Construction method and device of low-code operation semantic library

Abstract

The invention provides a low-code operation semantic library construction method and device, and relates to the field of low-code and knowledge engineering cross technologies, and the method comprises the steps: obtaining historical operation flow data of a user on a low-code platform, and converting the historical operation flow data into a semantic primitive related to a user operation component; performing rule mining on the plurality of semantic primitives to obtain an operation association rule; based on the semantic primitive and the operation association rule, carrying out fusion with knowledge of a preset vertical field, and generating an operation semantic library adaptive to the preset vertical field; and updating the semantic primitives and / or the operation association rules in the operation semantic library based on the newly added operation flow data and / or the user feedback data. According to the method and the device provided by the invention, through deep fusion of universal operation and vertical domain knowledge and continuous optimization by using feedback data, the accuracy and business fitting degree of functions such as intelligent recommendation are greatly improved, the semantic library can automatically adapt to changes of components and businesses, and the development efficiency of a low-code platform is improved.

Description

Technical Field

[0001] This invention relates to the field of low-code and knowledge engineering intersection technology, and in particular to a method and apparatus for constructing a low-code operation semantic library. Background Technology

[0002] As low-code platforms are increasingly used across various industries, including telecommunications, manufacturing, finance, and government, a wealth of operational data has accumulated. Currently, in low-code platforms, user actions such as dragging, connecting, and configuring components contain business logic, but existing technologies only record these actions as logs, failing to extract them into reusable semantic knowledge. Furthermore, the disconnect between operations and business logic means low-code platforms cannot understand the mapping from operation sequences to business intent, limiting intelligent recommendation and automated assembly capabilities. Finally, poor semantic reusability requires new users to relearn, hindering effective development efficiency.

[0003] Therefore, how to improve the intelligent assistance capabilities and development efficiency of low-code platforms has become a technical problem that the industry urgently needs to solve. Summary of the Invention

[0004] This invention provides a method and apparatus for constructing a low-code operation semantic library, which addresses the shortcomings of existing low-code platforms such as poor semantic usability, insufficient domain adaptation, and low intelligent assistance capabilities, thereby improving the intelligent assistance capabilities and development efficiency of low-code platforms.

[0005] This invention provides a method for constructing a low-code manipulation semantic library, comprising: Obtain the user's historical operation flow data on the low-code platform and transform the historical operation flow data into semantic primitives related to the user's operation components; Rule mining is performed on multiple semantic primitives to obtain operation association rules; Based on the semantic primitives and the operation association rules, respectively, they are fused with knowledge of a preset vertical domain to generate an operation semantic library adapted to the preset vertical domain; the operation semantic library is used to assist users in performing operations on the low-code platform and generate new operation flow data. Based on the newly added operation flow data and / or user feedback data, update the semantic primitives and / or operation association rules in the operation semantic library.

[0006] In some embodiments, acquiring historical operation flow data of the user on a low-code platform and converting the historical operation flow data into semantic primitives related to user operation components includes: Based on the historical operation flow data, determine the components related to user operations, as well as the component attributes and timing markers of each component; Parse the component properties of each component to generate the component semantics for each component; Analyze the timing markers of each component to generate the timing relationships between the components; Based on the component semantics of each component and the temporal relationships between each component, the semantic primitives related to the user operation components are generated.

[0007] In some embodiments, the step of performing rule mining on multiple semantic primitives to obtain operation association rules includes: Based on the contextual features of the multiple semantic primitives, the operational intent weight corresponding to each semantic primitive is determined; Frequent item mining is performed on the multiple semantic primitives, and the operation association rules are determined based on the mined frequent itemsets; The support of the frequent itemsets is determined based on the operational intent weights corresponding to each semantic primitive.

[0008] In some embodiments, the step of fusing the semantic primitives and the operation association rules with knowledge from a preset vertical domain to generate an operation semantic library adapted to the preset vertical domain includes: Determine the knowledge graph of the preset vertical domain; the knowledge graph includes business entities and the business logic between the business entities. Establish a mapping relationship between the semantic primitives and the business entities; Establish a mapping relationship between the operation association rules and the business logic.

[0009] In some embodiments, the method further includes: Based on the bidirectional attention weights between the semantic primitives and the business entities in the preset vertical domain, the mapping probability between the semantic primitives and the business entities is determined. The mapping probability is used to characterize the stability of the mapping relationship.

[0010] In some embodiments, updating the semantic primitives and / or operation association rules in the operation semantic library based on the newly added operation flow data and / or user feedback data includes: Based on the newly added operation flow data and / or user feedback data, determine the accuracy of each semantic primitive in the operation semantic library and the applicability of each operation association rule; Based on the accuracy of each semantic primitive and the applicability of each operation association rule, determine the semantic primitives and / or operation association rules to be optimized. Update the semantic primitives and / or operation association rules to be optimized.

[0011] This invention provides an apparatus for building a low-code manipulation semantic library, comprising: The acquisition module is used to acquire the user's historical operation flow data on the low-code platform and convert the historical operation flow data into semantic primitives related to the user's operation components. The mining module is used to perform rule mining on multiple semantic primitives to obtain operation association rules; The adaptation module is used to fuse the semantic primitives and operation association rules with knowledge of a preset vertical domain to generate an operation semantic library adapted to the preset vertical domain; the operation semantic library is used to assist users in performing operations on the low-code platform and generate new operation flow data; The update module is used to update the semantic primitives and / or operation association rules in the operation semantic library based on the newly added operation flow data and / or user feedback data.

[0012] The present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement a method for constructing a low-code operation semantic library.

[0013] The present invention provides a non-transitory computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the method for constructing the low-code operation semantic library.

[0014] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements a method for constructing the low-code operation semantic library.

[0015] The method and apparatus for constructing a low-code operation semantic library provided in this invention deeply integrates general operations with vertical domain knowledge and continuously optimizes using feedback data, thereby significantly improving the accuracy and business fit of functions such as intelligent recommendation and automatic process assembly. By constructing a closed loop of "feedback-analysis-optimization", the semantic library can automatically adapt to changes in new components, new businesses and user habits, providing efficient, accurate and sustainably evolving semantic support for low-code platforms, and improving the intelligence level and development efficiency of low-code platforms. Attached Figure Description

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

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the construction method of the low-code operation semantic library provided by the present invention.

[0019] Figure 2 This is an implementation architecture diagram of the method for building a low-code operation semantic library provided by the present invention.

[0020] Figure 3 This is a schematic diagram of the structure of the low-code operation semantic library construction device provided by the present invention.

[0021] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0023] It should be noted that the terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps, units, or modules is not necessarily limited to those explicitly listed, but may include other steps, units, or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0024] Figure 1 This is a flowchart illustrating the construction method of the low-code operation semantic library provided by the present invention, as shown below. Figure 1 As shown, the method includes steps 110, 120, 130 and 140.

[0025] Step 110: Obtain the user's historical operation flow data on the low-code platform, and convert the historical operation flow data into semantic primitives related to the user's operation components.

[0026] Specifically, the execution entity of the low-code operation semantic library construction method provided in this embodiment of the invention is a low-code operation semantic library construction device. This device can be implemented in software, such as a low-code operation semantic library construction program running on a computer; or it can be implemented in hardware, such as a computer or server executing the low-code operation semantic library construction method.

[0027] A low-code platform is a software development environment that allows developers to create, deploy, and manage applications through a visual graphical user interface, drag-and-drop components, and model-driven logic, thereby minimizing the need for hand-written code.

[0028] Components are reusable, functionally independent modules used in low-code platforms to build applications. They encapsulate specific functional logic and user interface elements, allowing developers to integrate them into applications through drag-and-drop, configuration parameters, and other methods, without having to write large amounts of code manually.

[0029] Semantic primitives are the basic units in low-code platforms for structurally representing user actions. They transform unstructured action data into structured data with clear semantic meaning by extracting the type of user action, the components involved, the temporal relationship of the action, and related semantic attributes.

[0030] In this embodiment of the invention, the user's historical operation flow data on the low-code platform is obtained, and the historical operation flow data is transformed into semantic primitives related to the user's operation components.

[0031] Specifically, it captures the user's raw operation flow on a low-code platform, extracts raw data containing operation types, component information, and temporal relationships, and transforms it into structured, reusable semantic primitives, solving problems such as unstructured operation data and semantic loss in existing technologies.

[0032] Step 120: Perform rule mining on multiple semantic primitives to obtain operation association rules.

[0033] Specifically, operation association rules refer to the patterns of association between operations with business value and statistical significance that are discovered by analyzing the temporal relationships, semantic connections, and business logic dependencies of user operations within the operation sequence of a low-code platform. These rules describe the sequence, dependencies, and semantic connections between operations in specific business scenarios, providing a basis for intelligent recommendation, automated assembly, and process optimization on the low-code platform.

[0034] In this embodiment of the invention, operational association rules with business value are extracted from the semantic primitive sequence, and a semantic dependency graph is constructed.

[0035] Semantic dependency graphs are structured models used to depict the semantic relationships between operations in low-code platforms. They describe the temporal constraints, semantic associations, and domain logic dependencies between operations in the form of directed graphs. They mainly use semantic primitives as nodes and dependency relationships as edges to form a graph structure.

[0036] Step 130: Based on the semantic primitives and the operation association rules, respectively, integrate them with the knowledge of the preset vertical domain to generate an operation semantic library adapted to the preset vertical domain; the operation semantic library is used to assist users in performing operations on the low-code platform and generate new operation flow data.

[0037] Specifically, a predefined vertical domain refers to a specific business domain or industry that is predefined and focused on when building a low-code operational semantic library. These domains typically have unique business logic, operational patterns, and terminology systems, requiring specialized semantic adaptation and optimization to meet their specific needs. By constructing domain knowledge graphs, semantic mappings, and dedicated semantic extensions, predefined vertical domains can provide domain-specific semantic support for low-code platforms, improving intelligence and development efficiency.

[0038] An operation semantic library is a structured knowledge base used to store and manage the semantic information of user operations in low-code platforms. It transforms user actions into structured semantic primitives and mines the association rules between these primitives, providing intelligent auxiliary capabilities to low-code platforms, such as intelligent recommendation, automated assembly, and process optimization. The core of the operation semantic library lies in transforming unstructured operation data into structured knowledge with clear semantics and business logic, thereby improving the development efficiency and intelligence level of low-code platforms.

[0039] In this embodiment of the invention, the operation semantic library is used to assist users in performing operations on the low-code platform and generate new operation stream data.

[0040] In this embodiment of the invention, semantic primitives and association rules are integrated with vertical domain knowledge to construct a domain-adaptive operational semantic library.

[0041] Step 140: Based on the newly added operation flow data and / or user feedback data, update the semantic primitives and / or operation association rules in the operation semantic library.

[0042] Specifically, in this embodiment of the invention, the primitives and association rules of the operation semantic library can be dynamically updated based on new operation flows and user feedback.

[0043] The method for constructing a low-code operation semantic library provided in this invention deeply integrates general operations with vertical domain knowledge and continuously optimizes using feedback data, thereby significantly improving the accuracy and business fit of functions such as intelligent recommendation and automatic process assembly. By constructing a closed loop of "feedback-analysis-optimization," the semantic library can automatically adapt to changes in new components, new businesses, and user habits, providing efficient, accurate, and sustainably evolving semantic support for low-code platforms, and improving the intelligence level and development efficiency of low-code platforms.

[0044] In some embodiments, acquiring historical operation flow data of the user on a low-code platform and converting the historical operation flow data into semantic primitives related to user operation components includes: Based on the historical operation flow data, determine the components related to user operations, as well as the component attributes and timing markers of each component; Parse the component properties of each component to generate the component semantics for each component; Analyze the timing markers of each component to generate the timing relationships between the components; Based on the component semantics of each component and the temporal relationships between each component, the semantic primitives related to the user operation components are generated.

[0045] Specifically, component attributes refer to the inherent characteristics, parameters, or data fields of a component in a low-code platform. These characteristics, parameters, or data fields are used to describe the component's functionality, behavior, state, or relationship with other components. Component attributes are an important part of component semantics, enabling standardized representation, semantic resolution, and dynamic adaptation of components.

[0046] Figure 2 This is an implementation architecture diagram of the method for building a low-code manipulation semantic library provided by this invention, as shown below. Figure 2 As shown, in this embodiment of the invention, the historical operation flow data of the user on the low-code platform is obtained, and the historical operation flow data is transformed into semantic primitives related to the user's operation components. This includes four parts: operation flow preprocessing, component semantic parsing, temporal relationship modeling, and semantic primitive generation.

[0047] First, based on historical operation flow data, the components related to user operations, as well as the component attributes and time-series markers of each component, are identified. Specifically, operation flow preprocessing refers to cleaning, standardizing, and extracting features from the raw operation logs. This involves converting the collected raw operation data, including operation type, component identification (ID), timestamp, coordinates, context parameters, etc., into standard operation feature vectors through methods such as outlier filtering, feature standardization, and incomplete context handling.

[0048] Among them, the operation types mainly refer to basic operation types such as creation, configuration, connection, deletion, movement, copying, pasting, pinning, pinning, alignment, and resizing.

[0049] Outlier filtering refers to removing erroneous operations and duplicate data, such as drag-and-drop actions with an operation duration of less than 100 milliseconds, and consecutive identical configurations of the same component.

[0050] Feature standardization refers to mapping component IDs to a unified type code, such as alarm_form identifying alarm form components, and normalizing coordinate data to the [0,1] interval.

[0051] Context completion refers to completing the context information of an operation by associating preceding and following operations through a sliding time window. For example, the preceding operation for configuring the "monitoring list" is "setting a filter".

[0052] Operation feature vectors include operation type encoding, component attribute vectors, timing markers, context features, etc.

[0053] Secondly, the component properties of each component are parsed to generate the component semantics for each component. Specifically, component semantic parsing refers to parsing the semantic attributes of the components and establishing a mapping between components and business entities.

[0054] Specifically, the process begins by creating a domain-local library to provide a standardized reference framework for component semantics. This is initially built using domain expert annotations and industry standard document parsing, and incorporates feedback from a dynamic update module to achieve automated incremental updates. The domain-local library is a localized knowledge base built for a specific domain, storing standardized semantic descriptions, entity definitions, attribute information, and relationships between components within that domain. This library provides a standardized semantic reference framework for components in low-code platforms, supporting alignment, semantic parsing, and dynamic adaptation between components and domain entities.

[0055] Its basic data model is as follows: { "domain": "telecom", "entities": [ { "entity_id": "TEL001", "entity_name": "Performance threshold rule", "aliases": ["Performance alert generation rules"], / / Aliases are used for fuzzy matching "parent_entity": "Business rules", / / Entity hierarchy "attributes": [ / / Entity attribute definitions { "attr_name": "threshold", "data_type": "numeric", "business_desc": "The threshold for triggering the rule", "possible_units": ["%", "Erl", "GB"] }, { "attr_name": "Applicable scenarios", "data_type": "enumeration", "enum_values": ["Performance alert generation", "Performance monitoring"] } ] } ] } Secondly, component-entity alignment is performed, mapping components to standard entities in the domain ontology library. This is mainly achieved through three aspects: multi-feature extraction, semantic similarity calculation, and alignment decision.

[0056] Among them, multi-feature extraction refers to extracting three key features from component metadata: name, type, and context.

[0057] Semantic similarity calculation involves concatenating the extracted component features into a vector, then using a sentence embedding model (such as Sentence-BERT) to calculate the matching degree between the component and the ontology entity, thus obtaining the similarity score. , For name matching degree, For type matching degree, For context matching degree, , , where is the coefficient. Sentence-BERT is a model based on the Bidirectional Encoder Representations from Transformers (BERT) architecture, specially fine-tuned to generate high-quality sentence embeddings.

[0058] Alignment decision refers to judging the similarity data calculated from semantic similarity. When the highest similarity is >0.8, it is directly aligned to the corresponding entity; when the highest similarity is between 0.6 and 0.8, the component's historical usage record is used to assist in the judgment; when the similarity is <0.6, it is marked as "unidentified component" and the manual annotation process is triggered.

[0059] Finally, an attribute extraction rule engine is established to parse business attributes from attributes, resolving the inconsistency between technical parameters and business terminology. For example, threshold=50000, threshold corresponds to the business term "threshold".

[0060] The rule engine consists of a two-layer structure: a basic rule layer and a domain rule layer. The basic rule layer mainly contains mapping rules for general parameters and attributes, such as "threshold" and "threshold". The domain rule layer mainly provides customized rules for vertical domains, such as "occur_time" corresponding to "alarm occurrence time" in the telecommunications field.

[0061] It should be noted that when extracting attributes, the domain rule layer is matched first. If there is no match, the basic rule layer is used. Technical parameter names are mapped to business attribute names, and technical values ​​are mapped to business values. For example, threshold=50000, threshold is mapped to the threshold value, and 50000 is the specific value of the threshold attribute.

[0062] Next, the temporal markers of each component are analyzed to generate temporal relationships between them. Specifically, temporal relationship modeling refers to analyzing the temporal dependencies of the operation flow and adding temporal markers to the primitives. Based on the time interval and operation type sequence, necessary dependencies and optional dependencies are identified, and relative position encoding is used to transform the temporal relationships into vector features that are integrated into the semantic primitives.

[0063] Among them, necessary dependency refers to the necessary and indispensable temporal relationship between the preceding and subsequent operations, such as the "configuration" operation must follow the "create" operation.

[0064] Optional dependencies have no strong constraints; for example, a "move" operation can occur before or after a "configure" operation. Temporal relationship characteristics include the preceding operation ID, the following operation ID, the time interval, and the dependency type.

[0065] Finally, based on the component semantics of each component and the temporal relationships between them, semantic primitives related to user-operated components are generated. Specifically, semantic primitive generation refers to generating structured semantic primitives by integrating the outputs of the above sub-modules. These structured semantic primitives are represented in JavaScript Object Notation (JSON) format and include meta-ID, operation type, component semantics, attribute key-value pairs, temporal markers, and confidence levels. The completeness and consistency of the primitives are verified by a rule engine; for example, the primitive for the "connect" operation must include both the source and target components, and the component for the "delete" operation cannot be a deleted component. An example of a structured semantic primitive is shown below: { "primitive_id": "P6", "op_type": "configuration", "entity_type": "Performance threshold rule", "business_property": "Performance metric threshold", "attribute": {"name": "threshold", "value": 50000, "data_type": "numeric value"}, "temporal_marker": {"prev_op_id": "3", "time_interval": 10, "dependency_type": "inevitable"}, "confidence": 0.91 } The method for constructing a low-code operation semantic library provided in this invention analyzes and extracts structured, reusable semantic primitives from the original operation flow, transforming unstructured, semantically ambiguous original operation logs into standardized, computable semantic knowledge units. This not only solves the fundamental problems of data semantic loss and difficulty in reuse, but also provides a high-quality, unified data foundation for subsequent deeper association rule mining and intelligent analysis.

[0066] In some embodiments, the step of performing rule mining on multiple semantic primitives to obtain operation association rules includes: Based on the contextual features of the multiple semantic primitives, the operational intent weight corresponding to each semantic primitive is determined; Frequent item mining is performed on the multiple semantic primitives, and the operation association rules are determined based on the mined frequent itemsets; The support of the frequent itemsets is determined based on the operational intent weights corresponding to each semantic primitive.

[0067] Specifically, such as Figure 2 As shown, in this embodiment of the invention, rule mining is performed on multiple semantic primitives to obtain operation association rules, which include three parts: operation intent recognition, association rule extraction, and rule evaluation and optimization.

[0068] First, based on the contextual features of multiple semantic primitives, the operational intent weights corresponding to each semantic primitive are determined. Specifically, operational intent recognition refers to identifying the user's operational intent in the semantic primitive sequence and distinguishing between valid operations and erroneous operations. A standardized primitive sequence and user operation context are received, including operation duration and component interaction frequency. Then, a classifier model is used to classify the operational intents. The intent weights are analyzed based on features such as primitive operation type, component type, and time interval. The classifier model can be constructed based on algorithms such as extreme gradient boosting (XGBoost). Simultaneously, contextual analysis is used to assist in judgment, and the intent weights are adjusted based on operation duration and component interaction frequency. For example, the intent weight of the Delete primitive is reduced by 40% when the operation duration is <300ms, while the intent weight of operations on core components is increased by 30%.

[0069] Secondly, frequent item mining is performed on the multiple semantic primitives, and the operation association rules are determined based on the mined frequent itemsets. The support of the frequent itemsets is determined based on the operation intent weights corresponding to each semantic primitive. Specifically, the association rule extraction mainly employs a sequence pattern mining algorithm to extract high-frequency, high-confidence association rules from a weighted primitive sequence. These association rules include attributes such as preorder primitives, postorder primitives, support, confidence, and lift. The sequence pattern mining algorithm, based on the Sequential Pattern Discovery using Equivalence classes (SPADE) algorithm, introduces intent weights to improve the accuracy of the association rules.

[0070] Specifically, first, a sequence library is built, which converts standardized primitive sequences into transactions. Each transaction is a primitive sequence in the project development process, such as: T1: Create(Form)->Configure(Form)->Link(Form,RiskRule)->Configure (RiskRule) T2: Create(Form)->Link(Form,Storage)->Configure(Storage) It should be noted that "->" indicates the order of operations of primitives in the primitive sequence.

[0071] Then, through intent analysis and context analysis, the intent weight of each primitive is determined. For example, in primitive sequence T1, P1 (Create) has a normal operation time and is a core form component, with a weight of 0.9; P2 (Configure) configures parameters as key fields, with a weight of 0.85; P3 (Link) connects to core business components, with a weight of 0.95; and P4 (Configure) configures risk rule information, with a weight of 0.8. That is, the weights of each primitive in T1 are [0.9, 0.85, 0.95, 0.8]. Similarly, in primitive sequence T2, the weights of each primitive are [0.92, 0.7, 0.88]. Then, an improved sequence pattern mining algorithm is used to mine and analyze frequent items, and primitive weights are added to the support calculation. For example, the support of the itemset {Create(Form), Configure(Form)} is (0.9×0.85 +0.92×0) / 2=0.38.

[0072] Finally, association rules are generated from the frequent itemset, and the confidence and lift are calculated. The support of the itemset {Create(Form),Configure(Form)} is (0.9 × 0.85 + 0.92 × 0) / 2 = 0.38, and the lift is 0.835 / 0.425 = 1.96. Rules with support > 0.3, confidence > 0.7, and lift > 1.2 are retained, and the final association rule set is generated in the following format: [ { "pre_primitive": "Create(Form)", "post_primitive": "Configure(Form)", "support": 0.38, / / Support level "confidence": 0.835, / / Confidence level "lift": 1.96, / / Lift "domain_tag": "Basic configuration of filter forms", "weight": 1.2 / / Weight } ] Furthermore, based on association rules and dependencies, semantic primitives are mapped to graph nodes, and temporal dependencies, semantic dependencies, data dependencies, and other relationships are mapped to edges, thereby generating a semantic dependency graph and enabling process optimization, conflict detection, and dynamic semantic transfer capabilities.

[0073] Rule evaluation and optimization mainly refers to the verification and weighting of rules based on the associated rule set and the domain knowledge graph.

[0074] For example, in the telecommunications field, associating "AlarmRule" with the "Approval" rule increases its weight by 20%. Simultaneously, rule weights are dynamically adjusted based on the rule's application effectiveness; for instance, the rule weight increases by 0.1 for every 10% increase in recommendation adoption rate.

[0075] The method for constructing a low-code operation semantic library provided in this invention extracts operation association rules with business value from semantic primitive sequences, and connects isolated and atomic operation primitives into process patterns with business logic. This makes the implicit business processes and operation habits of users explicit and structured, providing a core basis for subsequent accurate operation recommendations, automated process assembly and construction of semantic dependency graphs, and greatly improving the platform's understanding of users' true business intentions.

[0076] In some embodiments, the step of fusing the semantic primitives and the operation association rules with knowledge from a preset vertical domain to generate an operation semantic library adapted to the preset vertical domain includes: Determine the knowledge graph of the preset vertical domain; the knowledge graph includes business entities and the business logic between the business entities. Establish a mapping relationship between the semantic primitives and the business entities; Establish a mapping relationship between the operation association rules and the business logic.

[0077] Specifically, semantic primitives and association rules are integrated with vertical domain knowledge to construct a domain-adaptive operational semantic library. For example... Figure 2 As shown, in this embodiment of the invention, semantic primitives and operation association rules are fused with knowledge of a preset vertical domain to generate an operation semantic library adapted to the preset vertical domain, which includes three parts: domain knowledge graph, general-domain semantic mapping, and domain-specific semantic extension.

[0078] In this embodiment of the invention, firstly, a knowledge graph for a preset vertical domain is determined; wherein, the knowledge graph includes business entities and the business logic between the business entities. Specifically, the domain knowledge graph mainly involves constructing or introducing knowledge graphs for vertical domains to provide domain knowledge support for semantic adaptation, such as the operation and maintenance knowledge graph of the telecommunications industry.

[0079] Then, establish mapping relationships between semantic primitives and business entities, as well as between operation association rules and business logic. Specifically, general-domain semantic mapping refers to establishing mapping relationships between semantic primitives and domain entities and operations to achieve domain-specific semantic transformation. Mapping is mainly established from three dimensions: operation type, component attributes, and business logic. For example, the general "connection" operation is mapped to "data verification" in the financial domain, while it can be mapped to "signal transmission" in the industrial domain.

[0080] Domain-specific semantic extensions primarily supplement the domain-specific semantic library by adding domain-specific semantic primitives and association rules. At the same time, the semantic library content is dynamically adjusted based on feedback from domain applications to optimize mapping relationships and rules.

[0081] First, identify domain-specific operations not covered by the general semantic library from the operation logs, such as the "alarm-derived rule" operation in the telecommunications field.

[0082] Furthermore, domain-specific semantic primitives can be generated by combining domain terminology, such as equipment precision parameter configuration in the industrial field.

[0083] Then, a multi-indicator evaluation system is constructed based on semantic matching accuracy, rule usage rate, and development efficiency improvement value to evaluate application efficiency. Reinforcement learning and other algorithms are used to dynamically adjust mapping weights by using user feedback as reward signals. For example, if the recommendation adoption rate of a certain mapping rule increases by 10%, the weight increases by 0.1, thereby supporting continuous optimization of the domain semantic library.

[0084] The method for constructing a low-code operation semantic library provided in this invention integrates semantic primitives and association rules with vertical domain knowledge to build a domain-adaptive operation semantic library. This endows semantic primitives and association rules with precise industry connotations, enabling the low-code platform to provide truly business-specific and highly relevant intelligent recommendation and automation services, greatly enhancing its depth and value in vertical industry applications.

[0085] In some embodiments, the method further includes: Based on the bidirectional attention weights between the semantic primitives and the business entities in the preset vertical domain, the mapping probability between the semantic primitives and the business entities is determined. The mapping probability is used to characterize the stability of the mapping relationship.

[0086] Specifically, in this embodiment of the invention, when establishing the mapping relationship between semantic primitives and domain entities and operations from three dimensions—operation type, component attributes, and business logic—the mapping probability between semantic primitives and business entities can be determined based on the bidirectional attention weights between semantic primitives and business entities in a preset vertical domain; whereby the mapping probability is used to characterize the stability of the mapping relationship. Specifically, data can be labeled by domain experts, and the mapping weights can be learned through a logistic regression model. Higher mapping weights indicate a more stable mapping relationship. Furthermore, bidirectional attention can be used to capture the deep association between semantic primitives and domain knowledge, achieving accurate mapping.

[0087] Specifically, firstly, we focus on the domain-related parts of the semantic primitives to calculate the attention of the semantic primitives to the domain entities, and secondly, we focus on the parts of the domain entities that match the semantic primitives to calculate the attention of the domain entities to the semantic primitives.

[0088] Then, the mapping probability between semantic primitives and domain entities is calculated based on the attention weights, that is: mapping weight = sigmoid(semantic primitive vector × attention matrix × domain entity vector).

[0089] Finally, the attention matrix is ​​iteratively optimized by training the model using mapping samples annotated by domain experts, thereby optimizing the mapping relationship.

[0090] The method for constructing a low-code semantic library provided in this invention captures the deep association between semantic primitives and domain knowledge through bidirectional attention, achieving accurate mapping and surpassing the role of traditional shallow matching based on rules or keywords. By allowing the model to examine each other from both "semantic primitives" and "domain knowledge" simultaneously and dynamically focus on the most relevant parts, it can discover the deep intrinsic association between the two in complex or ambiguous scenarios, ultimately achieving highly accurate and robust semantic alignment, and significantly improving the quality and intelligence level of domain adaptation.

[0091] In some embodiments, updating the semantic primitives and / or operation association rules in the operation semantic library based on the newly added operation flow data and / or user feedback data includes: Based on the newly added operation flow data and / or user feedback data, determine the accuracy of each semantic primitive in the operation semantic library and the applicability of each operation association rule; Based on the accuracy of each semantic primitive and the applicability of each operation association rule, determine the semantic primitives and / or operation association rules to be optimized. Update the semantic primitives and / or operation association rules to be optimized.

[0092] Specifically, in this embodiment of the invention, the primitives and association rules of the operation semantic library are dynamically updated based on new operation flows and user feedback. For example... Figure 2 As shown, the dynamic update of the semantic library mainly includes three parts: multi-source feedback data analysis, incremental update execution, and update effect evaluation.

[0093] Multi-source feedback data analysis refers to collecting multi-dimensional data from low-code platform applications, user feedback, and domain knowledge updates, and identifying parts of the semantic library that need optimization, such as metawords with low accuracy and rules with low applicability.

[0094] Specifically, firstly, the collected multi-source feedback data is cleaned, and anomaly detection algorithms are used to filter out noisy data, such as mistakenly clicked recommendation feedback; Then, multi-source information is integrated through timestamp alignment and weight allocation. For example, the weight of explicit user feedback is greater than that of application logs, forming a standardized feedback dataset that includes feedback type, associated semantic ID, weight value, etc.

[0095] Then, based on the standardized feedback dataset and the current semantic library, low-quality semantic units are located through association analysis, and on this basis, the decision tree algorithm is used to explore the root causes of the problem, such as whether the low applicability of a certain rule is due to outdated mapping weights or changes in domain knowledge.

[0096] Finally, based on the "scope of impact × problem severity" scoring model, a list of items to be optimized is formed, which includes problem type, scope of impact, and priority score.

[0097] Incremental update execution refers to performing incremental updates on the items to be optimized, avoiding full reconstruction, thereby improving update efficiency.

[0098] Specifically, low-quality primitives are fine-tuned through primitive incremental optimization, such as revising the attribute definition of the "fault alarm" primitive in the telecommunications field; and rule weights are dynamically adjusted based on feedback data, such as using an online gradient descent algorithm to update rule confidence.

[0099] Furthermore, during updates, the semantic library is divided into a "core layer" (highly used primitives / rules) and an "extension layer," with the core layer being updated first to improve overall response speed.

[0100] Evaluation of update effectiveness refers to verifying the effect of the updated semantic database and ensuring the effectiveness of optimization measures. This mainly involves constructing a multi-dimensional indicator system. Core indicators may include semantic matching accuracy, rule recommendation adoption rate, and development efficiency improvement; extended indicators may include update time and system resource utilization.

[0101] The effects of the update are evaluated based on a multi-dimensional indicator system, and an update effect evaluation report is generated. When the indicators drop by more than 10% after the update, the rollback to the previous version is automatically triggered.

[0102] Furthermore, during this process, the semantic database can be dynamically optimized based on reinforcement learning algorithms.

[0103] Specifically, the current state of the semantic library is defined as a vector. ,in For the first The weight values ​​of each semantic unit (primitive / rule / mapping) are determined. Three types of update actions are defined: A1 adjusts the semantic unit weight, A2 corrects the semantic unit content, and A3 adds / deletes a semantic unit. Then, a reward value is set. , , , These are the weighting coefficients. Finally, the action selection is iteratively optimized through feedback data to maximize long-term cumulative rewards and continuously improve the recommendation performance of the semantic database.

[0104] The method for constructing a low-code operation semantic library provided in this invention dynamically updates the primitives and association rules of the operation semantic library based on new operation flows and user feedback. This allows for real-time optimization of semantic primitives, association rules, and general-domain mapping relationships, ensuring that the semantic library continuously meets actual development needs. It guarantees that the knowledge of the semantic library remains fresh and timely, automatically adapting to changes in platform features, business processes, and user habits, thus solving the problem that static semantic libraries cannot cope with business changes and the evolution of user operation habits.

[0105] The apparatus provided in the embodiments of the present invention will be described below. The apparatus described below can be referred to in correspondence with the method described above.

[0106] Figure 3 This is a schematic diagram of the structure of the low-code operation semantic library construction device provided by the present invention, as shown below. Figure 3 As shown, the device includes an acquisition module 310, a mining module 320, an adaptation module 330, and an update module 340 connected in sequence.

[0107] The acquisition module 310 is used to acquire the user's historical operation flow data on the low-code platform and convert the historical operation flow data into semantic primitives related to the user's operation components. The mining module 320 is used to perform rule mining on multiple semantic primitives to obtain operation association rules; The adaptation module 330 is used to fuse the semantic primitives and the operation association rules with knowledge of a preset vertical domain to generate an operation semantic library adapted to the preset vertical domain; the operation semantic library is used to assist users in performing operations on the low-code platform and generate new operation flow data. The update module 340 is used to update the semantic primitives and / or operation association rules in the operation semantic library based on the newly added operation flow data and / or user feedback data.

[0108] The low-code operation semantic library construction device provided in this embodiment of the invention deeply integrates general operations with vertical domain knowledge and continuously optimizes using feedback data, thereby significantly improving the accuracy and business fit of functions such as intelligent recommendation and automatic process assembly. By constructing a closed loop of "feedback-analysis-optimization", the semantic library can automatically adapt to changes in new components, new businesses and user habits, providing efficient, accurate and sustainably evolving semantic support for low-code platforms, and improving the intelligence level and development efficiency of low-code platforms.

[0109] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 4 As shown, the electronic device may include: a processor 410, a communications interface 420, a memory 430, and a communications bus 440, wherein the processor 410, the communications interface 420, and the memory 430 communicate with each other via the communications bus 440. The processor 410 can call logical commands stored in the memory 430 to execute the methods described in the above embodiments, for example: The system acquires historical operation flow data of users on a low-code platform and transforms this data into semantic primitives related to user operation components. It then performs rule mining on multiple semantic primitives to obtain operation association rules. Based on these semantic primitives and operation association rules, it integrates them with knowledge from a preset vertical domain to generate an operation semantic library adapted to that domain. This operation semantic library assists users in performing operations on the low-code platform, generating new operation flow data. Based on the new operation flow data and / or user feedback data, it updates the semantic primitives and / or operation association rules in the operation semantic library.

[0110] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a 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 a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0111] The processor in the electronic device provided in this embodiment of the invention can call logical instructions in the memory to implement the above method. Its specific implementation method is the same as the aforementioned method implementation method and can achieve the same beneficial effects, which will not be repeated here.

[0112] This invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the methods provided in the above embodiments.

[0113] The specific implementation method is the same as the aforementioned method implementation method and can achieve the same beneficial effects, so it will not be repeated here.

[0114] This invention provides a computer program product, including a computer program that, when executed by a processor, implements the method described above.

[0115] The system embodiments described above are merely illustrative. 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 modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0116] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a low-code operation semantic library, characterized in that, The method comprises the following steps: acquiring historical operation flow data of a user on a low-code platform, and converting the historical operation flow data into semantic primitives related to user operation components; performing rule mining on a plurality of semantic primitives to obtain operation association rules; fusing the semantic primitives and the operation association rules with knowledge of a preset vertical field respectively to generate an operation semantic library adapted to the preset vertical field; the operation semantic library is used to assist the user in performing operations on the low-code platform and generate new operation flow data; based on the new operation flow data and / or user feedback data, updating the semantic primitives and / or operation association rules in the operation semantic library.

2. The method of claim 1, wherein, The method comprises the following steps: based on the historical operation flow data, determining components related to user operations and component attributes and timing labels of each component; parsing the component attributes of each component to generate component semantics of each component; analyzing the timing labels of each component to generate timing relationships between each component; based on the component semantics of each component and the timing relationships between each component, generating the semantic primitives related to the user operation components.

3. The method of claim 1, wherein, The method comprises the following steps: based on the context features of the plurality of semantic primitives, determining operation intent weights corresponding to each semantic primitive; performing frequent item mining on the plurality of semantic primitives, and determining the operation association rules based on the mined frequent item sets; wherein the support degree of the frequent item set is determined based on the operation intent weights corresponding to each semantic primitive.

4. The method of claim 3, wherein, The method comprises the following steps: determining a knowledge graph of the preset vertical field; the knowledge graph comprises business entities and business logic between the business entities; establishing a mapping relationship between the semantic primitives and the business entities; establishing a mapping relationship between the operation association rules and the business logic.

5. The method of claim 4, wherein, The method further comprises the following steps: based on the bidirectional attention weights between the semantic primitives and the business entities in the preset vertical field, determining mapping probabilities between the semantic primitives and the business entities; the mapping probability is used to represent the stability degree of the mapping relationship.

6. The method of claim 1, wherein, The method comprises the following steps: based on the new operation flow data and / or user feedback data, determining the accuracy of each semantic primitive and the applicability of each operation association rule in the operation semantic library; based on the accuracy of each semantic primitive and the applicability of each operation association rule, determining semantic primitives and / or operation association rules to be optimized; updating the semantic primitives and / or operation association rules to be optimized.

7. A construction device of a low-code operation semantic library, characterized in that, The method comprises the following steps: An acquisition module is configured to acquire historical operation flow data of a user on a low-code platform, and convert the historical operation flow data into semantic primitives related to user operation components; A mining module is configured to perform rule mining on a plurality of semantic primitives to obtain operation association rules; An adaptation module is configured to fuse the semantic primitives and the operation association rules with knowledge of a preset vertical field respectively, and generate an operation semantic library adapted to the preset vertical field; The operation semantic library is used to assist the user in performing operations in the low-code platform, and generate new operation flow data; An update module is configured to update the semantic primitives and / or the operation association rules in the operation semantic library based on the new operation flow data and / or user feedback data.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, The processor executes the computer program to implement the construction method of the low-code operation semantic library according to any one of claims 1 to 6. 9.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the construction method of the low-code operation semantic library according to any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the construction method of the low-code operation semantic library according to any one of claims 1 to 6.

Images