A fast annotation method based on artificial intelligence large model

Abstract

The present invention relates to the field of artificial intelligence technology, and discloses a rapid annotation method based on an artificial intelligence large model, comprising: step 1, extracting entity sets and relationship sets in a specific field through a document set in a target field, wherein entity recognition technology is used to identify field-related nouns, terms and keywords in the document to form an entity set, and relationship extraction technology is used to analyze the logical and semantic associations between entities to generate a relationship set, and the entity set and the relationship set are combined to construct a knowledge triple, which contains the association relationship between entities. By introducing a domain knowledge graph, the entity and relationship information extracted from the target domain document is structured into a knowledge triple, and the semantic information of the annotation is matched and optimized with the data to be annotated, so as to achieve domain adaptability and semantic consistency of the annotation result, and obtain a higher accuracy and consistency effect of the annotated data in the professional field.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a fast labeling method based on an artificial intelligence large model. Background Art

[0002] Amid the rapid development of artificial intelligence and big data technologies, data annotation, as a fundamental step in deep learning model training, directly impacts model performance and widespread application. The widespread application of deep learning models in natural language processing, computer vision, and multimodal data processing has led to a rapid increase in the scale and complexity of annotation tasks, placing higher demands on both efficiency and quality. In the current technological landscape, traditional annotation methods are struggling to meet these growing demands.

[0003] Traditional manual labeling methods rely mainly on manual operations and have a certain degree of flexibility and accuracy, but they suffer from low efficiency, high time costs, and poor consistency. Manual labeling has certain advantages in small-scale data labeling. However, when labeling tasks are expanded to large-scale data and cross-domain and multimodal scenarios, the limitations of traditional methods become increasingly prominent, as shown below:

[0004] The knowledge of existing large models mainly comes from a wide range of general data and lacks deep semantic understanding for specific fields, resulting in insufficient accuracy and consistency of annotation results in professional fields;

[0005] Large models cannot flexibly adapt to the needs of different labeling tasks. Existing methods lack dynamic optimization methods for labeling task prompt templates, resulting in low efficiency and insufficient adaptability in multi-scene labeling tasks.

[0006] In labeling tasks, existing automatic labeling methods are not ideal for handling complex and ambiguous samples, requiring manual review, which increases workload and time costs. At the same time, they fail to fully utilize the results of manual review to optimize model performance;

[0007] The labeling task has expanded from a single modality to multimodality. Existing labeling technologies have limited capabilities for collaborative labeling of multimodal data. At the same time, there is a lack of optimization methods for the hierarchical and refined processing of labeling results.

[0008] Therefore, those skilled in the art provide a fast annotation method based on an artificial intelligence large model to solve the above-mentioned problems. Summary of the Invention

[0009] In view of the shortcomings of the existing technology, the present invention provides a fast labeling method based on an artificial intelligence large model to solve the problems raised in the above background technology.

[0010] To achieve the above objectives, the present invention is implemented through the following technical solutions: a rapid annotation method based on an artificial intelligence large model, comprising:

[0011] Step 1: Extract entity sets and relationship sets in a specific domain from the document collection of the target domain. Entity recognition technology is used to identify domain-related nouns, terms, and keywords in the document to form an entity set. Relation extraction technology is used to analyze the logical and semantic associations between entities to generate a relationship set. The entity set and the relationship set are combined to construct knowledge triples. The knowledge triples contain the association relationships between entities. The constructed knowledge triples are stored in a graph database to form a domain knowledge graph.

[0012] Step 2: Based on the knowledge graph, perform knowledge enhancement annotation on the data to be annotated, match the data to be annotated with the domain knowledge graph, and perform association matching based on the entities appearing in the data to be annotated and the corresponding entities in the knowledge graph. Optimize the annotated semantic information by querying the contextual association relationship of the entities in the knowledge graph. During the matching process, the logic constructed based on the entities and relationships in the knowledge triples provides domain information background for step 3.

[0013] Step 3: Based on the results of knowledge-enhanced annotation, a dynamic prompt template is designed for optimization. By combining historical data of the annotation task and analyzing the historical annotation results, the content and structure of the prompt template are dynamically adjusted. In the dynamic prompt template, the domain information provided by the knowledge-enhanced annotation is combined to generate prompt content for the specific annotation task for use by the artificial intelligence large model in the annotation task.

[0014] Step 4: Use dynamic prompt templates and large models to perform preliminary labeling on the unlabeled data, and screen samples in the unlabeled data through active learning. During the screening process, the uncertainty score of the sample is calculated based on the classification probability distribution in the labeling results. The samples are sorted according to the uncertainty score. Some samples are selected for manual review based on the set uncertainty threshold. After manual review, the corrected data is fed back to the model training set to update the model parameters.

[0015] Step 5: For multimodal labeling tasks, use the optimized large model to label the multimodal data, and extract text feature vectors and visual feature vectors from the multimodal data of text and images. Collaborative labeling of multimodal data is achieved through alignment and matching of feature vectors. At the same time, for the hierarchical structure of labels, a label system with different granularities is constructed. The labels are optimized according to the requirements of the labeling task, and the labeling probability is adjusted by calculating the correlation between the label hierarchies. The labeling results are generated and output to the automated labeling workflow to complete the overall rapid labeling task process.

[0016] Preferably, the generation of the relationship set in the knowledge triple in step 1 includes constructing relationship weights by counting the co-occurrence frequencies between entities in the target domain documents, and the relationship weights are calculated by the following formula:

[0017]

[0018] Among them, w ij For entity e i and e j The relationship weight between

[0019] Frequency (e i , e j ) is entity e i and e j The number of times they co-occur in the document,

[0020] Frequency (e i ) and frequency (e j ) is entity e i and e j The total number of times it occurs in the document.

[0021] Preferably, the relationship weight is used to adjust the priority of entity associations in the knowledge graph, and the priority calculation formula is:

[0022] Among them, p ij For entity e i To entity e j priority of entity e i The number of all directly related entities in the knowledge graph, w ij For entity e i and e j The relationship weight between

[0023] Represents entity e i The total relationship weight of all related entities is used as the denominator of normalization to normalize the relationship weight to the range of [0, 1].

[0024] k is the iterative calculation entity e i The weight of the relationships with all entities.

[0025] Preferably, the optimization of the dynamic prompt template in step 3 adopts a feedback learning-based approach to update the weight of the prompt template. The weight update formula is:

[0026] w t =w t-1 +a(Q t -Q t-1 ),

[0027] Among them, w t and w t-1 is the weight of the current and previous round prompt templates, Q t and Q t-1 is the accuracy of the current and previous rounds of labeling tasks, and a is the learning rate;

[0028] After the weight of the prompt template is adjusted, more targeted task descriptions are generated by screening frequently appearing keywords, which are used to improve the labeling performance of subsequent large models.

[0029] Preferably, the uncertainty score of the sample in step 4 is calculated using the entropy value formula:

[0030]

[0031] Among them, H(x) is the uncertainty score of sample x, K is the number of categories predicted by the model,

[0032] p(y i |x) is the sample belonging to category y i probability.

[0033] Preferably, the samples are selected for manual review based on a set uncertainty threshold r, and the threshold is dynamically adjusted by the following formula:

[0034] r t =r t-1 +y·(R target -R actual ),

[0035] Among them, r t and r t-1 is the uncertainty threshold of the current round and the previous round,

[0036] R target is the target manual review sample ratio, R actual is the current manual review sample ratio, and y is the learning rate.

[0037] Preferably, in step 5, the text and image data are annotated by a multimodal model, and the similarity between the text feature vector and the visual feature vector is calculated by cosine similarity:

[0038]

[0039] Among them, sin(v, t) is the similarity between the visual feature vector v and the text feature vector t,

[0040] ∥v∥ and ∥t∥ are the moduli of v and t.

[0041] Preferably, the hierarchical structure of the tags is optimized using a multimodal model, and the association between the tags is adjusted using the following formula:

[0042] Among them, p′(l i |x) is the sample x belonging to label l i The optimized probability of M is the number of all labels of sample x, p(l i |x) is the initial probability of the sample.

[0043] Preferably, after the similarity between the text feature vector and the visual feature vector is calculated, the feature alignment relationship of the multimodal model is optimized by contrastive learning, and the optimization target formula is:

[0044]

[0045] Among them, L contrastive is the loss function of contrastive learning, sin(v, t) is the similarity between the visual feature vector v and the text feature vector t, R is the temperature parameter, G is the number of negative samples,

[0046] sin(v,t i ) is the visual feature vector v and the i-th negative sample text feature vector t i similarity.

[0047] Preferably, when optimizing the association relationship between tags, the hierarchical structure depth of the tags is considered and the depth weight is adjusted. The optimization formula is:

[0048]

[0049] Among them, w′(l i ) indicates the label (l i ),

[0050] depth(l i ) indicates label l i The depth value in the hierarchical tag structure,

[0051] B is the adjustment coefficient, which is used to control the impact of depth on weight.

[0052] The present invention provides a rapid annotation method based on an artificial intelligence large model. It has the following beneficial effects:

[0053] 1. The present invention introduces a domain knowledge graph to structure the entity and relationship information extracted from the target domain documents into knowledge triples, and matches the semantic information of the annotations with the data to be annotated to achieve domain adaptability and semantic consistency of the annotation results, thereby obtaining higher accuracy and consistency of the annotated data in professional fields.

[0054] 2. The present invention uses dynamic prompt template optimization technology and combines historical data of labeling tasks to adjust the prompt content and weight in real time, thereby realizing the flexible adaptability of the large model to the requirements of different labeling tasks, significantly improving the labeling efficiency and enhancing the model's adaptability to multi-scene labeling tasks.

[0055] 3. The present invention combines active learning with uncertainty sampling technology to screen out uncertainty samples for manual review and dynamically update model parameters, thereby optimizing the performance of large models in complex sample labeling tasks, reducing the workload of manual review and improving the overall accuracy of labeling.

[0056] 4. The present invention combines multimodal feature alignment technology with a hierarchical label optimization method to achieve collaborative annotation of multimodal data such as text and images and refined processing of annotation results, so that the multimodal data annotation is more consistent, and the annotation results have a hierarchical and refined effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 Flowchart of the present invention. DETAILED DESCRIPTION

[0058] To help those skilled in the art understand the present invention, the following will provide a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. Obviously, the described embodiments are only partial embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0059] The present invention is described in detail below with reference to the accompanying drawings:

[0060] Example:

[0061] Please see the attached Figure 1 The embodiment of the present invention provides a rapid annotation method based on an artificial intelligence large model, comprising:

[0062] Step 1: Extract entity sets and relationship sets in a specific domain from the document collection of the target domain. Entity recognition technology is used to identify domain-related nouns, terms, and keywords in the document to form an entity set. Relation extraction technology is used to analyze the logical and semantic associations between entities to generate a relationship set. The entity set and the relationship set are combined to construct knowledge triples. The knowledge triples contain the association relationships between entities. The constructed knowledge triples are stored in a graph database to form a domain knowledge graph.

[0063] Step 2: Based on the knowledge graph, perform knowledge enhancement annotation on the data to be annotated, match the data to be annotated with the domain knowledge graph, and perform association matching based on the entities appearing in the data to be annotated and the corresponding entities in the knowledge graph. Optimize the annotated semantic information by querying the contextual association relationship of the entities in the knowledge graph. During the matching process, the logic constructed based on the entities and relationships in the knowledge triples provides domain information background for step 3.

[0064] Step 3: Based on the results of knowledge-enhanced annotation, a dynamic prompt template is designed for optimization. By combining historical data of the annotation task and analyzing the historical annotation results, the content and structure of the prompt template are dynamically adjusted. In the dynamic prompt template, the domain information provided by the knowledge-enhanced annotation is combined to generate prompt content for the specific annotation task for use by the artificial intelligence large model in the annotation task.

[0065] Step 4: Use dynamic prompt templates and large models to perform preliminary labeling on the unlabeled data, and screen samples in the unlabeled data through active learning. During the screening process, the uncertainty score of the sample is calculated based on the classification probability distribution in the labeling results. The samples are sorted according to the uncertainty score. Some samples are selected for manual review based on the set uncertainty threshold. After manual review, the corrected data is fed back to the model training set to update the model parameters.

[0066] Step 5: For multimodal labeling tasks, use the optimized large model to label the multimodal data, and extract text feature vectors and visual feature vectors from the multimodal data of text and images. Collaborative labeling of multimodal data is achieved through alignment and matching of feature vectors. At the same time, for the hierarchical structure of labels, a label system with different granularities is constructed. The labels are optimized according to the requirements of the labeling task, and the labeling probability is adjusted by calculating the correlation between the label hierarchies. The labeling results are generated and output to the automated labeling workflow to complete the overall rapid labeling task process.

[0067] Benefits of Step 1: Entity recognition and relationship extraction techniques are used to extract domain-specific entity and relationship information from documents in the target domain, effectively constructing knowledge triples and storing them in a graph database. Knowledge triples not only structure domain knowledge but also provide professional background information for subsequent annotation tasks. The construction of a domain knowledge graph improves the adaptability and semantic understanding of annotation tasks in professional scenarios, resulting in more accurate and consistent annotation results.

[0068] Benefits of Step 2: Matching the data to be annotated with the knowledge graph, leveraging the contextual relationships within the knowledge graph to optimize annotation semantics, effectively increasing the degree of domain knowledge reliance on annotation tasks. The matching process further strengthens the semantic consistency of the annotation results based on the logical relationships between knowledge triples, ensuring that the annotation results adhere to the logic of domain knowledge, especially in data scenarios involving complex semantic relationships.

[0069] Benefits of Step 3: Dynamic prompt templates can be optimized based on historical annotation task data. By adjusting prompt content and structure in real time, the adaptability of annotation tasks to dynamically changing requirements is improved. Combined with the domain information provided by knowledge-enhanced annotation, the generated prompt content can further enhance the task understanding capabilities of large models, making them adaptable to multi-scenario annotation requirements. By dynamically optimizing the weights of prompt templates, the efficiency of annotation tasks and the model's annotation performance are improved, while the complexity of human intervention is reduced.

[0070] Benefits of Step 4: Active learning technology is used to screen uncertain samples from unlabeled data, prioritizing manual review of complex and ambiguous samples, effectively reducing unnecessary manual review workload. The uncertainty scoring mechanism ensures scientific sample selection, focusing manual review on high-value samples and improving overall labeling efficiency. The reviewed sample data is fed back into the model training set to update model parameters and gradually optimize model performance, further improving the accuracy and efficiency of automatic labeling.

[0071] Benefits of Step 5: By extracting textual and visual feature vectors and utilizing feature vector alignment and matching, collaborative annotation of multimodal data such as text and images is achieved, addressing the problem of poor semantic consistency in multimodal tasks. A hierarchical labeling system is constructed within the annotation results, and hierarchical optimization strategies are used to meet the need for fine-grained label processing, resulting in a layered and refined annotation result. The integration of automated annotation workflows enables the automatic generation, storage, and export of annotation data, significantly reducing the complexity of manual intervention and improving the automation level of annotation tasks.

[0072] The generation of the relationship set in the knowledge triple in step 1 includes constructing the relationship weight by counting the co-occurrence frequency between entities in the target domain documents. The relationship weight is calculated by the following formula:

[0073]

[0074] Among them, w ij For entity e i and e j The relationship weight between

[0075] Frequency (e i , e j ) is entity e i and ej The number of times they co-occur in the document,

[0076] Frequency (e i ) and frequency (e j ) is entity e i and e j The total number of times it appears in the document;

[0077] Relationship weight is used to adjust the priority of entity association in the knowledge graph. The priority calculation formula is:

[0078]

[0079] Among them, p ij For entity e i To entity e j priority of entity e i The number of all directly related entities in the knowledge graph, w ij For entity e i and e j The relationship weight between

[0080] Represents entity e i The total relationship weight of all related entities is used as the denominator of normalization to normalize the relationship weight to the range of [0, 1].

[0081] k is the iterative calculation entity e i The weight of the relationships with all entities.

[0082] The benefits of constructing relationship weights: They convert semantic associations between entities into quantifiable weights, addressing the subjective nature of entity relationship representations. By using co-occurrence frequency statistics, we fully leverage the corpus data in domain documents, ensuring a data-driven, scientific basis for knowledge graph construction. They effectively identify core entity pairs with close relationships in the target domain, providing high-quality semantic support for subsequent annotation tasks.

[0083] The benefits of relationship weight-based priority calculation include: normalizing entity associations within the knowledge graph, facilitating subsequent annotation tasks. This ensures that the relationships most closely associated with a particular entity receive higher semantic weight during annotation, improving the effectiveness of knowledge augmentation annotation. Through iterative priority calculation, the associations between entities within the knowledge graph are dynamically optimized, making the knowledge graph's semantic expression more accurate.

[0084] Benefits of iterative optimization: Through iterative optimization, the priority of entity relationships is gradually adjusted, dynamically enhancing the expressive power of the knowledge graph. This avoids bias in the initial data, makes the knowledge graph more globally semantically consistent, and provides a more robust semantic context for subsequent tasks.

[0085] In step 3, the optimization of the dynamic prompt template adopts a feedback learning-based approach to update the weight of the prompt template. The weight update formula is:

[0086] w t =w t-1 +a(Q t -Q t -1),

[0087] Among them, w t and w t-1 is the weight of the current and previous round prompt templates, Q t and Q t-1 is the accuracy of the current and previous rounds of labeling tasks, and a is the learning rate;

[0088] After the weight of the prompt template is adjusted, more targeted task descriptions are generated by filtering out frequently appearing keywords, which is used to improve the labeling performance of subsequent large models.

[0089] The benefits of feedback-based weight updates: The weights of the prompt template are dynamically adjusted using historical annotation task data, ensuring that the prompt content closely matches the annotation task requirements. The learning rate a controls the magnitude of weight updates, ensuring a smooth and controllable prompt optimization process. This improves the dynamic prompt template's adaptability to changing scenarios within the annotation task, ensuring efficient and accurate annotation in complex and diverse situations.

[0090] Benefits of high-frequency keyword filtering: High-frequency keyword filtering focuses on the core semantic content of the annotation task, effectively avoiding redundant and inefficient information interference in the template content. The generated task description is more precise for the task scenario, quickly guiding large models to complete the annotation task. This reduces the time cost of manually designing prompt templates, improves the ability to automatically generate prompt templates, and provides more efficient support for complex annotation tasks.

[0091] Benefits of improving large-scale model annotation performance include: improving the model's ability to understand annotation task instructions and reducing annotation errors caused by inappropriate prompts. For multi-round annotation tasks, optimized prompt templates significantly reduce the need for manual intervention and improve annotation efficiency. This also enhances the model's generalization capabilities in complex scenarios, adapting to diverse task requirements and ensuring more reliable and consistent annotation results.

[0092] In step 4, the uncertainty score of the sample is calculated using the entropy formula:

[0093]

[0094] Among them, H(x) is the uncertainty score of sample x, K is the number of categories predicted by the model,

[0095] p(yi |x) is the sample belonging to category y i probability;

[0096] Samples are selected for manual review based on the set uncertainty threshold r, and the threshold is dynamically adjusted using the following formula:

[0097] r t =r t-1 +y·(R target -R actual ),

[0098] Among them, r t and r t-1 is the uncertainty threshold of the current round and the previous round,

[0099] R target is the target manual review sample ratio, R actual is the current manual review sample ratio, and y is the learning rate.

[0100] Benefits of entropy-based uncertainty scoring: Entropy provides a quantitative measure of sample prediction uncertainty. Samples with high entropy values ​​indicate high uncertainty in the model's predictions for that sample, helping to identify samples with problematic labeling results. By analyzing the sample's classification probability, it effectively screens out samples with poor model performance, avoiding wasted manual review resources on samples with high confidence in the model's predictions. It effectively focuses on complex, ambiguous, and ambiguous samples with high uncertainty, providing clear priority targets for manual review and improving review efficiency.

[0101] The benefits of dynamically adjusting the uncertainty threshold: The threshold adjustment mechanism compares the target manual review sample ratio with the current review sample ratio and dynamically adjusts the uncertainty threshold based on the deviation, ensuring that the number of review samples aligns with actual needs. If the current review sample ratio exceeds the target ratio, the threshold is lowered, reducing the frequency of screening uncertainty samples and controlling the workload of manual review. If the review ratio is insufficient, the threshold is increased to focus more high-value samples into the review process.

[0102] Benefits of reducing manual review workload: The total number of review samples is significantly reduced, reducing manual review workload while simultaneously focusing human effort on complex samples where annotation quality improvement is most needed, optimizing the efficiency of human resources. High-quality review sample data is fed back into the model to further optimize model parameters, continuously improving model performance in subsequent annotation tasks and gradually reducing reliance on manual review. A dynamic learning mechanism enables the model to gradually adapt to the distribution of complex samples, improving its ability to handle high-uncertainty samples and reducing overall annotation errors.

[0103] In step 5, the text and image data are annotated using a multimodal model, and the similarity between the text feature vector and the visual feature vector is calculated using cosine similarity:

[0104]

[0105] Among them, sin(v, t) is the similarity between the visual feature vector v and the text feature vector t,

[0106] ∥v∥ and ∥t∥ are the moduli of v and t;

[0107] After calculating the similarity between the text feature vector and the visual feature vector, the feature alignment relationship of the multimodal model is optimized through contrastive learning. The optimization objective formula is:

[0108]

[0109] Among them, L contrastive is the loss function of contrastive learning, sin(v, t) is the similarity between the visual feature vector v and the text feature vector t, R is the temperature parameter, G is the number of negative samples,

[0110] sin(v,t i ) is the visual feature vector v and the i-th negative sample text feature vector t i similarity.

[0111] The benefits of using cosine similarity to precisely align text and visual features: Cosine similarity measures the directional consistency of text and visual feature vectors, accurately capturing the semantic associations between modal features. By standardizing text and visual features, the impact of feature vector scale differences on similarity calculations is avoided, ensuring that similarity values ​​have stronger semantic interpretation capabilities. This improves feature alignment between text and image modalities, ensuring the collaborative consistency of multimodal data in annotation tasks, and making annotation results more reliable and accurate.

[0112] Benefits of optimizing feature alignment relationships in multimodal models based on contrastive learning: Contrastive learning enhances the multimodal model's ability to distinguish positive feature alignment by maximizing the similarity of positive samples while minimizing the similarity between visual features and textual features of negative samples. A temperature parameter is used to adjust the similarity distribution between positive and negative samples to ensure the stability of the optimization process. In multimodal annotation tasks, optimizing feature alignment relationships improves the model's understanding of complex text-image semantic relationships, ensuring more refined and semantically consistent annotation results.

[0113] The benefits of semantic consistency and refinement in multimodal annotation include: improving the model's understanding of multimodal input and enhancing cross-modal semantic consistency in annotation tasks. In practical applications, optimized multimodal models can accurately capture the semantic relationships between multimodal data and provide annotation results for complex scenarios. By optimizing feature alignment relationships, they address the common modality mismatch problem in multimodal annotation tasks, significantly improving the practicality and accuracy of annotation results.

[0114] The multimodal model is used to optimize the hierarchical structure of labels, and the relationship between labels is adjusted using the following formula:

[0115]

[0116] Among them, p′(l i |x) is the sample x belonging to label l i The optimized probability of M is the number of all labels of sample x, p(l i |x) is the initial probability of the sample;

[0117] When optimizing the association between tags, the hierarchical structure depth of the tags is considered and the depth weight is adjusted. The optimization formula is:

[0118]

[0119] Among them, w′(l i ) indicates the label (l i ),

[0120] depth(l i ) indicates label l i The depth value in the hierarchical tag structure,

[0121] B is the adjustment coefficient, which is used to control the impact of depth on weight.

[0122] Optimizing the relationships between labels improves the hierarchical nature of labeling results: Normalizing all label probabilities for a sample ensures that the label probability distribution falls within a specified range, preventing deviations in labeling results due to inconsistent probabilities between labels. Label optimization fully utilizes initial probabilities, and normalization improves the scientific nature and rationality of the label probability distribution. By optimizing the relationships between labels, semantic accuracy is ensured while preserving the relative hierarchy between labels, making the labeling more consistent with the hierarchical nature of the data structure.

[0123] The benefits of introducing depth weighting to enhance the hierarchical expression of the label structure include: The depth weighting of labels is dynamically adjusted based on their depth in the hierarchy. Deeper labels are given greater weight, prioritizing fine-grained semantic information, while shallower labels focus more on macro-classification, ensuring a global and local balance in the labeling results. The adjustment coefficient controls the impact of the depth weighting on the optimization process, allowing for flexible adaptation to the needs of different labeling tasks.

[0124] Supporting multi-granularity annotation improves the refinement of annotation results. Benefits include: Flexible adjustment of annotation granularity based on the requirements of the annotation task. This improves the practicality of annotation results, making the annotated data suitable for both macro-analysis and detailed research. It also ensures semantic consistency across multiple granularity levels, effectively avoiding conflicts and disconnections between annotation results at different granularities.

[0125] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A fast labeling method based on artificial intelligence large model, characterized in that: include: Step 1: Extract entity sets and relationship sets in a specific domain from the document collection of the target domain. Entity recognition technology is used to identify domain-related nouns, terms, and keywords in the document to form an entity set. Relation extraction technology is used to analyze the logical and semantic associations between entities to generate a relationship set. The entity set and the relationship set are combined to construct knowledge triples. The knowledge triples contain the association relationships between entities. The constructed knowledge triples are stored in a graph database to form a domain knowledge graph. Step 2: Based on the knowledge graph, perform knowledge enhancement annotation on the data to be annotated, match the data to be annotated with the domain knowledge graph, and perform association matching based on the entities appearing in the data to be annotated and the corresponding entities in the knowledge graph. Optimize the annotated semantic information by querying the contextual association relationship of the entities in the knowledge graph. During the matching process, the logic constructed based on the entities and relationships in the knowledge triples provides domain information background for step 3. Step 3: Based on the results of knowledge-enhanced annotation, a dynamic prompt template is designed for optimization. By combining historical data of the annotation task and analyzing the historical annotation results, the content and structure of the prompt template are dynamically adjusted. In the dynamic prompt template, the domain information provided by the knowledge-enhanced annotation is combined to generate prompt content for the specific annotation task for use by the artificial intelligence large model in the annotation task. Step 4: Use dynamic prompt templates and large models to perform preliminary labeling on the unlabeled data, and screen samples in the unlabeled data through active learning. During the screening process, the uncertainty score of the sample is calculated based on the classification probability distribution in the labeling results. The samples are sorted according to the uncertainty score. Some samples are selected for manual review based on the set uncertainty threshold. After manual review, the corrected data is fed back to the model training set to update the model parameters. Step 5: For multimodal labeling tasks, use the optimized large model to label the multimodal data, and extract text feature vectors and visual feature vectors from the multimodal data of text and images. Collaborative labeling of multimodal data is achieved through alignment and matching of feature vectors. At the same time, for the hierarchical structure of labels, a label system with different granularities is constructed. The labels are optimized according to the requirements of the labeling task, and the labeling probability is adjusted by calculating the correlation between the label hierarchies. The labeling results are generated and output to the automated labeling workflow to complete the overall rapid labeling task process.

2. A rapid labeling method based on an artificial intelligence large model according to claim 1, characterized in that: The generation of the relationship set in the knowledge triple in step 1 includes constructing relationship weights by counting the co-occurrence frequencies between entities in the target domain documents. The relationship weights are calculated using the following formula: Among them, w ij For entity e i and e j The relationship weight between Frequency (e i , e j ) is entity e i and e j The number of times they co-occur in the document, Frequency (e i ) and frequency (e j ) is entity e i and e j The total number of times it occurs in the document.

3. A rapid labeling method based on artificial intelligence large model according to claim 2, characterized in that: The relationship weight is used to adjust the priority of entity association in the knowledge graph. The priority calculation formula is: Among them, p ij For entity e i To entity e j priority of entity e i The number of all directly related entities in the knowledge graph, w ij For entity e i and e j The relationship weight between Represents entity e i The total relationship weight of all related entities is used as the denominator of normalization to normalize the relationship weight to the range of [0, 1]. k is the iterative calculation entity e i The weight of the relationships with all entities.

4. The rapid annotation method based on artificial intelligence large model according to claim 1 is characterized in that: The optimization of the dynamic prompt template in step 3 adopts a feedback learning-based approach to update the weight of the prompt template. The weight update formula is: w t =w t-1 +a(Q t -Q t-1 ), Among them, w t and w t-1 is the weight of the current and previous round prompt templates, Q t and Q t-1 is the accuracy of the current and previous rounds of labeling tasks, and a is the learning rate; After the weight of the prompt template is adjusted, more targeted task descriptions are generated by screening frequently appearing keywords, which are used to improve the labeling performance of subsequent large models.

5. The rapid annotation method based on artificial intelligence large model according to claim 1 is characterized in that: In step 4, the uncertainty score of the sample is calculated using the entropy value formula: Among them, H(x) is the uncertainty score of sample x, K is the number of categories predicted by the model, p(y i |x) is the sample belonging to category y i probability.

6. A rapid labeling method based on artificial intelligence large model according to claim 5, characterized in that: The samples are selected for manual review based on the set uncertainty threshold r, and the threshold is dynamically adjusted by the following formula: r t =r t-1 +y·(R target -R actual ), Among them, r t and r t-1 is the uncertainty threshold of the current round and the previous round, R target is the target manual review sample ratio, R actual is the current manual review sample ratio, and y is the learning rate.

7. The rapid annotation method based on artificial intelligence large model according to claim 1 is characterized in that: In step 5, the text and image data are annotated using a multimodal model, and the similarity between the text feature vector and the visual feature vector is calculated using cosine similarity: Among them, sin(v, t) is the similarity between the visual feature vector v and the text feature vector t, ||v|| and ||t|| are the moduli of v and t.

8. The rapid annotation method based on artificial intelligence large model according to claim 7 is characterized in that: The multimodal model is used to optimize the hierarchical structure of tags, and the relationship between tags is adjusted by the following formula: Among them, p′(l i |x) is the sample x belonging to label l i The optimized probability of M is the number of all labels of sample x, p(l i |x) is the initial probability of the sample.

9. The rapid labeling method based on artificial intelligence large model according to claim 7 is characterized in that: After the similarity between the text feature vector and the visual feature vector is calculated, the feature alignment relationship of the multimodal model is optimized by contrastive learning. The optimization target formula is: Among them, L contrastive is the loss function of contrastive learning, sin(v, t) is the similarity between the visual feature vector v and the text feature vector t, R is the temperature parameter, G is the number of negative samples, sin(v,t i ) is the visual feature vector v and the i-th negative sample text feature vector t i similarity.

10. The rapid annotation method based on artificial intelligence large model according to claim 8, characterized in that: When optimizing the association between tags, the hierarchical structure depth of the tags is considered and the depth weight is adjusted. The optimization formula is: Among them, w′(l i ) indicates the label (l i ), depth(l i ) indicates label l i The depth value in the hierarchical tag structure, B is the adjustment coefficient, which is used to control the impact of depth on weight.

Images