Civil aviation data field-based travel task type judgment method, device and medium

Abstract

The application provides a travel task type judgment method and device based on civil aviation data fields and a medium, relates to the travel task type judgment technical field, and the method comprises the following steps: obtaining the judgment method identifier QR of a target user; if the QR is a preset first identifier, then using a preset rule judgment method and the civil aviation data field corresponding to the current travel task to judge the type of the current travel task of the target user, so as to determine the type of the current travel task of the target user as a business travel type or a private travel type; if the QR is a preset second identifier, then using a preset behavior analysis method to judge the type of the current travel task of the target user, so as to determine the type of the current travel task of the target user as a business travel type or a private travel type; the application can save the consumption of computing power and improve the judgment efficiency while ensuring the accuracy of judgment.

Description

Technical Field

[0001] This invention relates to the field of travel task type determination technology, and in particular to a method, device and medium for determining travel task type based on civil aviation data fields. Background Technology

[0002] In the civil aviation travel service sector, accurately determining the type of user's travel mission (official or private) has significant practical value: on the one hand, by differentiating between official and private travel, more accurate user profiles can be constructed (such as the travel frequency and preferred routes of official users, and the travel destinations and time patterns of private users), providing a foundation for user demand analysis; on the other hand, differentiated business recommendations can be carried out for different travel types (for example, recommending business hotels and airport express transfer services to official users, and recommending accommodations and attraction tickets around tourist destinations to private users), improving service matching efficiency.

[0003] However, existing judgment methods have obvious limitations: traditional rule-based judgment methods, while relying on features such as corporate ticketing channels and payment identifiers in civil aviation data, have the advantages of low computational power consumption and high computational efficiency, but their accuracy is low for travel tasks lacking typical corporate-related data (such as official travel with tickets purchased through personal channels); behavioral analysis methods make judgments by mining users' historical travel patterns, which has a relatively higher accuracy, but it consumes a lot of computational power and is less efficient due to the need to process a large amount of historical data; how to save computational power consumption and improve judgment efficiency while ensuring accuracy in judging the type of travel task has become an urgent technical problem to be solved. Summary of the Invention

[0004] To address the aforementioned technical problems, the technical solution adopted by this invention is as follows: According to a first aspect of this application, a method for determining travel task type based on civil aviation data fields is provided, the method comprising the following steps: S100, Obtain the target user's judgment method identifier QR; QR is determined by using different judgment methods to judge the type of several historical travel tasks of the target user. S200, if QR is the preset first identifier, then the preset rule judgment method and the civil aviation data field corresponding to the current travel task are used to judge the type of the target user's current travel task, so as to determine whether the target user's current travel task is a business travel type or a private travel type. S300, if the QR is a preset second identifier, then a preset behavior analysis method is used to determine the type of the target user's current travel task, so as to determine whether the target user's current travel task is a business trip or a private trip; the first identifier and the second identifier are different; QR codes are determined through the following steps: S110, Obtain information on each historical travel task of the target user within a preset historical time period; the type of travel task corresponding to each historical travel task is known. S120, use a preset rule judgment method to judge the travel task type corresponding to each historical travel task information to obtain the first judgment accuracy. S130, if the accuracy of the first judgment is greater than the preset accuracy threshold, then QR is determined to be the preset first identifier; otherwise, QR is determined to be the preset second identifier.

[0005] According to another aspect of this application, a non-transitory computer-readable storage medium is also provided, wherein at least one instruction or at least one program is stored in the storage medium, and the at least one instruction or at least one program is loaded and executed by a processor to implement the above-described method for determining the travel task type based on civil aviation data fields.

[0006] According to another aspect of this application, an electronic device is also provided, including a processor and the aforementioned non-transitory computer-readable storage medium.

[0007] The present invention has at least the following beneficial effects: The travel task type determination method based on civil aviation data fields of this invention uses rule-based judgment only when the target user meets the criteria for rule-based judgment; otherwise, it uses behavioral analysis. This allows for accurate determination of whether a user's travel task is for official or private purposes, providing reliable support for building precise user profiles (such as travel frequency and preferred routes for official users, and destination preferences and time patterns for private users) and conducting differentiated business recommendations (such as recommending business hotels and airport express transfer services to official users, and tourist accommodations and attraction tickets to private users). This effectively improves the accuracy of user demand analysis and service matching efficiency. Furthermore, by selecting the corresponding judgment method based on the identifier determined by the user's historical travel judgment results, rule-based judgment is used for users with high rule-based judgment adaptability, fully leveraging its advantages of low computational consumption and high efficiency. For users with low rule-based judgment adaptability, behavioral analysis is used, utilizing its ability to mine historical patterns to ensure accuracy. This balances the relationship between judgment accuracy, computational consumption, and judgment efficiency, solving the problem in existing technologies where a single judgment method cannot simultaneously achieve both accuracy and efficiency. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A flowchart illustrating a method for determining travel task type based on civil aviation data fields, provided in an embodiment of the present invention. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] It should be noted that, based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Furthermore, this device and / or practice the method can be implemented using other structures and / or functionalities besides one or more of the aspects set forth herein.

[0012] Example 1: The following will refer to Figure 1 The flowchart shown illustrates a method for determining the type of travel task based on civil aviation data fields, which introduces such a method.

[0013] The method for determining the type of travel task based on civil aviation data fields may include the following steps: S100, Obtain the target user's judgment method identifier QR; QR is determined by using different judgment methods to judge the type of several historical travel tasks of the target user.

[0014] In this embodiment, in the civil aviation field, a travel task can be understood as the task of purchasing air tickets. A user's travel task can be divided into business travel and personal travel. Business travel and personal travel have different effects on related hotel booking behavior, ride-hailing behavior or other related behaviors. Therefore, accurately determining whether a user's travel task is business travel or personal travel is of great significance for subsequent business recommendations.

[0015] QR codes are labels "tailor-made" for the target user to determine the appropriate method (rule-based judgment / behavioral analysis) for that user to determine their travel type. This label is not fixed but is determined based on the performance (accuracy) of the rule-based judgment method in the user's historical travel data. Essentially, it tests the suitability of the rule-based judgment method for the user using historical data, and then records the suitability results using QR codes.

[0016] By setting up QR codes, the advantages of rule-based judgments—low computational consumption and fast response speed—are fully utilized to efficiently identify users with clear enterprise association characteristics (such as business people who regularly purchase tickets through enterprise channels). This ensures accuracy while reducing resource consumption and improving overall judgment efficiency.

[0017] Furthermore, QR is determined through the following steps: S110, Obtain information on each historical travel task of the target user within a preset historical time period; the type of travel task corresponding to each historical travel task is known.

[0018] Obtaining users' travel data over a past period (e.g., 3 months), and ensuring that the type of these historical trips (business / private) is clear (can be confirmed through expense records, user annotations, etc.), is equivalent to preparing "test samples" and "standard answers" for rule-based judgment methods.

[0019] By constructing a "test baseline" using real historical data, we can ensure that the subsequent evaluation of the accuracy of rule judgments closely reflects the actual travel characteristics of users and avoid the distortion of QR signs due to samples deviating from reality.

[0020] S120: Use a preset rule-based judgment method to judge the travel task type corresponding to each historical travel task information to obtain the first judgment accuracy.

[0021] The rule-based judgment method is used to simulate and judge each of the above historical trips, and then compared with the known real types to calculate the accuracy rate (e.g., if the rule judgment is correct 8 times out of 10 historical trips, the accuracy rate is 80%). In essence, it is to test the "natural fit" of the rule judgment to the user.

[0022] By using quantified accuracy data, we can objectively evaluate whether a rule is suitable for a user, avoiding biases caused by subjective method selection, and ensuring that the matching between the user and the method is supported by data.

[0023] S130, if the accuracy of the first judgment is greater than the preset accuracy threshold, then QR is determined to be the preset first identifier; otherwise, QR is determined to be the preset second identifier.

[0024] Set an accuracy threshold (e.g., 85%). If the rule's accuracy for a user's historical travel exceeds the threshold, the rule is considered sufficiently reliable and is marked as a first-class indicator; otherwise, it is marked as a second-class indicator. The threshold clearly defines the boundary between suitable and unsuitable rules. The accuracy threshold is an empirical value, which can be calculated based on the overall accuracy of the rule-based judgment method across all users' historical travel data (e.g., the average accuracy of rule judgment for all known types of historical travel tasks is 75%). This can serve as a basic reference for the threshold. If the accuracy of the rule judgment itself is stable at 70%-80% for most users, the threshold should not be too high (e.g., above 85% would lead to most users being classified as behavioral analysis, wasting the efficiency advantage of rule judgment), nor should it be too low (e.g., below 60% would lead to the abuse of rule judgment in low-accuracy scenarios, affecting the reliability of the results).

[0025] By using threshold quantification standards, the standardization of QR identifiers is achieved, ensuring that users with the first identifier can consistently enjoy the efficiency of rule-based judgment, while allowing users with the second identifier to switch to more reliable behavior analysis methods in a timely manner, thus balancing efficiency and accuracy from a mechanism perspective.

[0026] S200, if QR is the preset first identifier, then the preset rule judgment method and the civil aviation data field corresponding to the current travel task are used to judge the type of the target user's current travel task, so as to determine whether the target user's current travel task is a business travel type or a private travel type.

[0027] The first identifier indicates that "the rule-based judgment method is highly adaptable to this user" (accuracy meets the standard in historical tests). At this point, it can quickly determine the cause by directly relying on the current travel's civil aviation data fields (such as whether the sales channel is business-to-business, whether the payment method is a corporate account, etc.), without calling complex calculations, and directly output the result (business / private) through preset rules.

[0028] By fully leveraging the advantages of rule-based judgments—namely, low computational consumption and fast response speed—it enables efficient judgment of users with clear enterprise association characteristics (such as business people who regularly purchase tickets through enterprise channels). This ensures accuracy while reducing resource consumption and improving overall judgment efficiency.

[0029] Furthermore, step S200 may include the following steps: S210, obtain the key fields corresponding to the current travel task; the key fields include at least the task creation platform, key user number, value-related fields, user and trip attribute fields.

[0030] Focusing on core civil aviation data fields strongly correlated with travel types, four key information categories were extracted: Task establishment platform: that is, the ticket purchase channel (such as B2B enterprise procurement platform, B2C personal APP), which directly reflects whether the ticket purchaser is an enterprise or an individual.

[0031] Key user ID: such as "major customer ID" or "corporate travel account identifier", used to associate with corporate identity (if not empty, it usually corresponds to business travel).

[0032] Value-related fields include payment method (corporate account / personal payment), ticket price type (corporate negotiated price / retail price), etc., reflecting the entity responsible for bearing the cost.

[0033] User and itinerary attribute fields, such as passenger type (adult / child), flight segment type (international long-haul / domestic short-haul), and booking time (weekday / weekend), help determine the travel scenario.

[0034] For example, if the key fields of the current trip show "Task creation platform = B2B enterprise platform" and "Key user number = XXX enterprise code", it initially points to official travel.

[0035] By filtering core fields and avoiding interference from irrelevant data, we focus on the essential characteristics of determining whether a transaction is for official or private purposes, such as "enterprise association identifier" and "expense liability," laying the foundation for rapid judgment. At the same time, the field types are highly matched with the core logic of the rule judgment method (relying on explicit identifiers), ensuring that the data can directly support the application of the rules.

[0036] S220, if the task creation platform is a preset task creation platform, or the key user number is not empty, or the value-related field corresponds to a preset organizational structure field, then the type of the target user's current travel task is determined to be official travel; otherwise, a secondary judgment is made based on the auxiliary identifier corresponding to the current travel task; wherein, the auxiliary identifier includes the identifiers corresponding to the user and itinerary attribute fields.

[0037] The system employs a two-tiered judgment logic: "core identifier priority + auxiliary identifier supplementation". The core identifier directly indicates whether it is for official business: If the "Task Establishment Platform" is a pre-defined enterprise-specific platform (such as a B2B or OFC agency ticketing platform), it indicates that the ticket purchase was initiated by the enterprise. If the "Key User ID" is not empty (such as an associated enterprise travel account), it directly points to the binding relationship between the user and the enterprise. If the "Value-Related Fields" contain pre-defined organizational structure characteristics (such as payment method being an enterprise account and ticket price type being "company agreement price"), it indicates that the cost is borne by the enterprise. If any of the above conditions are met, it is directly determined to be a business trip (no further calculation is required).

[0038] Secondary judgment of auxiliary identifiers when there is no core identifier: If none of the core identifiers are met (e.g., the platform is a personal app, or there is no company ID), then auxiliary identifiers based on user and itinerary attributes will be used for inference. For example: if the passenger type is an adult, the booking time is a weekday from 9:00 to 18:00, and the flight segment is an international long-haul (which aligns with high-frequency business travel scenarios), the probability of a business trip is increased; if the passenger type includes a child, the flight segment is a domestic short-haul, and the booking time is a weekend (which aligns with high-frequency private travel scenarios), the probability of a private trip is increased.

[0039] This step allows core identifiers to directly skip complex calculations, fully leveraging the advantages of rule-based judgments in terms of "low computational consumption and fast response." For users with clear enterprise association characteristics (such as those who frequently purchase tickets through enterprise platforms), judgments can be completed in milliseconds, adapting to large-scale real-time business scenarios (such as real-time recommendations in apps).

[0040] The supplementary identification feature addresses scenarios where "core identification is missing but implicit business-related characteristics exist" (e.g., a company does not use a dedicated platform but a user travels long distances on a weekday). It avoids misjudgments due to the absence of a single identification feature, improving the coverage of rule-based judgments while ensuring efficiency.

[0041] The logic aligns with the essential differences between business and personal travel (corporate affiliation, cost sharing, and scenario characteristics), closely matches users' actual travel patterns, and the output can directly support user profiling (such as labeling "corporate affiliated users") and business recommendations (such as pushing business services to business users).

[0042] S300, if the QR is a preset second identifier, then a preset behavior analysis method is used to determine the type of the target user's current travel task, so as to determine whether the target user's current travel task is a business trip or a private trip; the first identifier and the second identifier are different.

[0043] The second identifier indicates that "the rule-based judgment method has low adaptability to this user" (accuracy rate did not meet the standard in historical tests). These users usually lack typical corporate-related data (such as business travelers who habitually purchase tickets through personal channels). In this case, behavioral analysis methods are used to mine the user's historical travel patterns (such as frequent trips to business cities, travel time concentrated on weekdays, etc.), and the type is determined by combining the current travel characteristics (such as whether the destination and time period match historical business travel patterns).

[0044] By leveraging the advantage of behavioral analytics in handling atypical data, we can compensate for the accuracy shortcomings of rule-based judgments on atypical users, ensuring the reliability of travel type judgments for these users and providing support for subsequent precise profiling and business recommendations.

[0045] Furthermore, step S300 may include the following steps: S310, Obtain the initial operation behavior feature vectors corresponding to each operation process of the target user from starting the target software to exiting the target software within the target time period, and obtain the initial operation behavior feature vector list A = (A1, A2, ..., A...). i A n ), i=1,2,…,n; A i The initial operation behavior feature vector corresponding to the i-th operation process is obtained, where n is the number of operation processes; the target time period is the time period between the time point when the current travel task is established and the time point when the previous travel task is established.

[0046] In this embodiment, during the process of creating a travel task, users typically engage in activities such as searching, comparing prices, and browsing travel guides, resulting in multiple launches and exits of the target software. Based on the target user's usage records of the target software, each operation within a target time period can be obtained. The target time period is the time between the creation time of the current travel task and the creation time of the previous travel task, specifically the time between the completion time of the current ticket purchase and the completion time of the previous ticket purchase, accurately pinpointing the user's software operation cycle before planning their current trip. During this period, the complete operation chain of each time the user opens and closes the target software (e.g., the ticket purchase software) is recorded, and the operational behavior is transformed into an initial feature vector A. i For example, A i It can include quantitative features such as "date range for flight search", "filtering conditions (such as whether to check 'on-time rate priority')", "spending time distribution of browsing pages", and "whether to add companions", and finally form a list A containing n operation processes.

[0047] By precisely defining the time frame (between two trips), we ensure that the collected operational behaviors are directly related to the user's travel decisions, avoiding interference from irrelevant operational data; the operation process is broken down into feature vectors, laying the foundation for subsequent quantitative analysis of user behavior patterns, and transforming abstract operational behaviors into computable structured data.

[0048] Furthermore, step S310 may include the following steps: S311, Obtain the operation process of the target user from starting the target software to exiting the target software each time within the target time period, to obtain the operation process list F = (F1, F2, ..., F...). i F n ); F i This is the i-th operation process.

[0049] The target time period is the time between the creation of the current travel task and the creation time of the previous travel task (such as the time interval between two ticket bookings). During this period, the complete behavioral cycle of the user from each launch to exit of the target software is recorded in real time, forming an operation process list F. For example, if the user opens the APP three times within this time period to plan the trip and completes the operation chain of "searching for flights → browsing cabin availability", "adding fellow travelers → saving the trip", and "confirming the order → exiting", then F = (F1, F2, F3), where F1 corresponds to the first complete operation process.

[0050] By dividing the operation process into a closed loop of "start-exit," fragmented user software usage behaviors are organized into structured behavioral units, ensuring the independence and integrity of each operation. This division not only avoids fragmenting individual operations but also provides a basis for subsequent analysis of "user behavior differences at different stages" (such as the change in operation between the first query and the final ticket booking), making the behavior analysis more logical.

[0051] After step S311 and before step S312, the method may further include the following steps: S31, obtain F v The corresponding exit time point t of the target software v and F v+1 The corresponding startup time point t of the target software v+1 v = 1, 2, ..., n-1.

[0052] For two consecutive operation procedures in the operation procedure list F (e.g., the vth F), v and the (v+1)th F v+1 Extract F v Exit time point t v (The moment the user closes the target software) and F v+1 Startup time point t v+1 (The moment the user reopens the target software), calculate the interval between the two time points (e.g., F). v Exit at 10:00, F v+1 It starts at 10:05, with intervals of 5 minutes.

[0053] S32, if |t v -t v+1 If |≤Δt, then F v and F v+1 Combined into one operation process; Δt is the preset time interval threshold for the operation process.

[0054] A preset operation time interval threshold Δt (e.g., 15 minutes, adjustable based on average user decision-making habits) is set. If t... v With t v+1 If the interval is ≤Δt (e.g., 5 minutes ≤ 15 minutes as mentioned above), then F is determined.v and F v+ 1. If it belongs to "continuous operation of the same travel decision" (withdrawal in the middle is a temporary interruption, not abandonment of the decision), the two are merged into one operation process; if the interval is > Δt (such as an interval of 1 hour), it is considered an independent operation process.

[0055] When users plan their trips, their operations may be temporarily interrupted due to phone calls, switching software, etc. (e.g., after exiting the APP, they reopen it 10 minutes later to continue booking tickets). Merging short interval processes can avoid splitting the same decision-making behavior into multiple independent units and ensure that the operation process F can fully reflect the user's continuous decision-making logic (e.g., "searching for flights → temporarily exiting → reopening to confirm cabin availability" should be regarded as the same process).

[0056] Failure to merge short-interval processes will result in an artificially inflated number of operations (e.g., the same decision being split into 3 processes), affecting subsequent feature extraction (such as the operation record list H of S312). i Fragmented data may introduce biases (such as misjudging "multiple queries" as independent actions). Merged data provides a more comprehensive view of the decision, improving subsequent feature vector A. i The accuracy.

[0057] The above steps resolve the issue of "fragmented operation process caused by temporary interruptions," ensuring that the operation process list F accurately reflects the user's travel decision-making cycle. Essentially, it restores behavioral integrity through time correlation, providing more reliable foundational data for subsequent operation record extraction and feature vector construction, thereby improving the accuracy of judgments.

[0058] S312, Obtain the target user's information in F i Record every operation of the target software to obtain F i Corresponding operation record list H i =(H i，1 H i，2 H i，u H i，g（i） ), u=1,2,…,g(i);H i，u For target users in F i In the context of the target software, the u-th operation record is the record of the target user in F. i The number of operation records corresponding to the target software; each operation record includes the operation time, operation page, page dwell time, and input fields.

[0059] Regarding the operation process F i Extract the detailed data of all specific operations to form an operation record list H. i Each operation record H i,u It includes four core elements: Operation time point: accurate to the minute (e.g., "2025-07-01 09:15"), recording the specific moment when the user performed the operation.

[0060] Operation pages: such as "flight search results page", "cabin class selection page", "order confirmation page", etc., clearly define the scenario in which the operation occurs.

[0061] Page dwell time: For example, a 2-minute dwell time on the "price comparison page" reflects the user's level of attention to the information on the page.

[0062] Input fields, such as "departure city = Beijing", "date = 2025-07-10", and "passenger type = adult", reflect the user's core needs.

[0063] In addition to the elements mentioned above, it may also include whether or not one browses guides or comments on them.

[0064] For example, H1 corresponding to F1 might contain: H 1,1 (Time = 09:15, Page = Flight Search Page, Stay = 1 minute, Input = Departure Point + Destination), H 1,2 (Time = 09:16, Page = Results Filter Page, Dwell Time = 2 minutes, Input = "On-Time Rate Priority" selected), etc., ultimately forming a list H containing g(i) records. i .

[0065] By comprehensively reconstructing user operation details through several elements, the abstract concept of "using software" is transformed into traceable and concrete behaviors. The timing of operations reflects the decision-making process (e.g., a weekday morning might indicate business planning), the duration of page dwell reflects the focus (e.g., prolonged dwell on the price page might indicate personal travel), and input fields directly relate to travel needs (e.g., "traveling with multiple people" might indicate family travel for personal reasons). This data provides rich raw material for subsequent feature extraction, ensuring that behavioral analysis has a concrete basis.

[0066] S313, for H i The operation records in the database are used to extract, encode, and unify the initial operation behavior features to obtain A. i .

[0067] In this embodiment, A i This can be achieved in the following way: Initial operation behavior feature extraction: from H i Extract key features from the operation records, such as "frequency of flight queries", "whether the 'business class' filter was selected", "total time spent on the price comparison page", and "number of companions entered", focusing on behaviors related to business / private travel (e.g., business travelers may frequently query fixed routes, while private travelers may frequently adjust dates).

[0068] Encoding: Convert non-quantitative features into machine-recognizable numerical values. For example, "Operation Page = Business Class Booking Page" is encoded as "1", "Economy Class Booking Page" is encoded as "0"; "Input field contains corporate travel code" is encoded as "1", otherwise it is "0".

[0069] Dimensional unification: Standardize the characteristics of different operation processes (e.g., convert dwell time into "minutes / total operation time percentage") to ensure that each A i The feature dimensions are consistent (e.g., all are 10-dimensional vectors), which facilitates subsequent concatenation and model input.

[0070] Finally, the operation process F i Corresponding to a structured initial feature vector A i For example, A i = (Query frequency = 3, Business class filter = 1, Price page dwell time = 0.2%, Number of companions = 0), intuitively reflects the user's behavior pattern in this process.

[0071] In this step, feature extraction focuses on core behaviors and filters out irrelevant information (such as the number of page swipes) to improve data relevance; encoding transforms unstructured data such as text and scenes into numerical values, enabling machines to "understand" user behavior; dimension unification solves the problem of feature format differences in different operation processes, ensuring data compatibility during subsequent fusion and splicing, providing standardized input for behavior analysis methods (such as model judgment), and avoiding analysis bias caused by format chaos.

[0072] By following the process of "dividing the operation process → recording operation details → transforming into structured vectors," the complete user behavior chain from "opening the software to planning a trip" is transformed into structured data that can be analyzed by machines. This process preserves the integrity of the behavior (through F-division) while capturing key details (through H-division). i (Record), ultimately through A i Achieving "behavioral digitization" provides a calculable basis for subsequent judgments on whether users are acting for business (such as frequently searching for business class tickets or focusing on low prices) or for personal reasons (such as frequently adjusting dates or focusing on low prices). At the same time, it ensures that the data format is compatible with the behavioral analysis model to support accurate judgment.

[0073] S320, based on the target user's historical query records, determine whether the target user engaged in comparing the cost of travel tasks during each operation, and obtain the list of comparison travel task cost identifiers TA = (TA1, TA2, ..., TA...). i , ...,TA n ), TA i For A i The corresponding comparative travel task value identifier.

[0074] Based on user historical query data, determine the A process for each operation. i Does the user engage in comparing travel value, i.e., comparing airfare prices? For example, if the target user engages in comparing travel value, then tag them. i A is marked as "1" (meaning there is a cost comparison); otherwise, it is marked as "0", forming a list TA that corresponds one-to-one with A.

[0075] Furthermore, TA i The corresponding value can be determined through the following steps: S321, Obtain several historical query records corresponding to each query sub-process of the target user in the i-th operation process, so as to obtain the initial historical query record list set C corresponding to the i-th operation process. i =(C i，1 C i，2 C i，j C i，f（i） ), j=1,2,…,f(i);C i，j Let f(i) be the initial historical query record list corresponding to the j-th query subprocess in the i-th operation process, and f(i) be the number of query subprocesses corresponding to the i-th operation process. The query subprocess is the query process from the start of entering the query interface of the target software to the end of exiting the query interface of the target software.

[0076] Furthermore, the historical query records include: the travel time, departure point, and destination corresponding to the travel task.

[0077] In the i-th operation process (such as the complete usage cycle of a single opening of a travel app), each closed loop of the user's "entering the query interface (such as the flight query page) → exiting the query interface (such as returning to the homepage or exiting the app)" is defined as a "query sub-process" (for example, when the user enters the query interface (such as the flight query page) → exits the query interface (such as returning to the homepage or exiting the app)). i The process involves entering and exiting the flight search page 3 times (i.e., f(i) = 3, corresponding to 3 query sub-processes). For each query sub-process (e.g., the j-th sub-process), its corresponding historical query records (such as the queried flight date, price range, cabin class, etc.) are extracted to form an initial list C. i,j Finally, the C of the i-th operation process i It consists of a list of records from f(i) query sub-procedures.

[0078] By dividing the process into "query sub-processes," the user's query behavior during the operation is broken down into smaller analytical units, accurately identifying the core behaviors directly related to travel decisions (query is a key step in comparing prices and selecting flights); collecting corresponding historical records provides raw data for subsequent analysis of "whether users repeatedly compare costs," ensuring that the analysis focuses on the core scenario of query.

[0079] S322, Ci，r Several consecutive historical query records before the cut-out time point and C i，r+1 A list of intermediate historical query records is formed by several consecutive historical query records following the entry point, in order to obtain C. i The corresponding intermediate historical query record list set D i =(D i，1 D i，2 D i，r D i，f（i）-1 ), r=1, 2,…, f(i)-1; D i，r C i，r and C i，r+1 The corresponding list of intermediate historical query records; each historical query record includes time information.

[0080] Based on the core idea of ​​"the correlation between behavior before and after the cutout", targeting C i Extract C from two consecutive query sub-processes (such as the r-th and r+1-th sub-processes). i,r The last few records before the query interface is displayed (e.g., the 10 query records before the display; if less than 10, add zeros) and C. i,r+1 The first few records after entering the query interface (e.g., 10 query records after entering, padded with 0s if less than 10) are concatenated in chronological order to form the intermediate historical query record list D. i,r For example, the user in C i,1 Before switching, I checked "lowest price for economy class flights from Beijing to Shanghai on July 10th". After switching (possibly comparing prices on other software), I switched back to C. i,2 The query for "Beijing-Shanghai economy class tickets available on July 10th" returned D. i,1 Including these two consecutive records, the final C i The corresponding D i It consists of f(i)-1 such lists.

[0081] This step captures the behavioral correlation between users' "switch-out" and "switch-in" actions. Users traveling for personal reasons often switch between different apps to compare prices (e.g., switching out of the target app to check prices on a third-party platform, then switching back to confirm), and their query records often revolve around the costs (price, availability) of the same flight. Users traveling for business are more direct (no need for repeated price comparisons), resulting in weaker correlations. By stitching together records before and after switching out, this approach overcomes the limitation of "only analyzing behavior within the target app," fully reconstructing the user's cost comparison logic across interfaces and avoiding gaps in behavioral analysis caused by switching between apps.

[0082] S323, for D i Query behavior features are extracted from each intermediate historical query record list to obtain D. i The corresponding query behavior feature vector E i=(E i，1 E i，2 , ..., E i，r , ..., E i，f（i）-1 ); E i，r D i，r The corresponding query behavior feature vector; E i Each query behavior in the dataset has the same dimension in its feature vector.

[0083] For D i Each intermediate historical query record list D i,r (e.g., related records of two consecutive query sub-processes), extract behavioral features related to "cost comparison", such as "whether the two queries are for the same route", "whether the price filter criteria have been adjusted (e.g., from 'lowest price priority' to 'highest price unlimited')", "whether the cabin class query type has changed (e.g., from economy class to business class)", "query interval duration (time from switchout to switchin)", etc.; encode these features into numerical values ​​(e.g., "same route" = 1, "different routes" = 0), and unify the dimensions (e.g., all are 8-dimensional vectors) to form E i,r Ultimately E i By D i It consists of all the feature vectors in the middle.

[0084] By transforming related query records into quantifiable feature vectors, the correlation between "cutout-cutout" and "cut-in" is converted from abstract behavior into machine-recognizable numerical values ​​(such as the feature combination of "same route + price adjustment," which directly points to cost comparison behavior). Unified dimensions ensure data compatibility during subsequent data concatenation, providing standardized foundational data for analyzing behavioral correlations.

[0085] S324, in chronological order, E i Each query behavior feature vector in the process is concatenated to obtain the target query behavior feature vector EA corresponding to the i-th operation. i .

[0086] In chronological order, E i All query behavior feature vectors are concatenated into a complete vector EA. i For example, if E i,1 This reflects the characteristic of "initially searching for the lowest price → switching out and then switching back to search for the same flight", E i,2 The EA, which incorporates the feature of "checking prices again after adjusting the date", is a composite document. i The system will fully record the continuous behavioral logic of the user during the i-th operation, which involves "multiple queries - switching out - switching in".

[0087] By preserving the behavioral sequence of "before switching out - after switching in" (e.g., comparing prices → confirming the switch out → switching back to book tickets), the continuous decision-making process of the user's query behavior is fully reconstructed. This temporal characteristic is key to judging "cost comparison" (the temporal characteristics of private users often manifest as "repeated price comparison → adjusting the choice," while business users are more "direct in their decision-making"). EA i As a feature carrier of the complete behavioral chain, it provides comprehensive behavioral basis for subsequent model judgment.

[0088] S325, EA i Input the data into a pre-defined comparative travel task value judgment model to obtain the TA. i The corresponding value.

[0089] The pre-defined comparative travel task value judgment model has been trained using historical data (input is similar to EA). i The feature vector is output as an indicator of whether a cost comparison exists. This can identify the association between related features and cost comparison behavior (e.g., "multiple queries on the same route + price filtering adjustments + short intervals between switching in and out" corresponds to "cost comparison exists"). EA i After inputting the model, the model outputs TA. i The value (e.g., "1" indicates that there is a cost comparison behavior, "0" indicates that there is no such behavior).

[0090] By leveraging deep learning of the "cutout-in correlation" model, it accurately captures cost comparison behaviors that are difficult for humans to identify (such as users indirectly comparing prices through related cutout-in queries even if they haven't explicitly stated the price), thus solving the problem that "it's impossible to determine the true intent based on a single query"; the output TA i By quantifying behavioral characteristics using binary identifiers, a clear basis for judgment is provided for subsequent integrated analysis, thereby improving the accuracy of determining whether a trip is for official or private purposes.

[0091] Furthermore, the pre-defined comparative travel task value judgment model can be trained in the following way: We selected historical operation data from a large number of users (covering users with different travel types and operating habits), focused on the "target time period between two travel tasks," and extracted the feature vector (i.e., EA) consistent with steps S321-S324. i The historical corresponding data is used as the sample input.

[0092] Label each sample manually or semi-automatically: If the user exhibits explicit cost comparison behavior during the corresponding operation (such as switching price sorting multiple times, comparing prices across platforms and then returning to the target software, adjusting query conditions to compare prices / refund / change costs of different flights, etc.), label it as "1"; if the behavior is straightforward (such as confirming the query in one go, not paying attention to the price field, only filtering airlines / slots), label it as "0". The labels need to be validated in conjunction with the user's actual travel type (business / private) (e.g., the proportion of private users labeled "1" should be higher than that of business users).

[0093] Based on the above training samples, those skilled in the art can use existing model training methods according to actual needs to obtain a preset comparative travel task cost judgment model, which will not be elaborated here; the initial training model can be a deep learning model such as LSTM or Transformer to capture the dynamic dependencies of behavior.

[0094] S330, Based on A and TA, obtain the list of fusion operation behavior feature vectors B = (B1, B2, ..., B...) corresponding to the target user. i B n ); B i B is the feature vector of the fusion operation behavior corresponding to the i-th operation process; i = (A i TA i ).

[0095] The initial feature vector A i Compare with the corresponding travel task value identifier TA i Combine them to form a fusion vector B i For example, if A i Features including "query period is weekdays 9:00-18:00" and "filter criteria include 'business class'" indicate that TA i If the value is "0" (no cost comparison), then B i This comprehensively reflects the characteristics of "efficient business travel operations + low cost sensitivity", ultimately forming List B.

[0096] Fusion Operation Details (A) i ) and behavioral motivation (TA) i This approach allows features to encompass both "what the user did" and "whether the user cared about the cost," providing a more comprehensive set of dimensions. This fusion can more accurately depict users' travel decision-making preferences (e.g., business travelers are often efficient and not concerned about price), providing the model with richer criteria for judgment.

[0097] S340, according to the chronological order, concatenate the fusion operation behavior feature vectors corresponding to each operation process of B to obtain the target operation behavior feature vector QA corresponding to the target user.

[0098] In chronological order (e.g., from the earliest operation to the most recent operation), all fused vectors in list B (B1 to B...) n These are concatenated sequentially to form a complete temporal feature vector QA. For example, if a user first searches for a "Beijing-Shanghai" flight on Monday (B1) and then searches for the refund and change policy for the same route on Wednesday (B2) within a target time period, then Q will retain the temporal relationship of "Monday operation → Wednesday operation" and completely record the behavior sequence.

[0099] Preserving the temporal correlation of behaviors captures the dynamic patterns of user decision-making. For example, business users may concentrate their ticket bookings on weekdays, while personal users may repeatedly compare prices on weekends. The inclusion of time-series information avoids the one-sidedness of analyzing single operations in isolation, enabling the model to identify more complex behavioral patterns.

[0100] S350 inputs QA into a preset travel task type judgment model to determine whether the target user's current travel task is a business trip or a private trip.

[0101] Input the target vector QA into a preset travel task type judgment model (such as a classification model trained based on user historical data). The model learns the association between "operation behavior sequence + cost sensitivity" and travel type (such as "efficient operation + low cost comparison → official business", "frequent price comparison + weekend operation → private business") and outputs the current travel type judgment result (official business / private business).

[0102] By leveraging the model's ability to uncover complex behavioral patterns, it accurately identifies scenarios that rule-based judgments struggle to cover (such as business travel where tickets are purchased using personal accounts), thus compensating for the accuracy shortcomings of rule-based judgments. Furthermore, the model is trained based on users' own behavioral data, allowing it to adapt to individual differences (such as a user's habit of booking business tickets through personal channels), further enhancing the reliability of its judgments.

[0103] Furthermore, the pre-defined travel task type judgment model can be trained in the following way: Acquire a large amount of historical user data and generate target operation behavior feature vectors (i.e., historical versions of QA) corresponding to "historical travel tasks" according to the process S310-S340. For example, historical QA for official business users may include "query periods are concentrated on weekdays" and "cost comparison identifier TA". i Features such as "mostly 0 (no price)" and "frequent queries of fixed business routes"; historical QA data for private users may include "query periods concentrated on weekends / holidays" and "TA". i Features include "mostly 1 (with price comparison)" and "frequent adjustments to destination and date".

[0104] Each historical QA sample is labeled with its corresponding travel type tag ("Official" or "Private"). Tag sources include: Authoritative data includes user travel expense records (business-related), user-defined notes (private-related), and travel purposes synchronized with the company's travel system.

[0105] Auxiliary verification: Cross-validate the labels by combining them with the types clearly defined in the rule judgment (such as official travel for which tickets are purchased through B2B channels) to ensure the accuracy of the labels.

[0106] Considering that QA is a "temporally concatenated behavioral feature vector" (containing the order of operations and related logic), models adapted to sequence data should be selected first: Transformer / LSTM: excels at capturing temporal features (such as the continuous behavior of "first querying business class → switching to the next page without a price → directly booking a ticket"), suitable for learning the behavioral sequences of "efficient decision-making" by business travelers; Gradient boosting tree (such as XGBoost): if structured features (such as the number of price comparisons, the proportion of business class queries) are high in QA, key differences can be identified through feature importance analysis (such as "business class query proportion > 60%" for business travelers), suitable for handling high-dimensional discrete features.

[0107] Based on the initial training model of the above training samples, those skilled in the art can use existing model training methods according to actual needs to obtain a preset travel task type judgment model, which will not be elaborated here.

[0108] In this embodiment, a highly accurate judgment scheme is provided for users with low rule-judgment adaptability (second-identified users) through the process of "precise behavior collection - cost feature extraction - multi-dimensional fusion - temporal modeling - intelligent judgment". Its advantages are: it can capture the essential differences in behavior between business and private travel (such as cost sensitivity), and it can adapt to complex scenarios through temporal features and model learning, complementing the rule judgment of S200, and ultimately achieving a balance between "efficiency" and "accuracy", providing reliable support for accurate user profiling and differentiated recommendations (such as recommending business services to business users and tourism products to private users).

[0109] This embodiment's travel task type determination method based on civil aviation data fields uses rule-based judgment only when the target user meets the criteria for rule-based judgment; otherwise, it uses behavioral analysis. This allows for accurate determination of whether a user's travel task is for official or private purposes, providing reliable support for building precise user profiles (e.g., travel frequency and preferred routes for official users, destination preferences and time patterns for private users) and conducting differentiated business recommendations (e.g., recommending business hotels and airport express transfer services to official users, and tourist accommodations and attraction tickets to private users). This effectively improves the accuracy of user demand analysis and service matching efficiency. Furthermore, by selecting the corresponding judgment method based on the identifier determined from the user's historical travel judgment results, rule-based judgment is used for users with high rule-based judgment adaptability, fully leveraging its advantages of low computational consumption and high efficiency. Behavioral analysis is used for users with low rule-based judgment adaptability, utilizing its ability to mine historical patterns to ensure accuracy. This balances the relationship between judgment accuracy, computational consumption, and judgment efficiency, solving the problem in existing technologies where a single judgment method struggles to balance accuracy and efficiency.

[0110] Furthermore, in the above embodiments, the preset behavior analysis method in step S300 can also be the method in Embodiment 2.

[0111] Example 2: Based on the method in Embodiment 1, the preset behavior analysis method in step S300 can also be the following method: Q100: Obtain the initial query behavior feature vector corresponding to each operation process of the target user from launching the target software to exiting the target software within the target time period, as well as the preset associated task identifier corresponding to the current travel task; the initial query behavior feature vector is obtained based on the query records of the target user in the travel task query interface of the target software; the target time period is the time period between the time point when the current travel task is established and the time point when the previous travel task is established.

[0112] Focusing on the interval between the "creation time of the current travel task" and the "creation time of the previous travel task," this approach precisely pinpoints the user's decision-making cycle for planning their current trip. It records the user's actions from opening to closing the target software (such as a travel app) within this timeframe, extracting features from the travel task query interface records (such as the type of route queried, date distribution, cabin class filtering preferences, query frequency, etc.) and converting them into structured vectors (e.g., quantitative features such as "percentage of international route queries" and "number of queries on weekdays"). It also collects pre-defined associated task information related to the current trip (such as whether business hotels were booked simultaneously, whether meeting schedules were linked, whether travel approval numbers were added, etc.), recording this information using binary identifiers (e.g., "1" indicates an association, "0" indicates no association).

[0113] In this step, the time range is focused to ensure that the data is strongly correlated with the current travel decision and to eliminate interference from irrelevant behaviors; the initial query vector captures the user's core operating habits, and the associated task identifiers supplement the scenario information (such as "associated meeting itinerary" directly pointing to official business), providing multi-dimensional basic data for subsequent judgment.

[0114] Furthermore, the initial query behavior feature vector corresponding to each operation process of the target user from launching the target software to exiting the target software within the target time period is obtained through the following steps: Q110, Obtain the operation process of the target user from launching the target software to exiting the target software every time within the target time period, to obtain the operation process list F = (F1, F2, ..., F...). i F n ); F i This is the i-th operation process.

[0115] In this embodiment, the implementation of Q110 is the same as the implementation of step S311 in Embodiment 1, and will not be described again here.

[0116] Q120, obtain in F i Each query record corresponding to the travel task query interface of the target software is used to obtain F. i The corresponding query record list W i =(W i，1 W i，2 ,…,W i，a ,…,W i，y（i） ), a=1, 2,…,y(i); W i，a For F i The a-th query record corresponding to the travel task query interface of the target software, y(i) is the value in F i The number of query records corresponding to the travel task query interface of the target software.

[0117] In this embodiment, Q120 differs from step S312 in Embodiment 1 in that S312 obtains all operation records of the target user, including operations such as querying and browsing; while Q120 only obtains the target user's query records, excluding other operation records.

[0118] Regarding the operation process F i Extract all specific query records from the user's travel task query interface (such as the flight query page and itinerary planning page) and form a list W. i Each query record W i,aIt includes core information such as the departure and destination points, date range, cabin class (e.g., economy / business), and filtering criteria (e.g., "on-time performance priority" or "price from low to high"). For example, an F1 ticket corresponding to a W1 ticket might include records such as "Search Beijing-Shanghai, 2025-08-10, business class, on-time performance priority" or "Search Beijing-Shanghai, 2025-08-11, business class".

[0119] Focusing on the core scenario of travel task query, this method accurately captures raw data directly related to travel decisions, eliminating interference from irrelevant operations within the software (such as browsing advertisements). Compared to basic behavioral analysis methods that only extract single operational features (such as query frequency), this step fully preserves query details, providing raw materials for subsequent extraction of deeper features such as "route preference," "cabin class selection," and "filtering logic," thereby improving the targeting of features.

[0120] Q130, for W i The query records in the database are used to extract, encode, and unify the initial query behavior features to obtain the initial query behavior feature vector corresponding to the i-th operation process.

[0121] Feature extraction: from W i Extract key features from the query records, such as "business class query percentage", "number of queries on weekdays", "frequency of queries on fixed routes", and "number of times the on-time rate filter is used", focusing on the differences between business and private travel (e.g., business users may frequently query business class, while private users may frequently adjust dates).

[0122] Encoding: Convert non-quantitative features into numerical values ​​(e.g., "Filtering condition = On-time rate priority" is encoded as "1", "Price priority" is encoded as "0").

[0123] Dimensional uniformity: Standardize the features of different operation processes (e.g., convert the number of queries into "the proportion of total queries") to ensure that the dimension of the feature vector of each initial query behavior is consistent (e.g., all are 12-dimensional).

[0124] In this step, feature extraction focuses on core differences and filters redundant information to solve the problem of insufficient feature targeting in traditional methods; encoding transforms textual features into machine-recognizable numerical values, realizing the "computability" of behavior; dimension unification eliminates feature format differences in different operation processes, ensuring data compatibility when fused with associated task identifiers and cost comparison identifiers, providing standardized input for the construction of the comprehensive feature vector XL1, and avoiding fusion deviations caused by format chaos.

[0125] Q200: Based on the target user's historical query records, determine whether the target user has engaged in comparing the value of travel tasks during each operation, and obtain the comparison travel task value identifier corresponding to each operation.

[0126] Quantifying users' sensitivity to travel costs reveals that business travelers are typically more concerned with efficiency (often marked with "0"), while personal travelers are more concerned with cost (often marked with "1"), providing key behavioral evidence for distinguishing between the two types of travel.

[0127] Furthermore, Q200 may include the following steps: Q210, obtain several historical query records corresponding to each query sub-process in the i-th operation process of the target user, so as to obtain the initial historical query record list set C corresponding to the i-th operation process. i =(C i，1 C i，2 C i，j C i，f（i） ), j=1,2,…,f(i);C i，j Let f(i) be the initial historical query record list corresponding to the j-th query subprocess in the i-th operation process, and f(i) be the number of query subprocesses corresponding to the i-th operation process. The query subprocess is the query process from the start of entering the query interface of the target software to the end of exiting the query interface of the target software.

[0128] Q220, C i，r Several consecutive historical query records before the cut-out time point and C i，r+1 A list of intermediate historical query records is formed by several consecutive historical query records following the entry point, in order to obtain C. i The corresponding intermediate historical query record list set D i =(D i，1 D i，2 D i，r D i，f（i）-1 ), r=1, 2,…, f(i)-1; D i，r C i，r and C i，r+1 The corresponding list of intermediate historical query records; each historical query record includes time information.

[0129] Q230, regarding D i Query behavior features are extracted from each intermediate historical query record list to obtain D. i The corresponding query behavior feature vector E i =(E i，1 E i，2 , ..., E i，r , ..., E i，f（i）-1 ); E i，r D i，r The corresponding query behavior feature vector; E i Each query behavior in the dataset has the same dimension in its feature vector.

[0130] Q240, in chronological order, place E i Each query behavior feature vector in the process is concatenated to obtain the target query behavior feature vector EA corresponding to the i-th operation. i .

[0131] Q250, EA i Input the data into the preset comparative travel task value judgment model to obtain the comparative travel task value identifier corresponding to the i-th operation process.

[0132] In this embodiment, the method for determining whether the target user has engaged in comparing the value of travel tasks during each operation and obtaining the comparison travel task value identifier corresponding to each operation is the same as the method in steps S321-S325 of Embodiment 1, and will not be described in detail here.

[0133] Q300 generates a comprehensive feature vector XL1 corresponding to the current travel task based on the feature vector of each initial query behavior, the identifier of each preset associated task corresponding to the current travel task, and the identifier of the comparative travel task value corresponding to each operation process.

[0134] In this embodiment, after obtaining the feature vector of each initial query behavior, the identifier of each preset associated task corresponding to the current travel task, and the identifier of the comparative travel task value corresponding to each operation process, the comprehensive feature vector XL1 corresponding to the current travel task is obtained by concatenation.

[0135] For example: the initial query vector contains "international route query percentage = 80%" and "number of queries on weekdays = 5"; the associated task identifier contains "meeting itinerary association = 1"; the cost comparison identifier is "0"; after fusion, XL1 comprehensively reflects the official characteristics of "high-frequency international queries + associated meetings + unbeatable price".

[0136] This step integrates three core features—operational habits, scenario association, and cost sensitivity—to comprehensively characterize the user's travel decision-making logic, avoiding the one-sidedness of a single feature and providing rich input for the first judgment model.

[0137] Q400: Obtain the text browsing feature vector XL2 corresponding to the target user within the target time period; the text browsing feature vector is obtained based on the target user's historical text browsing records within the target time period.

[0138] Extract users' historical text browsing records within the target time period (such as content viewed in the target software, such as "Business Travel Guides," "Airport VIP Lounge Services," and "Travel Attraction Guides"), extract features, and convert them into XL2 vectors. For example, features such as "Business-related text browsing time percentage = 90%" and "Travel guide browsing count = 0" are used to quantify text preferences.

[0139] By capturing users' implicit travel purposes—business users tend to browse business service content, while personal users tend to browse travel-related content—we can supplement decision-making clues beyond operational behaviors and enhance the comprehensiveness of judgment dimensions.

[0140] Furthermore, the historical text browsing records include: text type, browsing duration, whether text was entered, and the text information entered.

[0141] When acquiring the target user's historical text browsing records within the target time period, we focus on collecting four types of core information and performing feature transformation.

[0142] First, the text type, which is the attribute of the content that the user browses. For example, "Business Travel Guide" and "Airport VIP Service Introduction" are classified as business (coded as 1), while "Tourist Attraction Guide" and "Homestay Recommendation" are classified as tourism (coded as 0).

[0143] Secondly, browsing time is calculated by tracking the time users spend on various types of text and determining the proportion of browsing time for business texts to the total browsing time (e.g., if business texts are browsed for 30 minutes and the total browsing time is 40 minutes, the proportion is 0.75).

[0144] Third, whether text is entered: If keywords such as "meeting room booking" or "travel expense reimbursement" are entered on the business text page, it is marked as 1 (related input exists); if keywords such as "family-friendly attractions" or "food recommendations" are entered on the tourism text page, or if no input is entered, it is marked as 0.

[0145] Fourth, the input text information is used to extract keywords for semantic analysis. For example, the input "business class refund and rebooking" is determined to be business-related (coded as 1), and the input "self-driving tour route" is determined to be tourism-related (coded as 0). Finally, these quantified features are integrated to form the text browsing feature vector XL2 (e.g., XL2 = (text type = 1, business browsing time percentage = 0.75, input text = 1, input semantics = 1)).

[0146] By refining the core dimensions of historical text browsing records, we can deeply uncover clues about travel purposes hidden in user behavior. Text type and browsing duration directly reflect user content preferences, while whether text is entered and the information entered further strengthens the authenticity of preferences (e.g., actively entering business-related content is more indicative of a business-related inclination than simply browsing). This multi-dimensional feature extraction overcomes the limitations of traditional basic behavioral analysis methods that rely solely on query operation features, transforming text browsing behavior from "worthless information" into "judgment criteria." Simultaneously, the quantified features can be directly used to construct the text browsing feature vector XL2, providing precise input for the second judgment model in Q500. This enables the model to more accurately identify implicit travel types (e.g., business users who purchase tickets through personal channels but frequently browse business texts), significantly improving the comprehensiveness and accuracy of the judgment.

[0147] Q500: Input XL1 into the preset first judgment model to obtain the first confidence level η1 that the current travel task is for official business; and input XL2 into the preset second judgment model to obtain the second confidence level η2 that the current travel task is for official business.

[0148] The first judgment model: Taking XL1 as input, it outputs the probability η1 of the current travel being official (e.g., 0.85 means 85% probability of being official) through a trained classification model (such as Transformer, which is good at capturing multi-feature associations).

[0149] The second judgment model: taking XL2 as input, outputs the probability η2 of "cause of public" through a text classification model (such as BERT, adapted to text features) (e.g., 0.9 means 90% probability of "cause of public").

[0150] The two models focus on the features of "operation + scenario" and "text preference" respectively. The output confidence quantifies the degree of support for "official business" in different dimensions, providing a refined basis for subsequent integration.

[0151] Furthermore, the pre-defined first judgment model can be trained in the following way: Acquire a large amount of historical travel data from users and generate a comprehensive feature vector XL1 corresponding to the historical travel tasks according to the Q100-Q300 process (including historical initial query behavior features, associated task identifiers, and cost comparison identifiers). For example, the XL1 of historical business travel may include feature combinations such as "business class query ratio 80% + meeting association identifier 1 + cost comparison identifier 0".

[0152] Each XL1 sample was labeled with its actual travel type ("business" or "private"). The labels were sourced from user expense records, corporate travel system data, and user-generated labels to ensure accuracy.

[0153] Prioritize models that can handle multi-dimensional feature associations (such as gradient boosting trees XGBoost and deep learning models MLP), as they can effectively integrate heterogeneous features such as operational behavior (such as query frequency), scene association (such as hotel booking identifiers), and cost sensitivity (such as price comparison identifiers); then train the model using existing model training methods to obtain the first judgment model.

[0154] The pre-defined second judgment model can be trained in the following way: Acquire a large amount of historical data from users and generate text browsing feature vectors (XL2) corresponding to historical travel tasks according to the Q400 process (including text type, browsing time percentage, input text identifier, and semantic features). For example, the XL2 for historical business travel might include feature combinations such as "90% business text percentage + 0.8% business browsing time percentage + 1 semantic identifier for inputting 'travel reimbursement'"; the XL2 for personal travel might include "85% tourism text percentage + 0.7% tourism browsing time percentage + 1 semantic identifier for inputting 'homestay booking'".

[0155] Each XL2 sample is labeled with a real travel type label ("official" or "private"), and the label source is consistent with the first judgment model (such as expense records, corporate data, etc.) to ensure accurate correlation with text features.

[0156] Prioritize models that are well-suited to text features (such as a lightweight version of the text classification model BERT, or an attention-based MLP), as they can effectively capture the deep correlation between text type and semantic input (such as the synergy between browsing "business guide" and inputting "meeting booking" pointing to business travel); then train the model using existing model training methods to obtain the second judgment model.

[0157] Q600, weighted fusion of η1 and η2, yields the target confidence η of the current travel task being official travel.

[0158] Furthermore, the Q600 may include the following steps: Q610, based on the number of historical query records NUM1 and the number of historical text browsing records NUM2 of the target user within the target time period, determine the first weight α1 corresponding to η1 and the second weight α2 corresponding to η2; α1+α2=1.

[0159] Furthermore, α1 = NUM1 / (NUM1 + NUM2); α2 = NUM2 / (NUM1 + NUM2).

[0160] The system calculates the weights of the target user's historical query records (NUM1, such as the total number of travel-related queries like flights and hotels) and historical text browsing records (NUM2, such as the total number of times they browsed business guides and travelogues) within the target time period. The weights are then calculated using the formulas α1 = NUM1 / (NUM1 + NUM2) and α2 = NUM2 / (NUM1 + NUM2). For example, if a user queries 12 times and browses text 8 times within the target time period, then α1 = 12 / (12 + 8) = 0.6 and α2 = 8 / (12 + 8) = 0.4, and the sum of the two is always 1.

[0161] The weights are directly determined by the amount of user behavior data, reflecting the logic that "the richer the data, the more reliable the features." If there are more user query records (larger NUM1), the initial query behavior feature vector XL1 is more representative, and α1 increases accordingly. If there are more text browsing records (larger NUM2), the text browsing feature vector XL2 has higher reference value, and α2 increases accordingly. This dynamic weight allocation avoids the subjectivity of fixed weights (such as α1=0.5, α2=0.5), ensuring a high degree of match between the weights and the reliability of actual user behavior features, providing a more reasonable basis for subsequent fusion.

[0162] Q620, based on α1 and α2, weighted fusion of η1 and η2 is obtained, resulting in η = α1 × η1 + α2 × η2.

[0163] By using weighted fusion, the final target confidence η is made to take into account the contributions of both query behavior and text browsing behavior, and the contribution ratio is dynamically adjusted by the amount of data: For users with active query behavior (large NUM1) but little text browsing (small NUM2) (such as business people who frequently book but rarely browse guides), η is mainly dominated by query features to avoid bias caused by insufficient text data; For users with rich text browsing (large NUM2) but simple query behavior (small NUM1) (such as users who quickly book tickets after browsing a lot of business guides in advance), η refers more to text features to make up for the lack of query information.

[0164] This fusion approach retains the advantages of dual-model judgment while solving the problem of insufficient reliability of single features through data-driven weight allocation. The final output η can more objectively and comprehensively reflect the current travel type trend, providing accurate basis for Q700 judgment.

[0165] Q700, if η≥η', then the current travel task is determined to be official travel; otherwise, the current travel task is determined to be private travel; η' is the preset target confidence threshold.

[0166] The preset target confidence threshold η' (e.g., 0.7, which can be adjusted according to the business's need for accuracy) is: if η≥0.7 (e.g., 0.87 above), it is determined to be official travel; if η<0.7, it is determined to be private travel.

[0167] By clearly defining the boundaries through thresholds, we can ensure that the results meet the business's accuracy requirements for determining "official" reasons (e.g., setting a high threshold can reduce the probability of misjudging private reasons as official reasons). The output results can directly support user profile updates and differentiated service recommendations.

[0168] η' can be determined in the following way: Collect the η values ​​of all historical samples (with the real type labeled). For example, the η values ​​of 1000 real official samples are concentrated in the range of 0.6-0.9, and the η values ​​of 1000 real private samples are concentrated in the range of 0.1-0.4.

[0169] Plot the distribution curve of "η value - true type" to find the point where the two types of samples overlap the least (e.g., when η=0.5, 90% of the public service samples have η≥0.5, and 95% of the private service samples have η<0.5).

[0170] If this point is set to η' (e.g., η'=0.5), then most samples can be distinguished with a lower false positive rate.

[0171] This method ensures that the threshold aligns with the distribution patterns of actual data, avoiding judgment biases caused by subjective settings.

[0172] Furthermore, prior to step Q100, the method may further include the following steps: Q010, obtain the number of historical text browsing records of the target user within the target time period, NUM2.

[0173] Within the target time period, count the total number of historical text browsing records generated by the target user in the target software, i.e., NUM2. For example, if a user browsed texts such as "Business Travel Guide" and "Travel Attraction Guide" within this time period, and the total number of browsing records is 3, then NUM2=3.

[0174] By quantifying the number of text browsing records, we can quickly determine whether the text features are valuable. Text records are the core input of the second judgment model (Q500). If the number is too small, the XL2 features will be one-sided, thus affecting the reliability of η2. This step provides clear data basis for whether to use the backup judgment logic in the future, avoiding the forced use of dual-model fusion when there is insufficient text data, and reducing judgment bias from the source.

[0175] Q020, if NUM2 < NUM', then the preset behavior analysis method is used to determine whether the current travel task is for official or private purposes; NUM' is the preset threshold for the number of historical text browsing records.

[0176] NUM' is the minimum effective threshold for the number of text browsing records (e.g., NUM'=5, which can be set according to the minimum number of records in historical data that "text features can stably support the judgment"). If NUM2 < NUM' (e.g., the user only browses 2 texts, which cannot reflect the true preference), then the dual-model fusion process of Q400-Q600 is skipped, and the preset behavior analysis method is used directly; the preset behavior analysis method is the method in S300 of Example 1.

[0177] In this embodiment, by integrating initial query behavior features with associated task identifiers (such as business hotel bookings, conference itinerary associations, etc.) and combining cost comparison behavior analysis, the method comprehensively captures operational and scenario information related to travel decisions. Simultaneously, it incorporates text browsing features (such as browsing preferences in business guides or travel itineraries) to supplement implicit clues about travel purposes, thus addressing the problem of one-sided characterization in basic behavior analysis methods. Furthermore, by using two models to output the confidence scores of corresponding features and performing weighted fusion, the method fully considers the differences in the contribution of different features to the judgment, avoiding the limitations of a single model and significantly improving the accuracy of judgments for atypical scenarios (such as business travel with tickets purchased through personal channels). Ultimately, this method can provide reliable travel type judgment results for accurate user profile construction and differentiated business recommendations, balancing the comprehensiveness and reliability of the judgment.

[0178] Furthermore, although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0179] Embodiments of the present invention also provide a non-transitory computer-readable storage medium that can be disposed in an electronic device to store at least one instruction or at least one program related to implementing a method in the method embodiments, wherein the at least one instruction or the at least one program is loaded and executed by the processor to implement the method provided in the above embodiments.

[0180] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0181] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting programs for use by or in conjunction with an instruction execution system, apparatus, or device.

[0182] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0183] Program code for performing the operations of this application can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0184] Embodiments of the present invention also provide an electronic device, including a processor and the aforementioned non-transitory computer-readable storage medium.

[0185] The electronic device is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments in this application.

[0186] Electronic devices are manifested in the form of general-purpose computing devices. Components of an electronic device may include, but are not limited to: at least one processor, at least one memory, and a bus connecting different system components (including memory and processor).

[0187] The memory stores program code that can be executed by the processor, causing the processor to perform the steps in the various embodiments described in this specification.

[0188] The memory may include readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory, and may further include read-only memory (ROM).

[0189] The memory may also include programs / utilities having a set (at least one) of program modules, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0190] A bus can represent one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus that uses any of the various bus structures.

[0191] Electronic devices can also communicate with one or more external devices (e.g., keyboards, pointing devices, Bluetooth devices, etc.), one or more devices that enable user interaction with the electronic device, and / or any device that enables the electronic device to communicate with one or more other computing devices (e.g., routers, modems, etc.). This communication can be achieved through input / output (I / O) interfaces. Furthermore, electronic devices can communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via network adapters. The network adapter communicates with other modules of the electronic device via a bus. It should be understood that other hardware and / or software modules can be used in conjunction with the electronic device, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0192] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0193] Embodiments of the present invention also provide a computer program product including program code, which, when the program product is run on an electronic device, causes the electronic device to perform the steps of the methods described above in various exemplary embodiments of the present invention.

[0194] While specific embodiments of the invention have been described in detail by way of examples, those skilled in the art should understand that the examples are for illustrative purposes only and are not intended to limit the scope of the invention. Those skilled in the art should also understand that various modifications can be made to the embodiments without departing from the scope and spirit of the invention.

Claims

1. A method for determining travel task type based on civil aviation data fields, characterized in that, The method includes the following steps: S100, Obtain the target user's judgment method identifier QR; QR is determined by using different judgment methods to judge the type of several historical travel tasks of the target user. S200, if QR is the preset first identifier, then the preset rule judgment method and the civil aviation data field corresponding to the current travel task are used to judge the type of the target user's current travel task, so as to determine whether the target user's current travel task is a business travel type or a private travel type. Step S200 includes the following steps: S210, Obtain the key fields corresponding to the current travel task; the key fields include the task creation platform, key user ID, value-related fields, user and itinerary attribute fields; S220, if the task creation platform is a preset task creation platform, or the key user number is not empty, or the value-related field corresponds to a preset organizational structure field, then the type of the target user's current travel task is determined to be official travel; otherwise, a second judgment is made based on the auxiliary identifier corresponding to the current travel task; wherein, the auxiliary identifier includes the identifiers corresponding to the user and itinerary attribute fields; S300, if the QR is a preset second identifier, then a preset behavior analysis method is used to determine the type of the target user's current travel task, so as to determine whether the target user's current travel task is a business trip or a private trip; the first identifier and the second identifier are different; Step S300 includes the following steps: S310, Obtain the initial operation behavior feature vectors corresponding to each operation process of the target user from starting the target software to exiting the target software within the target time period, and obtain the initial operation behavior feature vector list A = (A1, A2, ..., A...). i A n ), i=1,2,…,n; A i The initial operation behavior feature vector corresponding to the i-th operation process is obtained, where n is the number of operation processes; the target time period is the time period between the time point when the current travel task is established and the time point when the previous travel task is established. S320, based on the target user's historical query records, determine whether the target user engaged in comparing the cost of travel tasks during each operation, and obtain the list of comparison travel task cost identifiers TA = (TA1, TA2, ..., TA...). i , ...,TA n ), TA i For A i Corresponding comparative travel task value identifier; S330, Based on A and TA, obtain the list of fusion operation behavior feature vectors B = (B1, B2, ..., B...) corresponding to the target user. i B n ); B i B is the feature vector of the fusion operation behavior corresponding to the i-th operation process; i = (A i TA i ); S340, according to the chronological order, concatenate the fused operation behavior feature vectors corresponding to each operation process of B to obtain the target operation behavior feature vector QA corresponding to the target user; S350: Input QA into the preset travel task type judgment model to determine whether the target user's current travel task is a business trip or a private trip. QR codes are determined through the following steps: S110, Obtain information on each historical travel task of the target user within a preset historical time period; the type of travel task corresponding to each historical travel task is known. S120, use a preset rule judgment method to judge the travel task type corresponding to each historical travel task information to obtain the first judgment accuracy. S130, if the accuracy of the first judgment is greater than the preset accuracy threshold, then QR is determined to be the preset first identifier; otherwise, QR is determined to be the preset second identifier.

2. The method for determining travel task type based on civil aviation data fields according to claim 1, characterized in that, TA i The corresponding value is determined through the following steps: S321, Obtain several historical query records corresponding to each query sub-process of the target user in the i-th operation process, so as to obtain the initial historical query record list set C corresponding to the i-th operation process. i =(C i，1 C i，2 C i，j C i，f（i） ), j=1,2,…,f(i);C i，j Let f(i) be the initial historical query record list corresponding to the j-th query sub-process in the i-th operation process, and f(i) be the number of query sub-processes corresponding to the i-th operation process. The query sub-process is the query process that starts from entering the query interface of the target software and ends when exiting the query interface of the target software. S322, C i，r Several consecutive historical query records before the cut-out time point and C i，r+1 A list of intermediate historical query records is formed by several consecutive historical query records following the entry point, in order to obtain C. i The corresponding intermediate historical query record list set D i =(D i，1 D i，2 D i，r D i，f（i）-1 ), r=1, 2,…, f(i)-1; D i，r C i，r and C i，r+1 The corresponding list of intermediate historical query records; each historical query record includes time information; S323, for D i Query behavior features are extracted from each intermediate historical query record list to obtain D. i The corresponding query behavior feature vector E i =(E i，1 E i，2 , ..., E i，r , ..., E i，f（i）-1 ); E i，r D i，r The corresponding query behavior feature vector; E i Each query behavior in the dataset has the same dimension in its feature vector. S324, in chronological order, E i Each query behavior feature vector in the process is concatenated to obtain the target query behavior feature vector EA corresponding to the i-th operation. i ; S325, EA i Input the data into a pre-defined comparative travel task value judgment model to obtain the TA. i The corresponding value.

3. The method for determining travel task type based on civil aviation data fields according to claim 1, characterized in that, Step S310 includes the following steps: S311, Obtain the operation process of the target user from starting the target software to exiting the target software each time within the target time period, to obtain the operation process list F = (F1, F2, ..., F...). i F n ); F i This is the i-th operation process; S312, Obtain the target user's information in F i Record every operation of the target software to obtain F i Corresponding operation record list H i =(H i，1 H i，2 H i，u H i，g（i） ), u=1,2,…,g(i);H i，u For target users in F i In the context of the target software, the u-th operation record is the record of the target user in F. i The number of operation records corresponding to the target software; each operation record includes the operation time, operation page, page dwell time, and input fields; S313, for H i The operation records in the database are used to extract, encode, and unify the initial operation behavior features to obtain A. i .

4. The method for determining travel task type based on civil aviation data fields according to claim 3, characterized in that, After step S311 and before step S312, the method further includes the following steps: S31, obtain F v The corresponding exit time point t of the target software v and F v+1 The corresponding startup time point t of the target software v+1 v = 1, 2, ..., n-1; S32, if |t v -t v+1 If |≤Δt, then F v and F v+1 Combined into one operation process; Δt is the preset time interval threshold for the operation process.

5. The method for determining travel task type based on civil aviation data fields according to claim 2, characterized in that, The historical query records include: the travel time, departure point, and destination corresponding to the travel task.

6. A non-transitory computer-readable storage medium, wherein the storage medium stores at least one instruction or at least one program segment, characterized in that, The at least one instruction or the at least one program segment is loaded and executed by the processor to implement the travel task type determination method based on civil aviation data fields as described in any one of claims 1-5.

7. An electronic device, characterized in that, Includes a processor and the non-transitory computer-readable storage medium as described in claim 6.

Images