An artificial intelligence-based knowledge graph question and answer interaction method and system

Abstract

This invention discloses an artificial intelligence-based knowledge graph question-answering interaction method and system, belonging to the field of knowledge graph technology. It includes four steps: temporal knowledge validity screening, reasoning path quality assessment, causal attribute consistency verification, and reliable result output and closed-loop optimization. After receiving a user's natural language query, the system first extracts the query time interval, filters valid triples matching the time dimension in the knowledge graph, calculates the path time matching degree and candidate answer attribute matching degree based on the valid triples, and then calculates the credibility after fusion before outputting the qualified answer. This invention integrates multiple technologies, using dual verification to ensure the accuracy of the answer in terms of temporality and logic, reducing errors. The system continuously learns and improves based on application feedback, enhancing the accuracy of the knowledge graph and the quality of question answering. Furthermore, the data interaction between modules is standardized and logically clear, enabling flexible responses to various queries and providing users with traceable, high-quality answers to meet diverse needs.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph technology, and in particular to a knowledge graph question-answering interaction method and system based on artificial intelligence. Background Technology

[0002] Knowledge graphs, as an important means of knowledge representation and organization, can store and manage entities and relationships between entities in the real world in an intuitive and efficient form due to their structured characteristics. This provides a solid knowledge foundation for many applications such as intelligent question answering and information retrieval. With the vigorous development of artificial intelligence technology, especially the continuous breakthroughs in natural language processing technology, question answering interaction systems based on knowledge graphs have become a hot topic in research and practice. The aim is to achieve natural, smooth and accurate information interaction between users and knowledge graphs, and meet users' needs for rapid acquisition and accurate understanding of knowledge.

[0003] However, current knowledge graph-based question-answering interaction technologies still have significant shortcomings in practical applications. Traditional methods struggle to quickly and effectively filter the time-sensitive validity of knowledge during the reasoning process, resulting in a large amount of invalid knowledge participating in reasoning calculations. This not only wastes computing resources but may also cause the reasoning results to become disconnected from reality, affecting the accuracy and reliability of question-answering. In terms of reasoning path selection, existing technologies lack an evaluation mechanism for the timeliness of reasoning paths. The knowledge covered by different reasoning paths varies in terms of time dimension, making it difficult to select high-quality reasoning paths based on the degree of matching between the knowledge in the path and the query time interval. This leads to a significant reduction in the credibility of reasoning results, making it difficult to meet the increasingly diverse needs of users. Summary of the Invention

[0004] To address the shortcomings mentioned in the background technology, this invention proposes a knowledge graph question-answering interaction method and system based on artificial intelligence.

[0005] The technical solution mainly consists of: an artificial intelligence-based knowledge graph question-answering interaction method, including the following steps:

[0006] Time-series knowledge validity filtering: After receiving a user query request, extract the time interval range corresponding to the query, traverse all triples associated with the query entity in the knowledge graph, compare the coverage relationship between the valid time range of the triples and the query time interval one by one, filter out the set of valid triples that match the time dimension, and filter out invalid knowledge that exceeds the query time range.

[0007] Inference path quality assessment: Based on the effective set of triples, multi-hop inference is performed to generate candidate inference paths. The overlap length ratio between all triples in each path and the query time interval is calculated to obtain the time matching degree of the path. The higher the matching degree, the stronger the timeliness of the path knowledge.

[0008] Causal attribute consistency verification: The candidate answers generated by reasoning are reverse parsed to obtain structured triples. The parsed relational attributes are matched with the pre-stored relational attributes in the knowledge graph, and the proportion of the number of items with consistent attributes to the total number of verified attributes is counted to obtain the attribute matching degree.

[0009] Credible result output and closed-loop optimization: The credibility score of the answer is calculated by multiplying the path time matching degree and the attribute matching degree. The answer is judged to meet the credibility requirements based on the score. For answers that do not meet the requirements, a re-reasoning process is triggered. At the same time, the triples corresponding to the low scores and the selected invalid triples are marked as pending review status for the content iteration and update of the knowledge graph.

[0010] Preferably, the time-series knowledge validity screening step specifically includes:

[0011] Each triple in the knowledge graph is pre-labeled with valid start and end time attributes. After receiving user queries, natural language processing technology is used to identify the time constraints in the query. Explicit time expressions are directly converted into standardized time intervals, and implicit time expressions are mapped to absolute time intervals by combining associated entity attributes and the current system time. The validity time range of each triple is checked to see if it intersects with the query time interval. Triples with intersections are determined to be valid and enter the inference path quality assessment step. Invalid triples without intersections are simultaneously marked as pending review and included in the knowledge graph iteration update process.

[0012] Preferably, the inference path quality assessment step specifically includes:

[0013] Based on the set of valid triples, path search is performed to generate all candidate inference paths. For each candidate path, the total number of triples contained in the path is counted, and the overlap length between each valid triple and the query time interval is calculated. The sum of all overlap lengths is then multiplied by the product of the total length of the query time interval and the total number of triples to obtain the time matching degree of the path. This ratio is used to reflect the degree of time dimension matching of the entire inference path. The higher the ratio, the more triples in the path completely cover the query time interval, and the stronger the overall timeliness. When there are multiple path conflicts, the path with the highest time matching degree is selected as the basis for inference.

[0014] Preferably, the causal attribute consistency verification step specifically includes:

[0015] Each type of relation in the knowledge graph is pre-labeled with three types of causal attribute tags: directionality, symmetry, and transitivity. After generating candidate answers through reasoning, the natural language answers are converted into structured triples using semantic parsing technology. The relation attributes corresponding to the answer triples are extracted and compared with the pre-stored attributes of the relation in the knowledge graph. The number of matching attributes is counted, and the ratio of the number of matching attributes to the total number of verified attributes is used as the attribute matching degree. Answers whose matching degree does not reach the preset attribute verification threshold are judged to have logical errors.

[0016] Preferably, the specific steps for the reliable result output and closed-loop optimization are as follows:

[0017] The comprehensive credibility score of the answer is calculated by multiplying the time matching degree and attribute matching degree of the reasoning path. Answers with a score that reaches the preset credibility threshold are simultaneously marked with the effective time range of the corresponding knowledge and the complete reasoning path and then output to the user. Answers with a score that does not reach the preset credibility threshold are automatically triggered to re-infer the process, and the triples involved in the reasoning path and the filtered invalid triples are marked as pending review.

[0018] The present invention also provides an artificial intelligence-based knowledge graph question-answering interaction system, including a time-aware knowledge graph construction module, an implicit time constraint parsing module, a time-aware multi-hop reasoning module, a causal-oriented anti-illusion verification module, and a reliable result output module;

[0019] The time-aware knowledge graph construction module is configured to store triple data with valid time attributes and causal attribute labels, receive low-quality triple label information from the causal-guided anti-illusion verification module and invalid triple label information from the time-aware multi-hop reasoning module, update the attribute labels and valid time range of the knowledge graph, and provide structured knowledge support for other modules.

[0020] The implicit time constraint parsing module is configured to receive user natural language queries, identify time constraint expressions in the queries, convert explicit time into standardized time intervals, map implicit time expressions into absolute time intervals, and output to the time-aware multi-hop inference module.

[0021] The time-aware multi-hop reasoning module is configured to filter the set of valid triples based on the parsed query time interval, entity and intent information, generate candidate reasoning paths and calculate the path time matching degree, select the optimal path to generate candidate answers and output them to the causal-oriented anti-illusion verification module, and at the same time mark the filtered invalid triples as pending review status and feed them back to the time-aware knowledge graph construction module.

[0022] The causal-guided anti-illusion verification module is configured to perform reverse parsing on candidate answers, verify the consistency between the answer relationship attributes and the knowledge graph attributes, calculate the attribute matching degree, calculate the credibility score by multiplying and fusing the time matching degree, feed back low-scoring answers to the time-aware multi-hop reasoning module to trigger re-reasoning, and output high-scoring answers to the credibility result output module.

[0023] The reliable result output module is configured to add knowledge validity time range and reasoning path tracing information to the verified answer, output it to the user in a formatted manner, and store all intermediate data in the reasoning process for subsequent optimization.

[0024] Preferably, the time-aware knowledge graph construction module includes a triplet storage unit, a time attribute annotation unit, a causal attribute annotation unit, and a knowledge update unit;

[0025] The triplet storage unit is configured to store basic triplet data consisting of head entity, relation, and tail entity. The time attribute labeling unit labels each triplet with valid start time and valid end time tags.

[0026] The causal attribute labeling unit is configured to label each type of relationship with three types of causal attribute tags: directionality, symmetry, and transitivity.

[0027] The knowledge update unit is configured to receive the triplet information to be verified from the causal-guided anti-illusion verification module and the invalid triplet information to be verified from the time-aware multi-hop reasoning module, and update the attribute tags and valid time range of the triplet after manual verification.

[0028] Preferably, the implicit time constraint parsing module includes an entity recognition unit, an intent classification unit, an explicit time recognition unit, and an implicit time mapping unit;

[0029] The entity recognition unit is configured to extract entity information from user queries;

[0030] The intent classification unit is configured to identify the type of user's query intent.

[0031] An explicit time recognition unit is configured to convert explicit time representations in a query into standardized time intervals;

[0032] The implicit time mapping unit is configured to convert relative time representations into absolute time intervals based on associated entity attributes and the current system time, and output them to the time-aware multi-hop inference module.

[0033] Preferably, the time-aware multi-hop reasoning module includes an effective knowledge filtering unit, a candidate path generation unit, a path matching degree calculation unit, and an optimal path selection unit;

[0034] The effective knowledge filtering unit is configured to filter invalid triples based on the query time interval to obtain a set of valid triples.

[0035] The candidate path generation unit is configured to perform a multi-hop search based on the set of valid triples to generate all candidate inference paths;

[0036] The path matching degree calculation unit is configured to calculate the time matching degree of each path;

[0037] The optimal path selection unit is configured to select the path with the highest time matching degree to generate candidate answers and output them to the causal-guided anti-illusion verification module.

[0038] Preferably, the causal-guided anti-hallucination verification module includes an answer reverse parsing unit, an attribute matching degree calculation unit, a credibility scoring unit, and an error handling unit;

[0039] The answer reverse parsing unit converts natural language candidate answers into structured triple format. The attribute matching degree calculation unit compares the consistency between the answer relation attributes and the pre-stored attributes of the knowledge graph, and calculates the attribute matching degree. The credibility scoring unit calculates the comprehensive credibility score by multiplying the path time matching degree and the attribute matching degree. The error handling unit feeds back answers whose scores do not reach the preset credibility threshold to the time sequence-aware multi-hop reasoning module to trigger re-reasoning, and marks the corresponding triple as pending review.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] This invention achieves efficient and accurate knowledge graph question-answering interaction by constructing a fully closed-loop processing flow that includes temporal knowledge validity screening, reasoning path quality assessment, causal attribute consistency verification, reliable result output, and closed-loop optimization. The temporal knowledge validity screening pre-processing accurately filters out invalid knowledge based on time information, effectively reducing computational cost during subsequent reasoning. Reasoning path quality assessment selects high-quality reasoning paths from a timeliness perspective, ensuring that reasoning results keep pace with time changes. Causal attribute consistency verification ensures logically sound answers and avoids logical errors. Reliable result output and closed-loop optimization comprehensively evaluate answer quality and continuously optimize the knowledge graph through a feedback mechanism. These closely coordinated steps significantly improve the performance and accuracy of the question-answering system in complex scenarios.

[0042] In this invention, through the collaborative operation of a time-aware knowledge graph construction module, an implicit time constraint parsing module, a time-aware multi-hop reasoning module, a causal-oriented anti-illusion verification module, and a reliable result output module, a comprehensive and high-quality knowledge question-and-answer service is provided to users. The time-aware knowledge graph construction module provides basic data support for the system and supports dynamic updates. The implicit time constraint parsing module accurately parses the time information in the user's query. The time-aware multi-hop reasoning module generates reasonable reasoning paths based on effective knowledge. The causal-oriented anti-illusion verification module ensures the logical correctness of the answer. The reliable result output module ensures the traceability of the answer and promotes continuous optimization of the system. The collaboration of these modules realizes the full-process automation and intelligence from query to answer output.

[0043] This invention significantly improves the reliability, accuracy, and adaptability of the knowledge graph question-answering interaction system by comprehensively utilizing multiple technical means and a closed-loop optimization mechanism. The dual verification mechanism effectively reduces timeliness errors and logical illusions, ensuring that answers meet requirements in both time and logic. The closed-loop optimization mechanism enables the system to continuously learn and improve based on actual application scenarios, thereby continuously enhancing the accuracy of the knowledge graph and the quality of question-answering interaction. At the same time, the standardized data interaction between the various modules of the system and the clear logical links enable it to flexibly respond to different user query needs, providing users with accurate, reliable, and traceable answers to meet diverse requirements. Attached Figure Description

[0044] Figure 1 This is an overall flowchart of the present invention;

[0045] Figure 2 This is a flowchart of the triplet validity determination and path matching degree calculation in this invention;

[0046] Figure 3 This is a flowchart of the causal attribute matching and credibility score calculation in this invention;

[0047] Figure 4 This is a flowchart of the knowledge graph closed-loop update process in this invention. Detailed Implementation

[0048] The present invention will now be described in further detail with reference to the accompanying drawings and preferred embodiments.

[0049] Example 1, refer to Figure 1-4As shown, an AI-based knowledge graph question-answering interaction method includes four steps: time-series knowledge validity screening, reasoning path quality assessment, causal attribute consistency verification, and reliable result output and closed-loop optimization. Specifically, it first receives the user's natural language query request and extracts the corresponding time interval. Then, it filters the triples associated in the knowledge graph by time dimension, filtering out invalid knowledge that exceeds the time range. Subsequently, it performs multi-hop reasoning based on the filtered valid triples to generate candidate paths, calculates the time matching degree of each path to evaluate the timeliness of the path, then performs reverse parsing on the generated candidate answers to verify the consistency between their relational attributes and the pre-stored attributes of the knowledge graph to obtain the attribute matching degree. Finally, it merges the two types of matching degrees to calculate the credibility of the answer, outputs the answer that meets the requirements, and marks low-quality knowledge for review to iteratively update the knowledge graph, forming a closed-loop processing flow of screening-reasoning-verification-optimization.

[0050] In this embodiment, the computational cost of invalid reasoning is greatly reduced by the pre-screening in the time dimension, the dual verification mechanism effectively reduces the timeliness errors and logical illusions that are common in knowledge graph question answering, and the closed-loop iteration mechanism enables the accuracy of the knowledge graph content to be continuously optimized as the frequency of use increases, significantly improving the question answering accuracy in complex time-constrained scenarios.

[0051] Example 2, refer to Figure 1-4 As shown, the time-series knowledge validity screening step is a key step in ensuring the timeliness of knowledge. Specifically, each triple in the knowledge graph is first labeled with a valid start time T. s and effective end time T e These time attributes can be generated automatically through manual annotation or knowledge-based time characteristics. Upon receiving a user query request, the system uses natural language processing technology to identify the time constraints in the query. For explicit time expressions, such as "January 1, 2024," they are directly converted into a standardized time interval. For implicit time expressions, such as "last month," they are mapped to an absolute time interval by combining the associated entity attributes and the current system time. Let T be the starting time of the query time interval. qs The end time is T. qe For each triple associated with the query entity in the knowledge graph, the validity of the triple is determined by the triple time validity determination formula, as follows:

[0052] ;

[0053] In the formula, T valid The result of the validity determination of a single triplet is a binary value. I(•) is an indicator function that returns 1 when the logical expression within the parentheses is true, indicating that the triplet is valid, and 0 otherwise, indicating that the triplet is invalid. T sT represents the effective start time of the current triplet pre-labeling. e T is the effective end time of the current triplet pre-labeling. qs T is the start and end time of the time range that the user queries. qe The system calculates the end time of the user's query time interval and can quickly and accurately filter out valid triples that match the query time interval, providing a reliable knowledge base for subsequent reasoning.

[0054] Specifically, firstly, time information of user queries is extracted and transformed to obtain standardized query time intervals. Then, for each triple in the knowledge graph, the time validity is determined using the above formula based on its pre-labeled valid start time and valid end time, as well as the query time interval. Based on the determination results, the set of valid triples is used for subsequent reasoning, and invalid triples are marked and incorporated into the knowledge graph iterative update process.

[0055] Through precise formula calculations, valid triples that meet the query time requirements can be accurately selected, while invalid knowledge is filtered out. This ensures the accuracy of the knowledge upon which subsequent reasoning is based in the time dimension, improves the reliability and accuracy of the entire question-and-answer interaction system, and provides users with answers that better meet their actual needs.

[0056] Example 3, referring to Figure 1-4 As shown, based on the set of valid triples obtained through effective knowledge filtering, path search is needed to generate all candidate inference paths. Let the total number of triples contained in the inference path be N, and the validity determination result of the i-th triple in the path be T. valid,i The result was calculated using the formula in the time series knowledge validity screening in Example 2, and the validity judgment result T was obtained. valid,i A value of 1 indicates that the triplet is valid, and a value of 0 indicates that the triplet is invalid. The start and end times of the validity of the triplet are T and T, respectively. s-i T e-i The start and end times of the query time interval are T and T, respectively. qs T qe After generating all candidate inference paths, the time matching degree of each candidate path is calculated using the following formula:

[0057] ;

[0058] In the formula, M p This represents the time matching degree of the current inference path, with a value ranging from 0 to 1. A higher value indicates a stronger overall timeliness of the path.

[0059] In the formula, This is used to calculate the overlap ratio between a single triple and the query time interval, reflecting the temporal matching degree of a single piece of knowledge. By calculating the ratio of the length of the overlap between each valid triple and the query time interval to the length of the query time interval, it is possible to intuitively understand the coverage of the triple within the query time range. The validity judgment result T valid,i As a result of the indicator function, it is ensured that only valid triples participate in the calculation, avoiding the influence of invalid triples on the result. The average of the overlap length ratios of all valid triples is used to obtain the time matching degree M of the entire inference path. p The higher the ratio, the more the triples in the path completely cover the query time interval, and the stronger the overall timeliness. When there are multiple path conflicts, choosing the path with the highest time matching degree as the reasoning basis can effectively improve the accuracy and reliability of the reasoning results.

[0060] Specifically, after determining the query time interval and the set of valid triples, for each candidate inference path, the overlap between each triple and the query time interval and related parameters are calculated in turn. The time matching degree of each path is obtained by step-by-step calculation according to the formula. The candidate paths are sorted according to the matching degree, and the path with the highest time matching degree is selected as the optimal inference path to generate candidate answers.

[0061] Through precise formula calculations, the matching degree between the triples in the reasoning path and the query time interval can be accurately measured, the reasoning path with the strongest timeliness can be selected, the quality and reliability of the reasoning results can be improved, and a strong guarantee can be provided for generating accurate and reliable candidate answers, thus meeting users' requirements for the timeliness of knowledge.

[0062] Example 4, refer to Figure 1-4 As shown, in the causal attribute consistency verification stage, the direction D is first pre-labeled for each type of relationship in the knowledge graph. k Symmetry S k and transitivity Tr k Three types of causal attribute labels are used. These labels can be generated manually or automatically based on knowledge-based logical relationships. After the reasoning generates candidate answers, a deep learning-based semantic parsing model is used to convert the natural language answers into structured triplet formats. The relational attributes corresponding to the answer triples are extracted, namely, directionality D. a Symmetry S a and transitivity Tr a The attribute matching degree is calculated by matching the attribute tags with the corresponding relationships in the knowledge graph, using the following formula:

[0063] ;

[0064] In the formula, C matchThe attribute matching degree of the candidate answer, with a value ranging from 0 to 1. The higher the value, the stronger the logical rationality of the answer. I(•) is an indicator function that returns 1 when the equation in the parentheses is true, and 0 otherwise.

[0065] After obtaining the path time matching degree M p Attribute matching degree C match Then, the time matching degree M of the inference path is... p Attribute matching degree C match The reliability score S of the answer obtained by performing product calculations cred The specific formula is as follows:

[0066] S cred =M p ×C match ;

[0067] S cred The answer credibility score is a comprehensive score, ranging from 0 to 1. The higher the score, the better the overall quality of the answer. The answer credibility score takes into account the timeliness of the reasoning path and the consistency of the logical attributes of the answer, and can more comprehensively evaluate the credibility of the answer.

[0068] Based on the scoring results, if the overall credibility score of the answer is S cred If the preset credibility threshold is reached, the corresponding knowledge's effective time range and complete reasoning path will be simultaneously labeled and output to the user. If the overall credibility score of the answer is S... cred If the preset credibility threshold is not reached, the re-inference process is automatically triggered. At the same time, the triples involved in the inference path and the selected invalid triples are marked as pending review status for the content iteration and update of the knowledge graph, so as to realize the closed-loop optimization of the system.

[0069] Specifically, the candidate answers are first checked for causal attribute consistency. The attribute matching degree is calculated using a formula to determine whether there are logical errors in the answer. Then, the answer credibility score is calculated by combining the time matching degree of the reasoning path and using a formula. Based on the score results, the answer is processed accordingly, such as outputting it to the user or triggering a re-reasoning process. The relevant triples are also marked for updating and optimizing the knowledge graph.

[0070] Through precise calculations, the consistency of causal attributes and overall credibility of the answers can be accurately assessed, ensuring that the answers output to users meet the requirements in terms of logic and timeliness. At the same time, the closed-loop optimization mechanism enables the system to detect and correct errors in a timely manner, continuously improving the system's performance and the accuracy of the knowledge graph, and providing users with higher-quality and more reliable knowledge question-and-answer interaction services.

[0071] Furthermore, all triples marked as pending review will be uniformly pushed to the knowledge update unit for subsequent processing: First, the system automatically summarizes the invalid triples and low-scoring association triples pending review to form a verification queue, which is then handed over to manual verification personnel for verification. If it is determined that the valid time tag or causal attribute tag of the triple is incorrectly labeled, the corresponding tag is directly corrected and re-entered into the database. If it is determined that the label is correct but the factual knowledge corresponding to the triple has expired, the triple is marked as a historical archive and will no longer participate in the subsequent reasoning process. If it is determined that the triple content and tag are both correct, its original attributes are retained and the pending review mark is removed. The triple data after the verification process is completed will be synchronously updated to the triple storage unit to provide more accurate knowledge support for the subsequent reasoning process.

[0072] Example 5, refer to Figure 1-4 As shown, the present invention also provides an artificial intelligence-based knowledge graph question-answering interaction system, comprising five components: a time-aware knowledge graph construction module, an implicit time constraint parsing module, a time-aware multi-hop reasoning module, a causal-oriented anti-illusion verification module, and a reliable result output module;

[0073] First, the time-aware knowledge graph construction module pre-stores triplet data with time and causal attribute labels. After receiving user queries, the implicit time constraint parsing module identifies the explicit and implicit time constraints in the query and converts them into standardized time intervals. Then, the time-aware multi-hop reasoning module filters effective knowledge based on the time intervals and generates candidate reasoning paths. The causal-guided anti-illusion verification module performs logical verification on the candidate answers and calculates their credibility. Finally, the credibility result output module adds source information and outputs the answer. At the same time, low-quality knowledge is fed back to the time-aware knowledge graph construction module to complete the content update.

[0074] Each module operates collaboratively according to the logical link of knowledge storage, query parsing, reasoning generation, verification and optimization, and result output. Data interaction between modules is completely standardized, and the verification results are fed back into the knowledge update process to form a system-level closed loop.

[0075] In this embodiment, through the collaborative work of various modules, the entire process from user query to answer output is automated, which can provide users with accurate, reliable and traceable answers. At the same time, the closed-loop feedback mechanism of the system helps to continuously optimize the knowledge graph and improve the quality of question-and-answer interaction, thus meeting users' diverse needs in knowledge-based question and answer.

[0076] It should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should also be within the scope of protection of this invention.

Claims

1. A knowledge graph question-answering interaction method based on artificial intelligence, characterized in that: Includes the following steps: Time-series knowledge validity filtering: After receiving a user query request, extract the time interval range corresponding to the query, traverse all triples associated with the query entity in the knowledge graph, compare the coverage relationship between the valid time range of the triples and the query time interval one by one, filter out the set of valid triples that match the time dimension, and filter out invalid knowledge that exceeds the query time range. Inference path quality assessment: Based on the effective set of triples, multi-hop inference is performed to generate candidate inference paths. The overlap length ratio between all triples in each path and the query time interval is calculated to obtain the time matching degree of the path. The higher the matching degree, the stronger the timeliness of the path knowledge. Causal attribute consistency verification: The candidate answers generated by reasoning are reverse parsed to obtain structured triples. The parsed relational attributes are matched with the pre-stored relational attributes in the knowledge graph, and the proportion of the number of items with consistent attributes to the total number of verified attributes is counted to obtain the attribute matching degree. Credible result output and closed-loop optimization: The credibility score of the answer is calculated by multiplying the path time matching degree and the attribute matching degree. The answer is judged to meet the credibility requirements based on the score. For answers that do not meet the requirements, a re-reasoning process is triggered. At the same time, the triples corresponding to the low scores and the selected invalid triples are marked as pending review status for the content iteration and update of the knowledge graph.

2. The knowledge graph question-answering interaction method based on artificial intelligence according to claim 1, characterized in that: The specific steps for screening the validity of time-series knowledge are as follows: Each triple in the knowledge graph is pre-labeled with valid start and end time attributes. After receiving user queries, natural language processing technology is used to identify the time constraints in the query. Explicit time expressions are directly converted into standardized time intervals, and implicit time expressions are mapped to absolute time intervals by combining associated entity attributes and the current system time. The validity time range of each triple is checked to see if it intersects with the query time interval. Triples with intersections are determined to be valid and enter the inference path quality assessment step. Invalid triples without intersections are simultaneously marked as pending review and included in the knowledge graph iteration update process.

3. The knowledge graph question-answering interaction method based on artificial intelligence according to claim 2, characterized in that: The inference path quality assessment steps are as follows: Based on the set of valid triples, path search is performed to generate all candidate inference paths. For each candidate path, the total number of triples contained in the path is counted, and the overlap length between each valid triple and the query time interval is calculated. The sum of all overlap lengths is then multiplied by the product of the total length of the query time interval and the total number of triples to obtain the time matching degree of the path. This ratio is used to reflect the degree of time dimension matching of the entire inference path. The higher the ratio, the more triples in the path completely cover the query time interval, and the stronger the overall timeliness. When there are multiple path conflicts, the path with the highest time matching degree is selected as the basis for inference.

4. The knowledge graph question-answering interaction method based on artificial intelligence according to claim 3, characterized in that: The specific steps for verifying the consistency of causal attributes are as follows: Each type of relation in the knowledge graph is pre-labeled with three types of causal attribute tags: directionality, symmetry, and transitivity. After generating candidate answers through reasoning, the natural language answers are converted into structured triples using semantic parsing technology. The relation attributes corresponding to the answer triples are extracted and compared with the pre-stored attributes of the relation in the knowledge graph. The number of matching attributes is counted, and the ratio of the number of matching attributes to the total number of verified attributes is used as the attribute matching degree. Answers whose matching degree does not reach the preset attribute verification threshold are judged to have logical errors.

5. The knowledge graph question-answering interaction method based on artificial intelligence according to claim 4, characterized in that: The specific steps for the reliable result output and closed-loop optimization are as follows: The comprehensive credibility score of the answer is calculated by multiplying the time matching degree and attribute matching degree of the reasoning path. Answers with a score that reaches the preset credibility threshold are simultaneously marked with the effective time range of the corresponding knowledge and the complete reasoning path and then output to the user. Answers with a score that does not reach the preset credibility threshold are automatically triggered to re-infer the process, and the triples involved in the reasoning path and the filtered invalid triples are marked as pending review.

6. A knowledge graph question-answering interaction system that implements the knowledge graph question-answering interaction method based on artificial intelligence as described in any one of claims 1-5, characterized in that: It includes a time-aware knowledge graph construction module, an implicit time constraint parsing module, a time-aware multi-hop reasoning module, a causal-oriented anti-illusion verification module, and a reliable result output module; The time-aware knowledge graph construction module is configured to store triple data with valid time attributes and causal attribute labels, receive low-quality triple label information from the causal-guided anti-illusion verification module and invalid triple label information from the time-aware multi-hop reasoning module, update the attribute labels and valid time range of the knowledge graph, and provide structured knowledge support for other modules. The implicit time constraint parsing module is configured to receive user natural language queries, identify time constraint expressions in the queries, convert explicit time into standardized time intervals, map implicit time expressions into absolute time intervals, and output to the time-aware multi-hop inference module. The time-aware multi-hop reasoning module is configured to filter the set of valid triples based on the parsed query time interval, entity and intent information, generate candidate reasoning paths and calculate the path time matching degree, select the optimal path to generate candidate answers and output them to the causal-oriented anti-illusion verification module, and at the same time mark the filtered invalid triples as pending review status and feed them back to the time-aware knowledge graph construction module. The causal-guided anti-illusion verification module is configured to perform reverse parsing on candidate answers, verify the consistency between the answer relationship attributes and the knowledge graph attributes, calculate the attribute matching degree, calculate the credibility score by multiplying and fusing the time matching degree, feed back low-scoring answers to the time-aware multi-hop reasoning module to trigger re-reasoning, and output high-scoring answers to the credibility result output module. The reliable result output module is configured to add knowledge validity time range and reasoning path tracing information to the verified answer, output it to the user in a formatted manner, and store all intermediate data in the reasoning process for subsequent optimization.

7. The knowledge graph question-answering interactive system based on artificial intelligence according to claim 6, characterized in that: The time-aware knowledge graph construction module includes a triplet storage unit, a time attribute annotation unit, a causal attribute annotation unit, and a knowledge update unit; The triplet storage unit is configured to store basic triplet data consisting of head entity, relation, and tail entity. The time attribute labeling unit labels each triplet with valid start time and valid end time tags. The causal attribute labeling unit is configured to label each type of relationship with three types of causal attribute tags: directionality, symmetry, and transitivity. The knowledge update unit is configured to receive the triplet information to be verified from the causal-guided anti-illusion verification module and the invalid triplet information to be verified from the time-aware multi-hop reasoning module, and update the attribute tags and valid time range of the triplet after manual verification.

8. The knowledge graph question-answering interactive system based on artificial intelligence according to claim 6, characterized in that: The implicit time constraint parsing module includes an entity recognition unit, an intent classification unit, an explicit time recognition unit, and an implicit time mapping unit; The entity recognition unit is configured to extract entity information from user queries; The intent classification unit is configured to identify the type of user's query intent. An explicit time recognition unit is configured to convert explicit time representations in a query into standardized time intervals; The implicit time mapping unit is configured to convert relative time representations into absolute time intervals based on associated entity attributes and the current system time, and output them to the time-aware multi-hop inference module.

9. The knowledge graph question-answering interactive system based on artificial intelligence according to claim 8, characterized in that: The time-aware multi-hop reasoning module includes an effective knowledge filtering unit, a candidate path generation unit, a path matching degree calculation unit, and an optimal path selection unit. The effective knowledge filtering unit is configured to filter invalid triples based on the query time interval to obtain a set of valid triples. The candidate path generation unit is configured to perform a multi-hop search based on the set of valid triples to generate all candidate inference paths; The path matching degree calculation unit is configured to calculate the time matching degree of each path; The optimal path selection unit is configured to select the path with the highest time matching degree to generate candidate answers and output them to the causal-guided anti-illusion verification module.

10. The knowledge graph question-answering interactive system based on artificial intelligence according to claim 9, characterized in that: The causal-guided anti-hallucination verification module includes an answer reverse parsing unit, an attribute matching degree calculation unit, a credibility scoring unit, and an error handling unit; The answer reverse parsing unit converts natural language candidate answers into structured triple format. The attribute matching degree calculation unit compares the consistency between the answer relation attributes and the pre-stored attributes of the knowledge graph, and calculates the attribute matching degree. The credibility scoring unit calculates the comprehensive credibility score by multiplying the path time matching degree and the attribute matching degree. The error handling unit feeds back answers whose scores do not reach the preset credibility threshold to the time sequence-aware multi-hop reasoning module to trigger re-reasoning, and marks the corresponding triple as pending review.

Images