Dialectical ontology fused entity identification method, apparatus and device, and storage medium
By combining large language models and dialectical ontology, entity recognition results are dynamically generated, which solves the limitations of static ontology and the lack of generalization ability in low-resource scenarios in existing technologies, and achieves accurate recognition and relationship understanding of emerging entities.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-26
- Publication Date
- 2026-04-10
AI Technical Summary
Existing entity recognition methods suffer from limitations due to static ontology, making dynamic updates difficult and exhibiting poor generalization ability in low-resource application scenarios, thus failing to effectively identify newly emerging entities.
Static entity recognition is performed using a large language model, combined with dialectical ontology and unsupervised rule technology to generate dynamic entities. Probabilistic causal reasoning and contradiction evolution are used to identify potential entities, generate rule entities and perform consistency verification, thus achieving dynamic closed-loop entity recognition.
It achieves accurate identification and understanding of emerging entities in low-resource scenarios, maintains the timeliness and integrity of the system, can identify and infer the relationships between entities, and dynamically update the entity identification results.
Smart Images
Figure CN121835671A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of entity recognition technology, specifically to an entity recognition method, apparatus, device, and storage medium that integrates dialectical ontology. Background Technology
[0002] In existing technologies, entity recognition is widely used in the field of artificial intelligence, including rule-based, feature-based machine learning, and Hidden Markov Model (HMM)-based entity recognition methods. These technologies rely on predefined rules, manually designed features, or statistical hypotheses to identify entities. For example, rule-based methods require predefined language and domain-specific rules; feature-based machine learning methods require manually designed features such as part-of-speech tagging and contextual information, followed by training and prediction using machine learning algorithms; and HMM-based methods require predicting the current label based on the previous label (entity type). However, these existing technologies generally suffer from limitations such as static ontology, difficulty in dynamic updates, and poor generalization ability in low-resource application scenarios. For example, in low-resource scenarios, existing technologies often fail to effectively identify newly emerging entities, resulting in poor recognition performance. Summary of the Invention
[0003] At least one embodiment of this application provides an entity recognition method, apparatus, device, and storage medium that integrates dialectical ontology, in order to solve the problems of static ontology limitations in the prior art, which make it difficult to dynamically update and have poor generalization ability in low-resource application scenarios.
[0004] To solve the above-mentioned technical problems, this application is implemented as follows: In a first aspect, embodiments of this application provide an entity recognition method that integrates dialectical ontology, including: Based on the input text, static entity recognition is performed using a large language model to obtain at least one static entity and the attribute values corresponding to each static entity. Based on the attribute values corresponding to each static entity and the preset attribute conflict quantification algorithm, the target static entity that will undergo entity transition is determined, as well as the first entity generated after the target static entity undergoes entity transition. Perform probabilistic causal reasoning on the first entity to obtain the contradictory entity pair corresponding to the first entity and the second entity in the contradictory entity pair that contradicts the first entity. By performing fusion reasoning on the contradictory entity pairs, a third entity is obtained; Based on the relationship rules between the contradictory entity pairs and the corresponding third entity, generate corresponding rule entities; A consistency check is performed on all the static entities, the first entity, the second entity, the third entity, and the rule entity to obtain an entity list.
[0005] Specifically, in the method described above, the step of performing static entity recognition based on the input text using a large language model to obtain at least one static entity and the attribute values corresponding to each static entity includes: The input text is processed according to the large language model to generate a candidate entity set, which includes at least one candidate entity. Based on the part-of-speech category of each candidate entity and its position in the corresponding sentence, the comprehensive weight of each candidate entity is obtained. Candidate entities whose overall weight is greater than a preset weight threshold are identified as static entities, and attribute values corresponding to each static entity are dynamically obtained.
[0006] Specifically, in the method described above, determining the target static entity that will undergo entity transition based on the attribute values corresponding to each static entity and a preset attribute conflict quantification algorithm includes: Based on the current attribute value, benchmark value, and standard deviation of the static entity, obtain the attribute conflict value of the static entity, wherein the benchmark value is the weighted average of the current attribute value and the historical attribute average value; If the attribute conflict value is greater than the preset attribute conflict threshold, then the corresponding static entity is determined to be the target static entity.
[0007] Specifically, in the method described above, the step of performing probabilistic causal reasoning on the first entity to obtain the contradictory entity pair corresponding to the first entity and the second entity in the contradictory entity pair that contradicts the first entity includes: According to a preset association rule algorithm, a pre-selected frequent 2-itemset of the first entity is identified, wherein the pre-selected frequent 2-itemset includes the first entity and a first pre-selected entity; Obtain the support and confidence information corresponding to each of the pre-selected frequent 2-itemsets, and determine the pre-selected frequent 2-itemsets whose support is greater than a preset support threshold as reference frequent 2-itemsets; A Bayesian network is constructed based on the reference 2-itemsets, and the conditional probability information corresponding to each reference 2-itemsets is obtained; Based on the support, the confidence information and the conditional probability information, entity conflict is determined for each of the reference frequent 2-itemsets to obtain the entity conflict determination result. The reference frequent 2-itemset that is determined to have a conflict is the contradictory entity pair, and the first pre-selected entity in the contradictory entity pair is the second entity.
[0008] Furthermore, in the method described above, the step of performing entity conflict judgment on each of the reference frequent 2-itemsets based on the support, the confidence information, and the conditional probability information to obtain the entity conflict judgment result includes: If the absolute difference between the positive confidence in the confidence information corresponding to the reference frequent 2-itemset and the positive conditional probability in the conditional probability information is greater than a first preset threshold, then the conflict judgment result is determined to be a conflict, wherein the positive direction represents the direction starting from the first entity and ending at the first preselected entity. If the positive confidence level corresponding to the reference frequent 2-itemset is greater than the preset confidence threshold and the support level is less than the preset support threshold, then the conflict judgment result is determined to be that there is a conflict. If the confidence direction determined based on the positive confidence and the reverse confidence in the confidence information is opposite to the network causal direction based on the positive conditional probability and the reverse conditional probability in the conditional probability information, then the conflict judgment result is determined to be a conflict, wherein the reverse direction refers to the direction starting from the first preselected entity and ending at the first entity.
[0009] Specifically, as described above, the step of performing fusion reasoning on the contradictory entity pairs to obtain the third entity includes: For the contradictory entity pairs, feature fusion and splicing are performed to obtain spliced features; Based on the efficient pointer network and the splicing features, multiple second pre-selected entities and their corresponding probability values are generated. The second pre-selected entity corresponding to the highest value among the probability values is determined as the third entity.
[0010] Specifically, in the method described above, generating the corresponding rule entity based on the relationship rules between the contradictory entity pair and the corresponding third entity includes: Construct the relationship path from the contradictory entity to the corresponding third entity; The relational path is processed using an unsupervised rule abstraction algorithm to obtain the rule entity.
[0011] Preferably, in the method described above, processing the relation path based on the unsupervised rule abstraction algorithm to obtain the rule entity includes: Based on high-dimensional mapping, each node in the relational path is mapped to a node embedding in a high-dimensional space; Based on the node embedding, obtain the attention coefficients of adjacent nodes; The attention coefficients are calculated in parallel according to the multi-head attention mechanism to obtain the spliced or averaged node embeddings. Integrate the node embeddings of the aforementioned relationship paths to generate a fixed-dimensional path embedding vector; The path embedding vector is subjected to density clustering to obtain the regular entity.
[0012] Specifically, as described above, the step of performing consistency checks on all the static entities, the first entity, the second entity, the third entity, and the rule entities to obtain an entity list includes: Obtain the entity model confidence score corresponding to the target entity, wherein the target entity is one of the static entity, the first entity, the second entity, the third entity, and the rule entity; Perform ontology consistency verification on each of the target entities to obtain logistic regression values; If the result of the exponential calculation of the entity model confidence score and the logistic regression value corresponding to the target entity is greater than or equal to the second preset threshold, then the target entity is added to the entity list.
[0013] The second aspect provides a control device, including: The static entity recognition module is used to perform static entity recognition based on the input text using a large language model, and obtain at least one static entity and the attribute values corresponding to each static entity. The dynamic entity recognition module is used to determine the target static entity that will undergo entity transition and the first entity generated after the target static entity undergoes entity transition, based on the attribute values corresponding to each static entity and a preset attribute conflict quantification algorithm. The causal reasoning module is used to perform probabilistic causal reasoning on the first entity to obtain the contradictory entity pair corresponding to the first entity and the second entity in the contradictory entity pair that contradicts the first entity. The fusion reasoning module is used to perform fusion reasoning on the contradictory entity pairs to obtain a third entity; The rule entity recognition module is used to generate corresponding rule entities based on the relationship rules between the contradictory entity pairs and the corresponding third entities; The verification module is used to perform consistency verification on all the static entities, the first entity, the second entity, the third entity, and the rule entity to obtain an entity list.
[0014] Thirdly, embodiments of this application provide an electronic device, including: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, it implements the steps of the entity recognition method of the fusion dialectical ontology as described above.
[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the entity recognition method for fusing dialectical ontology as described above.
[0016] Fifthly, embodiments of this application provide a computer program product, including computer instructions that, when executed by a processor, implement the steps of the entity recognition method for fusing dialectical ontology as described above.
[0017] Compared with existing technologies, the entity recognition method, apparatus, device, and storage medium integrating dialectical ontology provided in this application's embodiments achieve static entity recognition by initially screening static entities using a large language model, and then generating dynamic entities based on dialectical ontology (i.e., contradiction evolution) and unsupervised rule technology, thus achieving dynamic entity recognition with a dynamic closed loop, and comprehensive and accurate entity coverage. Specifically, the entity recognition method based on dialectical ontology and a large language model can effectively capture contextual information in text and identify potential entities based on contradiction evolution. Simultaneously, leveraging the powerful representation learning capabilities of the large language model and the structured knowledge framework provided by dialectical ontology, it can more effectively identify emerging entities and naturally embed them into the existing knowledge system, maintaining the system's timeliness, integrity, and scenario adaptability. Furthermore, the contradictory structure of dialectical ontology can integrate domain-specific professional knowledge into the entity recognition process. This method can not only identify entities but also understand and reason about the relationships between them. This deep knowledge integration provides strong support for ontology recognition. Attached Figure Description
[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is one of the flowcharts illustrating the entity recognition method based on the fusion of dialectical ontology in this application; Figure 2 This is the second flowchart illustrating the entity recognition method based on the fusion of dialectical ontology proposed in this application. Figure 3 This is the third flowchart illustrating the entity recognition method based on the fusion of dialectical ontology proposed in this application; Figure 4 This is the fourth flowchart illustrating the entity recognition method based on the fusion of dialectical ontology proposed in this application. Figure 5 This is the fifth flowchart illustrating the entity recognition method based on the fusion of dialectical ontology proposed in this application. Figure 6 This is the sixth flowchart illustrating the entity recognition method based on the fusion of dialectical ontology in this application; Figure 7 This is the seventh flowchart illustrating the entity recognition method based on the fusion of dialectical ontology in this application. Figure 8 This is the eighth flowchart illustrating the entity recognition method based on the fusion dialectical ontology proposed in this application. Figure 9 This is a schematic diagram of the control device of this application; Figure 10 This is a schematic diagram of the structure of the electronic device of this application. Detailed Implementation
[0019] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.
[0020] The terms “first,” “second,” etc., used in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. The terms “and / or” in the specification and claims indicate at least one of the connected objects.
[0021] See Figure 1 This application provides an entity recognition method that integrates dialectical ontology, comprising: Step S101: Based on the input text, perform static entity recognition through a large language model to obtain at least one static entity and the attribute values corresponding to each static entity. Step S102: Based on the attribute values corresponding to each static entity and the preset attribute conflict quantification algorithm, determine the target static entity that will undergo entity transition, and the first entity generated after the target static entity undergoes entity transition. Step S103: Perform probabilistic causal reasoning on the first entity to obtain the contradictory entity pair corresponding to the first entity and the second entity in the contradictory entity pair that contradicts the first entity. Step S104: Perform fusion reasoning on the contradictory entity pair to obtain a third entity; Step S105: Generate a corresponding rule entity based on the relationship rules between the contradictory entity pair and the corresponding third entity; Step S106: Perform consistency verification on all the static entities, the first entity, the second entity, the third entity, and the rule entity to obtain an entity list.
[0022] This embodiment proposes an entity recognition method that integrates dialectical ontology, also known as Dialectical Ontology Approach-Named Entity Recognition (DOA-NER). This method can recognize both static and dynamic entities. Specifically, it first performs static entity recognition based on the input text. This involves constructing an ontology filter using the language understanding capabilities of a large language model to identify static entities in the input text and obtain the attribute values corresponding to each static entity. Then, based on the attribute values of each static entity and a preset attribute conflict quantification algorithm, it identifies the target static entity that undergoes an entity transition due to attribute conflict and obtains the first entity generated after the transition. This allows for the dynamic identification of newly added entities resulting from qualitative changes caused by quantitative changes in attributes, thus improving the scope of entity recognition. After identifying the first entity, probabilistic causal reasoning is performed on the first entity based on the context in the input text to obtain a second entity that contradicts the first entity. The first entity and the second entity constitute a contradictory entity pair corresponding to the first entity. That is, through contradiction evolution, potential entities are identified, and the generalization ability of entity recognition is improved.
[0023] Furthermore, after obtaining the contradictory entity pair, the system will perform fusion reasoning on the contradictory entity pair to obtain a third entity used to reconcile and resolve the contradiction between the first entity and the second entity. That is, by utilizing the powerful representation learning ability of the large language model and the structured knowledge framework provided by dialectical ontology, emerging entities can be identified more effectively and naturally embedded into the existing knowledge system, maintaining the timeliness, integrity and scenario adaptability of the system.
[0024] Preferably, a constructed rule entity generator is used to identify the relationship rules between contradictory pairs and third entities based on an unsupervised rule abstraction algorithm, and corresponding rule entities are generated to carry the relationship between entities, thereby realizing dynamic entity recognition with dynamic closed loop.
[0025] Finally, the static and dynamic entities (including the first entity, the second entity, the third entity, and the rule entity) obtained above are subjected to consistency verification to determine whether the above entities satisfy the knowledge described by the ontology without contradictions and conform to logical rules, and based on this, an entity list is obtained, thereby ensuring that all entities in the final entity list are valid entities and ensuring the accuracy of entity recognition.
[0026] In summary, this application achieves static entity recognition by initially screening static entities using a large language model, and then generates dynamic entities based on dialectical ontology (i.e., contradiction evolution) and unsupervised rule technology, realizing a dynamic closed-loop entity recognition process with comprehensive and accurate entity coverage. The entity recognition method based on dialectical ontology and the large language model effectively captures contextual information in the text and identifies potential entities based on contradiction evolution. Furthermore, leveraging the powerful representation learning capabilities of the large language model and the structured knowledge framework provided by dialectical ontology, it can more effectively identify emerging entities and naturally embed them into the existing knowledge system, maintaining the system's timeliness, integrity, and scenario adaptability. In addition, the contradictory structure of dialectical ontology can integrate domain-specific expertise into the entity recognition process. This method not only identifies entities but also understands and infers the relationships between them. This deep knowledge integration provides strong support for ontology recognition.
[0027] See Figure 2 Specifically, in the method described above, the step of performing static entity recognition based on the input text using a large language model to obtain at least one static entity and the attribute values corresponding to each static entity includes: Step S201: Perform context processing on the input text according to the large language model to generate a candidate entity set, wherein the candidate entity set includes at least one candidate entity; Step S202: Obtain the comprehensive weight of each candidate entity based on its part-of-speech category and its position in the corresponding sentence. Step S203: Determine the candidate entities whose comprehensive weight is greater than the preset weight threshold as static entities, and dynamically obtain the attribute values corresponding to each static entity.
[0028] In this embodiment, the steps of the static entity recognition described above are illustrated. First, the input text is processed using a large language model to initially identify text fragments that may contain entities, and a candidate entity set is generated, which includes at least one candidate entity. It should be noted that, to reduce the workload of context processing, the original input text is cleaned beforehand to remove noise, errors, and inconsistencies, improving data quality and providing reliable and clean text input for subsequent processing.
[0029] Furthermore, each candidate entity is tagged with its part of speech to determine its category. Different part-of-speech weights are then assigned to each candidate entity based on its part of speech; for example, verbs have a lower weight than nouns. Simultaneously, each candidate entity is assigned a positional weight based on its importance within the corresponding sentence; for example, the weight at the beginning of a sentence is lower than the weight in the middle. The combined weights of the part-of-speech and positional weights yield the overall weight for each candidate entity. In a preferred embodiment, when obtaining the overall weight based on the part-of-speech and positional weights, it is preferable to multiply the part-of-speech weights and the positional weights.
[0030] Then, the obtained comprehensive weight is compared with a preset weight threshold obtained based on experience or model training. When the comprehensive weight is greater than the preset weight threshold, the corresponding candidate entity can be determined as the required static entity. Furthermore, the attribute types of entities are identified based on a hybrid framework, and the corresponding attribute values are dynamically obtained. This hybrid framework integrates rule-based initial screening, deep learning fine-tuning, and ontology verification. Specifically, rule-based initial screening involves building a rule base based on domain knowledge, such as matching entities or keywords in a specific format using regular expressions to initially extract candidate attributes. Deep learning enhancement involves inputting the rule-based initial screening results into a deep learning model, which optimizes the recognition effect through contextual understanding. Dynamic attribute acquisition involves dynamically updating the obtained attribute values by combining knowledge graphs or real-time data sources.
[0031] See Figure 3 Specifically, in the method described above, determining the target static entity that will undergo entity transition based on the attribute values corresponding to each static entity and a preset attribute conflict quantification algorithm includes: Step S301: Based on the current attribute value, benchmark value, and standard deviation of the static entity, obtain the attribute conflict value of the static entity, wherein the benchmark value is the weighted average of the current attribute value and the historical attribute average value. Step S302: If the attribute conflict value is greater than the preset attribute conflict threshold, then the corresponding static entity is determined to be the target static entity.
[0032] In this embodiment, when determining the target static entity, the attribute conflict value of the static entity is obtained based on the current attribute value, benchmark value, and standard deviation of the static entity. The attribute conflict quantification algorithm can be expressed as: ,in, This represents the baseline value at the current time i. This represents the attribute value at the current time i. This represents the standard deviation at time i. Preferably, this benchmark value is dynamically updated, specifically by calculating a weighted average of the current attribute value and the historical average using a recursive formula. ,in, This represents the baseline value at the previous time step (i-1) (a weighted average of the attribute value at the previous time step and the historical average). The attenuation factor is in the range (0, 1). When the value is close to 1, it indicates that the model is sensitive to the latest data and is suitable for quickly responding to changes in trends. A value close to 0 indicates that the model focuses more on long-term trends and reduces noise. Specific values can be obtained based on experience or model experiments.
[0033] Then, the obtained attribute conflict values are compared with the preset attribute conflict threshold. If the value is greater than the preset threshold, it indicates that a qualitative change has occurred due to a quantitative change, resulting in an entity transition. Therefore, the corresponding static entity is determined as the target static entity, and the first entity generated from it is obtained. The preset attribute conflict threshold can be set based on experience or obtained by comparing the experimental prediction results of the actual model with the actual data.
[0034] In one specific embodiment, the customer complaint work order is: User Z reports a scenario of weak cell signal and frequent network outages. The static entities involved include user Z, base station X, and cell Y. When the signal strength K of cell Y exceeds the threshold Q=1.5, a new entity will be generated: Weak signal area - cell Y(B).
[0035] See Figure 4 Specifically, in the method described above, performing probabilistic causal reasoning on the first entity to obtain the contradictory entity pair corresponding to the first entity and the second entity in the contradictory entity pair that contradicts the first entity includes: Step S401: According to a preset association rule algorithm, identify the pre-selected frequent 2-itemsets of the first entity, wherein the pre-selected frequent 2-itemsets includes the first entity and a first pre-selected entity; Step S402: Obtain the support and confidence information corresponding to each of the pre-selected frequent 2-itemsets, and determine the pre-selected frequent 2-itemsets with support greater than a preset support threshold as reference frequent 2-itemsets; Step S403: Construct a Bayesian network based on the reference 2-itemsets and obtain the conditional probability information corresponding to each reference 2-itemsets; Step S404: Entity conflict judgment is performed on each of the reference frequent 2-itemsets based on the support, the confidence information and the conditional probability information to obtain the entity conflict judgment result; Step S405: Determine that the reference frequent 2-itemset with conflict as determined by the conflict judgment result is the contradictory entity pair, and the first pre-selected entity in the contradictory entity pair is the second entity.
[0036] In this embodiment, the steps for obtaining contradictory entity pairs described above are illustrated. Specifically, after the attribute conflict value of the target static entity (A) exceeds a preset attribute conflict threshold, causing a qualitative change and generating a newly added first entity (B), frequent 2-itemsets of the first entity (B) are identified according to a preset association rule algorithm (Apriori). These frequent 2-itemsets include the first entity (B) and a first pre-selected entity (C1).
[0037] Then, the support and confidence information of frequent 2-itemsets are calculated. The support can be determined by the support function, which can be expressed as: The confidence information may include confidence in different directions, namely, the positive confidence along the first entity to the first pre-selected entity and the negative confidence along the first pre-selected entity to the first entity. Taking the positive confidence as an example, it is expressed as follows: Furthermore, based on the obtained support and the preset support threshold, frequent 2-itemsets with low support are cleaned to obtain pre-selected frequent 2-itemsets with support greater than the preset support threshold, which are then recorded as reference frequent 2-itemsets, thereby helping to reduce noise interference.
[0038] Furthermore, a Bayesian network is constructed based on reference 2-itemsets, nodes on the Bayesian network are determined based on each reference 2-itemset, and conditional probability information corresponding to each reference 2-itemset is obtained to represent the edge weights between nodes. By calculating the conditional probability information corresponding to each node, the probability distribution of the variable (i.e., C1) in a given parent node (i.e., B) state can be quantified.
[0039] Then, based on the support, confidence and conditional probability information obtained above, entity conflict judgment is performed on each reference frequent 2-item set, that is, whether the first entity and the first pre-selected entity in the reference frequent 2-item set are contradictory, and the entity conflict judgment result can be obtained; then, it can be determined that the reference frequent 2-item set with conflict judgment result is a contradictory entity pair corresponding to the first entity, and the first pre-selected entity in the contradictory entity pair is the second entity.
[0040] Furthermore, in the method described above, the step of performing entity conflict judgment on each of the reference frequent 2-itemsets based on the support, the confidence information, and the conditional probability information to obtain the entity conflict judgment result includes: If the absolute difference between the positive confidence in the confidence information corresponding to the reference frequent 2-itemset and the positive conditional probability in the conditional probability information is greater than a first preset threshold, then the conflict judgment result is determined to be a conflict, wherein the positive direction represents the direction starting from the first entity and ending at the first preselected entity. If the positive confidence level corresponding to the reference frequent 2-itemset is greater than the preset confidence threshold and the support level is less than the preset support threshold, then the conflict judgment result is determined to be that there is a conflict. If the confidence direction determined based on the positive confidence and the reverse confidence in the confidence information is opposite to the network causal direction based on the positive conditional probability and the reverse conditional probability in the conditional probability information, then the conflict judgment result is determined to be a conflict, wherein the reverse direction refers to the direction starting from the first preselected entity and ending at the first entity.
[0041] To facilitate understanding by those skilled in the art, this embodiment provides an example of the steps for determining entity conflicts. The determination can be based on whether there is a probability deviation, i.e., calculating the absolute difference between the positive confidence level and the positive conditional probability. The value is compared with a first preset threshold. If the absolute difference is greater than the first preset threshold, it can be determined that the confidence of the association rule deviates significantly from the conditional probability of the Bayesian network, indicating that the rule and the network structure are contradictory. The conflict judgment result is determined to be that there is a conflict, and the contradiction pair of the reference frequent 2-itemsets is established. The first preset threshold is a user-defined value or obtained through training.
[0042] Alternatively, we can determine whether there is a conflict between support and confidence. If the positive confidence is greater than the preset confidence threshold and the support is less than the preset support threshold, that is, high confidence and low support, it is a strong association according to the rules, but the actual probability of occurrence is extremely low. It may be a false association. Therefore, the conflict judgment result is determined to be a conflict. Refer to the contradiction pair of frequent 2-itemsets.
[0043] Alternatively, to determine whether there is a conflict in the causal relationships within the reference frequent 2-itemsets, compare the positive and negative confidence scores corresponding to the reference frequent 2-itemsets, and compare their corresponding positive and negative conditional probabilities. If the positive confidence score is greater than or equal to the negative confidence score, but the positive conditional probability in the Bayesian network is less than the negative conditional probability, or if the positive confidence score is less than the negative confidence score, but the positive conditional probability in the Bayesian network is greater than or equal to the negative conditional probability, then it is determined that the causal direction of the network is opposite to the direction of the rule confidence score, and the conflict judgment result is that there is a conflict, and the contradiction of the reference frequent 2-itemsets is valid.
[0044] See Figure 5Specifically, as described above, the process of performing fusion reasoning on the contradictory entity pairs to obtain a third entity includes: Step S501: Perform feature fusion and splicing on the contradictory entity pair to obtain spliced features; Step S502: Based on the efficient pointer network and the splicing features, generate multiple second pre-selected entities and their corresponding probability values; Step S503: Determine the second pre-selected entity corresponding to the highest value among the probability values as the third entity.
[0045] This implementation provides an example of how to obtain the third entity. After obtaining the contradictory entities, a reconciliation entity recognition model is constructed to acquire the third entity. First, the input contradictory entity pairs are fused and concatenated to obtain corresponding concatenated features. These concatenated features are then input into an Efficient Pointer Network (EPN) for inference. The EPN generates multiple second candidate entities and their corresponding probability values. The second pre-selected entity with the highest probability value is then identified as the third entity. For example, based on the contradictory entity pair (weak signal area (first entity B) ^ maintenance frequency < 2 times / month (second entity C), where "^" represents the contradiction operator), a priority temporary base station deployment (third entity D) is generated. By introducing the third entity D, the contradiction between the first entity B and the second entity C can be resolved.
[0046] See Figure 6 Specifically, in the method described above, generating the corresponding rule entity based on the relationship rules between the contradictory entity pair and the corresponding third entity includes: Step S601: Construct the relational path of the third entity corresponding to the contradictory entity; Step S602: Process the relationship path based on the unsupervised rule abstraction algorithm to obtain the rule entity.
[0047] In this implementation, an example is given of how to generate rule entities. After obtaining the third entity through fusion reasoning, the process of obtaining the third entity from contradictory entity pairs is path-based to obtain a relational path. This path-based approach can be represented as: ,in, Indicates the relational path. This indicates that the result is derived from the superposition of contradictions. Then, the relationship path obtained above is processed according to the unsupervised rule abstraction algorithm to generate new rule entities to carry the relationship rules between the first entity, the second entity, and the third entity.
[0048] See Figure 7Preferably, in the method described above, the step of processing the relation path based on the unsupervised rule abstraction algorithm to obtain the rule entity includes: Step S701: Based on high-dimensional mapping, map each node in the relationship path to a node embedding in a high-dimensional space; Step S702: Obtain the attention coefficients of adjacent nodes based on the node embedding; Step S703: Perform parallel calculations on the attention coefficients according to the multi-head attention mechanism to obtain the spliced or averaged node embeddings; Step S704: Integrate the node embeddings of the relationship path to generate a fixed-dimensional path embedding vector; Step S705: Perform density clustering on the path embedding vector to obtain the regular entity.
[0049] In this embodiment, the steps of processing the relation path based on the unsupervised rule abstraction algorithm to obtain the rule entity are illustrated. First, based on high-dimensional mapping, the nodes in the relation path are mapped to node embeddings in a high-dimensional space. Specifically, the input relation path is quantized, the nodes in the path are treated as graph nodes, and mapped to a high-dimensional space through a linear transformation.
[0050] Then, based on the node embeddings described above, the attention coefficients of adjacent nodes in the relationship path are calculated, which can be specifically expressed as: ,in, Indicates adjacent nodes and Attention coefficient Represents a node and nodes Feature splicing, Represents the weight matrix. This represents the feature vector corresponding to the node. This represents the feature vector corresponding to node j. Where, for the contradiction operator... This can be represented as follows: when calculating the weight of the weak signal region (first entity B) → base station maintenance lag (second entity C), It will be significantly higher than ordinary relationships (the intensity of conflict triggers attention focus).
[0051] Furthermore, the attention coefficients are calculated in parallel based on a multi-head attention mechanism to obtain concatenated or averaged node embeddings, thereby enhancing the expressive power for complex relationships. The parallel calculation based on the multi-head attention mechanism can be expressed as follows: ,in, This represents the node embedding after splicing or averaging. This represents the attention coefficient of the k-th independent group. This represents the weight matrix of the k-th attention head. This represents the total number of nodes adjacent to the node. This represents the activation function.
[0052] Then, the embeddings of all nodes along the entire relationship path are integrated to generate a vector with fixed dimensions to capture... The semantics, namely, the superposition of contradictions. The semantics of the result.
[0053] Finally, density clustering (DBSCAN) is performed on the embedding vectors of the above relationship paths to merge similar patterns and generate regular entities based on high-frequency relationship patterns. In a specific embodiment, signal optimization regular entity E: weak signal region (first entity B) ^ maintenance frequency < 2 times / month (second entity C) (high-frequency relationship pattern) Prioritize the deployment of temporary base stations (third entity D). The decision on whether to deploy temporary base stations can also be based on a comprehensive output rule entity that considers high-frequency mode and cost-effectiveness policy.
[0054] See Figure 8 Specifically, as described above, the consistency check performed on all the static entities, the first entity, the second entity, the third entity, and the rule entity to obtain the entity list includes: Step S801: Obtain the entity model confidence level corresponding to the target entity, wherein the target entity is one of the static entity, the first entity, the second entity, the third entity, and the rule entity; Step S802: Perform ontology consistency verification on each of the target entities to obtain logistic regression values; Step S803: If the exponential calculation result of the entity model confidence and the logistic regression value corresponding to the target entity is greater than or equal to the second preset threshold, then the target entity is added to the entity list.
[0055] In this embodiment, during consistency verification, the static entity, first entity, second entity, third entity, and rule entity obtained through the above entity identification are respectively taken as target entities, and the entity model confidence of the target entities is obtained. This entity model confidence is used to calculate the confidence distribution of the entity model's output results for a certain entity concept, which can be expressed as: This is a probability vector; then, ontology consistency verification is performed on the entity concept of the target entity, i.e. This ensures that the knowledge described by the ontology is consistent with logical rules and returns a logistic regression value. If the exponential calculation result of the entity model confidence score and logistic regression value corresponding to the target entity is greater than or equal to the second preset threshold, that is... Then the target entity is added to the entity list, wherein the second threshold The preferred value is 0.9, which can be adjusted based on the training results.
[0056] It should also be noted that the entity recognition method that integrates dialectical ontology in this application can dynamically, quickly and accurately identify static entities, potential entities (dynamic entities), and entity relationships.
[0057] The rule entities generated in this way can automatically generate execution strategies. For example, a signal optimization rule entity: weak signal area ^ maintenance frequency < 2 times / month (high-frequency relationship mode). Prioritize the deployment of temporary base stations. This rule entity can trigger the implementation of a strategy that prioritizes the deployment of temporary base stations, thereby improving customer satisfaction.
[0058] This model can improve the integrity and efficiency of databases, such as in database design and software development. It also promotes consistent data management and cross-system information sharing. Furthermore, it allows for the rapid construction of datasets to provide knowledge utilization, not only aiding in a better understanding of the knowledge structure in the field but also supporting applications such as information retrieval, data analysis, and intelligent question answering.
[0059] The various methods described above are based on embodiments of this application. Apparatus for implementing the above methods will now be provided.
[0060] See Figure 9 Another embodiment of this application also provides a control device, including: The static entity recognition module 901 is used to perform static entity recognition based on the input text through a large language model to obtain at least one static entity and the attribute values corresponding to each static entity. The dynamic entity recognition module 902 is used to determine the target static entity that will undergo entity transition and the first entity generated after the target static entity undergoes entity transition, based on the attribute values corresponding to each static entity and a preset attribute conflict quantification algorithm. The causal reasoning module 903 is used to perform probabilistic causal reasoning on the first entity to obtain the contradictory entity pair corresponding to the first entity and the second entity in the contradictory entity pair that contradicts the first entity. The fusion reasoning module 904 is used to perform fusion reasoning on the contradictory entity pair to obtain a third entity; The rule entity recognition module 905 is used to generate corresponding rule entities based on the relationship rules between the contradictory entity pairs and the corresponding third entities; The verification module 906 is used to perform consistency verification on all the static entities, the first entity, the second entity, the third entity, and the rule entity to obtain an entity list.
[0061] Specifically, in the apparatus described above, the static entity recognition module includes: The first processing unit is configured to perform contextual processing on the input text according to the large language model and generate a candidate entity set, wherein the candidate entity set includes at least one candidate entity. The second processing unit is used to obtain the comprehensive weight of each candidate entity based on the part-of-speech category of each candidate entity and its position information in the corresponding sentence. The third processing unit is used to determine the candidate entities whose comprehensive weight is greater than a preset weight threshold as static entities, and to dynamically obtain the attribute values corresponding to each static entity.
[0062] Specifically, in the apparatus described above, the dynamic entity recognition module includes: The fourth processing unit is used to obtain the attribute conflict value of the static entity based on the current attribute value, the benchmark value, and the standard deviation of the static entity, wherein the benchmark value is the weighted average of the current attribute value and the historical attribute average value. The fifth processing unit is used to determine the corresponding static entity as the target static entity if the attribute conflict value is greater than a preset attribute conflict threshold.
[0063] Specifically, in the apparatus described above, the causal reasoning module includes: The sixth processing unit is used to identify the pre-selected frequent 2-itemsets of the first entity according to a preset association rule algorithm. The pre-selected frequent 2-itemsets includes the first entity and a first pre-selected entity. The seventh processing unit is used to obtain the support and confidence information corresponding to each of the pre-selected frequent 2-itemsets, and to determine the pre-selected frequent 2-itemsets whose support is greater than a preset support threshold as reference frequent 2-itemsets; The eighth processing unit is used to construct a Bayesian network based on the reference 2-itemsets and obtain the conditional probability information corresponding to each reference 2-itemsets; The ninth processing unit is used to perform entity conflict judgment on each of the reference frequent 2-itemsets based on the support, the confidence information and the conditional probability information, and to obtain the entity conflict judgment result; The tenth processing unit is used to determine that the reference frequent 2-itemset with conflict as determined by the conflict judgment result is the contradictory entity pair, and the first pre-selected entity in the contradictory entity pair is the second entity.
[0064] Furthermore, in the apparatus described above, the ninth processing unit includes: If the absolute difference between the positive confidence in the confidence information corresponding to the reference frequent 2-itemset and the positive conditional probability in the conditional probability information is greater than a first preset threshold, then the conflict judgment result is determined to be a conflict, wherein the positive direction represents the direction starting from the first entity and ending at the first preselected entity. If the positive confidence level corresponding to the reference frequent 2-itemset is greater than the preset confidence threshold and the support level is less than the preset support threshold, then the conflict judgment result is determined to be that there is a conflict. If the confidence direction determined based on the positive confidence and the reverse confidence in the confidence information is opposite to the network causal direction based on the positive conditional probability and the reverse conditional probability in the conditional probability information, then the conflict judgment result is determined to be a conflict, wherein the reverse direction refers to the direction starting from the first preselected entity and ending at the first entity.
[0065] Specifically, in the method described above, the fusion inference module includes: The eleventh processing unit is used to perform feature fusion and splicing on the contradictory entity pairs to obtain spliced features; The twelfth processing unit is used to generate multiple second pre-selected entities and corresponding probability values based on the efficient pointer network and the splicing features; The thirteenth processing unit is used to determine that the second pre-selected entity corresponding to the highest value among the probability values is the third entity.
[0066] Specifically, in the apparatus described above, the rule entity recognition module includes: The fourteenth processing unit is used to construct the relationship path of the third entity corresponding to the contradictory entity; The fifteenth processing unit is used to process the relation path based on an unsupervised rule abstraction algorithm to obtain the rule entity.
[0067] Preferably, in the apparatus described above, the fifteenth processing unit includes: The first sub-processing unit is used to map each node in the relation path to a node embedding in a high-dimensional space based on high-dimensional mapping. The second sub-processing unit is used to obtain the attention coefficients of adjacent nodes based on the node embedding. The third sub-processing unit is used to perform parallel calculations on the attention coefficients according to the multi-head attention mechanism to obtain the spliced or averaged node embeddings. The fourth sub-processing unit is used to integrate the node embeddings of the relational path and generate a path embedding vector of fixed dimension. The fifth sub-processing unit is used to perform density clustering on the path embedding vector to obtain the rule entity.
[0068] Specifically, in the apparatus described above, the verification module includes: The sixteenth processing unit is used to obtain the entity model confidence level corresponding to the target entity, wherein the target entity is one of the static entity, the first entity, the second entity, the third entity, and the rule entity; The seventeenth processing unit is used to perform ontology consistency verification on each of the target entities and obtain logistic regression values. The eighteenth processing unit is used to add the target entity to the entity list if the exponential calculation result of the entity model confidence and the logistic regression value corresponding to the target entity is greater than or equal to a second preset threshold.
[0069] The apparatus embodiments of this application are apparatuses corresponding to the embodiments of the methods described above. All implementation means in the method embodiments described above are applicable to the apparatus embodiments and can achieve the same technical effects. The apparatus provided in this application embodiments can implement all the method steps implemented in the method embodiments described above and can achieve the same technical effects. Therefore, the parts and beneficial effects that are the same as those in the method embodiments in this embodiment will not be described in detail here.
[0070] See Figure 10 Another embodiment of this application provides an electronic device 100, including: a processor 1001, a memory 1002, and a program stored in the memory 1002 and executable on the processor 1001. When the program is executed by the processor 1001, it implements the steps of the entity recognition method of the fusion dialectical ontology as described above and can achieve the same technical effect. To avoid repetition, it will not be described again here.
[0071] Another embodiment of this application provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the steps of the entity recognition method for fusing dialectical ontology as described above, and achieves the same technical effect. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.
[0072] Another embodiment of this application provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the entity recognition method for fusing dialectical ontology as described above, and can achieve the same technical effect. To avoid repetition, it will not be described again here.
[0073] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0074] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.
[0075] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.
Claims
1. An entity recognition method integrating dialectical ontology, characterized in that, include: Based on the input text, static entity recognition is performed using a large language model to obtain at least one static entity and the attribute values corresponding to each static entity. Based on the attribute values corresponding to each static entity and the preset attribute conflict quantification algorithm, the target static entity that will undergo entity transition is determined, as well as the first entity generated after the target static entity undergoes entity transition. Perform probabilistic causal reasoning on the first entity to obtain the contradictory entity pair corresponding to the first entity and the second entity in the contradictory entity pair that contradicts the first entity. By performing fusion reasoning on the contradictory entity pairs, a third entity is obtained; Based on the relationship rules between the contradictory entity pairs and the corresponding third entity, generate corresponding rule entities; A consistency check is performed on all the static entities, the first entity, the second entity, the third entity, and the rule entity to obtain an entity list.
2. The method according to claim 1, characterized in that, The step of performing static entity recognition based on the input text using a large language model to obtain at least one static entity and the attribute values corresponding to each static entity includes: The input text is processed according to the large language model to generate a candidate entity set, which includes at least one candidate entity. Based on the part-of-speech category of each candidate entity and its position in the corresponding sentence, the comprehensive weight of each candidate entity is obtained. Candidate entities whose overall weight is greater than a preset weight threshold are identified as static entities, and attribute values corresponding to each static entity are dynamically obtained.
3. The method according to claim 1, characterized in that, The step of determining the target static entity that will undergo entity transition based on the attribute values corresponding to each static entity and a preset attribute conflict quantification algorithm includes: Based on the current attribute value, benchmark value, and standard deviation of the static entity, obtain the attribute conflict value of the static entity, wherein the benchmark value is the weighted average of the current attribute value and the historical attribute average value; If the attribute conflict value is greater than the preset attribute conflict threshold, then the corresponding static entity is determined to be the target static entity.
4. The method according to claim 1, characterized in that, The step of performing probabilistic causal reasoning on the first entity to obtain the contradictory entity pair corresponding to the first entity and the second entity in the contradictory entity pair that contradicts the first entity includes: According to a preset association rule algorithm, a pre-selected frequent 2-itemset of the first entity is identified, wherein the pre-selected frequent 2-itemset includes the first entity and a first pre-selected entity; Obtain the support and confidence information corresponding to each of the pre-selected frequent 2-itemsets, and determine the pre-selected frequent 2-itemsets whose support is greater than a preset support threshold as reference frequent 2-itemsets; A Bayesian network is constructed based on the reference 2-itemsets, and the conditional probability information corresponding to each reference 2-itemsets is obtained; Based on the support, the confidence information and the conditional probability information, entity conflict is determined for each of the reference frequent 2-itemsets to obtain the entity conflict determination result. The reference frequent 2-itemset that is determined to have a conflict is the contradictory entity pair, and the first pre-selected entity in the contradictory entity pair is the second entity.
5. The method according to claim 4, characterized in that, The step of performing entity conflict judgment on each of the reference frequent 2-itemsets based on the support, the confidence information, and the conditional probability information to obtain entity conflict judgment results includes: If the absolute difference between the positive confidence in the confidence information corresponding to the reference frequent 2-itemset and the positive conditional probability in the conditional probability information is greater than a first preset threshold, then the conflict judgment result is determined to be a conflict, wherein the positive direction represents the direction starting from the first entity and ending at the first preselected entity. If the positive confidence level corresponding to the reference frequent 2-itemset is greater than the preset confidence threshold and the support level is less than the preset support threshold, then the conflict judgment result is determined to be that there is a conflict. If the confidence direction determined based on the positive confidence and the reverse confidence in the confidence information is opposite to the network causal direction based on the positive conditional probability and the reverse conditional probability in the conditional probability information, then the conflict judgment result is determined to be a conflict, wherein the reverse direction refers to the direction starting from the first preselected entity and ending at the first entity.
6. The method according to claim 1, characterized in that, The process of fusing and reasoning through the contradictory entity pairs to obtain a third entity includes: For the contradictory entity pairs, feature fusion and splicing are performed to obtain spliced features; Based on the efficient pointer network and the splicing features, multiple second pre-selected entities and their corresponding probability values are generated. The second pre-selected entity corresponding to the highest value among the probability values is determined as the third entity.
7. The method according to claim 1, characterized in that, The step of generating corresponding rule entities based on the relationship rules between the contradictory entity pairs and the corresponding third entities includes: Construct the relationship path from the contradictory entity to the corresponding third entity; The relational path is processed using an unsupervised rule abstraction algorithm to obtain the rule entity.
8. The method according to claim 7, characterized in that, The process of the relation path based on the unsupervised rule abstraction algorithm to obtain the rule entity includes: Based on high-dimensional mapping, each node in the relational path is mapped to a node embedding in a high-dimensional space; Based on the node embedding, obtain the attention coefficients of adjacent nodes; The attention coefficients are calculated in parallel according to the multi-head attention mechanism to obtain the spliced or averaged node embeddings. Integrate the node embeddings of the aforementioned relationship paths to generate a fixed-dimensional path embedding vector; The path embedding vector is subjected to density clustering to obtain the regular entity.
9. The method according to claim 1, characterized in that, The consistency check is performed on all the static entities, the first entity, the second entity, the third entity, and the rule entity to obtain an entity list, including: Obtain the entity model confidence score corresponding to the target entity, wherein the target entity is one of the static entity, the first entity, the second entity, the third entity, and the rule entity; Perform ontology consistency verification on each of the target entities to obtain logistic regression values; If the result of the exponential calculation of the entity model confidence score and the logistic regression value corresponding to the target entity is greater than or equal to the second preset threshold, then the target entity is added to the entity list.
10. A control device, characterized in that, include: The static entity recognition module is used to perform static entity recognition based on the input text using a large language model, and obtain at least one static entity and the attribute values corresponding to each static entity. The dynamic entity recognition module is used to determine the target static entity that will undergo entity transition and the first entity generated after the target static entity undergoes entity transition, based on the attribute values corresponding to each static entity and a preset attribute conflict quantification algorithm. The causal reasoning module is used to perform probabilistic causal reasoning on the first entity to obtain the contradictory entity pair corresponding to the first entity and the second entity in the contradictory entity pair that contradicts the first entity. The fusion reasoning module is used to perform fusion reasoning on the contradictory entity pairs to obtain a third entity; The rule entity recognition module is used to generate corresponding rule entities based on the relationship rules between the contradictory entity pairs and the corresponding third entities; The verification module is used to perform consistency verification on all the static entities, the first entity, the second entity, the third entity, and the rule entity to obtain an entity list.
11. An electronic device, characterized in that, include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the entity recognition method of the fused dialectical ontology as described in any one of claims 1 to 9.
12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the entity recognition method for fusion dialectical ontology as described in any one of claims 1 to 9.
13. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the steps of the entity recognition method of the fused dialectical ontology as described in any one of claims 1 to 9.