An interactive system and method for visualizing the thinking analysis process of an artificial intelligence large model search engine

Abstract

This invention provides an interactive system and method for visually presenting the thinking and analysis process of an artificial intelligence large-scale model search engine. The system includes: a user terminal configured to receive natural language query requests input by a user and display a visual interactive interface; a backend service system configured to invoke the artificial intelligence large-scale model search engine and receive structured response data returned by the search engine based on the natural language query request, the structured response data including at least a first structure field and a second structure field; and a visual interactive rendering engine configured to parse the structured response data and render it as an interactive interface on the user terminal. The system proposed in this invention, based on structured field-driven full-process visual rendering and an interactive thought node graph, transparently, structurally, and visually displays the internal workflow of the artificial intelligence large-scale model search engine.

Description

Technical Field

[0001] This invention relates to the fields of human-computer interaction and artificial intelligence technology. Specifically, it relates to an interactive system and its implementation method that provides a structured, visualized, and interactive graphical representation of the internal thinking, retrieval, and analysis processes of a large-scale artificial intelligence search engine during deep knowledge exploration and question answering. This invention is particularly suitable for enhancing the transparency, credibility, and user control of results from next-generation search engines based on large language models. Background Technology

[0002] With the rapid development of artificial intelligence technology, especially the breakthroughs in natural language processing (NLP) and machine learning, profound changes have been driven in the way humans interact with computers.

[0003] The visualization of the thought process of an AI large-scale model search engine refers to the node-based, viewable, and interactive display of the thought process, information retrieval and analysis, and information reading and integration processes during the execution of the AI ​​large-scale model search engine. Traditional search engines only search for relevant content based on keywords, requiring users to manually identify and summarize the content into effective information, which consumes a significant amount of time and effort in reading and processing information. The AI ​​large-scale model search engine, on the other hand, performs targeted problem understanding and information search based on the user's specific questions, selects content deemed valuable by the large model for further in-depth reading, and then summarizes and refines a large amount of information into a concise answer for the user, rather than simply returning relevant webpage links. This brings unprecedented convenience to users in obtaining information.

[0004] However, existing technical solutions still have many limitations and shortcomings in visualizing the thinking and analysis process of large-scale AI model search engines: 1. The hidden and untraceable nature of the analytical process: Existing technical solutions typically only show a vague process of "understanding the problem - finding / reading information - generating an answer - completing the answer." This process disappears after the answer is generated, making the entire process like an unopenable black box, which users cannot view or interact with.

[0005] 2. Lack of Explanatory Power of Decision-Making Basis: Existing technical solutions do not support users in viewing the detailed thinking and analysis process of large-scale model search engines. Users do not know the details of how the large-scale artificial intelligence model understands the problem, how it finds relevant information, how it judges the reliability of the found information, and how it organizes the effective content from the information sources. The accuracy of the generated answers cannot be traced or explained.

[0006] 3. Limited interactive experience and lack of user control: Users can only see the rough execution steps of the large-scale search engine, and cannot perform in-depth interactive operations in the middle process. They lack a sense of control over the whole process and there are no more in-depth or exploratory interactive operations.

[0007] Therefore, there is an urgent need for a solution that can break the "black box" state, make the internal workflow of the AI ​​large model search engine transparent, structured, and visualized, and support in-depth user interaction, so as to improve the transparency of the search process, the credibility of the results, and the interpretability. Summary of the Invention

[0008] The objective of this invention is to provide an interactive system and method for visually presenting the thinking and analysis process of an artificial intelligence large-scale model search engine. Through the system and / or method, the problems of the model thinking process being invisible, opaque, untraceable, and uncontrollable by users in the prior art are solved. Based on structured field-driven full-process visualization rendering and interactive thought node diagrams, the internal workflow of the artificial intelligence large-scale model search engine is made transparent, structured, and visualized.

[0009] In a first aspect of the invention, the aforementioned task is solved by an interactive system that visualizes the thinking and analysis process of a large-scale artificial intelligence search engine, the system comprising: The user terminal is configured to receive natural language query requests input by the user and display a visual interactive interface; The backend service system is configured to invoke the AI ​​large-scale model search engine and receive structured response data returned by the AI ​​large-scale model search engine based on the natural language query request. The structured response data includes at least a first structure field and a second structure field. A visual interactive rendering engine is configured to parse the structured response data and render it as an interactive interface on the user terminal; wherein, the visual interactive rendering engine includes: The thought draft visualization module is configured to parse the first structure field and dynamically display the current thought content in the form of flowing text during the thinking process of the artificial intelligence big model. The Thinking Node Graph Visualization Module is configured to parse the second structure field, generate and display an interactive thinking node graph; The node content display module is configured to respond to user interaction with nodes in the thought node diagram and display the detailed content of the selected node in the interactive interface. Final Response Display Module: Configured to generate the final response content on the interactive interface.

[0010] In one embodiment of the present invention, the interactive system for visualizing the thinking and analysis process of the artificial intelligence large-scale model search engine further includes: The data storage module is configured to store user requests, large model response content, thought node information, and interaction records.

[0011] In one embodiment of the present invention, the visual interactive rendering engine further includes a node detailed content module, which is configured to lay out the node detailed content in the order of node title, thought process, information source, and information integration result when rendering the node detailed content; wherein, the thought process and information source are displayed in a collapsed state by default after the content is output, and one or more expand / collapse controls are provided to respond to user operations to switch the visibility of their content.

[0012] In one embodiment of the present invention, in the thinking node diagram generated by the thinking node diagram visualization module, the user's original question node and final reply node are set to be unclickable, while the intermediate nodes are set to be clickable. When the user clicks the intermediate node, the node detailed content module is triggered to load and display the detailed content of the node.

[0013] In one embodiment of the present invention, the node content display module is further configured to divide the interactive interface into a dialogue flow area on the left and a node content area on the right; wherein, the node content area is preset with control logic so that it is hidden by default before the dialogue begins, automatically expands when the visualization interactive rendering engine receives the first structure field, and automatically collapses after the large model completes its response.

[0014] In one embodiment of the present invention, the backend service system includes: Request parsing module: Configured to parse user-input natural language query requests and generate structured request packets; and Model Invocation Module: Configured to invoke the AI ​​Big Model Search Engine to process the structured request packet and receive the structured response data output by the AI ​​Big Model Search Engine.

[0015] In one embodiment of the present invention, the user terminal includes an interactive control module; The interactive control module is configured to respond to a user's click on a reference mark in the final reply content, call the corresponding information source webpage link and open it in a new window; and is configured to respond to a user's click on the expand / collapse control, control the display state of the node content display module.

[0016] In one embodiment of the present invention, the visualization interactive rendering engine uses a fault-tolerant processing strategy to parse the structured response data using null value judgment and default value rollback mechanism; when the structured response data is missing or its structure does not meet the rendering conditions, the thinking node diagram visualization module does not draw the thinking node diagram.

[0017] In a second aspect of the invention, the aforementioned task is further solved by an interactive method that visualizes the thinking and analysis process of an artificial intelligence large-scale model search engine. The interactive system for visualizing the thinking and analysis process of the artificial intelligence large-scale model search engine is used in this method, which includes the following steps: Receive natural language query requests from users; The natural language query request is sent to the backend service system to drive the artificial intelligence big model search engine to process it, and the structured response data returned by the backend service system is received. The structured response data includes at least a first structure field and a second structure field. The thought draft visualization module in the visualization interactive rendering engine parses the first structure field and dynamically displays the current thought content in the form of flowing text during the thinking process of the artificial intelligence big model. The visualization module of the visualization interactive rendering engine parses the second structure field to generate and display an interactive thinking node diagram. In response to the user's interactive operation on the nodes in the thought node diagram, the node content display module in the visualization interactive rendering engine displays the detailed content of the selected node in the interactive interface. The final response display module in the visualization interactive rendering engine generates the final response content on the interactive interface.

[0018] In one embodiment of the present invention, during the process of receiving the structured response data, the visualization interactive rendering engine continuously writes the data of each node into the local storage of the user terminal; when the user clicks on a node in the mind node diagram to switch node content, the node content display module preferentially reads the data of the corresponding node from the local storage for display.

[0019] In one embodiment of the present invention, the interactive method for visualizing the thinking and analysis process of an artificial intelligence large-scale model search engine further includes: The natural language query request, structured response data, and user interaction operation records are stored. In response to a user's request to switch and view historical dialogue rounds in a conversation, the visualization interaction rendering engine restores the corresponding dialogue content and node status based on local storage. Alternatively, when the local cache expires, historical data is requested from the backend service system based on the session identifier, and the thought process of the historical dialogue is reconstructed and displayed retrospectively based on the returned historical data.

[0020] In a third aspect, the present invention also provides an electronic device comprising: A processor, configured to execute machine-readable instructions; A graphics card configured to train an interactive method that visualizes the thought process and analysis of a large-scale artificial intelligence search engine; and A memory configured to store machine-readable instructions that, when executed by a processor and / or graphics card, perform the steps of the interactive method that visualizes the thinking and analysis process of a large-scale artificial intelligence search engine.

[0021] In a fourth aspect, the present invention also provides a computer-readable storage medium having stored thereon computer-readable instructions, which, when executed by a processor, perform the steps of the interactive method for visualizing the thinking and analysis process of an artificial intelligence large-scale model search engine.

[0022] The technical solution provided by this invention has the following beneficial effects: 1. Existing technologies only present a linear process of "understanding the problem - finding information - generating an answer," and the process is untraceable after the result is generated. In contrast, this invention visualizes the multi-level, multi-step thinking path within the large-scale AI model search engine through a structured thinking node diagram. Each node supports real-time interaction, allowing users to click to view search keywords, highlighted information from source web pages, and information integration logic. This fully reproduces the decision-making chain of the large model from problem analysis to answer generation, overcoming the shortcomings of existing technologies, such as process concealment and untraceability.

[0023] 2. The interactive system and method proposed in this invention, which visualizes the thinking and analysis process of a large-scale artificial intelligence model search engine, constructs a dual-dimensional explanation system at the node and path levels. On the one hand, users can stream the thought draft of each node through the thought field, allowing them to understand the specific reasoning logic of the model at that stage. On the other hand, by setting explanation controls on the nodes of the thought node diagram, users can view with one click why the model derives that node, revealing the logical connections across nodes, greatly enhancing the interpretability of the model's decisions, and enabling users to verify, question, and trust the search results.

[0024] 3. The interactive system and method proposed in this invention, which visualizes the thinking and analysis process of an AI large-scale model search engine, significantly enhances the user's active control and interactivity over this process. Through a dynamic, collapsible content architecture and rich manual controls (such as automatic node expansion / collapse strategies and manual control module display levels), users are given complete control over the granularity of the thinking path display and the depth of exploration. The design of multi-level interactive verification channels (such as clicking nodes to view search keywords, direct links to highlighted webpage information, and citation tagging for tracing) enables users to verify the authenticity and rationality of the large-scale model's decision-making basis in real time.

[0025] 4. The front-end rendering engine of this invention features a robust fault tolerance mechanism and a field-driven mode. Even if some fields returned by the backend are missing or delayed, the entire page will not crash, ensuring system stability. Furthermore, through the unified formatting and encapsulation of response data on the backend, this invention can adapt to large model engines from different vendors, exhibiting excellent technical scalability. Attached Figure Description

[0026] To further illustrate the above and other advantages and features of the various embodiments of the present invention, a more specific description of the various embodiments of the present invention will be presented with reference to the accompanying drawings. It is to be understood that these drawings depict only typical embodiments of the invention and are therefore not intended to limit its scope. In the drawings, identical or corresponding parts will be indicated by identical or similar reference numerals for clarity.

[0027] Figure 1 This diagram illustrates the overall architecture of an interactive system that visualizes the thinking and analysis process of a large-scale artificial intelligence search engine, according to an embodiment of the present invention. Figure 2 This diagram illustrates the correspondence between the functional modules of the visual interactive rendering engine and the front-end parsed fields according to an embodiment of the present invention. Figure 3 A schematic diagram illustrating the composition of multiple system functional modules and the cooperative relationship between the modules in one embodiment of the present invention is shown. Figure 4a This diagram illustrates the interface state of the draft thinking trigger step in the rendering triggering process according to an embodiment of the present invention. Figure 4b This diagram illustrates the interface state of the thinking node graph triggering step in the rendering triggering process according to an embodiment of the present invention. Figure 4c A schematic diagram of the interface state during the node detail triggering step in the rendering triggering of an embodiment of the present invention is shown. Figure 4d A schematic diagram of the interface state in the final answer triggering step during rendering triggering according to an embodiment of the present invention is shown; Figure 5 This diagram illustrates an interactive method for visualizing the thinking and analysis process of a large-scale artificial intelligence search engine, according to an embodiment of the present invention. Detailed Implementation

[0028] In the following description, the invention is described with reference to various embodiments. However, those skilled in the art will recognize that the embodiments may be practiced without one or more specific details or with other alternatives and / or additional methods or components. In other instances, well-known structures or operations are not shown or described in detail so as not to obscure the inventive points of the invention. Similarly, for illustrative purposes, specific numbers and configurations are set forth to provide a comprehensive understanding of the embodiments of the invention. However, the invention is not limited to these specific details.

[0029] In this specification, references to "an embodiment" or "this embodiment" mean that a particular feature, structure, or characteristic described in connection with that embodiment is included in at least one embodiment of the invention. The phrase "in one embodiment" appearing throughout this specification does not necessarily refer to the same embodiment in all instances.

[0030] It should be noted that the embodiments of the present invention describe the method steps in a specific order; however, this is only for illustrating the specific embodiment and not for limiting the order of the steps. On the contrary, in different embodiments of the present invention, the order of the steps can be adjusted according to actual needs.

[0031] In this invention, the "front-end" is not merely a "chat interface." The "front-end" displays results and the reasoning process, primarily including functions such as real-time display of thought drafts, generation of thought node diagrams, interactive node clicking, expansion / collapse of information sources, display of the final answer, and retrieval of historical thought processes. In this invention, the "front-end" includes a visual interactive rendering engine. In this invention, the "back-end" is not just a regular API service, but an "AI search and reasoning process orchestration center." The "back-end" primarily includes functions such as parsing user questions, calling large model search engines, driving model thinking, organizing search and reasoning results, and outputting structured fields.

[0032] In this invention, the modules of the system according to the invention can be implemented using software, hardware, firmware, or a combination thereof. When a module is implemented using software, its function can be implemented through computer program flow. For example, the module can be implemented using code segments (such as code segments in languages ​​like C and C++) stored in a storage device (such as a hard disk, memory, etc.), wherein the corresponding function of the module can be implemented when the code segment is executed by a processor. When a module is implemented using hardware, its function can be implemented by setting a corresponding hardware structure. For example, the module's function can be implemented by hardware programming a programmable device such as a field-programmable gate array (FPGA), or by designing an application-specific integrated circuit (ASIC) that includes multiple transistors, resistors, capacitors, and other electronic devices. When a module is implemented using firmware, the module's function can be written into a read-only memory such as an EPROM or EEPROM in the form of program code, and the corresponding function of the module can be implemented when the program code is executed by a processor. In addition, some functions of the module may need to be implemented by separate hardware or by working in cooperation with the hardware. For example, the detection function is implemented by a corresponding sensor (such as a proximity sensor, accelerometer, gyroscope, etc.), the signal transmission function is implemented by a corresponding communication device (such as a Bluetooth device, infrared communication device, baseband communication device, Wi-Fi communication device, etc.), the output function is implemented by a corresponding output device (such as a display, speaker, etc.), and so on.

[0033] To address the problems of existing technologies where the model's thinking process is invisible, opaque, untraceable, and uncontrollable by users, this invention aims to visualize the thinking content search and analysis process of an AI large-scale model search engine. This visualization enables a more natural and efficient interaction between users and the AI ​​large-scale model, helping users obtain more accurate and effective information. The interactive system provided by this invention, which visualizes the thinking and analysis process of an AI large-scale model search engine, presents the execution process of the large-scale model search engine in a structured, visual, and interactive interface. Users can flexibly view the specific search and thinking processes, reducing the difficulty for users to discern the large-scale model's responses and improving the transparency, credibility, and interpretability of search results.

[0034] Figure 1 This diagram illustrates the overall architecture of an interactive system that visualizes the thinking and analysis process of a large-scale artificial intelligence model search engine, according to an embodiment of the present invention. Figure 1 As shown, the interactive system that visualizes the thinking and analysis process of the AI ​​large model search engine includes: a user terminal 101, a backend service system 102, a visualization interactive rendering engine 103, and a data storage module 104.

[0035] An interactive system that visualizes the thinking and analysis process of a large-scale AI model search engine is based on specific structured fields. Through graphical, process-oriented, and hierarchical structures, it fully presents the search, judgment, analysis, and response generation process of the large-scale AI model on the user interface and supports various forms of interactive operations.

[0036] User terminal 101 is used to receive natural language query requests (such as search questions) input by users and display a visual interactive interface (where users can view the model's thought process, click on interactive nodes, and receive the final response). In one embodiment of the present invention, user terminal 101 can be a computer device or other terminal.

[0037] The backend service system 102 is used to call the AI ​​large model search engine to perform tasks such as problem understanding, information retrieval, judgment analysis, and answer generation, and to receive structured response data returned by the AI ​​large model search engine (hereinafter referred to as the large model) based on the query request. In one embodiment of the present invention, the structured response data includes structured fields, which at least include a first structured field formatted.thought and a second structured field formatted.adjacency_list.

[0038] The visualization and interactive rendering engine 103 is used to parse structured response data and present the model's internal reasoning process as a user-interactive graphical interface.

[0039] The data storage module 104 is used to store user requests, large model response content, information of each thought node, and interaction records, supporting traceability and reuse. The fields stored in the data storage module 104 include: user input, query_text, and intermediate model results: formatted.thought, formatted.adjacency_list, and fullInfo.response.content. In one embodiment of the invention, the data storage module 104 can use a database such as a relational database or a document database for persistent storage. In one embodiment of the invention, the interactive system supports tracing historical nodes of multi-turn dialogues. After each round of question-and-answer initiated by the user, the front end caches the question content, corresponding answer results, and associated thought node data of that round to local storage, so as to quickly display and trace historical node information during the current session. When the user switches to view historical rounds within the same session, the front end can directly restore the corresponding dialogue content and node status based on the local cache. Furthermore, when the user closes the page or the local cache expires, the system can still send a request to the backend history interface based on the session identifier chatid to obtain the historical questions, answer results, and node data saved under that session, and complete the reconstruction and retrospective display on the front end. Therefore, this solution has both local fast backtracking and remote history recovery based on chatid capabilities, enabling continuous tracking and history reproduction in multi-turn dialogue scenarios.

[0040] Users input natural language query requests through their terminals into the system interface. The backend service system receives and parses the request, constructs a structured request packet, and sends it to the large model. The large model performs operations such as understanding the user request, structured thinking, information retrieval, judgment analysis, and answer generation. Simultaneously, it returns the intermediate states of the user's reasoning and processing to the backend service system in the form of structured response data. The backend service system generates interactive node data based on specific fields in the returned structured response data and initiates frontend visualization rendering.

[0041] The visualization and interactive rendering engine of the interactive system provided by this invention consists of multiple system functional modules. These modules collaborate with each other, completing interface rendering and user interaction based on the structured fields returned by the large model. The relationship between each module and its corresponding field is as follows: Figure 2 As shown. Figure 2 As shown, the visualization and interactive rendering engine includes: a thought draft visualization module 201, a thought node diagram visualization module 202, a node content display module 203, a node detailed content module 204, and a final response display module 205.

[0042] The Thought Draft Visualization Module 201 is used to parse the first structural field, formatted.thought. The rendering method is as follows: during the large model's thinking process, the current thought content of the large model is dynamically displayed in the dialogue flow area as flowing text. The display includes a loading indicator (i.e., a thinking process in progress), and automatically hides after the thinking is complete, simulating the dynamic thinking process of the large model.

[0043] The Thinking Node Graph Visualization Module 202 is used to parse the second structural field `formatted.adjacency_list`. The rendering method is as follows: it abstracts the complex processes performed by the large AI model when processing user questions—including task decomposition, planning, information retrieval, and logical reasoning—into a tree or graph structure, and then renders the tree structure returned by the large model as an interactive thinking node graph. In one embodiment of the invention, in the thinking node graph generated by the Thinking Node Graph Visualization Module, the user's original question node and final response node are set to non-clickable, while intermediate nodes are set to clickable. When the user clicks an intermediate node, the node details module loads and displays the detailed content of that node; clicking the small circle in the upper right corner of the node allows viewing the explanation of why the model generated that node. It should be noted that `formatted.adjacency_list` uses an adjacency list to describe the structure of the thinking node graph, which is essentially a mapping relationship from "parent node name to child node array". It typically contains a root node `root`, whose value is several child node objects; each child node object contains at least a node identifier `id`, a node name `name`, and a node state `state`. The front-end recursively constructs a tree-like mind map based on the adjacency list; its accuracy is mainly guaranteed by the back-end service system: the back-end service system maintains the node and edge relationships uniformly during the execution of large model planning, search, and reasoning, and updates the status of each node in real time during streaming push. The front-end does not infer the thinking path itself, but uses the graph structure returned by the server as the sole basis for visualization, thus accurately reflecting the current thinking process of the model. In one embodiment of the invention, the visualization interactive rendering engine uses a fault-tolerant processing strategy. For optional fields returned by the interface, null value judgment, optional chain access, and default value fallback mechanisms are generally used to avoid the overall page abnormality caused by missing fields. For key steps such as JSON parsing and streaming message processing, an exception capture mechanism is used for protection; once a format error or parsing failure occurs, the system will stop further processing of the current abnormal segment and ensure that the page can continue to be used through prompt information or empty state display. For structural fields such as adjacency_list, if they are missing or the structure does not meet the rendering conditions, the front-end does not draw a mind map, but only retains the existing dialogue or blank placeholder areas.

[0044] The node content display module 203 responds to user interactions with nodes in the thought graph, displaying the detailed content of the selected node in the interactive interface. The display structure of the node content display module 203 is as follows: the interactive interface is divided into a dialogue flow area on the left and a node content area on the right. The control logic of the node content display module 203 is as follows: it is hidden by default before the dialogue begins, i.e., node content is not displayed; once the first structure field `formatted.thought` is generated, the node content is automatically expanded, and automatically collapsed after the main model completes its response; users can manually click on nodes in the thought graph to re-expand and view them.

[0045] The fields corresponding to the node detailed content module 204 include current_node1 to current_nodeN, where node1 represents node 1, nodeN represents node N, and N represents the number of nodes in the mind node diagram. Node detailed content module 204 is used to display the content of all N nodes. The content of each node in node detailed content module 204 is displayed in the following four parts in order and logic: node title, thought process, information source, and information integration result. The node title field is fullInfo.content. The thought process content field is fullInfo.response.formatted.thought. The search keyword field for the thought process is query. The information source includes the source description field fullInfo.response.formatted.thought, the webpage link field fullInfo.response.content, and the webpage highlight information field fullInfo.response.formatted.action.parameters.select_ids. The information integration result field is also fullInfo.response.formatted.thought. It should be noted that the thought process and information sources are displayed in a collapsed state by default after the content is output, and one or more expand / collapse controls are provided to respond to user actions to switch the visibility of their content.

[0046] The final response display module 205 generates the final response content in the interactive interface. The field used by the final response display module 205 is `formatted.thought`, and its content is rendered in the dialogue flow area. If there is any cited information, the system will mark the citation icon in the content; clicking the citation icon will redirect the user to view the source page.

[0047] Figure 3This diagram illustrates the composition of multiple system functional modules and the collaborative relationships among them, according to an embodiment of the present invention. In terms of functional modules, in addition to the visualization interactive rendering engine 103 and the data storage module 104, the interactive system that visualizes the thinking and analysis process of the artificial intelligence large model search engine also includes: a request parsing module 301, a model invocation module 302, and an interactive control module 303.

[0048] In one embodiment of the present invention, the backend service system 102 includes a request parsing module 301 and a model invocation module 302. The request parsing module 301 is used to parse the natural language query request input by the user, perform word segmentation, intent recognition and semantic parsing, and generate a structured request package. The input of the request parsing module 301 is the user's natural language keywords; the output fields are: query_text (user input text); parsed_intent (parsed intent information, used for large model invocation).

[0049] The model invocation module 302 is used to invoke the large model to process structured request packages, perform problem understanding, information retrieval, content filtering, and answer generation, and output structured intermediate results at each stage. The inputs of the model invocation module 302 are query_text and parsed_intent; the output fields include: formatted.thought (formatted thought draft, used to render the thought process), formatted.adjacency_list (adjacency list, describing the tree-like reasoning structure), fullInfo.response.content (content of the source webpage), and fullInfo.response.formatted.action.parameters.select_ids (highlighted information markers of the source webpage).

[0050] In one embodiment of the present invention, the user terminal 101 includes an interaction control module 303. The interaction control module 303 is used to implement user operations such as expanding, collapsing, and jumping to nodes. Specifically, clicking a node invokes `fullInfo.content` and `fullInfo.response.formatted.thought`; in response to the user clicking a reference mark in the final response content, `select_ids` is invoked and the user is redirected to the source page; and in response to the user clicking the expand / collapse control, the display state of the node content display module is controlled.

[0051] In one embodiment of the present invention, when a user clicks on a node in the thought node graph, the front end first records the name of the currently active node and reads the corresponding content from the locally stored node details data accordingly. Then, it dynamically refreshes the node content area on the right. The content displayed in the right area includes sub-question titles, thought instructions, search keywords, information sources, and integrated conclusions. The system has a lightweight caching mechanism: during the streaming reception of backend node data, the front end continuously writes node details to the browser's local storage for quick retrieval when switching nodes. Therefore, the loading of node details mainly relies on the front end's local cache, rather than re-initiating a backend request after each click.

[0052] In one embodiment of the present invention, the front end uses a tree-based recursive rendering method based on React components to implement the thought node graph, and gradually displays the node state using a streaming update mechanism. Under normal scale, this method can meet real-time display requirements. Existing implementations control rendering overhead by reducing unnecessary secondary requests, reconstructing the visual structure according to the adjacency list, and uniformly adjusting the layout size in the completed state.

[0053] The front-end rendering logic of this invention adopts a technical path of "field-driven + state machine + incremental rendering". Specifically, the back-end (including large models) returns intermediate processes and final results in a streaming manner using structured fields. The front-end converts the fields into interface states and performs visualization and interactive control by module. Overall, the front-end rendering logic follows a layout strategy of dialogue flow (left side) - node content (right side) and maintains bidirectional linkage with the thinking node diagram (i.e., mind map). The front-end rendering logic is explained below from six aspects: rendering triggering, layout and display, node content rendering, interaction and re-rendering, exception and fault tolerance, performance optimization and security compliance.

[0054] Rendering triggers follow a driven pattern of receiving fields, event triggers, and state transitions. Rendering triggers include: draft thinking trigger 401, thought node graph trigger 402, node detail trigger 403, and final answer trigger 404. For example... Figure 4a As shown, a 401 error is triggered when the `formatted.thought` field is received: the front-end displays the thought process in a streaming manner in the dialogue flow area; after this stage is completed, the thought process is automatically hidden, and the process moves to the subsequent node / answer stage; this field is used for both the thought draft and the rendering of the final response text. Figure 4bAs shown, the thought node graph triggers a 402 error: When the formatted.adjacency_list (adjacency list field) is received, the front-end parses it into a tree-like node data structure and draws an interactive thought node graph to represent the multi-layered paths of reasoning / retrieval / integration in the large model. For example... Figure 4c As shown, a 403 error is triggered by node details: When a user clicks a node in the middle of the mind map, the front end loads the node content on the right as needed based on the fields carried by the node, and displays sub-blocks such as thinking, information source, and information integration in a collapsed / expanded manner. The field mapping relationships are as follows: Node title: fullInfo.content; Thinking content: fullInfo.response.formatted.thought; Search keywords: query; Source webpage: fullInfo.response.content; Webpage highlight information: fullInfo.response.formatted.action.parameters.select_ids; Information integration: fullInfo.response.formatted.thought. Figure 4d As shown, the final answer triggers a 404 error: When the large model completes generation and returns the final answer, it continues to render in the dialogue flow area as formatted.thought; if there is a citation, a citation icon is displayed after the corresponding paragraph, and the user can click it to jump to the source page.

[0055] The layout and display adopt a column layout. The interactive interface is divided into a dialogue flow area on the left and a node content area on the right; the node content area on the right is not displayed when the dialogue has not started. When formatted.thought appears for the first time, the node content area on the right automatically expands; after the final reply is completed, the node content area on the right automatically collapses to maintain the continuity of dialogue reading. The clickable range of nodes includes: the original question and the final reply node in the thought node diagram are not clickable; only the middle thought node is clickable to ensure that the user focuses on the middle reasoning chain; the small circle in the upper right corner of the node is used to show the reason for the generation of the node. The collapse / expand strategy is as follows: in the node content area on the right, the thought and information source are automatically collapsed by default after the output is completed, and the user can manually control the visibility of all sub-blocks (including information integration) through the expand / collapse controls.

[0056] Node content rendering includes presentation order, source, and highlighting. The presentation order is as follows: when a user clicks on an intermediate node, the front end renders it in the order of title, reflection, information source, and information integration; the reflection / information source is a collapsed block, displaying a summary and status by default, and providing an interactive handle to expand and view more. The source and highlighting are as follows: the source webpage is displayed in a list or card format, opening in a new window when clicked; for the highlighted information specified by select_ids, the front end marks it with location / highlighting, facilitating quick tracing of the evidence chain for users.

[0057] Interaction and re-rendering follow an event-triggered and state-transition driven pattern. Interaction and re-rendering include clicking on mind map nodes, clicking on reference icons, and controlling display status. Clicking on mind map nodes triggers rendering of the node content area on the right or switches to the cached content of the corresponding node; clicking again collapses the content; clicking the small circle on the node displays the reason for its generation. Specifically, in the front-end display logic, the content closest to this semantic meaning mainly comes from the `response.formatted.thought` field returned by the backend. This field carries the thought process explanation for the current node during generation, retrieval, and integration, and is displayed as thought or stage explanation text in the node content area on the right. If the reason for node generation refers to the node title or sub-question itself in business semantics, the corresponding source is the node's `content` field, which the front-end will organize into a `subQuestion` for display. Clicking on reference icons involves clicking the reference icon in the dialogue flow area to call the corresponding source and jump; clicking the source entry or highlighted information in the node content area also supports jumping. The display state is controlled as follows: users click the expand / collapse control to change the visibility of the sub-block; closing the control will collapse the entire node content panel; this interaction and display state are kept consistent in the front-end state machine, which facilitates state recovery when switching between nodes.

[0058] Exception handling and fault tolerance are part of the robust design. For missing / delayed fields: if a field is not returned or returns late, the front-end displays a skeleton screen or placeholder; incremental merging and updates are performed when the data is complete to avoid frequent backflows affecting the user experience. For scenarios without a source: when a node does not contain a source list, only the "Thinking / Integration" sub-block is displayed, and the reference icon is not shown. For unreachable links: if a source link fails to open, a notification bar is displayed, and the option to copy the link is retained.

[0059] Performance optimization and security compliance are the engineering guarantees. Performance optimization includes: using virtualized rendering and progressive layout for the thought node diagram (prioritizing the rendering of visible nodes and then filling in non-visible nodes) to support a large number of nodes; using incremental concatenation and batch DOM (Document Object Model) updates for the structure field `formatted.thought` in the dialogue flow area; and using throttle to reduce computational overhead for the mind map layout and right panel scroll positioning. Security compliance includes: XSS filtering and decoding whitelist rendering of source webpage summaries and highlighted content; opening external links in new windows / tabs with `rel="noopener"`; escaping user-readable input areas; maintaining consistent field naming and semantics, and avoiding front-end semantic rewriting of core fields such as `formatted.thought` and `formatted.adjacency_list`.

[0060] It is worth noting that the interactive system proposed in this invention, which visualizes the thinking and analysis process of a large-scale AI model search engine, is deeply coupled to a specific response format on the front end, but not directly coupled to any particular LLM. If the back end can output the above format uniformly, theoretically any LLM can be integrated, but the front end lacks any dynamic adaptation logic—if the format changes slightly, it will silently fail or degrade the display (because optional chains and try-catch are used everywhere). GenerationParams is a request structure defined based on Pydantic BaseModel. After FastAPI receives the JSON Body from POST / solve, it automatically performs type validation and field mapping, generating a structured request package containing three fields: inputs: the user's natural language input, supporting single-turn strings or multi-turn dialogue lists (List[Dict]); session_id: a session isolation identifier, randomly generated by default, used to distinguish the context of different users in the Agent's memory, and automatically cleaned up after the request ends; agent_cfg: Agent extended configuration, currently empty by default, serving as a reserved entry point for future feature expansion. After parsing, inputs and session_id are directly passed to agent() to drive the MindSearch search inference chain, and the inference process is returned to the client in real time in SSE streaming mode.

[0061] This invention also provides an interactive method for visualizing the thinking and analysis process of a large-scale artificial intelligence search engine. Figure 5 This diagram illustrates an interactive method for visualizing the thinking and analysis process of a large-scale artificial intelligence search engine, according to an embodiment of the present invention. Figure 5As shown, the interactive system that visualizes the thinking and analysis process of the AI ​​large-scale model search engine is used as an interactive method for visualizing the thinking and analysis process of the AI ​​large-scale model search engine. The method includes the following steps: Step 501: Receive the natural language query request input by the user.

[0062] Step 502: Send the natural language query request to the backend service system to drive the artificial intelligence big model search engine to process it, and receive the structured response data returned by the backend service system. The structured response data includes at least a first structure field and a second structure field.

[0063] Step 503: The thought draft visualization module in the visualization interactive rendering engine parses the first structural field and dynamically displays the current thought content in the form of flowing text during the thinking process of the artificial intelligence big model.

[0064] Step 504: The visualization module of the visualization interactive rendering engine parses the second structure field and generates and displays an interactive thinking node graph.

[0065] Step 505: In response to the user's interactive operation on the nodes in the thought node diagram, the node content display module in the visualization interactive rendering engine displays the detailed content of the selected node in the interactive interface.

[0066] Step 506: The final response display module in the visualization interactive rendering engine generates the final response content on the interactive interface.

[0067] In one embodiment of the present invention, during the process of receiving structured response data, the visualization interactive rendering engine continuously writes the data of each node into the local storage of the user terminal; when the user clicks on a node in the mind node diagram to switch node content, the node content display module first reads the data of the corresponding node from the local storage for display.

[0068] In one embodiment of the present invention, the interactive method for visualizing the thinking and analysis process of the artificial intelligence large model search engine further includes: storing natural language query requests, structured response data, and user interaction operation records; responding to a user's request to switch and view historical dialogue rounds in a conversation, the visualization interactive rendering engine restores the corresponding dialogue content and node state based on local storage; or, when the local cache expires, it requests historical data from the backend service system based on the session identifier, and reconstructs and retrospectively displays the thinking process of the historical dialogue based on the returned historical data.

[0069] It is worth noting that the interactive system and method proposed in this invention, which visualizes the thinking and analysis process of a large-scale AI model search engine, can successfully run and clearly demonstrate the model's thinking process by calling the MindSearch open-source AI search engine framework.

[0070] This invention proposes an interactive system and method for visualizing the thinking and analysis process of a large-scale artificial intelligence model search engine, constructing a dual-dimensional explanation system at the node and path levels. On the one hand, users can stream the thought draft of each node through the `thought` field, enabling them to understand the specific reasoning logic of the model at that stage. On the other hand, by setting explanation controls on the nodes of the thought node diagram, users can view with one click why the model derived that node, revealing the logical connections across nodes, greatly enhancing the interpretability of the model's decisions, and enabling users to verify, question, and trust the search results.

[0071] This invention proposes an interactive system and method that visualizes the thinking and analysis process of an AI large-scale model search engine, significantly enhancing users' active control and interactivity over this process. Through a dynamic, collapsible content architecture and rich manual controls (such as automatic node expansion / collapse strategies and manual control of module display levels), users are given complete control over the granularity of the thinking path display and the depth of exploration. The design of multi-level interactive verification channels (such as clicking nodes to view search keywords, direct links to highlighted webpage information, and citation tagging for tracing) enables users to verify the authenticity and rationality of the large-scale model's decision-making basis in real time.

[0072] In one embodiment of the present invention, an electronic device is also provided, comprising a processor, a graphics card, and a memory. The memory is configured to store machine-readable instructions, the graphics card is configured to train the interactive method for visualizing the thinking and analysis process of an AI large-scale model search engine, and the processor is configured to execute the machine-readable instructions. When the processor and / or graphics card executes the machine-readable instructions, the following processing steps are implemented: receiving a natural language query request input by a user; sending the natural language query request to a backend service system to drive the AI ​​large-scale model search engine to process it, and receiving structured response data returned by the backend service system, the structured response data including at least a first structure field and a second structure field; a thinking draft visualization module in the visualization interactive rendering engine parses the first structure field and dynamically displays the current thinking content in the form of streaming text during the AI ​​large-scale model's thinking process; a thinking node graph visualization module in the visualization interactive rendering engine parses the second structure field, generates and displays an interactive thinking node graph; in response to user interaction with nodes in the thinking node graph, a node content display module in the visualization interactive rendering engine displays the detailed content of the selected node in the interactive interface; and a final response display module in the visualization interactive rendering engine generates the final response content in the interactive interface.

[0073] The graphics card used can preferably be a model with a GPU computing power higher than 5.0. Since the amount of data to be trained is large, providing a graphics card configuration can significantly improve the training speed.

[0074] The memory includes various media capable of storing machine-readable instructions, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0075] It is understood that, in addition to the memory and processor mentioned above, the computer system described above also includes other hardware and software components not listed in this specification. The specific components can be determined according to the model of the specific data processing equipment in different application scenarios, and will not be listed and described in detail in this specification.

[0076] In one embodiment of the present invention, a computer-readable storage medium is also provided, on which machine-readable instructions are stored. When executed by a processor, the machine-readable instructions perform the following processing steps: receiving a natural language query request input by a user; sending the query request to a backend service system to drive an artificial intelligence large-scale model search engine to process it, and receiving structured response data returned by the backend service system, the structured response data including at least a first structure field and a second structure field; a thought draft visualization module in a visualization interactive rendering engine parses the first structure field and dynamically displays the current thought content in the form of streaming text during the thinking process of the artificial intelligence large-scale model; a thought node graph visualization module in the visualization interactive rendering engine parses the second structure field and generates and displays an interactive thought node graph; in response to the user's interactive operation on the nodes in the thought node graph, a node content display module in the visualization interactive rendering engine displays the detailed content of the selected node in the interactive interface; and a final response display module in the visualization interactive rendering engine generates the final response content in the interactive interface.

[0077] Although various embodiments of the present invention have been described above, it should be understood that they are presented by way of example only and not as limitations. It will be apparent to those skilled in the art that various combinations, modifications, and alterations can be made without departing from the spirit and scope of the invention. Therefore, the breadth and scope of the invention disclosed herein should not be limited by the exemplary embodiments disclosed above, but should be defined according to the technical solutions of the invention and their equivalents.

Claims

1. An interactive system for visualizing a thinking analysis process of an artificial intelligence large model search engine, characterized by, include: The user terminal is configured to receive natural language query requests input by the user and display a visual interactive interface; The backend service system is configured to call the AI ​​large model search engine and receive structured response data returned by the AI ​​large model search engine based on the natural language query request. The structured response data includes at least a first structure field and a second structure field. as well as A visual interactive rendering engine is configured to parse the structured response data and render it as an interactive interface on the user terminal; wherein, the visual interactive rendering engine includes: The thought draft visualization module is configured to parse the first structure field and dynamically display the current thought content in the form of flowing text during the thinking process of the artificial intelligence big model. The Thinking Node Graph Visualization Module is configured to parse the second structure field, generate and display an interactive thinking node graph; The node content display module is configured to respond to user interaction with nodes in the thought node diagram and display the detailed content of the selected node in the interactive interface. Final Response Display Module: Configured to generate the final response content on the interactive interface.

2. The interactive system of claim 1, wherein, Also includes: The data storage module is configured to store user requests, large model response content, thought node information, and interaction records.

3. The interactive system of claim 1, wherein, The visualization interactive rendering engine also includes a node detailed content module, which is configured to lay out the node title, thought process, information source, and information integration result in the order of rendering the node detailed content. The thought process and information source are displayed in a collapsed state by default after the content is output, and one or more expand / collapse controls are provided to respond to user operations to switch the visibility of their content.

4. The interactive system of claim 1, wherein, In the thought node diagram generated by the thought node diagram visualization module, the user's original question node and final response node are set to be unclickable, while the intermediate nodes are set to be clickable. When the user clicks the intermediate node, the node details module is triggered to load and display the detailed content of the node.

5. The interactive system of claim 1, wherein, The node content display module is further configured to divide the interactive interface into a dialogue flow area on the left and a node content area on the right; wherein, the node content area has preset control logic so that it is hidden by default before the dialogue begins, automatically expands when the visualization interactive rendering engine receives the first structure field, and automatically collapses after the large model completes its response.

6. The interactive system of claim 1, wherein, The backend service system includes: Request parsing module: Configured to parse user-input natural language query requests and generate structured request packets; and Model Invocation Module: Configured to invoke the AI ​​Big Model Search Engine to process the structured request packet and receive the structured response data output by the AI ​​Big Model Search Engine.

7. The interactive system of claim 1, wherein, The user terminal includes an interactive control module; The interactive control module is configured to respond to a user's click on a reference mark in the final reply content, call the corresponding information source webpage link and open it in a new window; In addition, it is configured to control the display state of the node content display module in response to the user's click on the expand / collapse control.

8. The interactive system of claim 1, wherein, The visualization and interactive rendering engine uses a fault-tolerant processing strategy to parse the structured response data using null value judgment and default value rollback mechanism; when the structured response data is missing or its structure does not meet the rendering conditions, the thinking node diagram visualization module does not draw the thinking node diagram.

9. An interactive method for visualizing a thinking analysis process of an artificial intelligence large model search engine, applied to the interactive system of any one of claims 1 to 8. Includes the following steps: Receive natural language query requests from users; The natural language query request is sent to the backend service system to drive the artificial intelligence big model search engine to process it, and the structured response data returned by the backend service system is received. The structured response data includes at least a first structure field and a second structure field. The thought draft visualization module in the visualization interactive rendering engine parses the first structure field and dynamically displays the current thought content in the form of flowing text during the thinking process of the artificial intelligence big model. The visualization module of the visualization interactive rendering engine parses the second structure field to generate and display an interactive thinking node diagram. In response to the user's interactive operation on the nodes in the thought node diagram, the node content display module in the visualization interactive rendering engine displays the detailed content of the selected node in the interactive interface. The final response display module in the visualization interactive rendering engine generates the final response content on the interactive interface.

10. The interaction method of claim 9, wherein, During the process of receiving the structured response data, the visualization interactive rendering engine continuously writes the data of each node into the local storage of the user terminal; When a user clicks on a node in the mind map to switch node content, the node content display module first reads the data of the corresponding node from the local storage for display.

11. The interaction method of claim 9, wherein, Also includes: The natural language query requests, structured response data, and user interaction records are stored. In response to a user's request to switch and view historical conversation rounds in a session, the visualization and interactive rendering engine restores the corresponding conversation content and node state based on local storage; Alternatively, when the local cache expires, historical data is requested from the backend service system based on the session identifier, and the thought process of the historical conversation is reconstructed and displayed based on the returned historical data.

12. An electronic device, comprising: include: A processor, configured to execute machine-readable instructions; The graphics card is configured to train the interactive method for visualizing the thinking and analysis process of a large-scale artificial intelligence model search engine as described in any one of claims 9 to 11; and A memory configured to store machine-readable instructions that, when executed by a processor and / or a graphics card, perform the steps of the interactive method for visualizing the thinking and analysis process of an artificial intelligence large-scale model search engine according to any one of claims 9 to 11.

13. A computer-readable storage medium having stored thereon computer-readable instructions, wherein, When the computer-readable instructions are executed by the processor, they perform the steps of the interactive method described in any one of claims 9 to 11, which visualizes the thinking and analysis process of the artificial intelligence large-scale model search engine.

Images