Project application and enterprise information interaction method and system based on artificial intelligence

Abstract

The present application relates to the field of data processing technology, and discloses a method and system for project application and enterprise information interaction based on artificial intelligence. The method includes: obtaining historical project application data from multiple target data sources, and preprocessing and data fusion of the historical project application data to obtain a fused enterprise project data set; performing semantic feature extraction and text entity recognition on the fused enterprise project data set through a Transformer model to obtain entity and attribute relationship data; constructing a knowledge graph for entity and attribute relationship data based on a graph attention network and an RDF graph model to obtain an enterprise-project matching relationship graph; receiving a target query request sent by a user terminal, and querying the target query request and the enterprise-project matching relationship graph based on a knowledge graph question-answering model of a deformed graph matching to obtain a target query result, and then using an artificial intelligence algorithm to improve the accuracy of project application and enterprise information interaction matching and query.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method and system for project application and enterprise information interaction based on artificial intelligence. Background Art

[0002] Enterprises demonstrate their technical capabilities and potential for project implementation through project applications. However, due to the uneven distribution of resources, many enterprises often find it difficult to quickly and accurately find projects that are highly matched to their needs. Similarly, it is difficult for project parties to efficiently screen out the most suitable applicants. This mismatch not only reduces the efficiency of resource allocation, but also increases the time and economic costs of enterprises in finding suitable projects. Therefore, how to effectively use artificial intelligence technology to improve the matching between projects and enterprises and achieve efficient allocation of resources is still a problem that needs to be solved.

[0003] Traditional manual matching methods can no longer meet the rapidly growing market demand. There is an urgent need for an intelligent system that can automatically process large-scale data and provide accurate matching suggestions. Summary of the invention

[0004] The present application provides a method and system for project application and enterprise information interaction based on artificial intelligence, and further adopts artificial intelligence algorithm to improve the accuracy of project application and enterprise information interaction matching and query.

[0005] The first aspect of the present application provides a method for project application and enterprise information interaction based on artificial intelligence, and the method for project application and enterprise information interaction based on artificial intelligence includes:

[0006] Acquire historical project application data from multiple target data sources, and perform preprocessing and data fusion on the historical project application data to obtain a fused enterprise project data set;

[0007] Performing semantic feature extraction and text entity recognition on the fused enterprise project data set through a Transformer model to obtain entity and attribute relationship data;

[0008] Based on the graph attention network and RDF graph model, a knowledge graph is constructed for the entity and attribute relationship data to obtain an enterprise-project matching relationship graph;

[0009] A target query request sent by a user terminal is received, and the target query request and the enterprise-project matching relationship graph are queried based on a knowledge graph question-answering model of deformation graph matching to obtain a target query result.

[0010] The second aspect of the present application provides a project application and enterprise information interaction system based on artificial intelligence, and the project application and enterprise information interaction system based on artificial intelligence includes:

[0011] An acquisition module is used to acquire historical project application data from multiple target data sources, and to preprocess and fuse the historical project application data to obtain a fused enterprise project data set;

[0012] A recognition module, used for performing semantic feature extraction and text entity recognition on the fused enterprise project data set through a Transformer model to obtain entity and attribute relationship data;

[0013] A construction module is used to construct a knowledge graph for the entity and attribute relationship data based on a graph attention network and an RDF graph model to obtain an enterprise-project matching relationship graph;

[0014] The query module is used to receive a target query request sent by a user terminal, and query the target query request and the enterprise-project matching relationship graph based on the knowledge graph question-answering model of deformation graph matching to obtain a target query result.

[0015] The third aspect of the present application provides a computer device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the computer device executes the above-mentioned artificial intelligence-based project declaration and enterprise information interaction method.

[0016] The fourth aspect of the present application provides a computer-readable storage medium, which stores instructions. When the computer-readable storage medium is run on a computer, it enables the computer to execute the above-mentioned artificial intelligence-based project declaration and enterprise information interaction method.

[0017] In the technical solution provided by the present application, by automatically acquiring historical project declaration data from multiple target data sources, and preprocessing and data fusion of these data, the scattered data sources can be efficiently integrated to generate a fusion enterprise project data set. This efficient data processing capability significantly reduces the time and effort of manual data collation, and improves the efficiency and accuracy of data processing. By applying the advanced Transformer model to extract semantic features and identify text entities on the fusion enterprise project data set, the text content can be deeply understood, and key entities and attributes and their relationships can be identified. This deep semantic understanding provides a solid foundation for accurately matching enterprises and projects, and improves the accuracy and relevance of matching. Based on the graph attention network and RDF graph model, not only can a rich enterprise-project matching relationship map be constructed, but also the map can be dynamically updated according to real-time data. This dynamic knowledge graph construction and application mechanism makes the matching relationship between enterprises and projects more accurate and timely, and improves the flexibility and effectiveness of resource allocation. The knowledge graph question-answering model using deformed graph matching can accurately respond to the user's target query request and quickly and accurately provide the target query results. This intelligent query response mechanism greatly improves the user experience, allowing users to easily obtain the required matching information, thereby improving the efficiency and quality of decision-making. By defining the matching logic of enterprise features to projects and project features to enterprises, a two-way matching logic is realized, which enhances the interactivity and comprehensiveness of matching between enterprises and projects. This enhanced interactive logic can not only meet the needs of different users more comprehensively, but also promote the two-way flow and optimal allocation of resources, and then use artificial intelligence algorithms to improve the accuracy of interactive matching and query between project applications and enterprise information. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a schematic diagram of an embodiment of a method for project declaration and enterprise information interaction based on artificial intelligence in an embodiment of the present application;

[0019] Figure 2 This is a schematic diagram of an embodiment of a project declaration and enterprise information interaction system based on artificial intelligence in an embodiment of the present application. DETAILED DESCRIPTION

[0020] The embodiments of the present application provide a method and system for project application and enterprise information interaction based on artificial intelligence, thereby using artificial intelligence algorithms to improve the accuracy of project application and enterprise information interaction matching and query.

[0021] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0022] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the method for project application and enterprise information interaction based on artificial intelligence includes:

[0023] Step 101: Acquire historical project application data from multiple target data sources, and perform preprocessing and data fusion on the historical project application data to obtain a fused enterprise project data set;

[0024] It is understandable that the execution subject of the present application can be a project declaration and enterprise information interaction system based on artificial intelligence, or a terminal or a server, which is not limited here. The present application embodiment takes the server as the execution subject as an example for explanation.

[0025] Specifically, multiple target data sources are obtained, including project public websites, third-party service agencies, and enterprise management platforms, covering public project information released by the government, professional service data in the industry, and the project management records of the enterprise itself. To collect historical data from diversified target data sources, the system needs to have efficient data capture capabilities and compatible processing capabilities for different data formats, so as to ensure that the historical project application data collected from various data sources are complete. These data are formatted, including converting the data into a unified format, structured processing, and necessary data cleaning work to standardize the data set for subsequent processing and analysis. In order to improve the quality and accuracy of the data, the formatted project application data of each target data source is deduplicated to eliminate duplicate records that may occur during the data collection process and ensure the uniqueness and accuracy of the data. The deduplicated project application data of each target data source are fused to merge the project application data sets from different sources into a unified fused enterprise project data set. Through data fusion technology, the same or similar data entries in different data sources are matched and merged according to specific rules and logic to obtain a fused enterprise project data set.

[0026] Step 102: perform semantic feature extraction and text entity recognition on the fused enterprise project dataset through a Transformer model to obtain entity and attribute relationship data;

[0027] Specifically, fine-grained text classification is performed on the fusion enterprise project dataset, and the text data in the huge dataset is effectively classified according to its content features, and decomposed into text paragraph data that is easy to manage and analyze. Based on the sequence-to-sequence model, key information sentences are extracted from the classified text paragraph data to form a set of key information sentences. Semantic features are extracted from the set of key information sentences through the Transformer model. The Transformer model includes a full-layer normalization layer, a weighted residual connection layer, a reinforcement learning position encoding layer, and a zero-masking self-attention mechanism layer. These components work together to capture the deep semantic relationship within the sentence and can effectively handle long-distance dependency problems. Through this model, the text data is converted into a high-quality semantic feature set. Entity recognition is performed on the fusion enterprise project dataset based on the semantic feature set. Through pattern recognition and data matching technology, specific entities and their related attribute data are identified from the text. Entity relationship analysis and relationship mapping are performed on text entities and attribute data, and an accurate relationship model is established between entities. The potential connection between entities is obtained through logical reasoning to form a comprehensive and accurate entity and attribute relationship dataset.

[0028] Furthermore, the key information sentence set is input into the full-layer normalization layer for preliminary semantic understanding and data standardization, and a standardized semantic data set is obtained to ensure that the data processed subsequently reaches the standardized level in format and scope. The standardized semantic data set is input into the weighted residual connection layer for feature enhancement. By using residual connection, this layer can effectively solve the gradient vanishing problem that may occur in deep networks, and at the same time highlight important features through the weighting mechanism to ensure that the model can capture more in-depth semantic information. The enhanced semantic feature data is encoded by the reinforcement learning position encoding layer. In natural language processing, the position information of words or sentence components is crucial to understanding the overall semantics. The reinforcement learning position encoding layer effectively incorporates the position information into the model through the learned position encoding, so that the model can understand the semantic role of the element according to its position in the sentence. The position-encoded data is deeply semantically understood and feature extracted based on the zero-masking self-attention mechanism layer. The self-attention mechanism enables the model to consider the relationship with all other elements when processing each word or sentence component, thereby capturing the complex dependencies and deep semantic connections in the text, and finally obtaining a semantic feature set.

[0029] Step 103: construct a knowledge graph for the entity and attribute relationship data based on the graph attention network and the RDF graph model to obtain an enterprise-project matching relationship graph;

[0030] It should be noted that two core matching logics are defined: one is the enterprise feature to project matching logic, and the other is the project feature to enterprise matching logic. The enterprise feature to project matching logic aims to query the project feature information that matches the enterprise based on the specific feature information of the enterprise, while the project feature to enterprise matching logic seeks the enterprise feature information that matches the project based on the specific feature information of the project. Based on the defined matching logic, the knowledge graph is initialized for the entity and attribute relationship data, and the initial knowledge graphs from enterprise to project and from project to enterprise are obtained respectively. The entity and attribute relationship data are preliminarily organized and linked according to the defined matching logic, corresponding nodes are created for each enterprise and project, and links are established according to the matching logic between them, so as to ensure that the knowledge graph can reflect the actual relationship between enterprises and projects in the initialization stage. The initial knowledge graphs from enterprise to project and from project to enterprise are fused through the graph attention network. The graph attention network can identify and emphasize the most important nodes and edges in the graph through its attention mechanism, thereby optimizing the entity relationship representation in the knowledge graph. The graph attention network effectively fuses the two initial knowledge graphs into a more comprehensive and accurate enterprise and project fusion knowledge graph by analyzing and evaluating the relationship strength between entities in the graph. The enterprise and project fusion knowledge graph is structured and dynamically updated through the RDF (Resource Description Framework) graph model. The RDF graph model is a model used to describe Web resources. It represents data in the form of subject-predicate-object triples, making the structured representation of the knowledge graph clearer and more standardized. The RDF graph model is used to update the graph in real time according to new data or information to ensure that the knowledge graph can continuously reflect the latest matching relationship between enterprises and projects, and obtain the enterprise-project matching relationship graph.

[0031] Among them, node feature extraction is performed on the initial knowledge graph from enterprise to project and the initial knowledge graph from project to enterprise respectively, and the feature information of specific enterprises and projects is extracted from the two initial knowledge graphs. By analyzing the initial knowledge graph from enterprise to project, the feature set of each enterprise node is obtained; by analyzing the initial knowledge graph from project to enterprise, the feature set of each project node is obtained. Edge feature extraction is performed on the two initial knowledge graphs to identify and extract the edge features between enterprise to project and project to enterprise. These edge features include various attributes of the relationship between enterprise and project, such as cooperation frequency, project success rate, etc. Through edge feature extraction, the edge feature set from enterprise to project and the edge feature set from project to enterprise are obtained. Similarity calculation is performed on the feature sets of enterprise nodes and project nodes. By analyzing the similarity between different node features, a connection logic based on content similarity is provided for subsequent fusion. At the same time, weight calculation is performed on the edge feature set from enterprise to project and the edge feature set from project to enterprise to evaluate the importance of these edges and the degree of influence on the entire graph structure, and a weighted edge feature set is obtained. The node feature similarity data and the weighted edge feature set are processed by the graph attention network. The attention mechanism can effectively distribute the importance of different nodes and edges, and specifically emphasize the features and relationships that are most critical to the matching of enterprises and projects, thus obtaining the weighted graph of attention from enterprises to projects and the weighted graph of attention from projects to enterprises. The two attention-weighted graphs are fused into a comprehensive enterprise and project fusion knowledge graph. The fusion process considers the similarity between nodes and the weight of edges, and optimizes the structure of the entire graph through the attention mechanism, ensuring that the matching relationship between enterprises and projects is accurately and effectively reflected in the fusion knowledge graph, and finally obtaining the enterprise and project fusion knowledge graph.

[0032] Step 104: Receive a target query request sent by a user terminal, and query the target query request and the enterprise-project matching relationship graph based on a knowledge graph question-answering model of deformation graph matching to obtain a target query result.

[0033] Specifically, the target query request sent by the user terminal is received and the request is semantically parsed. The user's natural language query request is converted into structured query conditions to ensure the accuracy and efficiency of subsequent processing. Semantic parsing includes in-depth understanding of the actual intent and needs of the user's query, and decomposing it into specific conditions that the system can understand and process. Keyword extraction is performed on the structured query conditions. A set of keywords that can represent the core content of the query conditions is selected, and these keywords will serve as the main basis for the subsequent query process. The extracted query keyword set directly affects the recognition accuracy of the query intent and the relevance of the query results. Query intent recognition is performed based on the keyword set to understand the user's query target and the needs behind it. Query intent recognition ensures that the user's actual query purpose can be accurately grasped by analyzing the semantics and context information of the keywords. Based on the knowledge graph question-answering model of deformed graph matching, a detailed query is performed on the enterprise-project matching relationship graph according to the identified query intent. Using the entity relationship and structured data in the knowledge graph, the deformed graph matching technology is used to accurately match the user's query needs and generate multiple initial query results. The correlation between each initial query result and the target query intent is calculated to obtain the matching degree of each result with the user's query intent. The level of relevance index directly reflects the accuracy of the query results and the degree to which the user's needs are met. Based on the relevance index, all initial query results are sorted and analyzed. The sorting process takes into account the comprehensiveness and relevance of the query results to ensure that the optimal options that best match the query intent are obtained.

[0034] In the embodiment of the present application, by automatically acquiring historical project declaration data of multiple target data sources, and preprocessing and data fusion of these data, it is possible to efficiently integrate scattered data sources and generate a fusion enterprise project data set. This efficient data processing capability significantly reduces the time and effort of manual data collation, and improves the efficiency and accuracy of data processing. By applying the advanced Transformer model to extract semantic features and identify text entities on the fusion enterprise project data set, the text content can be deeply understood, and key entities and attributes and their relationships can be identified. This deep semantic understanding provides a solid foundation for accurately matching enterprises and projects, and improves the accuracy and relevance of matching. Based on the graph attention network and RDF graph model, not only can a rich enterprise-project matching relationship map be constructed, but also the map can be dynamically updated according to real-time data. This dynamic knowledge graph construction and application mechanism makes the matching relationship between enterprises and projects more accurate and timely, and improves the flexibility and effectiveness of resource allocation. The knowledge graph question-answering model using deformed graph matching can accurately respond to the user's target query request and quickly and accurately provide the target query results. This intelligent query response mechanism greatly improves the user experience, allowing users to easily obtain the required matching information, thereby improving the efficiency and quality of decision-making. By defining the matching logic of enterprise features to projects and project features to enterprises, a two-way matching logic is realized, which enhances the interactivity and comprehensiveness of matching between enterprises and projects. This enhanced interactive logic can not only meet the needs of different users more comprehensively, but also promote the two-way flow and optimal allocation of resources, and then use artificial intelligence algorithms to improve the accuracy of interactive matching and query between project applications and enterprise information.

[0035] In a specific embodiment, the process of executing step 101 may specifically include the following steps:

[0036] (1) Obtain multiple target data sources, including project public websites, third-party service agencies, and enterprise management platforms;

[0037] (2) Collect historical data from multiple target data sources to obtain historical project application data from each target data source;

[0038] (3) Performing data formatting on the historical project application data of each target data source to obtain the formatted project application data of each target data source;

[0039] (4) Performing data deduplication processing on the formatted project application data of each target data source to obtain deduplicated project application data of each target data source;

[0040] (5) Perform data fusion on the deduplicated project application data of each target data source to obtain a fused enterprise project data set.

[0041] Specifically, multiple target data sources are obtained, which are distributed on different platforms. For example, the project disclosure website provides public project information released by the government or other organizations, third-party service agencies may provide data on industry dynamics, and the enterprise management platform records the project application history of the enterprise itself. Historical data collection work is carried out on the diverse target data sources to obtain historical data related to project application in each target data source. Through a large number of web crawler technologies and API calls, it is ensured that as comprehensive and detailed data as possible is collected from various channels. In the process of collecting environmental protection project data, key information such as project name, application enterprise, project time, and project status are captured from various platforms. The collected data contains various formats and structures and needs to be formatted. Convert data in different formats into a unified format and organize the data according to the predetermined structure for subsequent processing and analysis. De-duplicate the formatted data. Since different data sources may repeatedly record the same project application information, de-duplicate can ensure data accuracy. Duplicates are identified by comparing key fields such as project name, application enterprise, and project time to ensure that the information of each project is unique in the final data set. The de-duplicated data is merged to form a comprehensive and integrated enterprise project data set. Data fusion not only involves merging data from different data sources for the same project, but also uses information from different data sources to complement each other to enrich the data records of each project. For example, in the data fusion process of environmental protection projects, it may be found that a third-party service agency provides additional details of a project, such as a specific environmental impact assessment of the project. This information will be integrated with the basic project declaration information to provide a more comprehensive view of the project.

[0042] In a specific embodiment, the process of executing step 102 may specifically include the following steps:

[0043] (1) Perform text classification processing on the integrated enterprise project data set to obtain classified text paragraph data;

[0044] (2) Extract key sentences from the classified text paragraph data based on the sequence-to-sequence model to obtain a set of key information sentences;

[0045] (3) The semantic features of the key information sentence set are extracted through the Transformer model to obtain a semantic feature set. The Transformer model includes: a full-layer normalization layer, a weighted residual connection layer, a reinforcement learning position encoding layer, and a zero-masked self-attention mechanism layer;

[0046] (4) Perform entity recognition on the fused enterprise project dataset based on the semantic feature set to obtain text entities and attribute data;

[0047] (5) Perform entity relationship analysis and relationship mapping on text entities and attribute data to obtain entity and attribute relationship data.

[0048] Specifically, the fusion enterprise project data set is subjected to text classification processing. The text data containing various information is divided into different categories according to its content characteristics. For example, the text in the project application document is divided into several paragraphs such as project introduction, project objectives, implementation plan, and expected effects. The process of text classification usually relies on machine learning models, such as support vector machines (SVM) and deep learning models, or convolutional neural networks (CNN) and recurrent neural networks (RNN). Through these models, the characteristics of the text are learned and text paragraphs of different categories are automatically identified. For example, by analyzing the text data of the project application document, the model can identify which paragraphs describe the project objectives and which paragraphs introduce the implementation plan. Key sentences are extracted from the classified text paragraph data. Sequence-to-sequence model-based technologies, such as long short-term memory networks (LSTM) or Transformer models, process the sequence characteristics of text data and extract sentences carrying key information from them. The most representative and information-valuable sentences are selected from a large amount of text. For example, sentences describing key technologies or innovations are extracted from the paragraphs of the project implementation plan. Semantic features are extracted from the key information sentence set through the Transformer model. The Transformer model includes a full-layer normalization layer, a weighted residual connection layer, a reinforcement learning position encoding layer, and a zero-masked self-attention mechanism layer. For example, by processing the extracted sentences about the key technologies of the project, the Transformer model can understand the specific application, advantages, and relationship between the technology and the project goals. Entity recognition is performed on the fused enterprise project dataset based on the semantic feature set to extract entities in the text (such as technology name, project name, enterprise name, etc.) and their attributes (such as technology field, project cycle, enterprise size, etc.). Different entities and their attributes are accurately labeled from the semantic feature set through named entity recognition (NER) technology. Entity relationship analysis and relationship mapping are performed on text entities and attribute data. The logical relationships and interactions between entities are analyzed, such as how a specific technology supports the realization of project goals, or how an enterprise can improve its technical strength by participating in a specific project. The process of entity relationship analysis and mapping can be constructed with the help of graph database technology, such as Neo4j, to construct an entity-relationship graph, i.e., a knowledge graph. For example, by analyzing entity and attribute data, a graph is constructed in which nodes represent entities such as projects, technologies, and enterprises, and edges represent the relationships between them, such as "adoption", "support", and "participation".

[0049] In a specific embodiment, the execution step extracts semantic features from the key information sentence set through a Transformer model to obtain a semantic feature set, and the Transformer model includes: a full-layer normalization layer, a weighted residual connection layer, a reinforcement learning position encoding layer, and a zero-masking self-attention mechanism layer. The process can specifically include the following steps:

[0050] (1) Input the key information sentence sets into the full-layer normalization layer for preliminary semantic understanding and data standardization to obtain a standardized semantic data set;

[0051] (2) Input the standardized semantic dataset into the weighted residual connection layer for feature enhancement processing to obtain enhanced semantic feature data;

[0052] (3) The enhanced semantic feature data is encoded with position information through the reinforcement learning position encoding layer to obtain position encoded data;

[0053] (4) Based on the zero-masked self-attention mechanism layer, deep semantic understanding and feature extraction are performed on the position-encoded data to obtain a set of semantic features.

[0054] Specifically, the key information sentence set is input into the full-layer normalization layer for preliminary semantic understanding and data standardization. In natural language processing tasks, text data often contains various noises and inconsistencies, such as different text lengths, language styles, and expression habits. Full-layer normalization standardizes the input of each layer so that each layer of the network can learn features at the same scale, accelerate the training process of the model and improve its stability. For example, when processing key sentences, full-layer normalization can ensure that the lexical contribution in each sentence is evenly considered without being dominated by certain high-frequency words. The standardized semantic data set is input into the weighted residual connection layer for feature enhancement. The residual network (ResNet) helps the flow of information in the network through weighted residual connections, solving the problem of gradient vanishing or exploding that may be encountered during the training of deep networks. For example, the weighted residual connection layer can help the model better understand the nuances in the text, such as the connection and distinction between different information. The enhanced semantic feature data is encoded with position information through the reinforcement learning position encoding layer. The reinforcement learning position encoding layer can dynamically learn how to encode position information, and the model can adaptively adjust its understanding and representation of position information according to the context. For example, when analyzing key sentences, the reinforcement learning position encoding layer can ensure that the model accurately captures the semantic role and syntactic structure of vocabulary, thereby better understanding the meaning of the text. The zero-masked self-attention mechanism layer performs deep semantic understanding and feature extraction on the position-encoded data. The self-attention mechanism can capture long-range dependencies by considering the information of all elements in the sequence when processing each element. The zero-masking part ensures that irrelevant information will not interfere when calculating the attention weight. When analyzing a set of key sentences about technological innovation, the zero-masked self-attention mechanism layer can accurately identify the technical concepts mentioned in the text, the relationship between them, and their impact on project goals.

[0055] In a specific embodiment, the process of executing step 103 may specifically include the following steps:

[0056] (1) Define the project feature information corresponding to the enterprise feature information query as the enterprise feature to project matching logic, and at the same time, define the enterprise feature information corresponding to the project feature information query as the project feature to enterprise matching logic;

[0057] (2) Based on the enterprise feature to project matching logic, the knowledge graph of entity and attribute relationship data is initialized to obtain the initial knowledge graph from the enterprise to the project, and based on the project feature to enterprise matching logic, the knowledge graph of entity and attribute relationship data is initialized to obtain the initial knowledge graph from the project to the enterprise;

[0058] (3) The initial knowledge graph from enterprise to project and the initial knowledge graph from project to enterprise are fused through the graph attention network to obtain the fused knowledge graph of enterprise and project;

[0059] (4) The enterprise and project fusion knowledge graph is structured and dynamically updated through the RDF graph model to obtain the enterprise-project matching relationship graph.

[0060] Specifically, the project feature information corresponding to the enterprise feature information query is defined as the enterprise feature to project matching logic, and the enterprise feature information corresponding to the project feature information query is defined as the project feature to enterprise matching logic. The enterprise feature to project matching logic focuses on how to query the project feature information (such as project requirements, project scale, required technical fields, etc.) that matches the enterprise feature information (such as enterprise scale, industry classification, technical capabilities, etc.), while the project feature to enterprise matching logic queries the enterprises that may be interested or meet the conditions based on the project feature information. Initialize the knowledge graph based on the matching logic. Organize the entities (enterprises and projects) and their attributes (feature information) according to the matching logic to form a preliminary knowledge structure. According to the enterprise feature to project matching logic, extract the enterprise entities and their related features from the existing data, and link these features with the potential project features to construct the initial knowledge graph from the enterprise to the project; similarly, according to the project feature to enterprise matching logic, link the project entities and their features with the potential enterprise features to construct the initial knowledge graph from the project to the enterprise. The initial knowledge graph is fused through the Graph Attention Networks (GAT). The graph attention network can automatically assign different weights to nodes (enterprises or projects) in the graph according to the importance of their connections, and dynamically adjust their expressions in the graph according to the strength of the relationship between entities, thereby achieving a more refined and dynamic representation of entity relationships. Through GAT processing, the initial knowledge graphs from enterprises to projects and from projects to enterprises are effectively integrated to generate a comprehensive fused knowledge graph that can fully reflect the matching relationship between enterprises and projects. The fused knowledge graph is structured and dynamically updated through the Resource Description Framework (RDF) graph model. RDF is a model for describing Web resources. It provides a flexible and standardized representation method for knowledge graphs by using triples of subject, predicate, and object. On this basis, the fused knowledge graph of enterprises and projects can be stored in a structured form for easy query, analysis, and update. With the addition of new enterprise information and project data, the knowledge graph can be dynamically updated through the RDF model, continuously optimizing and expanding its content and structure to ensure that the graph can continuously reflect the latest matching relationship between enterprises and projects.

[0061] In a specific embodiment, the execution step performs knowledge graph fusion on the initial knowledge graph from the enterprise to the project and the initial knowledge graph from the project to the enterprise through the graph attention network to obtain the fused knowledge graph of the enterprise and the project, which may specifically include the following steps:

[0062] (1) Extracting node features from the initial knowledge graph from the enterprise to the project and the initial knowledge graph from the project to the enterprise respectively, and obtaining the enterprise node feature set of the initial knowledge graph from the enterprise to the project and the project node feature set of the initial knowledge graph from the project to the enterprise;

[0063] (2) Extracting edge features from the initial knowledge graph from the enterprise to the project and the initial knowledge graph from the project to the enterprise respectively, and obtaining the edge feature set from the enterprise to the project of the initial knowledge graph from the enterprise to the project and the edge feature set from the project to the enterprise of the initial knowledge graph from the project to the enterprise;

[0064] (3) Calculate the node feature similarity of the enterprise node feature set and the project node feature set to obtain node feature similarity data;

[0065] (4) Calculate the edge weights of the edge feature set from the enterprise to the project and the edge feature set from the project to the enterprise to obtain a weighted edge feature set;

[0066] (5) The attention allocation process is performed on the node feature similarity data and the weighted edge feature set through the graph attention network to obtain the corresponding weighted graph of attention from the enterprise to the project and the weighted graph of attention from the project to the enterprise;

[0067] (6) The weighted graph of attention from enterprises to projects and the weighted graph of attention from projects to enterprises are fused to obtain a fused knowledge graph of enterprises and projects.

[0068] Specifically, node features are extracted from the initial knowledge graph from enterprise to project and the initial knowledge graph from project to enterprise. For each enterprise entity and project entity, their attribute information is analyzed, such as the scale, industry, technical strength of the enterprise, and the type, demand, budget, etc. of the project, to obtain the enterprise node feature set and the project node feature set. Edge features are extracted from these two initial knowledge graphs. Edge features refer to the relationship features connecting enterprises and projects, such as cooperation frequency, success rate, cooperation field, etc. These features can reveal the interaction mode and cooperation potential between enterprises and projects, and help understand the degree of matching between the two parties, and obtain the edge feature set from enterprise to project and the edge feature set from project to enterprise. Feature similarity calculation is performed on the enterprise node feature set and the project node feature set to find possible matching points between enterprises and projects. By comparing the features of different nodes, the similarity between them is quantified. Similarly, weight calculation is performed on the edge feature set to determine the importance of different edges, which helps to reveal the key factors of cooperation between enterprises and projects. Attention allocation processing is performed on the node feature similarity data and the weighted edge feature set through the graph attention network. The graph attention network can automatically assign attention weights according to the importance of nodes and edges, highlight the most relevant enterprise-project matching relationships, and maintain a dynamic balance of information in the entire knowledge graph, obtaining attention-weighted enterprise-to-project and project-to-enterprise graphs. The two attention-weighted graphs are fused to form a comprehensive enterprise and project fusion knowledge graph. The fusion process finds the optimal fusion method through intelligent analysis to ensure that the information in the graph is both comprehensive and accurate. During the fusion process, the matching logic and relationship coherence between entities are maintained to ensure that the final knowledge graph can truly and effectively reflect the matching relationship between enterprises and projects.

[0069] In a specific embodiment, the process of executing step 104 may specifically include the following steps:

[0070] (1) Receive the target query request sent by the user terminal, and perform semantic analysis on the target query request to obtain structured query conditions;

[0071] (2) Extract keywords from structured query conditions to obtain a query keyword set;

[0072] (3) Identify the query intent of the query keyword set to obtain the target query intent;

[0073] (4) A knowledge graph question answering model based on deformable graph matching queries the enterprise-project matching relationship graph according to the target query intent and obtains multiple initial query results;

[0074] (5) Calculate the relevance of multiple initial query results and target query intent to obtain a relevance index between each initial query result and the target query intent;

[0075] (6) The multiple initial query results are sorted and analyzed according to the relevance index to obtain a query result sequence, and the query result sequence is screened for the best matching items to obtain the target query result.

[0076] Specifically, the user's target query request is received and semantic analysis is performed on these requests. Semantic analysis refers to the use of natural language processing technology to parse the natural language expression of the user's query and convert it into structured query conditions that can be understood by the machine. The purpose is to understand the user's real needs, including the query subject, specific attribute requirements, etc. Keyword extraction is performed based on the structured query conditions. Core words are extracted from the user's query, and these keywords will serve as the basis for subsequent queries. The process of extracting keywords includes analyzing the semantic structure of the query conditions and identifying nouns, verbs, and adjectives, etc. These words can often accurately represent the user's query intent. Query intent recognition is performed on the extracted keyword set, and the user's query purpose is inferred by analyzing the keywords and their combinations. Based on the knowledge graph question-answering model of deformed graph matching, the enterprise-project matching relationship graph is deeply queried according to the query intent. Deformed graph matching is a graph query technology that dynamically adjusts the entities and relationships in the knowledge graph according to the query intent to find the most matching query results. Using the entity relationships in the knowledge graph, a series of initial query results are provided to the user through an intelligent matching algorithm. The correlation between the initial query results and the user's target query intent is calculated. By evaluating the match between the query results and the user's query intent, a relevance index is assigned to each result. Relevance calculation usually involves multiple factors, including the degree of attribute matching of entities, the consistency of relationships between entities, etc. All initial query results are sorted and analyzed according to the relevance index, and the sorted query result sequence is screened for the best matching items. From the numerous query results, the items that best meet the user's needs are selected to provide the user with the final query results. The sorting and screening process takes into account the relevance, accuracy, and possible diversity of the results to ensure that users can obtain the most valuable information from them.

[0077] The above describes the project application and enterprise information interaction method based on artificial intelligence in the embodiment of the present application. The following describes the project application and enterprise information interaction system based on artificial intelligence in the embodiment of the present application. Please refer to Figure 2 In the embodiment of the present application, an embodiment of the project application and enterprise information interaction system based on artificial intelligence includes:

[0078] The acquisition module 201 is used to acquire historical project application data from multiple target data sources, and pre-process and fuse the historical project application data to obtain a fused enterprise project data set;

[0079] The recognition module 202 is used to extract semantic features and recognize text entities on the fused enterprise project dataset through a Transformer model to obtain entity and attribute relationship data;

[0080] A construction module 203 is used to construct a knowledge graph for the entity and attribute relationship data based on a graph attention network and an RDF graph model to obtain an enterprise-project matching relationship graph;

[0081] The query module 204 is used to receive a target query request sent by a user terminal, and query the target query request and the enterprise-project matching relationship graph based on the knowledge graph question-answering model of deformation graph matching to obtain a target query result.

[0082] Through the collaboration of the above components, by automatically acquiring historical project application data from multiple target data sources, and preprocessing and fusion of these data, the scattered data sources can be efficiently integrated to generate a fusion enterprise project data set. This efficient data processing capability significantly reduces the time and effort of manual data collation and improves the efficiency and accuracy of data processing. By applying the advanced Transformer model to extract semantic features and identify text entities on the fusion enterprise project data set, the text content can be deeply understood, and key entities and attributes and their relationships can be identified. This deep semantic understanding provides a solid foundation for accurately matching enterprises and projects, and improves the accuracy and relevance of matching. Based on the graph attention network and RDF graph model, not only can a rich enterprise-project matching relationship graph be constructed, but the graph can also be dynamically updated according to real-time data. This dynamic knowledge graph construction and application mechanism makes the matching relationship between enterprises and projects more accurate and timely, and improves the flexibility and effectiveness of resource allocation. The knowledge graph question-answering model using deformable graph matching can accurately respond to users' target query requests and quickly and accurately provide target query results. This intelligent query response mechanism greatly improves the user experience, allowing users to easily obtain the required matching information, thereby improving the efficiency and quality of decision-making. By defining the matching logic of enterprise features to projects and project features to enterprises, a two-way matching logic is realized, which enhances the interactivity and comprehensiveness of matching between enterprises and projects. This enhanced interactive logic can not only meet the needs of different users more comprehensively, but also promote the two-way flow and optimal allocation of resources, and then use artificial intelligence algorithms to improve the accuracy of interactive matching and query between project applications and enterprise information.

[0083] The present application also provides a computer device, which includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the artificial intelligence-based project application and enterprise information interaction method in the above-mentioned embodiments.

[0084] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the artificial intelligence-based project declaration and enterprise information interaction method.

[0085] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0086] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0087] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A project application and enterprise information interaction method based on artificial intelligence, characterized in that: The method for project declaration and enterprise information interaction based on artificial intelligence includes: Acquire historical project application data from multiple target data sources, and perform preprocessing and data fusion on the historical project application data to obtain a fused enterprise project data set; Performing semantic feature extraction and text entity recognition on the fused enterprise project data set through a Transformer model to obtain entity and attribute relationship data; Based on the graph attention network and RDF graph model, a knowledge graph is constructed for the entity and attribute relationship data to obtain an enterprise-project matching relationship graph; specifically, the following steps are performed: defining the project feature information corresponding to the enterprise feature information query as the enterprise feature to project matching logic, and at the same time, defining the enterprise feature information corresponding to the project feature information query as the project feature to enterprise matching logic; initializing the knowledge graph of the entity and attribute relationship data based on the enterprise feature to project matching logic to obtain an initial knowledge graph from the enterprise to the project, and initializing the knowledge graph of the entity and attribute relationship data based on the project feature to the enterprise matching logic to obtain an initial knowledge graph from the project to the enterprise; performing node feature extraction on the initial knowledge graph from the enterprise to the project and the initial knowledge graph from the project to the enterprise, respectively, to obtain an enterprise node feature set of the initial knowledge graph from the enterprise to the project and a project node feature set of the initial knowledge graph from the project to the enterprise; performing node feature extraction on the initial knowledge graph from the enterprise to the project and the initial knowledge graph from the project to the enterprise, respectively, to obtain an enterprise node feature set of the initial knowledge graph from the enterprise to the project and a project node feature set of the initial knowledge graph from the project to the enterprise; performing node feature extraction on the initial knowledge graph from the enterprise to the project and the initial knowledge graph from the project to the enterprise, respectively. The initial knowledge graph of the enterprise is extracted to obtain the edge feature set from the enterprise to the project of the initial knowledge graph from the enterprise to the project and the edge feature set from the project to the enterprise of the initial knowledge graph from the project to the enterprise; the node feature similarity of the enterprise node feature set and the project node feature set is calculated to obtain node feature similarity data; the edge weight is calculated for the edge feature set from the enterprise to the project and the edge feature set from the project to the enterprise to obtain a weighted edge feature set; the node feature similarity data and the weighted edge feature set are processed by the graph attention network to allocate attention to the node feature similarity data and the weighted edge feature set to obtain the corresponding weighted graph of attention from the enterprise to the project and the weighted graph of attention from the project to the enterprise; the weighted graph of attention from the enterprise to the project and the weighted graph of attention from the project to the enterprise are fused to obtain the fused knowledge graph of the enterprise and the project; the fused knowledge graph of the enterprise and the project is structured and dynamically updated by the RDF graph model to obtain the enterprise-project matching relationship graph; A target query request sent by a user terminal is received, and the target query request and the enterprise-project matching relationship graph are queried based on a knowledge graph question-answering model of deformation graph matching to obtain a target query result.

2. The method for project declaration and enterprise information interaction based on artificial intelligence according to claim 1 is characterized in that: The obtaining of historical project application data from multiple target data sources, and preprocessing and data fusion of the historical project application data to obtain a fused enterprise project data set includes: Acquire multiple target data sources, including project public websites, third-party service agencies, and enterprise management platforms; Collecting historical data from the multiple target data sources to obtain historical project declaration data from each target data source; Perform data formatting on the historical project declaration data of each target data source to obtain the formatted project declaration data of each target data source; Performing data deduplication processing on the formatted project declaration data of each target data source to obtain deduplicated project declaration data of each target data source; Data fusion is performed on the deduplicated project application data of each target data source to obtain a fused enterprise project data set.

3. The method for project declaration and enterprise information interaction based on artificial intelligence according to claim 1 is characterized in that: The method of performing semantic feature extraction and text entity recognition on the fused enterprise project dataset through the Transformer model to obtain entity and attribute relationship data includes: Performing text classification processing on the fused enterprise project data set to obtain classified text paragraph data; Extract key sentences from the classified text paragraph data based on a sequence-to-sequence model to obtain a set of key information sentences; Extracting semantic features from the key information sentence set through a Transformer model to obtain a semantic feature set, wherein the Transformer model includes: a full-layer normalization layer, a weighted residual connection layer, a reinforcement learning position encoding layer, and a zero-masked self-attention mechanism layer; Performing entity recognition on the fused enterprise project data set according to the semantic feature set to obtain text entities and attribute data; Performing entity relationship analysis and relationship mapping on the text entities and attribute data to obtain entity and attribute relationship data.

4. The method for project declaration and enterprise information interaction based on artificial intelligence according to claim 3 is characterized in that: The semantic feature extraction of the key information sentence set is performed through the Transformer model to obtain a semantic feature set, wherein the Transformer model includes: a full-layer normalization layer, a weighted residual connection layer, a reinforcement learning position encoding layer, and a zero-masking self-attention mechanism layer, including: Inputting the key information sentence sets into the full-layer normalization layer for preliminary semantic understanding and data standardization processing to obtain a standardized semantic data set; Inputting the standardized semantic data set into the weighted residual connection layer for feature enhancement processing to obtain enhanced semantic feature data; Performing position information encoding processing on the enhanced semantic feature data through a reinforcement learning position encoding layer to obtain position-encoded data; Based on the zero-masked self-attention mechanism layer, deep semantic understanding and feature extraction are performed on the position-encoded data to obtain a semantic feature set.

5. The method for project declaration and enterprise information interaction based on artificial intelligence according to claim 1 is characterized in that: The receiving of the target query request sent by the user terminal and querying the target query request and the enterprise-project matching relationship graph based on the knowledge graph question-answering model of deformation graph matching to obtain the target query result include: Receiving a target query request sent by a user terminal, and performing semantic analysis on the target query request to obtain a structured query condition; Extracting keywords from the structured query conditions to obtain a query keyword set; Performing query intent recognition on the query keyword set to obtain a target query intent; A knowledge graph question answering model based on deformed graph matching queries the enterprise-project matching relationship graph according to the target query intent to obtain multiple initial query results; Calculating the relevance between the multiple initial query results and the target query intent to obtain a relevance index between each initial query result and the target query intent; The multiple initial query results are sorted and analyzed according to the correlation index to obtain a query result sequence, and the query result sequence is screened for the best matching item to obtain a target query result.

6. A project declaration and enterprise information interaction system based on artificial intelligence, characterized in that: The artificial intelligence-based project application and enterprise information interaction system includes: An acquisition module is used to acquire historical project application data from multiple target data sources, and to preprocess and fuse the historical project application data to obtain a fused enterprise project data set; A recognition module, used for performing semantic feature extraction and text entity recognition on the fused enterprise project data set through a Transformer model to obtain entity and attribute relationship data; A construction module is used to construct a knowledge graph for the entity and attribute relationship data based on a graph attention network and an RDF graph model to obtain an enterprise-project matching relationship graph; specifically comprising: defining the project feature information corresponding to the enterprise feature information query as the enterprise feature to project matching logic, and at the same time, defining the enterprise feature information corresponding to the project feature information query as the project feature to enterprise matching logic; initializing the knowledge graph of the entity and attribute relationship data based on the enterprise feature to project matching logic to obtain an initial knowledge graph from the enterprise to the project, and initializing the knowledge graph of the entity and attribute relationship data based on the project feature to the enterprise matching logic to obtain an initial knowledge graph from the project to the enterprise; extracting node features from the initial knowledge graph from the enterprise to the project and the initial knowledge graph from the project to the enterprise, respectively, to obtain an enterprise node feature set of the initial knowledge graph from the enterprise to the project and a project node feature set of the initial knowledge graph from the project to the enterprise; extracting node features from the initial knowledge graph from the enterprise to the project and the initial knowledge graph from the project to the enterprise, respectively. Perform edge feature extraction on the initial knowledge graph of the enterprise to obtain an edge feature set from the enterprise to the project of the initial knowledge graph from the enterprise to the project and an edge feature set from the project to the enterprise of the initial knowledge graph from the project to the enterprise; perform node feature similarity calculation on the enterprise node feature set and the project node feature set to obtain node feature similarity data; perform edge weight calculation on the edge feature set from the enterprise to the project and the edge feature set from the project to the enterprise to obtain a weighted edge feature set; perform attention allocation processing on the node feature similarity data and the weighted edge feature set through the graph attention network to obtain corresponding weighted graphs of attention from the enterprise to the project and weighted graphs of attention from the project to the enterprise; fuse the weighted graphs of attention from the enterprise to the project and the weighted graphs of attention from the project to the enterprise into knowledge graphs to obtain a fused knowledge graph of the enterprise and the project; perform structured representation and dynamic graph update on the fused knowledge graph of the enterprise and the project through the RDF graph model to obtain an enterprise-project matching relationship graph; The query module is used to receive a target query request sent by a user terminal, and query the target query request and the enterprise-project matching relationship graph based on the knowledge graph question-answering model of deformation graph matching to obtain a target query result.

7. A computer device, characterized in that: The computer device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the computer device executes the artificial intelligence-based project declaration and enterprise information interaction method as described in any one of claims 1-5.

8. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the artificial intelligence-based project declaration and enterprise information interaction method as described in any one of claims 1 to 5 is implemented.

Images