Project information recommendation method and system

Abstract

The invention relates to the technical field of information recommendation, in particular to an engineering information recommendation method and system, and the method comprises the following steps: S1, constructing an engineering domain knowledge graph; s2, user engineering project background analysis; s3, aligning the recommended contents with the technical nodes of the knowledge graph: preliminarily aligning the recommended contents extracted from the engineering information database with the matched technical nodes to obtain preliminary recommended contents, screening again to obtain final recommended contents, and pushing the final recommended contents to the user; and S4, semantic extension and technical completion: completing the technical background of the recommended content to ensure that the recommended content not only covers the dominant demand of the user, but also includes the technical scheme related to the potential demand of the project. According to the method, reasoning is carried out through the associated edges in the knowledge graph, the matching degree of the recommended content and the technical nodes required by the user is judged, the recommended content is screened and optimized based on the matching degree, and it is ensured that the content pushed to the user not only meets the current requirement, but also can adapt to the future project requirement.

Description

Technical Field

[0001] The present invention relates to the technical field of information recommendation, and in particular to a method and system for recommending engineering information. Background Art

[0002] As the scale and complexity of modern engineering projects continue to increase, the technologies, equipment, processes and standards involved in the projects are showing a trend of highly diversified and specialized development. In this context, the amount of technical information required at each stage of the engineering project is large and complex. Engineers and project managers are faced with the challenge of quickly obtaining project-related technical solutions from massive amounts of information. At present, the information recommendation systems on the market are mostly concentrated in the fields of e-commerce and social media, and mainly recommend based on user behavior data and historical preferences. However, this type of general recommendation system is difficult to meet the multi-level and complex technical requirements in the engineering field, especially when the project involves multiple disciplines and the requirements change at different stages. The existing recommendation systems often show the following shortcomings:

[0003] Unable to accurately match professional technical needs: Traditional recommendation systems are mostly based on keyword matching or simple user behavior analysis. They have difficulty understanding the complex technical terms and upstream and downstream technical relationships in the engineering field, resulting in low relevance of recommended content to users' actual needs.

[0004] Lack of technical expansion and foresight: The technical requirements in engineering projects are not limited to current tasks, but also involve technical problems and solutions that may arise in the future. However, existing recommendation systems often only focus on current explicit needs and cannot provide alternative paths or potential related technologies for technical solutions. They lack expansion and foresight support for technical solutions.

[0005] Failure to fully utilize knowledge graphs for in-depth analysis: Knowledge graphs are increasingly used in various fields, and are particularly suitable for processing technical associations and dependencies in complex fields. However, existing engineering information recommendation systems rarely use knowledge graphs as a tool and are unable to perform multi-level reasoning through technical association edges to provide technical solutions that meet user needs.

[0006] Therefore, there is an urgent need for an engineering information recommendation method that can accurately analyze user needs, use the knowledge graph in the engineering field, perform semantic expansion and complete technical solutions, so as to meet users' needs for multiple stages and technical levels in engineering projects. Summary of the invention

[0007] The present invention provides a method and system for recommending engineering information.

[0008] A method for recommending engineering information includes the following steps:

[0009] S1, Construction of knowledge graph in engineering field: Based on the actual needs in the engineering field, using engineering project data, technical documents, and industry standards, a knowledge graph covering the technology, equipment, process, and standards in the engineering field is constructed. The knowledge graph includes multiple technology nodes, each of which represents a technical element or process flow in an engineering field. The technology nodes are connected by technology-related edges, which include technology dependencies, collaborative working relationships, and development paths.

[0010] S2, user project background analysis: based on user input, obtain the user's current project information, including project progress and task content, analyze the user's potential needs based on the user's current project information, preliminarily match the user's potential needs with the technical nodes in the knowledge graph, and identify the technical nodes in the user's project that match the knowledge graph;

[0011] S3, Aligning recommended content with technical nodes in the knowledge graph: Perform preliminary alignment of the recommended content extracted from the engineering information database with the matching technical nodes. The preliminary alignment process includes semantic analysis of the technical terms and technical solutions in the recommended content to obtain preliminary recommended content, and reasoning on the associated edges of the knowledge graph to determine the matching degree between the recommended content and the technical nodes in the user's project. Based on the matching results, the preliminary recommended content is screened again, and the final recommended content is screened and pushed to the user;

[0012] S4, semantic extension and technical completion: Based on the technology nodes in the knowledge graph and other associated technology nodes, the semantic extension mechanism is used to generate completed recommended content. Semantic extension includes finding potential related technical solutions or alternative technical paths for the user's current needs through the associated edges of the technology nodes in the knowledge graph, and completing the technical background of the recommended content to ensure that the recommended content not only covers the user's explicit needs, but also includes technical solutions related to the project's potential needs.

[0013] Optionally, the construction of the engineering domain knowledge graph in S1 specifically includes:

[0014] S11, collect engineering project data (project planning, construction logs, progress reports) and technical documents, use named entity recognition (NER) model to automatically extract technical elements and process flows, including equipment, materials, process flows, and store them in a structured manner, and generate technical nodes based on the extracted technical elements and process flows. Each technical node represents a specific technical concept, process flow or equipment;

[0015] S12, Industry Standard Docking: Combine industry standards and specification documents (ISO, national standards, standards issued by industry associations), verify whether the generated technology nodes comply with industry specifications through industry standard docking, and use the BM25 text matching algorithm to compare the identified technical terms with the terms in the industry standards to ensure that the technology nodes comply with the current industry specifications in the engineering field;

[0016] S13, technology-related edge construction: Use graph neural network GNN to model the relationship between technology nodes, and generate different types of technology-related edges through upstream and downstream dependencies, collaborative working relationships, and technology development paths between technology nodes. The technology-related edges include:

[0017] Technology dependency edge: indicates the dependency of a technology node on other technology nodes when applied;

[0018] Collaborative working edge: represents the relationship between two or more technology nodes being used collaboratively in a specific project;

[0019] Development path edge: represents the technological evolution or alternative path between certain technology nodes, which is used to reflect the trend of technological development.

[0020] Optionally, the S2 specifically includes:

[0021] S21, user task content analysis: Use the word vector model to perform semantic analysis on the task content and project progress input by the user, extract the technical requirements in the project and the key technical terms in the current task. The technical terms include equipment names, process flows, and technical standards. Convert the technical terms into corresponding vector representations to generate semantic vector representations of the user's potential needs.

[0022] S22, semantic similarity calculation: preliminarily match the semantic vector representation of the user's potential needs with the technical nodes in the knowledge graph, calculate the semantic similarity between the technical terms in the user's task and the technical nodes in the knowledge graph through similarity calculation, and determine the technical nodes closest to the user's needs.

[0023] Optionally, the recommended content in S3 is aligned with the technical nodes of the knowledge graph as described above, using the same method, including semantic analysis of the recommended content extracted from the engineering information database, converting the technical terms in the extracted recommended content into word vector representations using a word vector model, generating a semantic vector representation of the recommended content, and then, as above, again using semantic similarity calculations to preliminarily align the technical terms in the recommended content with the matched technical nodes in the user project to obtain preliminary recommended content.

[0024] Optionally, the S3 determines the matching degree between the preliminary recommended content and the technical nodes in the user project by reasoning about the associated edges of the knowledge graph, specifically including:

[0025] Based on the graph neural network GNN, the associated edge reasoning in the knowledge graph is performed to infer the dependencies, collaborations and development paths between technology nodes, analyze the impact of neighboring nodes on the current technology nodes, and determine the matching depth between the recommended content and the technology nodes in the user's project.

[0026] Optionally, the graph neural network GNN performs multi-layer reasoning on the technology nodes and their associated edges, analyzes the multi-layer relationship between the technology nodes in the recommended content and the matching technology nodes in the user project, and through multi-layer reasoning, captures not only the direct semantic match between the recommended content and the technology nodes in the user project, but also understands the technology chain, collaborative work plan and technology development path on which they depend;

[0027] After the graph neural network GNN is inferred, a matching depth score is assigned to the matching relationship between each recommended content and the technical node of the user project. The matching depth score is used to measure the relevance and distance of the recommended content in the technical chain.

[0028] Optionally, the matching depth score is calculated as: Among them, Score depth is the final matching depth score, L is the number of layers of the graph neural network, indicating the depth of reasoning, and α k is the weight of each layer, indicating the contribution of a certain layer of reasoning, The feature vector representation of the k-th layer technology node v is generated after the associated edge reasoning, and the recommended content is filtered based on the matching depth score.

[0029] Optionally, the semantic extension and technology completion in S4 include associated edge reasoning based on the graph neural network GNN, and perform multi-layer reasoning to perform semantic extension on the matching technology nodes in the user project to find potential related technical solutions or alternative technical paths associated with the technology node, and the associated edges include technology dependencies, collaborative working relationships and technology development paths. By analyzing neighbor nodes, technology nodes that are helpful to current project needs are identified, and through the neighbor node set of the current technology node, through the neighbor nodes, extended information of the technology node is inferred, including alternative solutions or collaborative working nodes. Through semantic extension, neighbor nodes are found, and completed recommended content is generated, and the technology nodes in the knowledge graph are combined with the expanded neighbor nodes to complete the technical background of the recommended content.

[0030] Optionally, S4 also includes determining the downstream technical requirements that will be triggered by the technical nodes in the user's project, and providing relevant information to the user in advance, directly using the matching depth score as the triggered downstream technical requirement score, thereby ensuring that the recommended content obtained by the user is forward-looking in the technology chain. In the downstream technical demand prediction, by analyzing the associated edges in the knowledge graph, the subsequent technical path of the current technical node is determined, and relevant information is provided to the user in advance.

[0031] A recommendation system for engineering information, used to implement the above-mentioned recommendation method for engineering information, includes the following modules:

[0032] Knowledge graph construction module: According to the actual needs of the engineering field, it collects and processes engineering project data, technical documents and industry standards to build a knowledge graph covering the technology, equipment, process and standards in the engineering field. The knowledge graph includes multiple technology nodes, each of which represents a technical element or process flow in the engineering field, and the nodes are connected by technology-related edges;

[0033] User project background analysis module: Based on the information input by the user, it obtains relevant data of the current project, including project progress and task content, parses the information and analyzes the user's potential needs, preliminarily matches the potential needs with the technical nodes in the knowledge graph, and identifies the technical nodes related to the user's project;

[0034] Recommended content and technology node alignment module: extract recommended content from the engineering information database and preliminarily align it with the matched technology nodes. The alignment process includes semantic analysis of technical terms and technical solutions in the recommended content, generating preliminary recommended content, and reasoning by analyzing the associated edges in the knowledge graph to determine the matching degree between the recommended content and the technology nodes in the user's project. The preliminary recommended content is screened according to the matching degree to generate recommended content that meets user needs.

[0035] Semantic expansion and technology completion module: Based on the technology nodes in the knowledge graph and other related technology nodes, the semantic expansion mechanism generates completed recommended content. The semantic expansion finds potential technical solutions or alternative technical paths related to user needs through the associated edges of the technology nodes in the knowledge graph to complete the technical background of the recommended content.

[0036] Push module: pushes the final filtered and completed recommended content to the user. The pushed content is screened to ensure that it matches the technical requirements of the user's current engineering project.

[0037] Beneficial effects of the present invention:

[0038] By building a knowledge graph covering technologies, equipment, processes and standards in the engineering field, it is possible to accurately analyze users' potential needs and match them with technical nodes in the knowledge graph after users input project progress and task content. This matching method ensures the high relevance of recommended content to user projects through the association of semantic analysis and technical nodes, and improves the accuracy and reliability of recommendations. After the recommended content is initially matched, further reasoning is performed through the associated edges in the knowledge graph to determine the degree of match between the recommended content and the technical nodes of user needs, and the recommended content is screened and optimized based on the degree of match, ensuring that the content pushed to users not only meets current needs, but also adapts to future project needs.

[0039] The present invention, by constructing an exclusive knowledge graph in the engineering field and combining the user's current project progress and task content, can accurately match the user's explicit technical needs, parse the technical terms in the project through natural language processing technology, and calculate the semantic similarity between them and the technical nodes in the knowledge graph, so as to quickly and accurately identify the technical nodes required for the user's current project. This semantic matching mechanism improves the accuracy of recommended content, especially in scenarios where a variety of professional terms are used in the engineering field, avoiding the problem of inaccurate keyword matching common in traditional recommendation methods, and ensuring that the recommended content is highly relevant to user needs.

[0040] The present invention, through the semantic extension mechanism, can not only meet the user's current explicit needs, but also identify the user's potential technical needs. By using the associated edges in the knowledge graph, it can find potential technical solutions or alternative paths related to the user's project technology nodes, and complete the technical background of the recommended content. Through multi-layer graph neural network (GNN) reasoning, the relationship between neighbor nodes is taken into consideration to generate more comprehensive technical recommendations, ensuring that the technical suggestions obtained by users in actual projects not only cover the current tasks, but also provide forward-looking technical support related to future project needs, thereby improving the breadth and depth of recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 A schematic diagram of a recommended method flow chart of an embodiment of the present invention;

[0043] Figure 2 Schematic diagram of the functional modules of the recommendation system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0045] It should be noted that the references to "one embodiment", "an embodiment", "an exemplary embodiment", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0046] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0047] like Figure 1 As shown, a method for recommending engineering information includes the following steps:

[0048] S1, Construction of knowledge graph in engineering field: Based on the actual needs in the engineering field, using engineering project data, technical documents, and industry standards, a knowledge graph covering the technology, equipment, process, and standards in the engineering field is constructed. The knowledge graph includes multiple technology nodes, each of which represents a technical element or process flow in an engineering field. The technology nodes are connected by technology-related edges, which include technology dependencies, collaborative working relationships, and development paths.

[0049] S2, user project background analysis: based on user input, obtain the user's current project information, including project progress and task content, analyze the user's potential needs based on the user's current project information, preliminarily match the user's potential needs with the technical nodes in the knowledge graph, and identify the technical nodes in the user's project that match the knowledge graph;

[0050] S3, Aligning recommended content with technical nodes in the knowledge graph: Perform preliminary alignment of the recommended content extracted from the engineering information database with the matching technical nodes. The preliminary alignment process includes semantic analysis of the technical terms and technical solutions in the recommended content to obtain preliminary recommended content, and reasoning on the associated edges of the knowledge graph to determine the matching degree between the recommended content and the technical nodes in the user's project. Based on the matching results, the preliminary recommended content is screened again, and the final recommended content is screened and pushed to the user;

[0051] S4, semantic extension and technical completion: Based on the technology nodes in the knowledge graph and other associated technology nodes, the semantic extension mechanism is used to generate completed recommended content. Semantic extension includes finding potential related technical solutions or alternative technical paths for the user's current needs through the associated edges of the technology nodes in the knowledge graph, and completing the technical background of the recommended content to ensure that the recommended content not only covers the user's explicit needs, but also includes technical solutions related to the project's potential needs.

[0052] The construction of the engineering domain knowledge graph in S1 specifically includes:

[0053] S11, collect engineering project data (project planning, construction logs, progress reports) and technical documents, use named entity recognition (NER) model to automatically extract technical elements and process flows, including equipment, materials, process flows, and store them in a structured manner, and generate technical nodes based on the extracted technical elements and process flows. Each technical node represents a specific technical concept, process flow or equipment;

[0054] The entity recognition model is expressed as:

[0055]

[0056] in;

[0057] x: input text sequence, which comes from engineering project data and technical documents. These documents are the data source for generating technology nodes;

[0058] y i : The label corresponding to the current word indicates which type of technical term the word belongs to. The labels include different types of technical entities, such as equipment, materials, processes, and technical terms. For example, "concrete mixer" will be marked as an equipment term, and "steel strength grade" will be marked as a material term;

[0059] φ(x,y i-1 ): The first feature function represents the current text sequence x and the previous label y i-1The joint feature of is the core of the conditional random field model. It can capture the context information of the current word and the previously identified label relationship to determine the label of the current word. The output of the feature function includes the word's part of speech, word frequency and other context-related features.

[0060] φ(x,y′): The second feature function, φ(x,y′) is a feature related to a label y′ in the label candidate set, which is used to calculate the score of the candidate label y′. By comparing all possible candidate labels, the label y′ with the highest score is selected as the final label of the current word;

[0061] W: weight matrix, which represents the model parameters and is used to weight the second feature function φ(x,y′). Through training, the model learns the importance of different features in predicting labels. The weight matrix will be continuously updated during the training process and will eventually be used to predict the labels of new text sequences.

[0062] Y: The set of all possible labels, that is, the set of technical term categories that may appear in the entire engineering field.

[0063] S12, Industry Standard Docking: Combine industry standards and specification documents (ISO, national standards, standards issued by industry associations), verify whether the generated technology nodes comply with industry specifications through industry standard docking, and use the BM25 text matching algorithm to compare the identified technical terms with the terms in the industry standards to ensure that the technology nodes comply with the current industry specifications in the engineering field;

[0064] The BM25 text matching algorithm is expressed as follows:

[0065]

[0066] in;

[0067] q: query words, which are technical nodes generated by named entity recognition. These nodes represent the technical terms used in the user's engineering project.

[0068] d: Standard documents refer to a collection of industry standards or technical specifications, such as ISO standards or national standards. The generated technical nodes need to be compared with these standard documents to ensure compliance;

[0069] f(q i ,d):Query word q i The frequency of occurrence in document d, that is, the frequency of occurrence of the technology node in the industry standard document. The higher the frequency, the higher the authority of the node in the standard.

[0070] IDF(q i ): Inverse document frequency, representing the query word q iThe rarity among all standard documents. The higher the rarity (i.e., rarely appears in other documents), the more unique the term is, and the weight increases accordingly: N is the total number of standard documents, n(q i ) is a string containing the query word q i The number of documents;

[0071] k1,b: adjustment parameters of the BM25 model. k1 controls the effect of query word frequency on the matching score and is set between 1.2 and 2.0. b controls the effect of document length and is set to 0.75. Adjusting these parameters helps balance the effect of word frequency and document length on the matching score.

[0072] |d|: The length of the document, which refers to the total number of words in the industry standard document. The longer the document, the smaller the impact of the number of occurrences of a single word on the matching score;

[0073] avgdl: The average length of the document, which is used to standardize the impact of document length on the score and avoid short documents having too low a score due to low word frequency.

[0074] S13, technology-related edge construction: Use graph neural network GNN to model the relationship between technology nodes, and generate different types of technology-related edges through the upstream and downstream dependencies, collaborative working relationships, and technology development paths between technology nodes. Technology-related edges include:

[0075] Technology dependency edge: indicates the dependency of a technology node on other technology nodes when applied;

[0076] Collaborative working edge: represents the relationship between two or more technology nodes being used collaboratively in a specific project;

[0077] Development path edge: represents the technological evolution or alternative path between certain technology nodes, which is used to reflect the trend of technological development.

[0078] The graph neural network GNN models the relationship between technology nodes as follows:

[0079]

[0080] in;

[0081] The feature vector representation of node v in the k+1th layer. The GNN model gradually learns the high-dimensional feature representation of the node through the propagation and update of each layer, reflecting its relationship with neighboring nodes. In the present invention, node v represents a certain technology node, such as "high-strength concrete" or "smart substation equipment";

[0082] The feature vector representation of the neighbor nodes at the kth layer is used to contribute to the update of the node;

[0083] N(v): the set of neighbor nodes of node v, that is, other technology nodes that are associated with technology node v. For example, in the knowledge graph, the neighbors of the "rebar" node may be related technology nodes such as "rebar connection process" or "rebar cover design";

[0084] c vu : The normalization coefficient between node v and its neighbor node u. This coefficient is used to balance the impact of node u on the update of node v. It is calculated based on the degree of the neighbor nodes to ensure that the contribution of each neighbor node is reasonable: Among them, |N(v)| and |N(u)| are the degrees of node v and node u, that is, the number of other nodes they are directly connected to. This normalization coefficient ensures that nodes with larger degrees will not have too much influence on their neighbors;

[0085] W (k) : The weight matrix of the kth layer is a learnable parameter in the GNN model, which is used to adjust the contribution of neighbor nodes to the target node in each layer. Through training, these weights can capture which neighbor nodes have a greater impact on the target node;

[0086] b (k) : Bias term, used to adjust the feature value of the node when updating each layer to avoid overfitting or deviation of the model;

[0087] σ: Activation function, used to introduce nonlinearity so that the model can learn complex relationships between nodes. Activation functions include ReLU, expressed as: σ(x) = max(0,x). This activation function enables the output of the model to learn in a nonlinear space, thereby enhancing the model's performance.

[0088] S2 specifically includes:

[0089] S21, user task content analysis: Use the word vector model to perform semantic analysis on the task content and project progress input by the user, extract the technical requirements in the project and the key technical terms in the current task. The technical terms include equipment names, process flows, and technical standards. Convert the technical terms into corresponding vector representations to generate semantic vector representations of the user's potential needs.

[0090] The word vector model transforms the word vector representation: v(w) = f(w), where v(w) is the vector representation of word w, and f(w) is the vector generated by the word vector model. Each word is mapped to a multidimensional vector space, and the distance between vectors represents the semantic similarity between words.

[0091] In the present invention, the project task content (such as construction plan, equipment selection) provided by the user is processed to extract key technical terms, which are converted into corresponding vector representations to form a semantic representation of the user's needs.

[0092] For example, the project task description contains the terms "reinforced concrete" and "prestressed structure". These words are processed by the BERT model to generate corresponding vectors v(reinforced concrete) and v(prestressed structure).

[0093] S22, semantic similarity calculation: Preliminary matching of the semantic vector representation of the user's potential needs with the technical nodes in the knowledge graph, and the semantic similarity between the technical terms in the user's task and the technical nodes in the knowledge graph is calculated by similarity, to determine the technical nodes closest to the user's needs as the basis for subsequent recommendations. After the semantic similarity calculation, the user's potential needs can be derived by parsing the key terms in the project progress and task content, and the technical support that the user may need, such as specific equipment or processes, can be inferred through the construction phase and technical requirements in the task. Based on the project progress (such as the construction phase) and task content (such as equipment selection) input by the user, the technical requirements that the user may not express explicitly but may need can be identified through semantic similarity. For example, if the user mentions "high-strength concrete", the system can infer that the user may need related "prestressed structure" technical support.

[0094] After obtaining the vector representation of technical terms in user tasks, it is necessary to calculate the semantic similarity between these terms and the technical nodes in the knowledge graph. The commonly used semantic similarity calculation method uses cosine similarity: Among them, v1 represents the vector representation of the technical term in the user task, such as v(reinforced concrete), and v2 represents the vector representation of the technical node in the knowledge graph, such as v(prefabricated panel); ‖v1‖‖v2‖ are the moduli of vectors v1 and v2 respectively, and v1·v2 is the inner product of the two vectors, indicating their directional similarity.

[0095] In the present invention, when the user inputs the current project task (such as "reinforced concrete structure is used during the construction phase"), the technical terms in the task are first extracted and converted into vectors, and then the related technical nodes (such as "precast concrete") are found from the knowledge graph. By calculating the cosine similarity between these terms and the technical nodes of the knowledge graph, the system can determine which technical nodes are closest to the user's task requirements.

[0096] Example: The term "reinforced concrete" in the user project task description is mapped to the vector v(reinforced concrete), and the node "prestressed structure" in the knowledge graph is mapped to the vector v(prestressed structure) by calculating the cosine similarity:

[0097] The similarity scores of the two terms are obtained, and if the similarity is high, it is considered that "prestressed structure" is related to the needs of "reinforced concrete" in the user task.

[0098] The alignment of the recommended content in S3 with the technical nodes of the knowledge graph is the same as above, using the same method, including semantic analysis of the recommended content extracted from the engineering information database, using the word vector model to convert the technical terms in the extracted recommended content into word vector representations, and generating semantic vector representations of the recommended content. Then, as above, the technical terms in the recommended content are preliminarily aligned with the matched technical nodes in the user project through semantic similarity calculation again, using the cosine similarity formula: Among them, v3 represents the technical term vector in the recommended content, and v4 represents the technical node vector matched by the user project in the knowledge graph. Their similarity is calculated to determine the correlation between the recommended content and user needs, and obtain preliminary recommended content.

[0099] S3 determines the matching degree between the initial recommended content and the technical nodes in the user project by reasoning about the associated edges of the knowledge graph, including:

[0100] Based on the graph neural network GNN, the associated edge reasoning in the knowledge graph is performed to infer the dependencies, collaborations and development paths between technology nodes, analyze the impact of neighboring nodes on the current technology nodes, and determine the matching depth between the recommended content and the technology nodes in the user's project.

[0101] The graph neural network GNN performs multi-layer reasoning on the technology nodes and their associated edges, analyzing the multi-layer relationship between the technology nodes in the recommended content and the matching technology nodes in the user's project. After multi-layer reasoning, it captures not only the direct semantic match between the recommended content and the technology nodes in the user's project, but also understands the technology chain, collaborative work plan, and technology development path on which they depend.

[0102] Inference example:

[0103] Assume that "reinforced concrete structure" is matched in the user project, and there are multiple nodes associated with it in the knowledge graph, such as "reinforcement production process" (dependency relationship) and "concrete mixing equipment" (collaborative working relationship). Through the multi-layer reasoning of the graph neural network, it is identified that "prestressed concrete" and "reinforced concrete structure" in the recommended content are similar technologies in development paths, and there is a collaborative working and dependency relationship between them. Therefore, the system determines that the recommended content is highly matched with the technical nodes in the user project in the technical chain, and the recommended content is technologically forward-looking;

[0104] After the graph neural network GNN inference, a matching depth score is assigned to the matching relationship between each recommended content and the technical node of the user project. The matching depth score is used to measure the relevance and distance of the recommended content in the technical chain.

[0105] The matching depth score is calculated as: Among them, Score depth is the final matching depth score, L is the number of layers of the graph neural network, indicating the depth of reasoning, and α k is the weight of each layer, indicating the contribution of a certain layer of reasoning, The feature vector representation of the k-th layer technology node v is generated after the associated edge reasoning, and the recommended content is filtered based on the matching depth score.

[0106] The matching depth score of the recommended content should be above 0.8, or the top 20% of the content should be selected based on the distribution of the overall score to ensure that the recommended content has a high correlation with the technical node of the user's project.

[0107] The semantic extension and technology completion in S4 include associated edge reasoning based on graph neural network GNN. For the matching technology nodes in the user project, multi-layer reasoning is performed to perform semantic extension to find potential related technology solutions or alternative technology paths associated with the technology node. The associated edges include technology dependencies, collaborative working relationships, and technology development paths. By analyzing neighbor nodes, technology nodes that are helpful to current project needs are identified. Through the neighbor node set of the current technology node, that is, other technology nodes directly associated with the current technology node, through the neighbor nodes, the extended information of the technology node is inferred, including alternative solutions or collaborative working nodes. Through semantic extension, neighbor nodes are found, and completed recommended content is generated. The technology nodes in the knowledge graph are combined with the expanded neighbor nodes to complete the technical background of the recommended content. These solutions include but are not limited to alternative processes, innovative equipment, or optimized processes.

[0108] S4 also includes judging the downstream technical requirements that will be triggered by the technical nodes in the user's project, and providing relevant information to the user in advance. The matching depth score is directly used as the triggered downstream technical demand score, thereby ensuring that the recommended content obtained by the user is forward-looking in the technology chain. In the downstream technical demand prediction, by analyzing the related edges in the knowledge graph, the subsequent technical path of the current technology node is judged, and relevant information is provided to the user in advance.

[0109] like Figure 2 As shown, a recommendation system for engineering information is used to implement the above-mentioned recommendation method for engineering information, and includes the following modules:

[0110] Knowledge graph construction module: According to the actual needs of the engineering field, it collects and processes engineering project data, technical documents and industry standards to build a knowledge graph covering the technology, equipment, process and standards in the engineering field. The knowledge graph includes multiple technical nodes, each of which represents a technical element or process flow in the engineering field. The nodes are connected through technical association edges.

[0111] User project background analysis module: Based on the information input by the user, it obtains relevant data of the current project, including project progress and task content, parses the information and analyzes the user's potential needs, preliminarily matches the potential needs with the technical nodes in the knowledge graph, and identifies the technical nodes related to the user's project;

[0112] Recommended content and technology node alignment module: extract recommended content from the engineering information database and preliminarily align it with the matched technology nodes. The alignment process includes semantic analysis of technical terms and technical solutions in the recommended content, generating preliminary recommended content, and reasoning by analyzing the associated edges in the knowledge graph to determine the matching degree between the recommended content and the technology nodes in the user's project. The preliminary recommended content is screened according to the matching degree to generate recommended content that meets user needs.

[0113] Semantic expansion and technology completion module: Based on the technology nodes in the knowledge graph and other related technology nodes, the semantic expansion mechanism generates completed recommended content. The semantic expansion finds potential technical solutions or alternative technical paths related to user needs through the associated edges of the technology nodes in the knowledge graph to complete the technical background of the recommended content.

[0114] Recommended content push module: pushes the final screened and completed recommended content to the user. The pushed content is screened to ensure that it matches the technical requirements of the user's current engineering project.

[0115] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0116] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for recommending engineering information, characterized in that: The following steps are involved: S1, Construction of knowledge graph in engineering field: Based on the actual needs in the engineering field, using engineering project data, technical documents, and industry standards, a knowledge graph covering the technology, equipment, process, and standards in the engineering field is constructed. The knowledge graph includes multiple technology nodes, each of which represents a technical element or process flow in an engineering field. The technology nodes are connected by technology-related edges, which include technology dependencies, collaborative working relationships, and development paths. S2, user project background analysis: based on user input, obtain the user's current project information, including project progress and task content, analyze the user's potential needs based on the user's current project information, preliminarily match the user's potential needs with the technical nodes in the knowledge graph, and identify the technical nodes in the user's project that match the knowledge graph; S3, Aligning recommended content with technical nodes in the knowledge graph: Perform preliminary alignment of the recommended content extracted from the engineering information database with the matching technical nodes. The preliminary alignment process includes semantic analysis of the technical terms and technical solutions in the recommended content to obtain preliminary recommended content, and reasoning on the associated edges of the knowledge graph to determine the matching degree between the recommended content and the technical nodes in the user's project. Based on the matching results, the preliminary recommended content is screened again, and the final recommended content is screened and pushed to the user; S4, semantic extension and technical completion: Based on the technology nodes in the knowledge graph and other associated technology nodes, the semantic extension mechanism is used to generate completed recommended content. Semantic extension includes finding potential related technical solutions or alternative technical paths for the user's current needs through the associated edges of the technology nodes in the knowledge graph, and completing the technical background of the recommended content to ensure that the recommended content not only covers the user's explicit needs, but also includes technical solutions related to the project's potential needs.

2. The method for recommending engineering information according to claim 1, characterized in that: The construction of the engineering domain knowledge graph in S1 specifically includes: S11, collect engineering project data and technical documents, use named entity recognition model to automatically extract technical elements and process flow, including equipment, materials, process flow, and store them in a structured manner, and generate technology nodes based on the extracted technical elements and process flow; S12, Industry Standard Docking: Combine industry standards and specification documents to verify whether the generated technical nodes comply with industry specifications through industry standard docking. Use the BM25 text matching algorithm to compare the identified technical terms with the terms in the industry standards. S13, technology-related edge construction: Use graph neural network GNN to model the relationship between technology nodes, and generate different types of technology-related edges through upstream and downstream dependencies, collaborative working relationships, and technology development paths between technology nodes. The technology-related edges include: Technology dependency edge: indicates the dependency of a technology node on other technology nodes when applied; Collaborative working edge: represents the relationship between two or more technology nodes being used collaboratively in a specific project; Development path edge: represents the technological evolution or alternative path between certain technology nodes, which is used to reflect the trend of technological development.

3. The method for recommending engineering information according to claim 1, characterized in that: The S2 specifically includes: S21, user task content analysis: Use the word vector model to perform semantic analysis on the task content and project progress input by the user, extract the technical requirements in the project and the technical terms in the current task. The technical terms include equipment names, process flows, and technical standards. Convert the technical terms into corresponding vector representations to generate semantic vector representations of the user's potential needs. S22, semantic similarity calculation: preliminarily match the semantic vector representation of the user's potential needs with the technical nodes in the knowledge graph, calculate the semantic similarity between the technical terms in the user's task and the technical nodes in the knowledge graph through similarity calculation, and determine the technical nodes closest to the user's needs.

4. The method for recommending engineering information according to claim 3, characterized in that: The recommended content in S3 is aligned with the technical nodes of the knowledge graph, including semantic analysis of the recommended content extracted from the engineering information database, converting the technical terms in the extracted recommended content into word vector representation using a word vector model, generating a semantic vector representation of the recommended content, and then calculating the semantic similarity to preliminarily align the technical terms in the recommended content with the matched technical nodes in the user project to obtain preliminary recommended content.

5. The method for recommending engineering information according to claim 4, characterized in that: The S3 determines the matching degree between the preliminary recommended content and the technical nodes in the user project by reasoning about the associated edges of the knowledge graph, specifically including: Based on the graph neural network GNN, the associated edge reasoning in the knowledge graph is performed to infer the dependencies, collaborations and development paths between technology nodes, analyze the impact of neighboring nodes on the current technology nodes, and determine the matching depth between the recommended content and the technology nodes in the user's project.

6. A method for recommending engineering information according to claim 5, characterized in that: The graph neural network GNN performs multi-layer reasoning on the technology nodes and their associated edges, analyzes the multi-layer relationship between the technology nodes in the recommended content and the matching technology nodes in the user project, and through multi-layer reasoning, captures not only the direct semantic match between the recommended content and the technology nodes in the user project, but also understands the technology chain, collaborative work plan and technology development path on which they depend; After the graph neural network GNN is inferred, a matching depth score is assigned to the matching relationship between each recommended content and the technical node of the user project. The matching depth score is used to measure the relevance and distance of the recommended content in the technical chain.

7. The method for recommending engineering information according to claim 6, characterized in that: The matching depth score is calculated as: Among them, Score depth is the final matching depth score, L is the number of layers of the graph neural network, indicating the depth of reasoning, and α k is the weight of each layer, indicating the contribution of a certain layer of reasoning, The feature vector representation of the k-th layer technology node v is generated after the associated edge reasoning, and the recommended content is filtered based on the matching depth score.

8. The method for recommending engineering information according to claim 7, characterized in that: The semantic extension and technology completion in S4 include associated edge reasoning based on graph neural network GNN, and multi-layer reasoning is performed on the matching technology nodes in the user project to perform semantic extension, find potential related technology solutions or alternative technology paths associated with the technology node, and the associated edges include technology dependencies, collaborative working relationships and technology development paths. By analyzing neighbor nodes, technology nodes that are helpful to current project needs are identified. Through the neighbor node set of the current technology node, through the neighbor nodes, the extended information of the technology node is inferred, including alternative solutions or collaborative working nodes. Through semantic extension, neighbor nodes are found, and completed recommended content is generated. The technology nodes in the knowledge graph are combined with the expanded neighbor nodes to complete the technical background of the recommended content.

9. The method for recommending engineering information according to claim 8, characterized in that: The S4 also includes determining the downstream technical requirements that will be triggered by the technical nodes in the user's project, and providing the user with relevant information in advance, directly using the matching depth score as the triggered downstream technical requirement score.

10. A system for recommending engineering information, used to implement a method for recommending engineering information as claimed in any one of claims 1 to 9, characterized in that: Includes the following modules: Knowledge graph construction module: According to the actual needs of the engineering field, it collects and processes engineering project data, technical documents and industry standards to build a knowledge graph covering the technology, equipment, process and standards in the engineering field. The knowledge graph includes multiple technology nodes, each of which represents a technical element or process flow in the engineering field, and the nodes are connected by technology-related edges; User project background analysis module: Based on the information input by the user, it obtains relevant data of the current project, including project progress and task content, parses the information and analyzes the user's potential needs, preliminarily matches the potential needs with the technical nodes in the knowledge graph, and identifies the technical nodes related to the user's project; Recommended content and technology node alignment module: extract recommended content from the engineering information database and preliminarily align it with the matched technology nodes. The alignment process includes semantic analysis of technical terms and technical solutions in the recommended content, generating preliminary recommended content, and reasoning by analyzing the associated edges in the knowledge graph to determine the matching degree between the recommended content and the technology nodes in the user's project. The preliminary recommended content is screened according to the matching degree to generate recommended content that meets user needs. Semantic expansion and technology completion module: Based on the technology nodes in the knowledge graph and other related technology nodes, the semantic expansion mechanism generates completed recommended content. The semantic expansion finds potential technical solutions or alternative technical paths related to user needs through the associated edges of the technology nodes in the knowledge graph to complete the technical background of the recommended content. Push module: pushes the final filtered and completed recommended content to the user. The pushed content is screened to ensure that it matches the technical requirements of the user's current engineering project.

Images