Workflow node recommendation method and apparatus, storage medium, and electronic device

Abstract

The application relates to a workflow node recommendation method and device, a storage medium and an electronic device. The method comprises the following steps: collecting key data of each function node in a platform, and performing semantic analysis and technical relationship analysis on the key data of each function node to construct a node knowledge graph; in the process that a user constructs a workflow through a graphical interface, a complete state snapshot of the current workflow is collected in real time, and the complete state snapshot is analyzed to construct an abstract syntax tree; the abstract syntax tree is deeply traversed and analyzed to extract structure features, intention features and user features, and a context query vector is generated by fusion; based on the context query vector, multi-path parallel retrieval is respectively performed in the node knowledge graph, a plurality of candidate node sets are obtained, each candidate node in each candidate node set is comprehensively scored and sorted, and a node recommendation result is generated. The application solves the technical problem that it is difficult to accurately understand the construction intention and guide node combination.

Description

Technical Field

[0001] This application relates to the field of computer technology, and more particularly to a recommended method, apparatus, storage medium, and electronic device for a workflow node. Background Technology

[0002] With the widespread adoption of low-code / no-code development platforms, an increasing number of applications, automated processes, and data processing workflows are being built using graphical drag-and-drop functionality. These platforms typically have a large number of built-in functional nodes to perform various functions such as data acquisition, processing, transformation, judgment, and output, thereby lowering the development threshold and improving build efficiency. However, as platform functionality continues to expand and the number and types of nodes continue to increase, users often need to filter and combine nodes from a massive pool during actual workflow construction, significantly increasing the difficulty of node selection.

[0003] In existing technologies, node recommendation or search methods are mostly based on keyword matching, tag classification, or simple rules, which make it difficult to accurately understand the user's specific intent at the current construction stage. This results in low relevance of recommendation results, requiring users to repeatedly try and adjust, thus affecting construction efficiency. Furthermore, during node combination and configuration, users need to determine whether the input / output interfaces between nodes are compatible, whether the data flow can be correctly connected, and whether the node combination suits the actual business scenario. Especially for inexperienced users, this can easily lead to connection errors, improper configuration, or unreasonable process design, thereby affecting the stability and execution quality of the workflow. Summary of the Invention

[0004] This application provides a method, apparatus, storage medium, and electronic device for recommending workflow nodes to solve the technical problem of difficulty in accurately understanding the construction intent and guiding node combination.

[0005] Firstly, this application provides a method for recommending workflow nodes, comprising: collecting key data of each functional node in the platform, and performing semantic parsing and technical relationship analysis on the key data of each functional node to construct a node knowledge graph containing node entities, functional concept entities and their functional semantic relationships, technical compatibility relationships and statistical co-occurrence relationships; during the process of users constructing workflows through a graphical interface, collecting a complete state snapshot of the current workflow in real time, and parsing the above complete state snapshot to construct an abstract syntax tree containing node entities, edge connection relationships and focus states; performing depth traversal and analysis on the above abstract syntax tree to extract structural features, intent features and user features, and fusing them to generate a context query vector; based on the above context query vector, performing multi-way parallel retrieval of functional semantic matching, technical interface adaptation and historical practice association in the above node knowledge graph to obtain multiple candidate node sets, and comprehensively scoring and ranking each candidate node in each candidate node set according to functional relevance, technical adaptation, node popularity and user preference to generate node recommendation results.

[0006] Secondly, this application provides a workflow node recommendation device, comprising: a first construction module, used to collect key data of each functional node in the platform, and perform semantic parsing and technical relationship analysis on the key data of each functional node to construct a node knowledge graph containing node entities, functional concept entities and their functional semantic relationships, technical compatibility relationships and statistical co-occurrence relationships; a second construction module, used to collect a complete state snapshot of the current workflow in real time during the process of the user constructing the workflow through a graphical interface, and parse the above complete state snapshot to construct an abstract syntax tree containing node entities, edge connection relationships and focus states; a first generation module, used to perform depth traversal and analysis on the above abstract syntax tree to extract structural features, intent features and user features, and fuse them to generate a context query vector; and a second generation module, used to perform multi-way parallel retrieval of functional semantic matching, technical interface adaptation and historical practice association in the above node knowledge graph based on the above context query vector, to obtain multiple candidate node sets, and to comprehensively score and rank each candidate node in each candidate node set according to functional relevance, technical adaptation, node popularity and user preference, and generate node recommendation results.

[0007] As an optional example, the first construction module includes: an acquisition unit, used to acquire key data of each functional node in the platform, wherein the key data includes node metadata, interface technical specification data, and historical usage data; a first construction unit, used to construct corresponding node entities for each functional node based on the node metadata; a second construction unit, used to perform natural language processing on the functional description text in the node metadata to extract functional semantic features representing the core capabilities of the nodes, construct the functional concept entities, and establish functional semantic relationships between functional nodes based on the similarity relationship between the functional semantic features corresponding to different functional nodes; and a third construction unit, used to acquire key data of each functional node in the platform, including node metadata, interface technical specification data, and historical usage data; a first construction unit, used to acquire key data of each functional node in the platform, including node metadata, interface technical specification data, and historical usage data; a second construction unit, used to acquire key data of each functional node in the platform, including node metadata, interface technical specification data, and historical usage data; a third construction unit, used to acquire key data of each functional node in the platform, including node metadata, interface technical specification data, and historical usage data; a fourth construction unit, used to acquire key data of each functional node in the platform, including node metadata, interface technical specification data, and historical usage data; a fifth construction unit, used to acquire key data of each functional node in the platform, including node metadata, interface technical specification data, and historical usage data; and a sixth construction unit, used to acquire key data of each functional node in the platform, including node metadata, interface technical specification data, and historical usage data; and a seventh construction unit, used to acquire key data of each functional node in the platform, including node metadata, interface technical specification data, and historical usage data; and ... The system consists of four parts: a data parsing unit and a storage unit. The first part parses the input and output interfaces of each functional node, analyzes the data type matching relationships and interface dependencies between different functional nodes, and establishes a technical compatibility relationship to represent the connectivity between nodes. The second part constructs the data based on the historical usage data, statistically analyzes the frequency of multiple functional nodes being used in the same workflow, and establishes a statistical co-occurrence relationship to represent common combination patterns between nodes based on the statistical analysis results. The third part builds the data based on the historical usage data, statistically analyzes the frequency of multiple functional nodes being used in the same workflow, and establishes a statistical co-occurrence relationship to represent common combination patterns between nodes based on the statistical analysis results. The fourth part builds the data based on the historical usage data, statistically analyzes the frequency of multiple functional nodes being used in the same workflow, and establishes a statistical co-occurrence relationship to represent common combination patterns between nodes based on the statistical analysis results. The fifth part builds the data based on the historical usage data, statistically analyzes the node entities, functional concept entities, functional semantic relationships, technical compatibility relationships, and statistical co-occurrence relationships, and establishes a unified model to form the node knowledge graph. The sixth part builds the data based on the historical usage data, statistically analyzes the node entities, functional concept entities, functional semantic relationships, technical compatibility relationships, and statistical co-occurrence relationships, and establishes a unified model to form the node knowledge graph. The seventh part builds the data based on the historical usage data, statistically analyzes ...

[0008] As an optional example, the second building module includes: a collection unit, used to collect in real time the user's operation information on the current workflow in the graphical interface through the plug-in interface of the integrated development environment or the front-end event listening mechanism, and obtain a complete state snapshot representing the current workflow editing state; a mapping unit, used to perform structured parsing on the complete state snapshot to map each functional node placed in the current workflow to node entities in the abstract syntax tree, and to map the connection relationship between functional nodes to edge connection relationship in the abstract syntax tree; a determination unit, used to determine the target node, port or connection position currently in the editing focus based on the user interaction information recorded in the complete state snapshot, and mark it as the focus state in the abstract syntax tree; and a fifth building unit, used to uniformly encapsulate the node entities, edge connection relationships and focus state to construct an abstract syntax tree representing the current workflow structure and editing state.

[0009] As an optional example, the first generation module includes: a first generation unit, used to traverse the abstract syntax tree, identify input ports and output ports that have not been connected in the current workflow, and infer the expected data type and structural constraints of the ports that have not been connected based on the corresponding interface definitions and their upstream or downstream node information, so as to generate structural features representing structural gaps in the workflow; a second generation unit, used to aggregate and analyze the functional semantics of the connected nodes in the current workflow based on the functional description information corresponding to each node entity in the abstract syntax tree, so as to generate intent features representing the business scenario of the current workflow; a third generation unit, used to generate user features representing the user's proficiency level and functional preferences based on the user's session information and historical construction behavior data; and a fourth generation unit, used to standardize and encode the structural features, intent features, and user features, and fuse them to generate the context query vector.

[0010] As an optional example, the second generation module includes: a first retrieval unit, configured to perform functional semantic matching retrieval in the node knowledge graph based on the intent features in the context query vector, to obtain a first candidate node set semantically related to the business scenario of the current workflow; a second retrieval unit, configured to perform technical interface adaptation retrieval in the node knowledge graph based on the structural features in the context query vector, to obtain a second candidate node set that matches the structural gaps of the current workflow in terms of interface type, data structure, or connection direction; a third retrieval unit, configured to perform historical practice association retrieval in the node knowledge graph based on the features representing the local structure of the current workflow in the context query vector, to obtain a third candidate node set that has a high co-occurrence rate with the structural pattern of the current workflow in historical usage data; and a summarization unit, configured to summarize the first candidate node set, the second candidate node set, and the third candidate node set to form multiple candidate node sets.

[0011] As an optional example, the second generation module includes: a processing unit, configured to determine unprocessed candidate nodes as current candidate nodes, and perform the following processing on the current candidate nodes: calculating a functional relevance score for the current candidate nodes based on the degree of matching between the current candidate nodes and the intent features in the context query vector; calculating a technical suitability score for the current candidate nodes based on the degree of matching between the current candidate nodes and the structural features in the context query vector; calculating a node popularity score for the current candidate nodes based on the frequency of use or adoption ratio of the current candidate nodes in historical usage data; calculating a user preference score for the current candidate nodes based on the degree of matching between the current candidate nodes and the user features in the context query vector; and weightedly fusing the functional relevance score, the technical suitability score, the node popularity score, and the user preference score to obtain a comprehensive score for the current candidate nodes.

[0012] As an optional example, after generating the node recommendation results, the apparatus further includes: a display module for displaying the node recommendation results on the graphical interface; an insertion module for inserting the target candidate node into the corresponding position of the current workflow and connecting related nodes when the user selects to load the target candidate node into the current workflow through the graphical interface, wherein the target candidate node is any candidate node in the node recommendation results; and an update module for recording the user's feedback data on the node recommendation results and updating the node knowledge graph and the node recommendation results based on the feedback data.

[0013] Thirdly, this application provides a storage medium storing a computer program, wherein the computer program is executed by a processor to perform the recommended method of the above-mentioned workflow nodes.

[0014] Fourthly, this application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to execute the recommended method of the workflow node described above through the computer program.

[0015] The technical solutions provided in this application have the following advantages compared with the prior art: This application utilizes key data from each functional node in the data acquisition platform and performs semantic parsing and technical relationship analysis on the key data of each functional node to construct a node knowledge graph containing node entities, functional concept entities, and their functional semantic relationships, technical compatibility relationships, and statistical co-occurrence relationships. During the user's workflow construction process through a graphical interface, a complete state snapshot of the current workflow is collected in real time, and this snapshot is parsed to construct an abstract syntax tree containing node entities, edge connections, and focus states. A depth-first traversal and analysis are performed on the abstract syntax tree to extract structural features, intent features, and user features, which are then fused to generate a context query vector. Based on the context query vector, the node knowledge graph is then analyzed. This method employs multi-path parallel retrieval, performing functional semantic matching, technical interface adaptation, and historical practice association to obtain multiple candidate node sets. It then comprehensively scores and ranks each candidate node in each set based on functional relevance, technical adaptation, node popularity, and user preference to generate node recommendation results. This method constructs a node knowledge graph encompassing node functional semantics, interface compatibility relationships, and historical co-occurrence relationships. It also abstractly models the user's current workflow structure, editing focus, and context, extracting structural features, intent features, and user features to form a unified context query vector. This allows for multi-dimensional joint retrieval and scoring ranking within the knowledge graph, generating accurate node recommendation results. This achieves the goal of intelligently recommending suitable nodes based on business semantics and historical practice while ensuring technical connectivity, reducing user trial-and-error costs, improving workflow construction efficiency, correctness, and overall user experience, and ultimately solving the technical problem of accurately understanding construction intent and guiding node combination. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0019] Figure 1This is a flowchart of a recommended method for an optional workflow node according to an embodiment of this application; Figure 2 This is a flowchart illustrating a specific implementation of a recommended method for an optional workflow node according to an embodiment of this application. Figure 3 This is a schematic diagram of the structure of a recommended device for an optional workflow node according to an embodiment of this application; Figure 4 This is a schematic diagram of an optional electronic device according to an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] The following disclosure provides numerous different embodiments or examples for implementing various structures of this application. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the scope of this application. Furthermore, reference numerals and / or letters may be repeated in different examples. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0022] According to a first aspect of the embodiments of this application, a method for recommending workflow nodes is provided, optionally, as follows: Figure 1 As shown, the above method includes: S102, collect key data of each functional node in the platform, and perform semantic parsing and technical relationship analysis on the key data of each functional node to construct a node knowledge graph that includes node entities, functional concept entities and their functional semantic relationships, technical compatibility relationships and statistical co-occurrence relationships; S104: During the process of users building workflows through the graphical interface, a complete state snapshot of the current workflow is collected in real time, and the complete state snapshot is parsed to construct an abstract syntax tree containing node entities, edge connection relationships and focus states. S106, Perform deep traversal and analysis on the abstract syntax tree to extract structural features, intent features and user features, and fuse them to generate a context query vector; S108, based on the context query vector, performs multi-path parallel retrieval in the node knowledge graph, including functional semantic matching, technical interface adaptation, and historical practice association, to obtain multiple candidate node sets. Then, it comprehensively scores and ranks each candidate node in each candidate node set according to functional relevance, technical adaptation, node popularity, and user preference, and generates node recommendation results.

[0023] Optionally, in this embodiment, a workflow node recommendation method for low-code / no-code development platforms is proposed. The aim is to provide intelligent and accurate node recommendation services based on a comprehensive understanding of node capabilities, workflow context, and user behavior during the process of users building applications, automated processes, or data processing workflows through graphical drag-and-drop. This reduces the difficulty of node selection and combination configuration and improves workflow construction efficiency and quality.

[0024] Specifically, such as Figure 2 The flowchart shown first illustrates the process of collecting key data from each functional node in the platform. This key data includes node metadata, interface technical specifications, and historical usage data. Based on the node metadata, corresponding node entities are constructed for each functional node. Semantic parsing of the node's functional description text is then performed to extract functional semantic features representing the node's core capabilities, thereby constructing functional concept entities. Furthermore, based on the similarity relationships between the functional semantic features of different nodes, functional semantic relationships are established between nodes. Simultaneously, based on the interface technical specifications data, the input and output interfaces of each functional node are parsed to analyze the compatibility between nodes in terms of data type, structure, and connection direction, establishing technical compatibility relationships. Based on historical usage data, statistical analysis is performed on the common use of nodes in the same workflow, establishing statistical co-occurrence relationships representing common node combination patterns. Through these methods, a node knowledge graph is constructed, containing node entities, functional concept entities, functional semantic relationships, technical compatibility relationships, and statistical co-occurrence relationships, providing a structured knowledge foundation for subsequent intelligent recommendations.

[0025] During the process of users building workflows through a graphical interface, a complete state snapshot of the current workflow is captured in real time. The state snapshot is then parsed, and each functional node in the current workflow is mapped to a node entity. The connection relationships between nodes are mapped to edge connections. The current editing focus state is determined in conjunction with the user's current editing operation. This results in the construction of an abstract syntax tree containing node entities, edge connections, and focus states, which is used to structurally represent the overall structure and editing state of the current workflow.

[0026] Building upon this foundation, a deep traversal and analysis of the abstract syntax tree is performed to extract multi-dimensional contextual features. Specifically, structural features representing gaps in the workflow structure are generated by identifying unconnected input and output ports and combining interface definitions and upstream / downstream node information. Intent features representing the current business scenario or construction intent are generated through aggregated analysis of the functional semantics of connected nodes in the current workflow. Simultaneously, user features representing user proficiency and functional preferences are generated by combining user session information and historical construction behavior data. These structural features, intent features, and user features are then standardized, encoded, and fused to form a unified contextual query vector.

[0027] Subsequently, based on the context query vector, multi-path parallel retrieval is performed in the node knowledge graph: on the one hand, functional semantic matching is performed based on intent features to obtain candidate nodes semantically related to the current business scenario; on the other hand, technical interface adaptation is performed based on structural features to obtain candidate nodes that match the gaps in the current workflow structure in terms of interface type and data structure; simultaneously, historical practice association retrieval is performed based on the local structural patterns of the current workflow to obtain candidate nodes with high co-occurrence in historical usage data. The results of the above multi-path retrieval are summarized to form a candidate node set.

[0028] Furthermore, for each candidate node, its functional relevance, technical compatibility, node popularity, and user preference are calculated separately, and the scores of each dimension are weighted and fused to obtain the comprehensive score of the candidate node; the candidate nodes are ranked according to the comprehensive score to generate node recommendation results, which are then presented to the user in an interactive manner.

[0029] Optionally, in this embodiment, by constructing a node knowledge graph and combining it with workflow context awareness, a precise understanding of the user's construction intent is achieved. This enables the recommendation of technically connectable and practically validated node solutions at appropriate construction stages, avoiding blind searching and trial and error. At the same time, through a multi-dimensional comprehensive scoring and ranking mechanism, the accuracy and usability of the recommendation results are improved, thereby significantly enhancing the efficiency, correctness, and user experience of workflow construction in low-code / no-code platforms.

[0030] As an optional example, key data from each functional node in the platform is collected, and semantic parsing and technical relationship analysis are performed on the key data of each functional node to construct a node knowledge graph containing node entities, functional concept entities, and their functional semantic relationships, technical compatibility relationships, and statistical co-occurrence relationships. Obtain key data for each functional node in the platform, including node metadata, interface technical specifications, and historical usage data. Based on node metadata, construct corresponding node entities for each functional node; Natural language processing is performed on the functional description text in the node metadata to extract functional semantic features that represent the core capabilities of the nodes, construct functional concept entities, and establish functional semantic relationships between functional nodes based on the similarity relationship between the functional semantic features corresponding to different functional nodes. Based on the interface technical specification data, the input and output interfaces of each functional node are parsed, and the data type matching relationship and interface dependency relationship between different functional nodes are analyzed to establish a technical compatibility relationship that characterizes the connectivity between nodes. Based on historical usage data, statistical analysis is performed on the frequency of multiple functional nodes being used together in the same workflow, and based on the statistical analysis results, a statistical co-occurrence relationship representing common combination patterns among nodes is established. By unifying the modeling of node entities, functional concept entities, functional semantic relationships, technical compatibility relationships, and statistical co-occurrence relationships, a node knowledge graph is formed. Store the node knowledge graph in a graph database.

[0031] Optionally, in this embodiment, a node knowledge graph supporting intelligent recommendation is constructed by collecting and analyzing key data from each functional node in the platform. First, key data for each functional node is obtained from the low-code / no-code platform. This key data includes the node's basic metadata, interface technical specifications, and historical usage data. Specifically, the node metadata describes the node's name, category, functional description, and applicable scenarios; the interface technical specifications characterize the node's input / output interface forms, data types, and constraints; and the historical usage data reflects the frequency and combination of node usage in real workflows.

[0032] Building upon this foundation, a corresponding node entity is constructed for each functional node based on its metadata, serving as the fundamental entity unit in the knowledge graph. Further, natural language processing is applied to the functional description text within the node metadata. Through word segmentation, semantic encoding, and key capability extraction, functional semantic features characterizing the core functions of the nodes are extracted, and functional concept entities are constructed accordingly. By calculating the similarity between the functional semantic features corresponding to different functional nodes, functional semantic relationships are established between node entities and functional concept entities, as well as between node entities themselves, thereby characterizing the relevance of nodes at the business capability level.

[0033] Simultaneously, based on the interface technical specifications, the input and output interfaces of each functional node are structurally analyzed, focusing on the data types, data formats, and dependency constraints supported by the interfaces. By comparing the matching of input and output interfaces between different nodes, the feasibility of direct or indirect connections is identified, and a technical compatibility relationship characterizing the connectability between nodes is established to reflect the technical feasibility of node combinations.

[0034] In addition, based on historical usage data, statistical analysis is conducted on the situation where multiple functional nodes are used together in the same workflow. The frequency and stability of node combinations are calculated, and statistical co-occurrence relationships are established based on the statistical results to reflect common node combination patterns and implicit usage experience that have been verified in practice in the platform.

[0035] Finally, node entities, functional concept entities, functional semantic relationships, technical compatibility relationships, and statistical co-occurrence relationships are modeled in a unified manner to form a structured and scalable node knowledge graph. The node knowledge graph is then stored in a graph database to provide efficient data support for subsequent node retrieval, matching, and recommendation based on semantic understanding and technical constraints.

[0036] As an optional example, a complete state snapshot of the current workflow is captured in real time, and the snapshot is parsed to construct an abstract syntax tree containing node entities, edge connections, and focus states, including: By using the plugin interface of the integrated development environment or the front-end event listening mechanism, the operation information of the user on the current workflow in the graphical interface is collected in real time, and a complete state snapshot representing the current workflow editing state is obtained. The complete state snapshot is structured and parsed to map each functional node placed in the current workflow to a node entity in the abstract syntax tree, and the connection relationship between functional nodes to an edge connection relationship in the abstract syntax tree. Based on the user interaction information recorded in the complete state snapshot, determine the target node, port, or connection location that is currently in the editing focus, and mark it as the focus state in the abstract syntax tree; The node entities, edge connections, and focus states are uniformly encapsulated to construct an abstract syntax tree that represents the current workflow structure and editing state.

[0037] Optionally, in this embodiment, during the process of the user building a workflow through the graphical interface, a complete state snapshot of the current workflow is collected in real time, and the complete state snapshot is parsed to construct an abstract syntax tree containing node entities, edge connection relationships, and focus states, thereby providing a structured foundation for subsequent intent understanding and node recommendation.

[0038] Specifically, user actions within the visual modeling interface are collected in real-time through plugin interfaces or front-end event listener mechanisms provided by the integrated development environment. These actions include, but are not limited to, dragging and dropping functional nodes, deleting or copying nodes, configuring node parameters, creating or breaking connections between nodes, and changes in cursor or selection states. Based on these actions, a complete snapshot of the current workflow's overall structure, node configuration status, and editing progress can be continuously acquired, ensuring the real-time nature and consistency of the collected information.

[0039] After obtaining a complete state snapshot, a structured parsing process is performed, converting the workflow elements presented graphically in the snapshot into an abstract representation. Specifically, each functional node placed in the current workflow is parsed and mapped into a node entity in an abstract syntax tree, and each node entity is associated with its corresponding node type, functional attributes, and interface information. At the same time, the data flow or control flow relationships formed by the connections between functional nodes are parsed into edge connections in the abstract syntax tree to characterize the execution order, data dependencies, or control dependencies between nodes, thereby completely restoring the topology of the workflow.

[0040] Furthermore, based on user interaction information recorded in the complete state snapshot, the user's current editing focus state is further identified. For example, based on information such as cursor position, selected node or port, and input or output port currently in a pending connection state, the target node, target interface, or potential connection location that the user is currently focusing on or operating on is determined, and this information is marked as a focus state attribute in the abstract syntax tree. The focus state reflects the user's immediate operational intent and is an important basis for subsequent intent feature extraction.

[0041] Finally, node entities, edge connections, and focus states are uniformly encapsulated and organized to construct an abstract syntax tree that can simultaneously reflect the current workflow structure and the user's real-time editing status. This abstract syntax tree representation transforms the dynamically changing workflow editing process into a stable, computable, structured model.

[0042] As an optional example, a deep traversal and analysis of the abstract syntax tree is performed to extract structural features, intent features, and user features, and these are then fused to generate a contextual query vector, including: The abstract syntax tree is traversed to identify input and output ports that have not been connected in the current workflow. Based on the corresponding interface definitions and their upstream or downstream node information, the expected data types and structural constraints of the ports that have not been connected are inferred to generate structural features that characterize the structural gaps in the workflow. Based on the functional description information corresponding to each node entity in the abstract syntax tree, the functional semantics of the connected nodes in the current workflow are aggregated and analyzed to generate intent features that represent the business scenario of the current workflow. Based on user conversation information and historical building behavior data, user characteristics are generated to represent the user's proficiency level and functional preferences. Structural features, intent features, and user features are standardized and encoded, then fused to generate a context query vector.

[0043] Optionally, in this embodiment, by performing a deep traversal and analysis of the abstract syntax tree, features that can characterize the current workflow state and user intent are extracted from multiple dimensions and fused to generate a context query vector for intelligent node recommendation. Specifically, firstly, a structured traversal of the abstract syntax tree is performed to identify all input and output ports in the current workflow that have not established connections. Based on the node interface definition and information about its upstream or downstream nodes, the expected data type, structural constraints, and possible input / output patterns of the unconnected ports are inferred, thereby generating structural features that characterize the gaps in the workflow structure and providing accurate connection constraint information for subsequent node matching.

[0044] Simultaneously, the functional description text corresponding to each node entity in the abstract syntax tree is taken as input. Semantic analysis is performed using a pre-trained lightweight text classification model (such as FastText), and the functional semantics of the connected nodes in the current workflow are aggregated and analyzed to output a quantified business scenario probability distribution vector, thus obtaining intent features reflecting the current business scenario or construction goal. This intent feature not only considers the functional capabilities of individual nodes but also comprehensively analyzes the combination patterns and data flow logic between nodes, thereby inferring the user's macro-level business goals and potential operational needs in the current construction phase.

[0045] Furthermore, by combining user conversation information and historical building behavior data, user characteristics are generated, including user proficiency level, frequently used nodes, and functional preferences. These user characteristics can be used for personalized node recommendations, ensuring that the recommendations not only meet the current workflow requirements but also align with user habits and experience.

[0046] After extracting structural, intent, and user features, each feature is standardized and encoded, such as through normalization or vectorization, and then merged into a unified high-dimensional context query vector. This context query vector comprehensively reflects the workflow's structural state, business intent, and user preferences, and can be directly used as input for subsequent multi-dimensional retrieval, matching, and ranking in the node knowledge graph, thereby achieving high-precision and intelligent node recommendation.

[0047] As an optional example, based on the context query vector, multi-path parallel retrieval is performed in the node knowledge graph, including functional semantic matching, technical interface adaptation, and historical practice association, resulting in multiple candidate node sets, including: Based on the intent features in the context query vector, perform functional semantic matching retrieval in the node knowledge graph to obtain the first candidate node set that is semantically related to the business scenario of the current workflow; Based on the structural features in the context query vector, perform technical interface adaptation retrieval in the node knowledge graph to obtain a set of second candidate nodes that match the structural gaps in the current workflow in terms of interface type, data structure, or connection direction. Based on the features representing the local structure of the current workflow in the context query vector, a historical practice association retrieval is performed in the node knowledge graph to obtain a set of third candidate nodes that have a high co-occurrence with the structural pattern of the current workflow in historical usage data. The first set of candidate nodes, the second set of candidate nodes, and the third set of candidate nodes are aggregated to form multiple sets of candidate nodes.

[0048] Optionally, in this embodiment, multi-path parallel retrieval is performed in the node knowledge graph based on context query vectors to generate a candidate node set for recommendation from multiple dimensions, thereby achieving intelligent and accurate node matching. Specifically, firstly, functional semantic matching retrieval is performed in the node knowledge graph using intent features extracted from the context query vectors. This retrieval identifies nodes that can meet the user's macro-business goals and local functional needs by calculating the semantic similarity between the current workflow business scenario and the functional descriptions of each node, forming a first candidate node set, thereby ensuring that the candidate nodes are highly consistent with the user's intent at the functional semantic level.

[0049] Secondly, based on the structural features in the context query vector, a technical interface adaptation retrieval is performed in the node knowledge graph. This retrieval mainly evaluates the matching of the input interface, output interface, data type, and data structure of each candidate node with the unconnected ports in the current workflow, and filters out nodes that can seamlessly connect with the existing structure in terms of interface type, data format, and connection direction, forming a second set of candidate nodes. This ensures that the recommended nodes can be directly used in terms of technical connection, reducing user debugging costs.

[0050] Furthermore, information representing the local structural features of the current workflow in the context query vector is used to perform historical practice association retrieval in the node knowledge graph. This retrieval is based on analyzing the co-occurrence frequency of node combination patterns using historical usage data, identifying nodes with high co-occurrence rates with the local node patterns already built in the current workflow in the past, and forming a third candidate node set. This leverages empirical knowledge to improve the feasibility and rationality of recommended nodes.

[0051] Finally, the first, second, and third candidate node sets are aggregated and deduplicated to generate multiple candidate node sets. This multi-path parallel retrieval method considers recommendation strategies simultaneously from three dimensions: functional semantics, technical interfaces, and historical practices. It ensures the semantic relevance of nodes while also taking into account interface compatibility and the feasibility of combination patterns, thereby significantly improving the accuracy and intelligence of node recommendations, reducing user trial-and-error costs, and increasing workflow construction efficiency.

[0052] As an optional example, a comprehensive score is given to each candidate node in each candidate node set based on functional relevance, technical compatibility, node popularity, and user preference, including: The unprocessed candidate node is identified as the current candidate node, and the following processing is performed on the current candidate node: The functional relevance score of the current candidate node is calculated based on the degree of matching between the current candidate node and the intent features in the context query vector. Based on the degree of matching between the current candidate node and the structural features in the context query vector, calculate the technical suitability score of the current candidate node; Calculate the node popularity score of the current candidate node based on the frequency of use or the proportion of adoption of the current candidate node in historical usage data. Calculate the user preference score for the current candidate node based on the degree of matching between the current candidate node and the user features in the context query vector. The functional relevance score, technical compatibility score, node popularity score, and user preference score are weighted and fused to obtain the comprehensive score of the current candidate node.

[0053] Optionally, in this embodiment, a comprehensive scoring process is performed on each candidate node. Specifically, in terms of functionality, a matching analysis is conducted based on the intent features extracted from the context query vector and the current candidate node to calculate a functional relevance score. This score reflects the degree to which the node's semantic function aligns with the current workflow's business objectives and user intent, ensuring that the recommended nodes can meet the user's actual needs in the macro-business scenario.

[0054] From a technical perspective, based on the structural features contained in the context query vector, the interface type, data format, port direction, and data flow compatibility of the current candidate nodes are analyzed, and a technical suitability score is calculated. This score is used to evaluate whether the candidate nodes can seamlessly connect with existing workflow nodes, thereby reducing the risk of interface mismatch or connection errors.

[0055] In terms of popularity, a node popularity score is generated by referencing the frequency of use or adoption rate of candidate nodes in historical usage data. This score reflects the stability and reliability of nodes in practical applications, ensuring that the recommendation results have empirically validated feasibility.

[0056] In terms of user preferences, user preference scores are calculated by combining user feature information in the context query vector to evaluate whether candidate nodes conform to the current user's operating habits, commonly used node patterns, and historical preferences.

[0057] Finally, the functional relevance score, technical compatibility score, node popularity score, and user preference score are weighted and fused to generate a comprehensive score for the current candidate node. By performing the above process on all nodes in the candidate node set, a complete list of nodes with comprehensive scores can be formed, and they can be prioritized according to the comprehensive scores. This achieves intelligent node recommendation that meets business needs while also considering technical feasibility and user habits. This multi-dimensional weighted scoring method not only improves recommendation accuracy but also effectively reduces the trial-and-error costs for users in the workflow construction process, improving workflow design efficiency and user experience.

[0058] As an optional example, after generating the node recommendation results, the above method also includes: Display the node recommendation results in a graphical interface; If the system detects that the user has selected to load the target candidate node into the current workflow through the graphical interface, the target candidate node is inserted into the corresponding position in the current workflow and connected to the relevant nodes. The target candidate node is any candidate node in the node recommendation results. Record user feedback data on node recommendation results, and update the node knowledge graph and node recommendation results based on the feedback data.

[0059] Optionally, in this embodiment, as Figure 2 The flowchart shown further illustrates how an interactive visual interface enables a closed-loop process for displaying recommendation results and providing user feedback. Specifically, the node recommendation results, after multi-dimensional comprehensive scoring and ranking, are first presented graphically in the user's workflow editing interface. This allows users to intuitively browse the functional descriptions, interface types, and recommendation priorities of each candidate node, thereby assisting them in quickly selecting the most suitable node.

[0060] When a user selects a target candidate node through a graphical interface to load into the current workflow, the selected node is automatically inserted into its corresponding position within the workflow. Based on the interface definitions between nodes and the structural constraints of the workflow, relevant upstream and downstream nodes are intelligently connected, ensuring the compatibility of the newly inserted node with the existing workflow topology and the continuity of data flow. Context awareness and structural analysis guarantee the accuracy and seamless integration of the insertion operation, reducing the cost of manual debugging and repetitive operations for users.

[0061] Furthermore, the system automatically records user feedback data on node recommendation results during user operations, including which nodes were adopted, which were ignored, and the nodes ultimately manually selected. Based on this feedback data, the node knowledge graph can be dynamically updated in the background. This includes strengthening the functional or combined relationships between frequently adopted nodes, correcting interface adaptation rules, and updating historical usage statistics for nodes. Simultaneously, the node recommendation model can be optimized, such as adjusting the weights of various rating dimensions or optimizing semantic matching strategies, so that future recommendations better align with users' actual needs and usage habits.

[0062] Optionally, this embodiment not only realizes the visualization and convenient operation of node recommendation results, but also establishes an adaptive optimization mechanism based on user feedback, so that the node knowledge graph and recommendation model can continuously evolve, thereby improving the accuracy, intelligence level and workflow construction efficiency of recommendations, significantly improving user experience and reducing the error rate in the construction process.

[0063] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0064] According to another aspect of the embodiments of this application, a workflow node recommendation device is also provided, such as... Figure 3 As shown, it includes: The first construction module 302 is used to collect key data of each functional node in the platform, and to perform semantic parsing and technical relationship analysis on the key data of each functional node in order to construct a node knowledge graph containing node entities, functional concept entities and their functional semantic relationships, technical compatibility relationships and statistical co-occurrence relationships. The second construction module 304 is used to collect a complete state snapshot of the current workflow in real time during the process of the user building the workflow through the graphical interface, and to parse the complete state snapshot in order to build an abstract syntax tree containing node entities, edge connection relationships and focus states. The first generation module 306 is used to perform deep traversal and analysis of the abstract syntax tree to extract structural features, intent features and user features, and fuse them to generate a context query vector. The second generation module 308 is used to perform multi-path parallel retrieval in the node knowledge graph based on the context query vector, including functional semantic matching, technical interface adaptation, and historical practice association, to obtain multiple candidate node sets. Based on functional relevance, technical adaptation, node popularity, and user preference, the module comprehensively scores and ranks each candidate node in each candidate node set to generate node recommendation results.

[0065] It should be noted that the first building module 302 in this embodiment can be used to execute step S102 in this application embodiment, the second building module 304 in this embodiment can be used to execute step S104 in this application embodiment, the first generation module 306 in this embodiment can be used to execute step S106 in this application embodiment, and the second generation module 308 in this embodiment can be used to execute step S108 in this application embodiment.

[0066] As an optional example, the first building block includes: The acquisition unit is used to acquire key data of each functional node in the platform. The key data includes node metadata, interface technical specification data, and historical usage data. The first building unit is used to build corresponding node entities for each functional node based on node metadata; The second building unit is used to perform natural language processing on the functional description text in the node metadata to extract functional semantic features that represent the core capabilities of the node, construct functional concept entities, and establish functional semantic relationships between functional nodes based on the similarity relationship between the functional semantic features corresponding to different functional nodes. The third building unit is used to parse the input and output interfaces of each functional node based on the interface technical specification data, analyze the data type matching relationship and interface dependency relationship between different functional nodes, and establish a technical compatibility relationship that characterizes the connectivity between nodes. The fourth building unit is used to perform statistical analysis on the frequency of multiple functional nodes being used in the same workflow based on historical usage data, and to establish statistical co-occurrence relationships that characterize common combination patterns among nodes based on the statistical analysis results. The modeling unit is used to uniformly model node entities, functional concept entities, functional semantic relationships, technical compatibility relationships, and statistical co-occurrence relationships to form a node knowledge graph; Storage unit, used to store node knowledge graphs into graph database.

[0067] As an optional example, the second building block includes: The data acquisition unit is used to collect user operation information on the current workflow in the graphical interface in real time through the plugin interface of the integrated development environment or the front-end event listening mechanism, and to obtain a complete state snapshot representing the current workflow editing state. The mapping unit is used to perform structured parsing of the complete state snapshot, so as to map each functional node placed in the current workflow to node entities in the abstract syntax tree, and to map the connection relationship between functional nodes to edge connection relationship in the abstract syntax tree. The determination unit is used to determine the target node, port, or connection location currently under editing focus based on the user interaction information recorded in the complete state snapshot, and mark it as the focus state in the abstract syntax tree; The fifth building unit is used to encapsulate node entities, edge connections, and focus states in a unified manner, and to construct an abstract syntax tree that represents the current workflow structure and editing state.

[0068] As an optional example, the first generation module includes: The first generation unit is used to traverse the abstract syntax tree, identify the input and output ports that have not been connected in the current workflow, and infer the expected data type and structural constraints of the ports that have not been connected based on the corresponding interface definition and their upstream or downstream node information, so as to generate structural features that characterize the structural gaps in the workflow. The second generation unit is used to aggregate and analyze the functional semantics of the connected nodes in the current workflow based on the functional description information corresponding to each node entity in the abstract syntax tree, and generate intent features that represent the business scenario of the current workflow. The third generation unit is used to generate user characteristics that represent the user's proficiency level and functional preferences based on the user's session information and historical building behavior data. The fourth generation unit is used to standardize and encode structural features, intent features, and user features, and then fuse them to generate a context query vector.

[0069] As an optional example, the second generation module includes: The first retrieval unit is used to perform functional semantic matching retrieval in the node knowledge graph based on the intent features in the context query vector, so as to obtain the first candidate node set that is semantically related to the business scenario of the current workflow. The second retrieval unit is used to perform technical interface adaptation retrieval in the node knowledge graph based on the structural features in the context query vector, so as to obtain a set of second candidate nodes that match the structural gaps in the current workflow in terms of interface type, data structure or connection direction. The third retrieval unit is used to perform historical practice association retrieval in the node knowledge graph based on the features representing the local structure of the current workflow in the context query vector, so as to obtain a set of third candidate nodes that have a high co-occurrence with the structural pattern of the current workflow in historical usage data. The aggregation unit is used to aggregate the first candidate node set, the second candidate node set, and the third candidate node set to form multiple candidate node sets.

[0070] As an optional example, the second generation module includes: The processing unit is used to determine the unprocessed candidate node as the current candidate node and perform the following processing on the current candidate node: The functional relevance score of the current candidate node is calculated based on the degree of matching between the current candidate node and the intent features in the context query vector. Based on the degree of matching between the current candidate node and the structural features in the context query vector, calculate the technical suitability score of the current candidate node; Calculate the node popularity score of the current candidate node based on the frequency of use or the proportion of adoption of the current candidate node in historical usage data. Calculate the user preference score for the current candidate node based on the degree of matching between the current candidate node and the user features in the context query vector. The functional relevance score, technical compatibility score, node popularity score, and user preference score are weighted and fused to obtain the comprehensive score of the current candidate node.

[0071] As an optional example, after generating the node recommendation results, the above apparatus further includes: The display module is used to show the node recommendation results in a graphical interface; The insertion module is used to insert the target candidate node into the corresponding position in the current workflow and connect the relevant nodes when the user selects to load the target candidate node into the current workflow through the graphical interface. The target candidate node is any candidate node in the node recommendation results. The update module is used to record user feedback data on node recommendation results and update the node knowledge graph and node recommendation results based on the feedback data.

[0072] For other examples of this embodiment, please refer to the examples above, which will not be repeated here.

[0073] Figure 4 This is a schematic diagram of an optional electronic device according to an embodiment of this application, such as... Figure 4As shown, it includes a processor 402, a communication interface 404, a memory 406, and a communication bus 408. The processor 402, communication interface 404, and memory 406 communicate with each other via the communication bus 408. Memory 406 is used to store computer programs; When processor 402 executes a computer program stored in memory 406, it performs the following steps: Collect key data from each functional node in the platform, and perform semantic parsing and technical relationship analysis on the key data of each functional node to construct a node knowledge graph that includes node entities, functional concept entities and their functional semantic relationships, technical compatibility relationships and statistical co-occurrence relationships; During the process of users building workflows through the graphical interface, a complete state snapshot of the current workflow is collected in real time, and the complete state snapshot is parsed to construct an abstract syntax tree containing node entities, edge connection relationships, and focus states. The abstract syntax tree is deeply traversed and analyzed to extract structural features, intent features, and user features, and then fused to generate a context query vector. Based on context query vectors, multi-path parallel retrieval is performed in the node knowledge graph, including functional semantic matching, technical interface adaptation, and historical practice association, to obtain multiple candidate node sets. Then, each candidate node in each candidate node set is comprehensively scored and ranked according to functional relevance, technical adaptation, node popularity, and user preference to generate node recommendation results.

[0074] Optionally, in this embodiment, the communication bus can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 The symbol is represented by a single thick line, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0075] The memory may include RAM, or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0076] As an example, the memory 406 described above may include, but is not limited to, the first building module 302, the second building module 304, the first generation module 306, and the second generation module 308 in the workflow node recommendation device described above. Furthermore, it may include, but is not limited to, other module units in the workflow node recommendation device described above, which will not be elaborated upon in this example.

[0077] The processor mentioned above can be a general-purpose processor, including but not limited to: CPU (Central Processing Unit), NP (Network Processor), etc.; it can also be DSP (Digital Signal Processor), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0078] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0079] Those skilled in the art will understand that Figure 4 The structure shown is for illustrative purposes only. The device implementing the recommended method for the above workflow nodes can be a terminal device, such as a smartphone (e.g., Android phone, iOS phone), tablet computer, PDA, mobile Internet Devices (MID), PAD, etc. Figure 4 This does not limit the structure of the aforementioned electronic devices. For example, the electronic device may also include components that are more... Figure 4 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 4 The different configurations shown.

[0080] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, ROM, RAM, disk or optical disk, etc.

[0081] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, wherein a computer program is stored therein, wherein the computer program is executed by a processor to perform the steps in the recommended method of the above-described workflow node.

[0082] Optionally, in this embodiment, those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0083] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0084] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, 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. This computer software product is stored in a storage medium and includes several instructions to cause one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0085] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0086] In the several embodiments provided in this application, it should be understood that the disclosed client can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between units or modules, and may be electrical or other forms.

[0087] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0088] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0089] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for recommending workflow nodes, characterized in that, include: Collect key data from each functional node in the platform, and perform semantic parsing and technical relationship analysis on the key data of each functional node to construct a node knowledge graph that includes node entities, functional concept entities and their functional semantic relationships, technical compatibility relationships and statistical co-occurrence relationships; During the process of users building workflows through the graphical interface, a complete state snapshot of the current workflow is collected in real time, and the complete state snapshot is parsed to construct an abstract syntax tree containing node entities, edge connection relationships, and focus states. The abstract syntax tree is subjected to depth traversal and analysis to extract structural features, intent features, and user features, and these features are then fused to generate a context query vector. Based on the context query vector, multi-path parallel retrieval is performed in the node knowledge graph, including functional semantic matching, technical interface adaptation, and historical practice association, to obtain multiple candidate node sets. Then, each candidate node in each candidate node set is comprehensively scored and ranked according to functional relevance, technical adaptation, node popularity, and user preference to generate node recommendation results.

2. The method according to claim 1, characterized in that, Key data from each functional node in the platform is collected, and semantic parsing and technical relationship analysis are performed on the key data of each functional node to construct a node knowledge graph that includes node entities, functional concept entities, and their functional semantic relationships, technical compatibility relationships, and statistical co-occurrence relationships. Obtain key data for each functional node in the platform, including node metadata, interface technical specifications, and historical usage data; Based on the node metadata, construct corresponding node entities for each functional node; Natural language processing is performed on the functional description text in the node metadata to extract functional semantic features that represent the core capabilities of the node, construct the functional concept entity, and establish functional semantic relationships between functional nodes based on the similarity relationship between the functional semantic features corresponding to different functional nodes. Based on the interface technical specification data, the input and output interfaces of each functional node are parsed, and the data type matching relationship and interface dependency relationship between different functional nodes are analyzed to establish a technical compatibility relationship that characterizes the connectivity between nodes. Based on the historical usage data, a statistical analysis is performed on the frequency of multiple functional nodes being used together in the same workflow, and a statistical co-occurrence relationship characterizing common combination patterns among nodes is established based on the statistical analysis results. The node entities, functional concept entities, functional semantic relationships, technical compatibility relationships, and statistical co-occurrence relationships are modeled in a unified manner to form the node knowledge graph. The node knowledge graph is stored in a graph database.

3. The method according to claim 1, characterized in that, Real-time acquisition of a complete state snapshot of the current workflow, and parsing of the complete state snapshot to construct an abstract syntax tree containing node entities, edge connections, and focus states, including: Through the plugin interface of the integrated development environment or the front-end event listening mechanism, the operation information of the user on the current workflow in the graphical interface is collected in real time, and a complete state snapshot representing the current workflow editing state is obtained. The complete state snapshot is structured and parsed to map each functional node placed in the current workflow to a node entity in the abstract syntax tree, and the connection relationship between functional nodes to an edge connection relationship in the abstract syntax tree. Based on the user interaction information recorded in the complete state snapshot, determine the target node, port, or connection location that is currently in the editing focus, and mark it as the focus state in the abstract syntax tree; The node entities, edge connections, and focus states are uniformly encapsulated to construct an abstract syntax tree that represents the current workflow structure and editing state.

4. The method according to claim 1, characterized in that, The abstract syntax tree is subjected to depth-first traversal and analysis to extract structural features, intent features, and user features, which are then fused to generate a contextual query vector, including: The abstract syntax tree is traversed to identify input and output ports that have not been connected in the current workflow. Based on the corresponding interface definition and its upstream or downstream node information, the expected data type and structural constraints of the ports that have not been connected are inferred to generate structural features that characterize the structural gaps in the workflow. Based on the functional description information corresponding to each node entity in the abstract syntax tree, the functional semantics of the connected nodes in the current workflow are aggregated and analyzed to generate intent features that represent the business scenario of the current workflow. Based on the user's session information and historical construction behavior data, user characteristics representing the user's proficiency level and functional preferences are generated. The structural features, intent features, and user features are standardized and encoded, and then fused to generate the context query vector.

5. The method according to claim 1, characterized in that, Based on the context query vector, multi-path parallel retrieval is performed in the node knowledge graph, including functional semantic matching, technical interface adaptation, and historical practice association, to obtain multiple candidate node sets, including: Based on the intent features in the context query vector, a functional semantic matching retrieval is performed in the node knowledge graph to obtain a first candidate node set that is semantically related to the business scenario of the current workflow. Based on the structural features in the context query vector, a technical interface adaptation retrieval is performed in the node knowledge graph to obtain a second set of candidate nodes that match the structural gaps in the current workflow in terms of interface type, data structure, or connection direction. Based on the features representing the local structure of the current workflow in the context query vector, a historical practice association retrieval is performed in the node knowledge graph to obtain a set of third candidate nodes that have a high co-occurrence with the structural pattern of the current workflow in historical usage data. The first set of candidate nodes, the second set of candidate nodes, and the third set of candidate nodes are aggregated to form multiple sets of candidate nodes.

6. The method according to claim 1, characterized in that, The candidate nodes in each candidate node set are comprehensively scored based on functional relevance, technical compatibility, node popularity, and user preference, including: The unprocessed candidate node is identified as the current candidate node, and the following processing is performed on the current candidate node: Based on the degree of matching between the current candidate node and the intent features in the context query vector, the functional relevance score of the current candidate node is calculated; Based on the degree of matching between the current candidate node and the structural features in the context query vector, the technical suitability score of the current candidate node is calculated. The node popularity score of the current candidate node is calculated based on the frequency of use or the proportion of adoption of the current candidate node in historical usage data. Based on the degree of matching between the current candidate node and the user features in the context query vector, calculate the user preference score of the current candidate node; The functional relevance score, the technology compatibility score, the node popularity score, and the user preference score are weighted and fused to obtain the comprehensive score of the current candidate node.

7. The method according to any one of claims 1 to 6, characterized in that, After generating the node recommendation results, the method further includes: The node recommendation results are displayed in the graphical interface; When it is detected that the user selects to load the target candidate node into the current workflow through the graphical interface, the target candidate node is inserted into the corresponding position of the current workflow and connected to the relevant nodes, wherein the target candidate node is any candidate node in the node recommendation results; Record the user's feedback data on the node recommendation results, and update the node knowledge graph and the node recommendation results based on the feedback data.

8. A workflow node recommendation device, characterized in that, include: The first construction module is used to collect key data of each functional node in the platform, and to perform semantic parsing and technical relationship analysis on the key data of each functional node in order to construct a node knowledge graph containing node entities, functional concept entities and their functional semantic relationships, technical compatibility relationships and statistical co-occurrence relationships. The second construction module is used to collect a complete state snapshot of the current workflow in real time during the process of the user building the workflow through the graphical interface, and to parse the complete state snapshot to construct an abstract syntax tree containing node entities, edge connection relationships and focus states. The first generation module is used to perform deep traversal and analysis on the abstract syntax tree to extract structural features, intent features and user features, and fuse them to generate a context query vector. The second generation module is used to perform multi-path parallel retrieval of functional semantic matching, technical interface adaptation and historical practice association in the node knowledge graph based on the context query vector, to obtain multiple candidate node sets, and to comprehensively score and rank each candidate node in each candidate node set according to functional relevance, technical adaptation, node popularity and user preference, and generate node recommendation results.

9. A computer-readable storage medium storing a computer program, characterized in that, The computer program is executed by the processor to perform the method described in any one of claims 1 to 7.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the method described in any one of claims 1 to 7 through the computer program.

Images