Information processing methods, systems, devices, and media based on dynamic graphs

Abstract

This application relates to a dynamic graph-based information processing method, system, device, and medium, belonging to the field of artificial intelligence technology. The dynamic graph-based information processing method includes: acquiring user query information; determining predicted node intent information based on the user query information; searching for target intent nodes in a preset dynamic graph based on the predicted node intent information; and performing conversation processing based on the collected slot information corresponding to the target intent node to obtain query result information corresponding to the user query information. This maximizes the flexibility of the dialogue while ensuring the controllability of the dialogue flow, and improves user experience and satisfaction while reducing the complexity of task-oriented business script configuration.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an information processing method, system, device and medium based on dynamic graphs. Background Technology

[0002] With the continuous development of artificial intelligence technology, the application scope of dialogue systems based on artificial intelligence technology is becoming wider and wider, bringing great convenience to people's lives, work and study.

[0003] Currently, dialogue systems are mainly divided into open-domain dialogue systems and task-oriented dialogue systems. Task-oriented dialogue systems primarily involve multi-turn conversations to achieve a specific goal. Existing task-oriented dialogue systems require a high level of understanding of the context in their question-and-answer settings, such as... Figure 1 As shown, in a simple task dialogue flowchart for purchasing airline tickets, the task graph includes task nodes such as departure point, destination, and departure time. The dialogue between the task-based dialogue system and the user must strictly follow the task graph's flow. That is, the user must first enter the departure point, then the destination, and finally the departure time. Only after the task-based dialogue system displays the flight number based on the departure point, destination, and departure time can the user select a flight. If the user's input does not strictly follow the task graph's flow, the existing task-based dialogue system will pause multiple times for clarification, resulting in a poor user experience. Considering that sometimes the user's intent does not completely follow the flowchart, the existing task-based dialogue system's task graph may be configured with numerous node connections to satisfy various user intents. However, this not only increases the complexity of the task graph but also tests the operational experience and understanding of the business scenario when configuring task graphs, increasing the complexity of task-based business dialogue configuration and bringing operational complexity. Summary of the Invention

[0004] In view of this, this application provides an information processing method, system, device and medium based on dynamic graphs, so as to maximize the flexibility of the dialogue while ensuring the controllability of the dialogue process, and improve the user experience and satisfaction while reducing the complexity of task-oriented business script configuration.

[0005] In a first aspect, embodiments of this application provide an information processing method based on dynamic graphs, including:

[0006] Obtain user inquiry information;

[0007] Based on the user query information, determine the intention information of the predicted node;

[0008] Based on the predicted node intent information, the target intent node is searched in a preset dynamic graph;

[0009] Based on the collected slot information corresponding to the target intent node, session processing is performed to obtain the query result information corresponding to the user query information.

[0010] Optionally, determining the predicted node intent information based on the user query information includes:

[0011] The user query information is input into a pre-trained node intent classification model to obtain the predicted node intent information output by the node intent classification model; or,

[0012] Intent recognition is performed based on the question keywords in the user inquiry information to obtain intent recognition results, and the intent information of the predicted node corresponding to the first recognition score is extracted from the intent recognition results.

[0013] Optionally, the step of performing session processing based on the collected slot information corresponding to the target intent node to obtain the query result information corresponding to the user query information includes:

[0014] Based on the collected dialogue configuration information of the target intent node, the conversation guidance information is displayed;

[0015] Obtain user input information submitted in response to the session guidance information;

[0016] The user input information is identified as the collected slot information;

[0017] If the collected slot information includes the target collected slot corresponding to the target intent node, then the session processing is performed based on the target collected slot to obtain the query result information;

[0018] If the collected word slot information does not include the target collected word slot corresponding to the target intent node, then the jump intent information is determined based on the user input information;

[0019] Based on the jump intent information, the session jump node is determined, and the session is processed based on the collected slot information corresponding to the session jump node to obtain the query result information.

[0020] Optionally, the information processing method further includes:

[0021] For each target intent node, record the number of jumps corresponding to that session jump node;

[0022] If the number of jumps exceeds a preset threshold for the number of node jumps, the session jump node and the target intent node are merged to obtain a merged session node.

[0023] The dynamic graph is updated based on the merged session node.

[0024] Optionally, the step of performing conversation processing based on the target collected slots to obtain the query result information includes:

[0025] If the target collection slot meets the termination dialogue configuration conditions of the target intent node, then the conversation service information corresponding to the target collection slot is determined, and the query result information is generated based on the conversation service information;

[0026] If the target collection slot does not meet the termination dialogue configuration conditions of the target intent node, then based on the target collection slot and the collection dialogue configuration information, the target session guidance information is displayed, and the target input information submitted by the user for the target session guidance information is obtained.

[0027] Based on the target input information, session processing is performed to obtain the query result information.

[0028] Optionally, before inputting the user query information into the pre-trained node intent classification model, the method further includes:

[0029] The intent information of the sample nodes is determined as the intent information of the nodes to be trained;

[0030] Based on the intent information of the node to be trained, obtain the input sample information corresponding to the sample node;

[0031] The node intent classification model is obtained by training the model based on the input sample information and the intent information of the node to be trained.

[0032] Optionally, the step of searching for the target intent node in a preset dynamic graph based on the predicted node intent information includes:

[0033] For each session node in the dynamic graph, determine the matching degree between the intent information of the session node and the intent information of the predicted node;

[0034] The nodes are sorted according to the matching degree to obtain the node sorting order;

[0035] Based on the node sorting order, the session node corresponding to the first matching degree is determined as the target intent node.

[0036] Secondly, embodiments of this application provide an information processing system based on dynamic graphs, including:

[0037] The user inquiry module is used to obtain user inquiry information;

[0038] The node intent module is used to determine the predicted node intent information based on the user query information.

[0039] The node search module is used to search for target intent nodes in a preset dynamic graph based on the predicted node intent information.

[0040] The session processing module is used to perform session processing based on the collected slot information corresponding to the target intent node, and obtain the query result information corresponding to the user query information.

[0041] Thirdly, embodiments of this application provide an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor, when executing the program stored in the memory, implements the steps of the information processing method as described in any of the first aspects.

[0042] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the information processing method as described in any of the first aspects.

[0043] This application embodiment determines the intention information of the predicted node through the user's query information, and finds the target intention node in the preset dynamic graph based on the predicted node intention information. The conversation is then processed based on the collection slot information corresponding to the target intention node to obtain the query result information corresponding to the user's query information. This maximizes the flexibility of the dialogue while ensuring that the dialogue process is controllable. It solves the problem caused by the requirement in the prior art to strictly follow the process of the task graph of the task-oriented dialogue system. While reducing the complexity of the task-oriented business script configuration, it improves the user experience and satisfaction. Attached Figure Description

[0044] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0045] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart of a simple task session for purchasing airline tickets in the existing technology;

[0047] Figure 2 This is a flowchart of the steps of an information processing method based on dynamic graphs provided in an embodiment of this application;

[0048] Figure 3This is a schematic diagram of a task-oriented dialogue flow model in one example of this application;

[0049] Figure 4 A flowchart illustrating the steps of an information processing method based on a dynamic graph, provided as an optional embodiment of this application;

[0050] Figure 5 This is a structural block diagram of an information processing system based on dynamic graphs, provided in an embodiment of this application. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0052] There are two main approaches to implementing task-oriented dialogue systems: one is the pipeline approach, which mainstream models create as four processes: Natural Language Understanding (NLU), State Tracker, Policy Learning, and... There are two main approaches to pipelines in task-oriented dialogue: one is through learning and the other is through natural language generation (NLG). The other is an end-to-end approach. However, the end-to-end approach is less common in industry due to the complexity of task conversation data, difficulties in training and decoding models, high model training and deployment costs, and low maturity. Various pipeline methods exist for task-oriented dialogue, including the simplest slot-filling method, finite state machines, and methods combining deep learning with training. Among these, finite state machine models are widely popular because they offer stable performance in industry and have low development and maintenance costs. Finite state machines are characterized by excellent session state tracking and fixed conditional transitions, resulting in high stability. However, this also leads to fixed dialogue patterns based on finite state machines, preventing flexible switching of responses based on user feedback and dynamic transitions to appropriate state nodes. While policy learning combined with deep learning can achieve flexible node transitions, its current performance is unsatisfactory due to training complexity, implementation instability, and other factors.

[0053] Currently, the access task graph and the task graph itself are completely isolated. Flow within a task graph is achieved through task entry intents, and user switching intents are managed through the intents of the entire graph. Specifically, the existing task graph is a whole; intent recognition only involves hitting or not hitting this task graph. That is, the granularity of intent recognition is at the task graph level, not the node level. Therefore, current dialogue switching can only switch from one task graph to the starting node of another, initiating a new dialogue flow, but not from one task graph to a specific node within another. Once a user triggers a task graph, there are only two options: either continue flowing according to the task graph's process or exit the task graph. It's clear that all current task graph triggers begin at the starting node, creating coupling between intra-task and inter-task conversations. Switching only occurs from within a task to outside, not from within a task itself. This makes it difficult to switch from an intermediate state within one task graph to an intermediate state within another. Overall, the dialogue flow is strictly controlled, resulting in a poor user experience.

[0054] One of the core concepts of this application is to propose an information processing method based on dynamic graphs. By reasonably controlling the granularity of node slot collection and node intent, after obtaining user query information, the method determines the predicted node intent information based on the user query information. Based on the predicted node intent information, the method searches for the target intent node in the preset dynamic graph. Thus, the method can perform conversation processing based on the collected slot information corresponding to the target intent node, which solves the problem caused by the requirement in the prior art to strictly follow the task graph process of the task-oriented dialogue system for dialogue.

[0055] To facilitate understanding of the embodiments of this application, the following will provide further explanation and description with reference to the accompanying drawings and specific embodiments. These embodiments do not constitute a limitation on the embodiments of this application.

[0056] Reference Figure 2 This document illustrates a flowchart of a dynamic graph-based information processing method provided in an embodiment of this application. Specifically, the dynamic graph-based information processing method provided in this application may include the following steps:

[0057] Step 210: Obtain user inquiry information;

[0058] Step 220: Determine the intention information of the predicted node based on the user query information;

[0059] Step 230: Based on the predicted node intent information, search for the target intent node in the preset dynamic graph;

[0060] Step 240: Perform session processing based on the collected slot information corresponding to the target intent node to obtain the query result information corresponding to the user query information.

[0061] In this embodiment, the user query information may include various user questions input by the user, and these user questions can trigger node intents. Therefore, predicted node intent information can be determined based on the user query information, and the most suitable node entry point can be found through intent recognition scoring to enter the node session collection state. That is, the target intent node is searched in a preset dynamic graph based on the predicted node intent information to enter the session collection state corresponding to the target intent node. Here, the target intent node may refer to the most suitable task graph node determined based on the predicted node intent information; the predicted node intent information may refer to the node intent determined based on the user query information, which can be used to match session nodes in the dynamic graph.

[0062] In practical processing, node sessions primarily collect information in the form of slot collection. Operators configure the slot information that a node needs to collect and its corresponding collection scripts, thus driving user interaction through the configured slot collection scripts. Therefore, in this embodiment, after locating a target intent node, by obtaining the corresponding slot collection information, session processing can be performed based on this information to obtain the query result information corresponding to the user's query. The slot collection information corresponding to the target intent node can include slot information collected according to the configured slot collection scripts for that target intent node.

[0063] For example, when the target intent node is a node that needs to collect the user's departure and destination information, the collection dialogue corresponding to the departure point, "Which city do you choose to depart from?", and the collection dialogue corresponding to the destination, "Which city do you want to arrive at?", can be configured as the collection slot dialogues corresponding to the target intent node. This allows the collection of the user's input destination and departure slot information at the target intent node according to the collection slot dialogues. The collected destination and departure slot information can then be used as the collection slot information for the target intent node. Based on this destination and departure slot information, conversation processing can be performed to obtain the corresponding flight number information, which serves as the query result information for the user's query. This allows the user to input their destination and departure point at the same node, achieving the goal of quickly providing the user with flight number information from their departure point to their destination. The destination slot information can be used to determine the destination entered by the user, such as the destination address "A" entered by the user; the departure slot information can be used to determine the departure point entered by the user, such as the departure address "B" entered by the user, etc. This example does not impose specific restrictions on this.

[0064] As can be seen, after obtaining the user's query information, this embodiment can determine the predicted node intent information based on the user's query information, and use the predicted node intent information to find the target intent node in the preset dynamic graph. Then, based on the collected slot information corresponding to the target intent node, the session is processed to obtain the query result information corresponding to the user's query information. This solves the problem caused by the requirement to strictly follow the task graph process of the task-oriented dialogue system in the prior art. It maximizes the flexibility of the dialogue while ensuring the controllability of the dialogue process, thereby reducing the complexity of task-oriented business script configuration and improving user experience and satisfaction.

[0065] It should be noted that the task graph is a task-oriented session and flow model implemented based on a directed graph model; where the nodes (i.e., graph nodes) can represent the states that need to be collected in the current session, the incoming edges of the graph nodes represent the intentions that can trigger that node, and the outgoing edges of the graph nodes represent the next intentions predicted after the current node's states have been collected. For example, as... Figure 3As shown, the task graph may not have a distinction between start and finish nodes; all nodes can have incoming and outgoing edges, or only incoming edges without outgoing edges; incoming edges can be used to represent the contextual intent of entering the node in the dialogue, and entering the node's conversation can be achieved by satisfying the contextual intent corresponding to the node; outgoing edges can represent the contextual intent of the node in the dialogue, and by comparing and scoring this contextual intent with the contextual intents of the entire node, the most suitable match can be found for jump. Among them, the most important configuration of each node is the slots to be collected. One or more slots can be configured, and in order to ensure the normal flow of nodes, speech clarification, super-rotation termination, etc. can be configured for each node to predict the next intent after the node is completed.

[0066] Furthermore, this application embodiment performs session processing based on the collection slot information corresponding to the target intent node to obtain query result information corresponding to the user query information, including: displaying session guidance information based on the collection script configuration information of the target intent node; obtaining user input information submitted in response to the session guidance information; determining the user input information as the collection slot information; if the collection slot information contains the target collection slot corresponding to the target intent node, then performing session processing based on the target collection slot to obtain the query result information; if the collection slot information does not contain the target collection slot corresponding to the target intent node, then determining jump intent information based on the user input information; determining a session jump node based on the jump intent information, and performing session processing based on the collection slot information corresponding to the session jump node to obtain the query result information. The session guidance information can be used to prompt the user to input information as the collection slot information corresponding to the target intent node. The target collection slot corresponding to the target intent node can refer to the slots that the target intent node needs to collect, specifically including one or more slots required for session processing; this application embodiment does not impose specific limitations on this.

[0067] Specifically, in this embodiment, after locating the target intent node, conversation guidance information can be displayed based on the collection script configuration information of the target intent node to prompt the user to input corresponding information for the conversation guidance information. If the user inputs information after the conversation guidance information is displayed, the user's input information can be used as the user input information submitted for the conversation guidance information, and the user's input information can be determined as the collection slot information corresponding to the target intent node. It can be determined whether the user's input information contains the target collection slot corresponding to the target intent node. Thus, if the user's input information contains the target collection slot, conversation processing can be performed based on the target collection slot contained in the user's input information to obtain the query result information corresponding to the user's query information. If the user's input information does not contain the target collection slot corresponding to the target intent node, jump intent information is determined based on the user's input information, and a conversation jump node is determined based on the jump intent information to realize the jump of the intent node. Subsequently, conversation processing can be performed based on the collection slot information corresponding to the conversation jump node to obtain the query result information.

[0068] As an example of this application, if the predicted node intent information is determined to be "query flight information" based on the user's query information "What information is needed to book a flight", the predicted node intent information "query flight information" can be matched with the intent information of each session node in a preset dynamic graph. The session node in the dynamic graph whose intent information matches the predicted node intent information "query flight information" is then identified as the target intent node. For example, if the intent information of a session node in the dynamic graph is "query flight", that session node can be identified as the target intent node; that is, the session node with the intent information "query flight" is identified as the target intent node. If the collection script configuration information corresponding to the session node with the intent information "query flight" is "Where to depart from, where to arrive at". Based on the collected dialogue configuration information "where you depart from and where you are going," the system can display corresponding conversation guidance information, such as "Which city are you departing from?" or "Which city do you need to go to?" on the dialogue interface. This prompts the user to further input their departure and destination locations, thus enabling conversational communication based on the user's input. For example, in the case where the user inputs address "B" as the departure location and address "A" as the destination, the system can perform a flight search based on these addresses and provide the search results for flights from B to A as the query result to the user, such as displaying the flight information from B to A on the dialogue interface.

[0069] If a user enters information such as "Which city are you departing from?" or "Which city do you need to reach?" on the dialog interface, and then enters information that does not contain the target collection slots, such as "I want to take the train," it can be determined that the user's input does not contain the target collection slots corresponding to the target intent node. Subsequently, node intent can be predicted based on the user's input "I want to take the train," and the predicted node intent information can be identified as jump intent information. For example, if the predicted node intent information "search for train tickets" is identified as jump intent information, node jump can be performed based on the jump intent information. Thus, the jump node can be used as the conversation jump node. That is, the conversation jump node is determined based on the jump intent information. Then, conversation processing can be performed based on the collection slot information corresponding to the conversation jump node. When the user enters the departure address as address "B" and the destination address as address "A," a train ticket search can be performed based on the departure address "B" and the destination address "A." The searched train ticket information from B to A can be fed back to the user as the query result information, such as displaying the train ticket information from B to A on the dialog interface.

[0070] As can be seen, after finding the target intent node, this embodiment can display conversation guidance information based on the collection script configuration information of the target intent node. By obtaining user input information submitted in response to the conversation guidance information, the collection slot information corresponding to the target intent node can be determined, and it can be determined whether the collection slot information contains the target collection slot corresponding to the target intent node. When the collection slot information contains the target collection slot corresponding to the target intent node, conversation processing is performed based on the target collection slot. When the collection slot information does not contain the target collection slot corresponding to the target intent node, the jump intent information is determined based on the user input information, and the conversation jump node is determined based on the jump intent information. This realizes flexible jump of conversation nodes. Subsequently, conversation processing can be performed based on the collection slot information corresponding to the conversation jump node to obtain the query result information corresponding to the user query information. This realizes flexible and intelligent script jump, reduces the usage threshold of the conversation task graph, and improves the flexibility of the dialogue process.

[0071] In practical processing, after obtaining user inquiry information, this embodiment of the application can extract question keywords from the user inquiry information to perform intent recognition based on the extracted question keywords, thereby determining the intent information of the predicted node based on the intent recognition result. Further, based on the above embodiment, the determination of the predicted node intent information based on the user inquiry information in this embodiment of the application can specifically include: performing intent recognition based on the question keywords in the user inquiry information to obtain an intent recognition result, and extracting the predicted node intent information corresponding to the first recognition score from the intent recognition result. The question keywords can include keywords used to determine intent, such as keywords in the user's question used to trigger the node intent. For example, in the case of the above example, if the obtained user inquiry information is the user-inputted question "What information is needed to book a flight?", "book a flight" can be extracted from "What information is needed to book a flight?" as a question keyword in the user inquiry information, so that intent recognition can be performed subsequently based on the question keyword "book a flight".

[0072] Specifically, after obtaining user query information, this embodiment can perform intent recognition based on the keywords in the user query information to trigger intent recognition and obtain intent recognition results. The intent recognition results can include one or more node intents identified and a recognition score corresponding to each node intent. This embodiment can sort the recognition scores of each node intent in the intent recognition results, taking the highest recognition score as the first recognition score. Based on this first recognition score, the node intent with the highest recognition score in the intent recognition results can be extracted as the predicted node intent information corresponding to the first recognition score. Therefore, this embodiment can determine the node intent with the highest recognition score in the intent recognition results as the intent information matched by the user query information through intent recognition scoring, i.e., determine the predicted node intent information, so that the target intent node can be found in a preset dynamic graph for session processing based on the predicted node intent information.

[0073] Of course, in addition to extracting keywords from user query information for intent recognition, this application embodiment can also perform intent recognition in other ways, such as through a pre-trained node intent classification model. This application embodiment does not impose specific limitations on this.

[0074] Furthermore, based on the above embodiments, the method of determining the predicted node intent information based on the user query information in the embodiments of this application may further include: inputting the user query information into a pre-trained node intent classification model to obtain the predicted node intent information output by the node intent classification model.

[0075] For example, after obtaining user query information, the user query information can be input into a pre-trained node intent classification model to identify the intent of the user query information. That is, the node intent information is inferred based on the user query information through the node intent classification model, and the node intent information inferred by the node intent classification model can be determined as the predicted node intent information. In other words, the node intent information output by the node intent classification model is determined as the predicted node intent information corresponding to the user query information, so that the predicted node intent information can be used to find the target intent node in a preset dynamic graph. Here, the dynamic graph can refer to a task graph formed based on the dynamic connection of nodes, and can specifically include one or more session nodes in the dialogue task. This application embodiment does not make specific limitations in this regard.

[0076] In practical processing, the flexibility of dialogue depends on the control of node strength. A single conversation node can collect a series of slots, and most task-oriented dialogues are for information gathering, such as collecting conversation information from most nodes through slots. In the extreme case where a task is represented by a single node, if the task-oriented dialogue only collects information from a few simple nodes, the task graph structure is relatively simple. However, if the task-oriented dialogue needs to collect information from a large number of nodes, it will cause the individual node to swell, compromising the flexibility of the dialogue. Furthermore, in the extreme case where each node only collects one slot, configuring a node to collect dialogue through a single slot will lead to node overabundance. Too many nodes and too many intent recognitions not only increase operational complexity but also make it more difficult to debug dialogue effects. Therefore, to collect information more flexibly, in this embodiment, the nodes in the dynamic graph can support the configuration of clarification and termination dialogues as the collection dialogues for that node, and can also support the configuration of the slots that the node needs to collect as the target collection slots for that node. Moreover, the configured slots and their collection dialogue order are adjustable.

[0077] Specifically, in this embodiment, after collecting all the slot information required for a node, the next node intent can be scored based on the pre-configured intent in the dynamic graph and the collected slot information. The node entry intent with the highest matching degree is then selected for dialogue flow based on the score. The intent recognition and scoring can be designed using a deep learning classification model or set using question keywords; this embodiment does not impose specific limitations on this.

[0078] Specifically, in this embodiment, after finding the target intent node in the preset dynamic graph, it can start preparing to collect the slot information configured on the target intent node according to the slots to be collected and the collection scripts corresponding to the slots. This information is used as the collection slot information corresponding to the target intent node, so that the collection slot information corresponding to the target intent node can be used for conversation processing. Furthermore, when the collection slot information includes the target collection slot, conversation processing can be performed based on the target collection slot. After collecting all the slot information configured on the target intent node, it is determined that the conversation state of the target intent node is at the end position. When the user's input question information hits the next node intent, it jumps to the next conversation node and starts a new round of conversation collection. If, within the current target intent node, the user's input information matches the intent of another node, that is, if the collection slot information corresponding to the target intent node does not contain the target collection slot corresponding to the target intent node, it can be determined that the user's access information matches the intent of another node. Subsequently, the target intent node can be terminated early, and the jump intent information can be determined based on the user's input information. Based on the jump intent information, the session jump node can be determined, and the session processing can be performed based on the collection slot information corresponding to the session jump node to obtain the query result information.

[0079] In specific implementation, when the collection slot information corresponding to the target intent node does not contain the target collection slot, whether to terminate the current target intent node or continue clarification based on the collection script configured for the target intent node can be set according to the switching node threshold provided within the session task, such as a preset node jump count threshold. This application embodiment does not impose specific restrictions on this. Furthermore, the information processing method based on the dynamic graph provided in this application embodiment may further include: for the target intent node, recording the jump count corresponding to the session jump node; if the jump count exceeds the preset node jump count threshold, merging the session jump node and the target intent node to obtain a merged session node; updating the dynamic graph based on the merged session node. If the jump count does not exceed the preset node jump count threshold, merging the session jump node and the target intent node is not required, i.e., updating the dynamic graph is not required.

[0080] For example, when a user visits a task related to booking a flight ticket, considering that booking a flight naturally requires a departure point and a destination, the two nodes in the dynamic graph can be merged into a single node. For instance, if the number of jumps corresponding to the destination node exceeds a preset threshold, the departure point node and the destination node can be merged into a merged session node. Subsequently, the dynamic graph of the flight ticket booking task can be updated based on the merged session node, realizing the dynamic update of the session nodes in the dynamic graph. By collecting slots for the departure point node and the destination node through the merged session node, the cohesion of the intent node can be guaranteed, which also helps with disambiguation in the session. In a single node, it is easier to ensure that the departure point and destination addresses are not the same, thus preventing ambiguity.

[0081] In summary, this embodiment can flexibly adjust the node granularity by controlling it, thereby forming the intent cohesion of nodes based on the word slot correlation in the nodes, ensuring greater decoupling between nodes, improving node reusability, realizing a flexible task graph model, and thus enabling more flexible and controllable task-oriented dialogues. It also has a further impact on the unification of the dialogue domain.

[0082] In practical processing, this application embodiment utilizes the intent information of sample nodes for model training to train a classification model for determining node intent, which serves as the node intent classification model in this application embodiment. Furthermore, before inputting the user query information into the pre-trained node intent classification model, this application embodiment may further include: determining the intent information of the sample nodes as node intent information to be trained; obtaining input sample information corresponding to the sample nodes for the node intent information to be trained; and training the model based on the input sample information and the node intent information to be trained to obtain the node intent classification model. Here, sample nodes can refer to collected session nodes, specifically including session nodes of different task types; the intent information of the sample nodes can be used to characterize the node intent of the collected session nodes.

[0083] Specifically, in the actual industrial production modeling process, if the granularity of node slot collection and node intent can be reasonably controlled, more flexible dialogue can be achieved. This avoids being confined to a fixed intent graph and allows for the flow of dialogue through node configuration context and intent prediction mechanisms, leading to better dialogue effects and more efficient product operation and development. Specifically, in the model training and construction phase of this application, the intent information of all acquired sample nodes can be used as the intent information of the nodes to be trained. Then, for the intent information of the nodes to be trained, the corresponding input sample information can be obtained, i.e., the information input by the user for the sample nodes can be obtained as the corresponding input sample information. Thus, for each sample node, model training can be performed based on the input sample information and the intent information of the nodes to be trained. The trained model can then be identified as a node intent classification model, which can be used subsequently to determine the predicted node intent information corresponding to the user's query information.

[0084] As an example of this application, the intent information of conversation nodes in different domains can be determined as the node intent information to be trained. For example, conversation nodes such as "buy plane tickets" and "buy train tickets" in the travel ticketing domain can be used as sample nodes. Subsequently, the intent information of these sample nodes can be determined as the node intent information to be trained. The input information "flight to XX" corresponding to the "buy plane tickets" node can be used as the input sample information, and the input information "train to XX" corresponding to the "buy train tickets" node can be used as the input sample information. Then, the model is built based on the input sample information "flight to XX" and the node intent information "buy plane tickets". The system can train a node intent classification model based on the input sample information "train to XX" and the intent information of the node to be trained "buy train ticket". This model allows for subsequent training of the node intent classification model by inputting user query information into the pre-trained model. The predicted node intent information is then used to search for the target intent node in a preset dynamic graph. Conversation processing is then performed based on the slot information collected for the target intent node. This solves the problems of complex dialogue configuration and strict adherence to the task graph flow of task-oriented dialogue systems in existing dynamic graph dialogue systems.

[0085] Reference Figure 4 This illustration shows a flowchart of a dynamic graph-based information processing method provided in an optional embodiment of this application. Specifically, the dynamic graph-based information processing method provided in this optional embodiment can be applied to the model inference stage, such as the inference and prediction stage of a node intent classification model. Figure 4 As shown, this information processing method based on dynamic graphs can specifically include the following steps:

[0086] Step 410: Obtain user inquiry information;

[0087] Step 420: Input the user query information into the pre-trained node intent classification model to obtain the predicted node intent information output by the node intent classification model;

[0088] Step 430: Based on the predicted node intent information, search for the target intent node in the preset dynamic graph;

[0089] Step 440: Based on the collected dialogue configuration information of the target intent node, display the conversation guidance information;

[0090] Step 450: Obtain user input information submitted in response to the session guidance information;

[0091] Step 460: Determine the user input information as the collection slot information corresponding to the target intent node;

[0092] Step 470: If the collected slot information includes the target collected slot corresponding to the target intent node, then perform session processing based on the target collected slot to obtain the query result information;

[0093] Step 480: If the collected slot information does not contain the target collected slot corresponding to the target intent node, then determine the jump intent information based on the user input information;

[0094] Step 490: Determine the session jump node based on the jump intent information, and perform session processing based on the collection slot information corresponding to the session jump node to obtain the query result information.

[0095] Specifically, after obtaining the user's query information, this application embodiment can input the user's query information into a pre-trained node intent classification model, and use the node intent classification model to output the corresponding predicted node intent information. Then, the predicted node intent information can be used to find the target intent node in a preset dynamic graph. For example, the intent information of each session node in the dynamic graph can be compared and matched with the predicted node intent information to determine the matching degree between the intent information of each session node and the predicted node intent information. Thus, the session node with the highest matching degree can be determined as the target intent node.

[0096] Furthermore, in this embodiment of the application, the target intent node is located in a preset dynamic graph based on the predicted node intent information, which may specifically include the following sub-steps:

[0097] Sub-step 4301: For each session node in the dynamic graph, determine the matching degree between the intent information of the session node and the intent information of the predicted node;

[0098] Sub-step 4302: Sort the nodes according to the matching degree to obtain the node sorting order;

[0099] Sub-step 4303: Based on the node sorting order, determine the session node corresponding to the first matching degree as the target intent node.

[0100] Specifically, in this embodiment, after determining the predicted node information based on the user query information, the predicted node information can be compared and matched with the intent information of all preset dynamic graphs to confirm the matching degree between the intent information of each session node and the intent information of the predicted node. Then, the session nodes are sorted according to the matching degree to obtain the node sorting order corresponding to all session nodes. Thus, the session node corresponding to the intent information with the highest matching degree can be extracted based on the node sorting order as the session node corresponding to the first matching degree. The session node corresponding to the first matching degree can be determined as the target intent node corresponding to the intent information of the predicted node, so that subsequent session processing can be performed based on the collected slot information corresponding to the target intent node to obtain the query result information corresponding to the user query information. This solves the problem in the prior art that the dialogue needs to be strictly carried out according to the task graph flow of the task-oriented dialogue system. It maximizes the flexibility of the dialogue while ensuring the controllability of the dialogue flow and improves the user experience.

[0101] Specifically, after determining the target intent node, this application embodiment can display conversation guidance information based on the collection script configuration information of the target intent node to guide the user to input further information. In this way, the information input by the user in response to the conversation guidance information can be determined as the collection slot information corresponding to the target intent node. That is, the user input information submitted in response to the conversation guidance information is determined as the collection slot information corresponding to the target intent node, so that subsequent conversation processing can be performed based on the collection slot information corresponding to the target intent node.

[0102] As an example of this application, when the user's query information is "Flights to Beijing?", the user's query information "Flights to Beijing?" can be input into a pre-trained node intent classification model. The predicted node intent information output by the node intent classification model is "Buy a ticket". Then, using the predicted node intent information "Buy a ticket", a node with the intent information "Buy a ticket" can be found in a preset dynamic graph as the target intent node. Based on the collected dialogue configuration information of the target intent node, conversation guidance information can be displayed to guide the user to complete the ticket purchase process. For example, if the collection script configuration information for a node with the intent information "buy a plane ticket" includes: the collection script corresponding to the departure city is "Which city do you choose to depart from?", the collection script corresponding to the destination is "Which city do you want to arrive at?", and the collection script corresponding to the departure time is "When do you want to depart?", then based on this collection script configuration information, conversation guidance information such as "Please enter the departure city", "Please enter the destination city", and "Please enter the departure time" can be displayed to guide the user to further input slot information such as departure city, destination city, and departure time. In this way, the information submitted by the user in response to the above conversation guidance information can be obtained as the collection slot information corresponding to the target intent node, and then the conversation can be processed based on the collection slot information corresponding to the target intent node.

[0103] In one optional implementation, this application embodiment can determine whether the collection slot information corresponding to the target intent node contains the target collection slot corresponding to the target intent node. If the collection slot information corresponding to the target intent node contains the target collection slot, the session processing is performed based on the target collection slot contained in the collection slot information, thereby obtaining query result information. If the collection slot information corresponding to the target intent node does not contain the target collection slot corresponding to the target intent node, the jump intent information can be determined based on the obtained user input information. The session jump node can be determined based on the jump intent information, and the session processing is performed based on the collection slot information corresponding to the session jump node, thereby obtaining query result information. As can be seen, in this embodiment of the application, when the collection slot information corresponding to the target intent node does not include the target collection slot corresponding to the target intent node, that is, when the information input by the user for the conversation guidance information does not include the target collection slot corresponding to the target intent node, the jump intent information can be determined based on the user input information, and the conversation jump node can be determined based on the jump intent information. This realizes flexible jump of conversation nodes, so that conversation processing can be carried out based on the collection slot information corresponding to the conversation jump node, realizing flexible and intelligent speech jump, reducing the usage threshold of the conversation task graph, and improving the flexibility of the dialogue process.

[0104] Furthermore, in the embodiments of this application, when processing the conversation for the target collection slot, it can determine whether the target collection slot meets the termination dialogue configuration conditions of the target intent node. If the target collection slot contained in the collection slot information meets the termination dialogue configuration conditions of the target intent node, query result information is generated based on the conversation service information corresponding to the target collection slot. If the target collection slot contained in the collection slot information does not meet the termination dialogue configuration conditions of the target intent node, target conversation guidance information is further displayed based on the collection dialogue configuration information of the target collection slot and the target intent node. The conversation is then processed based on the target input information submitted by the user for the target conversation guidance information to obtain the final query result information. Optionally, in this embodiment of the application, the session processing based on the target collection slot to obtain the query result information may specifically include: if the target collection slot meets the termination script configuration conditions of the target intent node, then determine the session service information corresponding to the target collection slot, and generate the query result information based on the session service information; if the target collection slot does not meet the termination script configuration conditions of the target intent node, then display the target session guidance information based on the target collection slot and the collection script configuration information, and obtain the target input information submitted by the user for the target session guidance information; perform session processing based on the target input information to obtain the query result information.

[0105] In specific implementation, when the user input information contains all the slots required by the target intent node, as in the example above, if the user input information includes the departure address "B", destination address "A", and departure time "C", it can be determined that the target slots included in the collected slot information meet the termination dialogue configuration conditions of the target intent node. Subsequently, query result information can be generated based on the session service information corresponding to the target collected slots, such as flight information from B to A with a departure time of "C", and returned to the user as query result information, so that the user can obtain flight information from B to A with a departure time of "C". However, when the user input information does not contain all the slots required by the target intent node, as in the example above, if the user input information only contains the departure address "B" and destination address "A", it can be determined that the target slots included in the collected slot information do not meet the target intent node's termination dialogue configuration conditions. The graph node's termination dialogue configuration conditions are then set. Based on the target collection slots (i.e., origin address "B" and destination address "A") and the collection dialogue configuration information, the session guidance information "Please enter departure time" is displayed as the target session guidance information. This allows the user to input a departure time "C" in response to this guidance. The system then obtains the user's input time "C" as the target input information. Based on this target input information and the target collection slots, it can be determined that the user's input contains all the slots required by the target intent node. Therefore, the system can determine that the user's input meets the termination dialogue configuration conditions of the target intent node. Based on this target input information and the target collection slots, session processing is performed to return flight information from B to A with a departure time of "C" as the query result to the user, allowing them to obtain flight information from B to A with a departure time of "C".

[0106] When the user input does not contain all the slots required for the target intent node, such as when the user input is "buy train tickets", it can be determined that the user input triggered another node intent. In this case, the speech collection for the target intent node can be ended early. For example, it is not necessary to collect speech for the "buy plane tickets" intent node. The jump intent information can be determined directly based on the user input "buy train tickets". The conversation jump node can be determined as the "buy train tickets" node based on the jump intent information. Then, the conversation can be processed based on the collected slot information corresponding to the "buy train tickets" node.

[0107] As can be seen, the embodiments of this application can effectively improve the flexibility of the dialogue while ensuring the controllability of the task-oriented dialogue through node granularity control strategy and weakly associated dynamic graph model. Specifically, in this application embodiment, for each target intent node, the number of times the target intent node jumps to another session node can be recorded as the corresponding jump count for the other session node. If the recorded jump count exceeds a preset node jump count threshold, the other session node can be merged into the target intent node, achieving merging of the two nodes and realizing intent cohesion within the node. When a target intent node contains two or more session nodes, the jump counts corresponding to two of the session nodes can be recorded for that target intent node. If the jump count between two session nodes is less than the preset node jump count threshold (i.e., the two sub-session nodes within a target intent node do not frequently jump), such as exiting the current process directly after a sub-session node within the target intent node without jumping to the next sub-session node, the next sub-session node can be split from the target intent node, achieving node splitting. This ensures greater decoupling between nodes, allowing for flexible adjustment of node granularity and realizing a flexible task graph model for jumps.

[0108] In summary, after obtaining user query information, the embodiments of this application can determine the intention information of the predicted node based on the user query information. By using the intention information of the predicted node, the target intention node can be found in the preset dynamic graph. Thus, the collection slot information corresponding to the target intention node can be used for conversation processing to obtain the query result information corresponding to the user query information. This solves the problem caused by the requirement in the prior art to strictly follow the process of the task graph of the task-oriented dialogue system for dialogue.

[0109] Furthermore, the embodiments of this application can achieve a flexible task graph model by flexibly adjusting the node granularity and using a weakly correlated task graph model. This enables more flexible and controllable task-oriented dialogues, better generalizes open-domain dialogue and task-oriented dialogue models, and unifies the design of task-oriented and open-domain dialogues. This has a further impact on the unification of task graph models in the dialogue domain. Moreover, the flexible collection of configuration information by nodes can achieve cohesion and reuse of the dialogue process, reduce the complexity of task-oriented business script configuration, and maximize the flexibility of the dialogue while ensuring the controllability of the dialogue process through independent intent management of each node. This can greatly improve the user experience and satisfaction.

[0110] It should be noted that, for the sake of simplicity, the method embodiments are described as a series of actions. However, those skilled in the art should know that the embodiments of this application are not limited to the described order of actions, because according to the embodiments of this application, some steps may be performed in other orders or simultaneously.

[0111] This application also provides an information processing system based on dynamic graphs, such as... Figure 5 As shown, the dialogue information processing system 500 based on dynamic graphs provided in this application embodiment may include the following modules:

[0112] User inquiry module 510 is used to obtain user inquiry information;

[0113] The node intent module 520 is used to determine the predicted node intent information based on the user query information;

[0114] The node search module 530 is used to search for target intent nodes in a preset dynamic graph based on the predicted node intent information.

[0115] The session processing module 540 is used to perform session processing based on the collected slot information corresponding to the target intent node, and obtain the query result information corresponding to the user query information.

[0116] Optionally, the node intent module 520 includes the following sub-modules:

[0117] The node intent classification model submodule is used to input the user query information into a pre-trained node intent classification model to obtain the predicted node intent information output by the node intent classification model.

[0118] The node intent recognition submodule is used to perform intent recognition based on the question keywords in the user query information and obtain the intent recognition result.

[0119] The predicted node intent extraction submodule is used to extract the predicted node intent information corresponding to the first recognition score from the intent recognition result.

[0120] Optionally, the node lookup module 530 includes the following sub-modules:

[0121] The matching degree determination submodule is used to determine the matching degree between the intent information of the session node and the intent information of the predicted node for each session node in the dynamic graph.

[0122] The node sorting submodule is used to sort the nodes according to the matching degree to obtain the node sorting order.

[0123] The target intent node determination submodule is used to determine the session node corresponding to the first matching degree as the target intent node based on the node sorting order.

[0124] Optionally, the session processing module 540 includes the following sub-modules:

[0125] The conversation guidance submodule is used to collect the dialogue configuration information based on the target intent node and display the conversation guidance information;

[0126] The user information acquisition submodule is used to acquire user input information submitted in response to the session guidance information.

[0127] The word slot information collection and determination submodule is used to determine the user input information as the word slot information to be collected;

[0128] The collection slot information determination submodule is used to determine whether the collected slot information is the target collection slot corresponding to the target intent node;

[0129] The target slot conversation processing submodule is used to perform conversation processing based on the target collection slot when the collection slot contains the target collection slot corresponding to the target intent node, and obtain the query result information.

[0130] The jump intent information determination submodule is used to determine jump intent information based on the user input information when the collected slot information does not contain the target collected slot corresponding to the target intent node;

[0131] The session jump node determination submodule is used to determine the session jump node based on the jump intent information;

[0132] The node session processing submodule is used to perform session processing based on the collected slot information corresponding to the session jump node, and obtain the query result information.

[0133] Optionally, the session processing module 540 may also include the following modules:

[0134] The jump count recording module is used to record the number of jumps corresponding to the target intent node in the session jump;

[0135] The jump count determination module is used to determine whether the jump count exceeds a threshold;

[0136] The node merging module is used to merge the session jump node and the target intent node when the number of jumps exceeds a preset node jump count threshold, so as to obtain a merged session node.

[0137] The dynamic graph update module is used to update the dynamic graph based on the merged session node.

[0138] Optionally, the target slot conversation processing submodule may include the following units:

[0139] The termination dialogue judgment unit is used to determine whether the target collected word slot meets the termination dialogue configuration conditions of the target intent node;

[0140] The conversation service information unit is used to determine the conversation service information corresponding to the target collection slot when the target collection slot meets the termination dialogue configuration conditions of the target intent node, and to generate the query result information based on the conversation service information.

[0141] The conversation guidance information unit is used to display target conversation guidance information based on the target collection slot and the collection script configuration information when the target collection slot does not meet the termination script configuration conditions of the target intent node.

[0142] The target input information acquisition unit is used to acquire the target input information submitted by the user in response to the target session guidance information;

[0143] A session processing unit is used to perform session processing based on the target input information to obtain the query result information.

[0144] Optionally, information processing systems based on dynamic graphs also include:

[0145] The training node intent determination module is used to determine the intent information of the sample nodes as the intent information of the nodes to be trained.

[0146] The input sample acquisition module is used to acquire the input sample information corresponding to the sample node based on the intent information of the node to be trained.

[0147] The model training submodule is used to train the model based on the input sample information and the intent information of the node to be trained, so as to obtain the node intent classification model.

[0148] It should be noted that the information processing system based on dynamic graphs provided above can execute the information processing method based on dynamic graphs provided in any embodiment of this application, and has the corresponding functions and beneficial effects of the method.

[0149] Furthermore, embodiments of this application also provide an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor, when executing the program stored in the memory, implements the steps of the information processing method described in any of the above method embodiments.

[0150] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the information processing method steps described in any of the above method embodiments.

[0151] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For system, device, and storage medium embodiments, since they are basically similar to the method embodiments, the descriptions are relatively simple, and relevant parts can be referred to the descriptions of the method embodiments.

[0152] In this document, relational terms such as “first” and “second” are used merely to distinguish one entity or operation from another, without necessarily requiring or implying any such actual relationship or order between these entities or operations. Furthermore, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase “comprising one…” does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0153] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. An information processing method based on dynamic graphs, characterized in that, include: Obtain user inquiry information; Based on the user query information, determine the intention information of the predicted node; Based on the predicted node intent information, the target intent node is searched in a preset dynamic graph; Based on the collected slot information corresponding to the target intent node, the session is processed to obtain the query result information corresponding to the user query information; The step of performing session processing based on the collected slot information corresponding to the target intent node to obtain the query result information corresponding to the user query information includes: Based on the collected dialogue configuration information of the target intent node, the conversation guidance information is displayed; Obtain user input information submitted in response to the session guidance information; The user input information is identified as the collected slot information; If the collected slot information includes the target collected slot corresponding to the target intent node, then the session processing is performed based on the target collected slot to obtain the query result information; If the collected word slot information does not include the target collected word slot corresponding to the target intent node, then the jump intent information is determined based on the user input information; Based on the jump intent information, the session jump node is determined, and the session is processed based on the collection slot information corresponding to the session jump node to obtain the query result information; This also includes: For each target intent node, record the number of jumps corresponding to that session jump node; If the number of jumps exceeds a preset threshold for the number of node jumps, the session jump node and the target intent node are merged to obtain a merged session node. The dynamic graph is updated based on the merged session node; The step of performing conversation processing based on the target collected slots to obtain the query result information includes: If the target collection slot meets the termination dialogue configuration conditions of the target intent node, then the conversation service information corresponding to the target collection slot is determined, and the query result information is generated based on the conversation service information; If the target collection slot does not meet the termination dialogue configuration conditions of the target intent node, then based on the target collection slot and the collection dialogue configuration information, the target session guidance information is displayed, and the target input information submitted by the user for the target session guidance information is obtained. Based on the target input information, session processing is performed to obtain the query result information.

2. The information processing method according to claim 1, characterized in that, The step of determining the predicted node intent information based on the user query information includes: The user query information is input into a pre-trained node intent classification model to obtain the predicted node intent information output by the node intent classification model; or, Intent recognition is performed based on the question keywords in the user inquiry information to obtain intent recognition results, and the intent information of the predicted node corresponding to the first recognition score is extracted from the intent recognition results.

3. The information processing method according to claim 2, characterized in that, Before inputting the user query information into the pre-trained node intent classification model, the method further includes: The intent information of the sample nodes is determined as the intent information of the nodes to be trained; Based on the intent information of the node to be trained, obtain the input sample information corresponding to the sample node; The node intent classification model is obtained by training the model based on the input sample information and the intent information of the node to be trained.

4. The information processing method according to any one of claims 1 to 3, characterized in that, The step of searching for the target intent node in a preset dynamic graph based on the predicted node intent information includes: For each session node in the dynamic graph, determine the matching degree between the intent information of the session node and the intent information of the predicted node; The nodes are sorted according to the matching degree to obtain the node sorting order; Based on the node sorting order, the session node corresponding to the first matching degree is determined as the target intent node.

5. An information processing system based on dynamic graphs, characterized in that, include: The user inquiry module is used to obtain user inquiry information; The node intent module is used to determine the predicted node intent information based on the user query information. The node search module is used to search for target intent nodes in a preset dynamic graph based on the predicted node intent information. The session processing module is used to perform session processing based on the collected slot information corresponding to the target intent node, and obtain the query result information corresponding to the user query information; The step of performing session processing based on the collected slot information corresponding to the target intent node to obtain the query result information corresponding to the user query information includes: Based on the collected dialogue configuration information of the target intent node, the conversation guidance information is displayed; Obtain user input information submitted in response to the session guidance information; The user input information is identified as the collected slot information; If the collected slot information includes the target collected slot corresponding to the target intent node, then the session processing is performed based on the target collected slot to obtain the query result information; If the collected word slot information does not include the target collected word slot corresponding to the target intent node, then the jump intent information is determined based on the user input information; Based on the jump intent information, the session jump node is determined, and the session is processed based on the collection slot information corresponding to the session jump node to obtain the query result information; This also includes: For each target intent node, record the number of jumps corresponding to that session jump node; If the number of jumps exceeds a preset threshold for the number of node jumps, the session jump node and the target intent node are merged to obtain a merged session node. The dynamic graph is updated based on the merged session node; The step of performing conversation processing based on the target collected slots to obtain the query result information includes: If the target collection slot meets the termination dialogue configuration conditions of the target intent node, then the conversation service information corresponding to the target collection slot is determined, and the query result information is generated based on the conversation service information; If the target collection slot does not meet the termination dialogue configuration conditions of the target intent node, then based on the target collection slot and the collection dialogue configuration information, the target session guidance information is displayed, and the target input information submitted by the user for the target session guidance information is obtained. Based on the target input information, session processing is performed to obtain the query result information.

6. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the steps of the information processing method based on dynamic graphs as described in any one of claims 1-4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the information processing method based on dynamic graphs as described in any one of claims 1-4.

Images