System for intelligent interaction of software as service platform

Abstract

The invention provides a system for intelligent interaction of a software as a service platform. The system comprises a large language model LLM module, a voice processing module, a model context protocol MCP communication layer and a context management module. According to the system, function operation of the SaaS platform is directly driven through natural language interaction, combination of a menu or button control of a traditional graphical user interface (GUI), an LLM module and historical data of a context management module is not needed, the intention recognition accuracy is larger than or equal to 92%, and repeated correction of a user is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of software as a service platforms, and more specifically relates to a system for intelligent interaction of software as a service platforms. Background Art

[0002] With the widespread adoption of SaaS (Software as a Service) platforms, users are increasingly demanding convenient and natural interactions. Traditional SaaS platforms primarily rely on graphical user interfaces (GUIs), enabling user operations through visual elements such as menus, buttons, and forms. However, existing technologies suffer from the following significant drawbacks:

[0003] Inefficient interaction: Users must manually navigate menus and fill out forms, and complex operations require multiple clicks (for example, creating a work order in a CRM system requires 5-7 steps), resulting in a high learning cost and time-consuming.

[0004] Fragmented user experience: The layout of interface elements (such as buttons and lists) is limited by screen space, has poor adaptability to mobile devices, and cannot be dynamically adjusted based on user preferences.

[0005] Inadequate accessibility: The visual interaction method is not user-friendly for users with disabilities (such as the visually impaired), and auxiliary tools such as voice assistants have low functional integration with the SaaS platform;

[0006] Lack of context awareness: Traditional systems have difficulty understanding users' ambiguous intentions (for example, "Export last week's customer follow-up records as a table"), requiring users to repeatedly correct their input, resulting in poor interaction fluency.

[0007] Existing improvements, such as patent CN119541492A, rely on a combination of a large language model (LLM) and speech recognition to analyze user intent. However, its limitations lie in its relatively simple user intent analysis process, primarily relying on the speech recognition system to convert spoken language into text and then processing it through the LLM. This can lead to ambiguous or inaccurate intent analysis, especially for complex or non-standard queries, where it may not accurately understand the user's multi-dimensional needs. Furthermore, the patent primarily focuses on the recognition and processing of voice input, which is limited in certain scenarios. The accuracy of its speech recognition and semantic analysis is constrained by model training and speech quality. Furthermore, the patent does not explicitly mention integration with SaaS (Software as a Service) platforms, instead focusing on traditional speech recognition and database query processes. This design lacks flexibility and struggles to meet the challenges of rapid expansion and customization required by modern enterprises. While the patent has made some improvements in speech recognition and data query, it still faces stability and accuracy issues across multiple steps, including speech-to-text, semantic understanding, SQL generation, and data validation. Generated SQL statements may require extensive verification and revision due to insufficient reasoning in the large language model, increasing system complexity and operational costs.

[0008] Therefore, there is an urgent need for a new interactive system that can directly understand user intentions, support multimodal interaction and deeply integrate SaaS functions. Summary of the Invention

[0009] Therefore, in order to solve the above technical problems, the present invention proposes a system for intelligent interaction of a software-as-a-service platform, comprising: a large language model (LLM) module, a speech processing module, a model context protocol (MCP) communication layer, and a context management module; the system directly drives the functional operations of the SaaS platform through natural language interaction, without relying on the menus or button controls of a traditional graphical user interface (GUI). The LLM module combines historical data from the context management module to achieve an intent recognition accuracy rate of ≥92%, reducing repeated corrections by users.

[0010] A system for intelligent interaction of a software-as-a-service platform, comprising the following steps:

[0011] Large Language Model (LLM) module, used to parse the user's voice processing module input and generate semantic intent containing SaaS platform operation instructions, wherein the voice processing module includes an automatic speech recognition (ASR) unit and a text-to-speech (TTS) unit. The ASR unit is used to convert natural speech input into text and input it into the LLM module, and the TTS unit is used to convert the text output of the LLM module into natural speech feedback;

[0012] A Model Context Protocol (MCP) communication layer includes an MCP client and an MCP server. The MCP client converts the semantic intent into a standardized protocol request, and the MCP server adapts the protocol request into an API call executable by the SaaS platform.

[0013] A context management module, which is used to record user session history and SaaS platform status data, and dynamically optimize the response logic of the LLM module;

[0014] The system directly drives the functional operations of the SaaS platform through natural language interaction without relying on menus or button controls of traditional graphical user interfaces (GUIs).

[0015] In some embodiments, the MCP communication layer uses the Protobuf protocol for data transmission and defines the following metadata fields:

[0016] Operation type field, used to identify the functional classification of the SaaS platform;

[0017] Parameter mapping field, used to bind entities in natural language to API parameters of the SaaS platform;

[0018] Context identifier field, used to associate the current session with the history.

[0019] Furthermore, the LLM module implements domain adaptation in the following ways:

[0020] Fine-tuning on vertical domain datasets based on the low-rank adaptation LoRA method;

[0021] Design intent recognition templates to map user input into structured instructions containing operation type, parameters, and priority.

[0022] Furthermore, the ASR unit of the speech processing module adopts a real-time transcription model with a transcription delay of less than 300 milliseconds; the TTS unit supports multilingual speech synthesis and integrates emotional intonation control parameters to adjust the emotional expression of the output speech. The system transmits operation feedback through TTS emotional intonation parameters (such as speed and volume) to improve accessibility.

[0023] Furthermore, the context management module caches the conversation state through a distributed database and sets a dynamic expiration time TTL. The TTL is adaptively adjusted according to the session activity, and the adjustment range is 300 seconds to 1800 seconds. The context management module caches the conversation and dynamically adjusts the TTL, so that the LLM can continuously track user intentions in multiple rounds of interaction and improve response relevance.

[0024] Furthermore, the system also includes a high-frequency query optimization module to improve response efficiency by:

[0025] Establish a vector cache library for high-frequency requests and use a retrieval engine based on similarity matching for querying;

[0026] For requests with a matching threshold of 0.85 or above, cached results are returned directly;

[0027] High-frequency requests (such as the "Help Menu") achieve millisecond-level responses (latency <50ms) through the vector cache library (FAISS engine), which is 10 times more efficient than traditional database queries.

[0028] Furthermore, the MCP server integrates an OAuth2.0 authentication module and interacts with the SaaS platform through the open data protocol OData, supporting cross-platform function calls and data synchronization. The MCP communication layer integrates OAuth2.0 authentication to ensure that API calls comply with enterprise security policies. Penetration testing shows that the risk of unauthorized access is reduced by 99%. It is connected to mainstream SaaS platforms such as Salesforce and Pipedrive through the OData protocol, and function call compatibility covers more than 90% of enterprise application scenarios.

[0029] Furthermore, the natural language interaction paradigm supports multimodal input, including the fusion processing of text, voice and gesture commands, and transmits feedback information to the user terminal through a unified interface. It supports the fusion input of voice (ASR+TTS), text and gesture commands, is suitable for mobile terminals, desktop terminals and AR / VR devices, and reduces user churn rate by 30%; voice interaction provides a barrier-free operation path for visually impaired users.

[0030] Furthermore, the system is deployed in a cloud-based distributed architecture, wherein:

[0031] The LLM module runs on a GPU-accelerated cloud computing instance;

[0032] The MCP communication layer uses a load balancing strategy to dynamically allocate computing resources to respond to high-concurrency requests.

[0033] Furthermore, the interaction process of the system includes the following steps:

[0034] S1, the user inputs natural language instructions through voice or text;

[0035] S2, the ASR unit converts the voice command into text and transmits it to the LLM module;

[0036] In S3, the LLM module parses the instructions to generate semantic intent, and converts it into a protocol request through the MCP client;

[0037] S4, the MCP server adapts the protocol request to an API call of the SaaS platform and performs the operation;

[0038] S5, the operation results returned by the SaaS platform are processed by the LLM module and fed back to the user through TTS units or text forms.

[0039] Beneficial effects of the present invention: The present invention proposes a system for intelligent interaction on a software-as-a-service platform, comprising: a large language model (LLM) module, a speech processing module, a model context protocol (MCP) communication layer, and a context management module; the system directly drives the functional operations of the SaaS platform through natural language interaction, without relying on menus or button controls of a traditional graphical user interface (GUI). The LLM module combines historical data from the context management module to achieve an intent recognition accuracy rate of ≥92%, reducing repeated corrections by users. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 The present invention is a flow chart of a system for intelligent interaction on a software-as-a-service platform.

[0041] The following specific embodiments will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION

[0042] The following examples are described to assist in understanding the present application, and the examples are not and should not be interpreted in any way as limiting the scope of protection of the present application.

[0043] In the following description, those skilled in the art will recognize that throughout this discussion, components may be described as separate functional units (which may include sub-units), but those skilled in the art will recognize that various components or portions thereof may be divided into separate components or may be integrated together (including within a single system or component).

[0044] At the same time, the connections between components or systems are not intended to be limited to direct connections. Instead, data between these components may be modified, reformatted, or otherwise changed by intermediate components. In addition, additional or fewer connections may be used. It should also be noted that the terms "coupled," "connected," or "input" should be understood to include direct connections, indirect connections through one or more intermediate devices, and wireless connections. Example 1:

[0045] like Figure 1 1 is a system for intelligent interaction of a software as a service platform and a flow chart of the system according to the present invention.

[0046] A system for intelligent interaction of a software-as-a-service platform, comprising the following steps:

[0047] Large language model LLM module, used to parse the input of the user's voice processing module and generate semantic intent containing SaaS platform operation instructions, wherein the voice processing module includes an automatic speech recognition ASR unit and a text-to-speech TTS unit, the ASR unit is used to convert natural voice input into text and input it into the LLM module, and the TTS unit is used to convert the text output of the LLM module into natural voice feedback; the LLM module realizes domain adaptation in the following ways: fine-tuning on a vertical domain dataset based on the low-rank adaptation LoRA method; designing an intent recognition template to map the user input into a structured instruction containing operation type, parameters and priority; the voice The ASR unit of the processing module adopts a real-time transcription model with a transcription delay of less than 300 milliseconds; the TTS unit supports multi-language speech synthesis and integrates emotional intonation control parameters to adjust the emotional expression of the output speech. The system transmits operation feedback through TTS emotional intonation parameters (such as speed and volume) to improve accessibility; the natural language interaction paradigm supports multimodal input, including the fusion processing of text, voice and gesture commands, and transmits feedback information to the user terminal through a unified interface. It supports the fusion input of voice (ASR+TTS), text and gesture commands, adapts to mobile terminals, desktop terminals and AR / VR devices, and reduces user churn rate by 30%; voice interaction provides a barrier-free operation path for visually impaired users.

[0048] The Model Context Protocol (MCP) communication layer includes an MCP client and an MCP server. The MCP client converts the semantic intent into a standardized protocol request, and the MCP server adapts the protocol request into an API call executable by the SaaS platform. The MCP communication layer uses the Protobuf protocol for data transmission and defines the following metadata fields: an operation type field for identifying the functional classification of the SaaS platform; a parameter mapping field for binding entities in natural language to API parameters of the SaaS platform; and a context identifier field for associating the current session with historical records. The MCP server integrates an OAuth2.0 authentication module and interacts with the SaaS platform through the open data protocol OData, supporting cross-platform function calls and data synchronization. The MCP communication layer integrates OAuth2.0 authentication to ensure that API calls comply with enterprise security policies. Penetration testing shows that the risk of unauthorized access is reduced by 99%. Through the OData protocol, it connects to mainstream SaaS platforms such as Salesforce and Pipedrive, and function call compatibility covers more than 90% of enterprise application scenarios.

[0049] The context management module is used to record user session history and SaaS platform status data, and dynamically optimize the response logic of the LLM module. The context management module caches conversation status through a distributed database and sets a dynamic expiration time (TTL). The TTL is adaptively adjusted based on conversation activity within a range of 300 to 1800 seconds. By caching conversations and dynamically adjusting the TTL, the context management module enables the LLM to continuously track user intent across multiple rounds of interaction, improving response relevance.

[0050] The system directly drives the functional operations of the SaaS platform through natural language interaction without relying on menus or button controls of traditional graphical user interfaces (GUIs).

[0051] The system also includes a high-frequency query optimization module that improves response efficiency by:

[0052] Establish a vector cache library for high-frequency requests and use a retrieval engine based on similarity matching for querying;

[0053] For requests with a matching threshold of 0.85 or above, cached results are returned directly;

[0054] High-frequency requests (such as the "Help Menu") achieve millisecond-level responses (latency <50ms) through the vector cache library (FAISS engine), which is 10 times more efficient than traditional database queries.

[0055] The system is deployed in a cloud-based distributed architecture, where:

[0056] The LLM module runs on a GPU-accelerated cloud computing instance;

[0057] The MCP communication layer uses a load balancing strategy to dynamically allocate computing resources to respond to high-concurrency requests.

[0058] The interaction process of the system includes the following steps:

[0059] S1, the user inputs natural language instructions through voice or text;

[0060] S2, the ASR unit converts the voice command into text and transmits it to the LLM module;

[0061] In S3, the LLM module parses the instructions to generate semantic intent, and converts it into a protocol request through the MCP client;

[0062] S4, the MCP server adapts the protocol request to an API call of the SaaS platform and performs the operation;

[0063] S5, the operation results returned by the SaaS platform are processed by the LLM module and fed back to the user through TTS units or text forms.

[0064] Beneficial effects of the present invention: The present invention proposes a system for intelligent interaction on a software-as-a-service platform, comprising: a large language model (LLM) module, a speech processing module, a model context protocol (MCP) communication layer, and a context management module; the system directly drives the functional operations of the SaaS platform through natural language interaction, without relying on menus or button controls of a traditional graphical user interface (GUI). The LLM module combines historical data from the context management module to achieve an intent recognition accuracy rate of ≥92%, reducing repeated corrections by users.

[0065] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A system for intelligent interaction of a software-as-a-service platform, comprising the following steps: Large Language Model (LLM) module, used to parse the user's voice processing module input and generate semantic intent containing SaaS platform operation instructions, wherein the voice processing module includes an automatic speech recognition (ASR) unit and a text-to-speech (TTS) unit. The ASR unit is used to convert natural speech input into text and input it into the LLM module, and the TTS unit is used to convert the text output of the LLM module into natural speech feedback; A Model Context Protocol (MCP) communication layer includes an MCP client and an MCP server. The MCP client converts the semantic intent into a standardized protocol request, and the MCP server adapts the protocol request into an API call executable by the SaaS platform. The context management module is used to record user session history and SaaS platform status data, and dynamically optimize the response logic of the LLM module.

2. The system for intelligent interaction of a software-as-a-service platform according to claim 1, characterized in that: The MCP communication layer uses the Protobuf protocol for data transmission and defines the following metadata fields: Operation type field, used to identify the functional classification of the SaaS platform; Parameter mapping field, used to bind entities in natural language to API parameters of the SaaS platform; Context identifier field, used to associate the current session with the history.

3. The system for intelligent interaction of a software-as-a-service platform according to claim 1, characterized in that: The LLM module achieves domain adaptation in the following ways: Fine-tuning on vertical domain datasets based on the low-rank adaptation LoRA method; Design intent recognition templates to map user input into structured instructions containing operation type, parameters, and priority.

4. The system for intelligent interaction of a software-as-a-service platform according to claim 1, characterized in that: The ASR unit of the speech processing module adopts a real-time transcription model with a transcription delay of less than 300 milliseconds; the TTS unit supports multi-language speech synthesis and integrates emotional intonation control parameters to adjust the emotional expression of the output speech.

5. The system for intelligent interaction of a software-as-a-service platform according to claim 1, characterized in that: The context management module caches the conversation state through a distributed database and sets a dynamic expiration time TTL. The TTL is adaptively adjusted according to the activity of the conversation and the adjustment range is 300 seconds to 1800 seconds.

6. The system for intelligent interaction of a software-as-a-service platform according to claim 1, characterized in that: The system also includes a high-frequency query optimization module that improves response efficiency by: Establish a vector cache library for high-frequency requests and use a retrieval engine based on similarity matching for querying; For requests with a matching threshold set above 0.85, the cached result is returned directly.

7. The system for intelligent interaction of a software-as-a-service platform according to claim 1, characterized in that: The MCP server integrates the OAuth2.0 authentication module and interacts with the SaaS platform through the open data protocol OData, supporting cross-platform function calls and data synchronization.

8. The system for intelligent interaction of a software-as-a-service platform according to claim 1, characterized in that: The natural language interaction paradigm supports multimodal input, including the fusion processing of text, voice and gesture commands, and transmits feedback information to the user terminal through a unified interface.

9. The system for intelligent interaction of a software-as-a-service platform according to claim 1, characterized in that: The system is deployed in a cloud-based distributed architecture, where: The LLM module runs on a GPU-accelerated cloud computing instance; The MCP communication layer uses a load balancing strategy to dynamically allocate computing resources to respond to high-concurrency requests.

10. The system for intelligent interaction of a software-as-a-service platform according to claim 1, characterized in that: The interaction process of the system includes the following steps: S1, the user inputs natural language instructions through voice or text; S2, the ASR unit converts the voice command into text and transmits it to the LLM module; In S3, the LLM module parses the instructions to generate semantic intent, and converts it into a protocol request through the MCP client; S4, the MCP server adapts the protocol request to an API call of the SaaS platform and performs the operation; S5, the operation results returned by the SaaS platform are processed by the LLM module and fed back to the user through TTS units or text forms.

Images