Information analysis method and system based on knowledge graph and large model

Abstract

The invention provides an intelligence analysis method and system based on a knowledge graph and a large model, and relates to the technical field of artificial intelligence language and knowledge processing, and the method comprises the following steps: S1, constructing an intelligence knowledge graph; construction of the intelligence knowledge graph relates to a plurality of links including ontology modeling, knowledge extraction, knowledge fusion, knowledge storage and the like. Ontology modeling: on the basis of actual business demands, the knowledge graph adopts a top-down mode for modeling, entities are classified according to'people, objects, cases, events' and the like, and the'people, objects, cases and events' are connected in series through the relationship between the entities. According to the scheme, the knowledge graph is finally integrated with multi-source heterogeneous information, a clear knowledge structure is constructed, a unified intelligence analysis system is formed based on the powerful query capability of a graph database and the expansion capability of a technical and tactical model, and the multi-hop reasoning capability and the aggregation query capability based on the graph are provided; and the powerful semantic comprehension and generation capability of the large language model can be combined.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence language and knowledge processing technology, and in particular to an intelligence analysis method and system based on knowledge graph and large model. Background Art

[0002] With the increase in the scale of public security data, data formats, and the complexity of case relationships, traditional data retrieval functions can no longer meet the current public security business needs. At the same time, it is difficult to achieve semantic understanding and knowledge transformation of massive public security data, and the difficulty of mining and analyzing intelligence clues is becoming increasingly greater. Knowledge graphs are a structured knowledge base that expresses knowledge in the form of nodes and edges. Usually, "entities" are used to express nodes in knowledge graphs, and "relationships" are used to express edges in knowledge graphs. Entities refer to things in the real world, such as people, place names, events, organizations, etc. Relationships are used to express certain connections between different entities. Knowledge graphs can integrate intelligence data from different sources, formats, and fields, extract key information from these data through technologies such as entity recognition and relationship extraction, and present them in the form of graph structures. Association and display, thus providing an effective technical means for intelligence analysis methods in the public security big data environment. The knowledge graph acquires and integrates rich semantic knowledge from massive public security data to form a huge and interconnected public security intelligence knowledge network, and performs on-graph analysis and presentation through a visual interface to achieve automatic mining and reasoning of potential hidden information, providing strong support for intelligence analysts to conduct clue mining, analysis and judgment, etc. The intelligence analysis system based on the knowledge graph can integrate data from different sources, provide a unified view, and display complex relationships and data in a visual way, so that analysts can intuitively understand the information structure; it can provide various interactive functions, allowing users to explore data, filter views and deeply mine specific information, which helps to discover hidden associations and patterns between data, which is of great significance to intelligence judgment.

[0003] With the development of artificial intelligence technology, ChatGPT has emerged, which has not only achieved breakthroughs in technology, but also brought revolutionary changes in application models and user experience. The large language model has good context memory and semantic understanding capabilities, which can realize the perception and understanding of context, thereby providing more accurate and coherent answers. Traditional applications are often task-driven, and the interaction between users and applications is relatively limited. Applications based on large language models can interact with users more naturally in a conversational manner, which improves the user experience; at the same time, the large language model can understand the user's language habits and preferences, and can provide more personalized services and suggestions, making the application services more in line with user needs; the large language model can process a large amount of natural language input, automate many traditional manual processes, and provide detailed information retrieval and interpretation services. Therefore, text dialogue and question-and-answer will become an increasingly popular form of interaction for application systems with its friendliness and convenience. It can provide a more intelligent and humanized interactive interface through a deep understanding of the user's language, reduce the cost of users learning the operating system, and greatly improve the user experience.

[0004] Although the intelligence visualization analysis system based on knowledge graph can provide an intuitive analysis interface for intelligence analysts, the complexity and inherent uncertainty of intelligence analysis work require the visualization analysis system to have a high degree of flexibility and complexity to adapt to diverse analysis needs. Although this design enhances the functionality of the system, it also means that users need to master specific technical knowledge. The high learning threshold and operation difficulty may have a very negative impact on the user experience. Although the traditional question-and-answer system based on the knowledge graph can also achieve the purpose of interactive question-and-answer, it often lacks the ability to understand the context and cannot give accurate answers based on the context information. It also lacks the ability to conduct multiple rounds of dialogue with users and cannot deeply explore user needs.

[0005] Therefore, it is necessary to provide an intelligence analysis method and system based on knowledge graphs and big models to solve the above technical problems. Summary of the invention

[0006] The present invention provides an intelligence analysis method and system based on knowledge graph and large model, which solves the problem that the learning threshold and operation difficulty in related technologies may have a very bad impact on user experience. Although the traditional question-and-answer system based on the knowledge graph can also achieve the purpose of interactive question-and-answer, it often lacks the ability to understand the context and cannot give accurate answers based on the context information. At the same time, it also lacks the ability to conduct multiple rounds of dialogue with users and cannot deeply explore the problem of user needs.

[0007] In order to solve the above technical problems, the intelligence analysis method based on knowledge graph and large model provided by the present invention comprises the following steps:

[0008] Step S1: intelligence knowledge graph construction;

[0009] Building an intelligence knowledge graph involves multiple links, including ontology modeling, knowledge extraction, knowledge fusion, knowledge storage, etc.

[0010] Ontology modeling: Based on actual business needs, the knowledge graph is modeled in a top-down manner. Entities are classified according to "people, things, cases, events", etc., and "people, things, cases, events" are connected in series through the relationship between entities, so as to achieve the connection from people to people, people to things, things to people, people to cases, people to things, etc., and form an intelligence knowledge graph model;

[0011] Structured data extraction: Based on the ontology model, a mapping relationship is established between structured data and the intelligence knowledge graph model, and a data extraction task is constructed to achieve full or enhanced extraction of structured data;

[0012] Unstructured text extraction: Based on the ontology model, the pre-trained model Bert is used in combination with business rules to extract entities, relationships, attributes, events and other information from intelligence clue texts. Information such as names, ID numbers, and events can be extracted; event information can also be deduplicated and classified;

[0013] Knowledge fusion: Integrate and fuse knowledge from different sources and formats, eliminate duplicate and redundant information, solve semantic differences and consistency issues between different data sources, and ensure the consistency and integrity of knowledge. For example, knowledge extracted from structured data is integrated with knowledge extracted from unstructured text to form an overall graph.

[0014] Storage and indexing of knowledge graphs: Select Neo4j graph data to store the constructed graph information, and create indexes for the graphs to improve query efficiency;

[0015] Step S2: Construction of the technical and tactical model library;

[0016] Based on personnel trajectory data and business rules, we build technical and tactical models such as traveling together, living together, on the same railway, and on the same plane. We provide personnel relationship query viewing functions in the form of interface services, which are integrated with knowledge graph data as dynamic relationships and presented in a unified manner.

[0017] Step S3: fine-tuning the large language model;

[0018] Use instruction fine-tuning technology to fine-tune open source LLMs to enable them to have language understanding capabilities in the field of intelligence analysis, enhance their domain Cypher code generation capabilities and the ability to call specific interface services;

[0019] Data preparation: For intelligence analysis knowledge graphs, collect and prepare data sets specific to knowledge graph queries, including natural language questions and their corresponding Cypher query statements. The scope of the data set must cover all schemas of the constructed knowledge graph. If available data is limited, consider using data enhancement techniques, such as increasing data diversity through grammatical transformation, synonym replacement, etc.

[0020] NL2Cypher training: Based on the prepared data set, the open source LLM is fine-tuned through instruction fine-tuning technology to give it the NL2Cypher capability for specific fields: natural language questions are converted into Cypher, an executable language for Neo4j graph data. The model is trained through supervised learning, with natural language queries as input and Cypher query statements as output. During the training process, the model parameters are optimized by minimizing the error between the prediction and the actual results, so that it can accurately map natural language queries to corresponding Cypher query statements;

[0021] Step S4: analysis assistant;

[0022] An intelligent intelligence analysis assistant is built based on a fine-tuned large language model. Users can use it to process natural language queries, understand user intent, and generate natural language answers, and call the graph visualization analysis engine interface for visualization.

[0023] The queried relationships include static relationships and dynamic relationships. Static relationships are mainly queried from the graph database; dynamic relationships refer to the trajectory relationships of personnel. Whether the relationship is established is closely related to the set conditional rules, which needs to be queried through the technical and tactical service interface;

[0024] Intent recognition: User intent recognition is performed based on a large model to convert the user's natural language query into a structured query representation, including identifying keywords, operators, and query constraints, etc., to determine the user's query intent and perform step splitting to determine whether to query data from the graph database or from the interface service.

[0025] Preferably, firstly, a pre-built technique and tactic library is matched through semantic matching and other technologies. If a certain question is found to be similar to a type of question in the library, such as asking about the people who traveled with or lived with a person during a certain event, the techniques and tactic models in the technique and tactic model library can be called, and based on the key information extracted from the large model, the interface parameters are filled in for query, so that the particularity of the intelligence analysis scenario can be introduced;

[0026] If no matching technique is found, data is queried from the knowledge graph through nl2cypher. Entities (such as names, places, organizations, etc.) are extracted from the user's query. These entities will be used as the starting point of the query to determine the relationship between entities, which may include "is", "belongs to", "is located in" and other relationships. According to the identified entities and relationships, the corresponding Cypher query statement is constructed. The generated query statement may need to be further optimized to ensure that it can be executed effectively and return the correct results. Finally, the generated Cypher query statement is sent to the graph database for execution and the results are obtained;

[0027] The information obtained can be integrated to allow the large model to complete reasoning analysis and answer user questions. The thinking chain method can be used to improve the overall accuracy.

[0028] Finally, based on the result of reasoning, the answer to the question is generated. At the same time, the graph visualization component is called, and the system automatically generates graph information, so that users can intuitively understand the relationship between the answer and the graph. The interactive functions of the graph can be further called, such as allowing users to zoom, filter, search, and switch the graph layout according to attribute values, relationship types and other conditions, to further explore the data.

[0029] An intelligence analysis system based on a knowledge graph and a large model, wherein the intelligence analysis system based on a knowledge graph and a large model is used for an intelligence analysis method based on a knowledge graph and a large model, and is characterized in that it comprises an intelligence analysis chassis and a dust collection box;

[0030] The intelligence analysis chassis is internally installed with a language collection component and a dust removal component;

[0031] The language collection component comprises a back shell, a front shell is installed in front of the back shell by bolts, a notch is provided inside the front shell, a circuit board is installed inside the back shell by bolts, a language collector is installed on the outer wall of the circuit board and inside the front shell, and bayonet pins are fixed at both ends of the back shell;

[0032] The dust removal assembly includes a mounting plate, the outer walls of which are respectively fixed with a dust removal plate and a protective plate, the outer walls of the two mounting plates are fixed with sleeves, the inner bottom of the intelligence analysis chassis is fixed with a positioning plate, and a motor is installed inside the intelligence analysis chassis and on the outer wall of the positioning plate through bolts, the output shaft keyway of the motor is connected to a first gear, and the bottom of the first gear is meshingly connected to a second gear.

[0033] Preferably, the bayonet pin passes through the interior of the sleeve and does not contact the sleeve, and the bayonet pin is fixedly connected to the positioning plate.

[0034] Preferably, the inner wall of the second gear is fixedly connected to the outer wall of the sleeve, and the second gear is rotatably connected to the positioning plate.

[0035] Preferably, the inner walls of the protective plate and the dust removal plate fit with the outer walls of the back shell and the front shell, and the sleeve passes through the mounting plate.

[0036] Preferably, it also includes a cleaning component;

[0037] The cleaning assembly includes a brush roller and an annular pulley, a ratchet is fixedly provided at one end of the brush roller, a ratchet is installed on the inner wall of the annular pulley, a driving pulley is fixedly provided on the outer wall of the sleeve and on one side of the second gear, and a belt is sleeved on the outer walls of the driving pulley and the annular pulley.

[0038] Preferably, the ratchet teeth and the ratchet wheel are meshingly connected, and one end of the brush roller is rotatably connected to the positioning plate.

[0039] Preferably, it also includes a sealing component;

[0040] The sealing assembly comprises a side frame fixedly mounted on the side wall of the positioning plate, an empty bag is installed inside the side frame, a hose is sealedly mounted on the outer wall of the empty bag, a sealing ring is sealed on the outer wall of the front shell and around the notch, and a contact plate is fixedly mounted on the side wall of the mounting plate;

[0041] The outlet end of the hose is sealed with a sealing ring.

[0042] Compared with related technologies, the intelligence analysis method and system based on knowledge graph and large model provided by the present invention have the following beneficial effects:

[0043] Compared with the prior art, the question-answering intelligence analysis method and system of the present invention, by combining knowledge graphs and large language models, provides an accurate, comprehensive, real-time and highly interactive intelligence analysis solution, which is convenient for users to automatically generate analysis results in the field of intelligence analysis through question-answering, and assists in intuitively presenting the basis for analysis in the form of graph visualization, maintaining the continuity and fluency of the analysis ideas, ensuring the intuitiveness and interpretability of the intelligence analysis results, and is conducive to improving the accuracy of intelligence analysis and judgment, and improving the efficiency of intelligence analysis and judgment;

[0044] The present invention is aimed at complex analysis scenarios in the field of intelligence analysis. It makes full use of knowledge graphs to integrate multi-source heterogeneous information, constructs a clear knowledge structure, and forms a unified intelligence analysis system based on the powerful query capabilities of graph databases and the expansion capabilities of technical and tactical models. It provides multi-hop reasoning capabilities and aggregate query capabilities based on graphs. It can also combine the powerful semantic understanding and generation capabilities of large language models, automatically realize the intelligence analysis process based on question-and-answer interaction, assist in providing a knowledge graph visualization display interface, realize an intelligent and user-friendly interactive interface, reduce the cost of users learning the operating system, and greatly improve the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0046] Figure 1 The overall technical framework diagram provided by the present invention;

[0047] Figure 2 Constructing an architecture diagram for the intelligence knowledge graph provided by the present invention;

[0048] Figure 3 The workflow diagram of the analysis assistant provided by the present invention;

[0049] Figure 4 A schematic diagram of the overall structure of the intelligence analysis chassis provided by the present invention;

[0050] Figure 5 for Figure 4 The schematic diagram of the cross-sectional structure of the intelligence analysis chassis shown;

[0051] Figure 6 for Figure 5 The schematic diagram of the split structure of the language collection component and the dust removal component shown;

[0052] Figure 7 for Figure 5 and Figure 6 Detailed structural diagram of dust removal assembly and cleaning assembly shown;

[0053] Figure 8 for Figure 7 The enlarged structural diagram of point A shown;

[0054] Fig. 9 for Figure 7 The front structure schematic diagram shown;

[0055] Fig.10 for Figure 7 and Figure 8 The schematic diagram of the dust removal component and the cleaning component shown;

[0056] Fig.11 for Figure 7 and Figure 8 A schematic diagram of the detailed connection structure of the dust removal component and the cleaning mechanism shown;

[0057] Fig.12 for Figure 5 and Figure 7 Schematic diagram of the sealing component structure shown.

[0058] Description of Figure Numbers:

[0059] 1. Intelligence analysis chassis;

[0060] 2. language collection component, 21. back shell, 22. front shell, 23. notch, 24. circuit board, 25. language collector, 26. latch;

[0061] 3. dust removal assembly, 31. mounting plate, 32. dust removal plate, 33. protection plate, 34. sleeve, 35. contact plate, 36. positioning plate, 37. motor, 38. first gear, 39. second gear;

[0062] 4. cleaning assembly, 41. brush roller, 42. annular pulley, 43. ratchet, 44. belt, 45. ratchet, 46. driving pulley;

[0063] 5. sealing assembly, 51. side frame, 52. empty bag, 53. hose, 54. sealing ring;

[0064] 6. Dust box.

[0065] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0066] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not 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.

[0067] The present invention provides an intelligence analysis method and system based on knowledge graph and large model.

[0068] First embodiment:

[0069] Combination Figures 1 to 3: An intelligence analysis method based on knowledge graph and big model, comprising the following steps:

[0070] Step S1: intelligence knowledge graph construction;

[0071] Building an intelligence knowledge graph involves multiple links, including ontology modeling, knowledge extraction, knowledge fusion, knowledge storage, etc.

[0072] Ontology modeling: Based on actual business needs, the knowledge graph is modeled in a top-down manner. Entities are classified according to "people, things, cases, events", etc., and "people, things, cases, events" are connected in series through the relationship between entities, so as to achieve the connection from people to people, people to things, things to people, people to cases, people to things, etc., and form an intelligence knowledge graph model;

[0073] Structured data extraction: Based on the ontology model, a mapping relationship is established between structured data and the intelligence knowledge graph model, and a data extraction task is constructed to achieve full or enhanced extraction of structured data;

[0074] Unstructured text extraction: Based on the ontology model, the pre-trained model Bert is used in combination with business rules to extract entities, relationships, attributes, events and other information from intelligence clue texts. Information such as names, ID numbers, and events can be extracted; event information can also be deduplicated and classified;

[0075] Knowledge fusion: Integrate and fuse knowledge from different sources and formats, eliminate duplicate and redundant information, solve semantic differences and consistency issues between different data sources, and ensure the consistency and integrity of knowledge. For example, knowledge extracted from structured data is integrated with knowledge extracted from unstructured text to form an overall graph.

[0076] Storage and indexing of knowledge graphs: Select Neo4j graph data to store the constructed graph information, and create indexes for the graphs to improve query efficiency;

[0077] Step S2: Construction of the technical and tactical model library;

[0078] Based on personnel trajectory data and business rules, we build technical and tactical models such as traveling together, living together, on the same railway, and on the same plane. We provide personnel relationship query viewing functions in the form of interface services, which are integrated with knowledge graph data as dynamic relationships and presented in a unified manner.

[0079] Step S3: fine-tuning the large language model;

[0080] Use instruction fine-tuning technology to fine-tune open source LLMs to enable them to have language understanding capabilities in the field of intelligence analysis, enhance their domain Cypher code generation capabilities and the ability to call specific interface services;

[0081] Data preparation: For intelligence analysis knowledge graphs, collect and prepare data sets specific to knowledge graph queries, including natural language questions and their corresponding Cypher query statements. The scope of the data set must cover all schemas of the constructed knowledge graph. If available data is limited, consider using data enhancement techniques, such as increasing data diversity through grammatical transformation, synonym replacement, etc.

[0082] NL2Cypher training: Based on the prepared data set, the open source LLM is fine-tuned through instruction fine-tuning technology to give it the NL2Cypher capability for specific fields: natural language questions are converted into Cypher, an executable language for Neo4j graph data. The model is trained through supervised learning, with natural language queries as input and Cypher query statements as output. During the training process, the model parameters are optimized by minimizing the error between the prediction and the actual results, so that it can accurately map natural language queries to corresponding Cypher query statements;

[0083] Step S4: analysis assistant;

[0084] An intelligent intelligence analysis assistant is built based on a fine-tuned large language model. Users can use it to process natural language queries, understand user intent, and generate natural language answers, and call the graph visualization analysis engine interface for visualization.

[0085] The queried relationships include static relationships and dynamic relationships. Static relationships are mainly queried from the graph database; dynamic relationships refer to the trajectory relationships of personnel. Whether the relationship is established is closely related to the set conditional rules, which needs to be queried through the technical and tactical service interface;

[0086] Intent recognition: User intent recognition is performed based on a large model to convert the user's natural language query into a structured query representation, including identifying keywords, operators, and query constraints, etc., to determine the user's query intent and perform step splitting to determine whether to query data from the graph database or from the interface service.

[0087] First, semantic matching and other technologies are used to match the pre-built technical and tactical method library. If a question is found to be similar to a type of question in the library, such as asking about the people who traveled with or lived with someone during a certain event, the technical and tactical method in the technical and tactical method model library can be called. Based on the key information extracted from the large model, the interface parameters are filled in for query. This way, the particularity of the intelligence analysis scenario can be introduced.

[0088] If no matching technique is found, data is queried from the knowledge graph through nl2cypher. Entities (such as names, places, organizations, etc.) are extracted from the user's query. These entities will be used as the starting point of the query to determine the relationship between entities, which may include "is", "belongs to", "is located in" and other relationships. According to the identified entities and relationships, the corresponding Cypher query statement is constructed. The generated query statement may need to be further optimized to ensure that it can be executed effectively and return the correct results. Finally, the generated Cypher query statement is sent to the graph database for execution and the results are obtained;

[0089] The information obtained can be integrated to allow the large model to complete reasoning analysis and answer user questions. The thinking chain method can be used to improve the overall accuracy.

[0090] Finally, based on the result of reasoning, the answer to the question is generated. At the same time, the graph visualization component is called, and the system automatically generates graph information, so that users can intuitively understand the relationship between the answer and the graph. The interactive functions of the graph can be further called, such as allowing users to zoom, filter, search, and switch the graph layout according to attribute values, relationship types and other conditions, to further explore the data.

[0091] This embodiment: The method and system can make full use of the knowledge graph to integrate multi-source heterogeneous information, build a clear knowledge structure, and provide graph-based multi-hop reasoning capabilities based on the powerful query capabilities of the graph database; it can also combine the powerful semantic understanding and generation capabilities of the large language model, automate the intelligence analysis process based on the question-answering interaction method, and assist in providing a knowledge graph visualization display interface to help users better understand complex information.

[0092] Second embodiment:

[0093] See also Figures 4 to 11 , an intelligence analysis system based on a knowledge graph and a large model, the intelligence analysis system based on a knowledge graph and a large model is used for an intelligence analysis method of a knowledge graph and a large model, comprising an intelligence analysis chassis 1 and a dust collection box 6;

[0094] The intelligence analysis chassis 1 is internally installed with a language collection component 2 and a dust removal component 3;

[0095] The language collection component 2 includes a back shell 21, a front shell 22 is installed in front of the back shell 21 by bolts, a notch 23 is opened inside the front shell 22, a circuit board 24 is installed inside the back shell 21 by bolts, a language collector 25 is installed on the outer wall of the circuit board 24 and inside the front shell 22, and bayonet pins 26 are fixed at both ends of the back shell 21;

[0096] The dust removal assembly 3 includes a mounting plate 31, the outer walls of which are respectively fixed with a dust removal plate 32 and a protective plate 33, the outer walls of the two mounting plates 31 are fixed with a sleeve 34, a positioning plate 36 is fixed to the inner bottom of the intelligence analysis chassis 1, and a motor 37 is installed inside the intelligence analysis chassis 1 and on the outer wall of the positioning plate 36 by bolts, the output shaft keyway of the motor 37 is connected to a first gear 38, and the bottom of the first gear 38 is meshingly connected to a second gear 39.

[0097] See also Figure 1 : The workflow and implementation methods in the first embodiment are all implemented in the intelligence analysis chassis 1;

[0098] See also Figure 6 : When the user collects the original voice of the target person, the user needs to input the language of the target person through the language collector 25. The specific installation method is as follows:

[0099] The circuit board 24 with the language collector 25 installed and the back shell 21 are installed by bolts, and after the installation is completed, the front shell 22 and the back shell 21 need to be installed;

[0100] After the installation is completed, the target person can speak directly to the slot 23 , and the sound will enter the slot 23 through the front dust removal plate 32 , and finally the voice of the target person will be collected by the language collector 25 .

[0101] See also Figure 7 , Fig.10 and Fig.11 : After long-term use, due to the large number of target personnel in the public security system and the large amount of voice input, there will definitely be external dust and floating objects sticking to the dust removal plate 32. In this case, the dust removal plate 32 needs to be processed. During processing, the user can start the motor 37 to drive the first gear 38 to rotate. When the first gear 38 rotates, it will engage and control the second gear 39 to rotate. When the second gear 39 rotates, it will control the sleeve 34 to drive the mounting plate 31 to rotate. The mounting plate 31 can control the protective plates 33 and the dust removal plates 32 on both sides to rotate 180 degrees clockwise, so that the protective plates 33 rotate to the initial dust removal plates 32, and the dust removal plates 32 rotate to the position of the initial protective plates 33.

[0102] The bayonet pin 26 passes through the interior of the sleeve 34 and does not contact the sleeve 34 . The bayonet pin 26 is fixedly connected to the positioning plate 36 .

[0103] The inner wall of the second gear 39 is fixedly connected to the outer wall of the sleeve 34 , and the second gear 39 is rotatably connected to the positioning plate 36 .

[0104] The inner walls of the protection plate 33 and the dust removal plate 32 are in contact with the outer walls of the back shell 21 and the front shell 22 , and the sleeve 34 passes through the mounting plate 31 .

[0105] Understandable: From Fig.11 It can be seen that, since the bayonet 26 and the sleeve 34 do not interfere with each other, the second gear 39 will not affect the bayonet 26 in the fixed state during the rotation. If the bayonet 26 is not affected, the entire language acquisition component 2 will not be interfered.

[0106] This embodiment: Compared with the traditional language collection design, this case is used in the public security system, so the number of target personnel must be large, and the more people there are, the greater the workload of recording voice. Therefore, under long-term use, the surface of the language collector 25 will be contaminated with a lot of dust and external floating objects. Then the entire language collection component 2 is designed in a circular shape, and is divided into a protective plate 33 and a dust removal plate 32. The inner walls of the protective plate 33 and the dust removal plate 32 are also designed as arcs. At the same time, the curvature of the arc and the circular language collection component 2 are on the same axis, so that the dust removal plate 32 and the protective plate 33 can be continuously attached to the outer wall of the language collection component 2. Therefore, such a design not only removes dust and floating objects under long-term use, but also protects the language collector 25 during the processing process.

[0107] Also includes a cleaning component 4;

[0108] The cleaning assembly 4 includes a brush roller 41 and an annular pulley 42, a ratchet 43 is fixedly provided at one end of the brush roller 41, a ratchet 45 is installed on the inner wall of the annular pulley 42, a driving pulley 46 is fixedly provided on the outer wall of the sleeve 34 and on one side of the second gear 39, and a belt 44 is sleeved on the outer wall of the driving pulley 46 and the annular pulley 42.

[0109] The ratchet teeth 45 and the ratchet wheel 43 are meshedly connected, and one end of the brush roller 41 is rotatably connected to the positioning plate 36 .

[0110] See also Fig.10 and Fig.11 : Due to the provision of the ratchet 43 and the ratchet 45, when the dust removal plate 32 rotates clockwise, it will drive the driving pulley 46 on the outer wall of the sleeve 34 to rotate clockwise, and the clockwise rotating driving pulley 46 will drive the control belt 44 to drive the annular pulley 42 to rotate clockwise. At this time, the clockwise rotating annular pulley 42 will control the ratchet 45 to drive the ratchet 43 to rotate clockwise, and the rotating ratchet 43 will drive the brush roller 41 to rotate clockwise;

[0111] See also Fig. 9: The dust removal plate 32 in the rotating process will contact the brush roller 41 in the automatic rotating state, and the rotating brush roller 41 will contact and roll on the outer wall of the dust removal plate 32, thereby rolling the dust and impurities on the dust removal plate 32 into the dust collection box 6 for collection;

[0112] See also Fig.10 : When the cleaning is completed, the dust removal plate 32 is in the process of resetting, and the belt 44 controls the annular pulley 42 to rotate counterclockwise. The counterclockwise rotating annular pulley 42 will control the internal ratchet 45 to avoid on the ratchet 43. In this state, the brush roller 41 stops working, so there is no dust on the outer wall of the dust removal plate 32, so the brush roller 41 does not need to work.

[0113] This embodiment: Compared with the traditional design, in this case, when the dust removal plate 32 rotates clockwise, the rotating dust removal plate 32 will control the brush roller 41 at the bottom to rotate, the dust removal plate 32 and the brush roller 41 will contact each other, and the brush roller 41 will roll and brush the dust in the dust removal plate 32. Therefore, this design can automatically and linkedly remove stubborn dust and impurities. At the same time, the brush roller 41 will stop rotating during the resetting process. This design further increases the service life of the brush roller 41 and can avoid excessive repetitive work of the brush roller 41.

[0114] Third embodiment:

[0115] See also Fig.12 , further comprising a sealing component 5;

[0116] The sealing assembly 5 includes a side frame 51 fixed to the side wall of the positioning plate 36, a hollow bag 52 is installed inside the side frame 51, a hose 53 is sealed on the outer wall of the hollow bag 52, a sealing ring 54 is sealed on the outer wall of the front shell 22 and around the notch 23, and a contact plate 35 is fixed on the side wall of the mounting plate 31;

[0117] The outlet end of the hose 53 is sealed with the sealing ring 54 .

[0118] See also Fig.12 : In the second embodiment, when the brush roller 41 cleans the dust removal plate 32, the rotation of the dust removal plate 32 will definitely affect the rotation of the protective plate 33. When the protective plate 33 rotates, it will drive the resistance plate 35 to rotate. If the protective plate 33 rotates to a horizontal position, the resistance plate 35 will rotate to the inside of the side frame 51. The arc-shaped resistance plate 35 will resist the squeezing control air bag 52 for compression. The air bag 52 compresses the internal gas into the sealing ring 54 through the hose 53. The gas will control the sealing ring 54 to expand, and the expanded sealing ring 54 will fill the inner wall of the closed protective plate 33.

[0119] It can be understood that: since the hose 53 is made of flexible material and the rotation trajectory of the protective plate 33 is only from left to right, the hose 53 can avoid the mounting plate 31 after the rotation is successful to avoid interference. Secondly, the contact plate 35 is longer, and the empty bag 52 inside the side frame 51 will not be affected during the rotation of the dust removal plate 32.

[0120] In this embodiment, during the process of the first embodiment, the dust removal plate 32 will drive the protective plate 33 to rotate in conjunction during the rotation and cleaning process, and during the rotation process, the resistance plate 35 outside the protective plate 33 will resist and squeeze the air bag 52 to compress the gas to control the expansion of the sealing ring 54. The expanded sealing ring 54 will quickly fill the gap between the protective plate 33 and the slot 23, thereby further improving the sealing effect, achieving excellent waterproof and dustproof performance, and better protecting the language collection component 2.

[0121] Please refer to Figures 1 to 12 The working principle of the intelligence analysis method and system based on knowledge graph and big model provided by the present invention is as follows:

[0122] Step S1, the public security system collects the voice of the target person;

[0123] The target person can speak directly to the slot 23, and the sound will enter the slot 23 through the dust removal plate 32 at the front, and finally the speech of the target person will be collected by the language collector 25;

[0124] Through semantic matching and other technologies, the pre-built technical and tactical method library is matched. If a certain question is found to be similar to a type of question in the library, such as asking about the people who traveled with or lived with a person during a certain event, the technical and tactical method in the technical and tactical method model library can be called. Based on the key information extracted from the large model, the interface parameters can be filled in for query. In this way, the particularity of the intelligence analysis scenario can be introduced;

[0125] If no matching technique is found, data is queried from the knowledge graph through nl2cypher. Entities (such as names, places, organizations, etc.) are extracted from the user's query. These entities will be used as the starting point of the query to determine the relationship between entities, which may include "is", "belongs to", "is located in" and other relationships. According to the identified entities and relationships, the corresponding Cypher query statement is constructed. The generated query statement may need to be further optimized to ensure that it can be executed effectively and return the correct results. Finally, the generated Cypher query statement is sent to the graph database for execution and the results are obtained;

[0126] After long-term use, due to the large number of target personnel in the public security system and the large amount of voice input, there will definitely be external dust and floating objects sticking to the dust removal plate 32. In this case, the dust removal plate 32 needs to be processed. During the processing, the user can start the motor 37 to drive the first gear 38 to rotate. When the first gear 38 rotates, it will mesh and control the second gear 39 to rotate. When the second gear 39 rotates, it will control the sleeve 34 to drive the mounting plate 31 to rotate. The mounting plate 31 can control the protective plates 33 and the dust removal plates 32 on both sides to rotate 180 degrees clockwise, so that the protective plates 33 rotate to the initial dust removal plates 32, and the dust removal plates 32 rotate to the position of the initial protective plates 33;

[0127] Step S2, since the ratchet 43 and the ratchet 45 are provided, when the dust removal plate 32 rotates clockwise, the driving pulley 46 on the outer wall of the sleeve 34 will be driven to rotate clockwise, and the clockwise rotating driving pulley 46 will drive the control belt 44 to drive the annular pulley 42 to rotate clockwise, and at this time, the clockwise rotating annular pulley 42 will control the ratchet 45 to drive the ratchet 43 to rotate clockwise, and the rotating ratchet 43 will drive the brush roller 41 to rotate clockwise;

[0128] The dust removal plate 32 in the rotating process will contact the brush roller 41 in the automatic rotating state, and the rotating brush roller 41 will contact and roll on the outer wall of the dust removal plate 32, thereby rolling the dust and impurities on the dust removal plate 32 into the dust collecting box 6 for collection.

[0129] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. All equivalent structural changes made by using the contents of the present invention specification and drawings under the concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. An intelligence analysis method based on knowledge graph and big model, characterized in that: The following steps are involved: Step S1: intelligence knowledge graph construction; Building an intelligence knowledge graph involves multiple links, including ontology modeling, knowledge extraction, knowledge fusion, knowledge storage, etc. Ontology modeling: Based on actual business needs, the knowledge graph is modeled in a top-down manner. Entities are classified according to "people, things, cases, events", etc., and "people, things, cases, events" are connected in series through the relationship between entities, so as to realize the connection from people to people, people to things, things to people, people to cases, people to things, etc., and form an intelligence knowledge graph model; Structured data extraction: Based on the ontology model, a mapping relationship is established between structured data and the intelligence knowledge graph model, and a data extraction task is constructed to achieve full or enhanced extraction of structured data; Unstructured text extraction: Based on the ontology model, the pre-trained model Bert is used in combination with business rules to extract entities, relationships, attributes, events and other information from intelligence clue texts. Information such as names, ID numbers, events, etc. can be extracted; And deduplicate and classify event information; Knowledge fusion: Integrate and fuse knowledge from different sources and formats, eliminate duplicate and redundant information, solve semantic differences and consistency issues between different data sources, and ensure the consistency and integrity of knowledge. For example, knowledge extracted from structured data is integrated with knowledge extracted from unstructured text to form an overall graph. Storage and indexing of knowledge graphs: Select Neo4j graph data to store the constructed graph information, and create indexes for the graphs to improve query efficiency; Step S2: Construction of the technical and tactical model library; Based on personnel trajectory data and business rules, we build technical and tactical models such as traveling together, living together, on the same railway, and on the same plane. We provide personnel relationship query viewing functions in the form of interface services, which are integrated with knowledge graph data as dynamic relationships and presented in a unified manner. Step S3: fine-tuning the large language model; Use instruction fine-tuning technology to fine-tune open source LLMs to enable them to have language understanding capabilities in the field of intelligence analysis, enhance their domain Cypher code generation capabilities and the ability to call specific interface services; Data preparation: For intelligence analysis knowledge graphs, collect and prepare data sets specific to knowledge graph queries, including natural language questions and their corresponding Cypher query statements. The scope of the data set must cover all schemas of the constructed knowledge graph. If available data is limited, consider using data enhancement techniques, such as increasing data diversity through grammatical transformation, synonym replacement, etc. NL2Cypher training: Based on the prepared data set, the open source LLM is fine-tuned through instruction fine-tuning technology to give it the NL2Cypher capability for specific fields: natural language questions are converted into Cypher, an executable language for Neo4j graph data. The model is trained through supervised learning, with natural language queries as input and Cypher query statements as output. During the training process, the model parameters are optimized by minimizing the error between the prediction and the actual results, so that it can accurately map natural language queries to corresponding Cypher query statements; Step S4: analysis assistant; An intelligent intelligence analysis assistant is built based on a fine-tuned large language model. Users can use it to process natural language queries, understand user intent, and generate natural language answers, and call the graph visualization analysis engine interface for visualization. The queried relationships include static relationships and dynamic relationships. Static relationships are mainly queried from the graph database; dynamic relationships refer to the trajectory relationships of personnel. Whether the relationship is established is closely related to the set conditional rules, which needs to be queried through the technical and tactical service interface; Intent recognition: User intent recognition is performed based on a large model to convert the user's natural language query into a structured query representation, including identifying keywords, operators, and query constraints, etc., to determine the user's query intent and perform step splitting to determine whether to query data from the graph database or from the interface service.

2. The intelligence analysis method based on knowledge graph and large model according to claim 1 is characterized in that: First, semantic matching and other technologies are used to match the pre-built technical and tactical method library. If a question is found to be similar to a type of question in the library, such as asking about the people who traveled with or lived with someone during a certain event, the technical and tactical method in the technical and tactical method model library can be called. Based on the key information extracted from the large model, the interface parameters are filled in for query. This way, the particularity of the intelligence analysis scenario can be introduced. If no matching techniques are found, data is queried from the knowledge graph through nl2cypher. Entities (such as names, places, organizations, etc.) are extracted from the user's query. These entities will be used as the starting point of the query to determine the relationship between entities, which may include "is", "belongs to", "is located in" and other relationships. According to the identified entities and relationships, the corresponding Cypher query statement is constructed. The generated query statement may need to be further optimized to ensure that it can be executed effectively and return the correct results. Finally, the generated Cypher query statement is sent to the graph database for execution and the results are obtained; The information obtained can be integrated to allow the large model to complete reasoning analysis and answer user questions. The thinking chain method can be used to improve the overall accuracy. Finally, based on the result of reasoning, the answer to the question is generated. At the same time, the graph visualization component is called, and the system automatically generates graph information, so that users can intuitively understand the relationship between the answer and the graph. The interactive functions of the graph can be further called, such as allowing users to zoom, filter, search, and switch the graph layout according to attribute values, relationship types and other conditions, to further explore the data.

3. An intelligence analysis system based on a knowledge graph and a large model, wherein the intelligence analysis system based on a knowledge graph and a large model is used for the intelligence analysis method of the knowledge graph and the large model as claimed in any one of claims 1-2, characterized in that: Includes intelligence analysis chassis and dust collection box; The intelligence analysis chassis is internally installed with a language collection component and a dust removal component; The language collection component comprises a back shell, a front shell is installed in front of the back shell by bolts, a notch is provided inside the front shell, a circuit board is installed inside the back shell by bolts, a language collector is installed on the outer wall of the circuit board and inside the front shell, and bayonet pins are fixed at both ends of the back shell; The dust removal assembly includes a mounting plate, the outer walls of which are respectively fixed with a dust removal plate and a protective plate, the outer walls of the two mounting plates are fixed with sleeves, the inner bottom of the intelligence analysis chassis is fixed with a positioning plate, and a motor is installed inside the intelligence analysis chassis and on the outer wall of the positioning plate through bolts, the output shaft keyway of the motor is connected to a first gear, and the bottom of the first gear is meshingly connected to a second gear.

4. The intelligence analysis system based on knowledge graph and large model according to claim 3 is characterized in that: The bayonet pin passes through the interior of the sleeve and does not contact the sleeve, and the bayonet pin is fixedly connected to the positioning plate.

5. The intelligence analysis system based on knowledge graph and large model according to claim 3 is characterized in that: The inner wall of the second gear is fixedly connected to the outer wall of the sleeve, and the second gear is rotatably connected to the positioning plate.

6. The intelligence analysis system based on knowledge graph and large model according to claim 3 is characterized in that: The inner walls of the protective plate and the dust removal plate are fitted with the outer walls of the back shell and the front shell, and the sleeve passes through the mounting plate.

7. The intelligence analysis system based on knowledge graph and large model according to claim 3 is characterized in that: Also includes a cleanup component; The cleaning assembly includes a brush roller and an annular pulley, a ratchet is fixedly provided at one end of the brush roller, a ratchet is installed on the inner wall of the annular pulley, a driving pulley is fixedly provided on the outer wall of the sleeve and on one side of the second gear, and a belt is sleeved on the outer walls of the driving pulley and the annular pulley.

8. The intelligence analysis system based on knowledge graph and large model according to claim 7 is characterized in that: The ratchet teeth and the ratchet wheel are meshed and connected, and one end of the brush roller is rotatably connected to the positioning plate.

9. The intelligence analysis system based on knowledge graph and large model according to claim 3 is characterized in that: Also included is a sealing assembly; The sealing assembly comprises a side frame fixedly mounted on the side wall of the positioning plate, an empty bag is installed inside the side frame, a hose is sealedly mounted on the outer wall of the empty bag, a sealing ring is sealed on the outer wall of the front shell and around the notch, and a contact plate is fixedly mounted on the side wall of the mounting plate; The outlet end of the hose is sealed with a sealing ring.

Images