A travel plan recommendation method, device and electronic equipment

Abstract

This disclosure relates to a travel itinerary recommendation method, apparatus, and electronic device, comprising: generating a set of candidate travel itineraries that bind transportation and accommodation based on a travel plan; conducting a first round of screening by calculating compliance scores for the travel itineraries using a compliance agent and experience scores using an experience agent; employing a sliding time window strategy for dynamic constraint relaxation and determining the weights of compliance and experience scores through multiple rounds of game theory; and performing weighted summation to obtain a comprehensive score for the travel itinerary; conducting a second round of screening by using a spatiotemporal correlation model to predict the probability of successful or unsuccessful connection between transportation and accommodation in the travel itinerary based on the joint probability distribution of traffic delays and commuting congestion between transportation and accommodation; and displaying the travel itinerary to the target user. This disclosure treats transportation and accommodation as a whole, effectively avoiding the omission of potentially high-quality travel itineraries and spatiotemporal mismatches, and achieving deep coupling recommendation of transportation and accommodation.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to a method, apparatus and electronic device for recommending travel plans. Background Technology

[0002] Travel management is a core aspect of corporate administration. Existing travel solution recommendation systems process flights and hotels independently using rule engines. For example, flight and hotel recommendation modules can be queried and filtered independently, or filtered through a game-theoretic process between the two modules, displaying matching flights and hotels side-by-side for users to choose from. However, this approach separates transportation and accommodation recommendations, overlooking potentially high-quality travel options and causing mismatches between transportation and accommodation in terms of time and space. Summary of the Invention

[0003] In view of this, this disclosure proposes a travel itinerary recommendation method, apparatus, system, electronic device, storage medium, and computer program product. This disclosure treats transportation and accommodation as a whole. First, it generates a set of candidate travel itineraries that combine transportation and accommodation. Then, it employs a sliding time window strategy with dynamic constraint relaxation in a multi-round game for the first round of selection. Next, it uses a pre-prediction mechanism based on the joint probability distribution of traffic delays and commuting congestion between transportation and accommodation for a second round of selection based on spatiotemporal correlation. This effectively avoids missing potentially high-quality travel itineraries, achieves deep coupling recommendation of transportation and accommodation, avoids spatiotemporal mismatch between transportation and accommodation, and balances corporate travel standards with user travel preferences, thus improving user experience.

[0004] In some examples, during the process of generating compliance scores for travel plans that combine transportation and accommodation, the compliance agent performs semantic parsing of travel standards to extract "hard constraints" and "soft constraints." The compliance agent can then calculate compliance scores based on the soft and hard constraints generated by the semantic parsing, fully considering the flexible clauses in the travel standard information, thereby further avoiding filtering out high-value travel plans.

[0005] According to one aspect of this disclosure, a method for recommending travel plans is provided, comprising:

[0006] Obtain the travel plans of target users; Based on the travel plan, a set of candidate travel schemes that combine transportation and accommodation is generated, wherein each travel scheme in the set of candidate travel schemes that combine transportation and accommodation is a travel scheme that combines transportation and accommodation. The first round of screening is conducted on the candidate travel plans that combine transportation and accommodation. In this first round, for each travel plan, a compliance agent calculates the compliance score, and an experience agent calculates the experience score. A sliding time window strategy is used for dynamic constraint relaxation. The weights corresponding to the compliance score and the experience score are determined through multiple rounds of game theory. The compliance score and the experience score are then weighted and summed based on their respective weights to obtain the comprehensive score of the travel plan. A second round of screening is conducted on the set of travel plans that combine transportation and accommodation after the first round of screening. In the second round of screening, for each travel plan, a spatiotemporal correlation model is used to predict the probability of successful or unsuccessful connection between transportation and accommodation in the travel plan based on the joint probability distribution of traffic delay and commuting congestion between transportation and accommodation. Show the target user travel options from the collection of travel options that combine transportation and accommodation, after the second round of filtering.

[0007] In one possible implementation, the strategy of using a sliding time window for dynamic constraint relaxation, which determines the weights corresponding to the compliance score and the experience score through multiple rounds of game theory, includes: For each travel plan, in the current round of the game, the dynamic constraint relaxation of the compliance score and / or the dynamic constraint relaxation of the experience score are calculated based on the constraint relaxation function. Based on the dynamic constraint relaxation of the compliance score and / or the dynamic constraint relaxation of the experience score, and the compliance score and the experience score, calculate the first utility score of the compliance agent for the travel plan and the second utility score of the experience agent for the travel plan. Based on the first utility score, the second utility score, the current weight corresponding to the compliance score, and the current weight corresponding to the experience score, calculate the combined utility gain of the compliance agent and the experience agent corresponding to the current round of the game; Enter the next round of the game, update the current weights corresponding to the compliance score and the experience score, and repeat the above operations of calculating the dynamic constraint relaxation of the compliance score and / or the dynamic constraint relaxation of the experience score based on the constraint relaxation function in the current round of the game, until the preset conditions are met. The current weights of the compliance score and the experience score in the round of the game with the highest overall utility gain are respectively used as the weights of the compliance score and the experience score.

[0008] In one possible implementation, the constraint relaxation function is shown in the following equation: (t)= 0 exp( λ ΔI(t)) in, (t) represents the dynamic constraint relaxation at the current moment in the t-th round of the game; 0 represents the baseline maximum relaxation boundary; exp() represents the exponential penalty function; λ represents the scenario sensitivity coefficient; ΔI(t) represents the inventory consumption rate of transportation and accommodation in the travel plan within the time window corresponding to the current moment in the t-th round of the game.

[0009] In one possible implementation, for each travel itinerary, a spatiotemporal correlation model is used to predict the probability of successful or unsuccessful connection between transportation and accommodation based on the joint probability distribution of traffic delays and commuting congestion between transportation and accommodation. This includes: For each travel itinerary, obtain the latest check-in time for accommodations within that travel itinerary. Determine the characteristic information of commuting time for transportation and accommodation in the travel plan, as well as the characteristic information of arrival time for transportation in the travel plan; Based on the commuting time characteristics of transportation and accommodation in the travel plan and the arrival time characteristics of transportation in the travel plan, the target probability of arriving at accommodation in the travel plan earlier than the latest check-in time is calculated; wherein, the target probability represents the probability of successful connection between transportation and accommodation in the travel plan.

[0010] In one possible implementation, calculating the target probability that the arrival time at the accommodation in the travel plan is earlier than the latest check-in time, based on the characteristic information of commuting time for transportation and accommodation in the travel plan and the characteristic information of arrival time for transportation in the travel plan, includes: Based on the probability density function of commuting time for transportation and accommodation in the travel plan and the probability density function of arrival time for transportation in the travel plan, the following formula is used to calculate the target probability that the arrival time at accommodation in the travel plan is earlier than the latest check-in time.

[0011] In the formula, P(t) arrival <t checkin ) represents the time t for accommodation in the travel itinerary. arrival Earlier than the latest check-in time for accommodation in the travel plan. checkin The probability of t; f Indicates arrival time of transportation; t cIndicates commuting time for transportation and accommodation; t d Indicates the latest check-in time for accommodation; f F Let f represent the probability distribution of traffic arrival times. F (t f ) represents the arrival time t of the traffic. f The probability density function; f C f represents the probability distribution of commuting time between transportation and accommodation. C (t c ) represents the commuting time between transportation and accommodation. c The probability density function; Indicates the region t f +t c <t d Perform a double integral on the t+1 to calculate the probability of the joint event occurring.

[0012] In one possible implementation, the method further includes: Obtain the travel standard information corresponding to the travel plan; The travel standard information is semantically parsed using a first preset model to obtain at least one travel constraint; wherein the travel constraint includes hard constraints and / or soft constraints, the hard constraints represent constraints with rigid limitations, and the soft constraints represent constraints with flexible ranges. The calculation of the compliance score for the travel plan using a compliance intelligence agent includes: Calculate the score of the travel plan for each of the at least one travel constraint, where the score represents the degree to which the travel plan satisfies the travel constraint. Based on the preset weights of each travel constraint, the scores of the travel plan for each travel constraint are weighted and summed to obtain the compliance score of the travel plan.

[0013] In one possible implementation, the method further includes: Obtain the target user's historical travel data; Based on the target user's historical travel data, generate the target user's travel preference vector; The process of calculating the experience score of the travel plan through the experience agent includes: Generate the feature vector of this travel plan; The feature vector is concatenated with the user's travel preference vector to obtain a concatenated vector; The spliced ​​vector is processed by the second preset model to output an experience score for the travel plan.

[0014] In one possible implementation, the method further includes: In response to the target user's request to update the recommended travel plans, based on the target user's behavioral data during the process of viewing the travel plans in the set of travel plans that are linked to transportation and accommodation after the second round of filtering and / or the travel requirement information entered by the target user, the target user's travel preferences are updated. Based on the updated travel preferences of the target user, the first round of screening and subsequent operations are re-executed on the set of candidate travel options that combine transportation and accommodation.

[0015] According to another aspect of this disclosure, a travel plan recommendation device is provided, comprising: The acquisition module is used to acquire the travel plans of target users; The candidate travel plan generation module is used to generate a set of candidate travel plans that combine transportation and accommodation based on the travel plan. Each travel plan in the set of candidate travel plans that combine transportation and accommodation is a travel plan that combines transportation and accommodation. The first-round screening module is used to perform a first-round screening of the candidate travel plans that combine transportation and accommodation. In this first-round screening, for each travel plan, a compliance agent calculates the compliance score of the travel plan, and an experience agent calculates the experience score of the travel plan. A sliding time window strategy is used for dynamic constraint relaxation, and the weights corresponding to the compliance score and the experience score are determined through multiple rounds of game theory. Based on the weights corresponding to the compliance score and the experience score, the compliance score and the experience score are weighted and summed to obtain the comprehensive score of the travel plan. The second-round screening module is used to conduct a second round of screening on the set of travel plans that combine transportation and accommodation after the first round of screening. In the second round of screening, for each travel plan, a spatiotemporal correlation model is used to predict the probability of successful or unsuccessful connection between transportation and accommodation in the travel plan based on the joint probability distribution of traffic delay and commuting congestion between transportation and accommodation. The display module is used to show the target user the travel options from the collection of travel options that combine transportation and accommodation after the second round of screening.

[0016] According to another aspect of this disclosure, a travel itinerary recommendation system is provided, comprising: The global perception layer is used to acquire the target user's travel plans and travel standard information; The cognitive memory layer is used to store the target user's historical travel behavior data and the target user's behavioral data during the process of viewing travel plans in the set of travel plans that are linked to transportation and accommodation after the second round of filtering; A collaborative reasoning planning layer is used to generate a set of candidate travel plans that combine transportation and accommodation based on the travel plan. Each travel plan in the set consists of one transportation and one accommodation arrangement. A first round of screening is performed on the candidate travel plans. In this first round, for each travel plan, a compliance agent calculates its compliance score, and an experience agent calculates its experience score. A sliding time window strategy is used for dynamic constraint relaxation, employing multiple... The weights corresponding to the compliance score and the experience score are determined through round-robin game analysis. Based on these weights, the compliance score and the experience score are weighted and summed to obtain a comprehensive score for the travel plan. A second round of screening is then conducted on the travel plans that link transportation and accommodation after the first round of screening. In this second round, for each travel plan, a spatiotemporal correlation model is used to predict the probability of successful or unsuccessful connection between transportation and accommodation based on the joint probability distribution of traffic delays and commuting congestion between transportation and accommodation. The interaction execution layer is used to display travel options from the set of travel options that combine transportation and accommodation after the second round of filtering to the target user through an interactive interface.

[0017] According to another aspect of this disclosure, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-described method.

[0018] According to another aspect of this disclosure, a non-volatile computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the above-described method.

[0019] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0020] Through various aspects of this disclosure, the travel plans of target users are obtained; based on the travel plans, a set of candidate travel schemes that combine transportation and accommodation is generated, wherein each travel scheme in the set of candidate travel schemes that combine transportation and accommodation is a travel scheme that combines transportation and accommodation; a first round of screening is performed on the set of candidate travel schemes that combine transportation and accommodation; wherein, in the first round of screening, for each travel scheme, a compliance agent calculates the compliance score of the travel scheme, and an experience agent calculates the experience score of the travel scheme; a sliding time window strategy for dynamic constraint relaxation is adopted, and the corresponding compliance score is determined through multiple rounds of game theory. The system calculates the weights corresponding to the compliance score and the experience score, and then performs a weighted sum of these weights to obtain a comprehensive score for the travel plan. A second round of screening is then conducted on the travel plans that combine transportation and accommodation after the first round of screening. In this second round, for each travel plan, a spatiotemporal correlation model is used to predict the probability of successful or unsuccessful connection between transportation and accommodation based on the joint probability distribution of traffic delays and commuting congestion between transportation and accommodation. The target user is then shown the travel plans in the set of travel plans that combine transportation and accommodation after the second round of screening. This approach treats transportation and accommodation as a whole; firstly, based on the user's travel plan, multiple candidate travel plans that combine transportation and accommodation are generated; then, two rounds of screening are conducted on these candidate travel plans, effectively avoiding the omission of potentially high-quality travel plans and achieving deep coupling recommendation of transportation and accommodation.In the first round of screening, for any generated travel plan that combines transportation and accommodation, compliance and experience agents are used to generate compliance and experience scores for the travel plan. A sliding time window strategy for dynamic constraint relaxation is employed. Based on the inventory information of transportation and accommodation within the sliding time window, the utility scores of the compliance and experience agents for the travel plan are dynamically adjusted. This transforms the static game into a dynamic constraint relaxation process based on a sliding time window, simulating the uncertainty in real-world games or exploring potentially better weight allocations. Through multiple rounds of game analysis, the weights corresponding to the compliance score and the experience score are determined, and then a weighted sum is performed to obtain the travel plan. The system employs a comprehensive scoring mechanism, transforming the static game theory into a dynamic constraint relaxation process based on a sliding time window. This process comprehensively considers the degree of compliance with corporate travel standards and the degree of match with user travel preferences, balancing corporate travel standards and user travel preferences to achieve collaborative reasoning programming. In the second round of screening, for travel plans that combine transportation and accommodation after the first round of screening, a spatiotemporal correlation model is used to predict the probability of successful or unsuccessful connections between transportation and accommodation based on the joint probability distribution of traffic delays and commuting congestion between transportation and accommodation. This pre-prediction mechanism achieves deep coupling recommendation of transportation and accommodation, avoiding spatiotemporal mismatches in recommended travel plans and improving user experience. Furthermore, in the process of generating compliance scores for travel plans that combine transportation and accommodation, the compliance agent performs semantic parsing to extract "hard constraints" and "soft constraints." The compliance agent can calculate compliance scores based on the soft and hard constraints generated by semantic parsing, fully considering the flexible clauses in the travel standard information, thereby further avoiding filtering out high-value travel plans.

[0021] Other features and aspects of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0022] The accompanying drawings, which are included in and form part of this specification, illustrate exemplary embodiments, features, and aspects of this disclosure together with the specification and serve to explain the principles of this disclosure.

[0023] Figure 1 A flowchart illustrating a travel itinerary recommendation method according to an embodiment of the present disclosure is shown.

[0024] Figure 2 The flowchart illustrates a first-round screening of a set of candidate travel options that combine transportation and accommodation, according to an embodiment of the present disclosure.

[0025] Figure 3The flowchart illustrates a second round of screening of a set of travel options that combine transportation and accommodation, which have undergone a first round of screening, according to an embodiment of the present disclosure.

[0026] Figure 4 A structural diagram of a travel plan recommendation device according to an embodiment of the present disclosure is shown.

[0027] Figure 5 A schematic diagram of a travel itinerary recommendation system according to an embodiment of the present disclosure is shown.

[0028] Figure 6 A block diagram of an electronic device 1900 according to an embodiment of the present disclosure is shown. Detailed Implementation

[0029] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0030] As used herein, the terms “comprising,” “including,” “having,” or variations thereof are open-ended and include one or more of the stated features, integrals, elements, steps, components, or functions, but do not exclude the presence or addition of one or more other features, integrals, elements, steps, components, functions, or groups thereof.

[0031] When an element is referred to as “connected,” “coupled,” “responding,” or a variation thereof relative to another element, it may be directly connected, coupled, or responding to another element, or there may be an intermediate element present.

[0032] Although the terms first, second, third, etc., may be used herein to describe various elements / operations, these elements / operations should not be limited by these terms. These terms are only used to distinguish one element / operation from another. Therefore, without departing from the teachings of the inventive concept, a first element / operation in some embodiments may be referred to as a second element / operation in other embodiments.

[0033] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0034] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0035] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant regions.

[0036] Travel management is a core aspect of corporate administration. Among related technologies, rule engines are used to recommend transportation and accommodation, such as enterprise-level online travel agency (OTA) integration tools. Their working principle is as follows: travel standard parsing, which transforms travel standards into Boolean logic rules (e.g., "hotel price ≤ 800 RMB / night"); independent recommendation modules, where flight and hotel recommendations are queried and filtered independently, or filtered through a game-theoretic approach between the two modules; and result concatenation, displaying matching flights and hotels side-by-side for user selection. The workflow is as follows: the user inputs their requirements, then the flight application programming interface (API) and hotel API are called to retrieve flight and hotel information respectively, then the enterprise travel standard rules are applied to filter flights and hotels, finally outputting a hotel list and a flight list. This method of handling transportation and accommodation independently has the following limitations: The separation of transportation and accommodation recommendations: Transportation and accommodation are handled independently, lacking coordination in time and space, and overlooking potentially high-quality travel options; furthermore, the failure to consider commuting time from the airport to the hotel leads to a mismatch in time and space between transportation and accommodation. Limitations of static filtering: The rigid Boolean filtering of travel standards cannot handle flexible clauses such as "permitted in principle," resulting in overly strict compliance checks that filter out potentially high-quality travel options.

[0037] While this approach ensures basic compliance, it lacks in-depth collaborative reasoning, resulting in a poor overall user experience for recommended solutions. As corporate travel standards become more complex (ambiguous semantics, differences across job levels) and user needs become more personalized, this approach can no longer meet the demands of intelligent travel management.

[0038] To address one or more of the aforementioned technical problems, this disclosure provides a travel plan recommendation method (detailed description below). This method treats transportation and accommodation as a whole, sequentially generating a set of candidate travel plans that combine transportation and accommodation. The first round of screening involves a multi-round game using a sliding time window strategy for dynamic constraint relaxation. The second round of screening employs a pre-prediction mechanism based on the joint probability distribution of traffic delays and commuting congestion between transportation and accommodation, thus effectively avoiding the omission of potentially high-quality travel plans. This achieves deep coupling recommendation of transportation and accommodation, preventing spatiotemporal mismatches between them. It also improves user experience and meets the intelligent travel management needs arising from the increasing complexity of corporate travel standards and the personalization of user requirements. In some examples, during the compliance agent's generation of compliance scores for candidate travel plans combining transportation and accommodation: semantic analysis of travel standards is performed to extract "hard constraints" and "soft constraints." The compliance agent can calculate compliance scores based on the soft and hard constraints generated by semantic analysis, fully considering the flexible clauses in the travel standard information, thereby further avoiding filtering out high-value travel plans.

[0039] For example, the travel plan recommendation method provided in this disclosure can be executed by an electronic device or a part of an electronic device (such as a processor), such as a terminal device or a server. The terminal device can be a desktop terminal or a mobile terminal, for example, a laptop, tablet, desktop computer, smartphone, smart speaker, smartwatch, smart TV, in-vehicle terminal, or other types of electronic devices. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks, and big data and artificial intelligence platforms.

[0040] The following is a detailed description of a travel plan recommendation method provided by an embodiment of this disclosure.

[0041] Figure 1 A flowchart illustrating a travel itinerary recommendation method according to an embodiment of the present disclosure is shown, such as... Figure 1 As shown, the following steps may be included: Step 101: Obtain the travel plans of the target users.

[0042] The target users are those who request travel recommendations. In some scenarios, target users can provide travel plans in the form of voice, text, or images through the interactive interface of electronic devices such as mobile phones and computers. For example, if the target user is a company employee, they can log in to the company's travel booking platform through electronic devices and enter their travel plans in the interactive interface according to their travel needs.

[0043] For example, a business trip plan may include information such as departure point, destination, travel dates, and purpose. The travel dates refer to the user's planned departure time, return time, and duration; the purpose refers to the activities the user plans to participate in, such as meetings, training sessions, or client visits; the destination refers to the city or location the user needs to reach; and the departure point refers to the city or location the user departs from. For instance, if the target user is P8-level employee M of a technology company, and their travel plan is "to attend the Artificial Intelligence Summit at the Beijing National Convention Center from June 15-17, 2025," then the travel dates are June 15-17, 2025, the destination is Beijing, and the purpose is to attend the Artificial Intelligence Summit at the National Convention Center.

[0044] For example, a pre-trained large language model based on the Transformer architecture can be used in conjunction with an enterprise travel business knowledge base to build an intent recognition model; the intent recognition model can be used to perform semantic parsing, entity extraction (such as time, location) and context association on the unstructured travel plan input by the target user; and the results of semantic parsing and entity extraction can be converted into a structured travel plan containing standard fields (such as travel time, travel destination, travel purpose, etc.).

[0045] For example, user identity information of the target user can also be obtained, such as the user's job level, position, name, employee number, etc.

[0046] Step 102: Based on the travel plan, generate a set of candidate travel schemes that combine transportation and accommodation, wherein each travel scheme in the set of candidate travel schemes that combine transportation and accommodation is a travel scheme that combines transportation and accommodation.

[0047] In this context, a travel plan that combines transportation and accommodation represents a travel scheme that integrates transportation from the departure point to the destination with accommodation at the destination. Each travel plan includes information on one transportation item and one accommodation item. The specific data included in the transportation and accommodation information can be set according to needs and is not limited. For example, if the transportation is a flight and the accommodation is a hotel, then the travel plan is a flight and hotel bundled plan. This plan can include information such as the flight, airline, cabin class, time, price, and baggage allowance, as well as information such as the hotel, brand, location, price, star rating, and review score.

[0048] For example, based on the business trip dates, departure point, and destination in the target user's travel plan, real-time transportation information from the departure point to the destination can be queried, such as flight information, train information, and car information. Simultaneously, based on the business trip dates in the target user's travel plan, real-time accommodation information at the destination can be queried, such as hotel information and rental information. As an example, real-time transportation and accommodation information can be obtained from third-party data interfaces based on the target user's travel plan; for example, flight schedules, fares, and ticket availability can be obtained in real-time through airline APIs; hotel inventory information such as room availability and pricing data can be accessed from mainstream OTA platforms. Furthermore, based on the obtained real-time transportation and accommodation information, transportation and accommodation can be combined to obtain multiple transportation-accommodation-linked travel plans. These transportation-accommodation-linked travel plans can serve as candidate travel plans, thus completing the initialization of transportation-accommodation-linked travel plans.

[0049] As an example, we can iterate through real-time traffic information and real-time accommodation information to combine traffic and accommodation. Each traffic and each accommodation can be combined into a travel plan that combines traffic and accommodation, thereby generating a set of candidate travel plans that combine traffic and accommodation. This set can be denoted as B=b1,b2,...,bn, where n is a positive integer representing the number of travel plans that combine traffic and accommodation.

[0050] As an example, real-time traffic and accommodation information can be filtered based on preset rules, and combinations of these filtered options can be created. Each piece of traffic information and each hotel information is combined into a travel plan that binds transportation and accommodation, thus generating a set of candidate travel plans that bind transportation and accommodation. For example, the preset rule could be: if the purpose of the business trip is a meeting and the meeting time is in the morning, then transportation and accommodation arriving the night before the meeting should be selected; thus, transportation and accommodation information arriving the night before the meeting should be filtered and combined. Another example is: if the distance between the departure point and the destination is less than a certain value (e.g., 800km), then rail transportation should be selected; otherwise, air transportation should be selected; if the distance between the departure point and the destination in the target user's travel plan is less than a certain value, then rail information and accommodation information should be filtered and combined. For example, the preset rules could be: if the purpose of the business trip is a meeting, then the hotel should be within 1km of the meeting location; if the purpose of the business trip in the target user's travel plan is a meeting, then the transportation information can be combined with the filtered hotel information within 1km of the meeting location to combine transportation and accommodation.

[0051] For example, each travel plan that is bound to transportation and accommodation in the candidate set of travel plans can be represented by a vector. For instance, for any travel plan that is bound to transportation and accommodation, the various features of the travel plan (such as flight time, hotel brand, whether breakfast is included, etc.) can be transformed into a low-dimensional, dense numerical vector through embedding technology, which is the feature vector of the travel plan. This process aims to capture the semantic information of the travel plan and provide structured input data for subsequent intelligent agent processing.

[0052] Step 103: Conduct a first round of screening on the candidate travel plans that combine transportation and accommodation. In this first round, for each travel plan, a compliance agent calculates the compliance score, and an experience agent calculates the experience score. A sliding time window strategy is used for dynamic constraint relaxation. The weights corresponding to the compliance score and the experience score are determined through multiple rounds of game theory. Based on the weights corresponding to the compliance score and the experience score, a weighted sum is calculated to obtain the comprehensive score of the travel plan.

[0053] The system comprises a Compliance Agent and an Experience Agent, forming a Multi-Agent System (MAS). The Compliance Agent predicts the degree to which a travel itinerary conforms to the company's travel standards, generating a compliance score. The Experience Agent predicts the degree to which a travel itinerary matches the user's travel preferences, generating an experience score. The compliance score indicates the degree to which the travel itinerary conforms to the company's travel standards; for example, a higher compliance score indicates a better conformity. The experience score indicates the degree to which the travel itinerary matches the user's travel preferences; for example, a higher experience score indicates a better match. The weights corresponding to the compliance scores represent the importance of the compliance score calculated by the Compliance Agent, and the weights corresponding to the experience scores represent the importance of the experience score calculated by the Experience Agent.

[0054] The strategy of using a sliding time window for dynamic constraint relaxation involves dynamically adjusting the utility scores of the compliance agent and the experience agent for travel plans at each round of the game, based on the inventory information of transportation and accommodation within the sliding time window. This transforms the static game into a dynamic constraint relaxation process based on a sliding time window, simulating the uncertainty in real-world games or exploring potentially better weight allocations. Thus, by employing this strategy, the weights corresponding to the compliance score and the experience score are determined through multiple rounds of game analysis, thereby flexibly generating a comprehensive score for the travel plan.

[0055] The overall score of a travel plan can comprehensively reflect the degree to which the travel plan meets the company's travel standards and the degree of match with the travel preferences of the target users. For example, in the first round of screening, after calculating the overall score for each travel plan, the plan can be filtered based on the overall scores of each travel plan in the candidate set of travel plans that combine transportation and accommodation, resulting in a set of travel plans that combine transportation and accommodation after the first round of screening.

[0056] For example, a predetermined number of travel plans with lower overall scores can be filtered out from the candidate set of travel plans that combine transportation and accommodation, based on their comprehensive scores from lowest to highest. Alternatively, travel plans with comprehensive scores below a predetermined threshold can be filtered out. This results in a set of travel plans that have undergone the first round of filtering. The predetermined number and threshold can be flexibly set according to requirements. In this way, filtering based on the comprehensive scores of each travel plan in the candidate set of travel plans that combine transportation and accommodation can filter out travel plans with lower overall scores. The resulting set of travel plans that combine transportation and accommodation after the first round of filtering contains the travel plans with higher overall scores, ensuring that each travel plan in this set meets both the target user's personalized travel preferences and the company's travel standards.

[0057] Corporate travel standards, or travel policies for short, indicate the rules governing a company's travel policies; for example, they may include: accommodation class, transportation type, and cost restrictions.

[0058] In one possible implementation, the method further includes: acquiring travel standard information corresponding to the travel plan; performing semantic parsing on the travel standard information using a first preset model to obtain at least one travel constraint; wherein the travel constraint includes hard constraints and / or soft constraints, the hard constraints representing constraints with rigid limitations, and the soft constraints representing constraints with flexible ranges. In this way, travel constraints can be parsed from the travel standard information corresponding to the target user's travel plan. These travel constraints may involve job level, city, spending limits, cabin class, star rating, etc., and can be used by the compliance agent to predict the degree to which the travel plan complies with the company's travel standards.

[0059] Considering that corporate travel standards typically cover different groups of people (such as different job levels) and different business trip scenarios (such as different business trip destinations, different business trip purposes, etc.), it is possible to obtain the travel standard information corresponding to the travel plans of target users, so that the compliance intelligence agent can more accurately and quickly predict the degree to which the travel plan complies with the corporate travel standards. For example, based on the target user's identity information and / or information in their travel plan (such as travel destination, purpose of travel, etc.), the travel standard matching this information can be extracted from the company's travel standards to obtain the travel standard information corresponding to the target user's travel plan. For instance, taking the target user as P8-level employee M of a technology company, whose travel plan is "attending the Artificial Intelligence Summit at the Beijing National Convention Center from June 15-17, 2025," based on employee M's job level of P8, the matching travel standard information can be extracted from the technology company's corporate travel standards: "P8-level employee hotel standard: 800 RMB / night, business class, generally allowed to exceed the standard by 10%, special cases require reporting." Alternatively, based on employee M's job level of P8 and travel destination being Beijing, the matching travel standard information can be extracted from the technology company's corporate travel standards: "P8-level employee hotel standard in Beijing: 800 RMB / night, business class, generally allowed to exceed the standard by 10%, special cases require reporting." As an example, the original corporate travel standards can be obtained, which can be in the form of documents (such as PDF / Word), audio, video, or images. The travel standard information corresponding to the target user's travel plan can be extracted from these documents. This travel standard information is unstructured natural language text.

[0060] The first preset model has the ability to semantically parse travel standard information to generate hard constraints and / or soft constraints. Its specific type and model structure can be configured according to needs and are not limited thereto. The training method of the first preset model can refer to the existing methods.

[0061] For example, the first preset model is a large language model based on the Transformer architecture; the step of semantically parsing the travel standard information through the first preset model to obtain at least one travel constraint includes: performing semantic analysis on the travel standard information through the attention mechanism of the Transformer in the large language model to extract the hard constraint and / or the soft constraint. For example, the extracted hard constraint and / or soft constraint can be further vectorized to generate a travel standard constraint vector. In this way, the pre-trained large language model based on the Transformer architecture transforms the unstructured natural language text of the travel standard information corresponding to the target user's travel plan into a structured vector representation that can be understood and computed by machines; and, based on the semantic parsing of the large language model, the generated soft constraint fully considers the flexible terms in the travel standard information, thereby avoiding misjudgment of high-cost-performance travel plans when filtering travel plans in the future.

[0062] For example, the travel standard information corresponding to the target user's travel plan can first undergo text preprocessing and word segmentation. This includes cleaning, removing formatting tags, standardization, and word segmentation. Then, named entity recognition technology is used to automatically identify and extract key constraint elements from the travel standard information, such as job title, city, maximum amount, cabin class, hotel star rating, and train seat class. Next, through a multi-layer self-attention mechanism of a large language model, the logical relationships, conditional statements, and potential intentions of fuzzy semantics, such as "in principle" and "depending on the situation," are analyzed to obtain semantic information after deep semantic understanding. Finally, this semantic information is mapped to a high-dimensional continuous vector space to generate the travel standard constraint vector corresponding to the target user's travel plan. It should be noted that the high-dimensional continuous vector space has sufficient dimensions to ensure effective capture and differentiation of the rich semantic features and constraint relationships in complex travel standards. In this way, the generated travel standard constraint vector comprehensively represents the core constraints of the travel standard information corresponding to the target user's travel plan after semantic parsing, thus providing structured and quantifiable input data for subsequent compliance agent processing.

[0063] As an example, the first preset model is the BERT (Bidirectional Encoder Representations from Transformers) model. This involves preprocessing and segmenting the travel standard information corresponding to the target user's travel plan, using the BERT model to obtain the embedding of each token, and extracting hard and / or soft constraints through an attention weight matrix. This utilizes the Transformer attention mechanism to extract fuzzy semantic weight coefficients from natural language, achieving semantic softening of the travel standard information. Furthermore, a Graph Convolutional Network (GCN) can be applied to construct a graph of relationships between constraints. Finally, mean pooling is used for vector fusion to generate a high-dimensional travel standard constraint vector.

[0064] The attention weight matrix A is shown in the following formula:

[0065] In the formula, Q, K, and V are the query, key, and value matrices, respectively, dk is the dimension of the key vector, and T represents the transpose of the matrix.

[0066] For example, taking the travel standard information as "P8 level employees' hotel standard in Beijing is 800 yuan / night, business class, and in principle, it can exceed the standard by 10%, but special cases need to be reported", the semantic parsing through the first preset model yields the following results: hard constraint condition [hotel price ≤ 800, cabin class = business], soft constraint condition [flexible range 10%], and correspondingly, the travel standard constraint vector obtained through vectorization can be C = [0.95, 0.1, 0.8, ...].

[0067] User travel preferences indicate a user's likes or habits regarding transportation, accommodation, and other aspects when traveling for business. Examples include preferred hotel types, preferred hotel brands, preferred flight times, preferred airlines, price sensitivity, and habitual selection patterns for different cities.

[0068] In one possible implementation, the method further includes: acquiring the target user's historical travel data; and generating a user travel preference vector for the target user based on the target user's historical travel data. For example, a long-term memory mechanism can be configured to store the user travel preference vector generated based on the target user's historical travel data in a vector database, which can then be used by the experience agent to predict the degree of matching between travel plans and user travel preferences.

[0069] The user's historical travel data represents their online activity records related to transportation and accommodation during past business trips, such as clicks, searches, browsing, selections, bookings, and reviews. This data can characterize a user's long-term, stable travel preferences, such as preferences for specific airlines, hotel locations, and hotel amenities (e.g., gyms, laundry facilities). For example, historical travel data of target users can be extracted from a corporate travel booking platform; specifically, travel data of target users within a certain time frame (e.g., one month, one year, three years) can be extracted from the platform.

[0070] For example, generating a user travel preference vector for the target user based on the target user's historical travel behavior data can include: analyzing the target user's historical travel behavior data using a travel preference identification model to generate the target user's travel preference vector, which can integrate and represent the user's long-term and stable travel preferences. The travel preference identification model is trained based on historical travel behavior data belonging to different users and has the ability to predict the user's travel preference vector; the specific type and structure of the travel preference identification model can be configured according to requirements and are not limited thereto; the training method of the model can refer to existing technologies.

[0071] Step 104: Conduct a second round of screening on the set of travel plans that combine transportation and accommodation after the first round of screening; in the second round of screening, for each travel plan, a spatiotemporal correlation model is used to predict the probability of successful or unsuccessful connection between transportation and accommodation in the travel plan based on the joint probability distribution of traffic delay and commuting congestion between transportation and accommodation.

[0072] The Spatiotemporal Linkage model represents a model that comprehensively considers travel plans in both time and space dimensions. It predicts the probability of successful or unsuccessful connections between transportation and accommodation within a travel plan based on the joint probability distribution of traffic delays and commuting congestion between transportation and accommodation. Traffic delays reflect the uncertainty of arrival time for transportation within the travel plan, while commuting congestion between transportation and accommodation reflects the uncertainty of commuting time between them. The Spatiotemporal Linkage model uses the joint probability distribution of traffic delays and commuting congestion within the travel plan to predict the probability of successful or unsuccessful connections between transportation and accommodation, thus achieving a proactive risk assessment that considers uncertainties.

[0073] For example, in the second round of screening, after predicting the probability of successful or unsuccessful connection of transportation and accommodation for each travel option, filtering can be performed based on the probability of successful or unsuccessful connection of transportation and accommodation for each travel option in the candidate set of travel options with connected transportation and accommodation. Travel options with low probability of successful connection or high probability of unsuccessful connection are filtered out, thereby obtaining the set of travel options with connected transportation and accommodation after the second round of screening.

[0074] Step 105: Show the target user the travel options from the set of travel options that combine transportation and accommodation after the second round of screening.

[0075] In this step, the travel options presented to the target user are those that combine transportation and accommodation, thus providing recommendations for transportation and accommodation combinations. Furthermore, the target user can select one of the presented travel options as the final travel plan and then book transportation and accommodation; alternatively, the target user can trigger real-time dynamic replanning based on their needs, resulting in new travel options being recommended.

[0076] For example, travel options from a set of transportation and accommodation bundled travel plans, after a second round of filtering, can be displayed to the target user through the user interface of an electronic device. The specific number of travel options displayed can be set according to needs and is not limited. For instance, all travel options from the set of transportation and accommodation bundled travel plans after the second round of filtering can be displayed, or only a portion of the set can be displayed. When displaying multiple travel options to the target user, they can be displayed sequentially based on their overall rating.

[0077] For example, travel packages can also be rated to target users, such as experience ratings, compliance ratings, and overall ratings.

[0078] For example, when presenting any travel option to a target user, key information corresponding to that travel option can be displayed simultaneously. This includes key flight information such as airline, cabin class, time, and price, and key hotel information such as brand, location, price, and star rating. A "View Details" button can also be configured, allowing users to click on the button to view more detailed information about the flight or hotel, such as passenger reviews and on-time performance for flights, and hotel facilities, guest reviews, and room information.

[0079] For example, a short-term memory mechanism can also be configured to record the behavioral data of the target user during the process of viewing travel plans in the set of travel plans that are linked to transportation and accommodation after the second round of filtering; and to build a short-term conversation graph of the user in real time, which serves as a short-term conversation context and can reflect the user's travel preferences, thereby further adjusting the travel plan recommendation strategy to better suit the user's travel preferences.

[0080] In one possible implementation, the method further includes: in response to the target user's request to update the recommended travel plans, updating the target user's travel preferences based on behavioral data during the target user's viewing of travel plans in the set of travel plans that are linked to transportation and accommodation after the second round of filtering and / or travel requirement information input by the target user; and re-executing the first round of filtering and subsequent operations on the candidate set of travel plans linked to transportation and accommodation based on the updated user's travel preferences. The behavioral data of the target user during the process of viewing the travel plans in the set of travel plans that are linked to transportation and accommodation after the second round of filtering, as well as the travel requirements information entered by the target user, can reflect the target user's travel preferences to a certain extent. Based on this information, the target user's travel preferences are updated, and the operations in steps 103-105 above can be executed. In this way, based on the memory mechanism, online learning mechanism, and real-time dynamic replanning capability, the travel plan recommendations are dynamically adjusted. It can automatically adapt to the user's request to update the recommended travel plans and recommend new travel plans that meet the user's needs, without the need for the user to manually search again. The booking time is reduced from minutes to seconds, which significantly improves efficiency and accuracy and provides a better user experience.

[0081] As an example, if a target user wants to view more travel options among the displayed travel plans, they can send a request to update the recommended travel plans. In response to the target user's request to update the recommended travel plans, the system updates the target user's travel preferences in real time based on the behavioral data of the target user during the process of viewing the travel plans in the set of travel plans that are linked to transportation and accommodation after the second round of filtering. For example, if the target user spends a long time viewing a particular hotel, or clicks on the hotel's details, the system updates the target user's travel preferences. Then, based on the updated target user's travel preferences, the system re-executes the operations of steps 103-105 above, regenerates and displays new travel plans to the target user.

[0082] As another example, if a target user does not find a satisfactory travel plan among the displayed travel options, they can input travel request information, such as "recommend some cheaper hotels," triggering a request to update the recommended travel options. In response to the target user's request to update the recommended travel options, the system updates the target user's travel preferences in real time based on the travel request information entered by the target user. Then, based on the updated target user's travel preferences, the system re-executes the operations described in steps 103-105 above, regenerates, and displays a new travel plan to the target user.

[0083] For example, based on a real-time attention mechanism, the weights of user behavior data in short-term sessions can be calculated through the self-attention of the Transformer, and the weights of different travel feature preferences of users can be updated, thereby updating the user's travel preference vector. For example, if the user is very price-sensitive, the weight of total cost is increased; if the user has strict time requirements, the weight of flight punctuality rate is increased.

[0084] In one possible implementation, when a target user's travel itinerary changes, such as a change in destination or travel dates, the user can input the new destination or travel dates in the interactive interface, triggering a request to update the recommended travel plans. This updates the user's travel plan and re-executes steps 102-105, re-recommending travel options. In this way, when a user inputs change information, the system automatically replans and recommends new travel options that meet the user's needs, eliminating the need for manual searching. Booking time is reduced from minutes to seconds, significantly improving efficiency and accuracy.

[0085] In one possible implementation, the method further includes: obtaining weather information corresponding to the travel plan; generating at least one alternative travel plan when the weather information meets preset conditions, and displaying the at least one alternative travel plan to the target user; wherein the type of transportation in the alternative travel plan is different from the type of transportation in the original travel plan.

[0086] The travel plan includes weather information indicating weather warnings for the target user's destination during the travel period. For example, weather warnings for the destination airport can be obtained through the civil aviation meteorological system. Transportation types can include airplane, rail, and car. As an example, the travel plan uses airplane as the transportation type, while the alternative travel plan uses rail. Preset conditions can include a probability exceeding a certain value (50%) of affecting flight operations or causing delays. This allows for real-time monitoring of the destination's weather. When weather conditions at the destination affect flight operations during the travel period, such as sandstorms, heavy rain, hail, strong winds, dense fog, or thunderstorms, alternative travel plans are generated for users to use at any time. Furthermore, the alternative travel plan uses rail, which offers more stable arrival times and lower costs compared to flights, thus improving user experience and reducing corporate costs.

[0087] For example, the generation of alternative travel options can refer to steps 102-104 above. The transportation type of each travel option in the set of candidate travel options that are tied to transportation and accommodation needs to be changed, for example, from flight to rail.

[0088] For example, let's consider a target user, M, a P8-level employee at a tech company, whose travel plan is to attend an AI summit at the Beijing National Convention Center from June 15th to 17th, 2025. The presented travel option would be "Flight XXX of a certain airline (19:00-20:30) + Hotel XX in Beijing." This would allow us to obtain weather warnings for Beijing from June 15th to 17th. If a thunderstorm is predicted for Beijing on June 14th, with a 70% probability of flight delays, alternative travel options would be generated, such as "Beijing-Shanghai High-Speed ​​Railway G1 (13:00-17:30) + Hotel XX near Beijing South Railway Station."

[0089] In one possible implementation, while showing target users the travel options from the set of travel plans that have undergone a second round of filtering and are bundled with transportation and accommodation, the reasons for recommending these travel options and / or the chain of thought (CoT) can also be presented to the target users. In this way, by showing the reasons for recommending travel options and the chain of thought, interpretable output is achieved, which improves the transparency and credibility of decision-making, thereby enhancing users' trust in the recommended travel options, effectively reducing the probability of users needing to update the recommended travel options, and improving the user experience.

[0090] For example, a method based on a combination of preset rules and neural networks can be used to generate a thought reasoning chain; wherein, the preset rules can be set according to needs and are not limited thereto.

[0091] For example, natural language generation technology based on the target template can be used to output recommended travel options and explanations.

[0092] For example, taking the target user as P8-level employee M of a technology company, whose travel plan is "attending the Artificial Intelligence Summit at the Beijing National Convention Center from June 15th to 17th, 2025," the presented travel option would be "Flight XXX of a certain airline + Hotel XX in Beijing." The reasons for recommending this travel option could be presented as follows: "This option is recommended because: 1) The flight time allows you to get sufficient rest and avoids an early morning rush; 2) The hotel is only a 5-minute walk from the venue, saving commuting time." The reasoning chain could also be presented: "If the purpose of the business trip is a conference and the conference time is in the morning, then arriving the previous night is recommended; if the weather at the destination is bad, then alternative travel options will be generated; if the user's historical preferences include 'hotel near the venue,' then hotels within 1km of the venue will be prioritized." Furthermore, the travel option's rating could be displayed: compliance rating of 0.94 (price 780 yuan, within the standard range), experience rating of 0.88 (500-meter walking distance, including breakfast), and overall rating... .

[0093] In this embodiment, the travel plans of the target user are obtained; based on the travel plans, a set of candidate travel schemes that combine transportation and accommodation is generated, wherein each travel scheme in the set of candidate travel schemes that combine transportation and accommodation is a travel scheme that combines transportation and accommodation; a first round of screening is performed on the set of candidate travel schemes that combine transportation and accommodation; wherein, in the first round of screening, for each travel scheme, a compliance score is calculated by a compliance agent, and an experience score is calculated by an experience agent; a sliding time window strategy for dynamic constraint relaxation is adopted, and the weights corresponding to the compliance scores are determined through multiple rounds of game theory. The weights corresponding to the experience scores are considered; and the compliance scores and experience scores are weighted and summed based on the weights corresponding to the compliance scores and experience scores to obtain a comprehensive score for the travel plan; a second round of screening is conducted on the travel plan set that combines transportation and accommodation after the first round of screening; in the second round of screening, for each travel plan, a spatiotemporal correlation model is used to predict the probability of successful or unsuccessful connection between transportation and accommodation in the travel plan based on the joint probability distribution of traffic delays and commuting congestion between transportation and accommodation; the travel plans in the set of travel plans that combine transportation and accommodation after the second round of screening are displayed to the target user. In this way, transportation and accommodation are treated as a whole; firstly, based on the user's travel plan, multiple candidate travel plans that combine transportation and accommodation are generated; then, two rounds of screening are conducted on the multiple candidate travel plans that combine transportation and accommodation, thereby effectively avoiding the omission of potentially high-quality travel plans and realizing a deep coupling recommendation of transportation and accommodation.In the first round of screening, for any generated travel plan that combines transportation and accommodation, compliance and experience agents are used to generate compliance and experience scores for the travel plan. A sliding time window strategy for dynamic constraint relaxation is employed. Based on the inventory information of transportation and accommodation within the sliding time window, the utility scores of the compliance and experience agents for the travel plan are dynamically adjusted. This transforms the static game into a dynamic constraint relaxation process based on a sliding time window, simulating the uncertainty in real-world games or exploring potentially better weight allocations. Through multiple rounds of game analysis, the weights corresponding to the compliance score and the experience score are determined, and then a weighted sum is performed to obtain the travel plan. The system employs a comprehensive scoring mechanism, transforming the static game theory into a dynamic constraint relaxation process based on a sliding time window. This process comprehensively considers the degree of compliance with corporate travel standards and the degree of match with user travel preferences, balancing corporate travel standards and user travel preferences to achieve collaborative reasoning planning. In the second round of screening, for travel plans that combine transportation and accommodation after the first round of screening, a spatiotemporal correlation model is used to predict the probability of successful or unsuccessful connections between transportation and accommodation based on the joint probability distribution of traffic delays and commuting congestion between transportation and accommodation. This pre-prediction mechanism achieves deep coupling recommendation of transportation and accommodation, avoiding spatiotemporal mismatches in recommended travel plans and improving user experience. Furthermore, during the compliance agent's generation of compliance scores for candidate travel plans combining transportation and accommodation, the travel standards are semantically parsed to extract "hard constraints" and "soft constraints." The compliance agent can calculate compliance scores based on the soft and hard constraints generated by semantic parsing, fully considering the flexible clauses in the travel standard information, thereby further avoiding filtering out high-value travel plans.

[0094] The following is a summary of the above. Figure 1 In step 103, the processing procedure for each travel option in the first round of screening of the candidate travel options that combine transportation and accommodation is illustrated by example.

[0095] Figure 2 This diagram illustrates a flowchart of a first-round screening of a set of candidate travel options that combine transportation and accommodation, according to an embodiment of the present disclosure. Figure 2 As shown, the following steps may be included: Step 10301: For each travel plan, calculate the compliance score of the travel plan through the compliance agent.

[0096] In one possible implementation, calculating the compliance score of the travel plan through a compliance agent includes: calculating the score of the travel plan for each of the at least one travel constraint, where the score represents the degree to which the travel plan meets the travel constraint; and weighting and summing the scores of the travel plan for each travel constraint based on preset weights of each travel constraint to obtain the compliance score of the travel plan. Since the constraints may include hard constraints and / or soft constraints, hard constraints reflect the Boolean logic rules in the travel standard information, while soft constraints reflect the understanding of ambiguous semantics such as "in principle" and "special circumstances" in the travel standard information, thereby enabling the determination of the compliance score of the travel plan based on a rule engine and fuzzy logic.

[0097] As an example, the objective function of a compliant intelligent agent can be expressed as follows:

[0098] in, This represents the compliance score of travel plan B; n represents the total number of travel constraints, and i represents the index of the travel constraint. It is a function used to evaluate the degree to which travel plan B satisfies the i-th travel constraint. Its output can be normalized to the interval [0, 1], where 1 indicates complete satisfaction and 0 indicates complete violation. The specific type of this function can be configured according to requirements. This represents the weight assigned to the i-th travel constraint, reflecting the relative importance of that travel constraint in the overall compliance assessment. Thus, for this travel plan, the overall compliance score of the travel plan is obtained by iterating through all n travel constraints from 1 to n using this formula.

[0099] For example, for different travel constraints, the preset weights of each travel constraint can be configured according to needs. For instance, the weight of "total budget not exceeding the standard" is usually much higher than the weight of "preferentially choosing a certain airline".

[0100] For example, the Lagrange multiplier method can be used to handle hard constraints, while the penalty function method can be used to handle soft constraints. The Lagrange multiplier method transforms a constrained problem into an unconstrained optimization problem by introducing multipliers, ensuring that the solution lies within the feasible region. The output is 0 or 1 based on whether travel plan B meets the hard constraints; the penalty function method adds a penalty term to the objective function to quantify the degree of violation and transforms it into an unconstrained problem. Based on the degree to which travel plan B satisfies the soft constraints, output a value in the range of 0 to 1.

[0101] For example, the compliance agent can process the travel standard constraint vector and the feature vector of the travel plan as input data to calculate the compliance score of the travel plan.

[0102] Step 10302: Calculate the experience score of the travel plan through the experience agent.

[0103] For example, step 10302 may also be performed before step 10301, and there is no limitation thereto.

[0104] In one possible implementation, the step of calculating the experience score of the travel plan through the experience agent includes: generating a feature vector of the travel plan; concatenating the feature vector with the user's travel preference vector to obtain a concatenated vector; and processing the concatenated vector through a second preset model to output the experience score of the travel plan.

[0105] The type and specific structure of the second preset model can be configured according to requirements and are not limited thereto. The second preset model is a pre-trained neural network model that has the ability to predict the experience rating of a travel plan based on the concatenation of the feature vector of the travel plan and the user's travel preference vector. Its training method can be implemented using existing technologies.

[0106] As an example, the second preset model can be a multilayer perceptron (MLP), and the objective function of the experiencing agent is as follows:

[0107] Here, U(B) represents the user's predicted experience rating for travel plan B, also known as fpreference(B); UserProfile is the user's travel preference vector. Embedding(B) represents the transformation of various features of travel plan B into a low-dimensional, dense numerical vector through embedding techniques, i.e., the feature vector of travel plan B. ⊕ represents the vector concatenation operation, which connects the embedding vector Embedding(B) of travel plan B and the user's travel preference vector UserProfile in a dimensional direction, combining them into a longer feature vector so that the MLP can simultaneously consider the characteristics of the travel plan and the characteristics of the user's travel preferences. MLP(...) represents a multilayer perceptron, i.e., a feedforward neural network, which receives the concatenated feature vector as input, performs complex calculations through multiple nonlinear transformations, and finally outputs a scalar value U(B), i.e., the predicted user experience rating for travel plan B. In this way, the feature vector of the travel plan is concatenated with the travel preference vector of the target user and then input into the multilayer perceptron, thereby predicting the target user's subjective preference satisfaction with the travel plan, i.e., the experience rating.

[0108] For example, a multilayer perceptron can have a 3-layer fully connected neural network and employ... As an activation function.

[0109] Step 10303: Adopt a sliding time window strategy for dynamic constraint relaxation, and determine the weights corresponding to the compliance score and the experience score through multiple rounds of game.

[0110] For example, a strategy of dynamically relaxing constraints using a sliding time window can be used to determine the weights corresponding to compliance scores and experience scores through multiple rounds of game playing by a coordinating agent.

[0111] In one possible implementation, the strategy of using a sliding time window for dynamic constraint relaxation, which determines the weights corresponding to the compliance score and the experience score through multiple rounds of game playing, includes: for each travel plan, in the current round of game playing, calculating the dynamic constraint relaxation degree of the compliance score and / or the dynamic constraint relaxation degree of the experience score based on the constraint relaxation function; based on the dynamic constraint relaxation degree of the compliance score and / or the dynamic constraint relaxation degree of the experience score, and the compliance score and the experience score, calculating the first utility score of the compliance agent for the travel plan and the second utility score of the experience agent for the travel plan; based on the first utility score, the second utility score, and the experience score, calculating the first utility score of the compliance agent for the travel plan and the second utility score of the experience agent for the travel plan; and based on the first utility score, the second utility score, and the experience score, calculating the first utility score of the compliance agent for the travel plan and the second utility score of the experience agent for the travel plan. The current weights corresponding to the compliance score and the experience score are used to calculate the combined utility gain of the compliance agent and the experience agent in the current round of the game. In the next round of the game, the current weights corresponding to the compliance score and the experience score are updated, and the above-mentioned operations of calculating the dynamic constraint relaxation of the compliance score and / or the dynamic constraint relaxation of the experience score based on the constraint relaxation function in the current round of the game are repeated until a preset condition is met. The current weights corresponding to the compliance score and the experience score in the round of the game with the highest combined utility gain are respectively used as the weights corresponding to the compliance score and the experience score. The preset condition can be set according to needs and is not limited thereto; for example, it can be reaching a preset number of game rounds, or it can be the convergence of the calculated weights, etc.

[0112] For example, the constraint relaxation function is shown in the following equation: (t)= 0 exp( λ ΔI(t)) in, (t) represents the dynamic constraint relaxation at the current moment in the t-th round of the game; 0 represents the baseline maximum relaxation boundary; exp() represents the exponential penalty function; λ represents the scenario sensitivity coefficient; ΔI(t) represents the inventory consumption rate of transportation and accommodation in the travel plan within the time window corresponding to the current moment in the t-th round of the game.

[0113] Based on the aforementioned constraint relaxation function, it can be seen that the dynamic constraint relaxation is a dynamic state quantity based on the inventory update of transportation and accommodation in the travel plan at the current moment; the dynamic constraint relaxation usually decreases as the number of game rounds increases; by adjusting the utility score of the corresponding agent for the travel plan through the dynamic constraint relaxation, the static game is transformed into a dynamic constraint relaxation process based on a sliding time window.

[0114] As an example, for each travel itinerary, in the current round of the game, the dynamic constraint relaxation degree of the compliance score is calculated based on the aforementioned constraint relaxation function, as shown in the following formula: = 0 exp( λ ΔI(t)) in, This represents the dynamic constraint relaxation of the compliance score at the corresponding moment in the t-th round of the game. The dynamic constraint relaxation of the compliance score represents the maximum tolerance ratio that the compliance agent is allowed to exceed the original target. In the t-th round of the game, the compliance agent can dynamically decide whether to make concessions to the experience agent based on this dynamic constraint relaxation. 0 represents the baseline maximum relaxation boundary. Zero is a constant, representing the absolute physical bottom line that corporate management rules can tolerate. By setting a maximum relaxation boundary for the benchmark, the company's travel compliance judgment is made flexible while avoiding loss of control. For example, setting... =0.15 indicates that no matter how urgent the situation, the allowable exceedance of corporate travel compliance cannot exceed 15%, thus ensuring that the game has a strict safety boundary. exp() represents the exponential penalty function, using a negative exponential function of the natural constant, which better fits the resource grabbing game psychology of "smooth sailing in normal times, but sudden changes at critical points" in real-world scenarios. For example, when transportation and accommodation inventory is sufficient, the compliance agent can adhere to compliant travel standards; however, if transportation and accommodation inventory falls below the safety line, the compliance agent can instantly make concessions to improve game efficiency, so as to recommend travel plans to users more quickly and facilitate faster booking. ΔI(t) represents the inventory consumption rate of transportation and accommodation in the travel plan within the sliding time window corresponding to time t in the t-th round of the game, for example, the inventory consumption rate of flights and hotels; the size of the sliding time window and the sliding step can be set according to needs, for example, the window size is 5 minutes and the sliding step is 10 seconds; the inventory consumption rate of transportation and accommodation in the travel plan can be changed in real time through external airline or OTA interfaces. λ represents the scene sensitivity coefficient; this coefficient, as an adjustment of the exponential decay rate, can be set based on different travel scenarios.

[0115] For example, the λ value can be configured based on the varying sensitivities of different travel scenarios to environmental changes. As an example, for highly sensitive travel scenarios, such as travel during the Spring Festival travel rush or large-scale exhibitions, λ can be set to a larger value. In this scenario, even if hotel and flight inventory is currently low, the compliance agent remains highly sensitive, amplifying the dynamic constraint relaxation to achieve a faster game-theoretic outcome and thus recommend travel options to users more quickly, securing flights and hotels. Conversely, for less sensitive travel scenarios, such as regular business trips during off-peak seasons, λ can be set to a smaller value. In this scenario, hotel and flight inventory is usually plentiful, and the compliance agent tends to adhere to the company's compliance standards, prioritizing compliance with the company's travel standards. In addition, scenario sensitivity coefficients λ can be configured based on the routes and cities corresponding to the travel plan. For example, based on historical order data of different routes and cities, it can be determined by analysis that different routes and cities have different sensitivities to environmental changes, thereby setting different scenario sensitivity coefficients λ.

[0116] As an example, a first utility score for the compliance agent regarding the travel plan can be calculated based on the dynamic constraint relaxation of the compliance score and the compliance score itself. This first utility score represents a quantitative value of the compliance agent's "satisfaction" with the current game outcome, considering the current game context.

[0117] For example, in the t-th round of the game, the formula for calculating the utility score of the compliance agent for the travel plan can be expressed as:

[0118] in, This represents the utility score of the compliant agent for the i-th travel option bi in the t-th round of the game; The compliance score of the travel plan bi calculated by the compliance agent can be obtained by the objective function of the compliance agent in step 10301 above. Let represent the dynamic constraint relaxation degree of the compliance score at the corresponding moment in the t-th round of the game, which can be obtained through the constraint relaxation function mentioned above. Thus, this formula can be used to calculate the first utility score of the compliance agent for the travel option (i.e., travel option bi) in the current round of the game.

[0119] As an example, a second utility score for the experience agent regarding the travel plan can be calculated based on the dynamic constraint relaxation of the experience score and the experience score itself. This second utility score can represent a quantitative value of the experience agent's "satisfaction" with the current game outcome after considering the current game context.

[0120] For example, in the t-th round of the game, the formula for calculating the utility score of the experience agent for the travel plan can be expressed as:

[0121] in, Let represent the utility score of the agent for the i-th travel option bi in the t-th round of the game. The experience score of the travel plan bi calculated by the experience agent can be obtained by the objective function of the experience agent in step 10302 above. Let represent the dynamic constraint relaxation of the experience score at the corresponding moment in the t-th round of the game, which can be obtained through the constraint relaxation function mentioned above. Thus, this formula can be used to calculate the second utility score of the experience agent for the travel option (i.e., travel option bi) in the current round of the game.

[0122] For example, the combined utility gain of the compliant agent and the experience agent in the current round of the game can be the weighted geometric average of the utility gain obtained by the compliant agent and the experience agent in the current round of the game.

[0123] For example, in each round of the game, the current weight of the compliance score and the current weight of the experience score can be calculated using the Nash bargaining solution function.

[0124] As an example, the Nash bargaining solution function is shown below:

[0125] Where α represents the weight variable to be optimized for the compliant agent in the t-th round of the game, and β represents the weight variable to be optimized for the experience agent in the t-th round of the game. Let be the solution to the Nash bargaining function, which represents the optimal combination of the weights corresponding to the compliance score and the experience score calculated in the t-th round of the game. This represents the current weight corresponding to the compliance score calculated in round t of the game. This represents the current weight corresponding to the experience score calculated in the t-th round of the game; Let represent the first utility score of the compliance agent for the travel plan in the t-th round of the game, which can be calculated using the above formula for calculating the utility score of the compliance agent for the travel plan. Let represent the second utility score of the experience agent for the travel plan in the t-th round of the game, which can be calculated using the above formula for calculating the utility score of the experience agent for the travel plan; and It reflects the state of the game in the t-th round; This represents the retention utility or divergence point utility of a compliant intelligent agent. This indicates the retention utility or divergence point utility of the experiencing intelligent agent; and This refers to the minimum acceptable utility level that each agent can obtain if the compliance agent and the experience agent cannot reach a cooperation agreement (i.e., they cannot find a solution that is acceptable to both parties). It sets the bottom line for the game. This represents the utility gain obtained by the compliant agent in the t-th round of the game, i.e., the benefits of cooperating with the experience agent; Let α represent the utility gain obtained by the experiential agent in round t, i.e., the benefit of cooperating with the compliant agent. In this Nash bargaining solution function, α is the exponent of the utility gain obtained by the experiential agent in round t, and β is the exponent of the utility gain obtained by the compliant agent in round t. The magnitudes of these two exponents directly affect the proportion of their respective utility gains in the overall utility gain. argmax(α,β) represents finding the weighted geometric average of the utility gains obtained by the compliant agent and the experiential agent. The set of (α,β) parameters whose values ​​are maximized is then used. In this way, the coordinating agent, through this Nash bargaining solution function, aims to maximize the weighted geometric average (i.e., the overall utility gain) of the utility gain obtained by the compliance agent and the experience agent in each round of the game. This allows them to find a fair and efficient balance point, ensuring that an increase in utility for either the compliance agent or the experience agent does not excessively harm the interests of the other. In each round of the game, the coordinating agent updates the weights corresponding to the compliance score and the experience score, obtaining an optimal combination of these weights. Through iterative updates over multiple rounds, a stable equilibrium point is eventually found—the optimal combination of the weights corresponding to the compliance score and the experience score that yields the highest overall utility gain. This optimal combination of weights can then be used to further calculate the overall score of the travel plan, which best balances the compliance of corporate travel standards and the travel preferences of users. Thus, a Nash equilibrium multi-round game strategy employing a sliding time window for dynamic constraint relaxation balances corporate travel standards and individual preferences.

[0126] For example, the following is an example of an algorithm (Python) for coordinating agents to conduct multi-round games: def nash_equilibrium_optimization(compliance_agent, experience_agent, candidate_set): # Initialize game parameters alpha, beta = 0.5, 0.5# Initial weights for iteration in range(MAX_ITERATIONS): # Calculate the utility score for each agent util_comp = compliance_agent.evaluate(candidate_set) util_exp = experience_agent.evaluate(candidate_set) # Solving Nash Equilibrium nash_weights = solve_nash_equilibrium(util_comp, util_exp) alpha, beta = nash_weights # Combination Rating combined_scores = alpha util_comp + beta util_exp best_solution = candidate_set[combined_scores.argmax()] return best_solution, (alpha, beta) Step 10304: Based on the weights corresponding to the compliance score and the experience score, perform a weighted sum of the compliance score and the experience score to obtain the comprehensive score of the travel plan.

[0127] For example, the optimal combination of weights obtained through the above multi-round game (the weights corresponding to the compliance score and the experience score) is used to perform a weighted sum of the compliance score and the experience score of the travel plan. Since the optimal combination of weights is the Nash optimal solution generated through multi-round game evolution, the resulting comprehensive score of the travel plan can best balance compliance and user experience.

[0128] For example, the product of the weight corresponding to the compliance score and the compliance score of the travel plan can be calculated, the product of the weight corresponding to the experience score and the experience score of the travel plan can be calculated, and then the two products can be added together to achieve a weighted summation and obtain the comprehensive score of the travel plan.

[0129] Thus, through steps 10301-10304 above, each travel option in the travel plan is traversed, and a comprehensive score for each travel option is calculated. Based on this comprehensive score, some travel options are filtered out, resulting in a set of travel options that combine transportation and accommodation after the first round of screening. Because the compliance agent can calculate compliance scores based on soft constraints generated by semantic parsing of a large language model, it fully considers the flexible clauses in the travel standard information, thereby avoiding the filtering out of cost-effective travel options and greatly improving the utilization rate of travel resources.

[0130] In this embodiment, for any candidate travel plan that combines transportation and accommodation, a compliance agent and an experience agent calculate the compliance score and experience score of the travel plan, respectively. A sliding time window is used to dynamically relax the constraints. Through multiple rounds of game theory (such as multi-round game theory based on Nash equilibrium), the weights corresponding to the compliance score and experience score are determined, thereby generating a comprehensive score for the travel plan. This comprehensively considers the degree of compliance with corporate travel standards and user preferences, balances corporate travel standards and user travel preferences, and realizes collaborative reasoning planning.

[0131] The following is a summary of the above. Figure 1In step 104, the process of processing each travel plan in the second round of screening of the travel plans that are linked to transportation and accommodation after the first round of screening is illustrated by an example.

[0132] Figure 3 This diagram illustrates a flowchart of a second round of screening of a set of travel options combining transportation and accommodation, which has undergone a first round of screening, according to an embodiment of the present disclosure. Figure 3 As shown, it includes the following steps: Step 10401: For each travel plan, obtain the latest check-in time for accommodations within that travel plan.

[0133] For example, the latest check-in time for accommodations in the travel plan can be obtained from a third-party data interface; for example, the latest check-in time for a certain hotel is 23:00.

[0134] Step 10402: Determine the characteristic information of commuting time for transportation and accommodation in the travel plan, as well as the characteristic information of arrival time of transportation in the travel plan.

[0135] In this travel plan, the commuting time for transportation and accommodation refers to the time spent commuting from the arrival at the destination city by the transportation in the travel plan to the accommodation in the travel plan, and the arrival time of the transportation in the travel plan refers to the time when the transportation in the travel plan arrives at the destination city; for example, the commuting time between a certain flight and a certain hotel refers to the time spent commuting from the arrival at the destination airport by the flight to the hotel, and the arrival time of the flight refers to the time when the flight arrives at the destination airport.

[0136] As an example, feature information may include statistical features such as mean, standard deviation, and variance; For example, spatiotemporal correlation models can be based on gated networks, such as Long Short-Term Memory (LSTM) networks, for sequence modeling. LSTM networks have memory units and gating mechanisms, which can effectively capture long-term dependencies, thereby dynamically predicting the average and variance of commuting times for transportation and accommodation in the travel plan, as well as the average and variance of arrival times for transportation in the travel plan. The input dimensions of the LSTM network are [temporal features, spatial features, and environmental features], such as temporal features like scheduled flight arrival times, spatial features like routes from the airport to the hotel, and environmental features like weather conditions. By learning the complex spatiotemporal relationships between these features, the LSTM network can predict and quantify uncertainty. For example, different flight delay data, traffic data at different times and along different routes, and weather data can be obtained to train the LSTM network, where existing methods can be used for training. Training the network using various flight delay and weather data enables the Long Short-Term Memory (LSTM) network to predict future flight delays based on historical delay data, such as the average, variance, and standard deviation of arrival times. Similarly, training it with traffic and weather data from different time periods and routes enables it to predict future traffic conditions for specific time periods and routes, such as the average, variance, and standard deviation of commuting time. For example, considering the target user as a P8-level employee M at a technology company, whose travel plan is to attend an AI summit at the Beijing National Convention Center from June 15th to 17th, 2025, for a specific hotel, the network can predict an average commuting time of 90 minutes and a standard deviation of 15 minutes from Beijing Capital International Airport to the hotel during the evening rush hour on June 15th.

[0137] As another example, feature information may include information describing the probability distribution, such as the probability density function.

[0138] For example, the probability density function of commuting time for transportation and accommodation in the travel plan can be fitted using methods such as kernel density estimation based on historical traffic data of the same time period and route in the travel plan. Similarly, the probability density function of arrival time for transportation in the travel plan can be fitted using methods such as kernel density estimation based on historical traffic delay data in the travel plan.

[0139] Step 10403: Based on the characteristic information of commuting time for transportation and accommodation in the travel plan and the characteristic information of arrival time of transportation in the travel plan, calculate the target probability that the arrival time of accommodation in the travel plan is earlier than the latest check-in time; wherein, the target probability represents the probability that the transportation and accommodation itineraries in the travel plan are successfully connected.

[0140] As an example, the spatiotemporal correlation model considers the probability distribution of commuting time for transportation and accommodation in the travel plan, as well as the probability distribution of arrival time for transportation in the travel plan, and performs risk assessment based on the joint probability distribution. For instance, based on the average and variance of commuting time for transportation and accommodation in the travel plan, and the average and variance of arrival time for transportation in the travel plan, the cumulative distribution function of the standard normal distribution can be used to calculate the target probability that the arrival time at accommodation in the travel plan is earlier than the latest check-in time.

[0141] Taking the travel plan where transportation is by air and accommodation is by hotel as an example; the commuting time between the airport and the hotel is set to follow a Gaussian analysis (normal distribution), and the arrival time of the flight is set to follow a Gaussian analysis.

[0142] Probability distribution of commuting time between airport and hotel:

[0143] In the formula, Δt1 represents the commuting time variable, which is a random variable. This means that for the same commuting route, its specific value may be different each time it is calculated, thus simulating the fluctuation of commuting time caused by factors such as traffic congestion and weather. It means "follows a distribution of..."; N represents a Gaussian distribution, used to model the uncertainty of commuting time; This represents the mean of the Gaussian distribution, which indicates the historical average commute time from the airport to the hotel during a specific time period (such as peak traffic hours when flights are landing). The standard deviation of this Gaussian distribution represents the degree of fluctuation or uncertainty in commuting time. The larger the value, the more unstable the road conditions and the greater the range of variation in commuting time; This represents the variance of the Gaussian distribution, which is the variance of commuting time.

[0144] Probability distribution of flight arrival times:

[0145] In the formula, Δt2 represents the flight arrival time, which is a random variable. This means that for the same flight, its specific value may be different each time it is calculated, thus simulating the fluctuation of flight arrival time caused by factors such as weather. It indicates "follows a distribution of..."; N represents a Gaussian distribution, used to model the uncertainty of flight arrival times; This represents the mean of the Gaussian distribution, and denoted as the average arrival time of the flights. The standard deviation of flight arrival time is used to quantify the on-time performance or uncertainty of flight delays. This represents the variance of the Gaussian distribution, specifically the variance of flight arrival times.

[0146] The cumulative distribution function of the standard normal distribution is shown in the following equation:

[0147] In the formula, Pfail represents the probability of a failed itinerary connection, that is, the probability that "flight arrival time + commuting time" is not earlier than the hotel's latest check-in deadline; tdeadline represents the hotel's latest check-in time, which is a fixed time constraint. This represents the average arrival time of flights; This represents the average commute time from the airport to the hotel. This represents the variance of commuting time from the airport to the hotel. This represents the variance of flight arrival times, due to the uncertainty of flight delays ( ) and uncertainty of commuting time ( The variables are independent random variables. According to probability theory, the variance of the total uncertainty is the sum of the two variables, and the standard deviation is the square root of the variance. Φ represents the cumulative distribution function of the standard normal distribution, which calculates the probability that a value falls within a certain interval of the standard normal distribution. The numerator tdeadline ( + The denominator represents the available time margin. + The total uncertainty is measured. The entire fraction calculates the "standardized" time margin, and the Φ(...) function yields the probability of a successful flight-hotel connection (e.g., through table lookup), i.e., the target probability. Subtracting this value from 1 gives the probability of a failed flight-hotel connection, Pfail. Thus, by substituting the variance of flight arrival times, the variance of airport-hotel commute time, the average airport-hotel commute time, and the average flight arrival times into the Φ formula, the probability that "the arrival time at the hotel (flight arrival + commute time) is earlier than the latest hotel check-in time" (i.e., 1 / 2) is calculated. Pfail).

[0148] As another example, the spatiotemporal correlation model can conduct risk assessment based on the probability density function of commuting time for transportation and accommodation in the travel plan and the probability density function of arrival time of transportation in the travel plan, using a joint probability distribution.

[0149] For example, based on the probability density function of commuting time for transportation and accommodation in the travel plan and the probability density function of arrival time of transportation in the travel plan, the following formula is used to calculate the target probability that the arrival time of accommodation in the travel plan is earlier than the latest check-in time.

[0150] In the formula, P(t) arrival <t checkin ) represents the time t for accommodation in the travel itinerary. arrival Earlier than the latest check-in time for accommodation in the travel plan. checkin The probability of t (i.e., the target probability); f Indicates arrival time of transportation; t c Indicates commuting time for transportation and accommodation; t d Indicates the latest check-in time for accommodation; f F Let f represent the probability distribution of traffic arrival times. F (t f ) represents the arrival time t of the traffic. f The probability density function; f C f represents the probability distribution of commuting time between transportation and accommodation. C (t c ) represents the commuting time between transportation and accommodation. c The probability density function; Indicates the region t f +t c <t d Perform a double integral on the t+1 to calculate the probability of the joint event occurring.

[0151] Taking this travel plan as an example, where transportation is by air and accommodation is by hotel; then in the above formula, P(t) arrival <t checkin The t represents the probability of a successful connection between the flight and the hotel itinerary, i.e., the total commuting time from the airport to the hotel. arrival Earlier than the hotel's latest check-in time checkin The probability of t. f The actual arrival time of the flight is represented by a random variable; t c The commuting time from the airport to the hotel is represented by a random variable; t d This indicates the latest check-in time at the hotel, which is a fixed point in time; f F It is the probability distribution of flight arrival times, fF (t f ) represents the actual arrival time t of the flight. f The probability density function is used to describe this probability distribution; f C This is the probability distribution of commuting time from the airport to the hotel, f C (t c ) represents the commute time t c The probability density function is used to describe this probability distribution; Indicates the region t f +t c <t d We perform a double integral to calculate the probability of this joint event occurring. This formula represents the joint probability distribution considering both flight delays and traffic congestion from the airport to the hotel. By substituting the probability density functions of the commute time from the airport to the hotel and the flight arrival time into this formula for a double integral, we can obtain the probability that the total commute time from the flight landing at the airport to the hotel is earlier than the latest check-in time at the hotel. This gives us the probability of a successful connection between the flight and the hotel itinerary.

[0152] Furthermore, if the target probability is less than a preset threshold, the travel plan can be filtered out. The specific value of the preset threshold can be set according to needs and is not limited thereto; for example, it could be 0.95. Thus, risk assessment using the target probability shows that if the target probability is less than the preset threshold, it indicates a high risk of travel connection failure if the travel plan is executed, and therefore it is not suitable for recommendation to the target user, and is thus filtered out. If the target probability is not less than the preset threshold, it indicates a low risk of travel connection failure if the travel plan is executed, and therefore, it is a suitable candidate travel plan combining transportation and accommodation for recommendation to the target user, and is thus retained.

[0153] Thus, through steps 10401-10403 above, each travel plan in the set of travel plans that are linked to transportation and accommodation after the first round of screening is traversed, and some travel plans are filtered out to obtain the set of travel plans that are linked to transportation and accommodation after the second round of screening. In order to further recommend some or all of the travel plans in the set of travel plans that are linked to transportation and accommodation after the second round of screening to the target users.

[0154] In this embodiment, during the second round of screening, for each travel plan, a spatiotemporal correlation model is used to predict the probability of successful or unsuccessful connection between transportation and accommodation based on the joint probability distribution of traffic delays and commuting congestion between transportation and accommodation. This spatiotemporal correlation model further filters out travel plans with potentially unsuccessful connections between transportation and accommodation from the set of travel plans linked to transportation and accommodation after the first round of screening, thereby achieving deep coupling recommendation of transportation and accommodation, improving user experience, and realizing autonomous and intelligent travel plan recommendation. Simultaneously, by considering the uncertainty of the spatiotemporal correlation model, the geographical radius of accommodation locations in the recommended travel plans can be dynamically adjusted based on flight delay information and commuting traffic conditions. For example, under favorable traffic conditions, hotels farther from the airport may have shorter commuting times, and the probability of successful connection between flights and hotels may still be higher, thus expanding the geographical radius of the recommended travel plan; conversely, under poor traffic conditions, hotels closer to the airport have a higher probability of successful connection between flights and hotels, making them more likely to meet the requirements, thus narrowing the geographical radius of the recommended travel plan.

[0155] Based on the same inventive concept in the above method embodiments, this disclosure also provides a travel plan recommendation device, which can be used to execute the technical solutions described in the above method embodiments.

[0156] Figure 4 This diagram illustrates a structural diagram of a travel plan recommendation device according to an embodiment of the present disclosure, as shown below. Figure 4As shown, the device may include: an acquisition module 401 for acquiring the travel plans of a target user; a candidate travel plan generation module 402 for generating a set of candidate travel plans that combine transportation and accommodation based on the travel plans, wherein each travel plan in the set of candidate travel plans that combine transportation and accommodation is a travel plan that combines transportation and accommodation; and a first-round screening module 403 for performing a first-round screening of the set of candidate travel plans that combine transportation and accommodation. In the first-round screening, for each travel plan, a compliance score is calculated by a compliance agent, and an experience score is calculated by an experience agent. A sliding time window strategy for dynamic constraint relaxation is adopted, and the results are determined through multiple rounds of game theory. The compliance score is assigned a weight, and the experience score is assigned a weight. The compliance score and the experience score are then weighted and summed to obtain a comprehensive score for the travel plan. A second-round filtering module 404 is used to perform a second-round filtering on the travel plans that combine transportation and accommodation after the first-round filtering. In this second-round filtering, for each travel plan, a spatiotemporal correlation model is used to predict the probability of successful or unsuccessful connection between transportation and accommodation in the travel plan based on the joint probability distribution of traffic delays and commuting congestion between transportation and accommodation. A display module 405 is used to display the travel plans in the set of travel plans that combine transportation and accommodation after the second-round filtering to the target user.

[0157] In this embodiment, transportation and accommodation are treated as a whole. First, based on the user's travel plan, multiple candidate travel plans combining transportation and accommodation are generated. Then, two rounds of screening are performed on these candidate travel plans to effectively avoid overlooking potentially high-quality travel plans and achieve deep coupling recommendation of transportation and accommodation. Specifically, in the first round of screening, for any generated travel plan combining transportation and accommodation, compliance and experience agents are used to generate compliance and experience scores. A sliding time window strategy for dynamic constraint relaxation is employed. Based on the inventory information of transportation and accommodation within the sliding time window, the utility scores of the compliance and experience agents for the travel plan are dynamically adjusted. This transforms the static game into a dynamic constraint relaxation process based on a sliding time window, simulating the uncertainty in real-world games or exploring potentially better weight allocations. Through multiple rounds of game analysis, the weights corresponding to the compliance score and experience score are determined, and then a weighted sum is performed to obtain the travel plan. The system employs a comprehensive scoring mechanism, transforming the static game theory into a dynamic constraint relaxation process based on a sliding time window. This process comprehensively considers the degree of compliance with corporate travel standards and the degree of match with user travel preferences, balancing corporate travel standards and user travel preferences to achieve collaborative reasoning planning. In the second round of screening, for travel plans that combine transportation and accommodation after the first round of screening, a spatiotemporal correlation model is used to predict the probability of successful or unsuccessful connections between transportation and accommodation based on the joint probability distribution of traffic delays and commuting congestion between transportation and accommodation. This pre-prediction mechanism achieves deep coupling recommendation of transportation and accommodation, avoiding spatiotemporal mismatches in recommended travel plans and improving user experience. Furthermore, during the compliance agent's generation of compliance scores for candidate travel plans combining transportation and accommodation, the travel standards are semantically parsed to extract "hard constraints" and "soft constraints." The compliance agent can calculate compliance scores based on the soft and hard constraints generated by semantic parsing, fully considering the flexible clauses in the travel standard information, thereby further avoiding filtering out high-value travel plans.

[0158] In one possible implementation, the first round screening module 403 is further configured to: for each travel plan, in the current round of the game, calculate the dynamic constraint slack of the compliance score and / or the dynamic constraint slack of the experience score based on the constraint slack function; based on the dynamic constraint slack of the compliance score and / or the dynamic constraint slack of the experience score, and the compliance score and the experience score, calculate the first utility score of the compliance agent for the travel plan and the second utility score of the experience agent for the travel plan; based on the first utility score, the second utility score, the current weight corresponding to the compliance score, and the experience score... Given the current weights, calculate the combined utility gain of the compliance agent and the experience agent in the current round of the game; proceed to the next round of the game, update the current weights corresponding to the compliance score and the experience score, and repeat the above operations of calculating the dynamic constraint relaxation of the compliance score and / or the dynamic constraint relaxation of the experience score based on the constraint relaxation function in the current round of the game, until a preset condition is reached; take the current weights corresponding to the compliance score and the experience score in the round of the game with the highest combined utility gain as the weights corresponding to the compliance score and the experience score, respectively.

[0159] In one possible implementation, the constraint relaxation function is shown in the following equation: (t)= 0 exp( λ ΔI(t)) in, (t) represents the dynamic constraint relaxation at the current moment in the t-th round of the game; 0 represents the baseline maximum relaxation boundary; exp() represents the exponential penalty function; λ represents the scenario sensitivity coefficient; ΔI(t) represents the inventory consumption rate of transportation and accommodation in the travel plan within the time window corresponding to the current moment in the t-th round of the game.

[0160] In one possible implementation, the second round of filtering module 404 is further configured to: for each travel plan, obtain the latest check-in time of the accommodation in the travel plan; determine the characteristic information of the commuting time of transportation and accommodation in the travel plan and the characteristic information of the arrival time of transportation in the travel plan; based on the characteristic information of the commuting time of transportation and accommodation in the travel plan and the characteristic information of the arrival time of transportation in the travel plan, calculate the target probability that the arrival time of the accommodation in the travel plan is earlier than the latest check-in time; wherein, the target probability represents the probability that the transportation and accommodation itineraries of the travel plan are successfully connected.

[0161] In one possible implementation, the second round of screening module 404 is further configured to: calculate the target probability that the time of arrival at the accommodation in the travel plan is earlier than the latest check-in time, based on the probability density function of the commuting time of transportation and accommodation in the travel plan and the probability density function of the arrival time of transportation in the travel plan, using the following formula.

[0162] In the formula, P(t) arrival <t checkin ) represents the time t for accommodation in the travel itinerary. arrival Earlier than the latest check-in time for accommodation in the travel plan. checkin The probability of t; f Indicates arrival time of transportation; t c Indicates commuting time for transportation and accommodation; t d Indicates the latest check-in time for accommodation; f F Let f represent the probability distribution of traffic arrival times. F (t f ) represents the arrival time t of the traffic. f The probability density function; f C f represents the probability distribution of commuting time between transportation and accommodation. C (t c ) represents the commuting time between transportation and accommodation. c The probability density function; Indicates the region t f +t c <t d Perform a double integral on the t+1 to calculate the probability of the joint event occurring.

[0163] In one possible implementation, the acquisition module 401 is further configured to: acquire travel standard information corresponding to the travel plan; perform semantic parsing on the travel standard information using a first preset model to obtain at least one travel constraint; wherein the travel constraint includes hard constraints and / or soft constraints, the hard constraints representing constraints with rigid limitations, and the soft constraints representing constraints with flexible ranges; the first round screening module 403 is further configured to: calculate the score of the travel plan for each of the at least one travel constraint, wherein the score represents the degree to which the travel plan meets the travel constraint; and, based on the preset weights of each travel constraint, perform a weighted summation of the scores of the travel plan for each travel constraint to obtain a compliance score for the travel plan.

[0164] In one possible implementation, the acquisition module 401 is further configured to: acquire the target user's historical travel data; generate the target user's travel preference vector based on the target user's historical travel data; the first round of filtering module 403 is further configured to: generate a feature vector of the travel plan; concatenate the feature vector with the user travel preference vector to obtain a concatenated vector; process the concatenated vector through a second preset model to output an experience score for the travel plan.

[0165] In one possible implementation, the apparatus further includes an update module, configured to: in response to a request from the target user to update recommended travel plans, update the target user's travel preferences based on behavioral data during the target user's viewing of travel plans in the set of travel plans that have undergone the second round of filtering and / or travel requirement information input by the target user; and based on the updated travel preferences of the target user, re-execute the first round of filtering and subsequent operations on the set of candidate travel plans that have undergone the second round of filtering.

[0166] This disclosure also provides a travel itinerary recommendation system.

[0167] Figure 5 This diagram illustrates a travel itinerary recommendation system according to an embodiment of the present disclosure, such as... Figure 5 As shown, it can include: Omni-Perception Layer, Cognitive Memory Layer, Collaborative Reasoning Layer, and Action Layer.

[0168] The system comprises the following layers: a holistic perception layer for acquiring the target user's travel plans and travel standard information; a cognitive memory layer for storing the target user's historical travel behavior data and behavioral data during the process of the target user viewing travel options in the second round of filtering (transportation and accommodation bundled travel options); and a collaborative reasoning and planning layer for generating a candidate set of transportation and accommodation bundled travel options based on the travel plans, wherein each travel option in the candidate set is a travel option that combines transportation and accommodation; and a first round of filtering is performed on the candidate set of transportation and accommodation bundled travel options. In this first round of filtering, for each travel option, a compliance agent calculates the compliance score of the travel option, and an experience agent calculates the experience score of the travel option. The system employs a sliding time window strategy for dynamic constraint relaxation, determining the weights corresponding to the compliance score and the experience score through multiple rounds of game theory. Based on these weights, a weighted sum is calculated to obtain a comprehensive score for the travel plan. A second round of screening is then conducted on the travel plans that combine transportation and accommodation after the first round of selection. In this second round, for each travel plan, a spatiotemporal correlation model is used to predict the probability of successful or unsuccessful connection between transportation and accommodation based on the joint probability distribution of traffic delays and commuting congestion between transportation and accommodation. An interactive execution layer is used to display the travel plans in the set of transportation and accommodation-linked travel plans after the second round of selection to the target user through an interactive interface.

[0169] For example, the travel plan recommendation system adopts a four-layer architecture: a global perception layer (acquiring the target user's travel plans and standard travel information), a cognitive memory layer (user profile modeling), a collaborative reasoning and planning layer (generating a set of candidate travel plans that bind transportation and accommodation, followed by a first round of selection through a multi-round game using a sliding time window strategy for dynamic constraint relaxation, and a second round of selection based on the spatiotemporal correlation of traffic delays and commuting congestion between transportation and accommodation), and an interactive execution layer. This constructs an AI intelligent agent system with a closed loop of "perception-reasoning-planning-execution".

[0170] For example, the overall perception layer can be configured with: a flight dynamics collector to obtain flight schedules, fares, and ticket availability information in real time via airline APIs; a hotel inventory monitor to connect to hotel room and price data from mainstream OTA platforms; a traffic situation analyzer to integrate with map APIs and calculate commuting time and congestion index in real time; and a weather data collector to obtain weather warnings for business trip destinations.

[0171] For example, the cognitive memory layer adopts a hierarchical storage architecture to realize user-preferred long and short-term memory management: a long-term memory system (Vector DB) and a short-term memory system (Session Context) can be configured.

[0172] The long-term memory system uses a vector database to store users' travel data from the past three years. Furthermore, it generates user preference fingerprints based on perceptual hashing technology, including: brand preference vectors—hotel brand and airline preference ratings; a spatiotemporal preference matrix—habitual selection patterns across different time periods and cities; and consumption preference distribution—price sensitivity and service quality ratings. The short-term memory system constructs a real-time short-term conversation graph, following a "display-click-skip-conversion" timeline.

[0173] For example, the interaction execution layer can be configured with: a solution generation engine: using template-based natural language generation technology to output recommended travel solutions and explanations; a transaction processing module: supporting atomic transaction operations for bundled bookings of air tickets and hotels; and a monitoring and replanning module: monitoring external changes in real time and performing conditional judgment logic for replanning based on user needs.

[0174] In some embodiments, the functions or modules of the travel plan recommendation device or travel plan recommendation system provided in this disclosure can be used to execute the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0175] This disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0176] This disclosure also provides a non-volatile computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.

[0177] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0178] Figure 6 A block diagram of an electronic device 1900 according to an embodiment of the present disclosure is shown. For example, the electronic device 1900 may be provided as a server or a terminal device. (Refer to...) Figure 6The electronic device 1900 includes a processing component 1922, which further includes one or more processors, and memory resources represented by memory 1932 for storing instructions, such as application programs, that can be executed by the processing component 1922. The application programs stored in memory 1932 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 1922 is configured to execute instructions to perform the methods described above.

[0179] Electronic device 1900 may also include a power supply component 1926 configured to perform power management of electronic device 1900, a wired or wireless network interface 1950 configured to connect electronic device 1900 to a network, and an input / output interface 1958 (I / O interface). Electronic device 1900 can operate on an operating system, such as Windows Server, stored in memory 1932. TM Mac OS X TM Unix TM Linux TM FreeBSD TM Or similar.

[0180] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions that can be executed by a processing component 1922 of an electronic device 1900 to perform the above-described method.

[0181] Computer-readable storage media can be tangible devices capable of holding and storing programs / instructions used by instruction execution devices. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0182] The computer program (or computer-readable program instructions) described herein can be downloaded from a computer-readable storage medium to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage medium in the respective computing / processing device.

[0183] The computer program (or computer program instructions) used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions to implement various aspects of this disclosure.

[0184] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0185] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0186] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0187] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0188] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for recommending travel itineraries, characterized in that, include: Obtain the travel plans of target users; Based on the travel plan, a set of candidate travel schemes that combine transportation and accommodation is generated, wherein each travel scheme in the set of candidate travel schemes that combine transportation and accommodation is a travel scheme that combines transportation and accommodation. The first round of screening is conducted on the candidate travel plans that combine transportation and accommodation. In this first round, for each travel plan, a compliance agent calculates the compliance score, and an experience agent calculates the experience score. A sliding time window strategy is used for dynamic constraint relaxation. The weights corresponding to the compliance score and the experience score are determined through multiple rounds of game theory. The compliance score and the experience score are then weighted and summed based on their respective weights to obtain the comprehensive score of the travel plan. A second round of screening is conducted on the set of travel plans that combine transportation and accommodation after the first round of screening. In the second round of screening, for each travel plan, a spatiotemporal correlation model is used to predict the probability of successful or unsuccessful connection between transportation and accommodation in the travel plan based on the joint probability distribution of traffic delay and commuting congestion between transportation and accommodation. Show the target user travel options from the collection of travel options that combine transportation and accommodation, after the second round of filtering.

2. The method according to claim 1, characterized in that, The strategy of using a sliding time window for dynamic constraint relaxation determines the weights corresponding to the compliance score and the experience score through multiple rounds of game theory, including: For each travel plan, in the current round of the game, the dynamic constraint relaxation of the compliance score and / or the dynamic constraint relaxation of the experience score are calculated based on the constraint relaxation function. Based on the dynamic constraint relaxation of the compliance score and / or the dynamic constraint relaxation of the experience score, and the compliance score and the experience score, calculate the first utility score of the compliance agent for the travel plan and the second utility score of the experience agent for the travel plan. Based on the first utility score, the second utility score, the current weight corresponding to the compliance score, and the current weight corresponding to the experience score, calculate the combined utility gain of the compliance agent and the experience agent corresponding to the current round of the game; Enter the next round of the game, update the current weights corresponding to the compliance score and the experience score, and repeat the above operations of calculating the dynamic constraint relaxation of the compliance score and / or the dynamic constraint relaxation of the experience score based on the constraint relaxation function in the current round of the game, until the preset conditions are met. The current weights of the compliance score and the experience score in the round of the game with the highest overall utility gain are respectively used as the weights of the compliance score and the experience score.

3. The method according to claim 2, characterized in that, The constraint relaxation function is shown in the following equation: (t)= 0 exp( λ ΔI(t)) in, (t) represents the dynamic constraint relaxation at the current moment in the t-th round of the game; 0 represents the baseline maximum relaxation boundary; exp() represents the exponential penalty function; λ represents the scenario sensitivity coefficient; ΔI(t) represents the inventory consumption rate of transportation and accommodation in the travel plan within the time window corresponding to the current moment in the t-th round of the game.

4. The method according to claim 1, characterized in that, For each travel itinerary, a spatiotemporal correlation model is used to predict the probability of successful or unsuccessful connection between transportation and accommodation in that travel itinerary based on the joint probability distribution of traffic delays and commuting congestion between transportation and accommodation. This includes: For each travel itinerary, obtain the latest check-in time for accommodations within that travel itinerary. Determine the characteristic information of commuting time for transportation and accommodation in the travel plan, as well as the characteristic information of arrival time for transportation in the travel plan; Based on the commuting time characteristics of transportation and accommodation in the travel plan and the arrival time characteristics of transportation in the travel plan, the target probability of arriving at accommodation in the travel plan earlier than the latest check-in time is calculated; wherein, the target probability represents the probability of successful connection between transportation and accommodation in the travel plan.

5. The method according to claim 4, characterized in that, The calculation of the target probability that the arrival time at the accommodation in the travel plan is earlier than the latest check-in time, based on the characteristic information of commuting time for transportation and accommodation in the travel plan and the characteristic information of arrival time for transportation in the travel plan, includes: Based on the probability density function of commuting time for transportation and accommodation in the travel plan and the probability density function of arrival time for transportation in the travel plan, the following formula is used to calculate the target probability that the arrival time at accommodation in the travel plan is earlier than the latest check-in time. In the formula, P(t) arrival <t checkin ) represents the time t for accommodation in the travel itinerary. arrival Earlier than the latest check-in time for accommodation in the travel plan. checkin The probability of t; f Indicates arrival time of transportation; t c Indicates commuting time for transportation and accommodation; t d Indicates the latest check-in time for accommodation; f F Let f represent the probability distribution of traffic arrival times. F (t f ) represents the arrival time t of the traffic. f The probability density function; f C f represents the probability distribution of commuting time between transportation and accommodation. C (t c ) represents the commuting time between transportation and accommodation. c The probability density function; Indicates the region t f +t c <t d Perform a double integral on the t+1 to calculate the probability of the joint event occurring.

6. The method according to claim 1, characterized in that, The method further includes: Obtain the travel standard information corresponding to the travel plan; The travel standard information is semantically parsed using a first preset model to obtain at least one travel constraint; wherein the travel constraint includes hard constraints and / or soft constraints, the hard constraints represent constraints with rigid limitations, and the soft constraints represent constraints with flexible ranges. The calculation of the compliance score for the travel plan using a compliance intelligence agent includes: Calculate the score of the travel plan for each of the at least one travel constraint, where the score represents the degree to which the travel plan satisfies the travel constraint. Based on the preset weights of each travel constraint, the scores of the travel plan for each travel constraint are weighted and summed to obtain the compliance score of the travel plan.

7. The method according to claim 1, characterized in that, The method further includes: Obtain the target user's historical travel data; Based on the target user's historical travel data, generate the target user's travel preference vector; The process of calculating the experience score of the travel plan through the experience agent includes: Generate the feature vector of this travel plan; The feature vector is concatenated with the user's travel preference vector to obtain a concatenated vector; The spliced ​​vector is processed by the second preset model to output an experience score for the travel plan.

8. The method according to any one of claims 1-7, characterized in that, The method further includes: In response to the target user's request to update the recommended travel plans, based on the target user's behavioral data during the process of viewing the travel plans in the set of travel plans that are linked to transportation and accommodation after the second round of filtering and / or the travel requirement information entered by the target user, the target user's travel preferences are updated. Based on the updated travel preferences of the target user, the first round of screening and subsequent operations are re-executed on the set of candidate travel options that combine transportation and accommodation.

9. A travel itinerary recommendation device, characterized in that, include: The acquisition module is used to acquire the travel plans of target users; The candidate travel plan generation module is used to generate a set of candidate travel plans that combine transportation and accommodation based on the travel plan. Each travel plan in the set of candidate travel plans that combine transportation and accommodation is a travel plan that combines transportation and accommodation. The first-round screening module is used to perform a first-round screening of the candidate travel plans that combine transportation and accommodation. In this first-round screening, for each travel plan, a compliance agent calculates the compliance score of the travel plan, and an experience agent calculates the experience score of the travel plan. A sliding time window strategy is used for dynamic constraint relaxation, and the weights corresponding to the compliance score and the experience score are determined through multiple rounds of game theory. Based on the weights corresponding to the compliance score and the experience score, the compliance score and the experience score are weighted and summed to obtain the comprehensive score of the travel plan. The second-round screening module is used to conduct a second round of screening on the set of travel plans that combine transportation and accommodation after the first round of screening. In the second round of screening, for each travel plan, a spatiotemporal correlation model is used to predict the probability of successful or unsuccessful connection between transportation and accommodation in the travel plan based on the joint probability distribution of traffic delay and commuting congestion between transportation and accommodation. The display module is used to show the target user the travel options from the collection of travel options that combine transportation and accommodation after the second round of screening.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.

Images