A navigation method for identifying and correcting place names, a device, an electronic device and a medium

Abstract

The present application relates to a kind of navigation identification place name correction method, device, electronic equipment and medium, belong to speech recognition navigation technical field, wherein, the method includes: based on the complete intent detection model of training, the classification of the language recognition result obtained, obtain user intent type, user intent type includes no correction intent and correction intent, intent detection model uses large language model;When user intent type is correction intent, extract initial place name from language recognition result, based on initial place name, in the matching of pre-set geographic knowledge base, obtain target place name, the target place name is as the final navigation identification place.This application carries out intent speculation to user speech recognition data by intent detection model, only when user has correction intent, subsequent rewriting operation is carried out, user no correction intent does not carry out any rewriting operation, avoid unnecessary computing overhead and additional time-consuming.

Description

Technical Field

[0001] This invention relates to the field of voice recognition and navigation technology, and in particular to a navigation recognition place name correction method, device, electronic device, and medium. Background Technology

[0002] With the advent of the era of intelligent vehicles, using in-car navigation via voice interaction has become a driving habit. Currently, users navigate to a wide variety of destinations, ranging from shopping malls and restaurants to tourist attractions and hotels. However, existing voice recognition systems often misrecognize unfamiliar place names, especially when the characters in the name are unrelated. They are easily misidentified as homophones or similar-sounding words, leading to a decrease in overall recognition accuracy. In such cases, the navigated destination often deviates from the user's expectations, causing dissatisfaction.

[0003] Existing solutions include map entity database matching technology, the main steps of which include: obtaining speech recognition results, extracting the domain and entities to be corrected from the speech recognition results, and when the domain belongs to the navigation domain and the number of characters in the entity to be corrected is greater than or equal to a preset number of characters, correcting the entity to be corrected according to the map entity database to complete navigation entity correction.

[0004] However, in practical applications, not all text needs to be corrected. Correction is only required when the user intends to correct it. The existing technology mentioned above cannot determine when correction is needed. Regardless of whether the user intends to correct it, as long as the number of characters of the entity to be corrected is greater than or equal to the preset number of characters, the map database matching and correction are triggered. The lack of real-time quality assessment of speech recognition results makes it impossible for the system to intelligently determine when correction is needed, resulting in unnecessary computational overhead and additional time consumption. Summary of the Invention

[0005] In view of this, it is necessary to provide a navigation identification place name correction method, device, electronic device and medium to solve the technical problem in the prior art that the system cannot intelligently determine when error correction is needed, resulting in unnecessary computational overhead and additional time consumption.

[0006] To address the aforementioned problems, in a first aspect, the present invention provides a navigation identification place name correction method, comprising: Based on a fully trained intent detection model, the acquired language recognition results are classified to obtain user intent types, which include no-correction intent and corrective intent. The intent detection model adopts a large language model. When the user intent type is to correct, an initial location name is extracted from the language recognition result. Based on the initial location name, a target location name is matched in a preset geographical knowledge base, and the target location name is used as the final navigation recognition location.

[0007] In one possible implementation, the corrective intent includes both user-initiated corrective intent and navigation-reflective corrective intent. When the user intent type is a corrective intent, the initial location name is extracted from the language recognition result, including: When the user intent type is a user-initiated correction intent, the language recognition result contains user correction information. The language recognition result is rewritten based on a well-trained correction rewriting model to obtain a rewritten language recognition result. The initial location name is extracted from the rewritten language recognition result based on a well-trained entity extraction model. When the user intent type is navigation reflection and correction intent, the language recognition result does not contain user correction information. The initial location name is extracted from the language recognition result based on the fully trained entity extraction model. The correction rewriting model and the entity extraction model adopt a large language model.

[0008] In one possible implementation, the fully trained intent detection model, the fully trained corrected rewriting model, and the fully trained entity extraction model are obtained through the following steps: Acquire historical accumulated data and classify the historical accumulated data into user-initiated correction data where the user has the intention to correct and provides correction text, navigation reflection correction data where the user has the intention to correct but does not provide correction text, and other irrelevant data where the user has no intention to correct. The user-initiated correction data, navigation reflection correction data, and other irrelevant data are generalized and amplified using a well-trained large language model to generate structured data for intent detection, structured data for correction and rewriting, and structured data for entity extraction. The structured data for intent detection includes intent prompts, multi-turn dialogues, and intent tags. The structured data for correction and rewriting includes rewritten prompts, multi-turn dialogues, and user correction text. The structured data for entity extraction includes extracted prompts, multi-turn dialogues, and navigation operations plus location entities. The initial intent detection model is trained based on the structured intent detection data to obtain a fully trained intent detection model. The initial correction and rewriting model is trained based on the structured correction and rewriting data to obtain a fully trained correction and rewriting model. The initial entity extraction model is trained based on the structured entity extraction data to obtain a fully trained entity extraction model.

[0009] In one possible implementation, a well-trained large language model is used to generalize and amplify the user-initiated correction data, the navigation reflection correction data, and the other irrelevant data. The large language model in this process is a large language model with a large number of parameters.

[0010] In one possible implementation, the intent detection model, the corrective rewriting model, and the entity extraction model employ a large language model with a small number of parameters.

[0011] In one possible implementation, based on the initial location name, the target location name is matched in a preset geographic knowledge base, including: Convert the initial location name into Pinyin data; Based on the user's current latitude and longitude, city, and the obtained pinyin data, a matching query is performed on the preset geographical knowledge base to obtain the target location name.

[0012] In one possible implementation, the geographic knowledge base is a dynamically updated online geographic knowledge base.

[0013] Secondly, the present invention also provides a navigation identification place name correction device, comprising: The intent detection module is used to classify the acquired language recognition results based on a fully trained intent detection model to obtain user intent types. The user intent types include uncorrected intent and corrected intent. The intent detection model adopts a large language model. The entity extraction module is used to extract the initial location name from the language recognition result when the user intent type is a correction intent; The database matching module is used to match the target location name in a preset geographical knowledge base based on the initial location name, and use the target location name as the final navigation identification location.

[0014] Thirdly, the present invention also provides an electronic device, including a memory and a processor; The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps of the navigation identification place name correction method described above.

[0015] Fourthly, the present invention also provides a computer-readable storage medium storing a program or instructions that, when executed by a processor, implement the steps of the navigation identification place name correction method described above.

[0016] The beneficial effects of the present invention are as follows: The navigation recognition place name correction method provided by the present invention uses an intent detection model to infer the intent of the user's speech recognition data, thereby achieving real-time quality assessment of the language recognition results. As a result, subsequent rewriting operations are only performed when the user has a correction intent. When the user has no correction intent, the original text is returned directly without any rewriting operation, thus avoiding unnecessary computational overhead and additional time consumption.

[0017] Furthermore, this invention applies a large model to the field of navigation recognition and place name correction, which can improve the ability to correct navigation and reduce misjudgments and incorrect corrections. It can also adapt well to complex scenarios. By introducing a large model, it can capture deeper semantic information, thereby improving the accuracy of recognizing and correcting navigation correction intentions. At the same time, the powerful generalization ability of the large model enables the system to learn and adapt quickly when faced with more complex intentions, reducing the possibility of misjudgments. Attached Figure Description

[0018] Figure 1 This is a schematic flowchart of an embodiment of the navigation identification place name correction method provided by the present invention; Figure 2 For the present invention Figure 1 A schematic diagram of an embodiment of step S102; Figure 3 A schematic flowchart of an embodiment of the model training steps provided by the present invention; Figure 4 For the present invention Figure 1 A schematic flowchart of an embodiment of step S103; Figure 5 A schematic diagram of an embodiment of the navigation recognition place name correction device provided by the present invention; Figure 6 A schematic diagram of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] It should be understood that the illustrative drawings are not drawn to scale. The flowcharts used in this invention illustrate operations implemented according to some embodiments of the invention. It should be understood that the operations in the flowcharts may be implemented out of order, and steps without logical contextual relationships may be reversed or performed simultaneously. Furthermore, those skilled in the art, guided by the content of this invention, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor systems and / or microcontroller systems.

[0021] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. "And / or" describes the relationship between related objects, indicating that three relationships may exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone.

[0022] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0023] This invention provides a navigation identification place name correction method, device, electronic device, and medium, which are described below.

[0024] Figure 1 This is a schematic flowchart of an embodiment of the navigation identification and place name correction method provided by the present invention, as shown below. Figure 1 As shown, the navigation recognition place name correction method includes: S101. Based on the fully trained intent detection model, the acquired language recognition results are classified to obtain the user intent type. The user intent type includes no correction intent and correction intent. The intent detection model adopts a large language model. It should be noted that large language models are deep learning models based on the Transformer architecture, such as LLM. They are trained on massive amounts of text and can perform natural language processing tasks such as generation, classification, and summarization. Their core capability lies in predicting the next word and understanding contextual relationships.

[0025] S102. When the user's intent type is "correction intent", extract the initial location name from the language recognition result. Considering that in practical applications, users may actively point out which character is incorrect and provide the correct corresponding character, for example, in the sentence "Navigate to Pingjin Center, Jin is the Jin of gold," if the corresponding character is not corrected, the entity extraction of the language recognition results may easily lead to errors in the model's context understanding, affecting the accuracy of entity extraction. Therefore, to improve the accuracy of entity extraction, in some embodiments of this invention, step S101 includes correction intent, which includes user-initiated correction intent and navigation reflection correction intent. Thus, a trained intent detection model is used to classify the intent of the user's input query statement to determine whether it contains a correction request. The model outputs three possible classification results: no correction intent, user-initiated correction intent, and navigation reflection correction intent. No correction intent refers to a user query that is a regular navigation request or irrelevant instruction, such as directly inputting "Navigate to Beijing West Railway Station"; user-initiated correction intent specifically refers to the situation where the user explicitly points out the error and provides the correct text, typically expressed as "Navigate to Pingjin Center, Jin is the Jin of gold"; while navigation reflection correction intent covers the user's expression of negative feedback on the navigation result or attempts to correct it.

[0026] Correspondingly, such as Figure 2 As shown, step S102 specifically includes: S1021. When the user intent type is user active correction intent, the language recognition result contains user correction information. The language recognition result is rewritten based on the well-trained correction rewriting model to obtain the rewritten language recognition result. The initial location name is extracted from the rewritten language recognition result based on the well-trained entity extraction model. Specifically, the system will invoke the correction and rewriting model for queries that are determined to be user-initiated corrections. This correction and rewriting model performs semantic reconstruction on the original text based on the error correction information provided by the user, such as automatically correcting "Pingjin Center" to "Pingjin Center", and finally outputting the rewritten standard text.

[0027] S1022. When the user intent type is navigation reflection and correction intent, the language recognition result does not contain user correction information. The initial location name is extracted from the language recognition result based on the fully trained entity extraction model. The correction rewriting model and the entity extraction model adopt a large language model.

[0028] Specifically, the system processes the query for navigation reflection and correction intent and the text rewritten by the correction and rewriting model. It analyzes the query content through the entity extraction model and separates the user's navigation operation instructions (such as "go" and "navigate to") and the location name to be corrected (such as "Golden Eagle International").

[0029] To reduce latency, in some embodiments of this invention, a large language model with a small number of parameters is used to train three specialized models: an intent detection model (outputting no correction intent / user-initiated correction intent / navigation reflection correction intent). Figure 3 The model includes three sub-models: a class model, a correction and rewriting model (handling explicit user corrections), and an entity extraction model (parsing navigation operations and location names).

[0030] It's important to note that the use of a large language model is specifically due to the fact that users' phrasing is often quite generalized, not limited to just "navigate to xxx." Therefore, rule-based entity extraction methods struggle to handle complex sentence structures. This invention introduces a large language model to extract entities within the context of the sentence. Utilizing the powerful reasoning capabilities of the large model, it extracts and processes complex entities, effectively extracting generalized entity content. After extensive training on human preferences, the large model possesses strong text understanding capabilities, easily and accurately extracting user navigation actions and location names from complex text. It can also understand whether the user needs to correct location names based on their input. Furthermore, through carefully designed prompts and fine-tuning strategies, the model's prior knowledge is more fully activated, leading to more accurate identification and correction of incorrect location names. This strengthens the large model's ability to recognize correction intentions and extract entities, enabling it to better understand the user's context and complete the task of navigation identification and location correction.

[0031] In some embodiments of the present invention, such as Figure 3 As shown, the fully trained intent detection model, the fully trained correction and rewriting model, and the fully trained entity extraction model are obtained through the following steps: S201. Obtain historical accumulated data and classify it into user-initiated correction data where the user has the intention to correct and provides correction text, navigation reflection correction data where the user has the intention to correct but does not provide correction text, and other irrelevant data where the user has no intention to correct. It should be noted that, firstly, the historical accumulated data is cleaned and categorized. A clustering model is used to classify and summarize the historical navigation data and correction data into three types: user-initiated correction data (users actively point out which word is wrong and provide the corresponding correct word, for example: "Navigate to Pingjin Center, Jin is the Jin in Jinzi"), navigation reflection correction data (users' negative feedback on navigation, such as negative feedback after navigation failure, or repeatedly saying the same navigation location in an attempt to correct it), and other irrelevant data (data that does not require navigation correction, such as: normal navigation data, casual conversation data, vehicle control command data, etc.). Corresponding definitions and prompts are written for each category through manual annotation to provide basic data support for subsequent steps.

[0032] S202. Use a fully trained large language model to generalize and amplify user-initiated correction data, navigation reflection correction data, and other irrelevant data, and generate structured data for intent detection, correction rewriting, and entity extraction. The structured data for intent detection includes intent prompts, multi-turn dialogues, and intent labels. The structured data for correction rewriting includes rewriting prompts, multi-turn dialogues, and user correction text. The structured data for entity extraction includes extraction prompts, multi-turn dialogues, and navigation operations plus location entities. It should be noted that, for intent detection, the structured data includes: the prompts from the intent detection model, the user's first-round question Q1, the system's first-round response A1, the user's second-round question Q2, the system's second-round response Q2, and so on up to the user's Nth-round question Qn, and the category of intent; for the correction and rewriting model, the structured data includes: the prompts from the correction and rewriting model, the user's first-round question Q1, the system's first-round response A1, the user's second-round question Q2, the system's second-round response Q2, and so on up to the user's Nth-round question Qn, and the corrected rewritten text; for the entity extraction model, the structured data includes: the prompts from the entity extraction model, the user's first-round question Q1, the system's first-round response A1, the user's second-round question Q2, the system's second-round response Q2, and so on up to the user's Nth-round question Qn, and the extracted navigation operations and location names.

[0033] It should be noted that the large language model used in the generalization amplification of user-initiated correction data, navigation reflection correction data, and other irrelevant data using a fully trained large language model is a large language model with a large number of parameters.

[0034] S203. Train the initial intent detection model based on the structured data of intent detection to obtain a fully trained intent detection model. Train the initial correction and rewriting model based on the structured data of correction and rewriting to obtain a fully trained correction and rewriting model. Train the initial entity extraction model based on the structured data of entity extraction to obtain a fully trained entity extraction model.

[0035] S103. Based on the initial location name, the target location name is matched in the preset geographical knowledge base, and the target location name is used as the final navigation identification location.

[0036] Considering that searching a geographic database using only Pinyin might result in duplicate names, making it impossible to select the correct entity, some embodiments of this invention employ a search method combining latitude and longitude, city, and Pinyin. This effectively achieves a generalized search for location names. Specifically, for example... Figure 4As shown in the figure, step S103 specifically includes: S1031. Convert the initial location name into pinyin data; Specifically, perform pinyin processing on the extracted location name, and use the pinyin library to convert Chinese characters into pinyin strings. For example, "Golden Eagle International" is converted into "jin ying guo ji".

[0037] S1032. Based on the obtained current longitude and latitude of the user, the city where the user is located, and the pinyin data, perform a matching query in the preset geographic knowledge base to obtain the target location name.

[0038] Considering that the fixed geographic knowledge base depends on the entity entries in the database and cannot recognize rare place names or emerging locations that are not included, in some embodiments of the invention, the geographic knowledge base is a dynamically updated online geographic knowledge base.

[0039] Specifically, perform geographic retrieval by combining spatio-temporal information. The system obtains the longitude and latitude coordinates of the user's current location and the city where the user is located, and submits them together with the pinyin of the location name to the dynamically updated online geographic knowledge base (such as a navigation API) to perform a POI (Point of Interest, specifically referring to the entity location with specific functions or meanings in a map or geographic information system) matching query to obtain the searched results. Among them, the navigation API (Application Programming Interface) can continuously and dynamically update new POIs.

[0040] It should be noted that the present invention only corrects the language recognition results with the intention of correction, directly returns the original text for those without the intention of correction without any rewriting operations. For those with the intention of correction, the user's initial navigation operation (such as "navigate to") is spliced with the finally corrected target location name (such as "Golden Eagle International Plaza") to generate a navigation instruction that conforms to the user's true intention and returns it.

[0041] Compared with the prior art, the present invention uses an intention detection model to speculate on the intention of the user's speech recognition data, realizes real-time quality assessment of the language recognition results, and then only performs subsequent rewriting operations when the user has the intention of correction. When the user has no intention of correction, the original text is directly returned without any rewriting operations, avoiding unnecessary computational overhead and additional time consumption.

[0042] Furthermore, this invention applies a large model to the field of navigation recognition and place name correction, which can improve the ability to correct navigation and reduce misjudgments and incorrect corrections. It can also adapt well to complex scenarios. By introducing a large model, it can capture deeper semantic information, thereby improving the accuracy of recognizing and correcting navigation correction intentions. At the same time, the powerful generalization ability of the large model enables the system to learn and adapt quickly when faced with more complex intentions, reducing the possibility of misjudgments.

[0043] The present invention is interested in better implementing a navigation identification place name correction method according to an embodiment of the present invention. Based on this method, correspondingly, as shown below... Figure 5 As shown, this embodiment of the invention also provides a navigation recognition place name correction device 300, comprising: The intent detection module 301 is used to classify the acquired language recognition results based on the fully trained intent detection model to obtain the user intent type. The user intent type includes no correction intent and correction intent. The intent detection model adopts a large language model. The entity extraction module 302 is used to extract the initial location name from the language recognition result when the user intent type is a correction intent; The database matching module 303 is used to match the target location name in a preset geographic knowledge base based on the initial location name, and use the target location name as the final navigation identification location.

[0044] The navigation identification place name correction device 300 provided in the above embodiments can realize the technical solutions described in the above navigation identification place name correction method embodiments. The specific implementation principles of each unit can be found in the corresponding content in the above navigation identification place name correction method embodiments, and will not be repeated here.

[0045] like Figure 6 As shown, the present invention also provides an electronic device 400. The electronic device 400 includes a processor 401, a memory 402, and a display 403. Figure 6 Only some components of the electronic device 400 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.

[0046] In some embodiments, memory 402 may be an internal storage unit of electronic device 400, such as a hard disk or memory of electronic device 400. In other embodiments, memory 402 may also be an external storage device of electronic device 400, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 400.

[0047] Furthermore, the memory 402 may include both internal storage units of the electronic device 400 and external storage devices. The memory 402 is used to store application software and various types of data installed on the electronic device 400.

[0048] In some embodiments, processor 401 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 402 or process data, such as the navigation identification place name correction method of the present invention.

[0049] In some embodiments, display 403 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 403 is used to display information from electronic device 400 and to display a visual user interface. Components 401-403 of electronic device 400 communicate with each other via a system bus.

[0050] In some embodiments of the present invention, when the processor 401 executes the navigation identification place name correction program in the memory 402, the following steps can be implemented: Based on the fully trained intent detection model, the acquired language recognition results are classified to obtain the user intent type, which includes no corrective intent and corrective intent. The intent detection model adopts a large language model. When the user's intent type is to correct, the initial location name is extracted from the language recognition results. Based on the initial location name, the target location name is matched in the preset geographical knowledge base and used as the final navigation recognition location.

[0051] It should be understood that when the processor 401 executes the navigation identification place name correction program in the memory 402, in addition to the functions mentioned above, it can also perform other functions, as can be found in the description of the corresponding method embodiments above.

[0052] Furthermore, this embodiment of the invention does not specifically limit the type of electronic device 400 mentioned. Electronic device 400 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the invention, electronic device 400 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0053] Accordingly, embodiments of the present invention also provide a computer-readable storage medium for storing computer-readable programs or instructions. When the programs or instructions are executed by a processor, they can implement the steps or functions of the navigation identification place name correction methods provided in the above-described method embodiments.

[0054] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0055] The above provides a detailed description of a navigation identification and place name correction method provided by the present invention. Specific examples have been used to illustrate the principle and implementation of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation and application scope based on the idea of ​​the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

[0056] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for correcting place names in navigation identification, characterized in that, include: Based on a fully trained intent detection model, the acquired language recognition results are classified to obtain user intent types, which include no-correction intent and corrective intent. The intent detection model adopts a large language model. When the user intent type is to correct, an initial location name is extracted from the language recognition result. Based on the initial location name, a target location name is matched in a preset geographical knowledge base, and the target location name is used as the final navigation recognition location.

2. The navigation identification place name correction method according to claim 1, characterized in that, The corrective intent includes both user-initiated corrective intent and navigation-reflective corrective intent. When the user intent type is corrective intent, the initial location name is extracted from the language recognition result, including: When the user intent type is a user-initiated correction intent, the language recognition result contains user correction information. The language recognition result is rewritten based on a well-trained correction rewriting model to obtain a rewritten language recognition result. The initial location name is extracted from the rewritten language recognition result based on a well-trained entity extraction model. When the user intent type is navigation reflection and correction intent, the language recognition result does not contain user correction information. The initial location name is extracted from the language recognition result based on the fully trained entity extraction model. The correction rewriting model and the entity extraction model adopt a large language model.

3. The navigation identification place name correction method according to claim 2, characterized in that, The fully trained intent detection model, the fully trained corrected rewriting model, and the fully trained entity extraction model are obtained through the following steps: Acquire historical accumulated data and classify the historical accumulated data into user-initiated correction data where the user has the intention to correct and provides correction text, navigation reflection correction data where the user has the intention to correct but does not provide correction text, and other irrelevant data where the user has no intention to correct. The user-initiated correction data, navigation reflection correction data, and other irrelevant data are generalized and amplified using a well-trained large language model to generate structured data for intent detection, structured data for correction and rewriting, and structured data for entity extraction. The structured data for intent detection includes intent prompts, multi-turn dialogues, and intent tags. The structured data for correction and rewriting includes rewritten prompts, multi-turn dialogues, and user correction text. The structured data for entity extraction includes extracted prompts, multi-turn dialogues, and navigation operations plus location entities. The initial intent detection model is trained based on the structured intent detection data to obtain a fully trained intent detection model. The initial correction and rewriting model is trained based on the structured correction and rewriting data to obtain a fully trained correction and rewriting model. The initial entity extraction model is trained based on the structured entity extraction data to obtain a fully trained entity extraction model.

4. The navigation identification place name correction method according to claim 3, characterized in that, The large language model used in the generalization amplification of the user-initiated correction data, the navigation reflection correction data, and the other irrelevant data using a fully trained large language model is a large language model with a large number of parameters.

5. The navigation identification place name correction method according to claim 3, characterized in that, The intent detection model, correction and rewriting model, and entity extraction model employ a large language model with small parameters.

6. The navigation identification place name correction method according to claim 1, characterized in that, Based on the initial location name, the target location name is obtained by matching it in a preset geographical knowledge base, including: Convert the initial location name into Pinyin data; Based on the user's current latitude and longitude, city, and the obtained pinyin data, a matching query is performed on the preset geographical knowledge base to obtain the target location name.

7. The navigation identification place name correction method according to claim 1, characterized in that, The geographical knowledge base is a dynamically updated online geographical knowledge base.

8. A navigation recognition place name correction device, characterized in that, include: The intent detection module is used to classify the acquired language recognition results based on a fully trained intent detection model to obtain user intent types. The user intent types include uncorrected intent and corrected intent. The intent detection model adopts a large language model. The entity extraction module is used to extract the initial location name from the language recognition result when the user intent type is a correction intent; The database matching module is used to match the target location name in a preset geographical knowledge base based on the initial location name, and use the target location name as the final navigation identification location.

9. An electronic device, characterized in that, Including memory and processor; The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps of the navigation identification place name correction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the navigation identification place name correction method according to any one of claims 1 to 7.

Images