Text semantic conflict detection method and system for constructing context association based on large model

Abstract

The invention relates to a text semantic conflict detection method for constructing context association based on a large model. The method comprises the following steps: S1, segmenting an original text by using a segmentation function; s2, performing triple information extraction on each segmented text by using a large model to generate a data set of an entity corresponding relationship; s3, extracting data groups of all entity pairs and relationships, and performing confidence evaluation; s4, aggregating all triads, constructing a affair graph, and pointing out a contradictory relationship with relatively low confidence in the affair graph; and S5, for the found contradictory relationship, performing semantic conflict repair in the text in combination with the large model. According to the method, the accuracy of information generation of the large language model is remarkably improved in a context association mode.

Description

Technical Field

[0001] The present invention belongs to the technical field of semantic error correction, and in particular, relates to a text semantic conflict detection method and system based on building context association based on a large model. Background Art

[0002] Traditional neural network-based language model semantic error correction methods are easily affected by local high-frequency or low-frequency sequences, and the results are not stable enough; when it is necessary to balance the accuracy and recall rates, it is not easy to control them through thresholds; they can be well applied to spelling checking tasks, but their support for grammatical error correction tasks with varying sentence lengths is relatively weak.

[0003] In the semantic error correction method based on large models, the large model is not capable of correcting spelling errors and is prone to omissions and incorrect corrections. In addition, some grammatical errors can be detected but corrected incorrectly.

[0004] In the semantic error correction method based on knowledge graph, the labor cost of knowledge graph construction is high, and the values ​​of nodes and edges have no industry-wide unified standards and are completely defined by users. At the same time, due to the lack of unified standards, multi-source data fusion is difficult. In addition, knowledge graphs have difficulty in distinguishing concept nodes from object nodes, and it is impossible to uniformly define node and edge labels. Summary of the invention

[0005] (I) Purpose of the invention

[0006] In order to overcome the above shortcomings, the purpose of the present invention is to provide a text semantic conflict detection method and system based on a large model to build context association, so as to solve the above technical problems.

[0007] (II) Technical solution

[0008] To achieve the above objectives, the technical solutions provided by this application are as follows:

[0009] A text semantic conflict detection method based on building context association based on a large model comprises the following steps:

[0010] S1: Split the original text using a segmentation function to prevent the text input from being too long and exceeding the token limit of the large model, or to prevent the large model from extracting triples with a reduced accuracy due to too long text;

[0011] S2: For each segmented text, use the big model to extract triple information and generate a data set of entity correspondence relationships. The data set is: <entity 1, relationship 1, entity 2>, <entity 1, relationship 1, entity 2>, <entity 1, relationship 2, entity 2>, <entity 1, relationship 3, entity 2>, <entity 1, relationship 4, entity 3>, …, <entity x, relationship y, entity z>;

[0012] S3: Extract all entity pairs and relationships from the data set. For the same entity pair, the entity relationships corresponding to different text contents may be the same or different. The same entity relationships are verified, and the confidence increases accordingly; different entity relationships are verified, and the confidence decreases accordingly. The confidence score calculation method for <entity i, relationship k, entity j> is:

[0013] Score <实体i，关系k，实体j> =∑ 关系l＝关系k α·1 {<实体i，关系1，实体j>} +∑ 关系l≈关系k β·1 {<实体i，关系1，实体j>} +∑ 关系l≠关系k γ·1 {<实体i，关系1，实体j>} ,

[0014] Among them, α is the scoring coefficient corresponding to the same entity relationship, β is the scoring coefficient corresponding to the similar entity relationship, and γ is the scoring coefficient corresponding to the inconsistent entity relationship;

[0015] S4: Aggregate all triples, construct a truth map, and point out the contradictory relationships with low confidence in the truth map;

[0016] S5: Based on the found contradictory relationships, the big model is used to repair semantic conflicts in the text.

[0017] Preferably, the segmentation function includes the following segmentation methods:

[0018] Based on sentences, the text is segmented according to sentence boundaries, and each sentence becomes a segmented text segment;

[0019] Based on paragraphs, segmentation is performed according to paragraph boundaries, and each paragraph is regarded as a text segment;

[0020] Based on the length, a maximum length limit is set. When the text exceeds this length, it is segmented until each segment does not exceed this length.

[0021] Based on semantics, syntactic analysis or semantic segmentation is used to identify natural breakpoints in the text and then segment it.

[0022] Preferably, when the large model in S2 extracts a data set of entity pairs and relationships, it first converts the input text into a format that the model can understand, usually including word segmentation and adding special tags. Then the model processes the tokens in the text one by one, predicting whether each token is part of a named entity and the type of entity. Finally, after identifying the entities, the model further identifies the relationships between these entities, including the following methods:

[0023] Use dependency parsing to determine syntactic relationships between entities;

[0024] Use span classification to identify the relationship between two entities;

[0025] Use graph neural networks or other models to analyze the global relationships of entities in text.

[0026] Preferably, the specific steps in S4 include:

[0027] Step 1: Initialize an empty graph containing entity nodes and relationship edges, add each triplet as an edge to the graph, and the weight of the edge is the confidence score of the triplet;

[0028] Step 2: Ensure that all entities in the graph are standardized, that is, instances with different names but referring to the same entity are merged into a single node;

[0029] Step 3: If there are multiple relationships for the same entity pair, aggregate the confidence scores of these relationships by taking the average, maximum value, or other logic;

[0030] Step 4: Traverse all edges in the graph and look for contradictory or inconsistent relationships;

[0031] Step 5: Set a confidence threshold. Only relationships above this threshold are considered reliable. Relationships below the threshold require further review or are marked as potential contradictions.

[0032] Step 6: Use visualization tools to display the event map to facilitate analysis and understanding of the relationship between entities;

[0033] Step 7: Users can modify and edit the generated causal map to enhance user participation and interactivity. Users can evaluate the quality of answers based on the displayed graph logic information and results.

[0034] Preferably, the specific steps in S5 include:

[0035] A1 analyzes the causal map and identifies the relationships with low confidence or logical contradictions from the constructed causal map;

[0036] A2 extracts relevant context, surrounding the entities of contradictory relationships, and extracts relevant contextual information from the original text, including sentences, paragraphs, or entire documents;

[0037] A3 uses large models to conduct in-depth analysis to understand the relationships between entities and their role in the text;

[0038] A4 generates candidate solutions. Based on the understanding of the big model, it generates possible solutions or relationship modification suggestions, including changing the type of entity, adjusting the relationship description, or proposing a new explanation.

[0039] A5 evaluates the solution, using a large model to evaluate each candidate solution and predict its confidence;

[0040] A6 Manual review and selection: For solutions with low confidence, manual review is required to select the final solution;

[0041] A7 updates the cause-and-effect map, applies the selected solution to the cause-and-effect map, updates or deletes conflicting relationships, and ensures the consistency and accuracy of the map.

[0042] A text semantic conflict detection system based on building context association based on a large model, comprising:

[0043] The text processing module is responsible for receiving and preprocessing the input text data, including word segmentation, stop word removal, and part-of-speech tagging;

[0044] The text segmentation module segments text according to sentences, paragraphs, length or semantics to generate text fragments suitable for model processing;

[0045] The entity and relationship extraction module uses a large model to process the segmented text fragments, extract entities and their relationships, and generate triple data;

[0046] The confidence scoring module calculates the confidence score of each entity-pair relationship based on the extracted triple data;

[0047] The event graph construction module aggregates the extracted triples to construct an event graph representing entities and relationships, and performs entity standardization and relationship aggregation;

[0048] The contradictory relationship identification and visualization module identifies low-confidence or logically contradictory relationships in the logic map and displays the map through visualization tools;

[0049] The conflict repair module uses a large model to conduct in-depth analysis of identified conflicting relationships, generate and evaluate candidate solutions, including a manual review process;

[0050] The user interaction module provides a user interface that allows users to modify, edit and review the graph, thus enhancing user participation;

[0051] The knowledge base and storage module stores the reasoning graph, extracted triples, solutions and other data for internal use and external query.

[0052] Beneficial effects:

[0053] 1. The present invention significantly improves the accuracy of information generated by large language models through context association. Users can rely on this system to generate more accurate and reliable text content, whether it is academic research, business reports or daily communication, to ensure the correctness of information transmission, thereby enhancing trust and professionalism.

[0054] 2. The present invention generates text that conforms to semantic logic by combining a large model with structured data. This method not only improves the logic of the text, but also ensures that the generated text is consistent with the existing knowledge structure. In this way, the system can display and repair the text of the user's semantic contradictions, helping users avoid potential misunderstandings or confusion. This is especially important in documents that require a high degree of accuracy, such as legal documents and technical manuals, to ensure the rigor and authority of the text.

[0055] 3. The present invention can also display the contradictions and error probabilities extracted from the user's text. The intuitive probability display allows users to quickly identify uncertainties and potential problems in the text, which is more user-friendly and improves the interpretability of the generated results. Users can not only see the problems in the text, but also understand the probability of these problems, so as to make more informed decisions. This transparent information display enhances the user's trust in the system and also improves the user's satisfaction with the generated text. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solutions and advantages of the present invention more clear, the following is a detailed description of the present invention in conjunction with the specific implementation methods and with reference to the attached drawings. Figure 1 , the present invention is further described in detail. It should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present invention. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concept of the present invention.

[0058] A text semantic conflict detection method based on building context association based on a large model, characterized by comprising the following steps:

[0059] S1: Split the original text using a segmentation function to prevent the text input from being too long and exceeding the token limit of the large model, or to prevent the large model from extracting triples with a reduced accuracy due to too long text;

[0060] S2: For each segmented text, use the big model to extract triple information and generate a data set of entity correspondence relationships. The data set is: <entity 1, relationship 1, entity 2>, <entity 1, relationship 1, entity 2>, <entity 1, relationship 2, entity 2>, <entity 1, relationship 3, entity 2>, <entity 1, relationship 4, entity 3>, …, <entity x, relationship y, entity z>;

[0061] S3: Extract all entity pairs and relationships from the data set. For the same entity pair, the entity relationships corresponding to different text contents may be the same or different. The same entity relationships are verified, and the confidence increases accordingly; different entity relationships are verified, and the confidence decreases accordingly. The confidence score calculation method for <entity i, relationship k, entity j> is:

[0062] Score <实体i，关系k，实体j> =∑ 关系l＝关系k α·1 {<实体i，关系1，实体j>} +∑ 关系l≈关系k β·1 {<实体i，关系l，实体j>} +∑ 关系l≠关系k γ·1 {<实体i，关系1，实体j>} ,

[0063] Among them, α is the scoring coefficient corresponding to the same entity relationship, β is the scoring coefficient corresponding to the similar entity relationship, and γ is the scoring coefficient corresponding to the inconsistent entity relationship;

[0064] S4: Aggregate all triples, construct a truth map, and point out the contradictory relationships with low confidence in the truth map;

[0065] S5: Based on the found contradictory relationships, the big model is used to repair semantic conflicts in the text.

[0066] The segmentation function includes the following segmentation methods:

[0067] Based on sentences, the text is segmented according to sentence boundaries, and each sentence becomes a segmented text segment;

[0068] Based on paragraphs, segmentation is performed according to paragraph boundaries, and each paragraph is regarded as a text segment;

[0069] Based on the length, a maximum length limit is set. When the text exceeds this length, it is segmented until each segment does not exceed this length.

[0070] Based on semantics, syntactic analysis or semantic segmentation is used to identify natural breakpoints in the text and then segment it.

[0071] When the large model in S2 extracts data sets of entity pairs and relationships, it first converts the input text into a format that the model can understand, usually including word segmentation and adding special tags. Then the model processes the tokens in the text one by one, predicting whether each token is part of a named entity and the type of entity. Finally, after identifying the entities, the model further identifies the relationships between these entities, including the following methods:

[0072] Use dependency parsing to determine syntactic relationships between entities;

[0073] Use span classification to identify the relationship between two entities;

[0074] Use graph neural networks or other models to analyze the global relationships of entities in text.

[0075] The specific steps in S4 include:

[0076] Step 1: Initialize an empty graph containing entity nodes and relationship edges, add each triplet as an edge to the graph, and the weight of the edge is the confidence score of the triplet;

[0077] Step 2: Ensure that all entities in the graph are standardized, that is, instances with different names but referring to the same entity are merged into a single node;

[0078] Step 3: If there are multiple relationships for the same entity pair, aggregate the confidence scores of these relationships by taking the average, maximum value, or other logic;

[0079] Step 4: Traverse all edges in the graph and look for contradictory or inconsistent relationships;

[0080] Step 5: Set a confidence threshold. Only relationships above this threshold are considered reliable. Relationships below the threshold require further review or are marked as potential contradictions.

[0081] Step 6: Use visualization tools to display the event map to facilitate analysis and understanding of the relationship between entities;

[0082] Step 7: Users can modify and edit the generated causal map to enhance user participation and interactivity. Users can evaluate the quality of answers based on the displayed graph logic information and results.

[0083] The specific steps in S5 include:

[0084] A1 analyzes the causal map and identifies the relationships with low confidence or logical contradictions from the constructed causal map;

[0085] A2 extracts relevant context, surrounding the entities of contradictory relationships, and extracts relevant contextual information from the original text, including sentences, paragraphs, or entire documents;

[0086] A3 uses large models to conduct in-depth analysis to understand the relationships between entities and their role in the text;

[0087] A4 generates candidate solutions. Based on the understanding of the big model, it generates possible solutions or relationship modification suggestions, including changing the type of entity, adjusting the relationship description, or proposing a new explanation.

[0088] A5 evaluates the solution, using a large model to evaluate each candidate solution and predict its confidence;

[0089] A6 Manual review and selection: For solutions with low confidence, manual review is required to select the final solution;

[0090] A7 updates the cause-and-effect map, applies the selected solution to the cause-and-effect map, updates or deletes conflicting relationships, and ensures the consistency and accuracy of the map.

[0091] A text semantic conflict detection system based on building context association based on a large model, comprising:

[0092] The text processing module is responsible for receiving and preprocessing the input text data, including word segmentation, stop word removal, and part-of-speech tagging;

[0093] The text segmentation module segments text according to sentences, paragraphs, length or semantics to generate text fragments suitable for model processing;

[0094] The entity and relationship extraction module uses a large model to process the segmented text fragments, extract entities and their relationships, and generate triple data;

[0095] The confidence scoring module calculates the confidence score of each entity-pair relationship based on the extracted triple data;

[0096] The event graph construction module aggregates the extracted triples to construct an event graph representing entities and relationships, and performs entity standardization and relationship aggregation;

[0097] The contradictory relationship identification and visualization module identifies low-confidence or logically contradictory relationships in the logic map and displays the map through visualization tools;

[0098] The conflict repair module uses a large model to conduct in-depth analysis of identified conflicting relationships, generate and evaluate candidate solutions, including a manual review process;

[0099] The user interaction module provides a user interface that allows users to modify, edit and review the graph, thus enhancing user participation;

[0100] The knowledge base and storage module stores the reasoning graph, extracted triples, solutions and other data for internal use and external query.

[0101] The present invention uses a large model to summarize the key points of the information in the text, and then extracts the triples <entity 1, relationship, entity 2> for the summarized key information to construct a matter map. At the same time, during the extraction, the confidence is scored, and the scoring can be determined by the co-occurrence probability of the triples in the large model. If the probability of the same triple appearing is large or the repetition frequency is high, the confidence of the triple is considered to be high. If there are contradictory elements in the triples, such as two elements in two triples are the same, and the last triple is inconsistent, the accuracy of the two triples is reduced. Finally, the large model can be used to extract the relevant map and score and display the probability of the map edge. The edge with a smaller probability of the map edge is considered to have a contradiction in the content and needs to be displayed to the customer.

[0102] The present invention can be applied to the task of large-model semantic conflict mining combined with a large language model, which can significantly solve the problem of large-model semantic hallucination and improve the accuracy of text generation.

[0103] The present invention can be applied to the scenario of analyzing and mining semantic contradictions in a single text, discovering the semantic contradictions and ambiguities in the text context, and displaying modification suggestions and judgment logic to users. It can improve the efficiency of user text writing and correction. It can also optimize the semantic contradictions in large model text generation and improve the large model output experience.

[0104] The present invention can be applied to scenarios where multiple texts verify each other, discover semantic contradictions in multiple texts, and solve the knowledge confidence problem in non-homologous texts. At the same time, the expansion can analyze and mine the accuracy of each data source and optimize the subsequent RAG generation.

[0105] The present invention significantly improves the accuracy of information generated by large language models through context association. Users can rely on this system to generate more accurate and reliable text content, whether it is academic research, business reports or daily communication, to ensure the correctness of information transmission, thereby enhancing trust and professionalism.

[0106] The present invention generates text that conforms to semantic logic by combining a large model with structured data. This method not only improves the logic of the text, but also ensures that the generated text is consistent with the existing knowledge structure. In this way, the system can display and repair the text of the user's semantic contradictions, helping the user to avoid potential misunderstandings or confusion. This is especially important in documents that require a high degree of accuracy, such as legal documents and technical manuals, to ensure the rigor and authority of the text.

[0107] The present invention can also display the contradictions and error probabilities extracted from the user's text. The intuitive probability display allows users to quickly identify uncertainties and potential problems in the text, which is more user-friendly and improves the interpretability of the generated results. Users can not only see the problems in the text, but also understand the probability of these problems, so as to make more informed decisions. This transparent information display enhances users' trust in the system and also improves users' satisfaction with the generated text.

[0108] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0109] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A text semantic conflict detection method based on building context association based on a large model, characterized in that: The following steps are involved: S1: Split the original text using a segmentation function to prevent the text input from being too long and exceeding the token limit of the large model, or to prevent the large model from extracting triples with a reduced accuracy due to too long text; S2: For each segmented text, use the big model to extract triple information and generate a data set of entity correspondence relationships. The data set is: <entity 1, relationship 1, entity 2>, <entity 1, relationship 1, entity 2>, <entity 1, relationship 2, entity 2>, <entity 1, relationship 3, entity 2>, <entity 1, relationship 4, entity 3>, …, <entity x, relationship y, entity z>; S3: Extract all entity pairs and relationships from the data set. For the same entity pair, the entity relationships corresponding to different text contents may be the same or different. The same entity relationships are verified, and the confidence increases accordingly; different entity relationships are verified, and the confidence decreases accordingly. The confidence score calculation method for <entity i, relationship k, entity j> is: Score <实体i，关系k，实体j> =∑ 关系l＝关系k a·1 {<实体i，关系1，实体j>} +∑ 关系l≈关系k b·1 {<实体i，关系1，实体j>} +∑ 关系l≠关系k c·1 {<实体i，关系1，实体j>} , Among them, α is the scoring coefficient corresponding to the same entity relationship, β is the scoring coefficient corresponding to the similar entity relationship, and γ is the scoring coefficient corresponding to the inconsistent entity relationship; S4: Aggregate all triples, construct a truth map, and point out the contradictory relationships with low confidence in the truth map; S5: Based on the found contradictory relationships, the big model is used to repair semantic conflicts in the text.

2. According to claim 1, a text semantic conflict detection method based on large model construction context association is characterized in that: The segmentation function includes the following segmentation methods: Based on sentences, the text is segmented according to sentence boundaries, and each sentence becomes a segmented text segment; Based on paragraphs, segmentation is performed according to paragraph boundaries, and each paragraph is regarded as a text segment; Based on the length, a maximum length limit is set. When the text exceeds this length, it is segmented until each segment does not exceed this length. Based on semantics, syntactic analysis or semantic segmentation is used to identify natural breakpoints in the text and then segment it.

3. According to claim 1, a text semantic conflict detection method based on building context association based on a large model is characterized in that: When the large model in S2 extracts data sets of entity pairs and relationships, it first converts the input text into a format that the model can understand, usually including word segmentation and adding special tags. Then the model processes the tokens in the text one by one, predicting whether each token is part of a named entity and the type of entity. Finally, after identifying the entities, the model further identifies the relationships between these entities, including the following methods: Use dependency parsing to determine syntactic relationships between entities; Use span classification to identify the relationship between two entities; Use graph neural networks or other models to analyze the global relationships of entities in text.

4. The text semantic conflict detection method based on building context association based on a large model according to claim 1 is characterized in that: The specific steps in S4 include: Step 1: Initialize an empty graph containing entity nodes and relationship edges, add each triplet as an edge to the graph, and the weight of the edge is the confidence score of the triplet; Step 2: Ensure that all entities in the graph are standardized, that is, instances with different names but referring to the same entity are merged into a single node; Step 3: If there are multiple relationships for the same entity pair, aggregate the confidence scores of these relationships by taking the average, maximum value, or other logic; Step 4: Traverse all edges in the graph and look for contradictory or inconsistent relationships; Step 5: Set a confidence threshold. Only relationships above this threshold are considered reliable. Relationships below the threshold require further review or are marked as potential contradictions. Step 6: Use visualization tools to display the event map to facilitate analysis and understanding of the relationship between entities; Step 7: Users can modify and edit the generated causal map to enhance user participation and interactivity. Users can evaluate the quality of answers based on the displayed graph logic information and results.

5. The text semantic conflict detection method based on large model construction context association according to claim 1 is characterized in that: The specific steps in S5 include: A1 analyzes the causal map and identifies the relationships with low confidence or logical contradictions from the constructed causal map; A2 extracts relevant context, surrounding the entities of contradictory relationships, and extracts relevant contextual information from the original text, including sentences, paragraphs, or entire documents; A3 uses large models to conduct in-depth analysis to understand the relationships between entities and their role in the text; A4 generates candidate solutions. Based on the understanding of the big model, it generates possible solutions or relationship modification suggestions, including changing the type of entity, adjusting the relationship description, or proposing a new explanation. A5 evaluates the solution, using a large model to evaluate each candidate solution and predict its confidence; A6 Manual review and selection: For solutions with low confidence, manual review is required to select the final solution; A7 updates the cause-and-effect map, applies the selected solution to the cause-and-effect map, updates or deletes conflicting relationships, and ensures the consistency and accuracy of the map.

6. A text semantic conflict detection system based on a large model to build context association, characterized in that: include: The text processing module is responsible for receiving and preprocessing the input text data, including word segmentation, stop word removal, and part-of-speech tagging; The text segmentation module segments text according to sentences, paragraphs, length or semantics to generate text fragments suitable for model processing; The entity and relationship extraction module uses a large model to process the segmented text fragments, extract entities and their relationships, and generate triple data; The confidence scoring module calculates the confidence score of each entity-pair relationship based on the extracted triple data; The event graph construction module aggregates the extracted triples to construct an event graph representing entities and relationships, and performs entity standardization and relationship aggregation; The contradictory relationship identification and visualization module identifies low-confidence or logically contradictory relationships in the logic map and displays the map through visualization tools; The conflict repair module uses a large model to conduct in-depth analysis of identified conflicting relationships, generate and evaluate candidate solutions, including a manual review process; The user interaction module provides a user interface that allows users to modify, edit and review the graph, thus enhancing user participation; The knowledge base and storage module stores the reasoning graph, extracted triples, solutions and other data for internal use and external query.

Images