CN-121981683-A - Travel business service method, system, medium and product based on AI intelligent agent

CN121981683ACN 121981683 ACN121981683 ACN 121981683ACN-121981683-A

Abstract

A travel service method, system, medium and product based on AI intelligent agent relates to big data resource service field, the method includes obtaining multi-source heterogeneous data and mapping to travel field knowledge graph, generating dynamic knowledge graph and inputting to intention prediction model, obtaining clue heat value and stage stay probability of target customer to determine follow-up strategy instruction of target customer, searching nodes meeting preset constraint condition in dynamic knowledge graph, obtaining candidate resource set and inputting to big language model to make text construction, generating initial travel text, executing inventory interface check and locking to initial travel text, generating travel scheme to be confirmed, pushing travel scheme to be confirmed to client terminal according to corresponding touch channel matched with stage stay probability, and updating operation configuration parameters based on interactive feedback data. By implementing the method, the whole process from clue intention identification and dynamic journey to resource locking can be optimized, and the conversion efficiency is improved.

Inventors

Tian Dalong
LI KUNYU
Luo Huaguang

Assignees

四川省乐途智行科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (10)

1. An AI agent-based travel service method, characterized by being applied to an intelligent decision-making system, comprising: The method comprises the steps of obtaining multi-source heterogeneous data containing client interactive behaviors, mapping the client interactive behaviors into preset knowledge patterns in the travel field, and generating dynamic knowledge patterns containing real-time interactive features; Inputting the dynamic knowledge graph to a preset intention prediction model for characteristic reasoning to obtain a clue heat value and a stage stay probability of a target client; Determining a follow-up strategy instruction of the target client according to the cue heat value and a preset response threshold; responding to the follow-up strategy instruction, and calling a constraint satisfaction solver to search nodes meeting preset constraint conditions in the dynamic knowledge graph to obtain a candidate resource set; Inputting the candidate resource set into a large language model for text construction, and generating an initial travel text containing graphic materials; performing inventory interface check and locking operation on the resource elements in the initial travel text, and generating a travel scheme to be confirmed carrying a resource locking identifier; According to the stage stay probability, matching a corresponding touch channel, and pushing the stroke scheme to be confirmed to a client terminal associated with the target client through the touch channel; And receiving interactive feedback data returned by the client terminal, and updating operation configuration parameters of the intelligent decision system based on the interactive feedback data.
2. The method according to claim 1, wherein the step of obtaining multi-source heterogeneous data including client interaction behavior, mapping the client interaction behavior into a preset knowledge-graph of travel domain, and generating a dynamic knowledge-graph including real-time interaction features, specifically includes: Identifying the association relation between the equipment source identifiers and the accounts in the multi-source heterogeneous data, and mapping the interaction behaviors belonging to different source identifiers to independent sub-graph areas of the knowledge graph respectively; performing a graph embedding operation on each independent sub-graph region, generating a plurality of independent intent feature vectors characterizing different decision body preferences; When cosine similarity among the independent intention feature vectors is lower than a preset similarity threshold, marking the current state as a multi-main-body conflict state, outputting a conflict state set containing the independent intention feature vectors, and constructing a dynamic knowledge graph based on the conflict state set.
3. The method according to claim 2, wherein the step of inputting the dynamic knowledge graph to a preset intention prediction model to perform feature reasoning to obtain a clue heat value and a stage stay probability of a target client specifically includes: extracting the conflict state set from the attribute layer of the dynamic knowledge graph, inputting the conflict state set into a multi-head attention mechanism network, and calculating the weight coefficient of each independent intention feature vector in the conflict state set in a global decision based on the self-attention layer of the multi-head attention mechanism network; Performing weighted fusion on the plurality of independent intention feature vectors based on the weight coefficients to generate a fusion feature vector representing a multi-party compromise result; And mapping the fusion feature vector to a preset conversion probability space to obtain a clue heat value and a stage stay probability for representing the multi-main body common decision tendency.
4. The method according to claim 1, wherein the step of calling a constraint satisfaction solver to retrieve nodes satisfying a preset constraint condition in the dynamic knowledge graph in response to the follow-up policy instruction to obtain a candidate resource set specifically includes: Calculating an information entropy value of a currently activated node in the dynamic knowledge graph, and when the information entropy value indicates a discrete distribution state, searching a node with the maximum attribute variance in the dynamic knowledge graph as a probe node, generating an interaction option instruction containing the probe node attribute, and sending the interaction option instruction to a client terminal; receiving selection data returned by the client terminal and corresponding to the interaction option instruction, and executing pruning on non-selected branches in the dynamic knowledge graph based on the selection data to obtain an updated knowledge graph; And executing constraint satisfaction retrieval in the updated knowledge graph to obtain a candidate resource set.
5. The method of claim 1, wherein after the step of pushing the to-be-validated trip scenario to the client terminal associated with the target client through the touchdown channel according to the phase dwell probability matching the corresponding touchdown channel, the method further comprises: monitoring a read receipt signal of the client terminal for the travel scheme to be confirmed through the touch channel; Responding to the reading receipt signal, and sending a latch live time extension instruction corresponding to the resource locking identifier in the trip scheme to be confirmed to an external inventory system; and when the reading receipt signal is not received within a preset time window, sending a release instruction corresponding to the resource locking identifier to the external inventory system so as to release the locked resource inventory.
6. The method of claim 5, wherein after the step of sending a latch liveness extension instruction to an external inventory system corresponding to the resource lock identification in the trip scenario to be validated in response to the read response piece signal, the method further comprises: Acquiring semantic anchor point characteristics of an initial travel text in the travel scheme to be confirmed, and calculating a real-time holding cost coefficient of a currently locked resource inventory; when the real-time holding cost coefficient exceeds a preset risk threshold, intercepting the latch live-time extension instruction, and searching for a replacement resource node meeting homomorphic mapping relation with the semantic anchor point characteristic in the dynamic knowledge graph; When the semantic consistency score between the replacement resource node and the initial travel text is higher than a preset non-inductive replacement threshold value, sending a release instruction corresponding to the current locked resource to the external inventory system, and synchronously sending a locking instruction corresponding to the replacement resource node; And updating the bottom resource mapping relation of the travel scheme to be confirmed, and keeping the image-text materials displayed by the client terminal unchanged.
7. The method according to claim 1, wherein the step of receiving interactive feedback data returned by the client terminal and updating the operation configuration parameters of the intelligent decision system based on the interactive feedback data specifically comprises: Receiving interactive feedback data returned by the client terminal, extracting negative semantic features in the interactive feedback data, and calculating Euclidean distance between the negative semantic features and feature vectors of the current highest-weight activated nodes in the dynamic knowledge graph; When the Euclidean distance exceeds a preset drift threshold value, setting a state vector of a long-term memory unit of a long-term memory network in the intention prediction model to zero so as to cut off a characteristic transfer path of the client interaction behavior to current reasoning; and injecting the negative semantic features into the intent prediction model as new initial input vectors, and reinitializing the node activation weights of the dynamic knowledge graph.
8. An intelligent decision system comprising one or more processors and memory coupled to the one or more processors, the memory to store computer program code comprising computer instructions that the one or more processors invoke to cause the intelligent decision system to perform the method of any of claims 1-7.
9. A computer readable storage medium comprising instructions which, when run on an intelligent decision system, cause the intelligent decision system to perform the method of any of claims 1-7.
10. A computer program product, characterized in that the computer program product, when run on an intelligent decision system, causes the intelligent decision system to perform the method according to any of claims 1-7.

Description

Travel business service method, system, medium and product based on AI intelligent agent Technical Field The application relates to the field of big data resource service, in particular to a travel business service method, a travel business service system, a travel business service medium and a travel business service product based on an AI intelligent agent. Background With the digital transformation of the travel industry, travel agencies, online travel platforms (OTAs), and destination operators are faced with massive customer consultation and clue processing requirements. In the current business process, enterprises generally need to process all-link business from customer acquisition, demand communication and scheme formulation to final subscription payment, and potential consultation clues are converted into actual travel orders through efficient service, so that the revenue and market competitiveness of the enterprises are improved. In the related art, travel business processes typically employ a "CRM System+live advisor+static template" mode of operation. The CRM system is responsible for recording basic relation ways and preliminary source channels of clients as a basic database of a clue pool, after obtaining clues, a travel advisor performs preliminary screening on the clients according to manual experience, and searches matched products from a preset fixed package library or Excel quotation by combining destinations and date requirements of the clients, the advisor manually inquires the current state of the inventory system, fills product information into a fixed document template, generates a travel form or a static PDF form, sends the document to the clients through telephone or instant messaging software, and then performs subsequent communication and ordering. However, the related art relies on the thread screening and follow-up method of manual experience, response delay is easy to occur when high concurrent threads (such as exploded products in a promotion period) are faced, and the loss rate of high quality threads is high due to lack of a policy dynamic adjustment mechanism, so that the overall conversion efficiency is difficult to further improve. Disclosure of Invention The application provides a travel business service method, a system, a medium and a product based on an AI intelligent agent, which are used for realizing the overall process optimization from clue intention identification and dynamic trip co-creation to resource locking and improving the overall conversion efficiency. The application provides a travel service method based on an AI intelligent agent, which is applied to an intelligent decision system and comprises the steps of obtaining multi-source heterogeneous data containing client interaction behaviors, mapping the client interaction behaviors to preset travel field knowledge patterns to generate dynamic knowledge patterns containing real-time interaction features, inputting the dynamic knowledge patterns to preset intention prediction models to conduct feature reasoning to obtain cue heat values and stage stay probabilities of target clients, determining follow-up strategy instructions of the target clients according to the cue heat values and preset response thresholds, calling nodes meeting preset constraint conditions in the dynamic knowledge patterns to obtain candidate resource sets in response to the follow-up strategy instructions, inputting the candidate resource sets to a large language model to conduct text construction to generate initial travel texts containing graphics and text, executing inventory interface check and locking operation on resource elements in the initial travel texts to generate a travel scheme to be confirmed carrying resource locking identifiers, pushing the travel scheme to be confirmed to client terminals associated with the target clients through the touch channels, receiving the client terminals to interact with the client terminals, and updating the intelligent decision system based on feedback operation parameters. In the embodiment, the intelligent decision system integrates the dynamic interaction behavior of the client into the decision process in real time by constructing a full-flow processing link from data acquisition, intention prediction, strategy generation, scheme construction and resource locking to closed-loop feedback, thereby realizing accurate capture of the intention of the client and dynamic adjustment of the service strategy, replacing the traditional low-efficiency follow-up mode depending on manual experience, and improving the conversion efficiency of clues to orders and the response speed of the system while ensuring the service quality. In combination with some embodiments of the first aspect, in some embodiments, the step of obtaining multi-source heterogeneous data including client interaction behavior, mapping the client interaction behavior to a preset travel domain knowledge graph, and generating a