A traffic knowledge graph construction method, device, equipment and medium

Abstract

The application relates to the technical field of intelligent transportation, and in particular to a traffic knowledge graph construction method and device, equipment and a medium. The application extracts preset-dimension features of traffic basic data to obtain multi-dimensional feature information; extracts triplets from traffic accident data to obtain triplet information; performs text matching on address data to obtain a first matching result; performs place name matching on the address data and a plurality of address information in a standard address library to obtain a second matching result; respectively performs standard address coding and place name location matching on the first matching result and the second matching result to obtain a first place name location similarity corresponding to the first matching result and a second place name location similarity corresponding to the second matching result; performs similarity calculation on the first place name location similarity and the second place name location similarity to obtain an address matching result; and constructs a traffic knowledge graph according to the multi-dimensional feature information, the triplet information and the address matching result.

Description

Technical Field

[0001] This application relates to the field of intelligent transportation technology, and in particular to a method, apparatus, device and medium for constructing a traffic knowledge graph. Background Technology

[0002] Traffic accidents, as complex system failure events, are often the result of the dynamic coupling of multiple factors, including people, vehicles, roads, environment, and management. Traditional single-factor analysis methods are insufficient to fully reveal the causal mechanisms of accidents, while knowledge graphs constructed from multi-dimensional features can provide data-driven decision-making support for road safety management.

[0003] Existing technologies for cross-modal data fusion and knowledge graph construction targeting multi-dimensional features of traffic accidents have the following shortcomings:

[0004] On the one hand, traditional methods rely on NLP (Natural Language Processing) named entity extraction and rule-based syntactic dependency extraction, which have weak understanding of complex text, resulting in low accuracy in entity relation extraction. On the other hand, the lack of systematic multimodal data standardization processing and risk association methods makes it difficult to achieve accurate identification and dynamic management of high-risk groups, road sections, etc.

[0005] Therefore, how to accurately extract entity relationships and construct a multidimensional knowledge graph between entities has become an urgent problem to be addressed. Summary of the Invention

[0006] This application provides a method, apparatus, device, and medium for constructing a traffic knowledge graph, which can accurately extract entity relationships and construct a multi-dimensional traffic knowledge graph between entities.

[0007] Firstly, this application provides a method for constructing a traffic knowledge graph, the method comprising:

[0008] Acquire traffic data of multiple data types; these data types include address data, basic traffic data, and traffic accident data.

[0009] Feature extraction of preset dimensions is performed on the traffic basic data to obtain multi-dimensional feature information of the traffic basic data; triple extraction is performed on the traffic accident data using a fine-tuned large model to obtain triple information;

[0010] The address data is subjected to text matching to obtain a first matching result; the address data is matched with multiple address information in the standard address database by place name to obtain a second matching result; the first matching result and the second matching result are respectively subjected to standard address encoding and place name location matching to obtain a first place name location similarity corresponding to the first matching result and a second place name location similarity corresponding to the second matching result; the first place name location similarity and the second place name location similarity are calculated to obtain an address matching result;

[0011] A traffic knowledge graph is constructed based on the multidimensional feature information, the triplet information, and the address matching results.

[0012] In one possible implementation, the triple extraction of the traffic accident data using a fine-tuned large model to obtain triple information includes:

[0013] The traffic accident data is parsed and entity recognition is performed to obtain multiple entities corresponding to the traffic accident data.

[0014] Relationships are extracted from multiple entities corresponding to the traffic accident data to obtain the causal relationships between the multiple entities;

[0015] The causal relationships between the multiple entities are constructed into triples to obtain triple information; the triple information includes the accident result, the accident address, and the accident cause.

[0016] In one possible implementation, the method further includes:

[0017] Perform semantic parsing on the incident address to extract the target address elements from the incident address;

[0018] The target address element is matched with multiple address information in the standard address database to obtain the confidence level corresponding to each of the multiple address information;

[0019] Address information with a confidence level greater than a preset threshold among the multiple address information is used as standard address information; and the standard address information is used as the accident address in the triplet information.

[0020] In one possible implementation, the method further includes:

[0021] Obtain traffic flow data of road status type;

[0022] The vehicle flow data is filtered and spatiotemporally unified to obtain preprocessed vehicle flow data;

[0023] Road condition indicators are extracted from the preprocessed traffic flow data to obtain road condition indicators;

[0024] A traffic knowledge graph is constructed based on the multidimensional feature information, the triplet information, the address matching results, and the road status indicators.

[0025] In one possible implementation, the method further includes:

[0026] Retrieve view data for road anomaly data types;

[0027] The view data is subjected to road anomaly feature extraction to obtain road anomaly features;

[0028] A traffic knowledge graph is constructed based on the multidimensional feature information, the triplet information, the address matching results, and the road anomaly features.

[0029] In one possible implementation, the step of performing text matching on the address data to obtain a first matching result includes:

[0030] The address data is segmented into multiple words;

[0031] For each word, the word frequency in the address data is determined based on the number of times the word appears in the address data and the number of words in the address data; the inverse document frequency of the word is determined based on the number of documents in the address data that include the word.

[0032] The first matching result is determined based on the word frequency and inverse document frequency of the multiple words in the address data.

[0033] In one possible implementation, the step of matching the address data with multiple address information in a standard address database to obtain a second matching result includes:

[0034] The addresses in the address data are split to determine the special name string and the common name string;

[0035] The special name string and the special name strings of multiple address information in the standard address database are matched for similarity to determine the special name similarity between the special name string in the address data and the special name string of each address information.

[0036] The common name string and the common name strings of multiple address information in the standard address database are matched for similarity to determine the common name similarity between the common name string in the address data and the common name string of each address information.

[0037] The second matching result is determined based on the proper name similarity and the common name similarity.

[0038] Secondly, this application provides an apparatus for constructing a traffic knowledge graph, the apparatus comprising:

[0039] The data acquisition module is used to acquire traffic data of multiple data types, including address data, basic traffic data, and traffic accident data.

[0040] The feature extraction module is used to extract features of a preset dimension from the traffic basic data to obtain multi-dimensional feature information of the traffic basic data;

[0041] The triplet extraction module is used to extract triplet information from the traffic accident data using a finely tuned large model.

[0042] The address matching module is used to perform text matching on the address data to obtain a first matching result; to perform place name matching on the address data with multiple address information in a standard address database to obtain a second matching result; to perform standard address encoding and place name location matching on the first matching result and the second matching result respectively to obtain a first place name location similarity corresponding to the first matching result and a second place name location similarity corresponding to the second matching result; and to perform similarity calculation on the first place name location similarity and the second place name location similarity to obtain an address matching result.

[0043] The graph construction module is used to construct a traffic knowledge graph based on the multidimensional feature information, the triplet information, and the address matching results.

[0044] Thirdly, this application also provides an electronic device including a processor, which executes a computer program stored in a memory to implement the steps of the traffic knowledge graph construction method as described above.

[0045] Fourthly, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the traffic knowledge graph construction method as described above.

[0046] Fifthly, this application provides a computer program product, including a computer program: when the computer program is executed by a processor, it implements the method for constructing a traffic knowledge graph as described in the first aspect above.

[0047] The technical solutions provided by the embodiments of this application bring at least the following beneficial effects:

[0048] In this embodiment, feature extraction of traffic infrastructure data with preset dimensions is performed to obtain multi-dimensional feature information of the traffic infrastructure data; triple extraction is performed on traffic accident data using a fine-tuned large model to obtain triple information; text matching is performed on address data to obtain a first matching result; place name matching is performed on address data with multiple address information in the standard address database to obtain a second matching result; standard address encoding and place name location matching are performed on the first matching result and the second matching result respectively to obtain the first place name location similarity corresponding to the first matching result and the second place name location similarity corresponding to the second matching result; similarity calculation is performed on the first place name location similarity and the second place name location similarity to obtain the address matching result; based on the multi-dimensional feature information, triple information and address matching results, a traffic knowledge graph is constructed.

[0049] Therefore, this application employs a large model to extract information from traffic data, enabling entity extraction of causal relationships within complex text timeframes. Based on a hybrid matching approach using multiple algorithms, it first performs text matching and place name matching on address data. Then, it conducts similarity analysis based on the matching results of text and place name matching, along with the place name location matching results, to obtain address matching results. Finally, it fuses the extracted entities and address matching results to construct a traffic knowledge graph. This knowledge graph, built from multi-dimensional features, can comprehensively reveal the causal mechanisms of accidents and provide data-driven decision-making support for road safety management.

[0050] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0051] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application.

[0052] Figure 1 A structural diagram illustrating a method for constructing a traffic knowledge graph, as provided in an embodiment of this application.

[0053] Figure 2 A schematic diagram of a traffic knowledge graph provided in an embodiment of this application;

[0054] Figure 3 A flowchart illustrating a method for constructing a traffic knowledge graph, as provided in an embodiment of this application;

[0055] Figure 4A flowchart illustrating a method for obtaining triplet information provided in an embodiment of this application;

[0056] Figure 5 A flowchart illustrating the standardization of accident addresses is provided for an embodiment of this application.

[0057] Figure 6 A mesh diagram illustrating an address composition structure provided in an embodiment of this application;

[0058] Figure 7 A flowchart illustrating a text matching method provided in an embodiment of this application;

[0059] Figure 8 A flowchart illustrating a place name matching method provided in an embodiment of this application;

[0060] Figure 9 A flowchart illustrating a method for processing vehicle traffic flow data provided in an embodiment of this application;

[0061] Figure 10 A flowchart illustrating a view data processing method provided in an embodiment of this application;

[0062] Figure 11 A schematic diagram of a device for constructing a traffic knowledge graph provided in an embodiment of this application;

[0063] Figure 12 This is a schematic diagram of an electronic device structure provided in an embodiment of this application. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0065] It should be noted that the terms "comprising" and "having" and their variations used in this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0066] The terms "first" and "second" used in this document are for descriptive purposes only and should not be construed as indicating relative importance or implying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0067] The word “exemplary” as used below means “serving as an example, embodiment, or illustration.” Any embodiment illustrated as an “exemplary” need not be construed as superior to or better than other embodiments.

[0068] Traffic accidents, as complex system failure events, are often the result of the dynamic coupling of multiple factors, including people, vehicles, roads, environment, and management. Traditional single-factor analysis methods are insufficient to fully reveal the causal mechanisms of accidents, while knowledge graphs constructed from multi-dimensional features can provide data-driven decision-making support for road safety management.

[0069] Existing technologies for cross-modal data fusion and knowledge graph construction targeting multi-dimensional features of traffic accidents have the following shortcomings:

[0070] On the one hand, traditional methods rely on NLP named entity extraction and rule-based syntactic dependency extraction, which have weak understanding of complex text, resulting in low accuracy in entity relationship extraction. On the other hand, the lack of systematic multimodal data standardization processing and risk association methods makes it difficult to achieve accurate identification and dynamic management of high-risk groups, road sections, etc.

[0071] Therefore, how to accurately extract entity relationships and construct a multidimensional knowledge graph between entities has become an urgent problem to be addressed.

[0072] In view of this, this application provides a method, apparatus, device and medium for constructing a traffic knowledge graph, which can accurately extract entity relationships and construct a multidimensional traffic knowledge graph between entities.

[0073] The inventive concept of this application can be summarized as follows: Feature extraction of traffic basic data with preset dimensions is performed to obtain multi-dimensional feature information of the traffic basic data; triple extraction is performed on traffic accident data using a fine-tuned large model to obtain triple information; text matching is performed on address data to obtain a first matching result; place name matching is performed between the address data and multiple address information in a standard address database to obtain a second matching result; standard address encoding and place name location matching are performed on the first and second matching results respectively to obtain a first place name location similarity corresponding to the first matching result and a second place name location similarity corresponding to the second matching result; similarity calculation is performed on the first and second place name location similarities to obtain an address matching result; and a traffic knowledge graph is constructed based on the multi-dimensional feature information, triple information, and address matching results.

[0074] Therefore, this application employs a large model to extract information from traffic data, enabling entity extraction of causal relationships within complex text timeframes. Based on a hybrid matching approach using multiple algorithms, it first performs text matching and place name matching on address data. Then, it conducts similarity analysis based on the matching results of text and place name matching, along with the place name location matching results, to obtain address matching results. Finally, it fuses the extracted entities and address matching results to construct a traffic knowledge graph. This knowledge graph, built from multi-dimensional features, can comprehensively reveal the causal mechanisms of accidents and provide data-driven decision-making support for road safety management.

[0075] The method for constructing a traffic knowledge graph provided in this application can be applied to electronic devices, such as PCs, mobile terminals, terminal devices, and servers. Furthermore, the method for constructing a traffic knowledge graph provided in this application can be applied to distributed software platforms, such as blockchain.

[0076] After introducing the main inventive ideas of the embodiments of this application, combined with Figure 1 The process of constructing the traffic knowledge graph provided in this application is explained.

[0077] like Figure 1 As shown, the method for constructing the traffic knowledge graph in this application includes the following steps:

[0078] In this application, in order to achieve cross-modal data fusion and knowledge graph construction, based on the multi-dimensional characteristics of traffic accidents involving "people-vehicles-roads-environment-events", the data types of traffic data are mainly divided into: address data, basic traffic data, traffic accident data, road status data type, and road anomaly data type.

[0079] Different processing methods are applied to traffic data of different types. First, the traffic data of different types is standardized. Then, multi-dimensional features, causal relationships, and entities are extracted. Finally, a graph of relationships between entities is constructed. The specific process is as follows: Figure 1 As shown:

[0080] For basic traffic data such as drivers and vehicles, data processing and feature extraction of preset dimensions are first performed to obtain multi-dimensional feature information of the basic traffic data; then, a basic map is constructed based on the multi-dimensional feature information.

[0081] For traffic accident data such as police reports, the system first uses a fine-tuned large model to extract triples to obtain triple information; then it standardizes the accident addresses in the triple information; and finally, it constructs an accident map based on the triple information.

[0082] For address data, the road address data is first standardized to obtain multiple address information in the standard address database; then, the address data is analyzed for intersection relationships; finally, an address relationship map is constructed based on the intersection relationship analysis.

[0083] For traffic flow data of the road status type, we first perform spatiotemporal unification, then extract road status indicators to obtain road status indicators; finally, we construct an address-road status map based on the road status indicators and the unified spatiotemporal data.

[0084] For view data of road anomaly type, firstly, a large model is used to extract road anomaly features to obtain road anomaly features; then, the addresses of the road anomaly features are standardized; finally, an address-road anomaly map is constructed based on the road anomaly features and addresses.

[0085] Finally, the basic knowledge graph, accident knowledge graph, address relationship knowledge graph, address-road status knowledge graph, and address-road anomaly knowledge graph are integrated to construct the final traffic knowledge graph.

[0086] This application describes the different types of traffic data required and the features to be extracted. Specific traffic data information is shown in Table 1.

[0087] Table 1

[0088]

[0089] Based on the data types and extracted features in Table 1 above, entities are constructed. The entity information to be constructed is shown in Table 2:

[0090] Table 2

[0091]

[0092] The core relationships between the entities constructed in Table 2 are shown in Table 3:

[0093] Table 3

[0094]

[0095] according to Figure 1 The process shown, the entity information in Table 2, and the core relationships between entities in Table 3 are used to construct the final traffic knowledge graph.

[0096] The specific traffic knowledge graph constructed is as follows: Figure 2 As shown. Correlation between traffic events and roads is extracted from traffic accident data such as police reports, for example, illegal parking causing traffic congestion. Potential hidden dangers at roads or intersections are analyzed based on view data of road anomaly data types. Traffic flow characteristics and road features, such as congestion and intersection saturation, are analyzed based on traffic flow data of road status data types. Intersection and road information are correlated with standard addresses. Environmental information related to intersections and roads is obtained through the relationship between standard addresses and regions. A correlation between people and violation records is established from basic traffic data such as traffic violations and accident information. Entities of drivers and vehicles are constructed based on basic traffic data such as personnel and vehicle information, medical treatment, and violations.

[0097] To further illustrate the technical solutions provided in the embodiments of this application, a detailed description is provided below in conjunction with the accompanying drawings and specific implementation methods. Although the embodiments of this application provide method operation steps as shown in the following embodiments or drawings, the method may include more or fewer operation steps based on conventional or non-inventive methods. In steps where there is no logically necessary causal relationship, the execution order of these steps is not limited to the execution order provided in the embodiments of this application.

[0098] See Figure 3 This is a flowchart illustrating a method for constructing a traffic knowledge graph according to an embodiment of this application. The method can be specifically executed as follows: Figure 3 The steps shown are as follows:

[0099] In step S301, traffic data of multiple data types are acquired; the data types include address data, basic traffic data, and traffic accident data.

[0100] In step S302, feature extraction of preset dimensions is performed on the traffic basic data to obtain multi-dimensional feature information of the traffic basic data; triple extraction is performed on the traffic accident data using a fine-tuned large model to obtain triple information.

[0101] In this application, the feature extraction of traffic infrastructure data with preset dimensions to obtain multidimensional feature information of the traffic infrastructure data is specifically implemented as follows:

[0102] To address the needs of driver risk assessment, this paper proposes a systematic method for standardized processing and risk association of multi-dimensional features, integrating various information such as basic attributes, social security records, traffic violations, illegal activities, and repeated use of psychotropic drugs. After standardizing the data, it stores it as individual entity attributes. This solution provides core evidence for driver risk assessment through the standardization of 11 preset dimensions of core driver features, risk labeling standardization, and risk association logic. This enables the accurate identification and dynamic management of high-risk individuals.

[0103] The core characteristics, standardized processing methods, and risk association logic of the 11 preset dimensions of drivers are shown in Table 4:

[0104] Table 4

[0105]

[0106] Furthermore, this application can conduct data analysis based on historical case data of accident-prone highway locations, classifying roads according to the number and severity of accidents occurring on road sections, and constructing road entity characteristics. The road classification rules are shown in Table 5:

[0107] Table 5

[0108]

[0109] In one possible implementation, triplet extraction is performed on traffic accident data using a finely tuned large model to obtain triplet information, and then the process is executed as follows: Figure 4 The steps shown are as follows:

[0110] In step S401, text parsing and entity recognition are performed on the traffic accident data to obtain multiple entities corresponding to the traffic accident data.

[0111] In step S402, the relationships between multiple entities corresponding to the traffic accident data are extracted to obtain the causal relationships between the multiple entities.

[0112] In step S403, triples are constructed for the causal relationships between multiple entities to obtain triple information; the triple information includes the accident result, the accident address, and the accident cause.

[0113] In practice, this application requires extracting the accident results, accident address, and accident cause from text data of traffic accident data such as police reports and accidents.

[0114] Traditional named entity extraction is performed using NLP methods, which extract entity relationships based on rule-based approaches and syntactic dependency relations. This approach has low accuracy and lacks the ability to understand complex text. Therefore, this application constructs inference data samples and extracts events and relationships based on a large model fine-tuning method. The logical and causal relationships in the text description are provided to the large model for training, improving the accuracy of graph triple recognition and identifying the main events and causal relationships between events in the text description of traffic accident data.

[0115] By fine-tuning the large model, the accuracy of triple extraction from text data can be improved. The focus of triple extraction is to extract the causal relationships of road traffic events.

[0116] Because traffic accident data involves descriptions of various aspects, it is necessary to first construct inference samples and fine-tune the model: First, collect textual data of traffic accidents, clean the data to remove duplicates and excessive data; then, screen the data samples, selecting balanced samples for each category based on the classification data of traffic accidents; perform inference annotation on the selected data, the annotation process includes five steps: entity element identification, causal relationship analysis, data standardization, triple concatenation, and triple extraction, to construct inference samples; then, fine-tune the model based on the constructed inference samples; finally, extract triples based on the fine-tuned large model to obtain triple information.

[0117] For example, traffic accident data could be: "At the intersection of Renmin Road and Jiefang Road, pedestrians frequently cross the road against the red light. I am concerned that this may lead to traffic accidents. I suggest strengthening management or adding facilities."

[0118] First, text parsing is performed, resulting in the following text structure breakdown: "At the intersection of Renmin Road and Jiefang Road" is a clear location description, serving as the background of the event. "Pedestrians frequently jaywalk" describes a recurring behavior (high-frequency event), which is the core of the problem. "I worry this will lead to traffic accidents" expresses a causal relationship; "this" refers to the preceding behavior, directly linking the cause (jaywalking) and the result (traffic accident). "It is recommended to strengthen management or add facilities" is a suggestion and is not directly used for triple extraction (because it is not a causal entity but a subsequent action). The text emphasizes that the behavior of "pedestrians jaywalking" may lead to "traffic accidents," which is a typical road traffic causal chain. Text parsing reveals the key elements as: Location = "Intersection of Renmin Road and Jiefang Road," Cause Event = "Pedestrians frequently jaywalk," and Result Event = "Traffic accident" (potential risk).

[0119] Scanning the key elements of the text, "the intersection of Renmin Road and Jiefang Road" clearly identifies the location and is categorized as the entity type "location". "Running a red light" is a dynamic behavior and is categorized as the entity type "event", representing a violation. "Traffic accident" is a potential consequence and is categorized as the entity type "event", representing a risk. Therefore, entity recognition of the key elements of the text yields multiple entities corresponding to the traffic accident data: the address is the intersection of Renmin Road and Jiefang Road, the cause is running a red light, and the result is a traffic accident.

[0120] Relationships were extracted from multiple entities corresponding to traffic accident data, and the causal relationships between the entities were found to be: pedestrians running red lights are worried about causing traffic accidents and are worried about the risk of traffic accidents due to running red lights.

[0121] In the original text “Pedestrians often run red lights to cross the road,” “often” indicates frequency, but the core is “pedestrians run red lights”. This should be optimized into a concise event: “pedestrians run red lights”. The text is “worried that it will lead to”, so the result should be labeled as “traffic accident risk” (distinguishing between actual and potential risks). The accident level is 1, resulting in a triplet of information.

[0122] In this application, the system is standardized to construct a risk level label. Based on the severity of the accident and its social impact, traffic accidents are divided into 5 levels (Level 1 is the least serious, and Level 5 is the most serious). The following is an explanation of the risk level standardization:

[0123] Level 1: Minor accidents or concerns about potential accidents, with no injuries or only minor scratches, and low losses, generally below 1,000 yuan. Typical scenarios include low-speed rear-end collisions, parking scrapes, and minor single-vehicle accidents, where there is concern about potential accidents.

[0124] Level 2: General accidents, with minor injuries to 1-2 people who require medical assistance but are not in life-threatening condition, and partial damage to the vehicle, with losses ranging from 1,000 to 30,000 yuan. Typical scenarios include minor collisions of multiple vehicles and minor injuries to people who are still able to move.

[0125] Level 3: Major accidents with casualties, generally involving 1-2 people seriously injured and incapacitated, requiring hospitalization, or more than 3 people with minor injuries. Vehicles are severely damaged or multiple vehicles are damaged, resulting in losses of approximately RMB 30,000 to 200,000. Typical scenarios include multi-vehicle rear-end collisions on highways and bus rollovers.

[0126] Level 4: Major accidents that result in 1-2 deaths or 3 or more serious injuries, with property damage between RMB 200,000 and RMB 1 million. Typical scenarios include hazardous chemical leaks, bus plunges off cliffs, and serious school bus collisions.

[0127] Level 5: Particularly serious accidents, resulting in 3 or more deaths or 10 or more serious injuries, with property losses exceeding 1 million yuan, requiring the activation of a national emergency response, such as tunnel / bridge collapses, chain explosions, and major traffic accidents.

[0128] Finally, triplet information is constructed and the output triplet information is standardized: the accident address is the intersection of Renmin Road and Jiefang Road, the cause of the accident is running a red light, and the result of the accident is a level 1 traffic accident risk.

[0129] In one possible implementation, the method provided in this application can also be performed as follows: Figure 5 The steps shown are as follows:

[0130] In step S501, semantic parsing is performed on the accident address to extract the target address elements from the accident address.

[0131] In step S502, the target address element is matched with multiple address information in the standard address database to obtain the confidence level corresponding to each of the multiple address information.

[0132] In step S503, address information with a confidence level greater than a preset threshold among multiple address information is used as standard address information; and the standard address information is used as the accident address in the triplet information.

[0133] In practice, a large model is used to perform semantic parsing on the accident address to extract target address elements, such as road names, landmarks, and directional terms like "south side / nearby," to resolve ambiguities in vague descriptions, such as the need to determine the specific range of "nearby" in conjunction with the context. The parsed target address elements are then matched with multiple address information in the standard address database in multiple dimensions, such as road name matching, landmark association matching, and directional term calibration. Priority is given to matching address fragments with high confidence, such as the explicit road name "Renmin Road." Address information with a confidence level greater than a preset threshold among multiple address information is used as the standard address information.

[0134] Standard address information uses a unified address format, such as "intersection of XX Road and XX Road" being standardized as "intersection of XX Road and XX Road". Redundant expressions, such as prefixes like "located at" or "at", are eliminated to ensure that the output address conforms to the address coding standard, such as the GB / T2260-2020 administrative division code.

[0135] The preset threshold can be set according to actual needs, and this application does not impose any restrictions on it.

[0136] In this application, after obtaining the standard address information, the standard address information is added to the standard address database to dynamically expand the coverage of the knowledge base.

[0137] In this application, the standard address database mainly includes administrative regions, roads, POIs, etc., and includes the following main elements:

[0138] Administrative division elements: These refer to address elements corresponding to administrative divisions at the county level and above, including provinces, cities, counties, townships, etc. They appear at the beginning of the address and include provinces, cities, counties (districts), development zones, townships, streets, communities, and administrative villages. Regional address elements: These refer to address information at the natural village, business district, road, house number, and community level. Location address elements: These include vulnerabilities, units, floors, room numbers, and address description information, such as "XX City XX District XX Road XX Number" and "XX District XX Street XX Community". They include hierarchical relationships, such as province-city-district-street-house number, geographical coordinates, such as WGS84 / GCJ-02, and address aliases, such as the official name of "Jiaodong Airport" being "Qingdao Jiaodong International Airport".

[0139] Address composition structure network diagram as follows: Figure 6 As shown in Table 6, the address information in the standard address database is composed of address elements arranged according to certain rules. The core attributes of the address information in the standard address database are shown in Table 6.

[0140] Table 6

[0141]

[0142] In this application, the geohash algorithm is used to geocode address information in the standard address database. Area boundaries are defined according to radii of 2m, 20m, 60m, 200m, 1km, and 5km for subsequent aggregation and analysis of area information based on different business scenarios. This method enables the correlation between address data in traffic accident data and other modal data, for example:

[0143] Vague descriptions in police reports (such as "south side of XX intersection") can be located to a 60m level area using Geohash, and a quantitative correlation can be established with the phase difference adjustment of signal timing data (within the same Geohash area), forming a mapping chain of "semantic description → spatial coding → signal parameters".

[0144] If it is necessary to analyze the microscopic impact of pedestrians running red lights (within a 10m radius), generate an 8-bit Geohash code "ws18g5vj", which corresponds to a 24m×24m rectangular area, covering a 10m range south of the intersection.

[0145] If it is necessary to associate the traffic flow within a 5km radius, a 4-digit code "ws18" is generated, corresponding to a 10km×5km rectangular area, covering the overall traffic situation of the street where the intersection is located.

[0146] Through multi-precision Geohash encoding, full-granularity data fusion from "single-point events" to "regional impacts" can be achieved, providing a unified spatial benchmark for traffic accident causation analysis.

[0147] Thus, the attribute information shown in Table 7 is obtained, and a standard address library is constructed based on the attribute information shown in Table 7.

[0148] Table 7

[0149]

[0150] Address information extracted from traffic data is mostly non-standardized and often lacks latitude and longitude information. Location information extracted from traffic flow data and view data, however, contains checkpoints or road segments and can be correlated with latitude and longitude. Therefore, this application requires first geohash encoding the location information extracted from traffic flow data and view data, then associating the encoded addresses with address codes in a standard address database, and finally constructing a matching relationship using a matching algorithm.

[0151] It should be noted that since the encoding length of geohash codes is calculated based on the level of precision, the encoding length used in this application is 7 bits of precision, which is approximately 152 meters.

[0152] In step S303, text matching is performed on the address data to obtain a first matching result; place name matching is performed on the address data and multiple address information in the standard address database to obtain a second matching result; standard address encoding and place name location matching are performed on the first matching result and the second matching result respectively to obtain the first place name location similarity corresponding to the first matching result and the second place name location similarity corresponding to the second matching result; similarity calculation is performed on the first place name location similarity and the second place name location similarity to obtain the address matching result.

[0153] In its specific implementation, this application employs a hybrid approach for address matching, integrating the TF-IDF and KMP matching algorithms, and further improving upon these algorithms. It comprehensively matches addresses by combining address description information and latitude / longitude location coding information, and then performs correlation and fusion. The correlation between address data is considered comprehensively from both place name and location information perspectives. First, place name similarity is calculated based on place name parsing; then, place name location similarity is calculated based on location information; finally, place name similarity and location similarity are combined to match and correlate place name data.

[0154] In one possible implementation, the text matching of address data in this application to obtain a first matching result can also be performed as follows: Figure 7 The steps shown are as follows:

[0155] In step S701, the address data is segmented to obtain multiple words.

[0156] In step S702, for each word, the word frequency in the address data is determined based on the number of times the word appears in the address data and the number of words in the address data; the inverse document frequency of the word is determined based on the number of documents in the address data that include the word.

[0157] In step S703, the first matching result is determined based on the term frequency and inverse document frequency of multiple words in the address data.

[0158] In practice, the text is segmented into words, and the TF-IDF algorithm is used to calculate the first matching result. The formula for calculating the first matching result is as follows:

[0159] TF-IDF=TF IDF.

[0160] TF (Term Frequency) is the frequency of a word in the address data, where TF = the number of times the word appears in the document / the total number of words in the document. IDF is the inverse document frequency, where IDF = log(total number of documents / (number of documents containing the word + 1)). The denominator is incremented by 1 to prevent the denominator from being zero.

[0161] In one possible implementation, the method of matching address data with multiple address information in a standard address database to obtain a second matching result can also be performed as follows: Figure 8 The steps shown are as follows:

[0162] In step S801, the address in the address data is split to determine the special name string and the common name string;

[0163] In step S802, the special name string and the special name strings of multiple address information in the standard address database are matched for similarity to determine the special name similarity between the special name string in the address data and the special name string of each address information.

[0164] In step S803, the common name string and the common name strings of multiple address information in the standard address database are matched for similarity to determine the common name similarity between the common name string in the address data and the common name string of each address information.

[0165] In step S804, a second matching result is determined based on proper name similarity and common name similarity.

[0166] In specific implementation, first split the address into the structure of "specific name + general name". The specific name is the text segment used to distinguish each other in the place name. For example, in "No. 12 Tong'an Road, Shibei District, Qingdao City, Shandong Province", Shandong, Qingdao, Shibei, and Tong'an Road are specific names. The general name usually follows the specific name. For example, in "No. 12 Tong'an Road, Shibei District, Qingdao City, Shandong Province", province, city, and district are general names.

[0167] First, calculate the similarity of specific names. Drawing on the idea of the KMP (Knuth-Morris-Pratt) algorithm, calculate the similarity degree of the specific name strings in terms of character expression.

[0168] The specific name strings A = {a1, a2....ai} and B = (b1, b2..bj}, where ai (0 < i < m), bj (0 < j < n) respectively represent the Chinese characters in the specific name strings A and B, and m and n are the character lengths of the specific name strings A and B respectively. The calculation method of the specific name similarity is as follows:

[0169] First, compare the string lengths of the two specific name strings, and split the specific name string with the shorter string length into a set of single characters to form a target set. Assume that the string length of the specific name string B is shorter, and the character set formed by splitting is {b1, b2,..., bn}.

[0170] Then, take the character bt (0 < t ≤ n) from the character set of the specific name string B in the order from front to back, and search for the matching character in the specific name string A from the front. If there is a matching character ax = bt (0 < x ≤ m), then jump to ax and split the specific name string A, take the characters after ax to form a substring A2 of the specific name string A, which is used as the input string for the subsequent matching process, and then go to the next character bt+1 for matching; if no matching character is found, then jump to the next character in the character set of the specific name string B for matching until each character in the character set of the specific name string B completes the matching process. Record the number of matching characters between the specific name string A and the specific name string B as k, then the calculation formula for the specific name similarity Sspecific(A, B) between the specific name string A and the specific name string B is: Sspecific(A, B) = k / n × 100%.

[0171] Secondly, calculate the general name similarity Sgeneral(A, B). Analyze the similarity of place name types based on the general name information of place names, and assign corresponding weights when calculating the similarity of place name names. If the general names of two place names are exactly the same, the weight of the place name type similarity is 1; if the general names of the place names are different, assign values to the place name type similarity according to the differences in the classification attributes of the place names. If the sub-categories of the place name classification are the same, the place name type similarity is taken as 0.8; if the sub-categories of the place name classification are different but the middle categories are the same, the place name type similarity is taken as 0.6; if the middle categories of the place name classification are different, the place name type similarity is taken as 0.4.

[0172] Finally, the proper name similarity and common name similarity are combined to obtain the second matching result S(A,B) = Sproper(A,B)×Scommon(A,B).

[0173] After obtaining the first matching result based on the TF-IDF algorithm and the second matching result based on the KMP address matching algorithm, standard address encoding and place name location matching are performed on the first matching result and the second matching result respectively to obtain the first place name location similarity corresponding to the first matching result and the second place name location similarity corresponding to the second matching result.

[0174] In practice, the first and second matching results are first encoded using geohash codes to determine their addresses, and then place name location matching is performed to obtain the similarity score for each address code. The calculation method is as follows:

[0175]

[0176] in This represents the address code of the i-th element in the matching results after sorting by similarity from largest to smallest. This represents the similarity of the i-th position in the matching results.

[0177] The first place name location similarity of the first matching result obtained by the above formula is T={(GEO1,t1),(GEO2,t2),……,(GEOe,te)}; the second place name location similarity of the second matching result is: K={(GEO1,t1),(GEO2,t2),……,(GEOf,tf)};

[0178] Finally, the similarity between the first and second place names is calculated using the following formula:

[0179]

[0180] After similarity calculation, the Geohash code with the highest similarity weight that is greater than a preset threshold is taken as the address matching result. The preset threshold is set according to actual needs.

[0181] In step S304, a traffic knowledge graph is constructed based on multidimensional feature information, triplet information, and address matching results.

[0182] In one possible implementation, the method provided in this application can also be performed as follows: Figure 9 The steps shown are as follows:

[0183] In step S901, traffic flow data of road status data type is obtained;

[0184] In step S902, the traffic flow data is filtered and spatiotemporally unified to obtain preprocessed traffic flow data; road condition indicators are extracted from the preprocessed traffic flow data to obtain road condition indicators.

[0185] In step S903, a traffic knowledge graph is constructed based on multidimensional feature information, triplet information, address matching results, and road status indicators.

[0186] In practice, for traffic flow data of the road status type, the real-time status of the road at the intersection is analyzed by calculating the traffic flow, average speed, congestion index, and average flow rate in real time, and a real-time road entity relationship map is constructed.

[0187] First, perform data filtering: remove abnormal records in the traffic flow data where the speed exceeds 200% of the road section speed limit. For example, when the speed limit on urban roads is 60km / h, filter data with a speed >120km / h. Then, correct trajectory point jumps. If the distance between adjacent points exceeds 500m and the time difference is <1 second, mark it as an equipment failure and remove it.

[0188] Then, a unified spatiotemporal processing is performed: all data timestamps are converted to UTC+8 time zone, and the coordinates of each device (such as GCJ-02, BD-09) are converted to WGS84 standard coordinate system with the error controlled within ±2m. At the same time, multi-precision geohash codes are generated, and spatial association with police and accident data is achieved through the geolocation code here.

[0189] Finally, traffic flow data at intersections and along main roads is extracted from real-time traffic flow data, and road condition indicators such as real-time traffic volume, average speed, congestion index, and average traffic volume over the past hour are calculated.

[0190] The calculation methods for road condition indicators such as real-time traffic flow, average speed, congestion index, and average traffic flow over the past hour are existing technologies and will not be elaborated here.

[0191] The calculation method for road condition indicators is shown in Table 8:

[0192] Table 8

[0193]

[0194] In one possible implementation, the method provided in this application can also be performed as follows: Figure 10 The steps shown are as follows:

[0195] In step S1001, view data of road anomaly data type is obtained;

[0196] In step S1002, road anomaly features are extracted from the view data to obtain road anomaly features;

[0197] In step S1003, a traffic knowledge graph is constructed based on multidimensional feature information, triplet information, address matching results, and road anomaly features.

[0198] In practical implementation, for view data of road anomaly types, images captured by checkpoint videos and images from checkpoint inspection videos are labeled. Fine-tuning training is performed based on the InternVL multimodal large model to identify road anomaly features, such as missing guardrails, missing central dividers, conflicting signs and markings, long steep slopes, obscured traffic signs, damaged facilities, incorrect sign placement, excessive road openings, missing speed bumps, etc. Event identification is performed. At the same time, the level of events is classified by combining highway grade and event characteristics. In this application, existing technologies are combined to identify event types and event levels through multimodal large models, thereby constructing a road and event map.

[0199] The attribute descriptions of road anomaly features are shown in Table 9:

[0200] Table 9

[0201]

[0202] In one possible implementation, multidimensional feature information, triplet information, address matching results, road anomaly features, and road status indicators can be fused to construct a traffic knowledge graph.

[0203] Therefore, this application employs a large model to extract information from traffic data, enabling entity extraction of causal relationships within complex text timeframes. Based on a hybrid matching approach using multiple algorithms, it first performs text matching and place name matching on address data. Then, it conducts similarity analysis based on the matching results of text and place name matching, along with the place name location matching results, to obtain address matching results. Finally, it fuses the extracted entities and address matching results to construct a traffic knowledge graph. This knowledge graph, built from multi-dimensional features, can comprehensively reveal the causal mechanisms of accidents and provide data-driven decision-making support for road safety management.

[0204] Based on the same inventive concept, embodiments of this application also provide a device for constructing a traffic knowledge graph. Figure 11 A schematic diagram of a traffic knowledge graph construction device provided in this application embodiment, the device comprising:

[0205] The data acquisition module 1101 is used to acquire traffic data of multiple data types, including address data, basic traffic data, and traffic accident data.

[0206] Feature extraction module 1102 is used to extract features of preset dimensions from the traffic basic data to obtain multi-dimensional feature information of the traffic basic data;

[0207] The triplet extraction module 1103 is used to extract triplets from the traffic accident data using a finely tuned large model to obtain triplet information.

[0208] Address matching module 1104 is used to perform text matching on the address data to obtain a first matching result; to perform place name matching on the address data and multiple address information in a standard address database to obtain a second matching result; to perform standard address encoding and place name location matching on the first matching result and the second matching result respectively to obtain a first place name location similarity corresponding to the first matching result and a second place name location similarity corresponding to the second matching result; and to perform similarity calculation on the first place name location similarity and the second place name location similarity to obtain an address matching result.

[0209] The graph construction module 1105 is used to construct a traffic knowledge graph based on the multidimensional feature information, the triplet information, and the address matching results.

[0210] In one possible implementation, the triplet extraction module 1103 is specifically used for:

[0211] The traffic accident data is parsed and entity recognition is performed to obtain multiple entities corresponding to the traffic accident data.

[0212] Relationships are extracted from multiple entities corresponding to the traffic accident data to obtain the causal relationships between the multiple entities;

[0213] The causal relationships between the multiple entities are constructed into triples to obtain triple information; the triple information includes the accident result, the accident address, and the accident cause.

[0214] In one possible implementation, the triplet extraction module 1103 is specifically used for:

[0215] Perform semantic parsing on the incident address to extract the target address elements from the incident address;

[0216] The target address element is matched with multiple address information in the standard address database to obtain the confidence level corresponding to each of the multiple address information;

[0217] Address information with a confidence level greater than a preset threshold among the multiple address information is used as standard address information; and the standard address information is used as the accident address in the triplet information.

[0218] In one possible implementation, the device further includes an index extraction module 1106;

[0219] Data acquisition module 1101 is specifically used to acquire traffic flow data of road status type.

[0220] The indicator extraction module 1106 is used to perform data filtering and spatiotemporal unification on the traffic flow data to obtain preprocessed traffic flow data; and to extract road status indicators from the preprocessed traffic flow data to obtain road status indicators.

[0221] The graph construction module 1105 is used to construct a traffic knowledge graph based on the multidimensional feature information, the triplet information, the address matching results, and the road status indicators.

[0222] In one possible implementation, the data acquisition module 1101 is used to acquire view data of road anomaly data type;

[0223] Feature extraction module 1102 is used to extract road anomaly features from the view data to obtain road anomaly features;

[0224] The graph construction module 1105 is used to construct a traffic knowledge graph based on the multidimensional feature information, the triplet information, the address matching results, and the road anomaly features.

[0225] In one possible implementation, the address matching module 1104 is specifically used for:

[0226] The address data is segmented into multiple words;

[0227] For each word, the word frequency in the address data is determined based on the number of times the word appears in the address data and the number of words in the address data; the inverse document frequency of the word is determined based on the number of documents in the address data that include the word.

[0228] The first matching result is determined based on the word frequency and inverse document frequency of the multiple words in the address data.

[0229] In one possible implementation, the address matching module 1104 is specifically used for:

[0230] The addresses in the address data are split to determine the special name string and the common name string;

[0231] The special name string and the special name strings of multiple address information in the standard address database are matched for similarity to determine the special name similarity between the special name string in the address data and the special name string of each address information.

[0232] The common name string and the common name strings of multiple address information in the standard address database are matched for similarity to determine the common name similarity between the common name string in the address data and the common name string of each address information.

[0233] The second matching result is determined based on the proper name similarity and the common name similarity.

[0234] Based on the same inventive concept, this application also provides an electronic device. Figure 12 This application provides a schematic diagram of an electronic device structure, such as... Figure 12 As shown, it includes: processor 1201, communication interface 1202, memory 1203 and communication bus 1204, wherein processor 1201, communication interface 1202 and memory 1203 communicate with each other through communication bus 1204.

[0235] The memory 1203 stores a computer program, which, when executed by the processor 1201, causes the processor 1201 to perform the steps of any of the traffic knowledge graph construction methods provided in the embodiments of this application.

[0236] Since the problem-solving methods of the above-mentioned electronic devices are similar to those of constructing traffic knowledge graphs, the implementation of the above-mentioned electronic devices can be found in the embodiments of the method, and repeated details will not be repeated.

[0237] The communication bus mentioned in the aforementioned electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used in the figure, but this does not indicate that there is only one bus or one type of bus. The communication interface 1202 is used for communication between the aforementioned electronic device and other devices. The memory can include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0238] The processors mentioned above can be general-purpose processors, including central processing units, network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits, field-programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0239] Based on the above embodiments, this invention also provides a computer-readable storage medium storing a computer program executable by a processor. When the program runs on the processor, it causes the processor to perform the steps of any of the traffic knowledge graph construction methods provided in this application.

[0240] Based on the same inventive concept, embodiments of this application provide a computer program product, including a computer program, which, when executed by a processor, implements the steps of any of the traffic knowledge graph construction methods provided in embodiments of this application.

[0241] Since the principle of the computer-readable storage medium in solving the problem is similar to the method of constructing a traffic knowledge graph, the implementation of the computer-readable storage medium can be found in the embodiments of the method, and repeated details will not be repeated.

[0242] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0243] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0244] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0245] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0246] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for constructing a traffic knowledge graph, characterized in that, The method includes: Acquire traffic data of multiple data types; these data types include address data, basic traffic data, and traffic accident data. The method further includes: performing feature extraction on the traffic basic data in preset dimensions to obtain multi-dimensional feature information of the traffic basic data; the method also includes: performing standardization processing on the multi-dimensional feature information, risk label standardization processing and risk association logic, to provide the core basis for risk assessment for traffic data fusion and knowledge graph construction, and to be used for accurate identification and dynamic management of high-risk groups; The traffic accident data is processed using a fine-tuned large model for text parsing and entity recognition to obtain multiple entities corresponding to the traffic accident data; relationships are extracted from the multiple entities corresponding to the traffic accident data to obtain the causal relationships between the multiple entities; triples are constructed from the causal relationships between the multiple entities to obtain triple information; the triple information includes accident result, accident address, and accident cause. The address data is subjected to text matching to obtain a first matching result; the address data is matched with multiple address information in the standard address database to obtain a second matching result; the address data includes a proper name string and a common name string; the first matching result and the second matching result are respectively subjected to standard address encoding and place name location matching to obtain a first place name location similarity corresponding to the first matching result and a second place name location similarity corresponding to the second matching result; the first place name location similarity and the second place name location similarity are calculated to obtain an address matching result; A traffic knowledge graph is constructed based on the multidimensional feature information, the triplet information, and the address matching results.

2. The method according to claim 1, characterized in that, The method further includes: Perform semantic parsing on the incident address to extract the target address elements from the incident address; The target address element is matched with multiple address information in the standard address database to obtain the confidence level corresponding to each of the multiple address information; Address information with a confidence level greater than a preset threshold among the multiple address information is used as standard address information; and the standard address information is used as the accident address in the triplet information.

3. The method according to claim 1, characterized in that, The method further includes: Obtain traffic flow data of road status type; The traffic flow data is filtered and spatiotemporally unified to obtain preprocessed traffic flow data; road condition indicators are extracted from the preprocessed traffic flow data to obtain road condition indicators. A traffic knowledge graph is constructed based on the multidimensional feature information, the triplet information, the address matching results, and the road status indicators.

4. The method according to claim 1, characterized in that, The method further includes: Retrieve view data for road anomaly data types; The view data is subjected to road anomaly feature extraction to obtain road anomaly features; A traffic knowledge graph is constructed based on the multidimensional feature information, the triplet information, the address matching results, and the road anomaly features.

5. The method according to claim 1, characterized in that, The step of performing text matching on the address data to obtain a first matching result includes: The address data is segmented into multiple words; For each word, the word frequency in the address data is determined based on the number of times the word appears in the address data and the number of words in the address data; the inverse document frequency of the word is determined based on the number of documents in the address data that include the word. The first matching result is determined based on the word frequency and inverse document frequency of the multiple words in the address data.

6. The method according to claim 1, characterized in that, The step of matching the address data with multiple address information in the standard address database to obtain a second matching result includes: The addresses in the address data are split to determine the special name string and the common name string; The special name string and the special name strings of multiple address information in the standard address database are matched for similarity to determine the special name similarity between the special name string in the address data and the special name string of each address information. The common name string and the common name strings of multiple address information in the standard address database are matched for similarity to determine the common name similarity between the common name string in the address data and the common name string of each address information. The second matching result is determined based on the proper name similarity and the common name similarity.

7. A device for constructing a traffic knowledge graph, characterized in that, The device includes: The data acquisition module is used to acquire traffic data of multiple data types, including address data, basic traffic data, and traffic accident data. The feature extraction module is used to extract features from the traffic basic data in a preset dimension to obtain multi-dimensional feature information of the traffic basic data. The feature extraction module is also used to: perform standardization processing, risk label standardization processing and risk association logic on the multi-dimensional feature information, so as to provide the core basis for risk assessment for traffic data fusion and knowledge graph construction, and to accurately identify and dynamically manage high-risk groups. The triple extraction module is used to perform text parsing and entity recognition on the traffic accident data using a fine-tuned large model to obtain multiple entities corresponding to the traffic accident data; to extract relationships between the multiple entities corresponding to the traffic accident data to obtain causal relationships between the multiple entities; and to construct triples from the causal relationships between the multiple entities to obtain triple information; the triple information includes accident result, accident address, and accident cause. The address matching module is used to perform text matching on the address data to obtain a first matching result; to perform place name matching on the address data with multiple address information in a standard address database to obtain a second matching result; the address data includes a proper name string and a common name string; to perform standard address encoding and place name location matching on the first matching result and the second matching result respectively to obtain a first place name location similarity corresponding to the first matching result and a second place name location similarity corresponding to the second matching result; and to perform similarity calculation on the first place name location similarity and the second place name location similarity to obtain an address matching result. The graph construction module is used to construct a traffic knowledge graph based on the multidimensional feature information, the triplet information, and the address matching results.

8. An electronic device, characterized in that, The electronic device includes a processor that executes a computer program stored in a memory to implement the steps of the method for constructing a traffic knowledge graph as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the method for constructing a traffic knowledge graph as described in any one of claims 1-6.

Images