Scene-based voice intelligent recommendation method, system and terminal

Abstract

The invention relates to the technical field of intelligent recommendation, and discloses a scene-based voice intelligent recommendation method, system and terminal, and the method comprises the steps: obtaining corresponding in-scene comprehensive data according to the scene classification of a scene used by a specified user; inputting the comprehensive data in the scene into an intelligent interaction module to obtain a corresponding output result; selecting an action strategy for the specified user according to the output result; the scene classification comprises an active question-answer communication scene and a passive voice recommendation scene; the active question-answer communication scene is used for replying and guiding a consulting question when an interactive conversation with a specified user occurs; according to the passive voice recommendation scene, corresponding interaction actions are automatically initiated according to specified requirements of specified users, the user satisfaction can be improved, the service efficiency is improved by optimizing the interaction process, and the method is particularly suitable for being applied to the fields of intelligent customer service, voice assistants and the like needing to process complex interaction requirements. And a better solution is provided for constructing more intelligent and more humanized voice interaction.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent recommendation, and in particular to a scenario-based voice intelligent recommendation method, system and terminal. Background Art

[0002] With the rapid development of artificial intelligence (AI) technology, intelligent voice recommendation systems have been widely adopted in a variety of fields, including e-commerce, healthcare, education, and finance, becoming a key technology for improving user experience. The construction and use of intelligent voice recommendation systems require the prior accumulation and updating of knowledge bases in the corresponding fields. However, traditional knowledge base systems still have significant shortcomings when dealing with complex and changing user needs. During the initial construction phase, relying on manually curated knowledge bases is not only costly but also fails to fully cover all possible variants of questions in various fields. Furthermore, when user needs or business scenarios change, static knowledge bases are difficult to update in a timely manner, resulting in a rapid decline in system responsiveness and a serious impact on service quality and user experience.

[0003] This problem is particularly prominent in voice interaction scenarios. Different application scenarios have different requirements for voice intelligent recommendation: in the marketing field, it is necessary to accurately understand the user's colloquial description and associate historical purchase data or consultation data to make personalized recommendations or answer questions; in medical scenarios, it is required to accurately identify professional terms and provide reliable diagnosis and treatment suggestions based on the patient's symptoms; and in educational applications, the system needs to adapt to the expression methods of learners of different age groups and provide targeted knowledge explanations. Moreover, the current voice intelligent recommendation system cannot actively remind users without actively initiating communication, such as in scenarios where timely reminders are needed; at this stage, voice intelligent recommendation systems often use general models to handle various scenarios, lacking deep adaptation to domain characteristics, resulting in uneven recommendation results. In voice interaction scenarios, the system also needs to deal with complex factors such as speech recognition errors, dialect differences, and colloquial expressions. Summary of the Invention

[0004] In order to provide a method, system and terminal that can adapt to the needs of different scenarios for intelligent voice recommendation, the present application provides a scenario-based voice intelligent recommendation method, system and terminal.

[0005] In the first aspect, the present application provides a scenario-based voice intelligent recommendation method, system, and terminal, which adopt the following technical solutions:

[0006] A scenario-based voice intelligent recommendation method, comprising:

[0007] Step 1: Obtain the corresponding comprehensive data in the scene according to the scene classification of the specified user usage scene;

[0008] Step 2: Input the comprehensive data in the scene into the intelligent interaction module to obtain the corresponding output result;

[0009] Step 3: Selecting an action strategy for the designated user based on the output result;

[0010] The scenario classification includes active question-and-answer communication scenarios and passive voice recommendation scenarios;

[0011] The active question-and-answer communication scenario means: an interactive dialogue occurs with the designated user, and the user needs to respond to and provide guidance on the consultation questions;

[0012] The passive voice recommendation scenario indicates that a corresponding interactive action is automatically initiated according to the specified needs of the specified user.

[0013] By adopting the above technical solutions, through scenario-based data processing and intelligent analysis, it is possible to adapt to both active question-and-answer communication scenarios and passive voice recommendation scenarios, and realize intelligent recommendations for different interaction modes. It can not only properly handle users' immediate consultation needs, but also actively provide valuable recommendation information, forming a complete service closed loop. This scenario-aware intelligent recommendation mechanism not only significantly improves user satisfaction, but also improves service efficiency by optimizing the interaction process, providing reliable technical support for the in-depth application of intelligent voice services in various industries, and is particularly suitable for applications in intelligent customer service, voice assistants and other fields that need to handle complex interaction needs, providing innovative solutions for building smarter and more humane voice interaction systems.

[0014] Optionally, for the active question-and-answer communication scenario, the comprehensive data in the scenario includes the consulting question data of the designated user, and the intelligent interaction module includes a cluster mining module, a knowledge base, and a reply module;

[0015] The cluster mining module is used to automatically mine knowledge points for questions that the designated user is concerned about but has not responded to, and recommend valuable questions to humans in the form of knowledge points. If the questions are valuable, they can be directly added to the knowledge base by humans;

[0016] The reply module is used to reply to the question points and knowledge points that the designated user is concerned about but has responses to, and the reply output content is the output result; the action strategy corresponding to the output result is the associated guided teaching.

[0017] Optionally, the process of automatically mining the question knowledge points that the designated user is concerned about but has not addressed includes:

[0018] Collect original questions from users and obtain a corpus of uncovered questions;

[0019] Use NLP technology to perform semantic clustering on questions, identify high-frequency uncovered questions, and discover knowledge blind spots;

[0020] Analyze the clustering results, extract common knowledge points, and form expandable knowledge points;

[0021] Submit high-value questions for manual review to ensure accuracy and avoid incorrect knowledge being stored.

[0022] Optionally, the method of collecting original questions asked by users and obtaining a corpus of uncovered questions includes:

[0023] Continuously collect original consultation questions input by users through voice or text;

[0024] Preprocessing the original consultation question; the preprocessing includes removing irrelevant characters, filtering stop words and restoring lemmas;

[0025] Encoding the original consultation question into a high-dimensional semantic vector using a pre-trained language model;

[0026] Calculate the maximum cosine similarity between the high-dimensional semantic vector and all existing knowledge points in the knowledge base to evaluate the coverage of the original consultation question;

[0027] When the maximum cosine similarity is lower than a preset similarity threshold, the original consultation question is automatically grouped into a corpus of uncovered questions and marked with a label to be analyzed.

[0028] Optionally, the method of using NLP technology to perform semantic clustering on questions, identify high-frequency uncovered questions, and discover knowledge blind spots includes:

[0029] The high-dimensional semantic vector combines the contextual features based on the Transformer architecture and the optimized keyword features to form a hybrid semantic space with strong representation capabilities;

[0030] In the clustering process, we identify clusters of questions with high semantic relevance and determine the core questions through density peak detection algorithm;

[0031] The multi-dimensional indicators of each generated semantic cluster are calculated. When the frequency of occurrence of a specific problem cluster in a unit time exceeds the preset threshold, it is marked as a high-frequency uncovered question point.

[0032] Optionally, the method of analyzing the clustering results, extracting common knowledge points, and forming expandable knowledge points includes:

[0033] Perform dependency parsing and semantic role labeling on the questions in each semantic cluster to identify the core entities, actions and their relationships in the questions;

[0034] A neural network model based on the attention mechanism is used to extract frequently occurring common semantic patterns and convert them into structured knowledge triples.

[0035] Optionally, for the passive question-and-answer communication scenario, the comprehensive data within the scenario includes the real-time monitoring data of the designated user, scenario status information, and voice information within the scenario;

[0036] The working process of the intelligent interaction module includes:

[0037] Build a scene based on the comprehensive data in the scene and output the scene building result;

[0038] Obtain the corresponding recommendation strategy based on the scenario building results and perform voice recommendation.

[0039] Optionally, the process of setting up the scene includes:

[0040] Performing user status analysis on the designated user based on real-time monitoring data, and outputting a user status analysis index;

[0041] Perform scene application status analysis based on scene status information and output scene device status coefficients;

[0042] Perform scene state analysis based on the voice information in the scene and output the scene state coefficient.

[0043] In the second aspect, this application provides a scenario-based voice intelligent recommendation system, which adopts the following technical solutions:

[0044] A scenario-based voice intelligent recommendation system, comprising:

[0045] The data collection module is used to obtain the corresponding comprehensive data in the scene according to the scene classification of the specified user usage scene;

[0046] A data transmission module, used to input the comprehensive data in the scene into the intelligent interaction module to obtain corresponding output results;

[0047] a strategy selection module, configured to select an action strategy for the designated user according to the output result;

[0048] The scenario classification includes active question-and-answer communication scenarios and passive voice recommendation scenarios;

[0049] The active question-and-answer communication scenario means: an interactive dialogue occurs with the designated user, and the user needs to respond to and provide guidance on the consultation questions;

[0050] The passive voice recommendation scenario indicates that a corresponding interactive action is automatically initiated according to the specified needs of the specified user.

[0051] In a third aspect, the present application provides a scenario-based voice intelligent recommendation terminal, which adopts the following technical solutions:

[0052] A scenario-based voice intelligent recommendation terminal is used to execute a scenario-based voice intelligent recommendation method.

[0053] In summary, this application includes at least one of the following beneficial technical effects:

[0054] Through scenario-based data processing and intelligent analysis, this application can adapt to both active question-and-answer communication scenarios and passive voice recommendation scenarios, and realize intelligent recommendations for different interaction modes. It can not only properly handle users' immediate consultation needs, but also actively provide valuable recommendation information, forming a complete service closed loop. This scenario-aware intelligent recommendation mechanism not only significantly improves user satisfaction, but also improves service efficiency by optimizing the interaction process, providing reliable technical support for the in-depth application of intelligent voice services in various industries, and is particularly suitable for use in intelligent customer service, voice assistants and other fields that need to handle complex interaction needs, providing innovative solutions for building smarter and more humane voice interaction systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a flow chart of the scenario-based voice intelligent recommendation method of the present invention;

[0056] Figure 2 It is a module diagram of the scenario-based voice intelligent recommendation system of the present invention. DETAILED DESCRIPTION

[0057] Embodiments of the present application are described in detail below, examples of which are illustrated in the accompanying drawings.

[0058] Throughout this specification, reference to the terms "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with the embodiment or example is included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0059] The present application discloses a scenario-based voice intelligent recommendation method, referring to Figure 1 ,include:

[0060] Step 1: Obtain the corresponding comprehensive data in the scene according to the scene classification of the specified user usage scene;

[0061] Step 2: Input the comprehensive data in the scene into the intelligent interaction module to obtain the corresponding output result;

[0062] Step 3: Selecting an action strategy for the designated user based on the output result;

[0063] The scenario classification includes active question-and-answer communication scenarios and passive voice recommendation scenarios;

[0064] The active question-and-answer communication scenario means: an interactive dialogue occurs with the designated user, and the user needs to respond to and provide guidance on the consultation questions;

[0065] The passive voice recommendation scenario indicates that a corresponding interactive action is automatically initiated according to the specified needs of the specified user.

[0066] In this application, through scenario-based data processing and intelligent analysis, it can adapt to both active question-and-answer communication scenarios and passive voice recommendation scenarios, and realize intelligent recommendations for different interaction modes. It can not only properly handle users' immediate consultation needs (such as customer inquiries such as "Can I get a loan for my insurance policy?"), but also actively provide valuable recommendation information (such as automatically sending voice prompts based on customer status), forming a complete service closed loop. This scenario-aware intelligent recommendation mechanism not only significantly improves user satisfaction, but also improves service efficiency by optimizing the interaction process, providing reliable technical support for the in-depth application of intelligent voice services in various industries, and is particularly suitable for applications in intelligent customer service, voice assistants and other fields that need to handle complex interaction needs, providing innovative solutions for building smarter and more humane voice interaction systems.

[0067] Optionally, for the active question-and-answer communication scenario, the comprehensive data in the scenario includes the consulting question data of the designated user, and the intelligent interaction module includes a cluster mining module, a knowledge base, and a reply module;

[0068] The cluster mining module is used to automatically mine knowledge points for questions that the designated user is concerned about but has not responded to, and recommend valuable questions to humans in the form of knowledge points. If the questions are valuable, they can be directly added to the knowledge base by humans;

[0069] The reply module is used to reply to the question points and knowledge points that the designated user is concerned about but has responses to, and the reply output content is the output result; the action strategy corresponding to the output result is the associated guided teaching.

[0070] In this embodiment of the present application, for problems that already have solutions, the reply module can quickly provide accurate answers; and for newly emerging typical problems, knowledge precipitation and updates can be formed through the clustering mining module. This knowledge discovery mechanism not only greatly reduces the cost of manual maintenance, but also ensures that the knowledge base can continue to absorb new problems that users really care about, and always maintain the ability to update and iterate knowledge reserves.

[0071] Optionally, the process of automatically mining the question knowledge points that the designated user is concerned about but has not addressed includes:

[0072] Collect original questions from users and obtain a corpus of uncovered questions;

[0073] Use NLP technology to perform semantic clustering on questions, identify high-frequency uncovered questions, and discover knowledge blind spots;

[0074] Analyze the clustering results, extract common knowledge points, and form expandable knowledge points;

[0075] Submit high-value questions for manual review to ensure accuracy and avoid incorrect knowledge being stored.

[0076] Optionally, the method of collecting original questions asked by users and obtaining a corpus of uncovered questions includes:

[0077] Continuously collect original consultation questions input by users through voice or text;

[0078] Preprocessing the original consultation question; the preprocessing includes removing irrelevant characters, filtering stop words and restoring lemmas;

[0079] Using a pre-trained language model (such as BERT or RoBERTa) to encode the original consultation question into a high-dimensional semantic vector can break through the limitations of surface vocabulary and grasp the essence of the question from a deeper level;

[0080] Calculate the maximum cosine similarity between the high-dimensional semantic vector and all existing knowledge points in the knowledge base to evaluate the coverage of the original consultation question;

[0081] When the maximum cosine similarity is lower than a preset similarity threshold, the original consultation question is automatically grouped into a corpus of uncovered questions and marked with a label to be analyzed.

[0082] A multi-level verification mechanism is employed during the process: first, a rapid, coarse match is performed to screen out obvious uncovered issues, followed by a detailed semantic analysis of edge cases. A pre-dynamic similarity threshold adjustment module automatically optimizes the judgment criteria based on the knowledge density of different fields. For example, in fields densely populated with technical terminology, the threshold can be appropriately raised to 0.3, while in fields with everyday language, it can be lowered to 0.2, thereby ensuring the accuracy of the screening results. Identified uncovered issues are structured and stored according to timestamp, issue type, and field label, providing a high-quality data foundation for subsequent semantic analysis and knowledge mining.

[0083] Optionally, the method of using NLP technology to perform semantic clustering on questions, identify high-frequency uncovered questions, and discover knowledge blind spots includes:

[0084] The high-dimensional semantic vector combines the contextual features based on the Transformer architecture and the optimized keyword features to form a hybrid semantic space with strong representation capabilities;

[0085] In the clustering process, we identify clusters of questions with high semantic relevance and determine the core questions through density peak detection algorithm;

[0086] The multi-dimensional indicators (occurrence frequency, time distribution and user group characteristics) of each generated semantic cluster are calculated. When the frequency of occurrence of a specific question cluster in a unit time exceeds the preset threshold, it is marked as a high-frequency uncovered question point.

[0087] In addition, a dynamic weight adjustment mechanism can be configured to automatically optimize the judgment criteria based on the importance of business scenarios, ensuring that knowledge blind spots in key areas can be discovered first. At the same time, by analyzing the semantic association network between question clusters, the system can identify potential complex knowledge blind spots, that is, knowledge gaps pointed out by multiple related but differently expressed questions.

[0088] In this embodiment of the present application, the hybrid semantic space is used to simultaneously capture the contextual information and key elements of the problem, breaking through the limitations of single feature representation, enhancing the interpretability of features while maintaining semantic depth, and providing a more comprehensive and reliable data foundation for subsequent clustering analysis; the use of density peak detection technology to identify core questions effectively solves the noise interference and boundary ambiguity problems faced by traditional clustering algorithms in problem discovery scenarios. By analyzing the density distribution characteristics of semantic clusters, the system can accurately identify the core issues that truly represent the focus of user attention, rather than superficially similar marginal issues; this density peak-based analysis method is particularly suitable for processing common expression differences and semantic overlap phenomena in natural language, greatly improving the quality and usability of clustering results.

[0089] Optionally, the method of analyzing the clustering results, extracting common knowledge points, and forming expandable knowledge points includes:

[0090] Perform dependency parsing and semantic role labeling on the questions in each semantic cluster to identify the core entities, actions and their relationships in the questions;

[0091] A neural network model based on an attention mechanism is used to extract frequently occurring common semantic patterns and convert them into structured knowledge triples. For example, in the field of new energy vehicles, the system can automatically extract standardized knowledge units such as "new energy vehicle, winter range, decay rate" from multiple similar questions such as "how much does the driving range decrease in winter" and "battery decay rate in low temperature environments." Through the organic combination of deep semantic analysis and pattern recognition technology, an intelligent conversion from original questions to expandable knowledge points is achieved. During the analysis phase, the system uses dependency parsing and semantic role labeling techniques to deeply analyze the problem structure, accurately identifying core entities, actions, and their interrelationships. The parsing method based on linguistic rules ensures the accuracy and completeness of knowledge element extraction. Compared with traditional simple keyword extraction, it can accurately grasp the essential semantic relationships of the problem, laying a solid foundation for subsequent knowledge modeling.

[0092] Optionally, for the passive question-and-answer communication scenario, the comprehensive data within the scenario includes the real-time monitoring data of the designated user, scenario status information, and voice information within the scenario;

[0093] The working process of the intelligent interaction module includes:

[0094] Build a scene based on the comprehensive data in the scene and output the scene building result;

[0095] Obtain the corresponding recommendation strategy based on the scenario building results and perform voice recommendation.

[0096] For example, in the field of intelligent voice reminders for intelligent driving, on the basis of obtaining scene status information, by obtaining real-time monitoring data and voice information in the scene, it is possible to more comprehensively judge the risk of user fatigue driving, thereby improving the accuracy and timeliness of fatigue monitoring. At the same time, this embodiment presets multiple scenes and sets corresponding recommendation strategies according to the type of scene. Therefore, scenes are constructed based on real-time monitoring data, scene status information and voice information in the scene, and recommendation strategies are obtained and voice recommendations are made based on the constructed scenes. Through this process, corresponding strategies can be provided in advance according to different user states, thereby being able to adapt to the needs of different scenes.

[0097] In one embodiment, taking the intelligent voice reminder of intelligent driving as an example, a process of scene construction is provided, including: Specifically, the process of scene construction includes:

[0098] Performing user status analysis on the designated user based on real-time monitoring data, and outputting a user status analysis index;

[0099] Perform scene application status analysis based on scene status information and output scene device status coefficients;

[0100] Perform scene state analysis based on the voice information in the scene and output the scene state coefficient;

[0101] Determine the interval based on the user state analysis index, scene device state coefficient and scene state coefficient value respectively, compare the combined interval with the corresponding interval combination of the preset scene, determine the corresponding preset scene and use it as the construction scene.

[0102] Through the above technical solution, by quantifying real-time monitoring data, scene status information and voice information in the scene, the user state analysis index, scene device state coefficient and scene state coefficient are obtained. By comparing the intervals of the user state analysis index, scene device state coefficient and scene state coefficient, the corresponding preset scene is obtained as the construction scene.

[0103] In this embodiment, real-time monitoring data, scene status information and voice information within the scene are divided into two intervals respectively, so there are 8 combinations in total. The 8 combinations correspond to preset scenes, and each preset scene corresponds to a recommendation strategy suitable for the scene. Therefore, it can adapt to the needs of the scene and improve user satisfaction.

[0104] In one embodiment, the real-time monitoring data includes the user's heart rate variability, the user's blinking frequency, the duration of a single eye closure, the nodding frequency, the duration of the head exceeding a preset deviation angle, the duration of the skin conduction level being less than a preset value, and the value of the periocular temperature exceeding the average facial temperature; the above parameters can all be realized through common wearable devices and smart sensor modules on vehicles, among which the user's heart rate variability and the duration of the skin conduction level being less than a preset value are obtained through smart bracelets, and the user's blinking frequency, the duration of a single eye closure and the nodding frequency, and the duration of the head exceeding the preset deviation angle can be realized through a camera and a recognition algorithm, among which the recognition algorithm is realized using computer vision technology in the existing technology, which will not be described in detail here, and the value of the periocular temperature exceeding the average facial temperature can be realized through an infrared thermal imager and existing computer vision technology. Therefore, the above real-time monitoring data can not only accurately reflect the user's fatigue state, but also has better feasibility. In addition, the process of obtaining the user state analysis index includes: comparing the real-time monitoring data with the corresponding standards respectively, and the corresponding standards are set according to empirical data.

[0105] For example, for heart rate variability, if its value SDNN < 50ms, fatigue is determined, so 50ms is the corresponding standard. If the blink frequency is less than 8 times / minute, fatigue is determined, so 8 times / minute is the corresponding standard. Therefore, when there is a certain real-time monitoring data that meets the corresponding standard, the user state analysis index U = A, and U is determined to belong to the first physiological interval U1, indicating that the user's physiological state is in a state of fatigue.

[0106] Otherwise, by formula The user state analysis index U is calculated. Through this process, when the real-time monitoring data does not exceed the corresponding standard, the user's fatigue state can be judged by calculating the degree of proximity between its value and the corresponding standard. This can then be used to judge the user's fatigue state in advance and improve the timeliness of the monitoring process. Where i is a positive integer and satisfies i∈[1,7], p i is the value of the i-th item of real-time monitoring data, pt i is the ideal value of the i-th real-time monitoring data. This parameter is set according to empirical data. i is the quantization unit value of the i-th real-time monitoring data, which is determined according to the numerical type of each real-time monitoring data. For example, the value of heart rate variability has a corresponding quantization unit value of 1ms, α i is the influence coefficient of the i-th real-time monitoring data. When the value of the i-th real-time monitoring data is positively correlated with the user's fatigue state, α i > 0, when the value of the i-th real-time monitoring data is negatively correlated with the user's fatigue state, α i <0; This parameter is set according to the degree of influence of different real-time monitoring data in the test data. Therefore, the size of the user state analysis index U is used to judge the fatigue risk of the user. The user state analysis index U is compared with the preset physiological threshold At. The preset physiological threshold At is set according to the fitting of empirical data. If U≥At, the user state analysis index U is judged to belong to the first physiological interval U1; if U<At, the user state analysis index U is judged to belong to the second physiological interval U2. Through the calculation and comparison process of the above user state analysis index, the user's state can be judged accurately and timely, and then the subsequent scenario construction can be used to adapt to the user's needs, thereby improving user satisfaction while ensuring driving safety.

[0107] In one embodiment, the scene status information includes the number of vehicles passing per unit time, the cumulative driving time and the proportion of night driving, the curve of the total number of turns changing with driving time, the curve of the total number of brakes changing with driving time, and the curve of the vehicle speed changing with driving time. The above parameters are all obtained from vehicle driving data. The process of obtaining the scene device state coefficient includes:

[0108] The scene device state coefficient F is calculated by formula (1)-(5);

[0109]

[0110]

[0111] Since the factors that cause user fatigue during vehicle driving include time factors, operation factors, environmental factors and vehicle dynamics factors, the scene device state coefficient F is calculated by combining the influence of the four factors and then using formula (1); where j is a positive integer and j∈[1,4], λ j For S jThe corresponding influence coefficient is obtained after data fitting based on the influence degree of different factors. S1 is the driving coefficient, S2 is the environmental coefficient, S3 is the operation coefficient, S4 is the power coefficient, T is the cumulative driving time, η is the proportion of night driving, x is the adjustment coefficient, this parameter is obtained by fitting the test data, x>1; T0 is the time threshold, this data is set according to empirical data, and 2 hours is selected in this embodiment; N is the number of vehicles per unit time, N0 is the threshold of the number of vehicles per unit time, this parameter is set according to empirical data, and r(t) is the total number of turns as the driving time increases. Time variation curve, z(t) is the curve of the total number of braking times changing with driving time, t1 is the current time point, t0 is the starting time point, rt is the steering concentration coefficient, zt is the braking concentration coefficient, γ is the steering adjustment coefficient, τ is the braking adjustment coefficient, the above steering adjustment coefficient and braking adjustment coefficient are set according to the test data fitting, r0 is the steering number unit time threshold, z0 is the braking number unit time threshold, the steering number unit time threshold and the braking number unit time threshold are set according to the empirical data fitting, y1 and y2 are preset proportional coefficients, and their roots are According to the difference in the degree of influence, it is set by fitting the test data; v(t) is the curve of vehicle speed changing with driving time, v0 is the reference value of vehicle speed, which is set according to empirical data, κ is the speed adjustment coefficient, which is set according to the test data fitting, and vt is the speed concentration coefficient. Through the calculation process of formula (2), it is possible to judge the degree to which the user's driving time exceeds the standard. Through the calculation process of formula (3), it is possible to judge the traffic density of the vehicle in the past period. Through the calculation process of formula (4), it is possible to judge the user's state by comprehensively considering the steering and braking states during vehicle operation. Through the calculation process of formula (5), it is possible to judge the user's state according to the speed change factor during the user's driving process. Finally, through formula (1), a variety of factors are combined to realize the judgment process of the user's fatigue state, thereby improving the accuracy and timeliness of the judgment. In addition, in the calculation process of the operation coefficient and the power coefficient, it is not only judged by the size of the steering and braking times and the size of the speed, but also the corresponding concentration coefficient is obtained according to the overall degree of its data change in a short time to adjust the result, so that the obtained data is more in line with the actual state.

[0112] In one embodiment, the process of obtaining the steering concentration coefficient, the braking concentration coefficient, and the speed concentration coefficient includes: calculating the steering concentration coefficient rt, the braking concentration coefficient zt, and the speed concentration coefficient vt using formulas (7)-(9);

[0113]

[0114] in, is the average value of v(t) in the period t0~t1; through the calculation process of formula (7)-(8), the concentration state when steering and braking occur can be judged respectively. When the concentration is greater, the influence on the user fatigue is greater, and the calculated rt and zt values ​​are larger, and then the size of the operation coefficient is adjusted according to the obtained rt and zt. Similarly, through the calculation process of formula (9), the uniformity of the speed change can be judged. When the speed change is more uneven, the influence on the user fatigue is greater, and the calculated vt value is larger, and then the adjustment process of the power coefficient is realized.

[0115] In one embodiment, the process of obtaining the scene state coefficient includes: analyzing the acquired voice information in the scene based on AI to determine the conversation state in the car: if the conversation state is high density, the corresponding interval of the scene state coefficient is set to E1; if the conversation state is low density, the corresponding interval of the scene state coefficient is set to E2; the conversation content and conversation density are collected through the AI ​​model to determine the scene state during the user's driving process, and then obtain the corresponding interval of the scene state coefficient.

[0116] In one embodiment, the content of the recommendation strategy includes but is not limited to adjusting the voice type, voice fatigue prompt, adjusting the air conditioning temperature, adjusting the music playback rhythm, and recommending the nearest service area; corresponding to different scenarios, this embodiment combines the above-mentioned recommendation strategy content, and then determines the recommendation strategy according to the different adaptability of the user's fatigue level, thereby improving the user experience while ensuring the user's driving safety.

[0117] In one embodiment, a scenario-based voice intelligent recommendation system is provided, referring to Figure 2 ,include:

[0118] The data collection module is used to obtain the corresponding comprehensive data in the scene according to the scene classification of the specified user usage scene;

[0119] A data transmission module, used to input the comprehensive data in the scene into the intelligent interaction module to obtain corresponding output results;

[0120] a strategy selection module, configured to select an action strategy for the designated user according to the output result;

[0121] The scenario classification includes active question-and-answer communication scenarios and passive voice recommendation scenarios;

[0122] The active question-and-answer communication scenario means: an interactive dialogue occurs with the designated user, and the user needs to respond to and provide guidance on the consultation questions;

[0123] The passive voice recommendation scenario indicates that a corresponding interactive action is automatically initiated according to the specified needs of the specified user.

[0124] In one embodiment, a scenario-based voice intelligent recommendation terminal is provided, which is used to execute a scenario-based voice intelligent recommendation method.

[0125] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A scenario-based voice intelligent recommendation method, characterized in that: include: Step 1: Obtain the corresponding comprehensive data in the scene according to the scene classification of the specified user usage scene; Step 2: Input the comprehensive data in the scene into the intelligent interaction module to obtain the corresponding output result; Step 3: Selecting an action strategy for the designated user based on the output result; The scenario classification includes active question-and-answer communication scenarios and passive voice recommendation scenarios; For the proactive question-and-answer communication scenario, an interactive dialogue is conducted with the designated user, and the consultation questions need to be responded to and guided; For the passive voice recommendation scenario, the corresponding interactive action is automatically initiated according to the specified needs of the specified user. The working process of the intelligent interaction module includes: Build a scene based on the comprehensive data in the scene and output the scene building result; Obtain the corresponding recommendation strategy based on the scenario building results and perform voice recommendation.

2. The scenario-based voice intelligent recommendation method according to claim 1, characterized in that: For the active question-and-answer communication scenario, the comprehensive data in the scenario includes the consulting question data of the designated user, and the intelligent interaction module includes a cluster mining module, a knowledge base, and a reply module; The cluster mining module is used to automatically mine knowledge points for questions that the designated user is concerned about but has not responded to, and recommend valuable questions to humans in the form of knowledge points. If the questions are valuable, they can be directly added to the knowledge base by humans; The reply module is used to reply to the question points and knowledge points that the designated user is concerned about but has answers to, and the reply output content is the output result; The action strategy corresponding to the output result is the associated guided teaching.

3. The scenario-based voice intelligent recommendation method according to claim 2, characterized in that: The process of automatically mining the question points and knowledge points that the designated user is concerned about but has not responded to includes: Collect original questions from users and obtain a corpus of uncovered questions; Use NLP technology to perform semantic clustering on questions, identify high-frequency uncovered questions, and discover knowledge blind spots; Analyze the clustering results, extract common knowledge points, and form expandable knowledge points; Submit high-value questions for manual review to ensure accuracy and avoid incorrect knowledge being stored.

4. The scenario-based voice intelligent recommendation method according to claim 3, characterized in that: The method of collecting original questions asked by users and obtaining a corpus of uncovered questions includes: Continuously collect original consultation questions input by users through voice or text; Preprocessing the original consultation question; the preprocessing includes removing irrelevant characters, filtering stop words and restoring lemmas; Encoding the original consultation question into a high-dimensional semantic vector using a pre-trained language model; Calculate the maximum cosine similarity between the high-dimensional semantic vector and all existing knowledge points in the knowledge base to evaluate the coverage of the original consultation question; When the maximum cosine similarity is lower than a preset similarity threshold, the original consultation question is automatically grouped into a corpus of uncovered questions and marked with a label to be analyzed.

5. The scenario-based voice intelligent recommendation method according to claim 4, characterized in that: The method of using NLP technology to perform semantic clustering on questions, identify high-frequency uncovered questions, and discover knowledge blind spots includes: The high-dimensional semantic vector combines the contextual features based on the Transformer architecture and the optimized keyword features to form a hybrid semantic space with strong representation capabilities; In the clustering process, we identify clusters of questions with high semantic relevance and determine the core questions through density peak detection algorithm; The multi-dimensional indicators of each generated semantic cluster are calculated. When the frequency of occurrence of a specific problem cluster in a unit time exceeds the preset threshold, it is marked as a high-frequency uncovered question point.

6. The scenario-based voice intelligent recommendation method according to claim 4, characterized in that: The method of analyzing the clustering results, extracting common knowledge points, and forming expandable knowledge points includes: Perform dependency parsing and semantic role labeling on the questions in each semantic cluster to identify the core entities, actions and their relationships in the questions; A neural network model based on the attention mechanism is used to extract frequently occurring common semantic patterns and convert them into structured knowledge triples.

7. The scenario-based voice intelligent recommendation method according to claim 1, characterized in that: For the passive question-and-answer communication scenario, the comprehensive data in the scenario includes the real-time monitoring data of the designated user, the scenario status information, and the voice information in the scenario; The process of setting up the scene includes: Performing user status analysis on the designated user based on real-time monitoring data, and outputting a user status analysis index; Perform scene application status analysis based on scene status information and output scene device status coefficients; Perform scene state analysis based on the voice information in the scene and output the scene state coefficient.

8. A scenario-based voice intelligent recommendation system, characterized in that: The system adopts a scenario-based voice intelligent recommendation method according to any one of claims 1 to 7, comprising: The data collection module is used to obtain the corresponding comprehensive data in the scene according to the scene classification of the specified user usage scene; A data transmission module, used to input the comprehensive data in the scene into the intelligent interaction module to obtain corresponding output results; a strategy selection module, configured to select an action strategy for the designated user according to the output result; The scenario classification includes active question-and-answer communication scenarios and passive voice recommendation scenarios; The active question-and-answer communication scenario means: an interactive dialogue occurs with the designated user, and the user needs to respond to and provide guidance on the consultation questions; The passive voice recommendation scenario indicates that a corresponding interactive action is automatically initiated according to the specified needs of the specified user.

9. A scenario-based voice intelligent recommendation terminal, characterized in that: The terminal is used to execute a scenario-based voice intelligent recommendation method as described in any one of claims 1-7.

Images