A driving experience reminder navigation text generation system and method
Through driving experience reminder navigation text generation system, combining large language models and fine-tuning of traffic safety knowledge, detailed pre-reminder navigation text is generated, which solves the problem that existing navigation software cannot provide detailed risk warnings, and improves the driving safety of novice drivers and Ability to predict risks.
Patent Information
- Application Number
- CN202410336733.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-22
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-03-22
AI Technical Summary
Existing navigation software cannot provide sufficiently detailed driving risk warnings, especially in complex road situations, which are difficult to help novice drivers make the best path decisions, and lack professional traffic rules warning functions.
A driving experience reminder navigation text generation system was designed, and the pre-processing module for sharing information with driving experience, the planning path risk point information matching module, the big model prompt word module and the traffic safety knowledge fine-tuning big model module are generated to generate pre-reminder navigation text to help Drivers identify potential risks and make safer driving decisions.
It improves the driving safety of novice drivers in complex road situations, enhances the ability to predict potential risks, helps drivers to cope with driving challenges more confidently, and reduces the risk of traffic accidents.
Smart Images

Figure CN118278364B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of route navigation, and in particular relates to a driving experience reminder navigation text generation system and method. Background Art
[0002] For novice drivers who are not familiar with road conditions, navigation software is an important assistant for travel. These software rely on huge databases, real-time traffic updates and a large amount of traffic data to provide multiple optional routes based on travel mode and time. However, in actual driving, drivers often encounter challenges that navigation software cannot foresee: complex intersections and overpasses, road defects (such as potholes and cracks), crowded areas (such as shopping malls and near schools), road spills and construction areas, etc. These situations increase driving risks, especially for novice drivers.
[0003] If drivers can share their driving experience, especially details about specific road sections, navigation software can warn novice drivers of possible risk points when planning routes. This will not only help drivers choose better routes, but also enhance their overall understanding of the selected routes, thereby reducing the risk of traffic accidents and improving driving safety.
[0004] Although existing navigation software can provide basic navigation, real-time traffic updates and lane guidance, this information is often not detailed enough to fully help drivers make the best path decisions. In addition, there is a lack of navigation tools on the market that can provide professional traffic rules warnings.
[0005] With the advent of large language models (LLMs) such as ChatGPT, they have achieved remarkable results in multiple natural language processing (NLP) tasks such as text generation, translation, summary writing, and natural language understanding, and have attracted widespread attention. These technologies have shown excellent performance in many fields such as automatic customer service, content creation, educational tutoring, information retrieval, and data analysis. However, since LLMs are usually pre-trained on a wide range of text data, they are not enough to cover the language usage in certain professional fields. Therefore, it is a feasible and necessary approach to fine-tune the model through the collected driving experience data to improve its accuracy and efficiency in generating text in the fields of traffic safety and driving experience.
[0006] Accident risk points refer to specific areas or conditions on the road that may increase the risk of traffic accidents. These risk points can be composed of a variety of factors, including but not limited to road design, traffic flow, road surface conditions, visibility, weather conditions, and the activities of pedestrians and non-motor vehicles.
[0007] Lateral clearance: usually refers to the clearance requirement or space in the lateral (lateral) direction of a vehicle or building. In traffic planning or architectural design, lateral clearance is used to describe the lateral spacing or space between or around two objects to ensure that vehicles or pedestrians can pass safely and avoid potential collisions or obstructions.
[0008] LLM: large language model;
[0009] NLP: natural language process;
[0010] Prior art 1 proposes a navigation method, terminal, server, electronic device and storage medium, which relates to the field of map technology and is used in the field of autonomous driving. When a vehicle passes through a designated scene to be reminded, a prompt text information box of different prompt scenes is generated on the navigation map in combination with real-time road conditions, thereby improving driving safety. Specifically, it includes:
[0011] Step 1: Use the roadside sensing and computing unit to obtain the speed and position information of the object being measured.
[0012] Step 2: The measured information is sent to the roadside communication device, and the roadside communication device broadcasts the V2X message to the server.
[0013] Step 3: The terminal sends a navigation request to the server. The server provides navigation services based on the navigation request sent by the terminal. The specific navigation service refers to: providing a navigation map that displays the real-time location of the terminal and the navigation route.
[0014] Step 4: Determine whether the navigation object is in a reminder scene based on the location of the navigation object. When the navigation object is in a reminder scene, an information window will pop up in the navigation map to display the reminder content. The reminder scenes include: vulnerable traffic participant collision reminder scene, forward collision reminder scene, emergency braking reminder scene, blind spot reminder / lane change assist reminder scene, abnormal vehicle reminder scene, intersection collision reminder scene, left turn assist reminder scene, emergency vehicle reminder scene, green wave speed reminder scene, etc.
[0015] The disadvantages of this prior art are:
[0016] (1) The information window only provides simple reminders such as "Watch out for vehicles ahead" and "Watch out for pedestrians", and cannot provide professional information in the field of traffic safety knowledge.
[0017] (2) The content of the information window reminder is displayed in real time under the preset scenario, but in the selection of the entire driving path, it cannot provide decision-making assistance to the driver before selecting the driving path based on the collected driving experience information. Summary of the invention
[0018] The purpose of the present invention is to provide a driving experience reminder navigation text generation system and method to address the above-mentioned deficiencies in the prior art, so as to solve the problem that the existing large language model does not cover the professional navigation field and cannot provide professional information in the field of traffic safety knowledge.
[0019] In order to achieve the above object, the technical solution adopted by the present invention is:
[0020] In a first aspect, a driving experience reminder navigation text generation system and method includes a driving experience sharing information preprocessing module for extracting risk point information on a driving road based on historical traffic accident information and driving experience sharing information;
[0021] The planning path risk point information matching module is used to obtain the driving navigation path planning text according to the navigation starting point and end point input by the user, and to match the navigation path planning text with the risk point information in the driving experience sharing information preprocessing module to obtain the information items that need to be pre-reminded on the navigation path planning;
[0022] A large model prompt word module, used to extract information items that need to be pre-reminded on the route planning and use the information items as pre-remind point information, and generate guidance prompt words according to the pre-remind point information;
[0023] A traffic safety knowledge fine-tuning large model module is used to construct a traffic regulations large model, input the guidance prompt words into the traffic regulations large model, and output a pre-warning navigation text;
[0024] The pre-reminder navigation form module generates a pre-reminder navigation form for displaying the pre-reminder navigation text.
[0025] In a second aspect, a method for generating a driving experience reminder navigation text generation system comprises the following steps:
[0026] S1. Extract risk point information on the road based on historical traffic accident information and driving experience sharing information;
[0027] S2. According to the navigation starting point and end point input by the user, a driving navigation route planning text is obtained, and the navigation route planning text is matched with the risk point information to obtain information items that need to be pre-reminded on the navigation route planning;
[0028] S3, extracting information items that need to be pre-remindered on the route planning, and using the information items as pre-reminder point information, and generating guidance prompt words according to the pre-reminder point information;
[0029] S4, constructing a traffic regulations model, inputting the guidance prompt words into the traffic regulations model, and outputting a pre-warning navigation text;
[0030] S5. Generate a pre-reminder navigation window for displaying the pre-reminder navigation text information according to the pre-reminder navigation text.
[0031] Furthermore, step S1 includes the following steps:
[0032] S1.1. Collect and annotate historical traffic accident information and driving experience sharing information;
[0033] S1.2. Preprocess the annotated historical traffic accident information and driving experience sharing information to extract risk point information on the driving road.
[0034] Furthermore, the driving experience sharing information in step S1.1 includes: road surface disease information, roadside construction, road spillage, pedestrian-dense areas and high-incidence points of vehicle reversal.
[0035] Further, step S1.2 includes the following steps:
[0036] S1.2.1. Privacy protection and data cleaning of historical traffic accident information and driving experience sharing information;
[0037] S1.2.2. Standardize the historical traffic accident information and driving experience sharing information after privacy protection and data cleaning;
[0038] S1.2.3. Extract characteristic information from standardized historical traffic accident information and driving experience sharing information;
[0039] S1.2.4. Construct search keywords based on the location information and time information in the feature information to obtain risk point information.
[0040] Further, step S2 includes the following steps:
[0041] S2.1. Based on the navigation start point and end point input by the user, the map software generates a driving navigation route planning text;
[0042] S2.2, extracting location information from the driving navigation route planning text;
[0043] S2.3. Match the location information in the driving navigation route planning text with the risk point information to obtain information items that require advance reminders on the navigation route planning.
[0044] Furthermore, the guidance prompt word T in step S3 is:
[0045] T=S+P+X
[0046] Among them, S is the system message prompt part, P is the road pre-warning information, and X is the text input.
[0047] Furthermore, in step S4, a large traffic regulation model is constructed, including:
[0048] S4.1. Construct a traffic law dataset based on the collected traffic safety regulations, traffic driving rules, and traffic violation penalty points rules, and annotate the traffic law dataset to obtain a question-and-answer dataset corresponding to traffic safety knowledge;
[0049] S4.2. Standardize the format of the data in the question-and-answer dataset corresponding to traffic safety knowledge;
[0050] S4.3. Use a large language model to fine-tune the question-and-answer dataset corresponding to traffic safety knowledge after the format is standardized to obtain a large model of traffic regulations.
[0051] The driving experience reminder navigation text generation system and method provided by the present invention have the following beneficial effects:
[0052] The present invention assists novice drivers in making safer driving decisions. This method not only improves driving safety, but also helps novice drivers to face challenges on the road more confidently.
[0053] The present invention includes a driving experience sharing information preprocessing module, a planning path risk point information matching module, a large model prompt word module, a traffic safety knowledge fine-tuning large model module and a pre-warning navigation window module; combining the five modules, the driving experience data collected by fine-tuning the large language model is used to realize pre-warning and path auxiliary decision-making for the driver's path navigation.
[0054] The present invention matches the navigation text generated by the map navigation path with the risk point information shared by the driver's driving experience, and can generate pre-warning navigation text in real time; and uses the question-and-answer GPT in the field of traffic safety, Tra-GPT, which is fine-tuned by annotated traffic safety regulations information and violation penalty rules, to help novice drivers generate driving navigation pre-warning text to help drivers make auxiliary path decisions.
[0055] The present invention provides key safety warning information to the driver in advance by collecting experience sharing from other drivers; when the driver inputs the starting point and destination, the system can display a series of risk warning texts based on accumulated driving experience before the journey begins; in this way, the driver can have a comprehensive understanding of the road conditions to be driven on before departure, which significantly improves the safety during driving and enhances the driver's ability to predict potential risks. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1 This is a flow chart of a method for generating a driving experience reminder navigation text according to embodiment 2 of the present invention.
[0057] Figure 2 This is a module diagram of a driving experience reminder navigation text generation system according to Embodiment 1 of the present invention.
[0058] Figure 3 This is a navigation text display diagram of Example 2 of the present invention. DETAILED DESCRIPTION
[0059] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.
[0060] Example 1
[0061] This embodiment provides a driving experience reminder navigation text generation system, which uses a large language model to fine-tune the collected driving experience data to achieve pre-reminder and path auxiliary decision-making for the driver's path navigation, referring to Figure 2 , which specifically includes the following modules:
[0062] A driving experience sharing information preprocessing module is used to extract risk point information on the driving road based on historical traffic accident information and driving experience sharing information;
[0063] The planning path risk point information matching module is used to obtain the driving navigation path planning text according to the navigation starting point and end point input by the user, and to match the navigation path planning text with the risk point information in the driving experience sharing information preprocessing module to obtain the information items that need to be pre-reminded on the navigation path planning;
[0064] The large model prompt word module is used to extract the information items that need to be pre-reminded in the route planning and use the information items as the pre-remind point information, and generate the guidance prompt words according to the pre-remind point information;
[0065] A traffic safety knowledge fine-tuning large model module is used to construct a traffic regulations large model, input the guidance prompt words into the traffic regulations large model, and output a pre-warning navigation text;
[0066] The pre-reminder navigation form module generates a pre-reminder navigation form for displaying the pre-reminder navigation text.
[0067] Example 2
[0068] This embodiment is based on the system in embodiment 1, and uses the question-answering GPT in the field of traffic safety, Tra-GPT, which is fine-tuned by annotated traffic safety regulations and violation penalty rules, to help novice drivers generate driving navigation pre-reminder texts to help drivers make auxiliary path decisions, and provides a method for generating driving experience pre-reminder navigation texts, reference Figure 1 , which specifically includes the following steps:
[0069] Step S1: extract risk point information on the driving road based on historical traffic accident information and driving experience sharing information, which specifically includes the following contents:
[0070] Step S1.1, collecting and marking historical traffic accident information and driving experience sharing information;
[0071] The driving experience sharing information referred to in this embodiment is obtained by sharing by a large number of vehicle drivers, specifically including road disease information, roadside construction, road spillage, pedestrian-dense areas, and high-incidence points for vehicles driving in the wrong direction, etc.:
[0072] Historical accident information: historical traffic accident information on various roads in the city obtained from the traffic management department, including the time, location, and type of incident of the traffic accident.
[0073] Road surface damage information: refers to the location information, time information, picture information, text information, etc. of potholes, huge road cracks, etc. that affect normal driving or pose traffic hazards to normal driving.
[0074] Roadside construction information: refers to the location information, time information, image information and text information of road sections where the lanes become narrower and the lateral clearance is reduced due to roadside construction, which is likely to cause traffic congestion.
[0075] Pedestrians running red lights are more likely to occur in areas where there is a high incidence of red light violations: location information, time information, image information, and text information for intersections near university intersections or shopping malls with heavy traffic.
[0076] Traffic congestion hotspot information: location information, time information, image information, and text information at entrances and exits of large shopping malls, hospitals, schools, etc., where traffic congestion is severe due to frequent vehicle in and out.
[0077] High-incidence points for vehicle wrong-way traffic: locations, times, pictures and text information where vehicle wrong-way traffic is likely to occur due to unreasonable road structure or loose supervision.
[0078] Step S1.2: pre-process the annotated historical traffic accident information and driving experience sharing information to extract risk point information on the driving road, which specifically includes the following contents:
[0079] Step S1.2.1: Privacy protection and data cleaning of historical traffic accident information and driving experience sharing information, including:
[0080] Delete private information related to specific incidents, such as information about the parties involved in the accident and contact information of the person who uploaded the information;
[0081] Delete overly long Q&A content and Q&A information;
[0082] Delete irrelevant content.
[0083] Step S1.2.2, standardizing the historical traffic accident information and driving experience sharing information after privacy protection and data cleaning;
[0084] Specifically, accident information is standardized by setting filtering rules; for example, for live broadcast manuscript data, which has strong real-time characteristics and is mostly accident risk point information uploaded by listeners’ voice, it is necessary to delete spoken content and standardize risk point information.
[0085] Step S1.2.3: extracting characteristic information from the standardized historical traffic accident information and driving experience sharing information, which includes:
[0086] Extract characteristic information such as location information, time information, event summary, etc. of historical accident information.
[0087] For example: "At the intersection of xx Road and xx Road (location), on xx / xx / xx (time), xx event occurred (event description)".
[0088] Extract the location, time and event description of risk points for driving experience sharing.
[0089] For example: "At the intersection of xx Road and xx Road (location), xx / xx / xx (time), pedestrians jaywalking is serious and traffic accidents often occur / the road is under construction and is temporarily closed to traffic (event description)"
[0090] The extracted text is stored in jsonl file format, with each piece of data occupying one line. The storage format is as follows:
[0091] {"Location":[x,y],"Location text description","Time":"Time of event occurrence","description":"Event description"}.
[0092] Step S1.2.4: construct search keywords based on the location information and time information in the feature information to obtain risk point information.
[0093] Step S2: obtaining a driving navigation route planning text according to the navigation starting point and end point input by the user, and matching the navigation route planning text with the risk point information to obtain information items that need to be pre-reminded on the navigation route planning;
[0094] Specifically, this embodiment uses the map API interface to obtain the driving navigation route planning text, and then extracts the intersection section name and map information appearing in the planning text, and retrieves the matching risk point information data with the location name and longitude and latitude information to obtain the information items that need to be pre-reminder on the planned route, which include:
[0095] Step S2.1, the map software generates a driving navigation route planning text according to the navigation starting point and end point input by the user;
[0096] refer to Figure 3 ,This embodiment uses the API interface opened by Amap to obtain the text content corresponding to the driving navigation path;
[0097] After the user enters the starting and ending points of the navigation, the map software will automatically generate a navigation path, obtain the driving navigation path book through code, and output it as a txt file.
[0098] Step S2.2, extracting location information from the driving navigation route planning text;
[0099] Automatically extract information about locations in the navigation path, including intersection section names, map information points (such as a shopping mall, an overpass, etc.), and store the extracted information by entry.
[0100] Step S2.3: Match the location information in the driving navigation route planning text with the risk point information to obtain information items that require advance reminders on the navigation route planning.
[0101] The pre-warning message with location information and time information obtained by the driving experience sharing information module is accurately matched according to the place names appearing in the planned route and retrieved one by one.
[0102] If the same location information entry exists in the risk point information obtained in the driving experience sharing information module, it will be written into the "road warning information.csv" file;
[0103] If the same entry does not exist in the risk point information file, it will not be written to the "road pre-warning information.csv" file.
[0104] Finally, the matched road pre-warning information .csv file is obtained, which is used to generate large model prompt words.
[0105] Step S3: extract the information items that need to be pre-reminded on the route planning, and use the information items as pre-remind point information, and generate guidance prompt words according to the pre-remind point information, which specifically includes the following contents:
[0106] Setting refers to the large model prompt word template, which consists of three main parts: system message prompt part S, road pre-warning information P, and text input part X;
[0107] 1. System message prompt part S
[0108] The system message prompt words input to the large language model are used to reduce the diversity of the large language model's answer results and improve the accuracy of the model's answer. For example: "You are a route navigation pre-alert assistant. You can write a route navigation forecast based on the information provided by the following accident risk point information." Stored in the S set, such as S = {"You are a route navigation pre-alert assistant...", "XXX", ...}
[0109] Please make a route navigation forecast based on the following accident risk point information and the planned navigation route.
[0110] 2. Road warning information P
[0111] The road pre-warning information set generated by the planned path risk point information matching module, that is, the risk point information set on the planned path, is stored in the "csv" file format.
[0112] 3. Text input part X
[0113] The text set entered by the user is stored in the computer as a set x, such as: x = {"What should I pay attention to on the road?", "What are the traffic regulations and precautions on the route?"...};
[0114] Specifically, when the user enters the start and end point information in the navigation form, the map window will automatically draw the map, and the planned path risk point information matching module will automatically search for risk point information on the planned path. If there is accident risk point information, it will be extracted and stored in the "road pre-warning information.csv" file, and a prompt word T will be generated according to the program for interaction with the fine-tuned Tra-GPT;
[0115] The specific prompt word T is generated by combining the following contents:
[0116] T=S+P+X
[0117] Assume a scenario where a lively route navigation pre-reminder text is received before navigation begins. The prompt template is as follows:
[0118]
[0119] Step S4: construct a traffic regulations model, input the guidance prompt words into the traffic regulations model, and output the pre-warning navigation text, which specifically includes the following contents:
[0120] Step S4.1, construct a traffic law dataset based on the collected traffic safety regulations, traffic driving rules, and traffic violation penalty points rules, and annotate the traffic law dataset to obtain a question-and-answer dataset corresponding to traffic safety knowledge;
[0121] Specifically, the data sets required in this embodiment include traffic safety regulations data sets, traffic driving rules, Traffic violation penalty rule dataset and other data content related to traffic safety. Based on the collected information, the original text data is manually annotated to obtain the question-and-answer dataset corresponding to traffic safety knowledge. The dataset format used in this embodiment is the conservation format displayed on the OpenAI official website. The dataset format and the content of each dataset are as follows.
[0122] Traffic safety regulations dataset: A dataset of various traffic safety regulations that have been manually annotated, such as the Road Traffic Safety Law of the People's Republic of China and the Regulations for the Implementation of the Road Traffic Safety Law of the People's Republic of China. The annotation format of the dataset is:
[0123] {"messages":[{"role":"system","content":"Marv is a factual chatbotthat is also sarcastic."},{"role":"user","content":"What's the capital of France?"},{"role":"assistant","content":"Paris,as if everyone doesn't know that already."}]}
[0124] "role" is the role of the dialogue, which is divided into system, user, and assistant attributes;
[0125] "content" refers to the corresponding content. Different roles correspond to different contents.
[0126] When role=system, content corresponds to the fine-tuned system prompt word, which is used to help the large model identify the fine-tuned data set and narrow the text range of the answer;
[0127] When role=user, content corresponds to the question raised by the annotated data, which is the content entered by the user;
[0128] When role=assistant, content corresponds to the knowledge points corresponding to the annotated data, which is the content answered by the large model.
[0129] Each piece of data is stored in a csv file, and each piece of data occupies one line.
[0130] Among them, all data sets used for large model training are stored and filled in this format.
[0131] Traffic violation penalty rule annotation dataset: the penalty rules corresponding to various traffic driving violations, as well as other rules related to driver's license points.
[0132] Traffic driving rules dataset: various types of traffic driving rules datasets that have been manually annotated.
[0133] Step S4.2, performing format standardization processing on the data in the question-and-answer dataset corresponding to traffic safety knowledge;
[0134] For the generated dataset, error checking is performed to verify that each dialogue in the dataset conforms to the format of the fine-tuning dataset to improve the accuracy and precision of fine-tuning.
[0135] Format standardization processing includes: data type check, message missing check, role information check, and content format verification.
[0136] Data type checking refers to checking whether each entry in the data set is a dictionary (dict);
[0137] Message missing check refers to checking whether the message content corresponding to each role in the dataset message is missing;
[0138] Role information check refers to checking whether the role of each entry in the data set is one of system, user, and assistant;
[0139] Content format verification means checking that each entry in the data set has at least one message from the assistant.
[0140] The specific implementation is implemented in Python code.
[0141] Step S4.3: Use the large language model to fine-tune the question-answer data set corresponding to the traffic safety knowledge after the format is standardized to obtain a large model of traffic regulations:
[0142] Specifically, use the API provided by OpenAI to fine-tune the driving experience language model according to the process of "uploading datasets, creating fine-tuning tasks, and using fine-tuning models".
[0143] Specifically, uploading a dataset includes uploading a training dataset and a validation dataset;
[0144] Create a fine-tuning task: Use code to select different language models for fine-tuning and set appropriate hyperparameters (learning rate multiplier, number of training epochs, and training batch size).
[0145] Use the fine-tuned model: Call the trained fine-tuned model through Python code, observe the text generation effect of the fine-tuned model, and use different hyperparameter combinations to achieve the optimal training effect.
[0146] The model is fine-tuned to obtain the fine-tuned driving experience large model question and answer GPT, that is, the traffic regulations large model of this embodiment is obtained.
[0147] Step S5: generating a pre-alert navigation window for displaying the pre-alert navigation text information according to the pre-alert navigation text, which specifically includes:
[0148] Manually input the starting point and end point through the API and underlying code of Amap; or click the mouse on the map window on the right to determine the starting point and end point; or obtain the current vehicle GPS positioning information to input the starting point of the route;
[0149] Click "Generate pre-warning navigation text with one click" to generate pre-warning navigation text based on driving experience on the route in the text box.
[0150] Although the specific implementation of the invention is described in detail in conjunction with the drawings, it should not be understood as limiting the scope of protection of this patent. Within the scope described in the claims, various modifications and variations that can be made by those skilled in the art without creative work still fall within the scope of protection of this patent.
Claims
1. A method for generating a driving experience reminder navigation text generation system, It is characterized in that The following steps are involved: S1. Extract risk point information on the road based on historical traffic accident information and driving experience sharing information; S2. According to the navigation starting point and end point input by the user, a driving navigation route planning text is obtained, and the navigation route planning text is matched with the risk point information to obtain information items that need to be pre-reminded on the navigation route planning; S3, extracting information items that need to be pre-remindered on the route planning, and using the information items as pre-reminder point information, and generating guidance prompt words according to the pre-reminder point information; S4, constructing a traffic regulations model, inputting the guidance prompt words into the traffic regulations model, and outputting a pre-warning navigation text; S5. Generate a pre-reminder navigation window for displaying the pre-reminder navigation text information according to the pre-reminder navigation text; The step S1 comprises the following steps: S1.
1. Collect and annotate historical traffic accident information and driving experience sharing information; S1.
2. Preprocess the annotated historical traffic accident information and driving experience sharing information to extract risk point information on the driving road; The driving experience sharing information in step S1.1 includes: road surface disease information, roadside construction, road spillage, pedestrian-dense areas and vehicle reversal high-incidence points; The guidance prompt word T in step S3 is: T=S+P+X Among them, S is the system message prompt part, P is the road warning information, and X is the text input; The step S4 constructs a large traffic regulation model, including: S4.
1. Construct a traffic law dataset based on the collected traffic safety regulations, traffic driving rules, and traffic violation penalty points rules, and annotate the traffic law dataset to obtain a question-and-answer dataset corresponding to traffic safety knowledge; S4.
2. Standardize the format of the data in the question-and-answer dataset corresponding to traffic safety knowledge; S4.
3. Use the large language model to fine-tune the question-answering dataset corresponding to traffic safety knowledge after the format is standardized to obtain a large model of traffic regulations; Use the API provided by OpenAI to fine-tune according to the process of "upload dataset, create fine-tuning task, use fine-tuning model"; Uploading a dataset includes uploading a training dataset and a validation dataset; Create a fine-tuning task: Use code to select different language models for fine-tuning and set appropriate hyperparameters, including learning rate multiplier, number of training epochs, and training batch size. Use fine-tuned model: Call the trained fine-tuned language model through Python code, observe the text generation effect of the fine-tuned language model, and use different hyperparameter combinations to achieve the optimal training effect; Fine-tune the language model to obtain the fine-tuned driving experience large model question and answer GPT, that is, the traffic regulations large model.
2. The method for generating a driving experience reminder navigation text generation system according to claim 1, It is characterized in that The step S1.2 comprises the following steps: S1.2.
1. Privacy protection and data cleaning of historical traffic accident information and driving experience sharing information; S1.2.
2. Standardize the historical traffic accident information and driving experience sharing information after privacy protection and data cleaning; S1.2.
3. Extract characteristic information from standardized historical traffic accident information and driving experience sharing information; S1.2.
4. Construct search keywords based on the location information and time information in the feature information to obtain risk point information.
3. The method for generating a driving experience reminder navigation text generation system according to claim 1, It is characterized in that The step S2 comprises the following steps: S2.
1. Based on the navigation start point and end point input by the user, the map software generates a driving navigation route planning text; S2.2, extracting location information from the driving navigation route planning text; S2.
3. Match the location information in the driving navigation route planning text with the risk point information to obtain information items that require advance reminders on the navigation route planning.
4. A driving experience reminder navigation text generation system, It is characterized in that include: The driving experience sharing information preprocessing module is used to extract risk point information on the driving road based on historical traffic accident information and driving experience sharing information, which specifically includes: Collect and annotate historical traffic accident information and driving experience sharing information; driving experience sharing information includes: road surface disease information, roadside construction, road spillage, pedestrian-dense areas and vehicle reversal high-incidence points; Pre-process the annotated historical traffic accident information and driving experience sharing information to extract risk point information on the driving road; The planning path risk point information matching module is used to obtain the driving navigation path planning text according to the navigation starting point and end point input by the user, and to match the navigation path planning text with the risk point information in the driving experience sharing information preprocessing module to obtain the information items that need to be pre-reminded on the navigation path planning; The large model prompt word module is used to extract the information items that need to be pre-reminded on the path planning and use the information items as pre-remind point information, and generate guidance prompt words according to the pre-remind point information; the guidance prompt word T is: T=S+P+X Among them, S is the system message prompt part, P is the road warning information, and X is the text input; The traffic safety knowledge fine-tuning large model module is used to construct a traffic regulations large model, input the guidance prompt words into the traffic regulations large model, and output the pre-warning navigation text, which includes: Based on the collected traffic safety laws, traffic driving rules, and traffic violation penalty points rules, a traffic law dataset is constructed, and the traffic law dataset is annotated to obtain a question-and-answer dataset corresponding to traffic safety knowledge; Standardize the format of the data in the question-and-answer dataset corresponding to traffic safety knowledge; A large language model is used to fine-tune the question-answering dataset corresponding to traffic safety knowledge after the format is standardized to obtain a large model of traffic regulations; Specifically, use the API provided by OpenAI to perform fine-tuning according to the process of "upload dataset, create fine-tuning task, and use fine-tuning model"; Uploading a dataset includes uploading a training dataset and a validation dataset; Create a fine-tuning task: Use code to select different language models for fine-tuning and set appropriate hyperparameters, including learning rate multiplier, number of training epochs, and training batch size. Use fine-tuned model: Call the trained fine-tuned language model through Python code, observe the text generation effect of the fine-tuned language model, and use different hyperparameter combinations to achieve the optimal training effect; Fine-tune the language model to obtain the fine-tuned driving experience large model question-answer GPT, that is, the traffic regulations large model; The pre-reminder navigation form module generates a pre-reminder navigation form for displaying the pre-reminder navigation text.
