Business travel scene user demand prediction and service recommendation method based on AI

Abstract

The invention provides an AI-based business travel scene user demand prediction and service recommendation method, and relates to the technical field of AI recommendation, and the method comprises the steps: obtaining a user time sequence workflow and enterprise business travel historical data, and constructing a multi-modal data set; a workflow continuity loss coefficient WCLI model is constructed by using a two-stage hybrid strategy training model and a user dynamic knowledge graph, and potential impact of travel on working efficiency is quantitatively evaluated; according to the WCLI risk level, a differentiated intelligent recommendation scheme is generated, and workflow compensation and relief suggestions are provided for the high-risk travel; workflow compensation services such as schedule adjustment bound with the selected scheme are automatically executed; outputting a comprehensive decision support report and returning user feedback to the model; according to the method, the crossing from passive response to active perception and from isolated event optimization to global workflow continuity guarantee is realized, and the interruption risk of travel to the workflow is effectively reduced while the business travel arrangement efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of AI recommendation technology, and in particular to an AI-based method for predicting user demand and recommending services in business travel scenarios. Background Technology

[0002] In the business travel service sector, the iteration of information technology has always revolved around improving booking efficiency and service personalization. The industry has taken the lead in realizing online real-time retrieval and transactions of business travel services such as air tickets, hotels, and car rentals by building a full-category product database and transaction engine. The subsequent emergence of price comparison tools and rule-based recommendation functions has further helped users quickly filter options that meet preset conditions such as price, time, and location, laying the core foundation for modern online business travel services. However, as business activities become increasingly complex, users' needs for business travel services have shifted from simple information queries and transaction completion to an auxiliary experience that can be deeply adapted to workflows and intelligently adapted throughout the entire process. The limitations of existing technical solutions are becoming increasingly apparent. First, there is the passive and reactive service model. Currently, mainstream business travel platforms all take users' proactive initiation of queries as the starting point for service. Even with the introduction of preference recommendations based on historical orders, the essence is still a static summary of past behavior, lacking the ability to proactively predict users' future travel needs. The system can record and prioritize users' frequently selected hotels and flights, but it cannot anticipate unintended booking needs by incorporating users' latest work schedules, project milestones, and changes in corporate travel policies. Furthermore, it cannot dynamically generate travel plans suitable for all scenarios, requiring users to invest significant time in itinerary planning and comparisons. The intelligence remains superficial. Secondly, existing technologies lack depth and breadth in understanding user needs. Most recommendation models rely solely on structured historical transaction data, failing to capture the complex and dynamic unstructured background factors behind business travel, let alone interpret the deeper travel intentions hidden behind choices. Moreover, business travel is not an isolated event but a crucial link embedded in a user's continuous workflow. The purpose of a single trip, accompanying persons, and subsequent meeting arrangements collectively constitute the complete context of the decision-making process. However, existing technologies often view each booking behavior in isolation, failing to determine the priority and constraints of user decisions based on continuous context, making true precision difficult to achieve. Furthermore, existing service recommendations are limited in scope and cannot achieve cross-scenario resource integration and global optimization. Traditional recommendation engines operate with independent modules for flights, hotels, and transportation, resulting in recommendations that are often simple combinations of locally optimal solutions rather than globally optimal solutions based on the user's overall travel experience. Users are forced to manually coordinate information across multiple stages and repeatedly switch between platforms for verification. Focusing solely on service optimization during the "travel" phase can overlook the disruption business travel causes to the continuity of work before and after travel, as well as the resulting hidden efficiency losses. While recommending low-priced red-eye flights may achieve the lowest possible travel cost, it could lead to insufficient rest, impacting the user's work efficiency in key meetings the following day, requiring additional time to handle backlogged work—a hidden cost far exceeding the saved ticket price. Therefore, there is an urgent need for a novel AI-based method for predicting user demand and recommending services for business travel scenarios. Summary of the Invention

[0003] The purpose of this invention is to provide an AI-based method for predicting user demand and recommending services in business travel scenarios. This addresses the problems of existing traditional business travel service platforms, which can only passively process explicit travel requests and cannot proactively predict potential itineraries embedded in continuous workflows. Furthermore, these platforms struggle to identify deep-seated travel intentions and risks of work continuity disruption in complex business contexts, and lack effective perception and predictive mitigation measures for the hidden losses and recovery needs of work efficiency during non-travel periods caused by travel. The specific technical solution is as follows: This invention provides an AI-based method for predicting user demand and recommending services in business travel scenarios, including: S10: Obtain user's multi-dimensional and time-series native workflow data as model input data, combine it with de-identified corporate travel history data, and collect and construct a multimodal training and inference dataset covering user behavior, schedule, collaboration network and external environment according to preset data association rules. S20 sets up a two-stage hybrid training strategy, reads samples from the dataset, performs offline pre-training and online fine-tuning of the core prediction model, and obtains a prediction model that can accurately output potential travel demand and its characteristic parameters. S30 deploys the predictive model on a real-time data processing platform, connecting enterprise scheduling systems, travel management systems, and third-party data sources through standardized interfaces to build a dynamic knowledge graph of users and a real-time computing engine, enabling continuous perception and analysis of travel intentions embedded in workflows. S40, construct a quantitative assessment model for the impact of workflow continuity, and use the correlation features extracted from the knowledge graph and the output of the prediction model for comprehensive evaluation. Combine the core work conflict degree, collaboration network interruption degree and recovery load prediction amount related to the process to form a quantifiable workflow continuity loss coefficient WCLI, and conduct risk assessment and solution pre-drills in simulated environments and real-world scenarios. S50 uses differentiated intelligent recommendation and negotiation strategies to generate candidate travel service solutions based on the different risk ranges of the WCLI value, and automatically generates workflow compensation and mitigation suggestions for high-risk trips.

[0004] Furthermore, the method also includes: S60, executing automated workflow compensation services tied to the end-user's selected scheme, including automatically adjusting schedules, generating work handover summaries, and enabling a post-return-to-work focus mode to mitigate the hidden losses in work efficiency during non-travel periods caused by travel. S70 outputs a comprehensive decision support report that includes the basis for demand forecasting, risk assessment process, recommended solutions, and implemented compensation measures, and feeds user feedback back into the model training and knowledge graph update process.

[0005] Further, step S10 includes: S101 synchronizes structured calendar event data for a future preset time period from the user's enterprise office system through an authorized interface. The data fields include at least: event title, start and end timestamps, location, list of participants, event category tag, and project ID. It also synchronizes historical order data for a past preset time period from the user's enterprise travel management platform. The fields include at least: travel purpose description, departure / return time, origin and destination city pair, transportation and accommodation service provider and product code, and order creation and payment time. S102 collects user work context data through a security agent and privacy computing middleware, including: performing natural language processing on the subject lines of emails sent and received by the user within the most recent preset time period and the body of meeting invitations to extract high-frequency project names, customer names, and product terms as keywords; obtaining the current status, last update time, and next milestone date of the projects the user is responsible for in the collaboration tool; and obtaining a list of the user's direct team members and frequently used collaborating colleagues from the enterprise HR system.

[0006] Further, step S20 includes: S201, Set up the offline pre-training stage: Use the records of completed trips in the historical dataset as positive samples, use all relevant schedules, emails, and project status data within a preset number of days before the trip date as input feature sequences, use the trip order details as labels, and train a temporal Transformer network as the basic prediction model; S202, Set up an online incremental learning phase: After the system is deployed, all user interactions with the system's predictions and recommendations, as well as their subsequent actual travel results, are used as incremental training samples with reward signals. The basic prediction model is fine-tuned every time a preset number of new samples are accumulated or every preset number of days. Step S30 includes: S301, deploy the prediction model obtained in S20 as a real-time microservice, which runs automatically at a preset period, with the input being the user's latest calendar event sequence, the most recent work context snapshot, and the knowledge graph feature vector that is updated in real time; S302 correlates the model's output predictions with real-time flight and hotel inventory and pricing information queried from the global distribution system. S303 builds and maintains a user's personal dynamic knowledge graph based on graph database technology. The graph nodes include users, cities, airports, contacts, projects, and companies. The edge relationships include "expected destination", "historically visited", "project-related locations", and "close collaboration". The graph is updated once at a preset period.

[0007] Further, in step S40, the workflow continuity loss coefficient WCLI is calculated by weighted summation of core workflow conflict degree, collaborative network interruption degree, and recovery load prediction; wherein, The core job conflict level is used to quantify the severity of time overlap between the predicted travel time window and key scheduled events already existing in the user's calendar. The Collaboration Network Disruption Scale is used to quantify the risk of collaborative work disruption to a user's team or closely collaborative network due to user departure during the predicted travel period. The recovery load forecast is used to quantify the additional workload that is expected to be required to process backlogged transactions after users return from business trips. Based on the Workflow Continuity Loss Factor (WCLI) value, the defined risk level thresholds include: When WCLI is less than the first threshold, it is judged as a low workflow continuity risk; When WCLI is greater than or equal to the first threshold and less than the second threshold, it is judged as a medium workflow continuity risk; When WCLI is greater than or equal to the second threshold, it is judged as a high workflow continuity risk.

[0008] Further, step S50 includes: If WCLI is less than the first threshold, the proactive seamless recommendation mode is adopted: with the shortest total travel time and the lowest travel pressure index as multiple objectives, the optimization algorithm is used to solve the optimal multiple integrated packages under the premise of complying with the company's travel policy, and automatically generate the fine-tuning suggestions of the schedule that contributes the most to WCLI. If WCLI is greater than or equal to the first threshold, the risk warning and negotiation mode is activated: a high-risk alert is pushed to the user, and alternative solutions including local alternatives, trip splitting solutions, and pre-compensation solutions are provided. After the user clearly confirms their needs, a limited number of travel options are provided.

[0009] Further, step S60 includes: Once a user confirms and completes their travel booking, the system automatically triggers a preset compensation rule executor. The rules include: if the trip is a long-haul international flight, the system will automatically adjust the status of the meeting in the user's corporate calendar to a preset category on the day the user arrives in the destination city and the first day after returning to their place of residence. At a preset time one day before the user returns, key update summaries are automatically extracted from their project management system and email system, and a work briefing during their absence is generated and sent to their email address. The Do Not Disturb mode on the instant messaging software will be automatically activated during a preset time period on the first business day after the user returns.

[0010] Furthermore, step S60 includes, after the service loop is closed, the system automatically generates a comprehensive decision support report. The report includes demand forecast tracing: showing the core schedule events that triggered this forecast, project status, and collaboration relationships; risk assessment log: detailing the calculation process and values ​​of each WCLI item; recommended solution comparison: all previously provided solutions and their key parameters; user decision and feedback: the user's final choice and evaluation of the recommendation; executed compensation measures: a list of all workflow compensation operations automatically completed by the system; this report can be pushed to users and their managers for travel efficiency analysis; at the same time, user feedback is fed back into the model training and knowledge graph update process; the service loop includes two scenarios: the end of the trip or the user cancels the forecast demand.

[0011] Furthermore, the method also includes acquiring multi-source data and constructing a user workflow hypergraph model. Nodes in the workflow hypergraph represent specific work task units, and hyperedges represent a complete workflow, connecting all task unit nodes to which it belongs. Nodes are labeled with estimated standard time consumption, execution location preference, core collaborator set, and spatiotemporal flexibility attributes. The method also includes joint intent identification and impact deduction based on a hybrid expert network, which includes: The gated network is used to receive the state feature vector of the current workflow hypergraph and output a weight distribution vector to dynamically allocate the contribution of downstream expert sub-networks. The travel demand triggering expert is used to identify task nodes in the workflow hypergraph whose spatial elasticity coefficient Se is lower than a preset threshold and whose core collaborator set includes external entities. It determines the inevitability and urgency of triggering offline travel demand and outputs a set of potential travel events. Workflow resilience assessment expert is used to assess the impact and resilience of the original workflow hypergraph if a user leaves during the time window for each potential travel event. A joint strategy generation expert is used to synthesize the outputs of travel demand triggering experts and workflow resilience assessment experts to proactively generate a preliminary set of adaptive workflow adjustment strategies. The method also includes performing dynamic conjugate optimization to generate a business travel-workflow joint action plan. It defines conjugate travel variable sets X and workflow adjustment variable sets Y, and constructs an optimization problem aimed at minimizing the global disturbance index GPI. The global disturbance index GPI is a comprehensive indicator calculated by weighting and summing the total monetary cost of travel, total user travel time, total delay days on the workflow critical path, and the estimated number of new coordination sessions initiated due to adjustments. The optimization problem is iteratively solved using the alternating direction multiplier method until the convergence condition is met, and the optimal joint action plan is output. The method also includes path dependency strength analysis and comprehensive risk rating, including calculating the path dependency strength index PDS. The PDS is calculated based on the change in betweenness centrality of each node in the workflow hypergraph before and after the adjustment, and the importance weight of the task of that node, in the set of task nodes that have changed due to workflow adjustment. The risk level of the joint action plan is jointly determined by the global disturbance index GPI, the path dependency strength index PDS, and the scheme confidence, and projected into a three-dimensional decision space for regional division. The method also includes initiating joint review and dynamic revision of human-machine collaboration, showing users a complete interactive view of the joint action plan that includes travel options, workflow adjustment comparisons, risk analysis and collaboration status, and restarting local optimizations in response to users' modification commands.

[0012] Furthermore, the method also includes performing automated collaboration and workflow injection, automatically executing travel bookings after the user approves the joint action plan, automatically executing approved schedule changes, meeting mode conversions and task reassignment operations through the API of enterprise collaboration tools, sending change notifications, and inserting buffer and resume task events into the user's schedule according to the plan; The method also includes generating joint action source tracing reports and model evolution. The reports detail the optimization decision-making process and use user feedback and actual implementation data of the plan as reinforcement learning signals to feed back to the hybrid expert network of S2.

[0013] A computer device includes a processor and a memory, the memory storing instructions executable by the processor, characterized in that the instructions, when executed by the processor, implement the method described above.

[0014] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the method.

[0015] The beneficial effects of this invention are as follows: By deeply integrating the user's work context for proactive intent prediction, it transforms from traditional passive response to proactive perception; it quantitatively assesses the implicit impact of travel on work continuity, transforming ambiguous impacts into calculable workflow continuity loss coefficients, thus achieving risk management from isolated events to the global workflow; it further provides intelligent solutions that include pre-event risk mitigation and post-event automated compensation, such as automatically adjusting schedules, generating handover summaries, and enabling focus mode, substantially repairing the efficiency losses caused by travel; simultaneously, it adopts a hybrid strategy combining offline pre-training and online fine-tuning, enabling the model to continuously and personally adapt to the evolution of user behavior, improving prediction accuracy; finally, through the differentiated application of intelligent recommendation and negotiation strategies, it achieves human-machine collaborative decision-making, effectively preventing workflow interruption risks while improving the overall efficiency of business travel.

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of an AI-based method for predicting user demand and recommending services in a business travel scenario, according to Embodiment 1 of this application. Figure 2 This is a scatter plot showing the correlation between the WCLI value and the user's actual work impact score for an AI-based method for predicting user demand and recommending services in a business travel scenario, as described in Embodiment 1 of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1 like Figure 1 As shown in the figure, this invention proposes an AI-based method for predicting user demand and recommending services in business travel scenarios, including the following steps: S10: Obtain user's multi-dimensional and time-series native workflow data as model input data, combine it with de-identified corporate travel history data, and collect and construct a multimodal training and inference dataset covering user behavior, schedule, collaboration network and external environment according to preset data association rules. S20, set up a two-stage hybrid training strategy, read samples from the dataset, perform offline pre-training and online fine-tuning of the core prediction model, and obtain a prediction model that can accurately output potential travel demand and its characteristic parameters. S30, the prediction model is deployed on a real-time data processing platform, and the enterprise schedule system, travel management system and third-party data sources are connected through standardized interfaces to build a user's personal dynamic knowledge graph and real-time computing engine, so as to realize the continuous perception and analysis of travel intentions embedded in the workflow. S40, construct a quantitative assessment model for the impact of workflow continuity, and use the correlation features extracted from the knowledge graph and the output of the prediction model for comprehensive evaluation. Combine the core work conflict degree, collaboration network interruption degree and recovery load prediction amount related to the process to form a quantifiable workflow continuity loss coefficient WCLI, and conduct risk assessment and solution pre-drills in simulated environment and real scenario. S50, based on the different risk ranges of the WCLI value, adopt differentiated intelligent recommendation and negotiation strategies to generate candidate travel service solutions, and automatically generate workflow compensation and mitigation suggestions for high-risk trips; S60 executes automated workflow compensation services tied to end-user selected options, including automatically adjusting schedules, generating work handover summaries, and enabling a post-return-to-work focus mode to mitigate the hidden impact of travel on work efficiency during non-travel periods. S70 outputs a comprehensive decision support report that includes the basis for demand forecasting, risk assessment process, recommended solutions, and implemented compensation measures, and feeds user feedback back into the model training and knowledge graph update process.

[0020] The principle and effect of the above technical solution are as follows: This invention predicts from passive response to proactive perception, quantifies risks from isolated events to global workflows, and delivers comprehensive services including workflow repair from single travel plans. By deeply integrating the work context for intent prediction, it quantitatively assesses the hidden impact of travel on work continuity, and provides an intelligent solution that includes pre-event risk mitigation and post-event automated compensation. This overcomes the shortcomings of focusing only on travel itself while ignoring its external efficiency losses, improves the overall efficiency of business travel, and prevents the risk of workflow interruption.

[0021] In one embodiment, S10 includes: S101, through an authorized interface, synchronizes structured calendar event data for the next 90 days from the user's enterprise office system (such as Microsoft 365, Google Workspace) every 24 hours. The data fields include at least: event title, start and end timestamps (accurate to the minute), location (parsed into city name and specific address), list of participants, event category tag, and project ID; it also synchronizes the user's historical order data for the past 24 months from the enterprise travel management platform. The fields include at least: travel purpose description, departure / return time, origin and destination city pair, transportation and accommodation service provider and product code, and order creation and payment time. S102 collects user work context data through a secure agent and privacy computing middleware, including: performing natural language processing on the subject lines of emails sent and received by the user in the last 30 days and the body of meeting invitations to extract high-frequency project names, customer names, and product terms as keywords; obtaining the current status (e.g., "in progress / pending review"), last update time, and next milestone date of the projects the user is responsible for in collaboration tools (such as Jira, Asana); and obtaining a list of the user's direct team members and frequently used collaborating colleagues from the enterprise HR system.

[0022] The principle and effect of the above technical solution are as follows: By systematically acquiring and associating schedule, historical behavior, work content and collaboration data, a digital twin foundation for depicting the user's complete workflow is constructed, providing comprehensive and dynamic data support for subsequent in-depth understanding of the context of travel intentions, and overcoming the shortcomings of traditional methods such as single data source and information silos.

[0023] In one embodiment, S20 includes: S201, Set up the offline pre-training stage: Use records of completed trips in the historical dataset as positive samples, use all relevant schedules, emails, and project status data within 7 days before the trip date as input feature sequences, use trip order details as labels, and train a temporal Transformer network as the basic prediction model; This model learns to identify features that will trigger business trips from complex workflow sequence patterns; S202, Set up an online incremental learning phase: After the system is deployed, all user interactions with the system's predictions and recommendations (including acceptance, modification, or ignoring) and their subsequent actual travel results are used as incremental training samples with reward signals. Every 1,000 new samples or every 7 days, the basic prediction model is fine-tuned to continuously adapt to changes in user behavior patterns.

[0024] The principle and effect of the above technical solution are as follows: by adopting a training strategy that combines offline and online methods, a benchmark model with strong generalization ability is trained using large-scale historical data, and the model can be personalized to adapt to the evolution of user habits through continuous online learning, which ensures that the prediction accuracy improves with the use of time and solves the problem that traditional static models cannot adapt to individual changes.

[0025] In one embodiment, S30 includes: S301 deploys the prediction model obtained in S20 as a real-time microservice, which runs automatically every 4 hours. The input is the user's latest calendar event sequence (time window is the next 30 days), the most recent work context snapshot, and the knowledge graph feature vector that is updated in real time. S302, The prediction results output by the model, including destination city, travel purpose classification, confidence level, and suggested time window, are correlated with flight and hotel inventory and price information queried in real time from the global distribution system; S303 is a graph database technology-based system that builds and maintains a user’s personal dynamic knowledge graph. The graph nodes include users, cities, airports, contacts, projects, and companies. The edge relationships include “expected destination” (derived from model predictions), “historically frequented”, “project-related locations”, and “close collaboration” (calculated based on meeting co-occurrence frequency and email exchange density, with a threshold of ≥3 interactions per week). The graph is updated every 12 hours.

[0026] The principle and effect of the above technical solution are as follows: it realizes closed-loop real-time processing from data to decision-making, and connects discrete data points into a rich semantic network through dynamic knowledge graphs, enabling the system to understand deep logic such as "why go there" and "whose work is it related to", providing a relational reasoning basis for subsequent risk assessment and realizing contextual awareness of potential needs.

[0027] In one embodiment, S40 includes: constructing a workflow continuity loss coefficient calculation model, the calculation formula of which is: WCLI represents the workflow continuity loss coefficient, a dimensionless quantitative indicator used to comprehensively assess the total degree of potential disruption and efficiency loss caused by a predicted or planned business trip to a user's established workflow; α represents the first weighting coefficient assigned to the core work conflict degree C_c, a preset constant used to adjust the contribution ratio of C_c in the total loss coefficient WCLI; C_c represents the core work conflict degree, a dimensionless ratio ranging from [0, 1], used to quantify the severity of time overlap between the predicted travel time window and existing key schedule events in the user's calendar that are marked as high priority or must be attended by the user, specifically calculated as: the total length of conflict time divided by the total length of the core time of the predicted trip; β represents the second weighting coefficient assigned to the collaboration network interruption degree I_n, a preset constant used to adjust the contribution ratio of I_n in the total loss coefficient WCLI; I_n represents the collaboration network interruption degree, a value ranging from [0, 1]... γ is a dimensionless value between [0, 1] used to quantify the risk of collaborative work disruption to a user's team or closely collaborative network due to the user's departure during the predicted travel period. This value is based on the "close collaboration" relationship defined in the user's personal knowledge graph, and is calculated by counting the number of pending tasks that depend on the user's input or decision during the travel period, and then normalized. γ represents the third weighting coefficient assigned to the recovery load forecast R_l, which is a preset constant used to adjust the contribution ratio of R_l in the total loss coefficient WCLI. R_l represents the recovery load forecast, a dimensionless ratio between [0, 1] used to quantify the additional workload expected to be invested by the user after returning from business travel to handle backlogged tasks and reintegrate into the work rhythm. The specific calculation method is: the predicted additional processing hours required on the first working day after returning to work divided by the working hours of a standard working day (e.g., 8 hours).

[0028] Based on the value of the workflow continuity loss coefficient WCLI, the following risk level thresholds are defined: When WCLI < 0.3, it is judged as a low workflow continuity risk.

[0029] When 0.3 ≤ WCLI < 0.7, it is judged as a medium workflow continuity risk.

[0030] When WCLI ≥ 0.7, it is considered a high workflow continuity risk.

[0031] The principle and effect of the above technical solution are as follows: a comprehensive index WCLI was created to quantitatively assess the impact of travel on work continuity. The originally vague "whether the impact is large or small" was transformed into a calculable value. Through the scientific weighting of three dimensions, namely core work conflict degree, collaboration network interruption degree, and recovery load forecast, the potential impact of travel behavior on work efficiency was accurately characterized, providing a precise decision-making basis for subsequent differentiated response strategies.

[0032] In one embodiment, S50 includes: if WCLI < 0.3, adopting a proactive seamless recommendation mode: with "shortest total travel time" and "lowest travel pressure index" as multiple objectives, using a linear programming algorithm, and under the premise of complying with corporate travel policies, solving for the three optimal "flight + hotel + airport transfer" integrated packages from real-time inventory. Simultaneously, automatically generating minor adjustment suggestions for the 1-2 schedules that contribute the most to WCLI (e.g., postponing an internal discussion meeting by 2 hours), and calculating the adjusted new WCLI value for user reference.

[0033] If WCLI ≥ 0.3, initiate a risk warning and negotiation mode: First, push a high-risk alert to the user, clearly indicating the main source of risk (e.g., "conflict with the project A launch review meeting"). Second, provide alternative solutions: 1) Local alternative: Recommend a high-definition video conferencing solution and a list of required equipment; 2) Trip splitting solution: Suggest splitting a long trip into two short trips to reduce WCLI per trip; 3) Pre-trip compensation solution: Automatically generate a pre-trip work handover checklist template and post-trip schedule protection suggestions (e.g., automatically insert a 3-hour "do not disturb focus period" on the first morning after return). After the user clearly confirms their needs, provide a limited selection of travel options.

[0034] The principle and effect of the above technical solution are as follows: based on the quantified risk level, different interaction strategies are adopted. It can provide efficient and considerate seamless recommendations when the risk is low, thus improving the experience. When the risk is high, it can proactively warn, provide consultation and alternative solutions, return the decision-making power to the user and assist them in managing risks. It realizes the organic combination of intelligent assistance and human decision-making, and effectively prevents major interruptions in the workflow caused by improper intelligent recommendations.

[0035] In one embodiment, S60 includes: when a user confirms and completes a travel booking, the system automatically triggers a preset compensation rule executor. For example, Rule 1: If the trip is an international flight spanning 6 or more time zones, all meetings in the user's corporate calendar that are "tentative" or "movable" will be automatically postponed by 2 hours on the day the user arrives at the destination city and the first day after returning to their residence. Rule 2: At 18:00 the day before the user's return date, key update summaries will be automatically extracted from their project management system and email system, and a "Work Briefing During Leave" will be generated and sent to their email address. Rule 3: From 9:00 to 12:00 on the first working day after the user's return, the "Light Do Not Disturb Mode" of their instant messaging software will be automatically activated, temporarily blocking message pop-up notifications from non-direct superiors and contacts marked as "non-urgent".

[0036] The principle and effect of the above technical solution are as follows: it advances the restoration of workflow continuity from the "suggestion" level to the "automated execution" level. Through predefined intelligent rules, it automatically performs operations such as schedule protection, information aggregation, and attention protection at key time points before and after travel, which substantially reduces the cognitive load and coordination costs of users after returning to work, and puts the invention's compensation for "implicit efficiency loss" into practice.

[0037] In one embodiment, S70 includes: after each service loop (trip completion or user cancellation of predicted demand), the system automatically generates a structured report. The report includes: 1) Demand Forecast Origin: displaying the core schedule events, project status, and collaboration relationships that triggered the forecast; 2) Risk Assessment Log: detailing the calculation process and values ​​for each WCLI item; 3) Recommended Solution Comparison: all previously provided solutions and their key parameters; 4) User Decision and Feedback: the user's final choice and evaluation of the recommendation; 5) Implemented Compensation Measures: a list of all workflow compensation operations automatically completed by the system. This report can be pushed to the user and their manager for travel performance analysis.

[0038] We selected fully anonymized data from 12 large and medium-sized enterprises in China from January 2022 to December 2024, including 182,000 valid business travel orders, 2.17 million structured calendar events, 9.4 million anonymized work collaboration data, and 126,000 corporate travel policies and rules. This data was divided into training, validation, and test sets in a 7:2:1 ratio. From these enterprises, we selected 4 companies and recruited 240 high-frequency business travelers with an average of ≥2 business trips per month, dividing them into 3 groups (low-risk, medium-risk, and high-risk, corresponding to blue, orange, and red scatter dots, respectively). From the offline test set, we selected 1000 completed orders, including 200 low-risk orders (which users reported as "seriously impacting work") and 800 high-risk orders (which had "no significant impact"). We used the WCLI model from Example 1 to backtrack and calculate the WCLI value and risk level of each order, and statistically analyzed the high-risk identification accuracy (number of correctly identified high-risk orders / ...). (Actual total number of high-risk orders) and low-risk false alarm rate (number of low-risk orders mistakenly identified as high-risk / actual total number of low-risk orders). Figure 1 This is a scatter plot showing the correlation between WCLI values ​​and users' actual work impact ratings. The horizontal axis represents the calculated WCLI value, and the vertical axis represents the user's rating of the impact of travel on work from 1 to 5 (1 point for no impact, 5 points for severe impact). The goodness of fit R of the linear fitting curve for the medium-risk group in the figure is shown. 2 =0.892, the three sets of data together verify that the WCLI value is strongly positively correlated with the user's actual perceived impact on work.

[0039] Example 2 This invention provides an AI-based business travel decision-making method based on workflow-endogenous coupling and dynamic conjugate optimization, as a deepening and creative expansion of Embodiment 1, including the following steps: S1: Acquire multi-source data and build a hypergraph model of user workflow.

[0040] Specifically, in implementing data collection, in addition to acquiring the structured calendar, historical travel, work context, and corporate policy data described in Example 1, the key lies in performing higher-level semantic association and modeling on this data. The system automatically constructs a user-centric, time-series workflow hypergraph by parsing the task decomposition structure in the project management system, the pre- and post-dependent dependencies in the calendar event sequence, and the discussion context regarding task collaboration in emails and instant messaging. In this hypergraph, nodes represent specific work task units (such as "writing project weekly reports," "customer solution demonstration," and "code review"), and hyperedges represent a complete workflow (such as "completing Q2 product release"), connecting all its associated task unit nodes. Each task node is labeled with the following attributes: estimated standard execution time (based on historical similar task statistics, in minutes), execution location preference (such as "fixed office workstation," "remote," and "customer site"), core collaborator set (extracted from the participant list and communication records), and key spatiotemporal flexibility attributes.

[0041] In S1, the spatiotemporal flexibility attribute is defined by two quantitative indicators: the time flexibility coefficient Te and the spatial flexibility coefficient Se. The time flexibility coefficient Te represents the flexibility of the task to be postponed or advanced, with a value range of [0, 1.0]. For example, a "team weekly meeting" has high time flexibility (Te=0.8, meaning it can be adjusted by 2 working days), while the "contractually agreed final delivery date" has extremely low flexibility (Te=0.1). This coefficient is automatically determined by analyzing the deviation between the historical actual execution time and the planned time of the task, keywords in the task description (such as "deadline" vs. "catch-up"), and rules in the enterprise policy library. The spatial flexibility coefficient Se represents the degree of dependence of the task on the geographical location of execution, with a value range of [0, 1.0]. For example, "data center on-site inspection" has low spatial flexibility (Se=0.2), while "internal department training" allows online participation (Se=0.9). This coefficient is determined by analyzing the historical location of the task and the accessibility of the required equipment or resources.

[0042] S2: Joint intent identification and impact deduction based on hybrid expert networks.

[0043] Specifically, this step uses a specially designed hybrid expert network model instead of a single prediction model. This model consists of four core components: Gated network: Receives the state feature vector of the current workflow hypergraph (including the urgency, resilience, and dependencies of each task), and outputs a weight distribution vector [W1, W2, W3], which is used to dynamically allocate the contributions of the three downstream expert sub-networks.

[0044] Travel Demand Trigger Expert: This expert subnetwork specifically identifies task nodes in the workflow hypergraph that have a "spatial elasticity coefficient Se below the preset threshold of 0.4" and "the core collaborator set includes external entities." It determines the inevitability and urgency of triggering offline travel demand and outputs a preliminary set of potential travel events {PE}. Each event includes the predicted destination, the earliest and latest time windows, and the core reason.

[0045] Workflow Resilience Assessment Expert: This expert subnetwork assesses the impact and resilience of the existing workflow hypergraph for each potential travel event (PE) if a user leaves during that time window. It simulates the removal or postponement of user-related task nodes, calculates the estimated delay Delta_T of the entire workflow's critical path, and the number N_c of task nodes requiring external coordination (e.g., colleagues taking over, meeting rescheduling).

[0046] Joint Strategy Generation Expert: This expert subnetwork integrates the outputs of the previous two, and instead of passively accepting the influence, it actively generates a preliminary workflow adaptive adjustment strategy set {AS}. For example, for a task affected by PE, the strategies include: "Postpone task A (Te=0.8) by 2 days", "Switch meeting B (Se=0.9) to online mode", and "Delegate task C to colleague X (based on the success rate of collaboration history)".

[0047] The hybrid expert network is trained end-to-end, and the learning objective is to make the final output joint policy {PE,AS} minimize a comprehensive loss function, which includes travel cost prediction error, workflow delay prediction error, and a score of policy executability.

[0048] S3: Perform dynamic conjugate optimization to generate a business travel-workflow joint action plan.

[0049] Specifically, this step is the core optimization process. The system uses {PE} and {AS} output by S2 as initial solutions to construct a dynamic conjugate optimization problem.

[0050] In S3, a set of travel variables X and a set of workflow adjustment variables Y are defined, and the set of travel variables X and the set of workflow adjustment variables Y are conjugate decision variables: The travel variable group X selects specific flights (F), hotels (H), and ground transportation (G) for each PE. Its feasible domain is constrained by the company's travel policy, real-time inventory, and pricing.

[0051] The workflow adjustment variable group Y selects a specific adjustment strategy for each affected original task, such as a new start time, a new execution mode (online / offline), and a new responsible person. Its feasible domain is constrained by the task's spatiotemporal elasticity coefficients (Te, Se), colleague schedule availability, and inter-task dependencies. The objective function for minimizing global perturbation is defined as follows: + + Here, GPI (Global Perturbation Index) is the global perturbation index, and a lower value is better. Cost_Travel is the total monetary cost of travel. Budget_Norm is the normalized baseline value for travel budget, which is the upper limit of the per-person budget for the same level of travel products with the corresponding origin and destination city pairs and travel duration in the company's travel policy. The unit is consistent with Cost_Travel. If there is no corresponding company policy, the industry average travel cost for the same type of travel for the origin and destination city pairs is used. T_Total is the user's total travel time (including transportation, transfers, and travel to hotels). Delta_CriticalPath is the total delay days of the critical path in the workflow. N_Coordination is the estimated number of new coordination sessions (such as meeting rescheduling invitations and task handover requests) that need to be initiated due to adjustments. The weighting coefficients 0.35, 0.25, 0.30, and 0.10 sum to 1 and can be adjusted according to company preferences.

[0052] Specifically, the optimization process employs an iterative solution using the alternating direction multiplier method: With a fixed workflow adjustment Y, optimize travel plan X: Within the adjusted workflow timeframe, find the X that minimizes GPI. With a fixed travel plan X, optimize workflow adjustment Y: Under the constraint of a defined travel time window, rearrange the workflow to minimize GPI. Iterate repeatedly until the change in GPI is less than the threshold of 0.01 or the maximum number of iterations (50) is reached. Finally, the optimal joint action plan is output.

[0053] S4: Perform path dependency strength analysis and comprehensive risk assessment. Specifically, conduct a deeper long-term risk analysis on the optimal joint action plan generated in S3. In S4, calculating the path dependency strength index includes: identifying all task nodes that are directly changed due to the current workflow adjustment Y, forming a set ModifiedNodes. In the workflow hypergraph, calculate the betweenness centrality changes of these nodes in the original network and the adjusted network. Betweenness centrality reflects the importance of a node as a "bridge" in the network. The formula for calculating the path dependency strength index PDS is: Where W_node is the importance weight of the task at that node; BC_after(node) is the betweenness centrality of the node in the adjusted workflow hypergraph, and BC_before(node) is the betweenness centrality of the node in the workflow hypergraph before the adjustment. The betweenness centrality is calculated as the proportion of the number of shortest paths in the workflow hypergraph that pass through the node to the total number of shortest paths, reflecting the importance of the node as a "bridge" in the network; W_node is the importance weight of the task at that node, with a value range of [0.5, 0.3]. It is assigned a value based on two dimensions: whether the task is on the critical path of the workflow and the enterprise priority of the project to which the task belongs. For example, the W_node of a high-priority task on the critical path is assigned a value of 0.3, the W_node of a regular-priority task on a non-critical path is assigned a value of 1.0, and the W_node of a low-priority task on a non-critical path is assigned a value of 0.5; the higher the PDS value, the greater the potential chain reaction of the current adjustment on future unplanned workflows.

[0054] The risk level of the joint action plan is determined by three indicators and projected into a three-dimensional decision space: Global Perturbation Index (GPI) (reflecting immediate costs), Path Dependence Strength (PDS) (reflecting medium- to long-term risks), and Scheme Confidence Score_C (derived from the feasibility score of the plan given by the S2 hybrid expert network). The system's preset risk cube regions include: Green region (recommended): GPI < 0.3, PDS < 1.5, Score_C > 0.8; Yellow region (executable with caution): does not meet the conditions of the green region, GPI < 3.0, PDS < 3.0; Red region (recommended to refactor): GPI ≥ 3.0 or PDS ≥ 3.0.

[0055] S5: Initiate joint review and dynamic revision of human-machine collaboration.

[0056] Specifically, the system no longer presents users with isolated travel options, but rather a complete interactive view of the "Joint Action Plan." The interface also clearly presents the following content: Travel plan section: Flights, hotels, pick-up and drop-off details and costs.

[0057] Workflow Adjustment Section: Display the affected schedule changes in a comparison view (Timeline Before / After), clearly listing each change (e.g., "Design review meeting postponed to Friday at 10:00 AM and moved online").

[0058] Risk Analysis Panel: Displays GPI and PDS values ​​and their corresponding 3D spatial plots.

[0059] Collaboration Status: For adjustments involving colleagues (such as rescheduling a meeting), the status of the temporary invitation sent through the collaboration interface is displayed (e.g., "awaiting confirmation from colleague Li Si").

[0060] Users can propose modifications to any part of the plan (e.g., "I don't want to postpone the design review meeting, please replan"). The system will quickly restart the local optimization of S3 based on the new constraints and generate a revised plan for user review.

[0061] S6: Execute automated collaboration and workflow injection. Once the user finally approves the joint action plan, the system will automatically execute travel booking: calling the travel system API to complete the booking of air tickets and hotels; automatically execute workflow adjustments: through the API of enterprise collaboration tools (such as calendars, project management software), automatically execute user-approved schedule changes, meeting mode conversions, task reassignments, etc., and send change notifications with intelligent explanations to all affected personnel (e.g., "Because Zhang San needs to go to Beijing from X month Y day to Z day to attend a key negotiation with client P to ensure the Q milestone of the project, the design review meeting originally scheduled for Y day has been changed to be held online at 10 am on Z+1 day. Your participation is crucial"); inject buffer and recovery tasks: according to the plan, automatically insert "pre-trip work handover preparation" reminder events and "focus buffer period on the first morning after returning to work" events into the user's schedule.

[0062] S7: Generate joint action source tracing reports and model evolution.

[0063] Specifically, the system generates an in-depth source tracing report that not only records the decision-making results but also elaborates on the optimization process. For example, "By adopting the strategy of 'moving internal workshops online' (contributing to a 1.5-day reduction in Delta_CriticalPath), the negative impact of choosing a more cost-effective but time-consuming flight was offset, ultimately reducing the Global Perturbation Index (GPI) by 0.8." The report also records the PDS analysis conclusions. User satisfaction feedback on the plan and data on the actual execution effect of the plan (such as whether the workflow was actually completed as adjusted) are used as reinforcement learning signals and fed back to the hybrid expert network of S2 for fine-tuning of model parameters, enabling the continuous evolution of the system.

[0064] This second embodiment first constructs a workflow parsing engine, integrating natural language processing and graph algorithms to automatically extract tasks and dependencies from heterogeneous data and calculate spatiotemporal elasticity coefficients to build a workflow hypergraph. Second, it designs and trains a hybrid expert network, requiring a jointly labeled dataset containing historical travel records and corresponding workflow change records for supervised learning. Next, it develops a dynamic conjugate optimization solver, integrating linear programming and constraint solvers to achieve rapid iteration using the alternating direction multiplier method. Then, it establishes a path dependency analysis module, dynamically calculating network centrality indices based on a graph computing library. Finally, it develops an enterprise-level collaborative gateway to achieve bidirectional API integration with mainstream travel management, calendar, project management, and instant messaging systems to support automated execution. The entire system is deployed in a microservice architecture, with loosely coupled communication between modules via message queues to ensure high reliability and scalability.

[0065] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for predicting user demand and recommending services in business travel scenarios based on AI, characterized in that, include: S10: Obtain user's native workflow data as model input data, combine it with de-identified corporate travel history data, and collect and construct a multimodal training and inference dataset covering user behavior, schedule, collaboration network and external environment according to data association rules. S20: Set up a two-stage hybrid training strategy, read samples from the dataset, perform offline pre-training and online fine-tuning of the core prediction model to obtain the prediction model; S30 deploys the predictive model on a real-time data processing platform, connecting enterprise scheduling systems, travel management systems, and third-party data sources through standardized interfaces to build a dynamic knowledge graph of users and a real-time computing engine, enabling continuous perception and analysis of travel intentions embedded in workflows. S40, construct a quantitative assessment model for the impact of workflow continuity, use the correlation features extracted from the knowledge graph and the output of the prediction model for comprehensive evaluation, combine the core work conflict degree, collaboration network interruption degree and recovery load prediction amount related to the process to form the workflow continuity loss coefficient WCLI, and conduct risk assessment and solution pre-drills in simulated environment and real scenario. S50 uses differentiated intelligent recommendation and negotiation strategies to generate candidate travel service solutions based on the different risk ranges of the WCLI value, and automatically generates workflow compensation and mitigation suggestions for high-risk trips.

2. The method as described in claim 1, characterized in that, The method also includes: S60, which executes automated workflow compensation services tied to the end-user's selected options, including automatically adjusting schedules, generating work handover summaries, and enabling a return-to-work focus mode; S70 outputs a comprehensive decision support report that includes the basis for demand forecasting, the risk assessment process, recommended solutions, and implemented compensation measures, and feeds user feedback back into the model training and knowledge graph update process.

3. The method as described in claim 1, characterized in that, Step S10 includes: S101 synchronizes structured calendar event data for a future preset time period from the user's enterprise office system through an authorized interface. The data fields include event title, start and end timestamps, location, list of participants, event category tags, and project ID. It also synchronizes historical order data for a past preset time period from the user's enterprise travel management platform. The fields include travel purpose description, departure and return time, origin and destination city pair, transportation and accommodation service provider and product code, and order creation and payment time. S102 collects user work context data through a security agent and privacy computing middleware, including natural language processing of email titles and meeting invitation texts sent and received by the user within the most recent preset time period to extract high-frequency project names, customer names and product terms as keywords; obtains the current status, last update time and next milestone date of the projects the user is responsible for in the collaboration tool; and obtains a list of the user's direct team members and frequently used collaborating colleagues from the enterprise HR system.

4. The method as described in claim 1, characterized in that, Step S20 includes: S201, Set up the offline pre-training stage: Use the records of completed trips in the historical dataset as positive samples, use all relevant schedules, emails and project status data within a preset number of days before the trip date as input feature sequences, use the trip order details as labels, and train the temporal Transformer network as the basic prediction model; S202, Set up an online incremental learning phase: Use all user interactions with the system's predictions and recommendations, as well as their subsequent actual travel results, as incremental training samples with reward signals. Fine-tune the basic prediction model every time a preset number of new samples are accumulated or every preset number of days. Step S30 includes: S301, deploy the prediction model obtained in S20 as a real-time microservice, which runs automatically at a preset period, with the input being the user's latest calendar event sequence, the most recent work context snapshot, and the knowledge graph feature vector that is updated in real time; S302 correlates the model's output predictions with real-time flight and hotel inventory and pricing information queried from the global distribution system. S303 is based on graph database technology to build and maintain a user's personal dynamic knowledge graph. The graph nodes include users, cities, airports, contacts, projects, and companies.

5. The method as described in claim 1, characterized in that, In step S40, the workflow continuity loss coefficient WCLI is calculated by weighted summation of core job conflict degree, collaborative network interruption degree, and recovery load forecast; wherein the core job conflict degree is used to quantify the severity of time overlap between the predicted travel time window and the key schedule events already existing in the user's calendar. Collaboration network disruption is used to quantify the risk of collaborative work disruption to a user’s team or closely collaborative network due to user departure during a predicted trip. The recovery load forecast is used to quantify the additional workload that is expected to be required to process backlogged transactions after users return from business trips. Based on the Workflow Continuity Loss Factor (WCLI) value, the defined risk level thresholds include: When WCLI is less than the first threshold, it is judged as a low workflow continuity risk; When WCLI is greater than or equal to the first threshold and less than the second threshold, it is judged as a medium workflow continuity risk; When WCLI is greater than or equal to the second threshold, it is judged as a high workflow continuity risk.

6. The method as described in claim 1, characterized in that, Step S50 includes: If WCLI is less than the first threshold, the proactive seamless recommendation mode is adopted: with the shortest total travel time and the lowest travel pressure index as multiple objectives, the optimization algorithm is used to solve the optimal integrated package under the premise of complying with the company's travel policy, and automatically generate the fine-tuning suggestions of the schedule that contributes the most to WCLI. If WCLI is greater than or equal to the first threshold, the risk warning and negotiation mode is activated: a high-risk alert is pushed to the user, and alternative solutions including local alternatives, trip splitting solutions, and pre-compensation solutions are provided. After the user clearly confirms their needs, a limited number of travel options are provided.

7. The method as described in claim 2, characterized in that, Step S60 includes: Once a user confirms and completes their travel booking, the system automatically triggers a preset compensation rule executor. The rules include: if the trip is a long-haul international flight, the system will automatically adjust the status of the meeting in the user's corporate calendar to a preset category on the day the user arrives in the destination city and the first day after returning to their place of residence. At a preset time one day before the user returns, key update summaries are automatically extracted from their project management system and email system, and a work briefing during their absence is generated and sent to their email address. The Do Not Disturb mode on the instant messaging software will be automatically activated during a preset time period on the first business day after the user returns.

8. The method as described in claim 2, characterized in that, Step S60 includes the system automatically generating a comprehensive decision support report after the service loop is closed. The report includes demand forecasting and tracing, risk assessment logs, comparison of recommended solutions, user decisions and feedback, and implemented compensation measures. The demand forecasting traceability system displays the core schedule events, project status, and collaboration relationships that trigger the forecast; the risk assessment log details the calculation process and values ​​for each item in WCLI; the recommended solution comparison includes all previously provided solutions and their key parameters; user decision and feedback shows the user's final choice and evaluation of the recommendation; the executed compensation measures are a list of all workflow compensation operations automatically completed by the system; and user feedback is fed back into the model training and knowledge graph update process.

9. The method as described in claim 1, characterized in that, The method also includes acquiring multi-source data and constructing a user workflow hypergraph model. Nodes in the workflow hypergraph represent work task units, and hyperedges represent workflows, connecting the task unit nodes. Nodes are labeled with estimated standard time consumption, execution location preference, core collaborator set, and spatiotemporal flexibility attributes. The method also includes joint intent recognition and impact inference based on a hybrid expert network, which includes a gating network, a travel demand triggering expert, a workflow resilience assessment expert, and a joint strategy generation expert: the gating network is used to receive the state feature vector of the current workflow hypergraph and output a weight distribution vector to dynamically allocate the contribution of the downstream expert sub-networks; The travel demand triggering expert is used to identify task nodes in the workflow hypergraph whose spatial elasticity coefficient Se is lower than a preset threshold and whose core collaborator set includes external entities. It determines the inevitability and urgency of triggering offline travel demand and outputs a set of potential travel events. Workflow resilience assessment expert is used to assess the impact and resilience of the original workflow hypergraph if users leave during a potential travel event. A joint strategy generation expert is used to synthesize the outputs of travel demand triggering experts and workflow resilience assessment experts to proactively generate a preliminary set of adaptive workflow adjustment strategies. The method also includes performing dynamic conjugate optimization to generate a business travel-workflow joint action plan. It defines conjugate travel variable sets X and workflow adjustment variable sets Y, and constructs an optimization problem aimed at minimizing the global disturbance index GPI. The global disturbance index GPI is a comprehensive indicator calculated by weighting and summing the total monetary cost of travel, total user travel time, total delay days on the workflow critical path, and the estimated number of new coordination sessions initiated due to adjustments. The optimization problem is iteratively solved using the alternating direction multiplier method until the convergence condition is met, and the optimal joint action plan is output. The method also includes path dependence strength analysis and comprehensive risk classification, including the calculation of the path dependence strength index PDS. The PDS is calculated based on the change in betweenness centrality of each node in the workflow hypergraph before and after the adjustment, and the importance weight of the task of that node, in the set of task nodes that have changed due to workflow adjustment. The risk level of the joint action plan is determined by the global disturbance index GPI, the path dependence strength index PDS, and the scheme confidence, and projected onto the three-dimensional decision space for regional division. The method also includes initiating joint review and dynamic revision of human-machine collaboration, showing users a complete interactive view of the joint action plan that includes travel options, workflow adjustment comparisons, risk analysis and collaboration status, and restarting local optimizations in response to users' modification commands.

10. The method as described in claim 2, characterized in that, The method also includes performing automated collaboration and workflow injection, automatically executing travel bookings after users approve joint action plans, automatically executing approved schedule changes, meeting mode conversions and task reassignment operations through the API of enterprise collaboration tools, sending change notifications, and inserting buffer and resume task events into users' schedules according to the plan; The method also includes generating joint action source tracing reports and model evolution. The reports detail the optimization decision-making process and use user feedback and actual implementation data of the plan as reinforcement learning signals to feed back to the hybrid expert network of S2.

Images