Large-model dialogue multi-style interaction decision-making method and device under multi-service scene

Abstract

The invention discloses a large-model dialogue multi-style interaction decision-making method and device under a multi-service scene. The method comprises the following steps: extracting a service type and interaction demand characteristics in a dialogue request through semantic analysis; based on the service type obtained through analysis, a corresponding model working space is distributed for the service type, and the model working space is an independent logic area divided under the same large model; generating a file header identifier comprising a style identifier by using a Token consumption optimization model of a message header preposed decision mechanism according to the service type and the interaction demand characteristics, and pre-defining output content through the file header identifier; and the large model performs data processing according to a model working space corresponding to the file header identifier, selects a matched style from preset display styles in combination with the interaction demand characteristics and the style identifier in the file header identifier, and outputs interaction content in a streaming manner. According to the invention, the problems of high Token consumption and unsmooth dialogue of a low-computing-power terminal are solved; and the model resource utilization efficiency is low and the cost is high in a multi-service scene.

Description

Technical Field

[0001] This invention relates to the field of large model dialogue interaction technology, specifically to a method and apparatus for decision-making in multi-business scenarios using large model dialogue with multiple interaction styles. Background Technology

[0002] With the rapid development of artificial intelligence technology, large-scale model dialogue interaction technology has been widely penetrated into various business scenarios such as e-commerce, smart office, smart retail, and smart scenic spot guidance, becoming a core supporting technology for improving service efficiency and user experience. On e-commerce platforms, AI customer service needs to provide users with services such as product consultation and usage guidance; in smart office scenarios, AI office assistants undertake functions such as workflow planning and task allocation; in smart retail terminals and smart scenic spot guidance terminals, AI systems need to respond to users' queries and recommendation needs in real time. These application scenarios place extremely high demands on the interactive diversity of large-scale models, their adaptability to low computing power, and the efficiency of utilizing multiple business resources.

[0003] However, most existing large-scale model dialogue interactions are limited to traditional text-based output. This single display format cannot meet the diverse information presentation needs of users in different business scenarios. For example, in product recommendation scenarios, users need to intuitively understand the appearance, parameters, and other information of products; text descriptions alone cannot fully convey key content. In scenarios such as operation guidance and attraction introductions, dynamic videos and a combination of text and images can improve information delivery efficiency, but existing technologies cannot achieve such diversified displays, resulting in a rather rigid user interaction experience.

[0004] Furthermore, the efficient operation of large models typically relies on high computing power. However, in low-computing-power devices such as mobile phones, simple smart shopping guide terminals, and low-cost tourist guide terminals, large model dialogue output often suffers from stuttering and response delays. The core reason for this phenomenon is that existing technologies generally use the JSON protocol for communication. This protocol consumes a large number of tokens during data transmission, further increasing the processing burden on low-computing-power terminals. This prevents the terminal from responding to the output content of large models in a timely manner, severely impacting the user experience. When faced with multiple different business needs, traditional technical solutions often require loading independent large models for each business. This approach not only consumes a large amount of video memory, leading to a significant increase in hardware and operating costs, but also reduces the overall operating efficiency of the system due to frequent switching and loading of multiple models, and may even cause system instability. Summary of the Invention

[0005] To address these issues, this invention provides a method and apparatus for multi-style interactive decision-making in large-scale model dialogues under multiple business scenarios, solving the problems of monotonous display styles in existing large-scale model dialogues; high token consumption and unsmooth dialogue on low-computing-power terminals; and inefficient and costly utilization of model resources in multiple business scenarios.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for making multi-style interaction decisions in large-model dialogues under multiple business scenarios, comprising the following steps:

[0007] Step 1: Receive the dialogue request from the user terminal, and extract the business type and interaction requirement features from the dialogue request through semantic parsing;

[0008] Step 2: Based on the business type obtained from the parsing, allocate a corresponding model workspace for the business type. The model workspace is an independent logical region divided under the same large model.

[0009] Step 3: Based on the business type and the interaction requirements, use the Token consumption optimization model of the message header pre-decision mechanism to generate a file header identifier including style identifiers. The business affiliation, presentation format, theme and purpose of the output content are predefined through the file header identifier.

[0010] Step 4: The large model processes data based on the model workspace corresponding to the file header identifier, and selects a matching style from the preset display styles by combining the interaction requirement features with the style identifier in the file header identifier, and outputs the interactive content in a streaming manner.

[0011] As a preferred solution for a multi-mode interactive decision-making method for large-scale model dialogues in multi-business scenarios, in step 2, the session management mechanism of the model workspace satisfies: Let the number of currently ongoing sessions within a single workspace be... When N > 15, historical session cleanup is triggered, retaining the latest session for each session. The content of the next dialogue;

[0012] The priority ranking model for historical session cleanup is as follows:

[0013]

[0014] In the formula, Indicates the session retention priority; Time weighting for sessions; Weights for the frequency of interaction in the session; , These are the weighting coefficients, and This is used to adjust the proportion of influence of time factors and interaction frequency factors in priority ranking.

[0015] As a preferred solution for decision-making methods of large-scale dialogue with diverse interaction styles in multi-business scenarios, the token consumption optimization model of the message header pre-decision mechanism in step 3 is as follows:

[0016]

[0017] In the formula, This represents the total amount of tokens consumed in JSON protocol communication. This represents the total token consumption of the message header pre-decision mechanism. For protocol optimization coefficients, The basic consumption coefficient for identifiers. The number of tokens for the file header identifier.

[0018] As a preferred solution for a multi-model dialogue and multi-style interaction decision-making method in multi-business scenarios, in step 3, the number of tokens in the file header identifier satisfies... The generation logic is as follows:

[0019] The identifier format is "Business Space Identifier:Content Type Identifier:Core Theme Identifier:Purpose Identifier".

[0020] As a preferred solution for a multi-style interaction decision-making method for large-scale model dialogue in multiple business scenarios, in step 4, the display styles include at least text, video, products, and music, and the style selection strategy is: based on keyword matching degree of user interaction needs. ,when When the time comes, it will automatically switch to the corresponding style. This indicates the threshold for triggering style switching, which is dynamically adjusted based on the business scenario.

[0021] As a preferred solution for large-scale model dialogue and multi-style interaction decision-making methods in multi-business scenarios, in the context of intelligent office scenarios, the display styles also include workflows. When user interaction needs involve task allocation and process planning, keyword matching degree... The calculation model is as follows:

[0022]

[0023] In the formula, This refers to the number of workflow-related keywords in the user input. This refers to the total number of all valid keywords in the user input. For office scenarios, the adaptation coefficient is set.

[0024] In the context of intelligent scenic area navigation, the display style also includes recommendations. When user interaction requests involve recommendations for food or accommodation, the response time for style switching is considered. .

[0025] The present invention also provides a large-scale model dialogue multi-style interaction decision-making device for multi-service scenarios, comprising:

[0026] The request parsing module is used to receive dialogue requests from user terminals and extract the business type and interaction requirement features in the dialogue requests through semantic parsing.

[0027] The space management module is used to allocate a corresponding model workspace to the business type based on the parsed business type. The model workspace is an independent logical region divided under the same large model.

[0028] The identifier generation module is used to generate a file header identifier including style identifiers based on the business type and the interaction requirement characteristics, using the Token consumption optimization model of the message header pre-decision mechanism, and to predefine the business affiliation, presentation format, theme and purpose of the output content through the file header identifier;

[0029] The style output module is used to process data in the model workspace corresponding to the file header identifier of the large model, and select a matching style from the preset display styles by combining the interaction requirement features with the style identifier in the file header identifier, and output the interactive content in a streaming manner.

[0030] As a preferred solution for a large-scale model dialogue and multi-style interaction decision-making device in multi-service scenarios, the session management mechanism of the model workspace in the space management module satisfies the following: Let the number of currently ongoing sessions within a single workspace be... When N > 15, historical session cleanup is triggered, retaining the latest session for each session. The content of the next dialogue;

[0031] The priority ranking model for historical session cleanup is as follows:

[0032]

[0033] In the formula, Indicates the session retention priority; Time weighting for sessions; Weights for the frequency of interaction in the session; , These are the weighting coefficients, and This is used to adjust the proportion of influence of time factors and interaction frequency factors in priority ranking.

[0034] As a preferred solution for a large-scale dialogue and multi-style interaction decision-making device in multi-service scenarios, the token consumption optimization model of the message header pre-decision mechanism in the identifier generation module is as follows:

[0035]

[0036] In the formula, This represents the total amount of tokens consumed in JSON protocol communication. This represents the total token consumption of the message header pre-decision mechanism. For protocol optimization coefficients, The basic consumption coefficient for identifiers. The number of tokens for the file header identifier;

[0037] In the identifier generation module, the number of tokens for the file header identifier satisfies The generation logic is as follows:

[0038] The identifier format is "Business Space Identifier:Content Type Identifier:Core Theme Identifier:Purpose Identifier".

[0039] As a preferred solution for a large-scale model dialogue multi-style interaction decision-making device in multiple business scenarios, the display styles in the style output module include at least text, video, products, and music. The style selection strategy is based on keyword matching degree according to user interaction needs. ,when When the time comes, it will automatically switch to the corresponding style. This indicates the threshold for triggering style switching, which is dynamically adjusted based on the business scenario.

[0040] In smart office scenarios, the display style also includes workflows. When user interaction needs involve task allocation and process planning, keyword matching degree... The calculation model is as follows:

[0041]

[0042] In the formula, This refers to the number of workflow-related keywords in the user input. This refers to the total number of all valid keywords in the user input. For office scenarios, the adaptation coefficient is set.

[0043] In the context of intelligent scenic area navigation, the display style also includes recommendations. When user interaction requests involve recommendations for food or accommodation, the response time for style switching is considered. .

[0044] The present invention has the following advantages:

[0045] First, this invention introduces multiple style definitions through file header identifiers, supports multiple output formats such as text, video, product, and music, adapts to the information transmission needs of different business scenarios, makes the interaction more intuitive and vivid, and greatly improves the user experience.

[0046] Second, the present invention adopts a message header pre-decision mechanism to achieve streaming output with a file header identifier of no more than 5 tokens, which significantly reduces token consumption compared with the traditional JSON protocol, solves the problems of lag and response delay on low computing power terminals, and ensures smooth interaction.

[0047] Third, dedicated business workspaces are divided under the same large model, eliminating the need to load multiple independent models and reducing memory usage; combined with a refined session management mechanism, excess historical sessions are automatically cleaned up, reducing hardware and operating costs and ensuring stable and efficient system operation in multiple business scenarios.

[0048] Fourth, it allows for flexible switching of display styles and adjustment of resource allocation based on business type, adapting to the personalized needs of different industries, and demonstrating outstanding practicality and scalability. Attached Figure Description

[0049] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0050] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0051] Figure 1 This is a schematic diagram of the process of a large-model dialogue multi-style interaction decision-making method under multiple business scenarios provided in this embodiment of the invention;

[0052] Figure 2 This is a technical architecture diagram of the large-model dialogue multi-style interaction decision-making method in multi-service scenarios provided in this embodiment of the invention;

[0053] Figure 3 This is a schematic diagram of the architecture of a large-model dialogue multi-style interaction decision-making device for multiple business scenarios provided in an embodiment of the present invention. Detailed Implementation

[0054] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Example 1

[0056] See Figure 1 and Figure 2 Embodiment 1 of the present invention provides a method for making multi-style interaction decisions in large-model dialogue under multiple business scenarios, including the following steps:

[0057] S1: Receive dialogue requests from user terminals and extract business type and interaction requirement features from the dialogue requests through semantic parsing. Dialogue requests initiated by user terminals contain natural language data representing user intent. Semantic parsing employs intent recognition and keyword extraction techniques from natural language processing. Through a pre-trained semantic understanding model, it identifies core business types from the request text, such as e-commerce product inquiries, office workflow planning, and scenic area navigation queries. Simultaneously, it extracts interaction requirement features, such as information queries, operation guidance, and content recommendations, providing input for space allocation, identifier generation, and style selection to match user needs with business scenarios.

[0058] S2: Based on the parsed business type, allocate a corresponding model workspace for that business type. The model workspace is an independent logical region divided under the same large model. Traditional multi-business scenarios require loading multiple independent models, leading to excessive memory usage and inefficient switching. Step S2 adopts a single model and multiple logical spaces design. The same large model resides in the GPU memory, and independent workspaces are divided at the software level, with each space bound to a specific business type. The isolation of logical regions avoids mutual interference between different business data, and eliminates the need to repeatedly load models, reducing GPU memory consumption and model switching time, thus improving multi-business processing efficiency from an architectural perspective.

[0059] S3: Based on the business type and interaction requirements, a file header identifier, including style identifiers, is generated using a Token consumption optimization model based on a message header pre-decision mechanism. This file header identifier predefines the business affiliation, presentation format, theme, and purpose of the output content. The reason for lag on low-computing-power terminals is that traditional JSON protocol transmission consumes a large number of tokens. The message header pre-decision mechanism optimizes transmission efficiency through a logic of identifying the identifier first and then processing the content. Generating a file header identifier by combining business type and interaction requirements essentially pre-determines the transmission information, avoiding redundant data transmission. The Token consumption optimization model quantifies the optimization effect through coefficient adjustment, ensuring that the token consumption of the identifier itself is far lower than the redundant consumption of traditional protocols. The pre-defined function of the file header identifier allows the large model and the terminal to clearly define the interaction elements in advance, further reducing transmission and processing latency.

[0060] S4: The large model processes data based on the model workspace corresponding to the file header identifier. Combining the interaction requirement features with the style identifier in the file header identifier, it selects a matching style from preset display styles and streams the interactive content. The large model first locates the corresponding workspace through the file header identifier, calls the business-specific data and processing logic within that space, and ensures the targeting of data processing. The style selection stage is based on the mapping relationship between requirement-identifier-style. The style identifier locks in the range of candidate styles, and then the matching degree is verified in combination with the interaction requirement features to finally determine the optimal display form. Streaming output adopts block transmission and real-time rendering, transmitting the processed content to the terminal in batches. The terminal receives and renders one piece at a time, avoiding lag caused by overall loading, especially suitable for the processing capabilities of low-computing-power terminals.

[0061] In this embodiment, in step S2, the session management mechanism of the model workspace satisfies: Let the number of currently active sessions within a single workspace be... When N > 15, historical session cleanup is triggered, retaining the latest session for each session. The content of the next dialogue.

[0062] Specifically, an excessive number of sessions will continuously consume GPU memory resources, leading to a decrease in model running efficiency. Setting N=15 as the session number threshold can meet the concurrent session requirements in most scenarios while avoiding excessive GPU memory consumption. The latest k dialogue contents are retained because the recent content of the session is most relevant to the current interaction, while the earlier historical content has less reference value for subsequent responses. By triggering the number threshold and retaining recent content, the mechanism of dynamic release and efficient recycling of GPU memory resources can be achieved while ensuring the continuity of interaction.

[0063] The priority ranking model for historical session cleanup is as follows:

[0064]

[0065] In the formula, Indicates the session retention priority; Time weighting for sessions; Weights for the frequency of interaction in the session; , These are the weighting coefficients, and This is used to adjust the weighting of time and interaction frequency in priority ranking. When the number of sessions exceeds a threshold, priority ranking is used to determine the cleanup order and avoid mistakenly deleting high-value sessions. Time weighting The assignment logic of "higher for recent sessions and lower for longer sessions" is adopted to ensure that earlier sessions are cleaned up first; interaction frequency weighting. Weighting is applied proportionally to the number of interactions per unit time, giving higher retention priority to sessions with frequent interactions. , It can be dynamically adjusted according to business scenarios, and calculated through weighted summation. value, The smaller the value, the higher the priority of session cleanup, thus realizing the intelligence and scenario adaptability of the cleanup mechanism.

[0066] In this embodiment, the token consumption optimization model of the message header pre-decision mechanism in step S3 is as follows:

[0067]

[0068] In the formula, This represents the total amount of tokens consumed in JSON protocol communication. This represents the total token consumption of the message header pre-decision mechanism. For protocol optimization coefficients, The basic consumption coefficient for identifiers. The number of tokens for the file header identifier.

[0069] Specifically, the traditional JSON protocol requires the transmission of the complete key-value pair structure, resulting in a large number of redundant characters, leading to... The value is relatively high; As a protocol optimization coefficient (with a value range of [0.1, 0.3]), it reflects the degree of optimization achieved by this method in simplifying the transmission protocol and removing redundant data. The smaller the value, the more significant the optimization effect. This is the base token consumption coefficient for identifiers, used to correct the fixed token consumption of the file header identifier itself. The constraint T≤5 ensures that the identifier's own consumption is extremely low. This model can calculate token consumption in different scenarios, ensuring performance even on low-computing-power terminals. much smaller This ensures smooth dialogue from a quantitative perspective.

[0070] In this embodiment, in step S3, the number of tokens of the file header identifier satisfies The generation logic is as follows:

[0071] The identifier format is "Business Space Identifier:Content Type Identifier:Core Theme Identifier:Purpose Identifier".

[0072] Specifically, the number of tokens The resource consumption of the identifier itself is controlled to ensure that upfront decision-making does not add additional burden. The four-part format of the identifier adopts a minimalist coding logic: the business space identifier directly maps to the workspace allocated in step S2, clarifying the logical scope of data processing; the content type identifier is associated with the subsequent display style, locking the output format in advance; the core theme identifier focuses on the core dialogue object; and the purpose identifier clarifies the output purpose. The structured combination of the four parts of information allows the terminal and the large model to complete the synchronization of core information with only a small number of tokens, providing guidance for streaming output and style matching.

[0073] In this embodiment, in step S4, the display style includes at least text, video, product, and music, and the style selection strategy is: based on keyword matching degree of user interaction needs. ,when When the time comes, it will automatically switch to the corresponding style. This indicates the threshold for triggering style switching, which is dynamically adjusted based on the business scenario.

[0074] Specifically, multiple display styles are preset to adapt to the information delivery needs of different business scenarios, such as text adapting to text descriptions, videos adapting to dynamic demonstrations, and product styles adapting to graphic parameter displays. The keyword matching degree S is calculated by measuring the degree of overlap between the user's required keywords and the preset associated keywords for each style. The higher the S value, the stronger the adaptability between the requirement and the style. This trigger threshold is used to balance the accuracy and flexibility of style switching. When the value is high, switch the style only when the requirement is clearly matched to avoid accidental switching; When the value is lower, style switching is more flexible and adapts to ambiguous needs. Through matching degree calculation and threshold judgment strategies, automatic and precise style switching is achieved, improving the diversity and adaptability of interaction.

[0075] In one possible embodiment, in a smart office scenario, the display style further includes a workflow, where keyword matching is considered when user interaction needs involve task allocation and process planning. The calculation model is as follows:

[0076]

[0077] In the formula, This refers to the number of workflow-related keywords in the user input. This refers to the total number of all valid keywords in the user input. The adaptation coefficient for office scenarios.

[0078] Specifically, the workflow style in a smart office scenario needs to match specific requirements such as task allocation and process planning. Therefore, a scenario adaptation coefficient is introduced on top of the general matching degree calculation. . The keyword matching algorithm extracts workflow-related terms from the user input. The ratio of the two valid keywords in the statistical input reflects the basic matching degree. The introduction of this feature enhances the sensitivity of workflow-related needs recognition in office scenarios, enabling even needs with slightly lower basic matching degrees to trigger workflow style switching, adapting to the high-frequency need for process-oriented display in office scenarios, and improving office efficiency.

[0079] In one possible embodiment, in the intelligent scenic area guide scenario, the display style also includes recommendations. When user interaction requests involve recommendations for food or accommodation, the response time for style switching is [not specified]. .

[0080] Specifically, scenic area navigation terminals are mostly low-computing-power devices, and users in scenic areas have high requirements for the timeliness of information acquisition, such as instant searches for nearby restaurants and accommodations. Therefore, a response time threshold of τ≤0.5s is set. This threshold relies on pre-identification and a pre-loading caching mechanism. After the file header identifier pre-locks the recommended styles, the large model can process the recommended content data and style rendering instructions in parallel. Simultaneously, the terminal pre-loads rendering templates for commonly used recommended styles to avoid the time-consuming template loading during style switching. Through parallel processing and pre-loading optimization, the total time from identifying recommendation needs to completing style switching is ensured to be controlled within 0.5s, meeting the dual needs of low-computing-power terminals and high-timeliness scenarios, thus improving the user experience.

[0081] The application scenarios of this invention are as follows:

[0082] Scenario 1: E-commerce business scenario:

[0083] In AI customer service interactions on e-commerce platforms, this invention enables efficient end-to-end adaptation. When a user initiates a dialogue request such as product inquiry, usage guidance, or order inquiry (S1), the system quickly extracts the "e-commerce" business type and interaction requirements such as "product information inquiry" and "operation guidance" through semantic parsing; then, it allocates a dedicated "e-commerce business workspace" (S2), which contains built-in dedicated resources such as product database and order processing logic, without the need to load an additional independent model; next, it generates a file header identifier containing style identifiers (such as "e-commerce:product:mobile phone X:introduction"), with the number of tokens ≤ 5, and reduces transmission loss through a token consumption optimization model (S3); finally, it matches the display style according to the requirements. When inquiring about product parameters, it switches to the "product" style to intuitively present images, specifications, and prices; when asking about usage methods, it switches to the "video" style to play a demonstration video, ensuring smooth interaction on low-computing-power terminals such as mobile phones through streaming output (S4). At the same time, the workspace monitors the number of sessions in real time. When the number of sub-business sessions such as product inquiry and order processing N>15, it cleans up early low-value sessions according to priority sorting model, retaining only the latest 20 dialogues to ensure efficient concurrent operation of multiple businesses.

[0084] Scenario 2: Smart Office Scenario

[0085] Addressing the needs of enterprise team collaboration and individual office work, this invention provides customized interactive services. Employees initiate requests for task assignment, workflow planning, and document approval via office computers and smart office terminals (S1). Semantic parsing extracts the "smart office" business type and corresponding requirement characteristics. The system allocates an "office business workspace," dividing it into subspaces for task management, workflow planning, and scheduling (S2). A file header identifier (e.g., "office:workflow:project Y:task assignment") is generated, reducing token consumption through pre-decision processing (S3). During the style selection phase, when task assignment or workflow planning requirements are detected, the keyword matching degree S is calculated using the formula... The calculation (γ is set to 1.1 to enhance office scenario adaptation) automatically switches the "Workflow" style when S≥θ, clearly displaying task steps, time nodes, and responsible persons; when employees need to relax, the "Music" style is switched to play background music, and streaming output ensures smooth response under low computing load on office equipment (S4). The session management mechanism will promptly clean up excess historical sessions to ensure that multiple business processes such as schedule management and document approval are processed in parallel without lag.

[0086] Scenario 3: Smart Retail Terminal Scenario

[0087] In smart shopping guide terminals in offline retail scenarios such as shopping malls and convenience stores, this invention can enhance the customer interaction experience. Customers inquire about clothing styles, cosmetic efficacy, promotional activities, etc., through the terminal (S1). Semantic parsing extracts the "smart retail" business type and features such as "product search" and "activity information." A "retail business workspace" is allocated, covering subspaces such as product guidance, membership services, and promotional activities (S2). A file header identifier (e.g., "retail:product:clothing brand Z:style") is generated to adapt to the terminal's low computing power (S3). The display style is switched as needed; when searching for products, the "product" style is used to present images, colors, and sizes; when inquiring about materials and washing instructions, the "text" style is switched to output detailed text; when promoting new products, a "video" style is used to play promotional videos, with streaming output to avoid terminal lag (S4). When the number of sessions N in the membership service, product guidance, and other subspaces exceeds 15, early sessions are automatically cleaned up to ensure that the terminal can still efficiently respond to concurrent requests from multiple customers during peak hours.

[0088] Scenario 4: Intelligent Scenic Area Guide Scenario:

[0089] In low-computing-power navigation terminals at scenic area entrances and around attractions, this invention can meet the diverse query needs of tourists. Tourists request information such as attraction introductions, route planning, and recommendations for food and accommodation through the terminal (S1). Semantic parsing extracts the "scenic area navigation" service type and corresponding demand characteristics; a "scenic area navigation workspace" is allocated, dividing it into subspaces for attraction introductions, route planning, ticketing inquiries, and recommendation services (S2); a file header identifier is generated (e.g., "Scenic Area:Recommendation:Spot A Surrounding Area:Food"), controlling token consumption to adapt to the terminal's computing power (S3); the display style can be flexibly switched—using "text" style when introducing the history and culture of attractions; using "video" style when viewing live-action footage; switching to "recommendation" style when recommending food and accommodation, with a style switching response time τ≤0.5s to ensure tourists quickly obtain information; route planning, ticketing inquiries, and other requests are output as structured content according to the corresponding style (S4). The session management mechanism cleans up excess historical sessions, ensuring that the terminal can still operate stably and efficiently when multiple tourists query simultaneously.

[0090] Scenario 5: Education and Training Scenario

[0091] In online education platforms and smart learning terminals, this invention can adapt to the needs of teaching interaction. When a student or parent initiates a request for course consultation, knowledge point explanation, or homework tutoring (S1), semantic parsing extracts the "education and training" business type and features such as "knowledge query" and "tutoring needs"; an "education business workspace" is allocated, covering subspaces such as course recommendation, knowledge point explanation, and homework tutoring (S2); a file header identifier (e.g., "education:video:mathematics knowledge point M:explanation") is generated to reduce transmission token consumption (S3); the display style can be selected as needed, using the "video" style to play animated demonstrations when explaining abstract knowledge points; using the "text" style to output mind maps when organizing knowledge frameworks; and using the "product" style to present course outlines, prices, and class times when recommending courses, with streaming output ensuring smooth interaction on terminals such as mobile phones and tablets (S4). When multiple students simultaneously initiate homework tutoring or course consultation requests, resulting in a session count N>15, a priority sorting model is used to clean up low-value historical sessions, ensuring the stability of the platform's parallel processing of multiple teaching businesses.

[0092] It should be noted that the method of this embodiment can also be applied to distributed scenarios, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method in this embodiment, and the multiple devices will interact with each other to complete the large-model dialogue multi-style interaction decision-making method for multi-service scenarios.

[0093] It should be noted that the above description describes some embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0094] Example 2

[0095] See Figure 3 Embodiment 2 of the present invention also provides a large-model dialogue multi-style interaction decision-making device for multi-service scenarios, including:

[0096] The request parsing module 100 is used to receive dialogue requests from user terminals and extract the business type and interaction requirement features in the dialogue requests through semantic parsing.

[0097] The space management module 200 is used to allocate a corresponding model workspace for the business type based on the parsed business type. The model workspace is an independent logical region divided under the same large model.

[0098] The identifier generation module 300 is used to generate a file header identifier including style identifiers based on the business type and the interaction requirement characteristics, using the Token consumption optimization model of the message header pre-decision mechanism, and to predefine the business affiliation, presentation format, theme and purpose of the output content through the file header identifier;

[0099] The style output module 400 is used to process data in the model workspace corresponding to the file header identifier of the large model, and select a matching style from the preset display styles by combining the interaction requirement features and the style identifier in the file header identifier, and output the interactive content in a streaming manner.

[0100] In this embodiment, the session management mechanism of the model workspace in the space management module 200 satisfies the following: Let the number of currently active sessions within a single workspace be... When N > 15, historical session cleanup is triggered, retaining the latest session for each session. The content of the next dialogue;

[0101] The priority ranking model for historical session cleanup is as follows:

[0102]

[0103] In the formula, Indicates the session retention priority; Time weighting for sessions; Weights for the frequency of interaction in the session; , These are the weighting coefficients, and This is used to adjust the proportion of influence of time factors and interaction frequency factors in priority ranking.

[0104] In this embodiment, the token consumption optimization model of the message header pre-decision mechanism in the identifier generation module 300 is as follows:

[0105]

[0106] In the formula, This represents the total amount of tokens consumed in JSON protocol communication. This represents the total token consumption of the message header pre-decision mechanism. For protocol optimization coefficients, The basic consumption coefficient for identifiers. The number of tokens for the file header identifier;

[0107] In the identifier generation module 300, the number of tokens for the file header identifier satisfies The generation logic is as follows:

[0108] The identifier format is "Business Space Identifier:Content Type Identifier:Core Theme Identifier:Purpose Identifier".

[0109] In this embodiment, the display styles in the style output module 400 include at least text, video, product, and music, and the style selection strategy is based on keyword matching degree according to user interaction needs. ,when When the time comes, it will automatically switch to the corresponding style. This indicates the threshold for triggering style switching, which is dynamically adjusted based on the business scenario.

[0110] In smart office scenarios, the display style also includes workflows. When user interaction needs involve task allocation and process planning, keyword matching degree... The calculation model is as follows:

[0111]

[0112] In the formula, This refers to the number of workflow-related keywords in the user input. This refers to the total number of all valid keywords in the user input. For office scenarios, the adaptation coefficient is set.

[0113] In the context of intelligent scenic area navigation, the display style also includes recommendations. When user interaction requests involve recommendations for food or accommodation, the response time for style switching is considered. .

[0114] It should be noted that the information interaction and execution process between the modules of the above-mentioned device are based on the same concept as the method embodiment in Embodiment 1 of this application, and the resulting technical effects are the same as those in the method embodiment of this application. For details, please refer to the description in the method embodiment shown above in this application, and it will not be repeated here.

[0115] Example 3

[0116] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium storing program code for a multi-service scenario large model dialogue multi-style interaction decision method. The program code includes instructions for executing Embodiment 1 or any possible implementation thereof for the multi-service scenario large model dialogue multi-style interaction decision method.

[0117] Computer-readable storage media can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0118] Example 4

[0119] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0120] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor can call the program instructions to execute the large-model dialogue multi-style interaction decision method in multi-service scenarios under Embodiment 1 or any possible implementation thereof.

[0121] Specifically, a processor can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented in software, the processor can be a general-purpose processor that reads software code stored in memory. This memory can be integrated into the processor or located outside the processor and exist independently.

[0122] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0123] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing systems. They can be centralized on a single computing system or distributed across a network of multiple computing systems. Optionally, they can be implemented using program code executable by a computing system, thereby storing them in a storage system for execution by the computing system. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0124] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for making multi-style interactive decisions in large-scale model dialogues under multiple business scenarios, characterized in that, Includes the following steps: Step 1: Receive the dialogue request from the user terminal, and extract the business type and interaction requirement features from the dialogue request through semantic parsing; Step 2: Based on the business type obtained from the parsing, allocate a corresponding model workspace for the business type. The model workspace is an independent logical region divided under the same large model. Step 3: Based on the business type and the interaction requirements, use the Token consumption optimization model of the message header pre-decision mechanism to generate a file header identifier including style identifiers. The business affiliation, presentation format, theme and purpose of the output content are predefined through the file header identifier. Step 4: The large model processes data based on the model workspace corresponding to the file header identifier, and selects a matching style from the preset display styles by combining the interaction requirement features with the style identifier in the file header identifier, and outputs the interactive content in a streaming manner.

2. The method for large-scale model dialogue with diverse interaction styles in multi-service scenarios according to claim 1, characterized in that, In step 2, the session management mechanism of the model workspace satisfies the following: Let the number of currently active sessions within a single workspace be... When N > 15, historical session cleanup is triggered, retaining the latest session for each session. The content of the next dialogue; The priority ranking model for historical session cleanup is as follows: In the formula, Indicates the session retention priority; Time weighting for sessions; Weights for the frequency of interaction in the session; , These are the weighting coefficients, and This is used to adjust the proportion of influence of time factors and interaction frequency factors in priority ranking.

3. The method for large-scale model dialogue with diverse interaction styles in multi-service scenarios according to claim 1, characterized in that, In step 3, the token consumption optimization model for the message header pre-decision mechanism is as follows: In the formula, This represents the total amount of tokens consumed in JSON protocol communication. This represents the total token consumption of the message header pre-decision mechanism. For protocol optimization coefficients, The basic consumption coefficient for identifiers. The number of tokens for the file header identifier.

4. The method for large-scale model dialogue with diverse interaction styles in multi-service scenarios according to claim 3, characterized in that, In step 3, the number of tokens for the file header identifier satisfies The generation logic is as follows: The identifier format is "Business Space Identifier:Content Type Identifier:Core Theme Identifier:Purpose Identifier".

5. The method for large-scale model dialogue with diverse interaction styles in multi-service scenarios according to claim 1, characterized in that, In step 4, the display styles include at least text, video, products, and music. The style selection strategy is based on keyword matching degree according to user interaction needs. ,when When the time comes, it will automatically switch to the corresponding style. This indicates the threshold for triggering style switching, which is dynamically adjusted based on the business scenario.

6. The method for large-scale model dialogue with diverse interaction styles in multi-service scenarios according to claim 5, characterized in that, In smart office scenarios, the display style also includes workflows. When user interaction needs involve task allocation and process planning, keyword matching degree... The calculation model is as follows: In the formula, This refers to the number of workflow-related keywords in the user input. This refers to the total number of all valid keywords in the user input. For office scenarios, the adaptation coefficient is set. In the context of intelligent scenic area navigation, the display style also includes recommendations. When user interaction requests involve recommendations for food or accommodation, the response time for style switching is considered. .

7. A large-scale model dialogue multi-style interactive decision-making device for multiple business scenarios, characterized in that, include: The request parsing module is used to receive dialogue requests from user terminals and extract the business type and interaction requirement features in the dialogue requests through semantic parsing. The space management module is used to allocate a corresponding model workspace to the business type based on the parsed business type. The model workspace is an independent logical region divided under the same large model. The identifier generation module is used to generate a file header identifier including style identifiers based on the business type and the interaction requirement characteristics, using the Token consumption optimization model of the message header pre-decision mechanism, and to predefine the business affiliation, presentation format, theme and purpose of the output content through the file header identifier; The style output module is used to process data in the model workspace corresponding to the file header identifier of the large model, and select a matching style from the preset display styles by combining the interaction requirement features with the style identifier in the file header identifier, and output the interactive content in a streaming manner.

8. The large-scale model dialogue multi-style interactive decision-making device for multi-service scenarios according to claim 7, characterized in that, In the space management module, the session management mechanism of the model workspace satisfies the following: Let the number of currently active sessions within a single workspace be... When N > 15, historical session cleanup is triggered, retaining the latest session for each session. The content of the next dialogue; The priority ranking model for historical session cleanup is as follows: In the formula, Indicates the session retention priority; Time weighting for sessions; Weights for the frequency of interaction in the session; , These are the weighting coefficients, and This is used to adjust the proportion of influence of time factors and interaction frequency factors in priority ranking.

9. The large-scale model dialogue multi-style interactive decision-making device for multi-service scenarios according to claim 7, characterized in that, In the identifier generation module, the token consumption optimization model for the message header pre-decision mechanism is as follows: In the formula, This represents the total amount of tokens consumed in JSON protocol communication. This represents the total token consumption of the message header pre-decision mechanism. For protocol optimization coefficients, The basic consumption coefficient for identifiers. The number of tokens for the file header identifier; In the identifier generation module, the number of tokens for the file header identifier satisfies The generation logic is as follows: The identifier format is "Business Space Identifier:Content Type Identifier:Core Theme Identifier:Purpose Identifier".

10. The large-scale model dialogue multi-style interactive decision-making device for multi-service scenarios according to claim 7, characterized in that, In the style output module, the display styles include at least text, video, product, and music. The style selection strategy is based on keyword matching degree according to user interaction needs. ,when When the time comes, it will automatically switch to the corresponding style. This indicates the threshold for triggering style switching, which is dynamically adjusted based on the business scenario. In smart office scenarios, the display style also includes workflows. When user interaction needs involve task allocation and process planning, keyword matching degree... The calculation model is as follows: In the formula, This refers to the number of workflow-related keywords in the user input. This refers to the total number of all valid keywords in the user input. For office scenarios, the adaptation coefficient is set. In the context of intelligent scenic area navigation, the display style also includes recommendations. When user interaction requests involve recommendations for food or accommodation, the response time for style switching is considered. .

Images