A general knowledge extraction and spatiotemporal normalization expression method and related equipment

Abstract

The present invention provides a general knowledge extraction and spatiotemporal normalized expression method and related equipment, including: obtaining text data to be processed and defining a knowledge extraction ontology; extracting knowledge and normalizing the expression of the text data and the knowledge extraction ontology into a multi-agent hierarchical collaborative decision-making system constructed by inputting the text data to be processed and the knowledge extraction ontology to obtain the spatiotemporal knowledge expression result corresponding to the text data; decomposing complex tasks into sub-tasks with clear semantic boundaries through the master control agent in the multi-agent hierarchical collaborative decision-making system, using professional sub-agents in the multi-agent hierarchical collaborative decision-making system to deeply understand and normalize the complex spatiotemporal semantics, and using a decision-making process based on explicit reasoning to complete the processing, so that each link of the knowledge extraction can provide detailed processing basis and reasoning process, thereby solving the problem of insufficient explainability and controllability in the spatiotemporal knowledge extraction process in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a general knowledge extraction and spatiotemporal normalization expression method and related equipment. Background Art

[0002] With the vigorous development of new generation information technologies such as mobile Internet, ubiquitous text data widely distributed in various digital platforms (such as social media, news websites, professional document libraries, etc.) have shown the characteristics of rapid growth in quantity, fast update frequency, and wide coverage. These data contain rich spatiotemporal semantic information, have strong timeliness and relevance, and still become the key data source for obtaining spatiotemporal knowledge. However, ubiquitous text data composed of natural language often have inherent characteristics such as imprecise expression and inconsistent description. There are significant differences between the spatiotemporal features contained therein and the standardized expression based on the uncontrolled reference system required by spatiotemporal knowledge. For example, texts usually use place names and addresses to implicitly describe spatiotemporal features, and there are few accurate expressions of location or time. At the same time, different references to the same spatiotemporal context may have different forms of description, which greatly limits its application potential in downstream tasks such as spatiotemporal knowledge discovery and reasoning. Therefore, how to effectively recognize and process the ambiguity and diversity of semantic expression in ubiquitous text data, research and develop universal knowledge extraction and spatiotemporal normalization expression methods, and support high-quality spatiotemporal knowledge acquisition and services for practical applications, has important theoretical value and practical significance.

[0003] At present, the research methods for spatiotemporal knowledge extraction and standardized expression can be mainly divided into two categories: pipeline-based multi-stage processing methods and end-to-end unified modeling methods. The pipeline method adopts a modular serial processing paradigm, which usually includes core links such as knowledge extraction, spatiotemporal element recognition, and spatiotemporal semantic normalization. Its representative works include the Chinese address element parsing method based on rule templates and the address matching model DeepAM integrated with deep learning. The processing flow of this type of method is clear and has good interpretability and controllability, but its processing links are relatively independent, and it is difficult to fully utilize the semantic association information between links. At the same time, it shows weak generalization ability when facing uncovered task scenarios, and often requires continuous manual intervention and optimization. It is not effective in dealing with spatiotemporal semantic fuzziness and ambiguity problems that are common in massive texts. The end-to-end method directly models the task as a mapping learning problem from "spatial-temporal semantic text to standardized spatiotemporal expression". Typical representatives include the "geographic location-language" pre-training model ERNIE-GeoL and the multimodal geographic language model pre-training model MGeo. Although this type of method can achieve end-to-end automated processing, it is highly dependent on large-scale task-related labeled data, and the use of a black-box modeling approach leads to a lack of interpretability in the prediction results, making it difficult to ensure the robustness and reliability of the model in complex scenarios. Summary of the invention

[0004] The present invention provides a general knowledge extraction and spatiotemporal normalization expression method and related equipment, the purpose of which is to solve the problem of insufficient explainability and controllability in the spatiotemporal knowledge extraction process.

[0005] In order to achieve the above object, the present invention provides a general knowledge extraction and spatiotemporal normalization expression method, comprising:

[0006] Step 1: Obtain the text data to be processed and define the knowledge extraction ontology;

[0007] Step 2: extract knowledge and normalize the expression of the multi-agent hierarchical collaborative decision-making system constructed by inputting the text data to be processed into the knowledge extraction ontology, and obtain the spatiotemporal knowledge expression result corresponding to the text data;

[0008] The multi-agent hierarchical collaborative decision-making system includes a master agent and at least one professional sub-agent;

[0009] The master agent is used to analyze the input text data to determine the subtasks, integrate the processing results of the professional sub-agents, and output the spatiotemporal knowledge expression results corresponding to the input text data according to the knowledge extraction ontology;

[0010] Professional sub-agents are used to complete sub-tasks and output processing results.

[0011] Specifically, the master agent includes:

[0012] The intention understanding module is used to identify the intention of the received text data to be processed and obtain the subtasks and the execution process of the subtasks;

[0013] The task scheduling module is used to determine the task execution plan according to the execution process of the subtasks, and schedule the subtasks to the professional sub-agents according to the task execution plan;

[0014] Data recovery module, used to collect processing data of professional sub-agents;

[0015] A quality check module is used to check the quality of the processing results of the professional sub-agents;

[0016] An integration module is used to integrate the processing results after quality inspection;

[0017] The input end of the intention understanding module is the first input end of the main control agent, which is used to receive the text data to be processed and the knowledge extraction entity. The output end of the intention understanding module is connected to the input end of the task scheduling module. The output end of the task scheduling module is connected to the first output end of the main control agent and the input end of the professional sub-agent. The input end of the data recovery module is connected to the second input end of the main control agent and the output end of the professional sub-agent. The output end of the data recovery module is connected to the input end of the quality inspection module, and the output end of the quality inspection module is connected to the input end of the integration module. The output end of the integration module is the second output end of the main control agent, which is used to output the spatiotemporal knowledge expression results corresponding to the text data.

[0018] Furthermore, the collected extraction results are quality checked, including:

[0019] Conduct compliance checks on the format of collected processing results;

[0020] Perform a reasonableness check on the content of the collected processing results.

[0021] Furthermore, there are multiple professional sub-agents, namely:

[0022] The general knowledge extraction sub-agent is used to complete the general knowledge extraction sub-task, obtain multiple extraction results, and fuse the multiple extraction results to obtain the fused extraction result;

[0023] The time standardization sub-agent is used to complete the time description normalization sub-task and obtain the time standardization result;

[0024] Place name reference disambiguation sub-agent, used to complete the place name reference disambiguation sub-task and obtain the place name disambiguation result;

[0025] The spatial position reasoning sub-agent is used to complete the spatial position reasoning sub-task and obtain the spatial position reasoning result.

[0026] Furthermore, the general knowledge extraction sub-agent integrates three general knowledge extraction models: RexUIE, OneKE, and Qwen2.5;

[0027] The general knowledge extraction subtask is completed by calling three general knowledge extraction models, RexUIE, OneKE and Qwen2.5, in parallel.

[0028] Specifically, the place name reference disambiguation subagent is used to:

[0029] Call the POI data retrieval tool through the function calling mechanism to obtain the candidate place name entity item referred to by the place name;

[0030] In combination with the context of the text data to be processed, the place name entity item with the highest matching degree is selected from the candidate place name entity items;

[0031] Determine the type of the place name entity item with the highest matching degree and obtain the determination result;

[0032] According to the judgment result, the AOI information retrieval tool is called to refine the expression of the place name address boundary for the place name entity item with the highest matching degree, so as to complete the place name reference resolution subtask and obtain the place name resolution result.

[0033] Specifically, the spatial position reasoning sub-agent includes:

[0034] The feature grid modeling tool is used to discretize the features into a matrix grid structure to obtain the feature grid matrix;

[0035] The distance relationship quantification tool is used to expand the feature grid matrix step by step to generate a concentric ring matrix, and set the number of expansion layers to quantify the relative distance between the subject and the reference feature to obtain the distance relationship quantification result;

[0036] Azimuth relationship quantification tool, used to calculate the azimuth relationship between all feature grids and feature grid matrices, quantify the matching degree between each grid unit and the specified direction with normal distribution, and obtain the azimuth relationship quantification result;

[0037] The topological relationship quantification tool is used to obtain the spatial grid representation of the adjacent area of ​​the object through the expansion operator and the erosion operator based on the object grid matrix, and obtain the spatial grid representation of the external area of ​​the object through the matrix inversion operation to obtain the topological relationship quantification result;

[0038] The inference result fusion tool is used to fuse the feature grid matrix with the distance relationship quantification results, orientation relationship quantification results, and topological relationship quantification results to obtain a fusion matrix, and retain the non-zero weight values ​​of the overlapping areas in the fusion matrix to obtain the spatial position inference results.

[0039] The present invention also provides a general knowledge extraction and spatiotemporal normalization expression device, comprising:

[0040] The acquisition module is used to obtain the text data to be processed and define the knowledge extraction ontology;

[0041] The processing module is used to extract knowledge and normalize the expression of the multi-agent hierarchical collaborative decision-making system constructed by inputting the text data to be processed and the knowledge extraction ontology, and obtain the spatiotemporal knowledge expression result corresponding to the text data;

[0042] The multi-agent hierarchical collaborative decision-making system includes a master agent and at least one professional sub-agent;

[0043] The master agent is used to analyze the input text data to determine the subtasks, integrate the processing results of the professional sub-agents, and output the spatiotemporal knowledge expression results corresponding to the input text data according to the knowledge extraction ontology;

[0044] Professional sub-agents are used to complete sub-tasks and output processing results.

[0045] The present invention also provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a general knowledge extraction and spatiotemporal normalization expression method is implemented.

[0046] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the method for extracting general knowledge and expressing it in a time-space normalized manner is implemented.

[0047] The above scheme of the present invention has the following beneficial effects:

[0048] The present invention obtains text data to be processed and defines a knowledge extraction ontology; extracts knowledge and normalizes the expression of knowledge in a multi-agent hierarchical collaborative decision-making system constructed by inputting the text data to be processed and the knowledge extraction ontology, and obtains the spatiotemporal knowledge expression result corresponding to the text data; the multi-agent hierarchical collaborative decision-making system includes a main control agent and at least one professional sub-agent; the main control agent is used to analyze the input text data to determine the subtasks, the processing results of the professional sub-agents are integrated, and the spatiotemporal knowledge expression result corresponding to the input text data is output according to the knowledge extraction ontology; the professional sub-agents are used to Complete subtasks and output processing results; compared with the prior art, the present invention constructs a multi-agent hierarchical collaborative decision-making system, utilizes the master agent in the multi-agent hierarchical collaborative decision-making system to decompose complex tasks into subtasks with clear semantic boundaries, utilizes professional sub-agents in the multi-agent hierarchical collaborative decision-making system to deeply understand and standardize complex spatiotemporal semantics, and uses a decision-making process based on explicit reasoning to complete the processing, so that each link of knowledge extraction can provide detailed processing basis and reasoning process, thereby solving the problem of insufficient interpretability and controllability in the spatiotemporal knowledge extraction process in the prior art.

[0049] Other beneficial effects of the present invention will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A schematic diagram of a flow chart of an embodiment of the present invention;

[0051] Figure 2 This is an architecture diagram of a multi-agent hierarchical collaborative decision-making system in an embodiment of the present invention;

[0052] Figure 3 It is a structural block diagram of the general knowledge extraction and spatiotemporal normalization expression device;

[0053] Figure 4 A structural block diagram of a terminal device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] In the description of the present invention, it should be noted that the terms “first”, “second” and “third” are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0056] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installation", "connection" and "connection" should be understood in a broad sense, for example, it can be a locking connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0057] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0058] In view of the existing problems, the present invention provides a general knowledge extraction and spatiotemporal normalization expression method and related equipment.

[0059] like Figure 1 and Figure 2 As shown, an embodiment of the present invention provides a general knowledge extraction and spatiotemporal normalization expression method, including:

[0060] Step 1: Obtain the text data to be processed and define the knowledge extraction ontology;

[0061] Step 2: extract knowledge and normalize the expression of the multi-agent hierarchical collaborative decision-making system constructed by inputting the text data to be processed into the knowledge extraction ontology, and obtain the spatiotemporal knowledge expression result corresponding to the text data;

[0062] The multi-agent hierarchical collaborative decision-making system includes a master agent MA and at least one professional sub-agent SA;

[0063] The master agent MA is used to analyze the input text data to determine the subtasks, integrate the processing results of the professional sub-agents, and output the spatiotemporal knowledge expression results corresponding to the input text data according to the knowledge extraction ontology;

[0064] The professional sub-agent SA is used to complete the sub-task and output the processing results.

[0065] Most preferably, the master agent MA comprises:

[0066] The intention understanding module is used to identify the intention of the received text data to be processed and obtain the subtasks and the execution process of the subtasks;

[0067] The task scheduling module is used to determine the task execution plan according to the execution process of the subtasks, and schedule the subtasks to the professional sub-agents according to the task execution plan;

[0068] Data recovery module, used to collect processing data of professional sub-agents;

[0069] A quality check module is used to check the quality of the processing results of the professional sub-agents;

[0070] An integration module is used to integrate the processing results after quality inspection;

[0071] The input end of the intention understanding module is the first input end of the main control agent, which is used to receive the text data to be processed and the knowledge extraction entity. The output end of the intention understanding module is connected to the input end of the task scheduling module. The output end of the task scheduling module is connected to the first output end of the main control agent and the input end of the professional sub-agent. The input end of the data recovery module is connected to the second input end of the main control agent and the output end of the professional sub-agent. The output end of the data recovery module is connected to the input end of the quality inspection module, and the output end of the quality inspection module is connected to the input end of the integration module. The output end of the integration module is the second output end of the main control agent, which is used to output the spatiotemporal knowledge expression results corresponding to the text data.

[0072] Specifically, in the multi-agent hierarchical collaborative decision-making system, the master agent and the professional sub-agents are both based on the large language model. The master agent first receives the system input, analyzes the task features in the system input, infers the sub-tasks involved and the execution scheme of the sub-tasks, and assigns the specific sub-tasks to the task. Assigned to professional sub-agents, sub-tasks include general knowledge extraction, time attribute standardization, place name reference disambiguation, spatial location reasoning, etc. Finally, the main control agent collects the output of the professional sub-agents And conduct quality checks and integrate all content into standardized system output .

[0073] In the embodiment of the present invention, in order to realize the calling of professional sub-agents, the main control agent adopts an interface design based on the function calling mechanism. The professional sub-agents are registered in the main control agent as tool classes, providing functional interfaces of specific task types. After parsing the system input, the main control agent automatically selects and calls the corresponding professional sub-agent interface through the function calling mechanism according to the task requirements, passes the task parameters to it and receives its return.

[0074] In the embodiment of the present invention, the system input It includes the text data to be processed Text and the pre-defined knowledge extraction ontology Schema. The text data to be processed is the natural language description text that needs to be extracted and expressed in time and space normalization. The pre-defined knowledge extraction ontology is the knowledge ontology designed and organized with reference to the RexPrompt format. Its definition is shown in Table 1:

[0075] Table 1 Schema data structure definition table

[0076]

[0077] in, The representation form depends on the specific extraction task It depends on the task, and it is only one of the tasks of entity extraction (NER, NamedEntityRecognition), relationship extraction (RE, RelationExtraction) or event extraction (EE, EventExtraction). In the entity extraction task, is the entity type key, is a set of entity types; in the relation extraction task, is the head entity type key, for The tail entity key under the corresponding relationship, is a header type collection, is the tail entity type corresponding to the head entity; in the event extraction task, is the event type key, for Contains event argument keys, is a collection of event types, is the set of argument types corresponding to the event.

[0078] Among them, the system output It is a reference The predefined data structure is output as a string in a lightweight data exchange format (JSON, JavaScript Object Notation), containing three types of results: events, entities, and relationships.

[0079] Specifically, the collected extraction results are quality checked, including:

[0080] Conduct compliance checks on the format of collected processing results, such as the completeness of knowledge extraction results and the accuracy of address standardization;

[0081] Perform rationality checks on the content of the collected processing results, such as the rationality of spatial reasoning.

[0082] It should be noted that the embodiments of the present invention process the output results through regular matching and the semantic understanding ability of the large model. Check the format for compliance and process the output results Check the content for plausibility.

[0083] Most preferably, the professional sub-agent SA includes multiple, namely:

[0084] General knowledge extraction sub-agent SA1 is used to complete the general knowledge extraction sub-task, obtain multiple extraction results, and fuse the multiple extraction results to obtain the fused extraction result;

[0085] The time standardization sub-agent SA2 is used to complete the time description normalization sub-task and obtain the time standardization result;

[0086] Place name reference disambiguation sub-agent SA3 is used to complete the place name reference disambiguation subtask and obtain the place name disambiguation result;

[0087] The spatial position reasoning sub-agent SA4 is used to complete the spatial position reasoning sub-task and obtain the spatial position reasoning result.

[0088] Specifically, the general knowledge extraction sub-agent SA1 adopts an ensemble learning strategy, integrating three general knowledge extraction models: RexUIE, OneKE and Qwen2.5; by calling the three general knowledge extraction models RexUIE, OneKE and Qwen2.5 in parallel, the input text is parsed from different perspectives to complete the general knowledge extraction sub-task, and the processing results obtained after completing the task are integrated to form a preliminary extraction result containing specified knowledge elements such as entities, attributes, relationships, time, etc., thereby improving the accuracy and robustness of knowledge extraction.

[0089] It should be noted that RexUIE is an end-to-end model based on a unified information extraction framework. It performs stably on various structured extraction tasks and is particularly suitable for processing explicitly expressed knowledge elements. OneKE is a large language model that has been fine-tuned for knowledge extraction tasks. It has strong instruction comprehension capabilities and extraction accuracy, and performs well when processing complex sentences. Qwen2.5 is a general large language model with powerful language comprehension and reasoning capabilities. It is particularly suitable for processing implicit knowledge extraction that requires contextual reasoning, and can serve as a supplement to the other two special models.

[0090] The specific execution process of the general knowledge extraction sub-agent SA1 is as follows:

[0091] First, three general knowledge extraction models, RexUIE, OneKE and Qwen2.5, are called in parallel to extract general knowledge. Each model will output the extraction results based on a defined standardized format.

[0092] On input Based on the processing memory composed of the standardized output of all models, the general knowledge extraction sub-agent will rely on its own large language model to fuse the extraction results, that is, guide the large language model to perform semantic understanding and multi-source information integration through context learning technology, and finally integrate the general knowledge extraction results Feedback to the master agent MA.

[0093] Specifically, the temporal normalization sub-agent SA2 adopts the context learning strategy to guide the large model to normalize the temporal description in the text into a standard format by designing fine-grained prompt templates. <time> Pending time text snippet< / time> Input organized in the form of "context text" , and returns a JSON formatted result containing standardized time information .

[0094] Based on case prompts covering various time processing situations, the time standardization sub-agent SA2 can effectively process expressions such as exact time points (such as "Spring Festival in 2024") and time periods (such as "August 2023 to early 2024"), and use the time anchors in the text to perform contextual reasoning on relative time information (such as "three days later"); the processing results use a unified UTC standard time, where the time point is expressed as "YYYY / MM / DD-HH" and the time period is expressed as "[YYYY / MM / DD-HH, YYYY / MM / DD-HH]". After the processed results are verified for compliance through regular matching, a time standardization result in the format of {"standardized time": YYYY / MM / DD-HH"} is finally formed and fed back to the main control agent MA.

[0095] In the embodiment of the present invention, the regular matching process is to verify whether the processing result is in JSON format and whether the package does not contain the specified field.

[0096] In the embodiment of the present invention, the place name reference disambiguation sub-agent SA3 draws on the idea of ​​entity linking technology, and by retrieving the standardized place name address library of the online map service, ensures that the place name reference in the text can be mapped to the correct place name entity, thereby achieving accurate resolution of the place name reference; the sub-agent receives the " <address> Place name text segment to be processed< / address> Input organized in the form of "context text" , and returns a JSON format result containing place name reference resolution information .

[0097] Specifically, the place name reference disambiguation sub-agent SA3 is used to:

[0098] Call the POI data retrieval tool through the function calling mechanism to obtain the candidate place name entity item referred to by the place name;

[0099] In combination with the context of the text data to be processed, the place name entity item with the highest matching degree is selected from the candidate place name entity items;

[0100] Determine the type of the place name entity item with the highest matching degree and obtain the determination result;

[0101] According to the judgment result, the AOI information retrieval tool is called to refine the expression of the place name address boundary for the place name entity item with the highest matching degree, so as to complete the place name reference resolution subtask and obtain the place name resolution result.

[0102] The place name reference resolution sub-agent SA3 provided in the embodiment of the present invention has the ability to autonomously call the place name address retrieval tool set. In the specific execution process, the sub-agent first calls the point of interest (POI) data retrieval tool through the function calling mechanism to obtain the candidate place name entity items of the place name reference, and selects the most suitable place name entity items from the candidate items in combination with the context, so as to achieve the disambiguation and standardization of the place name reference. After determining the most relevant place name entity item, the sub-agent will further determine the type of the place name entity, and call the area of ​​interest (AOI) information retrieval tool for administrative divisions or specific place name addresses to achieve the detailed expression of the place name address boundary; on this basis, the sub-agent uses the coordinate batch conversion tool to convert the processing results from the map service's own coordinate system to the CGCS2000 geodetic coordinate system. The final place name resolution result will be checked for compliance through regular matching to be organized into a standard file that complies with the GeoJSON specification, and finally returned to the main control agent MA.

[0103] Spatial position reasoning sub-agent provided by an embodiment of the present invention Using the strategy of spatial semantic discretization modeling, by arranging and executing grid-based spatial location reasoning tools, the fuzzy expressed geographic spatial semantics is mapped to a discrete geographic grid, thus realizing the quantitative calculation of spatial semantics in text descriptions. <target> Subject location information< / target> <reference> Reference features< / reference> <coord_path> Reference figure coordinate file path< / coord_path> Input organized in the form of "context text" , and returns a JSON format result containing spatial information .

[0104] Most preferably, the spatial location reasoning sub-agent SA4 includes a variety of spatial grid modeling and relationship quantification tools, including:

[0105] The feature grid modeling tool is used to calculate the appropriate grid size according to the geographical span of the feature, and discretize the feature into a matrix grid structure in the form of a two-dimensional matrix to obtain a feature grid matrix. Each pixel point in the feature grid matrix represents the occupancy of the feature in the corresponding spatial position in the CGCS2000 coordinate system;

[0106] The distance relationship quantification tool is used to expand the feature grid matrix step by step based on the mathematical morphology expansion operator to generate a concentric ring matrix. Each layer of the buffer zone represents a specific distance from the feature, and the number of expansion layers is set to quantify the relative distance between the subject and the reference feature. The distance relationship is accurately expressed layer by layer in the feature grid matrix to obtain the distance relationship quantification result.

[0107] Azimuth relationship quantification tool, used to calculate the azimuth relationship between all feature grids and feature grid matrices using distance-weighted direction vector aggregation strategy, quantify the matching degree between each grid unit and the specified direction using normal distribution, with the highest confidence in the positive direction and gradually attenuating at the edge, thus taking the confidence distribution of the area of ​​interest under the specified orientation of the reference feature as the azimuth relationship quantification result;

[0108] The topological relationship quantification tool is used to obtain the spatial grid representation of the adjacent area of ​​the object through the expansion operator and the erosion operator based on the object grid matrix, and obtain the spatial grid representation of the external area of ​​the object through the matrix inversion operation to obtain the topological relationship quantification result;

[0109] The inference result fusion tool is used to resample the results based on the greatest common resolution, unify the coverage area and grid resolution of various relationship quantification results, and superimpose the terrain grid matrix with the distance relationship quantification results, orientation relationship quantification results, and topological relationship quantification results through element-by-element product at the same resolution to obtain a fusion matrix, and retain the non-zero weight values ​​of the overlapping areas in the fusion matrix to obtain the spatial position reasoning results expressed in confidence.

[0110] Specifically, during the specific execution process, the spatial location reasoning sub-agent SA4 dynamically calls the feature grid modeling and various relationship quantification tools based on the geographic description of the input text, and gradually realizes the quantitative reasoning of the spatial location. The process is as follows:

[0111] First, the sub-agent discretizes the ground objects into a matrix grid structure and determines the spatial position of each grid unit;

[0112] Next, spatial relationships at different levels are quantified using distance, orientation, and topological relationship quantification tools;

[0113] On this basis, the inference result fusion tool fuses the relationship quantization matrices by resampling, boundary expansion and element superposition, retaining only the non-zero confidence values ​​in the overlapping areas to form the final spatial position inference result, which is finally returned to the master intelligent agent MA.

[0114] In order to verify the feasibility of the method provided by the embodiment of the present invention, taking "heavy rain in a certain place on June 24, 2024" as an example, the input text data Text is "2024-06-24 14:59#Heavy rain in a certain place#The rain is too heavy, and I can't go out! I just walked about 500m west from the ×× high-speed railway station. The water has reached my calves and I can't move at all. The road ahead is flooded and many cars are turned off and parked on the roadside. I hope everyone is safe and don't run out! Safety first!"; the predefined knowledge extraction ontology is "{"disaster (time trigger word)":{"disaster location":None,"disaster event":None,"disaster rescue situation":None,"disaster affected situation":None,"disaster situation":None}}", and the obtained spatiotemporal knowledge expression result is as follows:

[0115] {[{"type":"Disaster (event trigger word)","span":"Heavy rain"},{"type":"Location of disaster","span":"500m west of ×× high-speed railway station"},{"type":"Time of disaster","span":"2024-06-2414:59"},{"type":"Disaster damage situation","span":"The accumulated water has reached the calf level, and many cars are parked on the roadside"},{"type":"Standardized time","span":"2024 / 06 / 24-14"},{"type":"Standardized location","span":"500m west of ×× High-speed Railway Station, ×× Road, ×× District, ×× City, ×× Province"},{"type":"Standardized longitude and latitude","span":"(113.05643447390906,28.147027575436226)"},{"type":"aoi coordinate file path","span":" / workspace / GeoCoder / tmp / ×× High-speed Railway Station_West500.json"}]}.

[0116] In the above case, the main control agent MA receives and parses the text data related to the "rainstorm event" and the pre-defined knowledge extraction ontology Schema, and then assigns specific tasks to each sub-agent according to task requirements; the general knowledge extraction sub-agent SA1 extracts key event information from the text, such as time, location and disaster situation; the time standardization sub-agent SA2 standardizes the time information in the text into a unified format; the place name reference disambiguation sub-agent SA3 standardizes the place name reference in the text to ensure that it can be accurately mapped to the relevant geographic information; the spatial position reasoning sub-agent SA4 infers and generates discrete spatial information of the area of ​​interest based on the spatial semantic description; the final integrated spatiotemporal knowledge expression result, in an embodiment of the present invention, the spatiotemporal knowledge expression result is a spatial visualization result.

[0117] The embodiment of the present invention obtains text data to be processed and defines a knowledge extraction ontology; extracts knowledge and normalizes the expression of knowledge in a multi-agent hierarchical collaborative decision-making system constructed by inputting the text data to be processed and the knowledge extraction ontology, and obtains a spatiotemporal knowledge expression result corresponding to the text data; the multi-agent hierarchical collaborative decision-making system includes a main control agent and at least one professional sub-agent; the main control agent is used to analyze the input text data to determine subtasks, the processing results of the professional sub-agents are integrated, and the spatiotemporal knowledge expression result corresponding to the input text data is output according to the knowledge extraction ontology; the professional sub-agents are used to Complete subtasks and output processing results; compared with the prior art, the embodiment of the present invention constructs a multi-agent hierarchical collaborative decision-making system, utilizes the master agent in the multi-agent hierarchical collaborative decision-making system to decompose complex tasks into subtasks with clear semantic boundaries, utilizes professional sub-agents in the multi-agent hierarchical collaborative decision-making system to deeply understand and standardize complex spatiotemporal semantics, and uses a decision-making process based on explicit reasoning to complete the processing, so that each link of knowledge extraction can provide detailed processing basis and reasoning process, thereby solving the problem of insufficient interpretability and controllability in the spatiotemporal knowledge extraction process in the prior art.

[0118] like Figure 3 As shown, the embodiment of the present invention further provides a general knowledge extraction and spatiotemporal normalized expression device 100, the general knowledge extraction and spatiotemporal normalized expression device 100 comprising:

[0119] The acquisition module 101 is used to acquire the text data to be processed and define the knowledge extraction ontology;

[0120] Processing module 102, used to extract knowledge and normalize the expression of the multi-agent hierarchical collaborative decision-making system constructed by inputting the text data to be processed and the knowledge extraction ontology, and obtain the spatiotemporal knowledge expression result corresponding to the text data;

[0121] The multi-agent hierarchical collaborative decision-making system includes a master agent and at least one professional sub-agent;

[0122] The master agent is used to analyze the input text data to determine the subtasks, integrate the processing results of the professional sub-agents, and output the spatiotemporal knowledge expression results corresponding to the input text data according to the knowledge extraction ontology;

[0123] Professional sub-agents are used to complete sub-tasks and output processing results.

[0124] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0125] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0126] The embodiment of the present invention also provides a terminal device, such as Figure 4 As shown, the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 4 Only one processor is shown in the figure), a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100. When the processor D100 executes the computer program D102, the above-mentioned general knowledge extraction and spatiotemporal normalization expression method is implemented.

[0127] The terminal device D10 may be a computing device such as a desktop computer, a notebook, a PDA, a server, a server cluster, a cloud server, etc. The terminal device may include, but is not limited to, a processor D100 and a memory D101. Those skilled in the art will appreciate that Figure 4 This is only an example of the terminal device D10 and does not constitute a limitation on the terminal device D10. The terminal device D10 may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, it may also include input and output devices, network access devices, etc.

[0128] The processor D100 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0129] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may also be an external storage device of the terminal device D10, such as a plug-in hard disk, a smart memory card (SMC, SmartMedia Card), a secure digital (SD, Secure Digital) card, a flash card (Flash Card), etc. equipped on the terminal device D10. Further, the memory D101 may also include both an internal storage unit of the terminal device D10 and an external storage device. The memory D101 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program, etc. The memory D101 may also be used to temporarily store data that has been output or is to be output.

[0130] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0131] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0132] An embodiment of the present invention further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the method for extracting general knowledge and expressing it in a spatiotemporal normalized manner is implemented.

[0133] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device that can carry the computer program code to the construction device / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, RandomAccess Memory), electric carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a disk or an optical disk.

[0134] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A general knowledge extraction and spatiotemporal normalization expression method, characterized in that: include: Step 1: Obtain the text data to be processed and define the knowledge extraction ontology; Step 2, extracting knowledge and normalizing the expression of the multi-agent hierarchical collaborative decision-making system constructed by inputting the text data to be processed into the knowledge extraction ontology, and obtaining the spatiotemporal knowledge expression result corresponding to the text data; The multi-agent hierarchical collaborative decision-making system includes a master control agent and at least one professional sub-agent; The master agent is used to analyze the input text data to determine the subtasks, and is also used to integrate the processing results of the professional sub-agents, and output the spatiotemporal knowledge expression results corresponding to the input text data according to the knowledge extraction ontology; The master control agent includes: An intention understanding module is used to identify the intention of the received text data to be processed, and obtain subtasks and the execution process of the subtasks; A task scheduling module, used to determine a task execution plan according to the execution process of the subtask, and schedule the subtask to the professional sub-agent according to the task execution plan; A data recovery module, used to collect the processing data of the professional sub-agent; A quality inspection module, used for performing quality inspection on the processing results of the professional sub-agent; An integration module is used to integrate the processing results after quality inspection; The input end of the intention understanding module is the first input end of the main control agent, which is used to receive the text data to be processed and the knowledge extraction ontology. The output end of the intention understanding module is connected to the input end of the task scheduling module. The output end of the task scheduling module is the first output end of the main control agent connected to the input end of the professional sub-agent. The input end of the data recovery module is the second input end of the main control agent connected to the output end of the professional sub-agent. The output end of the data recovery module is connected to the input end of the quality inspection module. The output end of the quality inspection module is connected to the input end of the integration module. The output end of the integration module is the second output end of the main control agent, which is used to output the spatiotemporal knowledge expression result corresponding to the text data. The professional sub-agent is used to complete the sub-task and output the processing result; The professional sub-agents include multiple, respectively: The general knowledge extraction sub-agent is used to complete the general knowledge extraction sub-task, obtain multiple extraction results, and fuse the multiple extraction results to obtain the fused extraction result; The time standardization sub-agent is used to complete the time description normalization sub-task and obtain the time standardization result; Place name reference disambiguation sub-agent, used to complete the place name reference disambiguation sub-task and obtain the place name disambiguation result; The spatial position reasoning sub-agent is used to complete the spatial position reasoning sub-task and obtain the spatial position reasoning result.

2. The general knowledge extraction and spatiotemporal normalization expression method according to claim 1 is characterized in that: Perform quality checks on the collected extraction results, including: Conduct compliance checks on the format of collected processing results; Perform a reasonableness check on the content of the collected processing results.

3. The general knowledge extraction and spatiotemporal normalization expression method according to claim 2 is characterized in that: The general knowledge extraction sub-agent integrates three general knowledge extraction models: RexUIE, OneKE and Qwen2.5; The general knowledge extraction subtask is completed by calling three general knowledge extraction models, RexUIE, OneKE and Qwen2.5, in parallel.

4. The general knowledge extraction and spatiotemporal normalization expression method according to claim 3 is characterized in that: The place name reference disambiguation sub-agent is specifically used for: Call the POI data retrieval tool through the function calling mechanism to obtain the candidate place name entity item referred to by the place name; Filtering out the place name entity item with the highest matching degree from the candidate place name entity items in combination with the context of the text data to be processed; Determine the type of the place name entity item with the highest matching degree and obtain the determination result; The AOI information retrieval tool is called according to the judgment result to refine the expression of the place name address boundary for the place name entity item with the highest matching degree, so as to complete the place name reference resolution subtask and obtain the place name resolution result.

5. The general knowledge extraction and spatiotemporal normalization expression method according to claim 4 is characterized in that: The spatial position reasoning sub-agent comprises: The feature grid modeling tool is used to discretize the features into a matrix grid structure to obtain the feature grid matrix; A distance relationship quantification tool is used to expand the feature grid matrix step by step to generate a concentric ring matrix, and set the number of expansion layers to quantify the relative distance between the subject and the reference feature to obtain a distance relationship quantification result; Azimuth relationship quantification tool, used to calculate the azimuth relationship between all feature grids and feature grid matrices, quantify the matching degree between each grid unit and the specified direction with normal distribution, and obtain the azimuth relationship quantification result; The topological relationship quantification tool is used to obtain the spatial grid representation of the adjacent area of ​​the object through the expansion operator and the erosion operator based on the object grid matrix, and obtain the spatial grid representation of the external area of ​​the object through the matrix inversion operation to obtain the topological relationship quantification result; The inference result fusion tool is used to fuse the terrain grid matrix with the distance relationship quantification result, the orientation relationship quantification result, and the topology relationship quantification result to obtain a fusion matrix, and retain the non-zero weight values ​​of the overlapping areas in the fusion matrix to obtain the spatial position reasoning result.

6. A general knowledge extraction and spatiotemporal normalization expression device, characterized in that: include: The acquisition module is used to obtain the text data to be processed and define the knowledge extraction ontology; A processing module, used to extract knowledge and normalize the expression of the multi-agent hierarchical collaborative decision-making system constructed by inputting the text data to be processed and the knowledge extraction ontology, so as to obtain the spatiotemporal knowledge expression result corresponding to the text data; The multi-agent hierarchical collaborative decision-making system includes a master control agent and at least one professional sub-agent; The master agent is used to analyze the input text data to determine the subtasks, and is also used to integrate the processing results of the professional sub-agents, and output the spatiotemporal knowledge expression results corresponding to the input text data according to the knowledge extraction ontology; The master control agent includes: An intention understanding module is used to identify the intention of the received text data to be processed, and obtain subtasks and the execution process of the subtasks; A task scheduling module, used to determine a task execution plan according to the execution process of the subtask, and schedule the subtask to the professional sub-agent according to the task execution plan; A data recovery module, used to collect the processing data of the professional sub-agent; A quality inspection module, used for performing quality inspection on the processing results of the professional sub-agent; An integration module is used to integrate the processing results after quality inspection; The input end of the intention understanding module is the first input end of the main control agent, which is used to receive the text data to be processed and the knowledge extraction ontology. The output end of the intention understanding module is connected to the input end of the task scheduling module. The output end of the task scheduling module is the first output end of the main control agent connected to the input end of the professional sub-agent. The input end of the data recovery module is the second input end of the main control agent connected to the output end of the professional sub-agent. The output end of the data recovery module is connected to the input end of the quality inspection module. The output end of the quality inspection module is connected to the input end of the integration module. The output end of the integration module is the second output end of the main control agent, which is used to output the spatiotemporal knowledge expression result corresponding to the text data. The professional sub-agent is used to complete the sub-task and output the processing result; The professional sub-agents include multiple, respectively: The general knowledge extraction sub-agent is used to complete the general knowledge extraction sub-task, obtain multiple extraction results, and fuse the multiple extraction results to obtain the fused extraction result; The time standardization sub-agent is used to complete the time description normalization sub-task and obtain the time standardization result; Place name reference disambiguation sub-agent, used to complete the place name reference disambiguation sub-task and obtain the place name disambiguation result; The spatial position reasoning sub-agent is used to complete the spatial position reasoning sub-task and obtain the spatial position reasoning result.

7. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the general knowledge extraction and spatiotemporal normalization expression method as described in any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the general knowledge extraction and spatiotemporal normalization expression method as described in any one of claims 1 to 5 is implemented.

Images