Page jump method of voice instruction based on NLP

Abstract

The invention provides an NLP-based voice instruction page jumping method, which comprises the following steps: receiving a voice instruction input by a user on a display device, and analyzing voice information carried by the instruction into text information through a voice recognition technology; matching the text information with a preset page keyword list, and skipping to the page when the text information contains keywords of any page; and when the text information is not matched with a preset page keyword list, sending the text information and the page configuration information to an LLM (Label Language Model), and outputting a result and a confidence score based on the LLM: if the confidence score is greater than or equal to a first score, skipping to a corresponding page. According to the page skipping method provided by the invention, a more convenient and intelligent voice interaction mode is provided for the user, the user demand can be responded more efficiently whether page skipping or dialogue chatting is performed, and diversified use scenes of the user are met.

Description

Technical Field

[0001] This application relates to the field of page navigation technology, and more particularly to a page navigation method based on NLP voice commands. Background Technology

[0002] In today's mobile internet era, smartphones have become deeply integrated into people's daily lives and work, becoming indispensable portable smart terminals. Users use smartphones to perform diverse operations such as information retrieval, social communication, mobile payment, and entertainment, placing higher demands on the convenience and efficiency of human-computer interaction. Among these, voice interaction, with its characteristic of requiring no manual operation, demonstrates significant advantages in scenarios such as driving, sports, and busy hands, and is gradually becoming an important way for users to interact with their phones, greatly enriching the usage scenarios and interactive experiences of mobile applications.

[0003] Existing mobile voice interaction systems typically implement basic functions through the following process: After receiving the user's voice input, the system converts it into text information using speech recognition technology. Then, based on preset keyword matching rules, it determines the user's intent and executes the corresponding operation (such as page navigation). For example, when a user issues the voice command "open settings," the system recognizes the keyword "settings," triggers a page navigation, and completes the user's desired operation.

[0004] However, existing technologies still have significant shortcomings in practical applications: On the one hand, the keyword matching rules relied upon for page navigation are relatively simple, usually only pre-setting a few fixed keywords, which is difficult to cover the diverse ways users express the same page, resulting in low matching accuracy. Slight differences in user voice commands (such as different wording or changes in expression) may prevent successful page navigation; on the other hand, when voice commands cannot match the preset page, the system lacks an effective follow-up processing mechanism, making it difficult to meet the diverse needs of users in complex scenarios, thus affecting the overall experience of voice interaction. Summary of the Invention

[0005] This application provides a page navigation method based on NLP voice commands to solve the problem that when traditional voice commands cannot match the preset page, the system lacks an effective follow-up processing mechanism, which makes it difficult to meet the diverse needs of users in complex scenarios and thus affects the overall experience of voice interaction.

[0006] This application provides a page navigation method based on NLP voice commands, the method comprising:

[0007] Receive voice commands input by the user on the display device, and parse the voice information carried by the commands into text information through voice recognition technology;

[0008] The parsed text information is matched with a preset list of page keywords. If the text information contains any page keyword, the user is redirected to that page.

[0009] When the text information does not match the preset list of page keywords, the text information and page configuration information are sent to the LLM large language model, and the output results and confidence scores are based on the LLM large language model:

[0010] If the confidence score is greater than or equal to the first score, then redirect to the corresponding page;

[0011] If the confidence score is greater than the second score but less than the first score, a confirmation prompt will be displayed to the user, and the user will be redirected after confirmation.

[0012] If the confidence score is lower than the second score, the text information is sent to the LLM large language model, which generates a response and displays the dialog box on the display device.

[0013] In some possible implementations, the method further includes:

[0014] The semantic relevance between the parsed text information and the page name in the page configuration information, the matching degree between the text information and the page function description, and the overlap rate between the text information and the corresponding preset keywords of the page are quantitatively evaluated to obtain multi-dimensional evaluation results.

[0015] The LLM large language model calculates a confidence score of 0-100 by weighting the multi-dimensional evaluation results.

[0016] In some possible implementations, the method further includes:

[0017] Before generating the response content using the LLM large language model, the mobile application interface is called to obtain the context information of the current application, which includes the current page name, user login status, and incomplete operation prompts.

[0018] A response tailored to the current scenario is generated based on the context information, and the response content includes operation guidance information.

[0019] In some possible implementations, matching the parsed text information with a preset list of page keywords includes:

[0020] The text information is matched precisely with the text; if the match fails, the pinyin match is performed.

[0021] If the text message contains any keyword from a certain page, it is considered a successful match.

[0022] In some possible implementations, the page configuration information also includes historical page jump data, which contains high-frequency time periods and associated operation sequences during which users triggered page jumps in the past 30 days; wherein the LLM large language model combines the historical jump data to perform intent reasoning in order to adjust the feature weights corresponding to the confidence score.

[0023] In some possible implementations, when matching the parsed text information with a preset list of page keywords, the method includes:

[0024] The text information and keywords are compared character by character. If no match is found, both the text information and keywords are converted into Pinyin format.

[0025] The text information in pinyin format is matched with keywords based on their similarity. A successful match is determined when the overlap of the pinyin sequence is greater than or equal to 80%.

[0026] In some possible implementations, the method further includes:

[0027] The speech information is input into the dialect adaptation model, and the speech is transcribed by the dialect-Mandarin conversion module to obtain the transcribed Mandarin text.

[0028] The transcribed Mandarin text is then matched for keywords.

[0029] If dialect conversion fails, matching is performed based on the pinyin features of the dialect speech.

[0030] In some possible implementations, the display device periodically collects statements from the user's actual voice commands that do not match successfully but point to a specific page, filters out statements with a semantic similarity of ≥0.85 with the target page, and adds a list of keywords to that page.

[0031] In some possible implementations

[0032] The speech recognition technology is the Vosk speech recognition tool. Vosk parses the user's voice input commands into text information. If the parsing fails, it sends an error message to the user and stops processing.

[0033] In some possible implementations, each page in the preset list of page keywords corresponds to multiple keywords:

[0034] The keywords are standardized before storage, specifically including: converting all keywords to lowercase and removing punctuation marks and redundant modifiers;

[0035] Keywords containing dialect words or internet slang are synchronously associated with their corresponding standard Mandarin expressions, forming a hierarchical structure of basic keywords combined with extended variations. The basic keywords are the standard page names, and the extended variations are diversified expressions after standardization.

[0036] As described above, this application provides a page navigation method based on NLP voice commands. This method includes receiving voice commands input by a user on a display device and parsing the voice information carried by the commands into text information using speech recognition technology; matching the text information with a preset list of page keywords, and navigating to the page when the text information contains any page keyword; when the text information does not match the preset list of page keywords, sending the text information and page configuration information to an LLM (Large Language Model), and outputting the results and confidence score based on the LLM: if the confidence score is greater than or equal to a first score, navigation to the corresponding page; if the confidence score is greater than a second score but less than the first score, a confirmation prompt is displayed to the user; if the confidence score is less than the second score, the text information is sent to the LLM, which generates a response and displays a dialog box on the display device. By setting multiple keywords for each page and using pinyin matching, the matching range is expanded, the matching failure rate due to differences in user expression is reduced, and the accuracy and success rate of page navigation are improved. This application introduces a Large Language Model (LLM) to assist in page navigation judgment and general dialogue interaction. Leveraging the powerful semantic understanding and reasoning capabilities of LLM, the system can handle more complex and diverse voice commands, enhancing its intelligence and flexibility. The page navigation method provided by this application offers users a more convenient and intelligent voice interaction method. Whether navigating between pages or engaging in conversation, it can respond to user needs more efficiently, meeting diverse usage scenarios and thus improving the user's voice interaction experience on mobile devices. Attached Figure Description

[0037] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This application provides a flowchart of a page navigation method based on NLP voice commands. Detailed Implementation

[0039] The embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described below do not represent all embodiments consistent with this application. They are merely examples of systems and methods consistent with some aspects of this application as detailed in the claims.

[0040] In today's mobile internet era, smartphones have become deeply integrated into people's daily lives and work, becoming indispensable portable smart terminals. Users use smartphones to perform diverse operations such as information retrieval, social communication, mobile payment, and entertainment, placing higher demands on the convenience and efficiency of human-computer interaction. Among these, voice interaction, with its characteristic of requiring no manual operation, demonstrates significant advantages in scenarios such as driving, sports, and busy hands, and is gradually becoming an important way for users to interact with their phones, greatly enriching the usage scenarios and interactive experiences of mobile applications.

[0041] Existing mobile voice interaction systems typically implement basic functions through the following process: After receiving the user's voice input, the system converts it into text information using speech recognition technology. Then, based on preset keyword matching rules, it determines the user's intent and executes the corresponding operation (such as page navigation). For example, when a user issues the voice command "open settings," the system recognizes the keyword "settings," triggers a page navigation, and completes the user's desired operation.

[0042] However, existing technologies still have significant shortcomings in practical applications: On the one hand, the keyword matching rules relied upon for page navigation are relatively simple, usually only pre-setting a few fixed keywords, which is difficult to cover the diverse ways users express the same page, resulting in low matching accuracy. Slight differences in user voice commands (such as different wording or changes in expression) may prevent successful page navigation; on the other hand, when voice commands cannot match the preset page, the system lacks an effective follow-up processing mechanism, making it difficult to meet the diverse needs of users in complex scenarios, thus affecting the overall experience of voice interaction.

[0043] In some embodiments, such as Figure 1 As shown, this application provides a page navigation method based on NLP voice commands, the method comprising:

[0044] Receive voice commands input by the user on the display device, and parse the voice information carried by the commands into text information through voice recognition technology;

[0045] The parsed text information is matched with a preset list of page keywords. If the text information contains any page keyword, the user is redirected to that page.

[0046] When the text information does not match the preset list of page keywords, the text information and page configuration information are sent to the LLM large language model, and the output results and confidence scores are based on the LLM large language model:

[0047] If the confidence score is greater than or equal to the first score, then redirect to the corresponding page;

[0048] If the confidence score is greater than the second score but less than the first score, a confirmation prompt will appear for the user, and the user will be redirected after confirmation; the first score is 85 points and the second score is 60 points.

[0049] If the confidence score is lower than the second score, the text information is sent to the LLM large language model, which generates a response and displays the dialog box on the display device.

[0050] By pre-setting a list of page keywords that cover diverse expressions and combining it with precise matching logic, the number of matching failures caused by differences in user expressions is greatly reduced. At the same time, when keyword matching fails, an LLM large language model is introduced to assist in the judgment. Its semantic understanding and reasoning capabilities are used to process complex instructions, which solves the limitations of single keyword matching in existing technologies and makes page redirection more in line with the user's true intention.

[0051] The LLM (Large Language Model) combines textual information with page configuration information to output a confidence score, enabling tiered processing of user intent. High confidence scores allow for direct navigation to improve efficiency, medium confidence scores require user confirmation to avoid accidental operations, and low confidence scores trigger dialogue interaction to avoid the "no response" dilemma. This tiered mechanism retains the convenience of automated operations while reducing error rates through human-computer collaboration, making the system more intelligently fault-tolerant when handling ambiguous and complex commands.

[0052] Clear feedback when speech recognition fails allows users to perceive the problem immediately; confirmation prompts with medium confidence respect user control and reduce frustration caused by operational errors; and the dialogue interaction mechanism with low confidence transforms "cannot jump" into "proactive assistance." The contextualized responses generated by the LLM model can guide users to complete their goals, preventing them from getting stuck in a loop of repetitive instructions, making the entire interaction process more coherent and user-friendly.

[0053] In some embodiments, the method further includes:

[0054] The semantic relevance between the parsed text information and the page name in the page configuration information, the matching degree between the text information and the page function description, and the overlap rate between the text information and the corresponding preset keywords of the page are quantitatively evaluated to obtain multi-dimensional evaluation results.

[0055] The LLM large language model calculates a confidence score of 0-100 by weighting the multi-dimensional evaluation results.

[0056] If the confidence score is not lower than 85 points (first score), it is judged as a high confidence match, and the system directly jumps to the corresponding page according to the page identifier returned by LLM;

[0057] If the confidence score is between 60 and 84, it is considered a medium confidence match. The system will then display a confirmation prompt to the user, asking "Do you want to go to [page name]?" The user will confirm before being redirected to the next page.

[0058] If the confidence score is below 60 (second score), it is judged as a low confidence match, and no page jump is performed. The process proceeds to the LLM normal dialogue interaction process in step 3.

[0059] By quantifying the evaluation across three core dimensions—semantic relevance, functional description matching, and keyword overlap—the previously vague concept of "user intent matching page" is transformed into a measurable numerical indicator, avoiding the bias of judging from a single dimension. For example, when a user's instruction "Find where to save photos" matches the "Album" page, the score is not only based on the keyword overlap rate between "photos" and "Album," but also comprehensively evaluated by combining the functional matching degree between "save photos" and the "Album function," as well as the semantic relevance between "where to save photos" and "Album." This makes the score more closely reflect the user's true intent and reduces subjective judgment errors.

[0060] LLM integrates multi-dimensional results through weighted calculations and can dynamically adjust the weights of each dimension according to different scenarios, making the scoring logic more aligned with page attributes and user interaction habits. This targeted weighting avoids a "one-size-fits-all" evaluation model, ensuring that high-confidence scores accurately point to core functional needs.

[0061] Multi-dimensional evaluation covers the entire chain from surface-level keywords to mid-level functional matching and deep semantic association, enabling LLM to more comprehensively parse user commands and understand implicit intent. This deep understanding capability reduces misjudgments caused by indirect or ambiguous user expressions, further improving the accuracy of page navigation and reducing the probability of users repeating commands.

[0062] In some embodiments, the method further includes:

[0063] Before generating the response content using the LLM large language model, the mobile application interface is called to obtain the context information of the current application, which includes the current page name, user login status, and incomplete operation prompts.

[0064] A response tailored to the current scenario is generated based on the context information, and the response content includes operation guidance information.

[0065] By acquiring contextual information such as the current page name, user login status, and incomplete operation prompts, LLM-generated responses can closely match the user's current application scenario, avoiding generic feedback that is detached from reality. For example, when a user issues a "How to buy" command on the "Shopping Cart" page and no corresponding page is found, the response will combine the context of "the current page is the shopping cart" and "there are 3 unchecked items" to generate a guide such as "You can click the 'Proceed to Checkout' button at the bottom of the current page to complete your purchase," rather than the general "Please find the checkout page to operate," making the response more targeted.

[0066] The operation guidance information included in the response of this application is generated based on the current context and is directly related to the user's real-time operating environment, so the user does not need to recall or search for entry points. For example, when a user says "I want to change my password" on the "Home" page, the response can combine information such as "the current page is the home page" and "the user is logged in" to suggest "you can click 'My' in the upper right corner of the current page, find 'Security Settings' in the personal center to change your password". The clear path guidance reduces the operation threshold for users, and is especially user-friendly for users who are not familiar with the application structure.

[0067] In some embodiments, matching the parsed text information with a preset list of page keywords includes:

[0068] The text information is matched precisely with the text; if the match fails, the pinyin match is performed.

[0069] If the text message contains any keyword from a certain page, it is considered a successful match.

[0070] In some embodiments, the page configuration information also includes historical page jump data, which includes high-frequency time periods and associated operation sequences during which users triggered page jumps in the past 30 days; wherein the LLM large language model combines the historical jump data to perform intent reasoning in order to adjust the feature weights corresponding to the confidence score.

[0071] Historical page navigation data (such as users' high-frequency navigation time periods and associated operation sequences within 30 days) directly reflects users' behavioral habits and interaction preferences. By incorporating this data and adjusting feature weights, LLM (Local Level Management) can better align confidence scores with individual users' usage logic. For example, if a user habitually navigates from the "Homepage" to the "Reading" page at 9 PM every night, when the user issues a command to "open the reading page" from the "Homepage" during that time, the system will increase the weights of "time period matching" and "associated page path," making it easier for the "Reading" page's confidence score to reach a high threshold. This ensures accurate navigation and avoids misjudgments caused by generic scoring logic ignoring individual user habits.

[0072] The associated operation sequence can provide the "context chain" of user behavior for the LLM. For example, in the user's historical operations, the sequence of "search for products → add to cart → checkout" appears multiple times. When the user issues the instruction "how to pay" on the "add to cart" page, the LLM combines the associated sequence of "cart → checkout", which will increase the weight of the "matching degree between text information and the function description of the 'checkout' page", making the confidence score of the "checkout" page more accurate, avoiding misjudging the instruction as other payment-related pages, and thus enhancing the understanding of the user's coherent operation intention.

[0073] Adjust the weight based on the historical data of the "past 30 days" to ensure that the scoring logic can be updated in real time according to the changes in user behavior habits. For example, if the user commonly used the "personal center" in the early stage, but in the past two weeks, frequently entered the personal information-related operations through the "my orders" page, the system will gradually reduce the weight of the "personal center" keyword and increase the weight of the "my orders" associated path, so that the confidence score can adapt to the user's latest interaction preferences in a timely manner, avoiding the "outdated judgment" caused by relying on static weights.

[0074] In some embodiments, when matching the parsed text information with the preset page keyword list, the method includes:

[0075] Perform a character-level comparison between the text information and the keyword. If the match fails, convert both the text information and the keyword into pinyin form.

[0076] Perform an approximate match between the pinyin-form text information and the keyword. When the coincidence degree of the pinyin sequence is greater than or equal to 80%, it is determined that the match is successful.

[0077] During the speech recognition process, the user's pronunciation may not be standard (such as confusion between flat and retroflex sounds, or confusion between front and back nasal sounds) or affected by dialects, which may lead to character differences between the recognized text and the actual intention (such as recognizing "photo album" as "相册"). By converting the text and the keyword into pinyin form for approximate matching, the character writing differences can be ignored, and the focus can be on the relevance at the pronunciation level. For example, the character-level comparison between "photo album" and "相册" fails, but their pinyin is both "xiangce", and the coincidence degree of the pinyin sequence is 100%, so it can be determined that the match is successful, avoiding the jump failure caused by recognition errors.

[0078] In some embodiments, the method further includes:

[0079] Input the speech information into the dialect adaptation model, and perform speech transcription through the dialect-Mandarin conversion module to obtain the transcribed Mandarin text.

[0080] Perform keyword matching on the transcribed Mandarin text.

[0081] When the dialect conversion fails, matching is performed based on the pinyin features of the dialect voice.

[0082] By transcribing the dialect voice into Mandarin text, the already constructed Mandarin page keyword list can be directly reused, which not only reduces the system's maintenance cost for multi-dialect keywords but also ensures the matching accuracy by leveraging the mature Mandarin keyword matching logic. For example, after the Sichuan dialect "jinqie shezhi yiha" (enter the settings) is transcribed into Mandarin, it can directly match the preset keywords on the "settings" page, avoiding matching failures caused by the disconnection between dialect expressions and the keyword system.

[0083] Regarding the characteristics of "similar pronunciation and same meaning" and "flexible expression" in dialects, through the mechanism of transcription first and pinyin fallback, it can not only handle standard dialect expressions (such as standard Cantonese, Mandarinized dialects) but also adapt to non-standard dialects (such as Mandarin mixed with dialect words, dialect oral expressions). For example, when a Shandong dialect user says "zhao ge cun zhaopian de dier" (including the dialect word "dier" for place), after being transcribed into Mandarin, it can match the keyword "album"; if the transcription fails, the core sequence overlap degree of its pinyin "zhao ge cun zhaopian de dier" and the pinyin of "album" meets the standard.

[0084] It can still achieve a jump, enabling the system to maintain a stable interaction ability in diverse dialect scenarios.

[0085] Dialect users do not need to deliberately adjust their pronunciation or switch languages and can maintain the natural state of daily communication (such as accented instructions, dialect-specific oral expressions). The system responds to their needs through accurate transcription or pinyin matching, avoiding the situation where users are forced to change their language habits because "dialects are not recognized", making the voice interaction more in line with real-life scenarios and enhancing users' acceptance and usage stickiness of the system.

[0086] In some embodiments, the display device regularly collects the expressions in the user's actual voice commands that do not match successfully but point to a specific page, and through semantic similarity calculation, filters out the expressions with a correlation degree ≥ 0.85 with the target page and supplements them to the keyword list of that page.

[0087] In some embodiments,

[0088] The voice recognition technology is the Vosk voice recognition tool, which analyzes the user's voice input command into text information. If the analysis fails, an error prompt is fed back to the user and the processing stops.

[0089] In some embodiments, in the preset page keyword list, each page corresponds to multiple keywords:

[0090] The keywords are standardized before storage, specifically including: converting all keywords to lowercase and removing punctuation marks and redundant modifiers;

[0091] Keywords containing dialect words or internet slang are synchronously associated with their corresponding standard Mandarin expressions, forming a hierarchical structure of basic keywords combined with extended variations. The basic keywords are the standard page names, and the extended variations are diversified expressions after standardization.

[0092] By associating dialectal vocabulary, internet slang, and standard Mandarin expressions (e.g., "album" with "photo album," "shopping spree page" with "shopping cart," and "personal homepage" with "personal center"), the keyword list covers users' everyday colloquial expressions while also being compatible with emerging expressions (such as internet slang) and regional language habits (such as dialects). For example, users who use "product recommendation page" can be matched with extended variations of "product recommendation page," and users who use "my homepage" can be matched with the basic keywords of "personal center," ensuring that users of different ages and language habits can complete the matching through natural expressions.

[0093] In this hierarchical structure combining basic keywords and extended variations, the basic keyword (the standard page name) serves as the core anchor, ensuring the stability of the keyword system (e.g., "album" always remains the basic keyword). Extended variations dynamically incorporate diverse expressions, adapting to new user habits without frequent adjustments to the basic keyword (e.g., adding extended variations like "photo library" or "album" doesn't require modifying the "album" basic keyword). This structure avoids keyword list clutter and reduces system maintenance costs (no need to set separate basic keywords for each new expression).

[0094] After standardization and removal of redundant modifiers, keywords focus on core semantics (e.g., "quickly enter the system settings interface" is standardized to "system settings"), which reduces interference from non-core words in the matching. At the same time, the strong association between extended variants and basic keywords avoids mismatches caused by vague expressions (e.g., internet slang) and weak associations between multiple pages, thus improving the accuracy of matching.

[0095] As can be seen from the above embodiments, this application provides a page navigation method based on NLP voice commands. This method includes receiving voice commands input by a user on a display device and parsing the voice information carried by the commands into text information using speech recognition technology; matching the text information with a preset list of page keywords, and navigating to the page when the text information contains any page keyword; when the text information does not match the preset list of page keywords, sending the text information and page configuration information to an LLM (Large Language Model), and outputting the results and confidence score based on the LLM: if the confidence score is greater than or equal to a first score, navigation to the corresponding page; if the confidence score is greater than a second score but less than the first score, a confirmation prompt is displayed to the user; if the confidence score is less than the second score, the text information is sent to the LLM, which generates a response and displays a dialog box on the display device. By setting multiple keywords for each page and using pinyin matching, the matching range is expanded, the matching failure rate due to differences in user expression is reduced, and the accuracy and success rate of page navigation are improved. This application introduces a Large Language Model (LLM) to assist in page navigation judgment and general dialogue interaction. Leveraging the powerful semantic understanding and reasoning capabilities of LLM, the system can handle more complex and diverse voice commands, enhancing its intelligence and flexibility. The page navigation method provided by this application offers users a more convenient and intelligent voice interaction method. Whether navigating between pages or engaging in conversation, it can respond to user needs more efficiently, meeting diverse usage scenarios and thus improving the user's voice interaction experience on mobile devices.

[0096] Similar parts between the embodiments provided in this application can be referred to mutually. The specific implementation methods provided above are only a few examples under the overall concept of this application and do not constitute a limitation on the scope of protection of this application. For those skilled in the art, any other implementation methods extended from the solution of this application without creative effort shall fall within the scope of protection of this application.

Claims

1. A method for page navigation based on NLP voice commands, characterized in that, The method includes: Receive voice commands input by the user on the display device, and parse the voice information carried by the commands into text information through voice recognition technology; The parsed text information is matched with a preset list of page keywords. If the text information contains any page keyword, the user is redirected to that page. When the text information does not match the preset list of page keywords, the text information and page configuration information are sent to the LLM large language model, and the output results and confidence scores are based on the LLM large language model: If the confidence score is greater than or equal to the first score, then redirect to the corresponding page; If the confidence score is greater than the second score but less than the first score, a confirmation prompt will be displayed to the user, and the user will be redirected after confirmation. If the confidence score is lower than the second score, the text information is sent to the LLM large language model, which generates a response and displays the dialog box on the display device.

2. The page navigation method based on NLP voice commands according to claim 1, characterized in that, The semantic relevance between the parsed text information and the page name in the page configuration information, the matching degree between the text information and the page function description, and the overlap rate between the text information and the corresponding preset keywords of the page are quantitatively evaluated to obtain multi-dimensional evaluation results. The LLM large language model calculates a confidence score of 0-100 by weighting the multi-dimensional evaluation results.

3. The page navigation method based on NLP voice commands according to claim 1, characterized in that, When the confidence score is less than the second score, the method further includes: Before generating the response content using the LLM large language model, the mobile application interface is called to obtain the context information of the current application, which includes the current page name, user login status, and incomplete operation prompts. A response tailored to the current scenario is generated based on the context information, and the response content includes operation guidance information.

4. The page navigation method based on NLP voice commands according to claim 3, characterized in that, The step of matching the parsed text information with a preset list of page keywords includes: The text information is matched precisely with the text; if the match fails, the pinyin match is performed. If the text message contains any keyword from a certain page, it is considered a successful match.

5. The page navigation method based on NLP voice commands according to claim 3, characterized in that, The page configuration information also includes historical page navigation data, which includes high-frequency time periods and associated operation sequences during which users triggered page navigation in the past 30 days. The LLM large language model combines the historical jump data to perform intent reasoning, thereby adjusting the feature weights corresponding to the confidence score.

6. The page navigation method based on NLP voice commands according to claim 5, characterized in that, When matching the parsed text information with a preset list of page keywords, the method includes: The text information and keywords are compared character by character. If no match is found, both the text information and keywords are converted into Pinyin format. The text information in pinyin format is matched with keywords based on their similarity. A successful match is determined when the overlap of the pinyin sequence is greater than or equal to 80%.

7. The page navigation method based on NLP voice commands according to claim 6, characterized in that, The method further includes: The speech information is input into the dialect adaptation model, and the speech is transcribed by the dialect-Mandarin conversion module to obtain the transcribed Mandarin text. The transcribed Mandarin text is then matched for keywords. If dialect conversion fails, matching is performed based on the pinyin features of the dialect speech.

8. The page navigation method based on NLP voice commands according to claim 1, characterized in that, The display device periodically collects unmatched but relevant statements from users' actual voice commands that point to a specific page. It then uses semantic similarity calculations to filter out statements with a relevance of ≥0.85 to the target page, along with a list of keywords added to that page.

9. The page navigation method based on NLP voice commands according to claim 1, characterized in that, The speech recognition technology is the Vosk speech recognition tool. Vosk parses the user's voice input commands into text information. If the parsing fails, it sends an error message to the user and stops processing.

10. The page navigation method based on NLP voice commands according to claim 1, characterized in that, In the preset list of page keywords, each page corresponds to multiple keywords: The keywords are standardized before storage, specifically including: converting all keywords to lowercase and removing punctuation marks and redundant modifiers; Keywords containing dialect words or internet slang are synchronously associated with their corresponding standard Mandarin expressions, forming a hierarchical structure of basic keywords combined with extended variations. The basic keywords are the standard page names, and the extended variations are diversified expressions after standardization.

Images