaction information to obtain preprocessed data;

[0041] The training module is used to train the initial model based on the preprocessed data to obtain the project transaction push model;

[0042] The output module is used to obtain the data to be processed, input the data to be processed into the project transaction push model, and output the project transaction push information.

[0043] The present invention proposes a project transaction push method and system based on artificial intelligence. Through big data and artificial intelligence technology, a push algorithm model for public resource transaction projects and services is constructed. The public service platform completes the information push to market entities, so that various types of markets can understand market trends and grasp transaction opportunities. By deeply mining and analyzing the user's historical transaction data, browsing behavior and other information, the model can accurately identify the user's needs and preferences and realize accurate personalized push. The model adopts advanced deep learning algorithms and distributed computing technologies, which can efficiently process large-scale project transaction data and extract valuable information from it. The model has iterative update capabilities, and can adjust and optimize push strategies according to market changes and user needs to achieve accurate push. Through accurate push, users can find transaction projects that meet their needs more quickly, shorten the transaction cycle, and improve transaction efficiency. Because the pushed projects are more in line with user needs, the user's time and energy investment in screening and comparing projects is reduced, thereby reducing transaction costs. Through accurate push and personalized services, user satisfaction and loyalty are improved, which helps enterprises establish long-term and stable customer relationships.

[0044] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0045] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0047] Figure 1 is a flow chart of a project transaction push method based on artificial intelligence according to an embodiment of the present invention;

[0048] Figure 2 is a schematic diagram of obtaining a first model according to an embodiment of the present invention;

[0049] Figure 3 is a schematic diagram of outputting project transaction push information according to an embodiment of the present invention;

[0050] Figure 4 It is a block diagram of a project transaction push system based on artificial intelligence according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0052] like Figure 1 As shown, the first embodiment of the present invention proposes a project transaction push method based on artificial intelligence, including steps S1-S4:

[0053] S1. Obtain project transaction information through project information sources;

[0054] S2. Preprocess the project transaction information to obtain preprocessed data;

[0055] S3. Train the initial model based on the preprocessed data to obtain a project transaction push model;

[0056] S4. Obtain the data to be processed, input the data to be processed into the project transaction push model, and output the project transaction push information.

[0057] The working principle of the above technical solution is as follows: data collection, obtaining project transaction information through project information sources. Data preprocessing, preprocessing the project transaction information to obtain preprocessed data; model training, training the initial model based on the preprocessed data to obtain the project transaction push model; using the trained model for real-time or quasi-real-time accurate information push, obtaining the data to be processed, and inputting the data to be processed into the project transaction push model. The model will associate the projects that the enterprise users may be interested in based on the enterprise's historical transaction behavior, preferences, current market demand and other information, and generate a corresponding push list. Then, the push module will send this list to the public service platform or other relevant channels to push information to market entities.

[0058] The beneficial effects of the above technical solutions are as follows: through big data and artificial intelligence technology, a push algorithm model for public resource transaction projects and services is constructed, and the public service platform completes the information push to market entities, so that various types of markets can understand market trends and seize transaction opportunities. By deeply mining and analyzing users' historical transaction data, browsing behavior and other information, the model can accurately identify users' needs and preferences and achieve accurate personalized push. The model uses advanced deep learning algorithms and distributed computing technologies to efficiently process large-scale project transaction data and extract valuable information from it. The model has iterative update capabilities and can adjust and optimize push strategies according to market changes and user needs to achieve accurate push. Through accurate push, users can find transaction projects that meet their needs more quickly, shorten the transaction cycle, and improve transaction efficiency. Because the pushed projects are more in line with user needs, users reduce the time and energy invested in screening and comparing projects, thereby reducing transaction costs. Through accurate push and personalized services, user satisfaction and loyalty are improved, which helps companies establish long-term and stable customer relationships.

[0059] According to some embodiments of the present invention, the sources of project information include government procurement projects, construction projects and sunshine procurement projects. Sunshine procurement projects refer to the behavior of enterprises and institutions to obtain products or services from the supply market as their own resources in accordance with the principles of "openness, fairness and justice" and "quality first, price first". Project transaction information includes project name, project type, transaction method, transaction amount and project description.

[0060] According to some embodiments of the present invention, data preprocessing is performed on project transaction information, including:

[0061] Remove duplicate, invalid and irregularly formatted data from project transaction information to obtain clean data;

[0062] Standardize and normalize the cleaned data to obtain the target data;

[0063] Feature extraction is performed on the target data to determine project features; wherein the project features include the manufacturer role, the number of historical completed projects of the supplier, the supplier's project type preference, and the purchaser's project type preference.

[0064] The working principle and beneficial effects of the above technical solution are as follows: data preprocessing is performed on project transaction information to facilitate the removal of duplicate, invalid and irregularly formatted data in project transaction information, to achieve standardized processing and normalization, and finally to extract project features to facilitate project analysis and push.

[0065] According to some embodiments of the present invention, the initial model is trained according to the preprocessed data to obtain the project transaction push model, including:

[0066] Divide the preprocessed data into training set, validation set and test set;

[0067] The initial model is trained through the training set based on the machine learning algorithm, the validation set is used to adjust the model parameters to prevent overfitting, and the test set is used to evaluate the model performance. During the training process, the initial model will continuously iterate and optimize the parameters to eventually obtain the project transaction push model.

[0068] The working principle and beneficial effects of the above technical solution are as follows: The data set is divided into 70% training set, 15% validation set, and 15% test set. The initial model is trained through the training set based on the machine learning algorithm to let the model learn the patterns and relationships in the data. The validation set is used during the training process to adjust the model's hyperparameters (such as learning rate, number of iterations, network structure, etc.) to prevent the model from overfitting on the training set. The test set is used after the model training is completed to evaluate the final performance of the model to ensure that the model can perform well on unseen data. The machine learning algorithm includes the Kmeans algorithm. The model iteratively learns the data in the training set and continuously optimizes internal parameters (such as weights and biases) to minimize the loss function (such as cross entropy loss, mean square error, etc.). Monitor the model performance on the validation set. If the performance of the model on the validation set begins to decline, stop training in advance to prevent overfitting. Find the best combination of model hyperparameters through methods such as grid search, random search, or Bayesian optimization. When evaluating the performance of the model, determine the evaluation indicators, analyze the performance of the model on the test set, identify possible errors and deviations, and consider whether it is necessary to further adjust the model or collect more data. Build an efficient and accurate project transaction push model to provide users with personalized transaction recommendation services.

[0069] According to some embodiments of the present invention, it also includes: online learning or updating of the project transaction push model based on the user's tracking feedback data; the tracking feedback data includes click-through rate and conversion rate.

[0070] The beneficial effect of the above technical solution is: the project transaction push model is online learned or updated according to the feedback data of users on the project transaction push information, thereby improving the accuracy of the project transaction push model.

[0071] According to some embodiments of the present invention, training the initial model through a training set based on a machine learning algorithm includes:

[0072] Based on the machine learning algorithm, the initial model is trained through the training set in the application scenario of the purchaser recommending suppliers and the application scenario of the supplier recommending the purchaser to obtain the first model;

[0073] Based on the machine learning algorithm, the initial model is trained through the training set in the supplier recommendation project application scenario and the project recommendation supplier application scenario to obtain the second model;

[0074] A project transaction push model is determined according to the first model and the second model.

[0075] The working principle and beneficial effects of the above technical solution: The project transaction push model is divided into four scenarios: purchaser recommends supplier, supplier recommends purchaser, supplier recommends project, and project recommends supplier. Among them, in the scenario of purchaser recommends supplier and supplier recommends purchaser, the first model can be used. In the scenario of supplier recommends project and project recommends supplier, the second model is used, and the project transaction push model is determined according to the first model and the second model.

[0076] like Figure 2 As shown, according to some embodiments of the present invention, the initial model is trained in the application scenario of the purchaser recommending suppliers and the application scenario of the supplier recommending the purchaser based on the machine learning algorithm through the training set to obtain the first model, including:

[0077] In the model training stage, the enterprise entity information and historical project information are determined based on the training set. The bidding unit is recorded as the purchaser role, and the winning bidder is recorded as the supplier role. The purchaser and historical project information, and the supplier and historical project information are output respectively.

[0078] Construct a purchaser portrait table based on the purchaser and historical project information; the purchaser portrait table includes the purchaser's basic information and the purchaser's historical purchase project preferences; train a purchaser clustering model based on the purchaser portrait table; the purchaser clustering model is used to output purchaser categories; cluster the purchaser categories based on the Kmeans clustering method to obtain purchaser clustering results; calculate the distance from the cluster category to the cluster center based on the purchaser clustering results, calculate the purchaser's basic score based on the distance to the cluster center, and obtain a purchaser category table;

[0079] Construct a supplier portrait table based on supplier and historical project information; the supplier portrait table includes supplier basic information, supplier historical winning project preferences, and supplier historical bidding project preferences; train a supplier clustering model based on the supplier portrait table; the supplier clustering model is used to output supplier categories; cluster the supplier categories based on the Kmeans clustering method to obtain supplier clustering results; calculate the distance from the cluster category to the cluster center based on the supplier clustering results, calculate the supplier basic score based on the distance to the cluster center, and obtain a supplier category table;

[0080] According to the purchaser category table, supplier category table and historical transaction project information, the correlation relationship and correlation degree between the categories are analyzed and learned from historical projects, and the first correlation relationship table is output to generate the first model.

[0081] The working principle and beneficial effects of the above technical solution: The basic information of the purchaser includes the company name, industry, scale, etc.; the purchaser's historical procurement project preferences include the purchased product type, price range, purchase frequency, etc., which are used to reflect the purchaser's procurement habits and preferences.

[0082] The purchaser clustering model is trained according to the purchaser portrait table, including: selecting features useful for clustering from the purchaser portrait table, such as the purchaser's basic information and the purchaser's historical purchase project preferences. Based on the K-means clustering algorithm, the initial cluster center is selected. The cluster division is continuously optimized through an iterative process to minimize intra-cluster differences (such as intra-cluster variance) and maximize inter-cluster differences. In the K-means algorithm, each purchaser is assigned to the nearest cluster center, and the cluster center of each cluster is recalculated. (Analyze the characteristics of each cluster, name each cluster or describe the type of purchaser it represents). After each iteration, check whether the clustering results converge (that is, whether the change in the cluster center is less than a certain threshold) or whether the preset number of iterations has been reached. After confirming convergence, the purchaser clustering model is obtained.

[0083] Calculate the basic score of the purchaser, including: Assuming there is clusters, each containing Purchasers, of which The set of buyers for a cluster is , each purchaser dimensional feature vector, then the The cluster centers of the clusters It can be calculated as:

[0084]

[0085] in, It is a cluster The feature vector of a purchaser in .

[0086] For clusters Each buyer in , calculate it to the cluster center This distance is the Euclidean distance.

[0087] Move the purchaser to the cluster center The distance is normalized or reversed (because the smaller the distance, the higher the similarity between the purchaser and the cluster center), and then mapped to a score range.

[0088]

[0089] in, and Cluster The minimum and maximum distances from all purchasers to the cluster center. Cluster The distance from any purchaser x to the cluster center. is the basic score of purchaser x.

[0090] The method for training the supplier clustering model is the same as that for training the purchaser clustering model, so it will not be repeated here.

[0091] The method for calculating the supplier's basic score is the same as that for calculating the purchaser's basic score, so it will not be repeated here.

[0092] Purchaser category table: contains the purchaser's unique identifier (such as ID) and the category to which it belongs.

[0093] Supplier category table: contains the supplier's unique identifier (such as ID) and the category to which it belongs.

[0094] Historical transaction project information: including detailed information of each transaction, such as purchaser ID, supplier ID, transaction time, transaction amount, etc.

[0095] Based on the historical transaction project information, the number of co-occurrences between each purchaser category and each supplier category is calculated. That is, the number of transactions each purchaser category and each supplier category have participated in together is counted. The rows of the matrix represent the purchaser category, the columns represent the supplier category, and the element value represents the number of co-occurrences. A variety of statistical or machine learning methods are used to quantify the degree of association, such as Pearson correlation coefficient, mutual information, chi-square test, etc. Here, point mutual information (PMI) is selected as an indicator of the degree of association. The correlation value is normalized to between 0 and 1, and the calculated correlation value is organized into a table form, namely the first correlation table. The table should include the purchaser category, the supplier category and the correlation value between them. Based on the first correlation table, a query model (first model) can be constructed to allow users to enter a specific purchaser category or supplier category and return a list of categories with the highest correlation. It is easy to obtain an accurate first model, which is suitable for the application scenario of purchasers recommending suppliers and suppliers recommending purchasers.

[0096] like Figure 3 As shown, the data to be processed is obtained, the data to be processed is input into the project transaction push model, and the project transaction push information is output, including:

[0097] Obtain increments through the time dimension, that is, add new enterprise entity information, that is, data to be processed;

[0098] The newly added bidding unit is the newly added purchaser, and the newly added winning bidder is the newly added supplier;

[0099] Generate a new purchaser profile table and a new supplier profile table based on the newly added purchasers and suppliers, call the trained purchaser clustering model and supplier clustering model to label the newly added purchasers and suppliers, and write the label results into the purchaser category table and supplier category table. This facilitates the update of the purchaser category table and supplier category table, and also facilitates the determination of project transaction push information.

[0100] According to some embodiments of the present invention, the initial model is trained in the supplier recommendation project application scenario and the project recommendation supplier application scenario based on the machine learning algorithm through the training set to obtain the second model, including:

[0101] Get the supplier category table;

[0102] Build a project portrait table based on the supplier's corresponding historical project information, which includes basic project information and the industry to which the project belongs; train a project classification model based on bid information, output a three-level industry classification, and output a project category table based on the project portrait table;

[0103] According to the supplier category table and the project category table, the correlation relationship and correlation degree between each subject category are analyzed and learned from the historical transaction projects, and the second correlation relationship table is output to generate the second model.

[0104] The working principle and beneficial effects of the above technical solution: Bidding information is the relevant information of bidding activities, including bidding announcements, bidding documents, winning results, etc. First, determine the supplier category table, and build a project portrait table based on the historical project information corresponding to the supplier, including: collecting detailed data of all historical projects in which the supplier participated, including project name, project type, project scale, project duration, project location, project budget, project team composition, and project results; ensuring that each project is associated with the correct supplier, and collecting basic information of the supplier, such as company name, industry field, main business scope, past performance, etc. Extract key features from historical project data, such as project type (such as construction engineering, software development, consulting services, etc.), project scale (such as investment amount, number of personnel, etc.), project duration (time from start to end), project success rate (proportion of successfully completed projects), project complexity (comprehensive evaluation based on project type, scale, duration, etc.). Also take into account the project-related features in the supplier information, such as the supplier's experience in the project type (the number of similar projects participated in), performance under the project scale (success rate, cost control ability, etc.), cooperation with the project team (communication efficiency, problem-solving ability, etc.), etc. Quantify these features so that they can be expressed in numerical or categorical form for subsequent analysis. Based on the extracted features, design a framework for project portraits, including but not limited to dimensions such as basic project information, project features, supplier performance, risks and opportunities. Fill each historical project according to the portrait framework to form a project portrait table. The table should contain the key information of the project, feature descriptions, and evaluation results based on supplier performance. Train the project classification model based on the bidding information, output a three-level industry classification, and output a project category table in combination with the project portrait table, specifically including: collecting bidding information data from government websites, corporate websites, third-party bidding information platforms, etc. The data contains key information such as project name, project description, bidding unit, budget, industry labels, etc. Select a machine learning model suitable for multi-classification tasks, such as random forest, gradient boosting tree (GBDT), multi-layer perceptron (MLP), and convolutional neural network (CNN, for text data). Divide the data set into training set, validation set, and test set. Use the training set to train the model, and adjust the model parameters on the validation set through cross-validation. Monitor the performance of the model on the test set to evaluate its generalization ability. Use the trained model to predict the bid information data and output the three-level industry classification for each project. The project portrait table includes the project's geographical location, scale, technical requirements, target user groups, etc. Based on the three-level industry classification output by the model, combined with the project portrait table, further refine and enrich the project categories. Design the table structure, including fields such as project ID, project name, three-level industry classification, geographical location, project scale, technical requirements, target user groups, etc. Fill in the table and generate a detailed category description for each project to accurately describe the project category table.

[0105] The historical transaction project data includes the supplier ID, project ID, supplier category, project category and other possible related information (such as transaction amount, transaction time, etc.).

[0106] Supplier Category Table: Ensure that each supplier ID has a corresponding supplier category.

[0107] Project category table: Ensure that each project ID has a corresponding project category.

[0108] Construct a two-dimensional matrix with supplier categories as rows and project categories as columns. Each element in the matrix represents the number of associations or the strength of associations between the corresponding supplier category and project category (such as the sum of transaction amounts, the number of transaction projects, etc.). Use the number of transaction projects as a direct indicator of the strength of association, or use the sum of transaction amounts for weighted calculation. Convert the association matrix into a tabular form, i.e., the second association table. The table should contain supplier categories, project categories, and the strength of associations between them. Based on the second association table, use graph neural networks (GNN) or network analysis methods to build a model to generate a second model, which is suitable for supplier recommendation project application scenarios and project recommendation supplier application scenarios.

[0109] In one embodiment, obtaining the data to be processed, inputting the data to be processed into the project transaction push model, and outputting the project transaction push information include:

[0110] In the scenario of supplier recommendation project, the increment is obtained through the time dimension, that is, the new project information, that is, the data to be processed.

[0111] Based on the newly added project information, the trained project classification model is called to label the newly added project information separately, and the label results are written into the valid project list to facilitate the output of project transaction push information.

[0112] In the scenario of project supplier recommendation, the increment is obtained through the time dimension, that is, the new project information is added.

[0113] Based on the newly added project information, the list of related suppliers is obtained through the project category and association relationship, and the recommended score is calculated by the supplier's basic score and association score (calculate the sum of the two), and finally the recommendation result is obtained.

[0114] According to some embodiments of the present invention, a purchaser portrait table is constructed based on the purchaser and historical project information, including:

[0115] Perform word frequency statistics on the purchaser and historical project information to obtain a purchaser text information vocabulary list; generate a first feature vector about the purchaser with word position information and context association information based on the purchaser text information vocabulary list;

[0116] Perform project relationship analysis on the purchaser and historical project information to obtain multi-level project objects located at the purchaser's project neighboring nodes as the purchaser's second feature vector; the multi-level project objects include project duration, steps, and project communication times;

[0117] The first feature vector and the second feature vector are input into a fully connected layer, and an intermediate vector is output. The fully connected layer is used to learn parameters describing the corresponding relationship between the input first feature vector and the second feature vector and the purchaser label; the intermediate vector is input into n label classifiers in a multi-dimensional label classifier, and the label prediction probability of the purchaser in n dimensions is output, where n is a positive integer; each of the n label classifiers includes different logistic regression function layers, which are used to output purchaser labels of different dimensions;

[0118] Sort the predicted probabilities of labels on n dimensions in descending order; select the top k purchaser labels from the purchaser labels on n dimensions according to the sorting results;

[0119] Construct a purchaser profile table based on the first k purchaser labels.

[0120] The working principle of the above technical solution is: conduct a comprehensive text analysis of the purchaser and historical project information, including project descriptions, procurement requirements, communication records, etc., perform word frequency statistics, and form a vocabulary list of the purchaser's text information, which helps to identify the keywords and topics that the purchaser is concerned about. Use natural language processing (NLP) technology, such as TF-IDF, Word2Vec, BERT and other models to capture the semantics and position information of the vocabulary and generate the first feature vector about the purchaser.

[0121] Analyze the historical projects that the purchaser has participated in, identify the project neighbor nodes that are closely related to the purchaser (such as suppliers, project types, partners, etc.), and obtain the multi-level characteristics of these projects, including project duration, steps (such as project establishment, bidding, implementation, acceptance, etc.) and project communication times. Convert the information of multi-level project objects into numerical features to form the second feature vector. These numerical features can quantify the behavior patterns and preferences of purchasers in projects.

[0122] The first feature vector and the second feature vector are input into the fully connected layer. The fully connected layer outputs an intermediate vector by learning the complex relationship between the input feature vector and the purchaser label, which integrates the information of both text and project relationship. The intermediate vector is input into the n label classifiers in the multi-dimensional label classifier. Each label classifier corresponds to a dimension (such as purchasing preference, cooperation attitude, project management ability, etc.) and contains a logistic regression function layer to output the label prediction probability on the dimension. The label prediction probabilities on the n dimensions are sorted, and the top k labels with the highest probability are selected as the main features of the purchaser according to the sorting results. The selection of k can be adjusted according to actual needs and scenarios. Based on the selected top k purchaser labels, a purchaser portrait table is constructed. The portrait table shows the characteristics and preferences of the purchaser in different dimensions in an intuitive way. Selecting the top k purchaser labels facilitates the extraction of effective information for constructing the purchaser portrait table, which is convenient for improving the efficiency and accuracy of constructing the purchaser portrait table.

[0123] The beneficial effects of the above technical solution are: taking into account text information and project relationships, a comprehensive and multi-dimensional portrait is constructed for the purchaser.

[0124] In one embodiment, the method of constructing a supplier portrait table based on supplier and historical project information is the same as the method of constructing a purchaser portrait table, which will not be repeated.

[0125] like Figure 4 As shown, the second embodiment of the present invention proposes an artificial intelligence-based project transaction push system, including:

[0126] An acquisition module is used to acquire project transaction information through project information sources;

[0127] A preprocessing module is used to perform data preprocessing on project transaction information to obtain preprocessed data;

[0128] The training module is used to train the initial model based on the preprocessed data to obtain the project transaction push model;

[0129] The output module is used to obtain the data to be processed, input the data to be processed into the project transaction push model, and output the project transaction push information.

[0130] The beneficial effects of the above technical solutions are as follows: through big data and artificial intelligence technology, a push algorithm model for public resource transaction projects and services is constructed, and the public service platform completes the information push to market entities, so that various types of markets can understand market trends and seize transaction opportunities. By deeply mining and analyzing users' historical transaction data, browsing behavior and other information, the model can accurately identify users' needs and preferences and achieve accurate personalized push. The model uses advanced deep learning algorithms and distributed computing technologies to efficiently process large-scale project transaction data and extract valuable information from it. The model has iterative update capabilities and can adjust and optimize push strategies according to market changes and user needs to achieve accurate push. Through accurate push, users can find transaction projects that meet their needs more quickly, shorten the transaction cycle, and improve transaction efficiency. Because the pushed projects are more in line with user needs, users reduce the time and energy invested in screening and comparing projects, thereby reducing transaction costs. Through accurate push and personalized services, user satisfaction and loyalty are improved, which helps companies establish long-term and stable customer relationships.

[0131] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A project transaction push method based on artificial intelligence, characterized in that: include: Obtain project transaction information through project information sources; Perform data preprocessing on project transaction information to obtain preprocessed data; The initial model is trained based on the preprocessed data to obtain the project transaction push model; Obtain the data to be processed, input the data to be processed into the project transaction push model, and output the project transaction push information; The initial model is trained based on the preprocessed data to obtain the project transaction push model, including: Divide the preprocessed data into training set, validation set and test set; The initial model is trained through the training set based on the machine learning algorithm. The validation set is used to adjust the model parameters to prevent overfitting, and the test set is used to evaluate the model performance. During the training process, the initial model will continuously iterate and optimize the parameters, and finally obtain the project transaction push model. The initial model is trained using a training set based on a machine learning algorithm, including: Based on the machine learning algorithm, the initial model is trained through the training set in the application scenario of the purchaser recommending suppliers and the application scenario of the supplier recommending the purchaser to obtain the first model; Based on the machine learning algorithm, the initial model is trained through the training set in the supplier recommendation project application scenario and the project recommendation supplier application scenario to obtain the second model; Determine a project transaction push model according to the first model and the second model; Based on the machine learning algorithm, the initial model is trained through the training set in the application scenario of the purchaser recommending suppliers and the application scenario of the supplier recommending the purchaser, and the first model is obtained, including: In the model training stage, the enterprise entity information and historical project information are determined based on the training set. The bidding unit is recorded as the purchaser role, and the winning bidder is recorded as the supplier role. The purchaser and historical project information, and the supplier and historical project information are output respectively. Construct a purchaser portrait table based on the purchaser and historical project information; the purchaser portrait table includes the purchaser's basic information and the purchaser's historical purchase project preferences; train a purchaser clustering model based on the purchaser portrait table; the purchaser clustering model is used to output purchaser categories; cluster the purchaser categories based on the Kmeans clustering method to obtain purchaser clustering results; calculate the distance from the cluster category to the cluster center based on the purchaser clustering results, calculate the purchaser's basic score based on the distance to the cluster center, and obtain a purchaser category table; Construct a supplier portrait table based on supplier and historical project information; the supplier portrait table includes supplier basic information, supplier historical winning project preferences, and supplier historical bidding project preferences; train a supplier clustering model based on the supplier portrait table; the supplier clustering model is used to output supplier categories; cluster the supplier categories based on the Kmeans clustering method to obtain supplier clustering results; calculate the distance from the cluster category to the cluster center based on the supplier clustering results, calculate the supplier basic score based on the distance to the cluster center, and obtain a supplier category table; According to the purchaser category table, supplier category table and historical transaction project information, the correlation relationship and correlation degree between the categories are analyzed and learned from historical projects, and the first correlation relationship table is output to generate the first model.

2. The project transaction push method based on artificial intelligence as claimed in claim 1, characterized in that: Sources of project information include government procurement projects, construction projects and open procurement projects; project transaction information includes project name, project type, transaction method, transaction amount and project description.

3. The project transaction push method based on artificial intelligence as claimed in claim 1, characterized in that: Data preprocessing of project transaction information, including: Remove duplicate, invalid and irregularly formatted data from project transaction information to obtain clean data; Standardize and normalize the cleaned data to obtain the target data; Feature extraction is performed on the target data to determine project features; wherein the project features include the manufacturer role, the number of historical completed projects of the supplier, the supplier's project type preference, and the purchaser's project type preference.

4. The project transaction push method based on artificial intelligence as claimed in claim 1, characterized in that: Also includes: The project transaction push model is learned or updated online based on the user's tracking feedback data; the tracking feedback data includes click-through rate and conversion rate.

5. The project transaction push method based on artificial intelligence as claimed in claim 1, characterized in that: Based on the machine learning algorithm, the initial model is trained in the supplier recommendation project application scenario and the project recommendation supplier application scenario through the training set to obtain the second model, including: Get the supplier category table; Build a project portrait table based on the supplier's corresponding historical project information, which includes basic project information and the industry to which the project belongs; train a project classification model based on bid information, output a three-level industry classification, and output a project category table based on the project portrait table; According to the supplier category table and the project category table, the correlation relationship and correlation degree between each subject category are analyzed and learned from the historical transaction projects, and the second correlation relationship table is output to generate the second model.

6. The project transaction push method based on artificial intelligence as claimed in claim 1, characterized in that: Construct a purchaser profile table based on the purchaser and historical project information, including: Perform word frequency statistics on the purchaser and historical project information to obtain a purchaser text information vocabulary list; generate a first feature vector about the purchaser with word position information and context association information based on the purchaser text information vocabulary list; Perform project relationship analysis on the purchaser and historical project information to obtain multi-level project objects located at the purchaser's project neighboring nodes as the purchaser's second feature vector; the multi-level project objects include project duration, steps, and project communication times; The first feature vector and the second feature vector are input into a fully connected layer, and an intermediate vector is output. The fully connected layer is used to learn parameters describing the corresponding relationship between the input first feature vector and the second feature vector and the purchaser label; the intermediate vector is input into n label classifiers in a multi-dimensional label classifier, and the label prediction probability of the purchaser in n dimensions is output, where n is a positive integer; each of the n label classifiers includes different logistic regression function layers, which are used to output purchaser labels of different dimensions; Sort the predicted probabilities of labels on n dimensions in descending order; select the top k purchaser labels from the purchaser labels on n dimensions according to the sorting results; Construct a purchaser profile table based on the first k purchaser labels.

7. An artificial intelligence-based project transaction push system, characterized in that: include: An acquisition module is used to acquire project transaction information through project information sources; A preprocessing module is used to perform data preprocessing on project transaction information to obtain preprocessed data; The training module is used to train the initial model based on the preprocessed data to obtain the project transaction push model; The output module is used to obtain the data to be processed, input the data to be processed into the project transaction push model, and output the project transaction push information; The initial model is trained based on the preprocessed data to obtain the project transaction push model, including: Divide the preprocessed data into training set, validation set and test set; The initial model is trained through the training set based on the machine learning algorithm. The validation set is used to adjust the model parameters to prevent overfitting, and the test set is used to evaluate the model performance. During the training process, the initial model will continuously iterate and optimize the parameters, and finally obtain the project transaction push model. The initial model is trained using a training set based on a machine learning algorithm, including: Based on the machine learning algorithm, the initial model is trained through the training set in the application scenario of the purchaser recommending suppliers and the application scenario of the supplier recommending the purchaser to obtain the first model; Based on the machine learning algorithm, the initial model is trained through the training set in the supplier recommendation project application scenario and the project recommendation supplier application scenario to obtain the second model; Determine a project transaction push model according to the first model and the second model; Based on the machine learning algorithm, the initial model is trained through the training set in the application scenario of the purchaser recommending suppliers and the application scenario of the supplier recommending the purchaser, and the first model is obtained, including: In the model training stage, the enterprise entity information and historical project information are determined based on the training set. The bidding unit is recorded as the purchaser role, and the winning bidder is recorded as the supplier role. The purchaser and historical project information, and the supplier and historical project information are output respectively. Construct a purchaser portrait table based on the purchaser and historical project information; the purchaser portrait table includes the purchaser's basic information and the purchaser's historical purchase project preferences; train a purchaser clustering model based on the purchaser portrait table; the purchaser clustering model is used to output purchaser categories; cluster the purchaser categories based on the Kmeans clustering method to obtain purchaser clustering results; calculate the distance from the cluster category to the cluster center based on the purchaser clustering results, calculate the purchaser's basic score based on the distance to the cluster center, and obtain a purchaser category table; Construct a supplier portrait table based on supplier and historical project information; the supplier portrait table includes supplier basic information, supplier historical winning project preferences, and supplier historical bidding project preferences; train a supplier clustering model based on the supplier portrait table; the supplier clustering model is used to output supplier categories; cluster the supplier categories based on the Kmeans clustering method to obtain supplier clustering results; calculate the distance from the cluster category to the cluster center based on the supplier clustering results, calculate the supplier basic score based on the distance to the cluster center, and obtain a supplier category table; According to the purchaser category table, supplier category table and historical transaction project information, the correlation relationship and correlation degree between the categories are analyzed and learned from historical projects, and the first correlation relationship table is output to generate the first model.

Images