An application scenario-based standard element semantic integration method

Abstract

This invention relates to the field of semantic integration, and more particularly to a standard element semantic integration method based on application scenarios. Applied to an enterprise management platform with a standard element semantic integration strategy, the method includes: assigning corresponding task attribute values ​​to the standard element semantic integration model according to task attributes; adjusting the training weights of the standard element semantic integration model by formulating weight allocation rules; collecting several semantically ambiguous words and preprocessing them to form corresponding fuzzy pre-data; transmitting this data to the standard element semantic integration model for learning; comparing the fuzzy integration accuracy with an accuracy threshold to determine whether to retain the corresponding standard element semantic integration model. By improving the efficiency and accuracy of semantic processing, the method enhances the integration accuracy of different semantic elements, optimizes various business processes in the enterprise management platform, reduces business delays and errors caused by semantic misunderstandings or inaccuracies, and improves the overall operational efficiency of the enterprise management platform.

Description

Technical Field

[0001] This invention relates to the field of semantic integration, and more particularly to a method for semantic integration of standard elements based on application scenarios. Background Technology

[0002] In modern enterprise management, enterprise management platforms typically need to process large amounts of standardized document data, including but not limited to national standards, industry standards, local standards, group standards, as well as internal rules and regulations, enterprise standards, and work manuals. This data contains rich semantic information, which is crucial for enterprise decision support, business process optimization, and customer service. However, existing enterprise management platforms suffer from several problems in semantic processing. First, standard data often contains many semantically ambiguous terms with no clear semantic boundaries, making it difficult for traditional semantic processing methods to accurately understand their meanings. This significantly reduces the accuracy and completeness of semantic understanding, posing a significant challenge to semantic analysis.

[0003] Furthermore, the semantic processing models in existing enterprise management platforms typically lack flexibility and adaptability. They often employ fixed model structures and parameter settings, making it difficult to dynamically adjust them according to different application scenarios and task requirements. This makes it difficult for existing models to meet practical requirements when facing complex and ever-changing enterprise management needs.

[0004] Chinese Patent Publication No. CN104361017B discloses a traffic information processing method based on unified semantic understanding. This method achieves unified semantic understanding by recording traffic information vocabulary in a central information register. The specific steps include: 1) establishing a central data register and recording traffic information vocabulary in it; 2) establishing an ontology database and converting the traffic information vocabulary into a human-computer readable form; 3) establishing a traffic information cloud service platform and publishing traffic information services based on the traffic information vocabulary in the central data register and ontology database; 4) establishing a service semantic retrieval module and providing information retrieval services to service requesters. Compared with existing technologies, this invention has advantages such as ensuring the uniformity of data within the traffic status information platform.

[0005] Chinese patent application publication number CN114492463A discloses a unified semantic Chinese text polishing method based on adversarial multi-task learning, which includes the following steps: S1, determining the polishing range through a polishing range segmentation model; S2, traversing and searching for the number of characters to be inserted within the polishing range, generating a series of new sentences using a masked language model, and scoring the resulting series of new sentences using a position scoring model; S3, selecting the best sentence based on the scoring results. This invention has the advantages of promoting the research progress of unified text proofreading methods, being suitable for practical application, and indeed improving people's text editing experience to a certain extent.

[0006] However, the above methods have the following problems: they often use fixed model structures and parameter settings, making it difficult to dynamically adjust them according to different application scenarios and task requirements. Summary of the Invention

[0007] To address this, the present invention provides a standard element semantic integration method based on application scenarios, which overcomes the problem that existing technologies often employ fixed model structures and parameter settings, making it difficult to dynamically adjust them according to different application scenarios and task requirements.

[0008] To achieve the above objectives, this invention provides a standard element semantic integration method based on application scenarios, which is applied to an enterprise management platform. The enterprise management platform is characterized by having a standard element semantic integration strategy, including:

[0009] Determine the task attributes of several standard feature semantic integration models, and assign corresponding task attribute values ​​to the standard feature semantic integration models according to the task attributes, wherein,

[0010] The task attributes include standard data source, model type, and task objective;

[0011] The task attribute values ​​include standard data source attribute values, model type attribute values, and task target attribute values;

[0012] Based on the task attribute values, weight allocation rules are formulated for the standard element semantic integration model;

[0013] The training weights of the standard element semantic integration model are adjusted according to the weight allocation rules, and the adjusted training weights are applied to the standard element semantic integration model.

[0014] Several semantically ambiguous words from the enterprise management platform are collected, and these semantically ambiguous words are preprocessed to form corresponding fuzzy pre-data.

[0015] The semantically ambiguous words refer to words in the enterprise management platform whose semantic scope has no clear boundaries;

[0016] The fuzzy pre-database formed by the fuzzy pre-data is used to train the standard element semantic integration model, and the fuzzy pre-data is transmitted to the standard element semantic integration model for learning to generate the corresponding integration result;

[0017] The fusion results are verified to obtain the fuzzy fusion accuracy rate corresponding to a single standard element semantic integration model. The fuzzy fusion accuracy rate is compared with an accuracy threshold. Based on the comparison result, it is determined whether to retain the corresponding standard element semantic integration model.

[0018] The accuracy threshold is the minimum standard for the fuzzy integration accuracy of the standard element semantic integration model in semantically integrating the semantically fuzzy words, and it is inversely correlated with the operating efficiency of the enterprise management platform.

[0019] Furthermore, the steps for determining the task attributes of the standard feature semantic integration model include:

[0020] The standard data source of the semantically ambiguous words is clearly defined, and the source range of the semantically ambiguous words is limited;

[0021] The type of the standard element semantic integration model is clearly defined, and the implementation method of the standard element semantic integration model is limited;

[0022] The task objectives of the standard element semantic integration model are clearly defined, and the application scenarios of the standard element semantic integration model are limited.

[0023] Furthermore, the steps for assigning task attribute values ​​include:

[0024] Based on the source range of the semantically ambiguous words, assign standard data source attribute values ​​to the standard element semantic integration model;

[0025] Based on the implementation method of the standard element semantic integration model, assign model type attribute values ​​to the standard element semantic integration model;

[0026] Based on the application scenario of the standard element semantic integration model, task target attribute values ​​are assigned to the standard element semantic integration model.

[0027] Furthermore, based on the task attribute values, the weight allocation rule of the standard element semantic integration model is calculated using linear rules, wherein,

[0028] The weight allocation rule is proportional to the task attribute value.

[0029] Furthermore, the steps of applying the training weights to the standard feature semantic ensemble model include:

[0030] The standard element semantic integration model is initialized by setting the training weights of the standard element semantic integration model to the initial training weights.

[0031] According to the weight allocation rule, the comprehensive training weight of the standard element semantic integration model is calculated, and the comprehensive training weight is used as the new training weight;

[0032] The training weights are applied to the training process of the standard element semantic integration model.

[0033] Furthermore, the steps for generating fuzzy predata include:

[0034] Several index features of the semantically fuzzy words are selected, among which,

[0035] The indicator features include the degree of ambiguity, semantic similarity, and distribution characteristics of the semantically ambiguous words;

[0036] The semantically ambiguous words are segmented according to a standard sampling rate to form corresponding fuzzy pre-data, wherein...

[0037] The standard sampling rate is the learning rate that the standard element semantic integration model can recognize, and for a single learning iteration, the corresponding standard sampling rate is a single learning rate.

[0038] Furthermore, the steps for training the standard feature semantic ensemble model using a fuzzy pre-database include:

[0039] The weights of the standard element semantic integration model are adjusted to the training weights, and the standard element semantic integration model is iteratively optimized until the loss value of the standard element semantic integration model converges to the preset loss value.

[0040] The current standard element semantic integration model was tested using the aforementioned fuzzy pre-database;

[0041] When the test results meet the preset test results, the learning parameters of the standard element semantic integration model are saved, and the learning parameters are used as standard parameters.

[0042] Furthermore, the learning parameters of the standard element semantic integration model are adjusted to the standard parameters, the fuzzy pre-data and the indicator features are input into the standard element semantic integration model, the standard element semantic integration model learns the fuzzy pre-data, generates the corresponding integration result, the integration result is labeled and verified according to the indicator features, and the corresponding fuzzy integration accuracy is obtained.

[0043] Furthermore, the accuracy of the fuzzy integration is compared with an accuracy threshold. When the accuracy of the fuzzy integration is greater than the accuracy threshold, the corresponding standard element semantic integration model is not adjusted.

[0044] Furthermore, when the fuzzy fusion accuracy is less than the accuracy threshold, the fuzzy pre-data is divided into several sub-fuzzy pre-data, and the sub-fuzzy pre-data is learned using the standard element semantic integration model to obtain the corresponding secondary fusion accuracy. When the secondary fusion accuracy is greater than the accuracy threshold, the corresponding standard element semantic integration model is not adjusted. When the secondary fusion accuracy is less than the accuracy threshold, the enterprise management platform determines that the corresponding standard element semantic integration model has a performance problem and filters the corresponding standard element semantic integration model.

[0045] Compared with existing technologies, this invention assigns corresponding task attribute values ​​to the standard element semantic integration model based on task attributes, adjusts the training weights of the standard element semantic integration model according to the weight allocation rules formulated based on the task attribute values, collects and preprocesses several semantically ambiguous words to form corresponding fuzzy pre-data, transmits it to the standard element semantic integration model for learning, compares the fuzzy integration accuracy with the accuracy threshold, and determines whether to retain the corresponding standard element semantic integration model. By improving the efficiency and accuracy of semantic processing, the integration accuracy of different semantic elements is improved, various business processes in the enterprise management platform are optimized, business delays and errors caused by semantic misunderstanding or inaccuracy are reduced, and the overall operational efficiency of the enterprise management platform is improved.

[0046] Furthermore, by clarifying the standard data sources for semantically ambiguous words, the types of standard element semantic integration models, and the task objectives, the relevance, efficiency, accuracy, and scalability of standard element semantic integration models are significantly improved. This not only helps optimize model performance but also promotes knowledge sharing and collaborative work within enterprises, providing strong support for the efficient operation of enterprise management platforms.

[0047] Furthermore, by assigning standard data source attribute values, model type attribute values, and task target attribute values ​​to the standard element semantic integration model, it is easier to select the most suitable model type, which can give full play to the advantages of the model, further improve the accuracy of semantic processing, enhance the efficiency and performance of semantic processing, avoid processing irrelevant standard data, and reduce the waste of computing resources.

[0048] Furthermore, by calculating the weight allocation rule of the standard element semantic integration model based on the task attribute value using linear rules, and making the weight allocation rule proportional to the task attribute value, the training process of the model can be optimized, the accuracy of semantic processing can be improved, and the adaptability and flexibility of the system can be enhanced. This provides a scientific and quantitative solution for the weight allocation of the standard element semantic integration model, which helps to build an efficient and accurate semantic processing system.

[0049] Furthermore, by applying training weights to the training process of the standard feature semantic ensemble model, the training of the model can be optimized, and the model's adaptability, generalization ability and performance can be improved. Using comprehensive training weights, the model can focus more on processing standard data with higher values ​​than its task attributes, thereby improving the model's performance.

[0050] Furthermore, by selecting the index features of semantically ambiguous words, segmenting them according to the standard sampling rate, and forming fuzzy pre-data, complex semantically ambiguous words can be transformed into a format suitable for learning by the standard element semantic integration model. This not only improves the quality and processability of standard data, but also optimizes the learning efficiency and generalization ability of the standard element semantic integration model, providing strong support for building an efficient and accurate semantic processing system.

[0051] Furthermore, through weight adjustment and iterative optimization, model testing, and parameter saving and standardization, the standard element semantic integration model can be systematically trained and optimized. This not only ensures the model's performance on the fuzzy pre-database but also improves the model's reliability and consistency, providing strong support for building an efficient and accurate semantic processing system and helping to enhance the overall effect of semantic processing in enterprise management platforms.

[0052] Furthermore, by adjusting learning parameters, inputting fuzzy pre-data and indicator features, labeling and verifying the fusion results, and comparing the fuzzy fusion accuracy with the accuracy threshold, the performance of the standard element semantic integration model can be systematically optimized. This not only ensures the accuracy and reliability of the model on fuzzy pre-data, but also supports the continuous optimization and improvement of the model.

[0053] Furthermore, by segmenting fuzzy pre-data into sub-fuzzy pre-data and performing secondary learning and verification, the performance of the standard element semantic integration model can be further optimized. This not only enhances the model's ability to process complex standard data, but also ensures that only high-quality models are retained and used through a strict verification and screening mechanism. This helps improve the overall effect of semantic processing in the enterprise management platform, while ensuring the efficient operation and reliability of the system. Attached Figure Description

[0054] Figure 1 This is a flowchart of a standard element semantic integration method based on application scenarios according to the present invention;

[0055] Figure 2 A flowchart illustrating the task attributes of the standard element semantic integration model in an embodiment of the present invention;

[0056] Figure 3 A flowchart illustrating the allocation of task attribute values ​​in embodiments of the present invention;

[0057] Figure 4 This is a flowchart illustrating how training weights are applied to a standard element semantic integration model, as described in an embodiment of the present invention. Detailed Implementation

[0058] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0059] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0060] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0061] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0062] Please see Figure 1 The diagram shows a flowchart of a standard element semantic integration method based on an application scenario, applicable to an enterprise management platform. The method is characterized by the enterprise management platform having a standard element semantic integration strategy, including:

[0063] Step S1: Determine the task attributes of several standard feature semantic integration models, and assign corresponding task attribute values ​​to the standard feature semantic integration models according to the task attributes, wherein...

[0064] Task attributes include standard data source, standard feature semantic integration model type, and task objective;

[0065] Task attribute values ​​include standard data source attribute values, standard feature semantic integration model type attribute values, and task objective attribute values;

[0066] Step S2: Based on the task attribute values, formulate weight allocation rules for the standard feature semantic integration model;

[0067] Step S3: Adjust the training weights of the standard element semantic integration model according to the weight allocation rules, and apply the adjusted training weights to the standard element semantic integration model.

[0068] Step S4: Collect several semantically ambiguous words from the enterprise management platform, preprocess the semantically ambiguous words to form corresponding fuzzy pre-data, among which...

[0069] Semantically ambiguous words refer to words in enterprise management platforms whose semantic scope is not clearly defined;

[0070] Step S5: Use the fuzzy pre-database formed by the fuzzy pre-data to train the standard element semantic integration model, and transmit the fuzzy pre-data to the standard element semantic integration model for learning to generate the corresponding integration result;

[0071] Step S6: Verify the fusion results, obtain the fuzzy fusion accuracy rate corresponding to the semantic fusion model of a single standard element, compare the fuzzy fusion accuracy rate with the accuracy rate threshold, and determine whether to retain the corresponding standard element semantic fusion model based on the comparison result.

[0072] The accuracy threshold is the minimum standard for the fuzzy integration accuracy of the standard element semantic integration model in semantically integrating fuzzy words, and it is inversely correlated with the operating efficiency of the enterprise management platform.

[0073] By assigning corresponding task attribute values ​​to the standard element semantic integration model based on task attributes, and adjusting the training weights of the standard element semantic integration model according to the weight allocation rules formulated based on the task attribute values, a number of semantically ambiguous words are collected and preprocessed to form corresponding fuzzy pre-data, which is then transmitted to the standard element semantic integration model for learning. The fuzzy integration accuracy is compared with the accuracy threshold to determine whether to retain the corresponding standard element semantic integration model. By improving the efficiency and accuracy of semantic processing, the integration accuracy of different semantic elements is improved, various business processes in the enterprise management platform are optimized, business delays and errors caused by semantic misunderstanding or inaccuracy are reduced, and the overall operational efficiency of the enterprise management platform is improved.

[0074] Please see Figure 2As shown, it is a flowchart of the task attributes for determining the standard element semantic integration model of the present invention, including:

[0075] Step S11: Clarify the standard data source for semantically ambiguous words and limit the source range of semantically ambiguous words;

[0076] Step S12: Clarify the type of the standard element semantic integration model and limit the implementation method of the standard element semantic integration model;

[0077] Step S13: Clarify the task objectives of the standard element semantic integration model and define the application scenarios of the standard element semantic integration model.

[0078] In practice, semantically ambiguous terms come from different modules of the enterprise management platform, such as customer feedback, internal documents, market research, and financial statements. By clearly defining the sources of standard data, the scope of standard data processed by the standard element semantic integration model can be limited, avoiding the processing of irrelevant standard data.

[0079] The above operations are highly targeted: the standard element semantic integration model can focus on standard data from specific sources, improving processing efficiency.

[0080] Standard data quality improvement: By limiting the source of standard data, targeted cleaning and preprocessing of standard data can be performed to improve its quality.

[0081] Reduce interference: Avoid irrelevant standard data from interfering with the training and application of the standard element semantic integration model, thereby improving the accuracy and reliability of the standard element semantic integration model.

[0082] Standard feature semantic integration models can be implemented in various ways, such as rule-based models, statistical models, and machine learning models (e.g., deep learning models). Once the model type is determined, the most suitable model architecture can be selected based on the specific task.

[0083] The above operations optimize model performance: Different types of models are suitable for different tasks and standard data types. Once the model type is clearly defined, its advantages can be fully utilized to improve model performance.

[0084] Rational resource allocation: Allocate computing resources appropriately according to model type. For example, deep learning models may require more computing resources, while rule-based models are relatively lightweight.

[0085] Enhanced scalability: Once the model type is defined, the system can quickly expand or adjust the model architecture as needed to adapt to new task requirements.

[0086] The task objectives of standard element semantic integration models are usually closely related to specific application scenarios. For example, in customer relationship management (CRM), the task objective might be to extract key information from customer feedback; in supply chain management, the task objective might be to optimize the semantic integration of logistics information.

[0087] The above operations focus on core needs: After clarifying the task objectives, the model can focus on solving specific problems and avoid wasting resources.

[0088] Enhance user experience: The model can better meet the needs of users in specific scenarios, thereby enhancing the user experience.

[0089] Enhanced maintainability: With clear task objectives, system maintenance personnel can better understand the model's functions and applicable scope, facilitating maintenance and optimization.

[0090] By clarifying the standard data sources for semantically ambiguous words, the types of standard element semantic integration models, and the task objectives, the relevance, efficiency, accuracy, and scalability of standard element semantic integration models are significantly improved. This not only helps optimize the performance of standard element semantic integration models but also promotes knowledge sharing and collaborative work within enterprises, providing strong support for the efficient operation of enterprise management platforms.

[0091] Please see Figure 3 As shown, it is a flowchart of the task attribute value allocation process of the present invention, including:

[0092] Step St1: Assign standard data source attribute values ​​to the standard element semantic integration model based on the source range of semantically ambiguous words;

[0093] Step St2: Based on the implementation method of the standard feature semantic integration model, assign the standard feature semantic integration model type attribute value to the standard feature semantic integration model;

[0094] Step St3: Assign task target attribute values ​​to the standard element semantic integration model according to the application scenario of the standard element semantic integration model.

[0095] In practice, the following are the specific rules for assigning task attribute values ​​to the standard feature semantic integration model:

[0096] 1. Assign standard data source attribute values ​​based on the source range of semantically ambiguous words.

[0097] The source range of semantically ambiguous words determines the allocation of standard data source attribute values. Common source ranges include internal standard data sources, external standard data sources, and user-generated standard data. Specific rules are as follows:

[0098] Internal standard data source: If the semantically ambiguous words mainly come from the company's internal standard database or system (such as ERP system), then assign a higher standard data source attribute value, because internal standard data usually has higher reliability and relevance.

[0099] External standard data source: If the semantically ambiguous words come from external standard data sources (such as industry reports, news information, etc.), then assign a medium standard data source attribute value, because although external standard data is abundant, its reliability and relevance may be low.

[0100] User-generated standard data: If semantically ambiguous words come from user input (such as user feedback, comments, etc.), a lower standard data source attribute value is assigned because the quality of user-generated standard data varies and requires additional preprocessing.

[0101] 2. Assign model type attribute values ​​according to the implementation method of the standard feature semantic integration model.

[0102] The implementation method of the standard feature semantic integration model determines the allocation of model type attribute values. Common implementation methods include rule-based models, machine learning models, and deep learning models. Specific rules are as follows:

[0103] Rule-based models: If the model is implemented based on predefined rules, assign a lower model type attribute value, because such models are simple but less flexible.

[0104] Machine learning model: If the model is implemented using a machine learning algorithm, assign a moderate model type attribute value, because this type of model performs well when dealing with complex standard data, but requires sufficient training standard data.

[0105] Deep learning models: If the model is implemented using deep learning algorithms, assign a higher model type attribute value because these models perform well in handling highly complex semantic tasks, but require a lot of computing resources and standard data.

[0106] 3. Assign task target attribute values ​​according to the application scenario of the standard element semantic integration model.

[0107] The application scenario of the standard feature semantic integration model determines the allocation of task objective attribute values. Common application scenarios include semantic matching, semantic classification, and semantic generation. Specific rules are as follows:

[0108] Semantic matching: If the main task of the model is to determine whether two semantic units have the same semantics (such as determining the relevance of a query to a document in a search engine), then assign a moderate task objective attribute value.

[0109] Semantic classification: If the main task of the model is to classify semantic units into predefined categories (such as classifying customer feedback), then a higher task objective attribute value is assigned, because semantic classification usually requires a high level of accuracy from the model.

[0110] Semantic generation: If the main task of the model is to generate new semantic content based on the input (such as automatic summarization, question answering, etc.), then a lower task objective attribute value is assigned, because semantic generation tasks are more difficult and the requirements for the model are more complex.

[0111] Based on the application scenario, assign corresponding task objective attribute values ​​to the standard element semantic integration model. For example:

[0112] When the application scenario is "customer relationship management", the task objective attribute value can be "customer feedback analysis".

[0113] When the application scenario is "supply chain management", the task objective attribute value can be "logistics information optimization".

[0114] When the application scenario is "market analysis", the task objective attribute value can be "market trend prediction".

[0115] By applying the above rules, appropriate task attribute values ​​can be assigned to the standard element semantic integration model, thereby optimizing the model's performance and applicability.

[0116] By assigning standard data source attribute values, model type attribute values, and task target attribute values ​​to the standard element semantic integration model, it is easier to select the most suitable model type, which can give full play to the advantages of the model, further improve the accuracy of semantic processing, enhance the efficiency and performance of semantic processing, avoid processing irrelevant standard data, and reduce the waste of computing resources.

[0117] Specifically, based on the task attribute values, the weight allocation rule for the standard feature semantic integration model is calculated using linear rules, where...

[0118] The weighting rules are proportional to the task attribute values.

[0119] In practice, weight allocation rules are defined based on task attribute values. For example, a weight coefficient can be assigned to each task attribute value, and then these weight coefficients are multiplied by the corresponding task attribute values ​​to obtain the total weight of each model.

[0120] Calculate the weights: Calculate the weights of each standard feature in the semantic integration model using linear rules. For example, if a model has a standard data source attribute value of 3, a model type attribute value of 4, and a task objective attribute value of 5, with weight coefficients of 0.2, 0.3, and 0.5 respectively, then the total weight of the model is:

[0121] Total weight = 3 × 0.2 + 4 × 0.3 + 5 × 0.5 = 0.6 + 1.2 + 2.5 = 4.3.

[0122] The weight allocation rule is proportional to the task attribute value, meaning that the training weights of the standard feature semantic ensemble model are proportional to its task relevance. This allows the standard feature semantic ensemble model to focus more on processing standard data with higher task attribute values, thereby improving the accuracy of semantic processing.

[0123] By adjusting the weight allocation rules, the system can quickly adapt to new task requirements. For example, if the target attribute value of a certain task suddenly becomes very important, its weight coefficient can be increased to make the standard feature semantic integration model pay more attention to that task.

[0124] By calculating the weight allocation rules of the standard element semantic integration model based on the task attribute values ​​using linear rules, and making the weight allocation rules proportional to the task attribute values, the training process of the model can be optimized, the accuracy of semantic processing can be improved, and the adaptability and flexibility of the system can be enhanced. This provides a scientific and quantitative solution for the weight allocation of the standard element semantic integration model, which helps to build an efficient and accurate semantic processing system.

[0125] Please see Figure 4 As shown, this is a flowchart of the present invention applying training weights to a standard feature semantic integration model, including:

[0126] Step Sp1: Initialize the standard feature semantic integration model by setting the training weights of the standard feature semantic integration model to the initial training weights.

[0127] Step Sp2: Calculate the comprehensive training weights of the standard element semantic integration model according to the weight allocation rules, and use the comprehensive training weights as the new training weights.

[0128] Step Sp3 involves applying the training weights to the training process of the standard feature semantic integration model.

[0129] In practice, before training the standard element semantic integration model begins, an initial training weight is assigned to it. This initial weight can be set empirically or calculated using a certain method. The initialization step provides a clear starting point for model training, ensuring the smooth progress of the training process. By setting initial training weights, it avoids the model starting with zero weights, thereby improving training efficiency.

[0130] Based on the weight allocation rules, the initial training weights of the standard element semantic integration model are combined with the task attribute values ​​to calculate the comprehensive training weights. For example, the weighted sum of the initial training weights and the task attribute values ​​can be used as the comprehensive training weights.

[0131] By applying training weights to the training process of a standard feature semantic ensemble model, the training of the model can be optimized, and the model's adaptability, generalization ability and performance can be improved. Using comprehensive training weights, the model can focus more on processing standard data with higher values ​​than its task attributes, thereby improving the model's performance.

[0132] Specifically, the steps to generate fuzzy predata include:

[0133] Several indicator features of semantically ambiguous words were selected, among which,

[0134] The indicator features include the degree of ambiguity, semantic similarity, and distribution characteristics of semantically ambiguous words;

[0135] Semantically ambiguous words are segmented according to a standard sampling rate to form corresponding fuzzy pre-data.

[0136] The standard sampling rate is the learning rate that the standard element semantic integration model can recognize, and for a single learning iteration, the corresponding standard sampling rate is the single learning rate.

[0137] In practical implementation, several indicator features of semantically ambiguous words are selected, including:

[0138] Fuzziness level: measures the degree of uncertainty or ambiguity of semantically vague words.

[0139] Semantic similarity: measures the semantic similarity between semantically ambiguous words and other words.

[0140] Distribution characteristics: Describe the distribution of semantically ambiguous words in the text, such as frequency of occurrence and contextual information.

[0141] By selecting multiple indicator features, the characteristics of semantically ambiguous words can be comprehensively described, providing rich information for subsequent processing. These indicator features help the model better understand the semantics and context of semantically ambiguous words, thereby improving the quality of standard data. By considering the degree of ambiguity, semantic similarity, and distribution characteristics, the model can better adapt to different types of semantically ambiguous words, improving the model's adaptability.

[0142] Based on a standard sampling rate, semantically ambiguous words are segmented into several fragments to form corresponding fuzzy pre-data. The purpose of segmentation is to decompose complex semantically ambiguous words into smaller units that the model can process. By segmenting semantically ambiguous words into smaller fragments, the complexity of model processing can be reduced, making them easier for the model to recognize and learn.

[0143] By selecting the index features of semantically ambiguous words, segmenting them according to the standard sampling rate, and forming fuzzy pre-data, complex semantically ambiguous words can be transformed into a format suitable for learning by the standard element semantic integration model. This not only improves the quality and processability of the standard data, but also optimizes the learning efficiency and generalization ability of the standard element semantic integration model, providing strong support for building an efficient and accurate semantic processing system.

[0144] Specifically, the steps for training a standard feature semantic ensemble model using a fuzzy pre-database include:

[0145] Adjust the weights of the standard element semantic integration model to the training weights, and iteratively optimize the standard element semantic integration model until the loss value of the standard element semantic integration model converges to the preset loss value.

[0146] The current standard feature semantic integration model was tested using a fuzzy pre-database;

[0147] When the test results meet the preset test results, the learning parameters of the standard element semantic integration model are saved and used as standard parameters.

[0148] In practice, the preset loss value is a threshold used to determine whether the model has reached a satisfactory performance level.

[0149] Check whether the test results meet the preset test standards. Preset test results typically include performance metrics such as accuracy, recall, and F1 score.

[0150] Through weight adjustment and iterative optimization, model testing, and parameter saving and standardization, the standard element semantic integration model can be systematically trained and optimized. This not only ensures the model's performance on the fuzzy pre-database but also improves the model's reliability and consistency, providing strong support for building an efficient and accurate semantic processing system and helping to improve the overall effect of semantic processing in enterprise management platforms.

[0151] Specifically, the learning parameters of the standard element semantic integration model are adjusted to standard parameters. Fuzzy pre-data and indicator features are input into the standard element semantic integration model. The standard element semantic integration model learns from the fuzzy pre-data, generates corresponding integration results, and labels and verifies the integration results according to the indicator features to obtain the corresponding fuzzy integration accuracy.

[0152] Specifically, the accuracy of fuzzy fusion is compared with an accuracy threshold. When the accuracy of fuzzy fusion is greater than the accuracy threshold, the corresponding standard element semantic integration model is not adjusted.

[0153] By adjusting learning parameters, inputting fuzzy pre-data and indicator features, labeling and verifying the fusion results, and comparing the fuzzy fusion accuracy with the accuracy threshold, the performance of the standard element semantic integration model can be systematically optimized. This not only ensures the accuracy and reliability of the model on fuzzy pre-data, but also supports the continuous optimization and improvement of the model.

[0154] Specifically, when the accuracy of fuzzy fusion is less than the accuracy threshold, the fuzzy pre-data is divided into several sub-fuzzy pre-data. The sub-fuzzy pre-data is then learned using the standard element semantic integration model to obtain the corresponding secondary fusion accuracy. When the secondary fusion accuracy is greater than the accuracy threshold, the corresponding standard element semantic integration model is not adjusted. When the secondary fusion accuracy is less than the accuracy threshold, the enterprise management platform determines that the corresponding standard element semantic integration model has a performance problem and filters it.

[0155] By segmenting fuzzy pre-data into sub-fuzzy pre-data and performing secondary learning and verification, the performance of the standard element semantic integration model can be further optimized. This not only enhances the model's ability to process complex standard data, but also ensures that only high-quality models are retained and used through a strict verification and screening mechanism. This helps improve the overall effect of semantic processing in the enterprise management platform, while ensuring the efficient operation and reliability of the system.

[0156] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0157] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An application scenario-based standard element semantic integration method applied to an enterprise management platform, characterized in that, The enterprise management platform has a standard element semantic integration strategy, including: Determine the task attributes of several standard feature semantic integration models, and assign corresponding task attribute values ​​to the standard feature semantic integration models according to the task attributes, wherein, The task attributes include standard data source, model type, and task objective; The task attribute values ​​include standard data source attribute values, model type attribute values, and task target attribute values; Based on the task attribute values, weight allocation rules are formulated for the standard element semantic integration model; The training weights of the standard element semantic integration model are adjusted according to the weight allocation rules, and the adjusted training weights are applied to the standard element semantic integration model. Several semantically ambiguous words from the enterprise management platform are collected, and these semantically ambiguous words are preprocessed to form corresponding fuzzy pre-data. The semantically ambiguous words refer to words in the enterprise management platform whose semantic scope has no clear boundaries; The fuzzy pre-database formed by the fuzzy pre-data is used to train the standard element semantic integration model, and the fuzzy pre-data is transmitted to the standard element semantic integration model for learning to generate the corresponding integration result; The fusion results are verified to obtain the fuzzy fusion accuracy rate corresponding to a single standard element semantic integration model. The fuzzy fusion accuracy rate is compared with an accuracy threshold. Based on the comparison result, it is determined whether to retain the corresponding standard element semantic integration model. The accuracy threshold is the minimum standard for the fuzzy integration accuracy of the standard element semantic integration model in semantically integrating the semantically fuzzy words, and is inversely correlated with the operating efficiency of the enterprise management platform. Assign standard data source attribute values ​​based on the source range of semantically ambiguous words; Assign model type attribute values ​​based on the implementation method of the standard element semantic integration model; Assign task target attribute values ​​according to the application scenario of the standard element semantic integration model; The steps for training a standard feature semantic ensemble model using a fuzzy pre-database include: The weights of the standard element semantic integration model are adjusted to the training weights, and the standard element semantic integration model is iteratively optimized until the loss value of the standard element semantic integration model converges to the preset loss value. The current standard element semantic integration model was tested using the aforementioned fuzzy pre-database; When the test results meet the preset test results, the learning parameters of the standard element semantic integration model are saved, and the learning parameters are used as standard parameters. The learning parameters of the standard element semantic integration model are adjusted to the standard parameters. The fuzzy pre-data and indicator features are input into the standard element semantic integration model. The standard element semantic integration model learns from the fuzzy pre-data and generates the corresponding integration result. The integration result is labeled and verified according to the indicator features to obtain the corresponding fuzzy integration accuracy. The steps for determining the task attributes of the standard feature semantic integration model include: The standard data source of the semantically ambiguous words is clearly defined, and the source range of the semantically ambiguous words is limited; The type of the standard element semantic integration model is clearly defined, and the implementation method of the standard element semantic integration model is limited; The task objectives of the standard element semantic integration model are clearly defined, and the application scenarios of the standard element semantic integration model are limited.

2. The method of claim 1, wherein the application scenario-based standard element semantic integration method is characterized by, The steps for assigning task attribute values ​​include: Based on the source range of the semantically ambiguous words, assign standard data source attribute values ​​to the standard element semantic integration model; Based on the implementation method of the standard element semantic integration model, assign model type attribute values ​​to the standard element semantic integration model; Based on the application scenario of the standard element semantic integration model, task target attribute values ​​are assigned to the standard element semantic integration model.

3. The standard element semantic integration method based on application scenarios according to claim 2, characterized in that, Based on the task attribute values, the weight allocation rule of the standard element semantic integration model is calculated using linear rules, wherein... The weight allocation rule is proportional to the task attribute value.

4. The standard element semantic integration method based on application scenarios according to claim 3, characterized in that, The steps for applying training weights to a standard feature semantic ensemble model include: The standard element semantic integration model is initialized by setting the training weights of the standard element semantic integration model to the initial training weights. According to the weight allocation rule, the comprehensive training weight of the standard element semantic integration model is calculated, and the comprehensive training weight is used as the new training weight; The training weights are applied to the training process of the standard element semantic integration model.

5. The standard element semantic integration method based on application scenarios according to claim 4, characterized in that, The steps to generate fuzzy predata include: Several index features of the semantically fuzzy words are selected, among which, The indicator features include the degree of ambiguity, semantic similarity, and distribution characteristics of the semantically ambiguous words; The semantically ambiguous words are segmented according to a standard sampling rate to form corresponding fuzzy pre-data, wherein... The standard sampling rate is the learning rate that the standard element semantic integration model can recognize, and for a single learning iteration, the corresponding standard sampling rate is a single learning rate.

6. The standard element semantic integration method based on application scenarios according to claim 5, characterized in that, The accuracy of the fuzzy integration is compared with the accuracy threshold. When the accuracy of the fuzzy integration is greater than the accuracy threshold, the corresponding standard element semantic integration model is not adjusted.

7. The standard element semantic integration method based on application scenarios according to claim 6, characterized in that, When the accuracy of the fuzzy fusion is less than the accuracy threshold, the fuzzy pre-data is divided into several sub-fuzzy pre-data. The sub-fuzzy pre-data is then learned using the standard element semantic integration model to obtain the corresponding secondary fusion accuracy. When the secondary fusion accuracy is greater than the accuracy threshold, the corresponding standard element semantic integration model is not adjusted. When the secondary fusion accuracy is less than the accuracy threshold, the enterprise management platform determines that the corresponding standard element semantic integration model has a performance problem and filters the corresponding standard element semantic integration model.

Images