Large model parallel interaction method and system based on tree topology structure

Abstract

The invention provides a large model parallel interaction method and system based on a tree topology structure, and relates to the technical field of artificial intelligence interaction.The method comprises the steps that each round of interaction between a user and a large model is defined as a node; according to the large model answer of the current node, a plurality of child nodes are generated through a knowledge point extractor, and a tree topology structure is constructed; when a user triggers any node, recursively tracing a path from the node to a root node, and aggregating node data according to a path sequence to generate a precise context; initiating a question to a large model based on the context, writing an answer into a current node and automatically generating a new child node; a user is supported to initiate interaction requests in parallel among different nodes, and the system asynchronously processes tasks of all the nodes. According to the technical scheme of large model parallel interaction based on a tree structure, scattered chat records are changed into a knowledge map which can be unfolded at will, details can be dug deeply in the longitudinal direction, different topics can be compared transversely, and multi-node simultaneous propulsion can be achieved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence interaction technology, and is particularly suitable for dialogue management, context control, and multi-task parallel processing scenarios of large-scale language models (LLMs). Specifically, it relates to a large-model parallel interaction method and system based on a tree topology structure. Background Art

[0002] The current mainstream large-scale interaction technologies (such as ChatGPT, deepseek, Wenxin Yiyan, etc.) mainly adopt a linear dialogue mode, that is, a question-and-answer format. In its information organization form, a single question and answer is an independent unit, and the dialogue history is arranged linearly in chronological order to form a continuous chat record text stream; the interaction logic is that the user manually constructs a logical chain by repeatedly asking questions. Each dialogue only supports a single path extension. When managing context, in order to imitate the human memory function and enable the large model to understand the current chat context, all dialogue records will be sent to the large model for reasoning in the form of a text stream. There is no dialogue content screening mechanism; the interaction timing is that the user needs to wait for the answer to the current question to be generated before initiating a new question, and the question-answering process is strictly serialized.

[0003] Based on the characteristics of the above existing technologies, the current mainstream large-scale model interaction technology has the following technical defects:

[0004] 1. Severe information fragmentation: The knowledge points users acquire are scattered across lengthy, continuous conversations. Users must manually organize the information (such as copying and pasting, and manually annotating) or use prompts to insert summary instructions after several rounds of conversation, significantly reducing cognitive efficiency.

[0005] 2. Single expansion dimension: It only supports one-way extended dialogue paths and cannot vertically expand and deepen multiple aspects of the generated answers.

[0006] 3. Low resource utilization: Loading the full historical context results in a large amount of irrelevant information being involved in the calculation. Unfiltered historical noise not only increases server load but also easily causes information hallucinations and even semantic drift.

[0007] 4. Lack of backtracking path: It is impossible to quickly locate specific content in historical conversations. Users need to browse through historical records one by one, which makes it easy to miss important information. There is a lack of visual navigation tools.

[0008] 5. Interaction efficiency bottleneck: The serial question-answering model forces users to wait for each answer in sequence, making it impossible to explore multiple knowledge branches simultaneously, resulting in limited processing efficiency.

[0009] 6. There is no way to return to the context of a certain moment in history. Summary of the Invention

[0010] In view of this, and in response to the above-mentioned defects, the purpose of the present invention is to propose a large-model parallel interaction method and system based on a tree-like topological structure, which expands the tree-like knowledge node based on the answer of the large model, and switches the context of the tree-like nodes, through multi-tasking parallelism, to solve the serious problems of information fragmentation caused by conversation records organized in chronological order, the problem of a single expansion dimension caused by linear conversations, the problem of low resource utilization caused by the full historical context, the problem of being unable to return to the context of a certain moment in history, and the problem of bottleneck in the efficiency of question-and-answer interaction.

[0011] To achieve the above object, the present invention provides the following technical solutions:

[0012] Based on the above objectives, in a first aspect, the present invention provides a large model parallel interaction method based on a tree topology structure, comprising the following steps:

[0013] Each round of interaction between the user and the big model is defined as a node. The data structure of each node includes: user question, big model answer, node number, and parent node number;

[0014] Based on the macro model answer of the current node, a number of child nodes are generated through the knowledge point extractor to build a tree topology structure;

[0015] When a user triggers any node, the path from the node to the root node is recursively traced, and node data is aggregated in the order of the path to generate an accurate context;

[0016] Ask the big model a question based on the context, write the answer to the current node and automatically generate a new child node;

[0017] It supports users to initiate interactive requests in parallel between different nodes, and the system processes tasks of each node asynchronously.

[0018] As a further solution of the present invention, each round of interaction between the user and the large model is defined as a node, wherein the definition of the node includes:

[0019] After the user asks a question, a root node is generated based on the question and associated with the answer of the large model;

[0020] The knowledge point extractor is used to extract several knowledge points from the answer of the large model, and a child node is generated for each knowledge point to build a tree topology structure.

[0021] As a further solution of the present invention, each node includes:

[0022] Node number, parent node number, user-asked question, and the answer from the large model;

[0023] After the answer from the large model is processed, the generated knowledge points are used to construct new child nodes and establish connections with the current node.

[0024] As a further solution of the present invention, the child node generation step includes:

[0025] Input the large model answer into the knowledge point extractor and output N knowledge points;

[0026] Create a child node for each knowledge point and associate the current node as the parent node.

[0027] As a further solution of the present invention, when aggregating node data in path order to generate a precise context, the context construction step includes:

[0028] Taking the trigger node as the endpoint, recursively trace back to the root node along the parent node number;

[0029] Concatenate the questions and answers of all nodes in the order of the root node and the current node to generate a path context.

[0030] As a further solution of the present invention, the node triggering method includes:

[0031] Click on the end node in the tree topology diagram to automatically generate a question: "Based on your answer just now, explain in detail: [node title]";

[0032] Enter a new question in the chat box to create a child node for the current node titled User Input.

[0033] As a further solution of the present invention, when clicking on the end node in the tree topology diagram, the question and answer content of all parent nodes is obtained by recursively tracing back from the selected node to the root node, and the question and answer content is arranged in order from root to end to build the context of the node; a new question is generated according to the title of the current node, and the question and the obtained context are submitted to the big model together.

[0034] As a further solution of the present invention, the parallel interaction is implemented as follows:

[0035] While waiting for an answer at the first node, the user can switch to the second node to raise a new question;

[0036] The system processes tasks of each node independently without blocking each other.

[0037] As a further solution of the present invention, when the system asynchronously processes tasks of each node, the front-end interface of the system includes:

[0038] A view showing a tree topology, where nodes show all interaction records from the root node to the current node;

[0039] Users can switch contexts by clicking on different nodes and view all conversation records related to the node in the chat box.

[0040] As a further solution of the present invention, the front-end interface of the system is provided with:

[0041] The historical session list and the new session button on the left;

[0042] The middle tree topology visualization area;

[0043] The complete conversation record display area of ​​the root node and the current node on the right;

[0044] Chat box and large model service selector at the bottom.

[0045] In a second aspect, the present invention further provides a large model parallel interaction system based on a tree topology structure, comprising the following components:

[0046] Tree topology building module: used to generate nodes that interact with the large model and their corresponding child nodes to form a tree structure;

[0047] Context management module: manages and switches the context of different nodes through node numbers and parent node numbers to ensure the accuracy of the question and answer context of each node;

[0048] Parallel interaction module: supports parallel interaction between multiple nodes, allowing users to switch between multiple nodes and ask questions.

[0049] Compared with existing technologies, the present invention proposes a large-model parallel interaction method and system based on a tree-like topology structure. This large-model parallel interaction technical solution based on a tree-like structure transforms scattered chat records into a knowledge map that can be expanded at will. It can not only dig deep into details vertically, but also compare different topics horizontally, and advance multiple nodes simultaneously. Compared with traditional dialogue systems, it can achieve the following beneficial effects:

[0050] 1. Revolutionary improvement in cognitive effects: Each question and answer is converted into a node in a tree structure. Child nodes are automatically created after the answer is generated. The visual tree topology allows users to intuitively grasp the overall picture of the knowledge system, fundamentally solving the problem of information fragmentation and improving information absorption efficiency.

[0051] 2. Breakthrough expansion of the interaction dimension: Supports initiating hierarchical questions (vertical deepening) or same-level expansion (horizontal broadening) for any node, solving the single extension problem of traditional interaction methods.

[0052] 3. Significantly optimized resource utilization: Based on the node tracing path, context information is dynamically adjusted, and irrelevant information from other paths is filtered out. This allows for precise filtering of historical conversation content, increasing the proportion of valid information, reducing the computational burden, and significantly alleviating semantic drift and information hallucination issues.

[0053] 4. Intelligent contextual navigation: The tree-like topology of conversations allows easy tracing back to any node, allowing easy access to important Q&A records at different nodes and the ability to continue asking questions at any node. This improves reuse and organization efficiency, preventing important content from being missed.

[0054] 5. Breaking through the bottleneck of interaction efficiency: Users can activate follow-up requests from multiple branches simultaneously. The system processes different requests asynchronously without interfering with each other, breaking the timing limitations of traditional serial interactions. While waiting for a response from a branch, users can continue to explore other nodes, greatly improving information acquisition efficiency.

[0055] These and other aspects of the present application will be more clearly understood in the following description of the embodiments. It should be understood that the above general description and the following detailed description are merely exemplary and explanatory and cannot limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the following briefly introduces the drawings required for the exemplary embodiments or related technical descriptions. The drawings are used to provide a further understanding of the present invention and constitute part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the drawings:

[0057] Figure 1 This is a first front-end flow diagram of a large model parallel interaction method based on a tree topology structure according to an embodiment of the present invention.

[0058] Figure 2 This is a second front-end flow diagram of a large model parallel interaction method based on a tree topology structure in an embodiment of the present invention.

[0059] Figure 3 This is a third front-end flow diagram of a large model parallel interaction method based on a tree topology structure according to an embodiment of the present invention. DETAILED DESCRIPTION

[0060] Below, the present application is further described in conjunction with the accompanying drawings and specific implementation methods. It should be noted that, under the premise of no conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.

[0061] To make the purpose, technical solutions and advantages of the present invention more clearly understood, the following is a further detailed description of the embodiments of the present invention in conjunction with specific embodiments and with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.

[0062] It should be noted that all expressions using "first" and "second" in the embodiments of the present invention are intended to distinguish two non-identical entities or non-identical parameters with the same name. Therefore, "first" and "second" are used for convenience of expression only and should not be understood as limitations on the embodiments of the present invention. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, other steps or units inherent to a process, method, system, product, or device that includes a series of steps or units.

[0063] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0064] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0065] The following describes some embodiments of the present application in detail with reference to the accompanying drawings. In the absence of conflict, the following embodiments and features therein may be combined with each other.

[0066] In response to the technical defects of the current mainstream large-scale model interaction technology, this paper proposes a large-scale model parallel interaction method and system based on a tree topology structure. Based on the answers of the large model, the tree-like knowledge nodes are expanded, and based on the context switching of the tree nodes, multi-tasking is carried out in parallel to solve the serious problems of information fragmentation caused by chronologically organized conversation records, the problem of single expansion dimension caused by linear conversations, the problem of low resource utilization caused by full historical context, the problem of being unable to return to the context of a certain moment in history, and the bottleneck of question-and-answer interaction efficiency.

[0067] See also Figure 1 As shown, an embodiment of the present invention provides a large model parallel interaction method based on a tree topology structure, including expanding tree knowledge nodes based on answers of the large model, context switching based on the tree nodes, and multi-task parallelism. The method specifically includes the following steps:

[0068] Step S10: define each round of interaction between the user and the big model as a node. The data structure of each node includes: user question, big model answer, node number, and parent node number.

[0069] In this step, each node includes: node number, parent node number, question raised by the user and answer of the large model; after the answer of the large model is processed, the generated knowledge points are used to construct a new child node and establish a connection with the current node.

[0070] Step S20: Based on the macro-model answer of the current node, a knowledge point extractor is used to generate several child nodes to construct a tree topology structure.

[0071] In this step, each round of interaction between the user and the big model is defined as a node, where the definition of the node includes: after the user asks a question, a root node is generated based on the question, and the node is associated with the answer of the big model; a knowledge point extractor is used to extract several knowledge points from the answer of the big model, and a child node is generated for each knowledge point to construct a tree topology structure.

[0072] The sub-node generation step includes:

[0073] Input the large model answer into the knowledge point extractor and output N knowledge points;

[0074] Create a child node for each knowledge point and associate the current node as the parent node.

[0075] The tree topology consists of several child nodes, each representing a round of interaction with the large model—one question, one answer. A node is a basic storage unit, storing core data including the user's question, the large model's answer, the node ID, and the parent node ID. The large model's answer for each node is processed to generate several knowledge points. Child nodes are created based on these knowledge points and connected to the current node.

[0076] If you click on a newly generated child node, it will be selected, and then the context and question will be automatically constructed to ask the backend model. The generated answer will be written to the current node, and several child nodes will be expanded based on this answer.

[0077] If you enter text in the chat box and press Enter, a child node titled "the user's question" will be created for the currently selected node. Once this child node is created, the user's question, along with its context, is automatically sent to the backend model. Once an answer is received, several child nodes are automatically generated based on the answer.

[0078] Through the above operations, after the user asks a question in text for the first time, he can expand a knowledge tree by just clicking the mouse.

[0079] The specific method flow for expanding the subnode is as follows:

[0080] (1) When a terminal node is clicked, the system will first start with the clicked node and recursively trace back all parent nodes according to the parent node number until the root node is reached. Then, the questions and answers in all nodes will be arranged in order from root to terminal to form the context information of the current node.

[0081] (2) Get the title of the current node, generate a question, add it to the context generated in the previous step, and submit it to the big model (for example, if the title of the current node section is "Classification of Psychology", the generated question may be "Based on your previous answer, explain in detail: Classification of Psychology").

[0082] (3) The answer of the large model is written into the current node and displayed in the conversation record window.

[0083] (4) Input the answer of the large model into the knowledge point extractor at the back end, output several knowledge points, and then construct a new terminal sub-node.

[0084] according to Figures 1 to 3 As shown, the process of building the tree topology includes:

[0085] like Figure 1 As shown in the figure, after the user creates a new session, he asks "Question A". At this time, A is the root node and has no context information. The large model answers question A and obtains "Answer BCDE". After the knowledge point extractor extracts the knowledge points, it constructs 4 child nodes BCDE.

[0086] like Figure 2 As shown, click node B. The context is [A Question, A Answer]. The model is asked, "Based on your previous answer, explain B in detail." It receives the answer, "Answer FGH." After further summarizing and inductive analysis, the three child nodes FGH are constructed. The chat history on the right now shows all the questions and answers from A to B: A Question, A Answer, B Question, B Answer.

[0087] like Figure 3 As shown, click on the other child node E under node A, and the PROMPT project automatically builds the question "Question E". Note that the context information at this time is still [A Question, A Answer], because the current path is A->E, and the chat record on the right will not show steps 2 and Figure 2 The B node is shown as a question and answer. Questions are asked to the large model and the answer is "Answer IJK". After further summarization, the three child nodes of IJK are constructed.

[0088] like Figure 3, Node A is the root node, its child nodes are BCDE, Node B has child nodes FGH, Node E has child nodes IJK, and the terminal node is CDFGHIJK. Node BE is a node with complete data, and node CDFGHIJK is the terminal node so it does not store the large model answer.

[0089] Step S30: When the user triggers any node, the path from the node to the root node is recursively traced, and the node data is aggregated in the order of the path to generate an accurate context.

[0090] In this step, when node data is aggregated in path order to generate accurate context, the context construction steps include:

[0091] Taking the trigger node as the endpoint, recursively trace back to the root node along the parent node number;

[0092] Concatenate the questions and answers of all nodes in the order of the root node and the current node to generate a path context.

[0093] Node triggering methods include:

[0094] Click on the end node in the tree topology diagram to automatically generate a question: "Based on your answer just now, explain in detail: [node title]";

[0095] Enter a new question in the chat box to create a child node for the current node titled User Input.

[0096] In this embodiment, when the end node in the tree topology diagram is clicked, the question and answer content of all parent nodes is obtained by recursively tracing back from the selected node to the root node, and the question and answer content is arranged in order from root to end to build the context of the node; a new question is generated according to the title of the current node, and the question and the obtained context are submitted to the big model together.

[0097] Step S40: Ask a question to the large model based on the context, write the answer into the current node and automatically generate a new child node.

[0098] In this step, context switching is based on the tree topology. Figure 1 As shown, an interaction with the large model is constructed into a node, and the horizontal and vertical extensions of the node form a tree-like topological structure. Based on the tree structure, this patent achieves precise control of the context.

[0099] In this embodiment, nodes are used as storage units, which store node numbers and parent node numbers. With these two numbers, the path data from the root node to the current node can be easily obtained. As described in point 1 of the child node expansion process, the questions and answers in all nodes are arranged in order from root to end, thereby constructing an accurate path context. Figure 3 As shown, if you click on node I, the tree-like topology shows that the node chain is A->E->I, and the context is Q&A A + Q&A E. Similarly, if you click on node G, the context is Q&A A + Q&A B. This shows that this patent manages and controls context through a tree-like topology node chain, allowing users to switch historical contexts at will.

[0100] Step S50: Support users to initiate interactive requests in parallel between different nodes, and the system asynchronously processes tasks of each node.

[0101] In this step, the parallel interaction is implemented as follows:

[0102] While waiting for an answer at the first node, the user can switch to the second node to raise a new question;

[0103] The system processes tasks of each node independently without blocking each other.

[0104] In this embodiment, when the system asynchronously processes tasks at each node, the front-end interface of the system includes:

[0105] A view showing a tree topology, where nodes show all interaction records from the root node to the current node;

[0106] Users can switch contexts by clicking on different nodes and view all conversation records related to the node in the chat box.

[0107] Among them, the front-end interface of the system is set up with:

[0108] The historical session list and the new session button on the left;

[0109] The middle tree topology visualization area;

[0110] The complete conversation record display area of ​​the root node and the current node on the right;

[0111] Chat box and large model service selector at the bottom.

[0112] See also Figure 1 As shown, the left side of the front-end page displays the user's historical conversation history, with a New Conversation button in the upper left corner. The center of the page displays a tree structure consisting of a root node and multiple levels of child nodes. The right side of the page displays all conversations from the root node to the current node. Below that, there's a chat box for users to ask questions and select different large model services.

[0113] In this embodiment, when navigating precisely in a tree-like topology, the front-end interface provides real-time tree-like topology rendering, which allows users to clearly perceive the knowledge structure of the entire question-and-answer exploration. Compared with the traditional linear scrolling search history, clicking on the nodes of the tree-like topology is of course a faster and more accurate navigation method. For example, when the user expands Figure 3 When viewing the tree-structured topology diagram shown, note that the chat box on the right only displays the Q&A record from the root node A to the node E, i.e., Q&A A + Q&A E. If the user is not interested in the content of the current node E but is interested in the content related to the previous node B, the user can easily find the location of node B on the topology diagram and click on node B. The chat record on the right switches to the context of node B, i.e., the Q&A record from the root node A to the node B is displayed. At this time, the content related to node E is hidden, making it convenient for the user to return to the context semantics at that time.

[0114] When multiple nodes interact in parallel, the traditional large model's interaction mode is a conversational chat, a question-and-answer format. While the large model is answering, users cannot ask additional questions and must wait for the large model to finish answering before asking the next question. However, the tree-like topology constructed by the present invention allows switching between multiple nodes, which naturally creates the need and corresponding advantages of asking questions simultaneously at different nodes. While one node is waiting for the backend to return the result, users can switch to other nodes to ask questions.

[0115] In summary, the large-scale parallel interaction method based on the tree topology of the present invention converts each question and answer into a node in the tree structure. After the answer is generated, child nodes are automatically created. The visual tree topology allows users to intuitively grasp the overall picture of the knowledge system, fundamentally solving the problem of information fragmentation and improving information absorption efficiency. It supports initiating hierarchical questions (vertical deepening) or same-level expansion (horizontal broadening) on ​​any node, solving the single extension problem of traditional interaction methods. According to the node tracing path, the context information is dynamically adjusted, and irrelevant information of other paths is filtered. The historical conversation content is accurately filtered, which increases the proportion of effective information, reduces the computational burden, and greatly alleviates the problems of semantic drift and information illusion. The tree topology structure composed of the conversation can be easily traced back to any node, making it convenient to view important question and answer records at different nodes, and to continue asking questions at any node. It improves reuse and organization efficiency and no longer misses important content. Users can activate multiple branches of question requests at the same time. The system processes different requests asynchronously without interfering with each other, breaking the timing limitations of traditional serial interaction. While waiting for a branch to respond, other nodes can continue to be explored, greatly improving information acquisition efficiency.

[0116] It should be noted that the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention and are not intended to be limiting. It is readily understood that the processes illustrated in the above figures do not indicate or limit the temporal order of these processes. Furthermore, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0117] It should be understood that, although the above is described in a certain order, these steps are not necessarily performed in sequence according to the above order. Unless clearly stated herein, the execution of these steps does not have strict order restrictions, and these steps can be performed in other orders. Moreover, a part of the steps of the present embodiment may include multiple steps or multiple stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of the steps or stages in other steps or other steps.

[0118] In a second aspect of the embodiments of the present invention, the present invention further provides a large model parallel interaction system based on a tree topology structure, comprising:

[0119] Tree topology building module: used to generate nodes that interact with the large model and their corresponding child nodes to form a tree structure;

[0120] Context management module: manages and switches the context of different nodes through node numbers and parent node numbers to ensure the accuracy of the question and answer context of each node;

[0121] Parallel interaction module: supports parallel interaction between multiple nodes, allowing users to switch between multiple nodes and ask questions

[0122] Through the above detailed steps, the large model parallel interaction system based on tree topology structure of the present invention is used to execute the steps of the large model parallel interaction method based on tree topology structure in the above embodiment, which will not be repeated here. The large model parallel interaction system based on tree topology structure of the present invention, based on the large model parallel interaction technical solution of tree structure, turns scattered chat records into a knowledge map that can be expanded at will, which can not only dig deep into details vertically, but also compare different topics horizontally, and can also advance multiple nodes simultaneously, compared with traditional dialogue systems.

[0123] The above are exemplary embodiments disclosed in the present invention, but it should be noted that various changes and modifications may be made without departing from the scope of the embodiments disclosed in the claims. The functions, steps and / or actions of the method claims according to the disclosed embodiments described herein do not need to be performed in any particular order. In addition, although the elements disclosed in the embodiments of the present invention may be described or required in individual form, they may also be understood as multiple unless expressly limited to the singular.

[0124] It should be understood that, as used herein, the singular form "a" or "an" is intended to include the plural form as well, unless the context clearly supports an exception. It should also be understood that, as used herein, "and / or" refers to any and all possible combinations of one or more of the items listed in association. The serial numbers of the embodiments disclosed in the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0125] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the disclosure of the embodiments of the present invention (including the claims) is limited to these examples. Within the spirit of the embodiments of the present invention, the technical features of the above embodiments or different embodiments may be combined, and there are many other variations of different aspects of the above embodiments of the present invention, which are not provided in detail for the sake of simplicity. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present invention should be included in the scope of protection of the embodiments of the present invention.

Claims

1. A large model parallel interaction method based on tree topology structure, characterized in that: The method comprises the following steps: Each round of interaction between the user and the big model is defined as a node. The data structure of each node includes: user question, big model answer, node number, and parent node number; Based on the macro model answer of the current node, a number of child nodes are generated through the knowledge point extractor to build a tree topology structure; When a user triggers any node, the path from the node to the root node is recursively traced, and node data is aggregated in the order of the path to generate an accurate context; Ask the big model a question based on the context, write the answer to the current node and automatically generate a new child node; It supports users to initiate interactive requests in parallel between different nodes, and the system processes tasks of each node asynchronously.

2. The large model parallel interaction method based on tree topology structure according to claim 1, characterized in that: Each round of interaction between the user and the big model is defined as a node, where the definition of a node includes: After the user asks a question, a root node is generated based on the question and associated with the answer of the large model; The knowledge point extractor is used to extract several knowledge points from the answer of the large model, and a child node is generated for each knowledge point to build a tree topology structure.

3. The large model parallel interaction method based on tree topology structure according to claim 2, characterized in that: Each node includes: Node number, parent node number, user-asked question, and the answer from the large model; After the answer from the large model is processed, the generated knowledge points are used to construct new child nodes and establish connections with the current node.

4. The large model parallel interaction method based on tree topology structure according to claim 3, characterized in that: The steps to generate child nodes include: Input the large model answer into the knowledge point extractor and output N knowledge points; Create a child node for each knowledge point and associate the current node as the parent node.

5. The large model parallel interaction method based on tree topology structure according to claim 1, characterized in that: When aggregating node data in path order to generate an accurate context, the context construction steps include: Taking the trigger node as the endpoint, recursively trace back to the root node along the parent node number; Concatenate the questions and answers of all nodes in the order of the root node and the current node to generate a path context.

6. The large model parallel interaction method based on tree topology structure according to claim 5, characterized in that: Node triggering methods include: Click on the end node in the tree topology diagram to automatically generate a question: "Based on your previous answer, explain in detail: [node title]"; Enter a new question in the chat box to create a child node for the current node titled User Input.

7. The large model parallel interaction method based on tree topology structure according to claim 6, characterized in that: When clicking on an end node in the tree topology, the system recursively traces back from the selected node to the root node, obtains the Q&A content of all parent nodes, and arranges them in order from root to end to build the context of the node; Generate a new question based on the title of the current node and submit the question and the obtained context to the big model.

8. The large model parallel interaction method based on tree topology structure according to claim 1, characterized in that: Parallel interaction is implemented as follows: While waiting for an answer at the first node, the user can switch to the second node to raise a new question; The system processes tasks of each node independently without blocking each other.

9. The large model parallel interaction method based on tree topology structure according to claim 8, characterized in that: When the system asynchronously processes tasks at each node, the system's front-end interface includes: A view showing a tree topology, where nodes show all interaction records from the root node to the current node; Users switch contexts by clicking different nodes and view all conversation records related to the node in the chat box.

10. A large model parallel interactive system based on a tree topology, characterized in that: The system is used to execute the large model parallel interaction method based on the tree topology structure according to any one of claims 1 to 9, comprising: Tree topology building module: used to generate nodes that interact with the large model and their corresponding child nodes to form a tree structure; Context management module: manages and switches the context of different nodes through node numbers and parent node numbers to ensure the accuracy of the question and answer context of each node; Parallel interaction module: supports parallel interaction between multiple nodes, allowing users to switch between multiple nodes and ask questions.

Images