Information processing method, electronic device, and storage medium

Abstract

Provided in the present disclosure are an information processing method, an electronic device, and a storage medium, which are applied to the technical field of artificial intelligence. The method comprises: determining information to be processed corresponding to an intelligent agent; and inputting the information to be processed and node sequence information into a large model, wherein the node sequence information is used for indicating a plurality of nodes included in a task flow and edges between the nodes, each node represents an action instruction for the intelligent agent, and each edge represents a transition between the nodes, such that the large model determines, on the basis of the information to be processed, a target node to be executed and controls the intelligent agent to execute the action instruction corresponding to the target node. The present disclosure enables the planning of a complex flow and controllability of an execution process, allowing the intelligent agent to complete tasks more efficiently and accurately. Furthermore, the present disclosure can support the customization of different task flows for different scenarios, thereby meeting usage requirements in the different scenarios and improving user experience.

Description

Information Processing Method, Electronic Device, and Storage Medium This disclosure claims the priority of a Chinese patent application filed with the Chinese Patent Office on January 12, 2024, with an application number of 202410044265.5 and an application title of "Information Processing Method, Electronic Device, and Storage Medium". The entire content is incorporated herein by reference. Technical Field This disclosure relates to artificial intelligence technology, and particularly to an information processing method, an electronic device, and a storage medium. Background Art The technology of large model-driven agents is one of the important research contents in the fields of large models and artificial intelligence, which helps to achieve the wide application of large models in various industries. An agent can complete tasks through autonomous planning. The processing of tasks can be simply described as: perception, planning, and action. Among them, perception means that the agent obtains information from the environment, planning means the decision-making process made by the agent to complete the task. For example, the task can be split to autonomously complete the process planning, and action means the actions taken based on the environment and planning. To improve the processing effect, a large model can be used as the core control module of the agent, and the task can be planned based on the capabilities of the large model, and the agent can be controlled to complete the task processing. However, currently, large model-driven agents are difficult to support complex tasks, are prone to uncontrollable problems, etc. The efficiency and accuracy of task processing are low, and it is difficult to meet the customized requirements for task processes in different scenarios, resulting in poor user experience. Summary of the Invention This disclosure provides an information processing method, an electronic device, and a storage medium to improve the task processing effect of large model-driven agents. In a first aspect, an embodiment of this disclosure provides an information processing method, including: Determine the information to be processed corresponding to the agent; Input the information to be processed and node sequence information into a large model, where the node sequence information is used to indicate multiple nodes included in the task process and the edges between the nodes. A node represents an action instruction for the agent, and an edge represents a jump between nodes, so that the large model determines the target node to be executed according to the information to be processed, and controls the agent to execute the action instruction corresponding to the target node. In a second aspect, an embodiment of this disclosure provides an information processing method, including: Obtain the input information of the user; Input the input information and the node sequence information into a large model, where the node sequence information is used to indicate multiple nodes included in the task process and the edges between the nodes. The nodes represent action instructions for the agent, and the edges represent jumps between the nodes, so that the large model determines the target node to be executed according to the input information, and determines the agent output information according to the action instruction corresponding to the target node; Output the agent output information to the user. In a third aspect, an embodiment of the present disclosure provides an electronic device, including: At least one processor; and A memory communicatively connected to the at least one processor; Wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the electronic device to execute the method described in any of the above aspects. In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium, in which computer-executable instructions are stored, and when the processor executes the computer-executable instructions, the method described in any of the above aspects is implemented. In a fifth aspect, an embodiment of the present disclosure provides a computer program product, including a computer program, and when the computer program is executed by a processor, the method described in any of the above aspects is implemented. The information processing method, electronic device, and storage medium provided by the embodiments of the present disclosure can determine the information to be processed corresponding to the agent, and input the information to be processed and the node sequence information into a large model, where the node sequence information is used to indicate multiple nodes included in the task process and the edges between the nodes. The nodes represent action instructions for the agent, and the edges represent jumps between the nodes, so that the large model determines the target node to be executed according to the information to be processed, and controls the agent to execute the action instruction corresponding to the target node, so that the processing process of the agent can be constructed by using the nodes and edges in the task process, enabling the large model to more accurately understand the task process, driving the agent to execute corresponding action instructions with reference to the task process, realizing the planning of complex processes and the controllability of the execution process, thus completing tasks more efficiently and accurately, and being able to support customizing different task processes for different scenarios, meeting the usage requirements in different scenarios, and improving the user experience. Description of the Drawings The drawings here are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure. FIG. 1 is a schematic diagram of an application scenario provided by an embodiment of the present disclosure; FIG. 2 is a schematic flowchart of an information processing method provided by an embodiment of the present disclosure; FIG. 3 is a flowchart of an ETC activation provided by an embodiment of the present disclosure; FIG. 4 is a flowchart of an ETC consumption record query provided by an embodiment of the present disclosure; FIG. 5 is a schematic diagram of an agent interaction framework provided by an embodiment of the present disclosure; FIG. 6 is a schematic flowchart of an agent training method provided by an embodiment of the present disclosure; FIG. 7 is a schematic diagram of the principle of constructing training data provided by an embodiment of the present disclosure; FIG. 8 is a schematic flowchart of an information processing method applied to an intelligent customer service scenario provided by an embodiment of the present disclosure; FIG. 9 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present disclosure. Through the above-mentioned drawings, specific embodiments of the present disclosure have been shown, and there will be more detailed descriptions hereinafter. These drawings and text descriptions are not intended to limit the scope of the concept of the present disclosure in any way, but to illustrate the concept of the present disclosure to those skilled in the art by referring to specific embodiments. Detailed Embodiments Here, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. It should be noted that the user information (including but not limited to user device information, user attribute information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present disclosure are all information and data that have been authorized by the user or fully authorized by all parties. Moreover, the collection, use, and processing of relevant data need to comply with relevant laws, regulations, and standards, and corresponding operation entrances are provided for users to choose to authorize or refuse. The embodiments of the present disclosure can be implemented through large models, for example, large language models. Among them, a large language model refers to a deep learning language model with a large number of language model parameters, usually including hundreds of millions, tens of billions, hundreds of billions, trillions, or even more than one hundred trillion language model parameters. A large language model can also be called a Foundation Model. Through pre-training of a large language model with a large amount of unlabeled corpus, a pre-trained language model with more than one hundred million parameters is produced. This kind of language model can adapt to a wide range of downstream tasks and has good generalization ability, such as Large Language Model (LLM), multi-modal pre-training model, etc. In actual application, only a small number of samples are needed to fine-tune the pre-trained language model and it can be applied to different tasks. The large language model can be widely used in natural language processing (NLP), computer vision and other fields. Specifically, it can be applied to computer vision tasks such as visual question answering (VQA), image caption (IC), image generation, as well as natural language processing tasks such as text-based sentiment classification, text summary generation, and machine translation. The main application scenarios of the large language model include digital assistants, intelligent robots, search, online education, office software, e-commerce, intelligent design, etc. First, the terms involved in this disclosure are explained: Agents: take actions autonomously to achieve goals and can improve their performance by learning or acquiring knowledge. Directed graph: A graph is a way to represent the relationship between objects and is the basic research object of graph theory. A graph appears to be composed of some small dots (called vertices or nodes) and straight lines or curves connecting these dots (called edges). If a direction is specified for each edge of the graph, the resulting graph is called a directed graph, and its edges are also called directed edges. The application scenarios of the present disclosure are first described below. Intelligent agents driven by large models can imitate the interaction process between humans and the external environment to process tasks. For example, based on the perception of the environment, combined with their own memory and knowledge of the world, they make plans, decisions and actions. After the actions act on the environment, the environment will give the intelligent agent new feedback. Combined with the observation of the feedback, they will take action again, and this cycle will be repeated, and finally the task can be processed through interaction with the environment. The above-mentioned large model-driven intelligent agent solution can be applied to various scenarios in various fields. Taking the intelligent customer service scenario as an example, traditional intelligent customer service often solves process-related tasks by defining complex state machines. Under this architecture, when interacting with users, intelligent customer service can only output fixed words to users according to the predefined state machine. The interaction is stiff and it is difficult to handle problems outside the state machine. When applying intelligent agents to intelligent customer service scenarios, the intelligent agents can autonomously split and plan tasks under the drive of the large model, and continuously make decisions and actions to process tasks through the perception of user input information. However, current intelligent agents still have difficulty solving complex process tasks. The main bottlenecks are: (1) Complex process construction: More complex tasks involve complex process knowledge, and these processes have strict sequential and branching logics. Currently, agents build processes autonomously through large models. However, in cases where the branches are relatively complex, it is difficult for large models to plan the entire process. Therefore, in complex situations, agents cannot truly understand the task process and complete the construction of complex process knowledge, resulting in poor task processing effects. (2) Controllable planning: Current agents autonomously plan the entire execution process based on large models, and humans cannot plan the execution process based on scenario knowledge. As a result, it is very difficult for the entire execution path to follow the custom scenario knowledge, and it is also impossible to intervene after an error, making it difficult to achieve controllable planning. In view of this, embodiments of the present disclosure provide an information processing method for an agent, which supports a customer to configure a flowchart, constructs the processing process of the agent according to the nodes and edges in the flowchart, and during the interaction between the agent and the environment, can determine the node where the current agent is located in the flowchart based on the large model, and control the agent to execute the corresponding action instruction according to the node where it is located, so as to achieve the planning of complex processes and the controllability of the execution process. FIG. 1 is a schematic diagram of an application scenario provided by an embodiment of the present disclosure. As shown in FIG. 1, the solution provided by the embodiment of the present disclosure may involve the following stages: 1. Offline training stage: Multiple industry-corresponding training data can be used to train an agent driven by a large model. Among them, the training data may include flowcharts of various industries, enabling the large model to have the ability to drive the agent to process tasks using flowcharts. 2. Customer configuration stage, which can also be called the process customization stage. In this stage, the customer can combine their own industry and give a flowchart corresponding to a specific scenario. For example, for a certain highway ETC (Electronic Toll Collection) enterprise, a flowchart corresponding to ETC-related tasks can be configured, and a corresponding API (Application Programming Interface) system can be configured. The API system may include the APIs that may be used in the flowchart, so that the agent can better implement task processing with the help of the API system. Specifically, the flowchart configured by the customer can be a directed graph including nodes and edges. In the directed graph, the nodes represent the action instructions corresponding to the agent, which are used to guide the actions of the agent, and the edges represent the jumps between the nodes. After obtaining the directed graph, it can be converted into a structured graph instruction, that is, node sequence information, and the node sequence information can be presented in the form of text, which is convenient for input into the large model for processing. 3. During the online usage stage, the agent can act as an intelligent customer service, handling relevant tasks through interactions with users. For example, it can guide users to handle ECT activation, etc. During the process of handling tasks, the node sequence information obtained from the flowchart will guide the large model, enabling the large model to determine the currently pending node based on the current information, so that the agent can execute the actions corresponding to the current node. For example, when the node contains an action instruction to call an API, the agent can call the relevant APIs in the API system. Among them, the "customer" in the embodiments of the present disclosure refers to the role that configures the flowchart corresponding to the agent, which can be an enterprise customer, such as an ETC enterprise, or other types of customers. The "user" refers to the role that interacts with the agent in actual applications. For example, when the agent acts as an intelligent customer service in the ECT field, the user can be a consumer with ECT handling needs. Through the above solution, the graph structure can be used to solve the problem of constructing complex task processes. With the help of the nodes and edges in the flowchart, the large model drives the agent to handle tasks, enabling the agent to execute step by step according to the graph instructions defined by the scenario, achieving controllable paths. Compared with the traditional intelligent customer service that outputs fixed reply words and phrases by means of a state machine, the present disclosure is more flexible. Compared with the method of the agent autonomously planning the process to implement the intelligent customer service function, the present disclosure guides the large model through the flowchart, enabling the large model to more easily understand the task process, driving the agent to act with reference to the flowchart, making the actions of the agent more controllable, so as to complete tasks more efficiently and accurately. Moreover, it supports customers to customize the flowcharts related to scenarios to meet the usage requirements in different scenarios, and can also adjust the flowcharts at any time according to the performance of the agent to improve the task processing effect and enhance the user experience. Since the solution of combining the large model and the agent is still in its infancy in various fields, there are almost no complex task processes and process-based training data. Therefore, the present disclosure will specifically describe the processing process based on the task process and the construction process of the training data after the combination of the large model and the agent. Next, some embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. Without conflict between the embodiments, the following embodiments and the features in the embodiments can be combined with each other. In addition, the step timings in the following method embodiments are only examples and are not strictly limited. FIG. 2 is a schematic flowchart of an information processing method provided by an embodiment of the present disclosure. The method in this embodiment can be implemented on any device with information processing capabilities, such as implemented in the cloud, locally deployed, implemented on a client, implemented on an IOT (Internet of Things) device, etc. As shown in FIG. 2, the method may include: Step 201: Determine the information to be processed corresponding to the agent. Among them, the agent can be an agent driven by a large model, or can be called an agent controlled by a large model. Here, the meaning of "driven" or "controlled" is that the large model can be the core module for making decisions on the actions of the agent. The agent can actively obtain the information output by the large model and execute corresponding actions. In the solution shown in FIG. 1, the agent can include a large model. In other alternative implementation manners, the large model and the agent can also be two separate parts. The embodiments of the present disclosure do not limit the relationship between the two, as long as the large model can drive the agent. When the agent includes a large model, a processing device can also be provided in the agent to execute the method provided by the embodiments of the present disclosure. The processing device can be implemented in the form of software, hardware, or a combination of software and hardware. When the agent does not include a large model, the method provided by the embodiments of the present disclosure can be executed by the agent. Of course, it can also be executed by other external devices, as long as the actions of the agent can be realized. The agent can be used to interact with the environment, and the information to be processed can include the information obtained by the agent from the environment. Among them, the environment can be any thing outside the agent. For example, the environment can include at least one of the following: user, API system, document system, other agents, etc. Correspondingly, the information obtained from the environment can include at least one of the following: user input information, information returned by the API system, information returned by the document system, information sent by other agents (including information generated or forwarded by other agents, etc.). Exemplarily, in the field of intelligent customer service, the information to be processed corresponding to the agent can include user input information. Step 202: Input the information to be processed and the node sequence information into the large model. Among them, the node sequence information is used to indicate multiple nodes included in the task process and the edges between the nodes. The node represents an action instruction for the agent, and the edge represents the jump between nodes, so that the large model determines the target node to be executed according to the information to be processed and controls the agent to execute the action instruction corresponding to the target node. Among them, an action can also be called a behavior or an operation, which is one of the important capabilities of the agent. In the embodiments of the present disclosure, the nodes in the node sequence information can be used to indicate the actions of the agent, and the edges can be used to indicate the jumps between multiple actions of the agent, so that the large model can drive the actions of the agent according to the node sequence information. Optionally, in the node sequence information, the action instructions included in each node can be used to instruct the agent to perform one or more actions. Each node can correspond to one or more edges. Each edge is used to connect two nodes, and there is at most one edge between any two nodes. The edge can be directional and is used to represent the jump between nodes. For example, there is an edge between node A and node B, which can indicate that after node A is executed, it enters node B. In addition, some edges can also be used to represent jump conditions. For example, the edge between node A and node B has a certain jump condition, and node B will be entered only when the condition is met. Optionally, the node sequence information can be presented in various forms, such as text form or image form. When the large model has the ability to process images, the flowchart can be directly input into the large model as the node sequence information. When the large model does not have the ability to process images or has high performance requirements, the flowchart can be converted into text-form node sequence information and input into the large model. After inputting the information to be processed and the node sequence information into the large model, the large model can determine the specific process of the task according to the node sequence information, and then combine the information to be processed to determine the current position, that is, the current node, so as to control the agent to execute the action instructions corresponding to the node. Optionally, the agent can achieve the processing of the task through multiple rounds of interaction with the environment. The interaction information for each round can include the agent output information, that is, the information output by the agent to the environment, and can also include the agent input information, that is, the information obtained by the agent from the environment. Among them, one round of interaction can include one output and one input of the agent. The object of interaction is the environment, and some information can be output to the environment and some information can be obtained from the environment to complete one round of interaction. As mentioned above, the environment can include users, API systems, document systems, other agents, etc. The specific interaction object for each round can be one of them, and the specific interaction objects for each round of interaction can be the same or different. For example, in the first round, the interaction between the agent and the environment is specifically the interaction with the user. The agent can output inquiry information to the user and obtain the requirements input by the user. In the second round, the interaction between the agent and the environment is specifically the interaction with the API system. The agent can output the name and input parameter values of the API to be called to the API system and obtain the output parameter values returned by the API system. In the third round, the agent can continue to interact with the user or the API system or other objects, and complete the processing of the task through multiple rounds of interaction. The above steps 201 to 202 can be steps executed separately in each round. In any round, the information to be processed includes the interaction information of the previous round, so that the large model determines the target node corresponding to the current round according to the interaction information of the previous round, the node sequence information, and the node corresponding to the previous round. Among them, there are multiple implementation manners for how the large model determines the node corresponding to the previous round. In one implementation manner, according to the interaction information of the previous round, it can be determined which node in the node sequence information has been executed currently, and this node is used as the node corresponding to the previous round, and the target node corresponding to the current round is determined according to the node corresponding to the previous round. Exemplarily, the nodes in the node sequence information successively include: Node 1 is used to ask the user which order to query, Node 2 is used to determine the specific item to be queried according to the order (such as querying how to use the commodity or querying the express information), Node 3 is used to call the corresponding API according to the specific item to be queried, and other nodes can also be included after Node 3. The large model has certain natural language understanding and analysis capabilities, and can determine which position in the node sequence information is currently executed according to the existing interaction situation. For example, if it is determined according to the interaction information of the previous round that the order to be queried and the specific item have been asked and determined currently, but the corresponding API has not been called yet, it can be determined that the node corresponding to the previous round is Node 2, and then Node 3 can be executed. In another implementation manner, the node corresponding to the previous round can be directly prompted to the large model. Optionally, the node corresponding to each round can be determined and output by the large model itself, and the information input to the large model in each round can include the node determined in the previous round, so that the large model determines the node corresponding to the current round according to the node corresponding to the previous round. Exemplarily, the information input to the large model can be: "The node sequence information is..., the node corresponding to the previous round is..., the interaction information of the previous round is..., please determine the node to be executed currently according to these information, and determine the information to be output to the environment (such as the information sent to the user or the API system)". Taking the intelligent customer service scenario as an example, the interaction with the environment includes at least one of the following: interacting with the user, calling the API, and querying the document. Among them, the API system includes multiple APIs, which are used to return the corresponding output parameter values according to the values of the input parameters of the API to be called; the document system includes multiple documents, which are used to output the corresponding documents according to the information to be queried. In each round, the intelligent agent can interact with any one of the user, the API system, and the document system. After multiple rounds of interaction, the task processing is completed. Specifically, multiple rounds of interaction can be achieved through multiple nodes in the node sequence information. The embodiments of the present disclosure do not limit the functions of each node. For any node in the node sequence information, the node can correspond to one round of interaction, or at least two rounds of interaction, or it may not involve interaction and only indicate its own calculation process, such as performing certain calculations or judgments based on existing information. Moreover, the number of interactions corresponding to the node is not limited either, and the large model can perform intelligent analysis based on the current interaction situation with the environment to determine how many interactions are required to execute the node. In an alternative approach, when a certain node instructs an interaction with the environment, the large model can drive the agent to complete the execution of the node through one or multiple interaction processes. Optionally, the large model determines the target node corresponding to the current round, which may include: based on the interaction information of the previous round, determining whether the node corresponding to the previous round has been executed. If it has been executed, the target node corresponding to the current round is the next node of the node corresponding to the previous round; if it has not been executed, the target node corresponding to the current round remains the node of the previous round. Exemplarily, if a certain node instructs to call an API, and the parameter values for calling the API need to be queried from the user to be determined, the large model can first control the agent to interact with the user, then determine whether all parameter values have been obtained, and then achieve the call of the API through interaction with the API system. In this case, in each round, the large model can determine whether the node of the previous round has been executed. For example, whether the action of calling the API instructed by the node has been completed. If not, the target node corresponding to the current round is still the node of the previous round until the action instruction instructed by the node of the previous round is completed, and then it enters the next node. Optionally, controlling the agent to execute the action instruction corresponding to the target node includes: the large model determines whether the condition for executing the action instruction corresponding to the target node is currently met. If it is met, the current round is used to execute the action instruction; if it is not met, the current round is used to perform the interaction for preparing for the action instruction. Exemplarily, after determining the current target node, the large model can determine whether the condition for executing the action instruction corresponding to the target node is currently met based on the interaction of the previous round. For example, whether the condition for API call is met. If it is met, it directly controls the agent to interact with the API system to complete the API call. If it is not met, it determines the preparatory work required for the API call. For example, if the parameter values need to be determined first, it can first determine the parameter values through interaction with the user. Optionally, when it is determined according to the action instruction corresponding to the target node that the interaction in the current round is to interact with the user, the information output by the agent includes the reply information to the user. Correspondingly, the information input by the agent includes the input information of the user. When it is determined according to the action instruction corresponding to the target node that the interaction in the current round is to call an API, the information output by the agent includes the name of the API to be called and the value of the input parameter, so that the agent can send an API call request to the API system according to the name of the API and the value of the input parameter. Correspondingly, the information input by the agent includes the value of the output parameter of the API returned by the API system. When it is determined according to the action instruction corresponding to the target node that the interaction in the current round is to query a document, the information output by the agent includes the information to be queried, so that the agent can send a query request to the document system according to the information to be queried. Correspondingly, the information input by the agent includes the queried document returned by the document system. Through the above solution, the agent can complete the task processing through the interaction with the user, the API system, and the document system, so as to better realize the service for the user with the help of the capabilities of the API system and the document system, improve the efficiency of task processing, and enhance the user experience. In another optional way, it can be set that each node corresponds to one interaction, that is, when the agent jumps to each node, it only needs to perform one interaction to complete the node and enter the next node. Correspondingly, the large model determines the target node corresponding to the current round, which may include: determining that the target node corresponding to the current round is the next node of the node corresponding to the previous round, so as to control the agent to execute the action instruction corresponding to the target node. In this way, the large model can more simply and accurately judge the current target node, so that the processing logic of the agent is more efficient and more controllable. Next, the processing process of the embodiments of the present disclosure will be described in detail in combination with the intelligent customer service process of a certain highway ETC. FIG. 3 is a flowchart of ETC activation provided by an embodiment of the present disclosure. FIG. 4 is a flowchart of ETC consumption record query provided by an embodiment of the present disclosure. As shown in FIGS. 3 and 4, both flowcharts include nodes and edges, which have different meanings. Among them, the nodes include the action instructions for the agent in each step, and can represent the action set of the agent to a certain extent. The edges represent the jumps between nodes, which also include various intentions and expressions of the user, judgment basis conditions, etc. It should be noted that, for the convenience of description, the ID (identifier) corresponding to each node is displayed in the text box, and it is not actually required to be displayed in practical applications. In the above flowchart, relevant information of documents and APIs can also be integrated into nodes, and interactions with the document system and API system can be achieved through the action instructions of the nodes. For example, the instructions in nodes B, F, and J in the figure include calls to APIs, indicating that the corresponding API needs to be called when reaching these nodes. The activated knowledge in node G is actually a document, and this node instructs to guide the user to activate according to the document knowledge. It can be seen that through the node design of the flowchart, documents, APIs, and process knowledge can be organically combined. After obtaining the flowchart, the flowchart can be processed to convert it into corresponding node sequence information. Specifically, referring to Figure 3, the node sequence information corresponding to the ETC activation flowchart can be: “flowchart TD A[Start]--ETC activation-->B[According to the system mobile phone number, call the API to obtain license plate information and get the user's license plate information] B-->C[Traverse the license plate numbers in the user's license plate information and confirm with the user the license plate number to be activated] C-->D{Whether the license plate number to be activated is obtained} D--Yes-->F[According to the license plate number to be activated, call the API to obtain vehicle type information and query the vehicle type and corresponding activation knowledge] F-->G[Guide the activation according to the activation knowledge] D--No-->E[Ask about the license plate number to be activated] E-->F G--Activation successful-->H[Congratulations on successful activation, express gratitude and goodbye]” Referring to Figure 4, the node sequence information corresponding to the ETC consumption record query flowchart can be: “flowchart TD I([Start])--ETC consumption record query-->J[According to the definition of the API for obtaining ETC consumption records, ask for parameters and call] J--Query return content normal-->K[Inform the user of the consumption records returned by the query] J--Query return content empty-->L[Inform that the query is abnormal and ask the user to confirm the license plate number and date] K&L-->M[Ask if there is anything else needed] M--No-->N[Express gratitude and goodbye]” Among them, flowchart TD can indicate that the direction of nodes in the flowchart is from top to bottom. The content in "[]" and "{}" is used to represent the action instructions corresponding to the nodes. The former is mainly used to represent the content that needs to be interacted with, such as the response to the user or the information of calling the API. The latter is mainly used to represent the process executed by the agent itself, such as judging certain parameters. "--" is used to represent the edge, and the jump situation of the entire flowchart can be described by using the edge. The content between "--" and "--" is used to represent the jump condition, and various jump conditions can be described in descriptive natural language, such as user intent, judgment branches based on variables or content, etc. The above flowchart describes the relevant task processes in the ETC scenario. For better display, start nodes, judgment nodes, etc. are also set in the above flowchart. At the start node, the agent can output fixed inquiry words to the user. At the judgment node, the agent can execute the corresponding judgment process according to the current situation. In actual applications, the flowchart can be adjusted according to the scenario requirements. For example, node D can be cancelled, and 2 edges can be directly set after node C, connecting to node E and node F respectively, and corresponding jump conditions (yes or no) are given based on the edges. Another example is that two flowcharts can be merged into one flowchart. Specifically, the two start nodes can be merged into one, and enter the ETC activation process and the ETC consumption record query process through different edges respectively. Combined with the above flowchart, the large model can drive the agent to assist the user to complete the ETC handling according to the scenario-related process. In order to better guide the large model, in the embodiments of the present disclosure, the agent can be assisted to implement task processing through prompt information, where the prompt information can include input prompt information and / or output prompt information. Among them, the input prompt information can include at least one of the following: agent portrait information, node sequence information, jump condition information, API information, task information, environment information. The following will explain each input prompt information separately from several aspects. In the first optional implementation manner, the input prompt information can include agent portrait information. Among them, the agent portrait information can be used to define the role and function description of the agent, etc. The function description can include the description of each flowchart. Exemplarily, the agent portrait information can be specifically: "You are the intelligent assistant of the highway ETC. Now you need to help the user activate the ETC device and query the ETC consumption record by answering the user's call. Please communicate with the user according to the following process." In a second optional implementation manner, the input prompt information may include node sequence information. Among them, the node sequence information may refer to the node sequence information corresponding to FIGS. 3 and 4 above. In a third optional implementation manner, the input prompt information may include jump condition information, and the jump condition information may be used to explain or describe the jump conditions involved in the edges. The jump conditions that may be involved may include, but are not limited to: the intention of the user, slots, and return values after calling the API, etc. Optionally, among multiple edges involved in the node sequence information, each edge may be used to indicate a jump between two nodes. In addition, at least some of the edges may also be used to represent jump conditions, so that the large model determines whether the current jump condition corresponding to the target edge is satisfied according to the information to be processed, where the target edge is an edge with the node in the previous round as the starting node, and when the jump condition corresponding to the target edge is satisfied, the target node is determined to be the ending node of the target edge. Exemplarily, there is an edge between node L and node M, indicating that after node L is executed, node M is executed, which is the function of the edge to indicate the jump between nodes. In addition, the edge may also correspond to a jump condition. For example, the jump condition between node I and node J is "ETC consumption record query", indicating that when the intention of the user is to query the ETC consumption record, node J will be entered. In any round of interaction, if the large model determines that the node in the previous round is node I, and the jump condition corresponding to the target edge with node I as the starting node is "ETC consumption record query", if the current jump condition is satisfied, the current target node can be determined to be the ending node of the target edge, that is, node J. The jump conditions can be set by the customer. The judgment of some jump conditions is relatively simple. For example, it only needs to judge yes or no. The judgment of some jump conditions may be more difficult. For example, when the customer defines the jump condition as certain intentions or slots involved in a specific scenario, the large model may have difficulty accurately judging the intentions and slots in this scenario. At this time, the description information of the intentions and slots can be added to the input prompt information to facilitate the large model to accurately judge whether the jump condition is satisfied. Optionally, the interaction between the agent and the environment includes the interaction with the user, and the agent input information includes the user's input information. When the jump condition of any node in the node sequence information includes an intention and / or a slot, the method further includes: inputting the jump condition information into the large model, where the jump condition information includes the description information corresponding to the intention and / or slot included in the node sequence information, so that when the large model determines that the user's input information matches the description information of the intention and / or slot corresponding to the target edge, it is determined that the jump condition corresponding to the target edge is satisfied. Among them, the intention can refer to the user's needs. For example, if the user's input information is "What's the weather like in Area A tomorrow", the corresponding intention can be determined as "weather query". The slot can be at least one parameter corresponding to the intention. For example, the slots corresponding to the intention of "weather query" can include "location", "time", etc. According to the values of the intention and the slots, the information output to the user can be determined. For example: "It will be cloudy turning to light rain in Area A tomorrow." Among them, the description information can be any information used to explain the intention and / or slot. For example, it can be common phrases corresponding to the intention or slot. Exemplarily, the jump condition between node A and node B is "activate ETC". To improve the understanding of this intention by the large model, the jump condition information can be used to explain it. For example, the jump condition information can be: "In the process ---- the middle represents the jump condition, and the user intentions and their example phrases included are: ETC activation: ['I want to activate ETC', 'How to use the ETC device after purchase']". According to the above jump condition information, the large model can determine the specific meaning of the intention corresponding to the jump condition between node A and node B. When it is determined that the user's input information matches the description information of the intention corresponding to the target edge, it is determined that the jump condition corresponding to the target edge is satisfied. Here, "matching" can mean the same or similar semantics. For example, when the user's input information is an expression similar to "I want to activate ETC" or "How to use the ETC device after purchase", it indicates that the user's intention is "ETC activation". At this time, it can jump from node A to node B and execute the process corresponding to node B. Similarly, the jump condition can also include the judgment of slots. Correspondingly, the jump condition information can include the description information of the slots. The specific implementation principle is similar to that of the intention and will not be elaborated here. Through the above solution, the large model can be used to judge the jump conditions between nodes. When it is determined that the current jump condition is satisfied, it enters the target node corresponding to the jump condition, thereby supporting the customer to restrict the jump conditions between nodes by setting the edges, and further restricting the jump between the actions of the intelligent agent to meet the requirements of the task. And when the edge includes the recognition of the intention and / or slot, the corresponding description information of the intention and / or slot can be input into the large model, so that the large model can accurately judge the current node to be entered according to the relevant information, reducing the situation that the large model cannot accurately execute the process due to insufficient understanding of the knowledge of specific scenarios, and improving the accuracy of the intelligent agent in processing tasks. In the fourth alternative implementation, the input prompt information can include API information. Corresponding API information can be given for the APIs involved in the node sequence information. The API information may include the name of the API, description information, at least one input parameter, and at least one output parameter. When the action instruction corresponding to the target node in the large model includes calling the API, it is determined whether the values of the at least one input parameter are currently obtained. When the values of at least some of the input parameters are not obtained, based on the description information, an interaction is performed with the user to determine the values of the at least some input parameters, and the agent output information corresponding to calling the API is determined to include: the name of the API and the values of the at least one input parameter of the API, so as to complete the call of the API through the agent output information and obtain the values of the at least one output parameter. Exemplarily, the API information may specifically be: "The API information involved in the process includes: API for obtaining ETC consumption records: Name: getETCRecord; Description information: Query ETC consumption records according to the date, and only one day's consumption records can be queried each time; Input parameters: {"plate_no (license plate number)": {"description": "License plate number, a 7-digit string, where the first digit is the provincial abbreviation, and the following are letters and numbers", "type": "string"}, "date (date)": {"description": "Date: year + month + day, such as May 6, 2022, July 1, 2021. Note that the date must completely include year, month, and day, and the user needs to be confirmed about the date", "type": "string"}}; Output parameters: {"record (record)": {"description": "Consumption record description text", "type": "string"}}." Referring to the above example, the input parameters corresponding to the API for obtaining ETC consumption records include the license plate number and the date. If the license plate number has been determined according to the context information or other information, but the date to be queried has not been determined yet, the date can be determined through interaction with the user. When interacting with the user, the description information can be referred to. For example, the user can be prompted to enter the date to be queried, and only one day's consumption records can be queried each time. After obtaining the date input by the user, the name of the API and the values corresponding to each input parameter can be used as the agent output information, so that the API system can return the values corresponding to the output parameter according to the name of the API and the values corresponding to each input parameter. Through the above solution, the information of each API involved in the node sequence information can be used as input prompt information and input into the large model, enabling the large model to drive the intelligent agent to interact with the API system based on the API information, complete the correct call of the API, and thus efficiently implement the processing of tasks with the help of the API system, further improving the task processing effect. In addition, for APIs that are not involved in the node sequence information but may be used by the large model through retrieval or other means in actual applications, the corresponding API information can also be provided. In this way, even if certain APIs do not directly appear in the action instructions of the nodes, but the information of these APIs is included in the input prompt information, the large model can also choose to use these APIs to implement the task processing based on the current situation. In the fifth optional implementation manner, the input prompt information may include task information, which can be used to describe the task to be processed by the intelligent agent and may also include precautions during the task processing, etc., so as to effectively guide the large model to drive the intelligent agent to process the task. Exemplarily, the task information may specifically be: "Please assist the user to complete the relevant tasks of ETC through communication with the user. Note: 1. When answering the user's questions, focus on handling issues related to the process and API to complete the guidance for the user. 2. When the user provides parameter information, ensure confirmation in the specified format. 3. Avoid repeating the user's questions, especially matters that have been confirmed." In the sixth optional implementation manner, the input prompt information may include environmental information. Optionally, the environmental information can be input into the large model so that when the action instruction corresponding to the target node includes calling an API, the large model determines the values of at least some input parameters of the API to be called based on the information to be processed and the environmental information, and when there are specific input parameters, determines the values of the specific input parameters based on the interaction with the user; where the specific input parameter is an input parameter whose value cannot be determined through the information to be processed and the environmental information. Among them, the environmental information may include information that can be determined without interacting with the user, such as the current date, time, etc. Exemplarily, the environmental information may specifically be: "Please remember that the current time is: 17:00:00 on October 21, 2023. The caller's mobile phone number is 13XXXX." Referring to the above example, assume that the action instruction corresponding to the target node includes calling an API to obtain the user's ETC consumption record, and the input parameters corresponding to the API include the date to be queried. If, according to the information to be processed, such as the interaction information with the user in the previous round, it is determined that the user wants to query the consumption record of yesterday, then the date to be queried can be determined as October 20, 2023 by combining the environmental information. If there are input parameters whose values cannot be determined based on the interaction information and environmental information of the previous round, the user can be further queried to determine the values of relevant parameters through interaction with the user. Through the above solution, the environmental information can be input as input prompt information into the large model, so that the large model can determine the values of at least some of the input parameters of the API corresponding to the target node according to the environmental information, reducing the number of interactions with the user and improving the user experience. In practical applications, the input prompt information can include one or more of the six implementation methods mentioned above. Exemplarily, the input prompt information can include the following information arranged in sequence: agent portrait information, node sequence information corresponding to the ETC activation process, jump condition information corresponding to the ETC activation process, API information corresponding to the ETC activation process, node sequence information corresponding to the ETC consumption record query process, jump condition information corresponding to the ETC consumption record query process, API information corresponding to the ETC consumption record query process, task information, environmental information. Optionally, these information can be input into the large model simultaneously. For example, these information can be input into the large model as a whole paragraph of text, or these information can also be input separately. These information can be used as the input information of the large model to guide the large model and help the large model better complete the task. Optionally, output prompt information can also be input into the large model; wherein, the output prompt information is used to prompt that the output information of the large model includes: the node corresponding to the previous round, the target node of the current round, and the agent output information determined according to the target node. The output prompt information will be described in detail below. Among them, the output prompt information can also be called the output thought chain, which is used to prompt the large model what content to output. By using the output prompt information, the output information of the large model can be restricted to enable the large model to find out which position it is in the flowchart and determine the corresponding action instruction. Optionally, the output information of the large model can include a thinking part and agent output information. The thinking part can include: the node corresponding to the previous round, the jump condition (if any), and the target node of the current round. Among them, the jump condition can also be omitted. Optionally, the thinking part can further include: if the action instruction corresponding to the target node includes calling an API, then consider whether the values of all input parameters of the API have been obtained and give a conclusion on whether the API can be called. Optionally, the agent output information can be distinguished according to the type of interaction. For example, when interacting with the user, the agent output information is the response information to the user, and when calling an API, the agent output information is the name of the API and the values of the input parameters. Optionally, the method further includes: determining the agent output information in the output information generated by the large model; outputting the agent output information through the agent. Exemplarily, the output prompt information input to the large model can specifically be: "Please answer in the following format: thought (thinking part): [previous node] + [jump condition] + [current jumped-in node] +...; response (agent output information): [answer content], please note to fill in according to the format." After inputting the above output prompt information into the large model, the large model can understand what content it needs to output and output according to the corresponding logic. In one example, when an API needs to be called currently, the output information of the large model can specifically be: "thought: previous node = A, current intention = ETC activation, currently should jump into node B, from the above text, it can be learned that the parameters and their values of getuserinfo (the API for obtaining license plate information), the mobile phone number is 13XXXX, all required parameters of getuserinfo are known, no need to ask again, can call getuserinfo. action (action to be executed, specifically the name of the API): getuserinfo; actioninput (input corresponding to the action, specifically the values of the API input parameters): mobile phone number 13XXXX." After the large model determines the name of the API and the values of the input parameters, the agent can, according to the name and values, complete the call of the relevant API through the API system, obtain the values of the corresponding output parameters, and feedback them to the large model, so that the large model can perform the next round of processing according to the values of the output parameters. In another example, when an interaction with the user is needed currently, the output information of the large model can specifically be: "thought: previous node = B, currently should jump into node C, traverse and confirm the license plate number. response: Okay, two license plates are queried according to your incoming call mobile phone number, namely Beijing AXXXX and Beijing BXXXX, which one do you activate?" The agent can output the content of the response to the user. If the user has subsequent input, the user's input information will be fed back to the large model, enabling the large model to perform the next round of processing based on the user's input information. In practical applications, in each round, the input prompt information, output prompt information, and information to be processed can be input into the large model, where the information to be processed includes the interaction information from previous rounds. For the sake of easy understanding, the following are reference examples of several rounds of interaction information: “<start(Start)>assistant(Assistant role, used to represent the agent): Hello, this is the intelligent assistant of the highway ETC center. How can I help you?<end(End)> <start>user: Hello, I just bought your ETC and still can't use it after activating it according to the instruction manual. <end> <start>assistant: thought: The previous node = A, the current intention = ETC activation, and the current node to jump to should be B. From the above context, the parameters of getuserinfo and their values can be obtained. The mobile phone number is 13XXXX. All required parameters of getuserinfo are known and there is no need to ask again. getuserinfo can be called. action: getuserinfo; actioninput: Mobile phone number 13XXXX. <end> <start>function (Functional role, used to represent the API system): Beijing AXXXX, Beijing BXXXX. <end> <start>assistant: thought: The previous node = B, currently should jump to node C, traverse and confirm the license plate number. response: Okay, two license plates are found according to your calling mobile phone number, namely Beijing AXXXX and Beijing BXXXX. Which one did you activate? <end>” Among them, <start>and <end>Used to represent the start and end positions of the information currently output by a certain role. Specifically, in the first round, the input prompt information and the output prompt information can be input into the large model. The input prompt information and the output prompt information can refer to the aforementioned examples. The input prompt information contains the relevant information of the flowchart, enabling the large model to plan according to the guidance of the flowchart. The output prompt information contains the output thought chain, enabling the large model to control the agent to execute specific action instructions by thinking about the jumps between nodes. Among them, the first round corresponds to the start node, that is, node A, which is used to greet the user. Optionally, the corresponding greeting information can be pre-configured for the start node as the output information of the agent. For example: "Hello, this is the intelligent assistant of the high-speed ETC center. How can I help you?" After the agent outputs the corresponding greeting information, the user will input a query (question), for example, "Hello, I just bought your ETC and it still doesn't work after activating it according to the instructions", as the input information of the agent. In the second round, the input prompt information, the output prompt information, and the information to be processed can be input into the large model. The information to be processed at this time includes the interaction information of the previous round, that is, the output information and the input information of the agent in the first round. This enables the large model to think based on this information, determine the node to be executed currently, and make a decision on the next action, such as calling an API. At this time, the large model will output the Action and Action Input corresponding to the API. Once it is detected that the output information of the large model includes the Action and Action Input corresponding to the API, the agent can interact with the API system through the API information to obtain the value of the output parameter. In the third round, the input prompt information, the output prompt information, and the information to be processed can be input into the large model. The information to be processed at this time includes the interaction information of the previous two rounds. The large model can drive the agent to interact with the user based on this information. And so on, until the task of ETC activation is completed. Through the above solution, the large model can determine the target node corresponding to the current round based on the interaction information of the previous rounds and the relevant information of the flowchart, and combine the node corresponding to the previous round, so that the agent can complete actions such as interacting with the user and calling the API according to the action instructions of the target node. As a result, the large model can better plan the actions of the agent according to the flowchart, improve the controllability of the agent's task execution, and better achieve the effect of controllable planning. In addition, the output prompt information can be input into the large model, and the output information of the large model is prompted to include the node corresponding to the previous round, the target node of the current round, the agent output information determined according to the target node, etc., so as to effectively guide the large model, utilize the capabilities of the large model to reasonably plan the current actions, and further improve the controllability of the agent. In practical applications, the input prompt information, the output prompt information, and the information to be processed can be input into the large model simultaneously. For example, the input prompt information and the output prompt information can be used as header information and concatenated with the current information to be processed before being input into the large model. Alternatively, they can also be input into the large model separately. In addition, the prompt information can also include other content. For example, the various methods provided by the embodiments of the present disclosure for the large model to determine the target node, the method for determining the interaction content of the current round, etc. can also be input into the large model as prompt information. The above gives an example of the agent achieving task processing through multi-round interactions with the user and the API system. In other alternative implementation manners, the agent can also interact only with the user or only with the API system. The flowchart can involve multi-round interactions or one-round interactions. The position of the one-round interaction in the flowchart is not limited. For example, in the previous one or several nodes, certain action instructions can be executed first using the environmental information, an interaction with the user or the API system can be performed at a certain intermediate node, and in the last one or several nodes, relevant action instructions can be continued to be executed according to the results of the interaction and other information until the task processing is completed. In summary, the information processing method provided in this embodiment can determine the information to be processed corresponding to the intelligent agent, and input the information to be processed and the node sequence information into the large model, where the node sequence information is used to indicate multiple nodes included in the task process and the edges between the nodes. The node represents an action instruction for the intelligent agent, and the edge represents the jump between the nodes, so that the large model can determine the target node to be executed according to the information to be processed, and control the intelligent agent to execute the action instruction corresponding to the target node. Therefore, the processing process of the intelligent agent can be constructed by using the nodes and edges in the task process, so that the large model can more accurately understand the task process, drive the intelligent agent to execute the corresponding action instruction with reference to the task process, realize the planning of complex processes and the controllability of the execution process, thereby completing tasks more efficiently and accurately. Moreover, it can support customizing different task processes for different scenarios, meet the usage requirements in different scenarios, and improve the user experience. The following will elaborate on the specific content corresponding to the action instruction of the node and the construction process of the node sequence information. As mentioned above, the action instruction corresponding to the node can include interacting with the user, calling the API, querying the document, judging the parameters, etc. Correspondingly, the multiple nodes in the node sequence information can include: reply node, call node, query node, judgment node, etc. In an alternative implementation, the multiple nodes include a reply node for replying to the user, and the action instruction corresponding to the reply node is any one of the following types: Ordinary reply, which is used to indicate replying according to the content included in the action instruction; Strict reply, which is used to indicate replying according to the content and format given in the action instruction that are prohibited from being modified; Step-by-step reply, which is used to indicate replying to the user through at least two interactions; Flexible reply, which is used to indicate replying according to the knowledge given in the action instruction; Among them, the action instruction contains keywords for indicating the type of reply, so that the large model can determine the type of reply according to the keywords and generate the corresponding reply information. Specifically, the ordinary reply means that when there is no strict key point information requirement for the reply content, the action instruction can be directly edited to inform how to reply. For example: [When telling the user that there is a housing purchase withdrawal record in place A, they cannot receive the subsidy in place B]. In the case of an ordinary reply, the content output by the intelligent agent can be the same as the content given in the action instruction, or can be slightly adjusted. For example, different styles can be configured for different intelligent agents, such as rigorous, lively, etc. The large model can adaptively adjust the content in the action instruction according to the corresponding style and output it. Strict reply means: when the reply content includes important and unmodifiable content, use strict reply. The content and format of the strict reply can be determined by the customer. For example, the format is [Strict reply: reply content title]. Since strict replies are generally longer, the specific reply content can be defined in the reply knowledge column. Step-by-step reply means: when the reply content is long and can be informed to the user step by step, use step-by-step reply. For example, the action instructions corresponding to the step-by-step reply can be [first determine the category of the coupon, and then reply based on the knowledge of store coupons and live broadcast coupons], or [break down into multiple steps based on the knowledge of ETC activation, and guide users to activate ETC step by step]. Flexible reply means: when the reply content can be selectively answered based on certain knowledge, flexible reply can be used. Specifically, corresponding knowledge can be given and the big model can make flexible replies based on the knowledge. Exemplarily, the action instruction can be [Flexible reply based on taxi money knowledge], wherein the taxi money knowledge can be written in the action instruction of the node, or input into the big model in other ways, for example, included in the input prompt information. Exemplarily, the input prompt information may include: {Taxi money knowledge: On the evaluation page of XX application, users will see taxi money red envelopes. After clicking, they can get taxi money of varying amounts, which can be used to directly deduct travel expenses. Taxi money can be used to deduct express consumption. It is available when it reaches 10 yuan, and can be used in combination with taxi coupons. Taxi money is valid for only one week.} The above different types of reply content can be written directly in the action instructions of the node. The big model can determine the corresponding type according to the keywords in the action instructions and reply according to the corresponding type. For example, if the action instructions contain keywords such as strict, flexible, and step-by-step, the reply information can be generated according to the corresponding type. If it does not contain any of the keywords such as strict, flexible, and step-by-step, the reply information can be generated according to the type of ordinary reply. Optionally, customers can directly specify the response type of each node through the flowchart and configure the corresponding action instructions. By setting different types of reply nodes, the large model can generate corresponding reply information according to the type, which can meet the reply needs in different scenarios. In addition, a node can be used to guide one or more interactions of the large model, which helps to use limited flowcharts to achieve more process interactions, so that customers do not have to tediously configure the nodes and jump conditions in the flowchart, thereby improving the customer experience. In another optional implementation, the multiple nodes include a calling node for calling an API, and the action instructions corresponding to the calling node include: The default part is used to indicate the name of the API to be called; wherein the API corresponds to at least one input parameter; A custom part for indicating the format of the values of all or some of the at least one input parameter, so that the large model can obtain the values that conform to the format according to the custom part. Optionally, the call node can be mainly used to describe the call process of the API. Among them, the custom part can be an optional item, that is, the action instruction corresponding to the call node can only include the default part, or can include the default part and the custom part. Exemplarily, the default part can be: [Ask for relevant parameters and call according to the definition of the XXX API]. The custom part is generally used to remind of content such as parameter formats, for example: [Note that the format of the parameter date must be year + month + day]. By providing the default part and the custom part, the call requirements of different APIs can be met. Moreover, both the default part and the custom part can be set by the customer. The customer can remind the large model to convert the parameters into the accurate format through the custom part, improve the accuracy of API calls, and ensure the smooth completion of the task. In another optional implementation manner, the multiple nodes can further include a judgment node, a query node for querying documents, etc. Among them, the judgment node can generally be used to make a judgment based on a certain content or variable to determine the subsequent branch. The judgment node can involve interaction with the user or not, for example, it can be to ask the user a certain information or make a judgment based on a certain variable. Exemplarily, the action instruction of the judgment node can be: {Ask the user whether the office location is place C or place E} or {Judge whether the number of vehicles is greater than 5}. The latter can also be directly represented by pseudocode, such as {if (if) the number of vehicles > 5} followed by True (true) and False (false) to enter different branches. The query node can be used to interact with the document system to implement the ability to query documents. For example, the action instruction corresponding to the query node can be: [Query relevant documents according to the activation method selected by the user]. The customer can complete the configuration of the above various types of nodes by drawing. Optionally, a flowchart input by the customer for indicating the task process can be obtained; according to the flowchart, the node sequence information can be generated. Among them, the meanings of the nodes and edges in the flowchart can refer to the foregoing content. The node sequence information can be in text form, and can specifically be in the form of pseudocode. The node sequence information can present the action instructions involved in the entire task process through the description of the graph structure. To facilitate the conversion into node sequence information in text form, a conversion protocol from flowchart to pseudocode can be designed. For example, in some technologies, the conversion from pseudocode to flowchart is provided. Embodiments of the present disclosure can refer to the corresponding conversion protocol of such technologies to implement the conversion from flowchart to pseudocode, so that the information contained in the flowchart can be converted into instructions in a text sequence. Optionally, generating the node sequence information according to the flowchart includes: determining multiple nodes in the flowchart and assigning corresponding IDs to each node; determining multiple edges in the flowchart, and generating a piece of text for each edge, where the text includes the ID of the starting node of the edge, the ID of the ending node, and the action instruction corresponding to the ending node. If the edge further includes a jump condition, the jump condition is inserted between the ID of the starting node and the ID of the ending node. Among them, the node sequence information includes multiple pieces of text and direction information, and the direction information is used to indicate the execution order of the multiple pieces of text, and the multiple pieces of text are the texts corresponding to the multiple edges. Taking Figure 3 as an example, when obtaining the flowchart input by the customer, the flowchart contains the action instructions corresponding to each node, but does not contain node IDs. All nodes in the flowchart can be traversed, and a corresponding ID can be assigned to each node. The customer is supported to input multiple flowcharts, and the node IDs in each flowchart can be non-repetitive to avoid problems during the task processing as much as possible. After assigning an ID to each node, all edges can be traversed. For each edge, a preset character is used to represent the edge, such as "--", and the ID of the starting node and the ID of the ending node of the edge are connected using this character. If the edge further includes a jump condition, the jump condition can be inserted. In addition, the action instruction corresponding to the ending node can be added, so that when the large model jumps to the ending node corresponding to the edge, the intelligent agent can be controlled to execute the corresponding action instruction. Among them, the flowchart can be a directed graph, the edge can correspond to a direction, and the starting node and the ending node can be the nodes at both ends of the edge. After generating multiple pieces of text according to multiple edges, the multiple pieces of text form the node sequence information. The direction information in the node sequence information can be used to indicate the execution order of the multiple pieces of text. For example, it is executed from bottom to top, or from top to bottom, etc. Optionally, the direction information can also be omitted, and it is default to be executed from top to bottom. Through the above solution, the customer can complete the process construction of complex tasks by drawing a daily flowchart, and through the conversion of the flowchart according to the pre-designed specification, the obtained node sequence information in text form is consistent with the content of the flowchart and can be understood by the large model, so that the task can be executed smoothly. FIG. 5 is a schematic diagram of an agent interaction framework provided by an embodiment of the present disclosure. As shown in FIG. 5, the agent interaction system may mainly involve three roles: user, agent, and system. Among them, the system may be an API system, etc. The customer can configure process knowledge and API knowledge for the agent. Among them, the process knowledge may include a flowchart. The process knowledge and API knowledge may each have a processing module for implementing the construction of the flowchart into graph instructions (sequence information of nodes in text form) and the definition and registration of the API. The API knowledge may be tools that need to be called in the field, including various operations such as common queries and modifications. Each API may include corresponding output parameters and input parameters. The definition and registration of the API can be completed according to the API knowledge. Among them, the parameters for definition and registration may be all the parameters included in the API, or only the parameters that need to interact with the user to obtain. Other parameters can be supplemented based on rules or environmental information, that is, the large model can only process some of the parameters. The graph instruction builder can be used to generate the information input to the large model according to the node sequence information corresponding to the flowchart and the definition and registration information of the API. For example, it can generate a prompt message and splice the user's input message, and give it to the large model for processing. The large model can plan according to the information output by the graph instruction builder and determine the actions of the agent at each step. Optionally, in addition to obtaining the information output by the graph instruction builder, the large model can also directly obtain knowledge from the graph instructions and the definition and registration information of the API. For example, the prompt message output by the graph instruction builder to the large model may only contain partial API information and graph instructions. The large model can analyze the current situation and obtain knowledge from all the constructed graph instructions and API information and process it when necessary. The solution shown in FIG. 5 can be applied to the online usage stage or the offline training stage. In the offline training stage, a user simulator can be used to replace the user, and a system simulator can be used to replace the API system. Among them, the user simulator and the system simulator can be respectively implemented using an LLM with strong processing capabilities to train the agent. Exemplarily, the prompt, the information of each API, and the name and input parameters of the API to be called can be input into the LLM. The prompt is "You are a system for processing API calls. Please refer to the relevant API knowledge and generate and return the values of the corresponding output parameters according to the name and values of the input parameters of the API to be called", so that the LLM can simulate the function of the API system and assist in implementing the API call. In this way, in the offline training phase, when it is difficult to directly obtain users and API systems in various industries, through the system simulator and user simulator, it is possible to simulate real users and API systems, so as to complete the training of the agent while saving resources as much as possible and improving the overall efficiency of training. Optionally, in order to alleviate the problem of lack of training data, in the embodiments of the present disclosure, training data can be automatically constructed based on multi-agent. FIG. 6 is a schematic flow chart of an agent training method provided by an embodiment of the present disclosure. As shown in FIG. 6, the agent training method may include: Step 601, generate portrait information of multiple users for interacting with the agent according to the portrait information of the agent. Step 602, respectively construct corresponding user agents according to the portrait information of each user, and construct a robot agent according to the portrait information of the agent. Wherein, both the user agent and the robot agent are implemented based on a trained model. The agent may be an agent for interacting with users. That is, in the process of completing tasks, the agent will involve interactions with users. Here, it is not excluded that the agent may also interact with other modules, such as an API system. Before training the agent, training data can be constructed first. Since there are few practical applications of large models and agents, and there is almost no corresponding training data, therefore, corresponding user agents and robot agents can be constructed according to the portrait information of the agent first, and then the construction of training data can be completed. FIG. 7 is a schematic diagram of the principle of constructing training data provided by an embodiment of the present disclosure. As shown in FIG. 7, portrait information of corresponding users can be generated first according to the portrait information of the agent to be trained. Exemplarily, the portrait information of the agent is: a smart assistant in the high-speed ETC center, which can help activate ETC, query ETC consumption records, etc. The portrait information of the agent can be input into a language model, and using the capabilities of the language model, portrait information of multiple users interacting with the agent can be generated. For example, the following information can be prompted to the language model: there is now a smart customer service, and its functions are... Who may be the users interacting with it? Describe their portraits. The language model can generate multiple portrait information according to the prompt. For example, User 1: newly bought a car and needs to activate ETC; User 2: has two cars and needs to activate first and then query. Traverse all the generated portrait information and construct corresponding user agents respectively. In addition, according to the portrait information of the agents, a robot agent can be constructed. Among them, the robot agent is used to simulate the agents in the foregoing embodiments, and the user agent is used to simulate the user. Both the robot agent and the user agent can be implemented based on an LLM or other models with strong capabilities, so as to simulate the interaction between the agent and the user based on the capabilities of the model and generate corresponding dialogue data as subsequent training data. Specifically, the relevant portrait information can be input into the LLM so that the LLM can implement the dialogue according to the portrait information. For example, the following prompt information can be input into the LLM corresponding to the robot agent: "You are an intelligent assistant at a high-speed ETC center, and you can help activate the ETC and query the ETC consumption records. Please assist the user to complete the relevant tasks of the ETC through communication with the user...". The following prompt information can be input into the LLM corresponding to the user agent: "You are a user who has just bought a car and needs to activate the ETC. Please complete the activation of the ETC through communication with the intelligent customer service...". Step 603, based on the node sequence information sample, control the robot agent to interact with each user agent to obtain a plurality of dialogue data, where the dialogue data includes multiple rounds of conversations between the user agent and the robot agent. After constructing multiple user agents and robot agents, any one user agent and robot agent can be selected from them, and the two interact according to their corresponding portrait information to construct dialogue data. For example, the user agent outputs: "I need to activate the ETC for my newly bought car", and the robot agent can output: "Okay, it is queried that your license plate information is Jing CXXXX. Do you want to activate this license plate?". When constructing the dialogue data, the processing process of the robot agent can refer to the processing process of the agent in the foregoing embodiments, that is, it can be processed in combination with the node sequence information sample. In addition, when the node sequence information sample involves calling the API, the system simulator can also be used to complete the API call. Step 604, train the agent according to the multiple dialogue data and the corresponding node sequence information sample to obtain the trained agent. After obtaining a plurality of dialogue data, the agent to be trained can be trained according to the dialogue data and the node sequence information sample, and the trained agent can be used to execute the method shown in Figure 2. Optionally, the dialogue data and the node sequence information samples can be constructed separately for multiple industries. Specifically, for each industry, the portrait information of the intelligent agent corresponding to the industry can be set, and the flow chart corresponding to the industry can be obtained. The node sequence information samples can be constructed according to the flow chart, and further the construction of the dialogue data of the industry can be realized, so that the trained intelligent agent has the task processing ability of multiple industries. The training method provided in this embodiment can simulate real users through the constructed user intelligent agent, so as to construct multi-round dialogue data based on scenario knowledge, form fine-tuning data that strictly corresponds to the flow chart, and conduct targeted training on the intelligent agent, further improving the effect of the intelligent agent in complex tasks. In addition, diverse user portrait information is constructed through the portrait information of the intelligent agent, and diverse user intelligent agents are constructed based on the diverse user portrait information, so that there is a high degree of diversity on the user side, and then high-quality and diverse training data can be constructed, effectively improving the training effect. Optionally, based on the node sequence information samples, controlling the interaction between the robot intelligent agent and each user intelligent agent may include: during the interaction between the user intelligent agent and the robot intelligent agent, inputting the interaction information between the user intelligent agent and the robot intelligent agent and the discrimination prompt information into the discrimination model, where the discrimination prompt information is used to prompt the discrimination model to judge whether the information output by the robot intelligent agent in the interaction information meets the requirements according to the portrait information and the node sequence information samples corresponding to the robot intelligent agent; if it is determined according to the discrimination model that the information output by the robot intelligent agent does not meet the requirements, the robot intelligent agent is instructed to regenerate the output information. Specifically, since the combination of the intelligent agent and the large model is a relatively new technology, even the existing models with strong capabilities may be difficult to complete task processing well in some scenarios. Therefore, in the embodiments of the present disclosure, a discrimination model can be added as an adversarial role to always pay attention to whether the robot intelligent agent outputs as expected. For example, the following prompt information can be input into the discrimination model: "There is now an intelligent customer service, its function is..., the processing flow of the current task is..., and its interaction information with the user is... Please view whether its reply is appropriate from the perspective of the manager." If the discrimination model believes that the information output by the robot intelligent agent does not meet the requirements, the robot intelligent agent can be allowed to re-output. Optionally, the reason for not meeting the requirements can also be prompted to be output by the discrimination model, so that the reason for not meeting the requirements can also be input into the robot intelligent agent, and the robot intelligent agent can re-output in combination with the reason. Among them, the discrimination model can also be implemented using a model with stronger capabilities, such as an LLM. The discrimination model is mainly used to judge the output of the robot agent. Since judgment is usually simpler than generation, the judgment ability of the discrimination model can be used to assist the generation ability of the robot agent, helping the robot agent better complete the generation of conversations, improving the accuracy of the constructed training data, and enhancing the training effect of the agent. Referring to FIG. 7, a robot response quality judgment agent can be designed. The robot response quality judgment agent is implemented based on the aforementioned discrimination model. Specifically, after the robot agent generates a response content, the response content can be sent to the robot response quality judgment agent first, and the robot response quality judgment agent can feedback the judgment on the quality of the response and the corresponding reasons. The robot agent can output a high-quality response to the user agent according to the feedback of the robot response quality judgment agent. Optionally, based on the solutions shown in FIGS. 6 and 7, training data can be constructed using user agents, robot agents, etc. Then, based on the training data, the agents can be trained through the framework shown in FIG. 5. After the training is completed, the customer can configure relevant process knowledge and API knowledge, and enter the online usage stage after release. In the online usage stage, the query of real users is used as the input information, and the real API system is used to implement the call of the API. In summary, the embodiments of the present disclosure solve the problem of constructing complex processes through a graph structure, enabling the agent to execute step by step according to the graph instructions defined by the scenario, and the implementation path is controllable. Moreover, the method of automatically constructing dialogue data using multiple agents is also used to further solve the problem of lack of training data. The embodiments of the present disclosure have at least the following effects: 1. Based on the graph structure and flowchart specifications, enable customers to complete the knowledge construction of complex task processes for agents just like drawing daily flowcharts; 2. Based on the edges in the flowchart representing jump conditions and the nodes representing current action instructions, convert the flowchart into a structured graph instruction. At the same time, a thought chain is designed for each output of the large model to infer the current node by adding the jump condition to the previous node and then perform actions based on the current node, ensuring that the large model performs controllable planning and execution based on the flowchart; 3. Through role-playing based on large models and multiple portraits, diverse user agents are constructed, and multi-turn dialogue data based on scenario knowledge is automatically constructed, simulating corresponding data in the absence of data, effectively alleviating the problem of lack of training data, and further conducting targeted training on the agents, ensuring the processing effect and controllability of the agents. The method provided by the embodiments of the present disclosure can be applied to any field and scenario. FIG. 8 is a schematic flowchart of an information processing method applied to an intelligent customer service scenario provided by the embodiments of the present disclosure. As shown in FIG. 8, the method includes: Step 801, obtain the input information of the user. Step 802, input the input information and node sequence information into a large model, where the node sequence information is used to indicate multiple nodes included in the task process and the edges between the nodes, the nodes represent action instructions for the intelligent agent, and the edges represent jumps between the nodes, so that the large model determines the target node to be executed according to the input information, and determines the intelligent agent output information according to the action instruction corresponding to the target node. Among them, the intelligent agent can implement the processing of the task through multiple rounds of interaction with the user. Each round can execute the above steps. Optionally, only the latest obtained input information of the user can be input into the large model, or the interaction information of the previous rounds can be input into the large model, or the prompt information can also be input into the large model to improve the task processing effect. Step 803, output the intelligent agent output information to the user. Among them, the intelligent agent output information can be the information sent by the intelligent agent to the user. After obtaining the intelligent agent output information through the large model, the intelligent agent output information can be output to the user. After multiple rounds of interaction, the processing of the task can be completed. The specific implementation principles and processes of each step in this embodiment can be referred to the foregoing embodiments, and will not be elaborated here. The information processing method provided in this embodiment can use a large model-driven intelligent agent architecture to implement the function of intelligent customer service, avoiding the problems of high configuration cost, high threshold, and lack of flexibility of traditional intelligent customer service. Moreover, it can also solve the problems of the intelligent agent in aspects such as the construction of complex process knowledge and the controllability of the planning part, enabling the intelligent agent to complete the processing of the task according to the guidance of the task process and improving the user experience. Corresponding to the above method, the embodiments of the present disclosure further provide an information processing device, including: A determination module, configured to determine the information to be processed corresponding to the intelligent agent; A processing module, configured to input the information to be processed and node sequence information into a large model, where the node sequence information is used to indicate multiple nodes included in the task process and the edges between the nodes, the nodes represent action instructions for the intelligent agent, and the edges represent jumps between the nodes, so that the large model determines the target node to be executed according to the information to be processed, and controls the intelligent agent to execute the action instruction corresponding to the target node. The embodiments of the present disclosure further provide an information processing device applied to an intelligent customer service scenario, including: An acquisition module for acquiring the input information of the user; A processing module for inputting the input information and node sequence information into a large model, where the node sequence information is used to indicate multiple nodes included in a task process and the edges between the nodes. A node represents an action instruction for an intelligent agent, and an edge represents a jump between nodes, so that the large model determines a target node to be executed according to the input information, and determines the intelligent agent output information according to the action instruction corresponding to the target node; An output module for outputting the intelligent agent output information to the user. For the specific implementation principle and effect of the device provided in the embodiments of the present disclosure, reference may be made to the foregoing embodiments, which will not be elaborated here. FIG. 9 is a schematic structural diagram of an electronic device provided in an embodiment of the present disclosure. As shown in FIG. 9, the electronic device in this embodiment may include: At least one processor 901; and a memory 902 communicatively connected to the at least one processor; wherein, the memory 902 stores instructions executable by the at least one processor 901, and the instructions are executed by the at least one processor 901 to enable the electronic device to execute the method described in any of the foregoing embodiments. Optionally, the memory 902 may be either independent or integrated with the processor 901. For the implementation principle and technical effect of the electronic device provided in this embodiment, reference may be made to the foregoing embodiments, which will not be elaborated here. The embodiments of the present disclosure further provide a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the method described in any of the foregoing embodiments is implemented. The embodiments of the present disclosure further provide a computer program product, including a computer program, which when executed by a processor implements the method described in any of the foregoing embodiments. In several embodiments provided by the present disclosure, it should be understood that the disclosed devices and methods may be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. The integrated modules implemented in the form of software function modules as described above may be stored in a computer-readable storage medium. The above-mentioned software function modules are stored in a storage medium, including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor to execute some steps of the methods described in the embodiments of the present disclosure. It should be understood that the above-mentioned processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the method disclosed in combination with the application can be directly embodied as being executed and completed by a hardware processor, or can be executed and completed by a combination of hardware and software modules in the processor. The memory may include high-speed random access memory (RAM), and may also include non-volatile memory (NVM), such as at least one disk memory, and may also be a USB flash drive, a portable hard disk, a read-only memory, a magnetic disk, or an optical disc, etc. The above storage medium may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, a magnetic disk, or an optical disc. The storage medium may be any available medium accessible by a general-purpose or special-purpose computer. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium may also be a component of the processor. The processor and the storage medium may be located in an application specific integrated circuit. Of course, the processor and the storage medium may also exist as discrete components in an electronic device or a master control device. It should be noted that in this text, the term "including", "comprising" or any other variants thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such a process, method, article or device. Through the description of the above embodiments, those skilled in the art can clearly understand that the above-described embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present disclosure, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions for causing a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in various embodiments of the present disclosure. The above are only the preferred embodiments of the present disclosure, and do not limit the patent scope of the present disclosure accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present disclosure, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present disclosure.< / end> < / start> < / end> < / start> < / end> < / start> < / end> < / start> < / end> < / start>

Claims

1. An information processing method, wherein Including: Determine the information to be processed corresponding to the agent; Input the information to be processed and the node sequence information into a large model, where the node sequence information is used to indicate multiple nodes included in the task process and the edges between the nodes. The nodes represent action instructions for the agent, and the edges represent jumps between the nodes, so that the large model determines the target node to be executed according to the information to be processed and controls the agent to execute the action instruction corresponding to the target node; Wherein, the edge is further used to represent a jump condition, so that the large model determines whether the current satisfies the jump condition corresponding to the target edge according to the information to be processed. The target edge is the edge starting from the node in the previous round. When the jump condition corresponding to the target edge is satisfied, determine the target node as the end node of the target edge; The information to be processed includes the user's input information; When the jump condition of any node in the node sequence information includes an intention and / or a slot, the method further includes: inputting jump condition information into the large model, where the jump condition information includes the description information corresponding to the intention and / or slot included in the node sequence information, so that the large model determines that the jump condition corresponding to the target edge is satisfied when it determines that the user's input information matches the description information of the intention and / or slot corresponding to the target edge.

2. The method according to claim 1, wherein, The agent realizes the processing of the task through multiple rounds of interaction with the environment; In any round, the information to be processed includes the interaction information of the previous round; wherein, the interaction information includes the agent output information and the agent input information obtained from the environment, so that the large model determines the target node corresponding to the current round according to the interaction information of the previous round, the node sequence information, and the node corresponding to the previous round.

3. The method according to claim 2, wherein The interaction with the environment includes at least one of the following: interacting with the user, calling an application programming interface API, querying a document; When it is determined according to the action instruction corresponding to the target node that the interaction in the current round is to interact with the user, the agent output information includes the reply information to the user; When it is determined according to the action instruction corresponding to the target node that the interaction in the current round is to call an API, the agent output information includes the name of the API to be called and the value of the input parameter; When it is determined according to the action instruction corresponding to the target node that the interaction in the current round is to query a document, the agent output information includes the information to be queried.

4. The method according to claim 2, wherein Also including: Input the API information corresponding to the API in the node sequence information into the large model; wherein, the API information includes the name, description information, at least one input parameter, and at least one output parameter of the API, so that when the action instruction corresponding to the target node includes calling an API, the large model determines whether the value of at least one input parameter of the API has been obtained. When the value of at least some input parameters has not been obtained, interact with the user based on the description information of the API to determine the value of the at least some input parameters, so as to complete the call of the API through the name of the API and the value of at least one input parameter and obtain the value of at least one output parameter.

5. The method according to claim 2, wherein, Also including: Input environmental information into the large model so that when the action instruction corresponding to the target node in the large model includes calling an API, determine the values of at least some input parameters of the API to be called according to the information to be processed and the environmental information, and when there are specific input parameters, determine the values of the specific input parameters based on the interaction with the user; wherein, the specific input parameter is an input parameter whose value cannot be determined through the information to be processed and the environmental information.

6. The method according to claim 2, wherein It further includes: Input the output prompt information into the large model; wherein, the output prompt information is used to prompt that the output information of the large model includes: the node corresponding to the previous round, the target node of the current round, and the intelligent agent output information determined according to the target node. After determining the intelligent agent output information in the output information generated by the large model, output the intelligent agent output information through the intelligent agent.

7. The method according to any one of claims 1-6, wherein, The multiple nodes include reply nodes for replying to the user, and the action instructions corresponding to the reply nodes are any one of the following types: Ordinary reply, used to indicate replying according to the content included in the action instruction. Strict reply, used to indicate replying according to the content and format given in the action instruction that are prohibited from being modified. Step-by-step reply, used to indicate replying to the user through at least two interactions. Flexible reply, used to indicate replying according to the knowledge given in the action instruction. Wherein, the action instruction includes keywords for indicating the type of reply, so that the large model determines the type of reply according to the keywords and generates corresponding reply information.

8. The method according to any one of claims 1-7, wherein The multiple nodes include call nodes for calling an API, and the action instructions corresponding to the call nodes include: The default part, used to indicate the name of the API to be called; wherein, the API corresponds to at least one input parameter. The custom part, used to indicate the format of the values of all or some of the at least one input parameter, so that the large model obtains values that conform to the format according to the custom part.

9. The method according to any one of claims 1-8, wherein, It further includes: Obtain the flow chart input by the customer for indicating the task process. Generate the node sequence information according to the flow chart, and the node sequence information is in text form.

10. The method according to any one of claims 1-9, wherein, When the intelligent agent is used to interact with the user, the intelligent agent is trained in the following manner: Generate the portrait information of multiple users for interacting with the intelligent agent according to the portrait information of the intelligent agent. Construct corresponding user intelligent agents according to the portrait information of each user, and construct a robot intelligent agent according to the portrait information of the intelligent agent; wherein, both the user intelligent agent and the robot intelligent agent are implemented based on a trained model. Based on the node sequence information samples, control the robot intelligent agent to interact with each user intelligent agent to obtain multiple conversation data, and the conversation data includes multiple rounds of conversations between the user intelligent agent and the robot intelligent agent. Train the intelligent agent according to the multiple conversation data and the corresponding node sequence information samples to obtain the trained intelligent agent.

11. An information processing method, wherein, It includes: Obtain the input information of the user. Input the input information and node sequence information into a large model, where the node sequence information is used to indicate multiple nodes included in a task process and the edges between the nodes. A node represents an action instruction for an agent, and an edge represents a jump between nodes, so that the large model determines a target node to be executed according to the input information, and determines agent output information according to the action instruction corresponding to the target node; Output the agent output information to the user; wherein, the edge is further used to represent a jump condition, so that the large model determines whether the current situation meets the jump condition corresponding to the target edge according to the user's input information, where the target edge is an edge starting from the node in the previous round, and when the jump condition corresponding to the target edge is met, determine the target node as the end node of the target edge; When the jump condition of any node in the node sequence information includes an intention and / or a slot, the method further includes: inputting jump condition information into the large model, where the jump condition information includes description information corresponding to the intention and / or slot included in the node sequence information, so that when the large model determines that the user's input information matches the description information of the intention and / or slot corresponding to the target edge, it is determined that the jump condition corresponding to the target edge is met.

12. An electronic device, wherein, Comprising: At least one processor; And A memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to cause the electronic device to execute the method according to any one of claims 1-11.

13. A computer-readable storage medium, wherein, Computer-executable instructions are stored in the computer-readable storage medium, and when the processor executes the computer-executable instructions, the method according to any one of claims 1-11 is implemented.

14. A computer program product, wherein, Comprising a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1-11 is implemented.

Images