An artificial intelligence-based travel scene advertisement accurate placement optimization system

Abstract

This invention discloses an AI-based system for optimizing precise advertising placement in business travel scenarios. The system includes: a multi-source data acquisition module for generating a business travel behavior observation sequence; a boundary segmentation processing module for outputting segmented sequences of business travel stages; a duration distribution modeling module for constructing a hidden semi-Markov model and establishing the initial duration distribution corresponding to each business travel stage; a duration elastic correction module for performing duration elastic correction processing and generating a corrected duration distribution; a state decoding and determination module for performing state decoding on the business travel behavior observation sequence and generating a dwell depth interval identifier; an advertising trigger control module for generating advertising placement decision results; and a feedback iterative update module for updating the processing results of stage boundary gating, duration elastic correction, and trigger control. This invention improves the accuracy of business travel stage identification and the matching degree of advertising placement.

Description

Technical Field

[0001] This invention relates to the field of digital technology for business travel services, and in particular to an artificial intelligence-based system for precise targeting and optimization of advertising in business travel scenarios. Background Technology

[0002] With the development of internet advertising technology, recommendation algorithms, and digital management technologies for corporate travel, advertising methods have gradually evolved from static rule-based targeting to data-driven delivery. In travel service scenarios, platforms typically push advertising content to users around aspects such as flight bookings, hotel check-ins, transportation, meeting arrangements, dining, and return travel services to improve advertising conversion efficiency and extend the value of travel services. Existing technologies commonly use user browsing history, long-term interest tags, order information, or location data to match and push advertising content; some solutions also combine time windows, geofencing, or simple behavioral sequences for scene recognition, thereby achieving advertising recommendations tailored to business travelers.

[0003] However, existing technologies still have significant shortcomings in business travel scenarios. Business travel behavior is characterized by significant stages and continuity. Users' needs change rapidly and vary significantly in duration at different stages, such as departure preparation, waiting for flights or trains, transit, arrival and transfer, check-in and stay, meetings and activities, and return journey waiting. Existing technologies usually focus more on identifying behaviors or static tags at a single point in time, lacking effective modeling of the duration of business travel stages. As a result, although advertising systems can roughly identify that users belong to the business travel population, they are unable to further accurately determine the current stage of the user's business travel and the intensity of their corresponding needs.

[0004] Therefore, how to provide an AI-based advertising precision targeting and optimization system for business travel scenarios is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an AI-based advertising precision targeting optimization system for business travel scenarios. This invention models the duration of business travel stages based on a hidden semi-Markov model, and combines elastic duration correction and dwell depth interval determination to achieve collaborative optimization of business travel stage identification and advertising trigger control. This can improve the accuracy of advertising timing determination, advertising content matching, and feedback iterative adjustment capabilities in business travel scenarios, and reduce the invalid reach rate.

[0006] An AI-based advertising precision targeting optimization system for business travel scenarios, according to an embodiment of the present invention, includes: The multi-source data acquisition module is used to collect multi-source observation data from business travelers in business travel scenarios, and to perform data preprocessing to generate business travel behavior observation sequences. The boundary segmentation processing module is used to perform phase boundary gating, buffer confirmation, and boundary rollback on the travel phase observation sequence, and output the travel phase segmented sequence; The duration distribution modeling module is used to construct a hidden semi-Markov model based on the segmented sequence of travel stages. The travel behavior observation sequence is input into the hidden semi-Markov model to establish the initial duration distribution corresponding to the travel stages. The duration elastic correction module is used to perform duration elastic correction processing on the hidden semi-Markov model. It generates duration elastic correction parameters based on multi-source observation data and corrects the initial duration distribution to generate the corrected duration distribution. The state decoding and determination module is used to perform state decoding on the travel behavior observation sequence based on the corrected duration distribution, obtain the current travel stage identification result and stage duration characteristics, and generate the stay depth interval identifier. The advertising trigger control module is used to perform trigger control processing on the candidate advertising set based on the current travel stage identification result and the dwell depth interval identifier, and generate advertising placement decision results; The feedback iteration update module is used to collect advertising feedback data corresponding to the advertising placement decision results, and update the processing results of stage boundary gating, duration elastic correction and trigger control based on the advertising feedback data.

[0007] Optionally, the multi-source data acquisition module includes: Collect multi-source observation data from business travelers in business travel scenarios, aggregate the data according to the observation time, and establish a correspondence between various types of data corresponding to the same observation time to form raw observation data units; Perform data preprocessing on the original observation data units, and then re-integrate the processed data to generate cleaned observation data units; The cleaned observation data units are processed by sequence arrangement and state association, and then connected in chronological order to form a difference travel observation sequence.

[0008] Optionally, the boundary segmentation processing module includes: The read difference travel is the observation sequence. Order status change information, location trajectory change information, page interaction change information, and external disturbance event change information are extracted sequentially between adjacent observation data units. The change information is weighted and summed to generate candidate boundary judgment values. At the same time, the candidate boundary judgment values ​​are compared with the preset gate threshold to generate candidate boundary judgment results. The candidate boundary determination results are buffered and confirmed. Candidate boundary positions that meet the preset gate threshold are marked as boundary positions to be confirmed. A preset number of observation data units are read after the current boundary position to be confirmed. The corresponding candidate boundary determination values ​​are averaged to generate buffered confirmation values. The buffered confirmation values ​​are compared with the preset confirmation threshold to confirm the effective stage boundary. Boundary rollback processing is performed on the effective stage boundaries. A preset number of observation data units are read from each effective stage boundary. The corresponding candidate boundary judgment values ​​are averaged to generate rollback verification values. The rollback verification values ​​are compared with preset rollback thresholds to cancel or retain the corresponding effective stage boundaries. Based on the retained effective stage boundaries, the travel behavior observation sequence is segmented to output the travel stage segment sequence.

[0009] Optionally, the duration distribution modeling module includes: Read the travel stage segment sequence and travel behavior observation sequence, number each stage segment in the travel stage segment sequence according to the time sequence, and arrange each observation data unit in the travel behavior observation sequence according to the time sequence. Align each stage segment with the observation data unit within the corresponding time range, so that each stage segment corresponds to a set of continuous observation data units, forming stage input data. Based on the stage input data, a hidden state system of a hidden semi-Markov model is constructed. According to the connection order of each stage segment in the travel stage segment sequence, the connection relationship between the previous stage segment and the next stage segment is extracted, and each connection relationship is written into the corresponding state transition relationship in the hidden semi-Markov model. At the same time, the connection relationships between all stage segments are summarized and organized to generate the state transition relationship between travel stages. Based on the stage input data, the continuous length of the stage segment corresponding to each hidden state is statistically determined. The number of observation data units that appear consecutively in the difference behavior observation sequence or the length of time that they are continuously covered is taken as the duration value of the current hidden state. All duration values ​​under the same hidden state are collected, sorted and frequency counted. Based on the statistical results, an initial duration distribution corresponding to each hidden state is generated. The state transition relationship and the initial duration distribution are written into the hidden semi-Markov model.

[0010] Optionally, the duration elastic correction module includes: Read the initial duration distribution of the hidden semi-Markov model and simultaneously read multi-source observation data. Extract order status change information, location trajectory change information, time node change information, page interaction change information, advertising feedback change information, and external disturbance event change information corresponding to each travel stage according to the observation time sequence. Collect and associate the extracted change information to generate duration correction input data corresponding to each travel stage. Based on the continuous duration correction input data, the continuous duration elastic correction parameter generation process is performed on each travel stage. For each travel stage, the corresponding delay disturbance intensity, stay fluctuation intensity, itinerary change intensity, interaction activity intensity, and feedback response intensity are extracted, and the intensity information is combined according to the preset correspondence to generate the continuous duration elastic correction parameters corresponding to each travel stage. Based on the flexible duration correction parameter, the initial duration distribution corresponding to each travel stage is elastically corrected. The duration intervals in the initial duration distribution are lengthened, shortened, or the distribution center of gravity is shifted to generate the corrected duration distribution corresponding to each travel stage.

[0011] Optionally, the state decoding and determination module includes: The travel journey observation sequence, state transition relationship, and corrected duration distribution are read. Each observation data unit in the travel journey observation sequence is input into the hidden semi-Markov model in chronological order. The hidden semi-Markov model matches the travel journey stage to which each observation data unit belongs at the current time position according to the state transition relationship and corrected duration distribution corresponding to each travel stage, and generates stage decoding results corresponding to each continuous observation interval. The path filtering process is performed on the stage decoding results. According to the sequential connection of each continuous observation interval in the travel behavior observation sequence, the continuity verification and stage affiliation verification of each stage decoding result are performed. The target stage decoding path corresponding to the current observation position is determined from all stage decoding results. The current travel stage identification result corresponding to the current observation position is determined based on the target stage decoding path. Based on the current travel phase identification results, the continuous coverage range of the current travel phase in the travel behavior observation sequence is extracted, the number of observation data units corresponding to the current travel phase and the duration of continuous coverage are counted, the actual duration result of the current travel phase is generated, and the duration values ​​in the corrected duration distribution are extracted sequentially and statistically distributed to generate the distributed duration result of the current travel phase. Based on the actual duration result and the distributed duration result, the phase duration feature is generated. The stage persistence features are processed to determine the dwell depth. The stage persistence features are mapped to the corresponding dwell depth intervals according to the preset dwell depth division rules, and a dwell depth interval identifier is generated based on the mapped dwell depth interval.

[0012] Optionally, the advertisement triggering control module includes: Extract the current travel stage identification result and the dwell depth interval identifier, match the current travel stage identification result with the stage adaptation information in the advertising resource pool, match the dwell depth interval identifier with the depth adaptation information in the advertising resource pool, and filter to obtain a candidate advertising set; Trigger control processing is performed on the candidate ad set. Based on the stage matching result, depth matching result and feedback constraint result corresponding to each ad item, suppression processing is performed on ad items that do not meet the trigger conditions, and retention processing is performed on ad items that meet the trigger conditions, generating a trigger release set and the corresponding trigger control result. Based on the trigger release set and trigger control results, the system performs placement sorting and placement configuration processing on each ad item to determine the placement order, timing, and content of each ad item, and then integrates these to generate ad placement decision results.

[0013] Optionally, the advertisement triggering control module includes: Collect advertising feedback data corresponding to the advertising placement decision results, and associate and sequentially collect various advertising feedback data to generate feedback data units that correspond one-to-one with each advertising placement decision result, forming an advertising feedback data sequence. Based on the advertising feedback data sequence, each feedback data unit is back-matched with the corresponding travel stage segment sequence and the corrected duration distribution, and correction processing is performed within the corresponding matching range to generate the updated travel stage segment sequence and the updated corrected duration distribution. Based on the advertising feedback data sequence, each feedback data unit is back-matched with the advertising entries in the corresponding advertising placement decision results, and correction processing is performed within the corresponding matching range to generate updated advertising placement decision results. The updated advertising placement decision results are then used as the basis for the next round of advertising placement decision processing.

[0014] The beneficial effects of this invention are: This invention performs stage boundary gating, buffer confirmation, and boundary rollback processing on the business travel behavior observation sequence, which can output the business travel stage segment sequence more accurately. This reduces the misjudgment caused by short-term location changes, page jumps, or occasional disturbances in stage identification, making the stage division in business travel scenarios more stable.

[0015] This invention establishes the initial duration distribution between travel stages based on a hidden semi-Markov model, and flexibly corrects the duration distribution by combining multi-source observation data. This allows the continuous process of travel stages to be included in the modeling scope, effectively improving the ability to identify the user's current travel stage and its continuous status.

[0016] In the state decoding process, this invention not only obtains the current travel stage identification result, but also further generates stage persistence features and dwell depth interval identifiers. This enables ad trigger control to no longer rely solely on static tags or single point-in-time behavior, but to dynamically adjust based on the user's dwell state during the travel stage, thereby improving the accuracy of ad trigger timing and ad content matching.

[0017] Based on the current travel stage identification results and the dwell depth interval identifier, this invention performs trigger control processing on the candidate advertisement set, which can generate more targeted advertising placement decision results and reduce ineffective reach and waste of placement resources.

[0018] This invention collects advertising feedback data corresponding to the advertising placement decision results, and iteratively updates the stage boundary gating, duration elastic adjustment and trigger control to form a closed-loop optimization mechanism for business travel scenarios, thereby improving the system's adaptive adjustment and continuous optimization capabilities in complex business travel environments. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of an AI-based advertising precision targeting optimization system for business travel scenarios proposed in this invention; Figure 2 This is a schematic diagram of the core algorithm of an AI-based advertising precision targeting optimization system for business travel scenarios proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figures 1-2 An AI-based advertising precision targeting and optimization system for business travel scenarios includes: The multi-source data acquisition module is used to collect multi-source observation data from business travelers in business travel scenarios, and to perform data preprocessing to generate business travel behavior observation sequences. The boundary segmentation processing module is used to perform phase boundary gating, buffer confirmation, and boundary rollback on the travel phase observation sequence, and output the travel phase segmented sequence; The duration distribution modeling module is used to construct a hidden semi-Markov model based on the segmented sequence of travel stages. The travel behavior observation sequence is input into the hidden semi-Markov model to establish the initial duration distribution corresponding to the travel stages. The duration elastic correction module is used to perform duration elastic correction processing on the hidden semi-Markov model. It generates duration elastic correction parameters based on multi-source observation data and corrects the initial duration distribution to generate the corrected duration distribution. The state decoding and determination module is used to perform state decoding on the travel behavior observation sequence based on the corrected duration distribution, obtain the current travel stage identification result and stage duration characteristics, and generate the stay depth interval identifier. The advertising trigger control module is used to perform trigger control processing on the candidate advertising set based on the current travel stage identification result and the dwell depth interval identifier, and generate advertising placement decision results; The feedback iteration update module is used to collect advertising feedback data corresponding to the advertising placement decision results, and update the processing results of stage boundary gating, duration elastic correction and trigger control based on the advertising feedback data.

[0022] In this embodiment, the multi-source data acquisition module includes: Collect multi-source observation data of business travelers in business travel scenarios. The multi-source observation data includes order status data, location trajectory data, time node data, page interaction data, advertising feedback data and external disturbance event data. The data is collected according to the observation time and a correspondence is established between the various types of data corresponding to the same observation time to form the original observation data unit. Data preprocessing is performed on the original observation data units. The data preprocessing includes unifying time records, filling missing fields, removing abnormal fields, removing duplicate fields, and standardizing data formats. The processed data are then re-integrated to generate cleaned observation data units. The cleaned observation data units are processed by sequence arrangement and state association, and then connected in chronological order to form a difference travel observation sequence.

[0023] In this embodiment, the boundary segmentation processing module includes: The read difference travel is the observation sequence. Order status change information, location trajectory change information, page interaction change information, and external disturbance event change information are extracted sequentially between adjacent observation data units. The change information is weighted and summed to generate candidate boundary judgment values. At the same time, the candidate boundary judgment values ​​are compared with the preset gate threshold to generate candidate boundary judgment results. The candidate boundary determination results are buffered and confirmed. Candidate boundary positions that meet the preset gate threshold are marked as boundary positions to be confirmed. A preset number of observation data units are read after the current boundary position to be confirmed. The corresponding candidate boundary determination values ​​are averaged to generate buffered confirmation values. The buffered confirmation values ​​are compared with the preset confirmation threshold to confirm the effective stage boundary. Boundary rollback processing is performed on the effective stage boundaries. A preset number of observation data units are read from each effective stage boundary. The corresponding candidate boundary judgment values ​​are averaged to generate rollback verification values. The rollback verification values ​​are compared with preset rollback thresholds to cancel or retain the corresponding effective stage boundaries. Based on the retained effective stage boundaries, the travel behavior observation sequence is segmented to output the travel stage segment sequence.

[0024] In this embodiment, the duration distribution modeling module includes: Read the travel stage segment sequence and travel behavior observation sequence, number each stage segment in the travel stage segment sequence according to the time sequence, and arrange each observation data unit in the travel behavior observation sequence according to the time sequence. Align each stage segment with the observation data unit within the corresponding time range, so that each stage segment corresponds to a set of continuous observation data units, forming stage input data. Based on the stage input data, a hidden state system of a hidden semi-Markov model is constructed. According to the connection order of each stage segment in the travel stage segment sequence, the connection relationship between the previous stage segment and the next stage segment is extracted, and each connection relationship is written into the corresponding state transition relationship in the hidden semi-Markov model. At the same time, the connection relationships between all stage segments are summarized and organized to generate the state transition relationship between travel stages. Based on the stage input data, the continuous length of the stage segment corresponding to each hidden state is statistically determined. The number of observation data units that appear consecutively in the difference behavior observation sequence or the length of time that they are continuously covered is taken as the duration value of the current hidden state. All duration values ​​under the same hidden state are collected, sorted and frequency counted. Based on the statistical results, an initial duration distribution corresponding to each hidden state is generated. The state transition relationship and the initial duration distribution are written into the hidden semi-Markov model.

[0025] In this embodiment, the duration elastic correction module includes: Read the initial duration distribution of the hidden semi-Markov model and simultaneously read multi-source observation data. Extract order status change information, location trajectory change information, time node change information, page interaction change information, advertising feedback change information, and external disturbance event change information corresponding to each travel stage according to the observation time sequence. Collect and associate the extracted change information to generate duration correction input data corresponding to each travel stage. Based on the continuous duration correction input data, the continuous duration elastic correction parameter generation process is performed on each travel stage. For each travel stage, the corresponding delay disturbance intensity, stay fluctuation intensity, itinerary change intensity, interaction activity intensity, and feedback response intensity are extracted, and the intensity information is combined according to the preset correspondence to generate the continuous duration elastic correction parameters corresponding to each travel stage. Based on the flexible duration correction parameter, the initial duration distribution corresponding to each travel stage is elastically corrected. The duration intervals in the initial duration distribution are lengthened, shortened, or the distribution center of gravity is shifted to generate the corrected duration distribution corresponding to each travel stage.

[0026] In this embodiment, the state decoding and determination module includes: The travel journey observation sequence, state transition relationship, and corrected duration distribution are read. Each observation data unit in the travel journey observation sequence is input into the hidden semi-Markov model in chronological order. The hidden semi-Markov model matches the travel journey stage to which each observation data unit belongs at the current time position according to the state transition relationship and corrected duration distribution corresponding to each travel stage, and generates stage decoding results corresponding to each continuous observation interval. The path filtering process is performed on the stage decoding results. According to the sequential connection of each continuous observation interval in the travel behavior observation sequence, the continuity verification and stage affiliation verification of each stage decoding result are performed. The target stage decoding path corresponding to the current observation position is determined from all stage decoding results. The current travel stage identification result corresponding to the current observation position is determined based on the target stage decoding path. Based on the current travel phase identification results, the continuous coverage range of the current travel phase in the travel behavior observation sequence is extracted, the number of observation data units corresponding to the current travel phase and the duration of continuous coverage are counted, the actual duration result of the current travel phase is generated, and the duration values ​​in the corrected duration distribution are extracted sequentially and statistically distributed to generate the distributed duration result of the current travel phase. Based on the actual duration result and the distributed duration result, the phase duration feature is generated. The stage persistence features are processed to determine the dwell depth. The stage persistence features are mapped to the corresponding dwell depth intervals according to the preset dwell depth division rules, and a dwell depth interval identifier is generated based on the mapped dwell depth interval.

[0027] In this embodiment, the advertisement triggering control module includes: Extract the current travel stage identification result and the dwell depth interval identifier, match the current travel stage identification result with the stage adaptation information in the advertising resource pool, match the dwell depth interval identifier with the depth adaptation information in the advertising resource pool, and filter to obtain a candidate advertising set; Trigger control processing is performed on the candidate ad set. Based on the stage matching result, depth matching result and feedback constraint result corresponding to each ad item, suppression processing is performed on ad items that do not meet the trigger conditions, and retention processing is performed on ad items that meet the trigger conditions, generating a trigger release set and the corresponding trigger control result. Based on the trigger release set and trigger control results, the system performs placement sorting and placement configuration processing on each ad item to determine the placement order, timing, and content of each ad item, and then integrates these to generate ad placement decision results.

[0028] In this embodiment, the advertisement triggering control module includes: Collect advertising feedback data corresponding to the advertising placement decision results, and associate and sequentially collect various advertising feedback data to generate feedback data units that correspond one-to-one with each advertising placement decision result, forming an advertising feedback data sequence. Based on the advertising feedback data sequence, each feedback data unit is back-matched with the corresponding travel stage segment sequence and the corrected duration distribution, and correction processing is performed within the corresponding matching range to generate the updated travel stage segment sequence and the updated corrected duration distribution. Based on the advertising feedback data sequence, each feedback data unit is back-matched with the advertising entries in the corresponding advertising placement decision results, and correction processing is performed within the corresponding matching range to generate updated advertising placement decision results. The updated advertising placement decision results are then used as the basis for the next round of advertising placement decision processing.

[0029] Example 1: To verify the feasibility of this invention in practice, it was applied to a business travel advertising scenario on an enterprise-level business travel service platform. This platform provides frequent travelers with services such as ticket booking, accommodation booking, itinerary management, pick-up and drop-off services, meeting support, and value-added business travel services. During the user's business trip, it pushes advertisements for services such as pick-up and drop-off, accommodation upgrades, business dining, meeting services, rest areas, insurance coverage, and return trip assistance. While existing systems can push advertisements based on order information, basic location boundaries, historical interest tags, and some browsing behavior, they still have significant shortcomings in practical applications. Because business travel has distinct stages and continuity, users constantly switch between states such as departure preparation, waiting, transit, arrival and transfer, check-in, meeting activities, and return trip preparation. Traditional advertising systems typically rely more on static tag recognition or single-point-of-time recognition, making it difficult to accurately identify the boundaries of business travel stages and effectively model the duration changes of different stages. Therefore, in actual deployment, problems such as premature ad triggering, delayed ad triggering, mismatch between ads and the user's current stage, and waste of ad resources often occur. These problems are even more prominent under conditions of disturbance such as delays, queues, congestion, rescheduling, and fluctuations in service supply.

[0030] In this embodiment, the multi-source data acquisition module first collects multi-source observation data from the travel scenario and performs preprocessing operations on the collected data to generate a travel-based observation sequence. The collected data is not limited to single order data, but rather unifies and correlates relevant behaviors of the same travel user during the travel process, enabling subsequent processing to revolve around the continuous travel process. Subsequently, the boundary segmentation processing module performs stage boundary gating, buffer confirmation, and boundary rollback processing on the travel-based observation sequence to output a travel stage segmentation sequence. In this process, the system does not directly switch stages based on a single page jump or single location change. Instead, it first generates candidate boundary judgments through stage boundary gating, then suppresses misjudgments caused by short-term fluctuations through buffer confirmation, and performs boundary rollback when subsequent observations do not meet the conditions, thereby obtaining a more stable travel stage segmentation result. In this way, the problem that stage boundaries are easily disturbed by occasional behaviors in the prior art can be solved.

[0031] After obtaining the segmented sequence of travel stages, the duration distribution modeling module constructs a hidden semi-Markov model based on this sequence and inputs the travel behavior observation sequence into the model. This allows the model to go beyond simply identifying which stage the user "belongs to," and further establish the initial duration distribution corresponding to each travel stage. In other words, the system not only identifies whether the user is currently in a waiting, transferring, or staying state, but also further characterizes how long these states typically last. Next, the duration elastic correction module generates duration elastic correction parameters based on multi-source observation data and corrects the initial duration distribution, generating a corrected duration distribution. This allows the system to dynamically correct the duration of the current stage based on real-time disturbances during the travel process, making the model's reflection of real-world scenarios more accurate. Compared to existing technologies that treat external disturbances merely as ordinary contextual features, this invention, through duration elastic correction, allows disturbance information to directly affect the stage's duration process itself, thereby improving the realism of subsequent state decoding.

[0032] After the corrected duration distribution is generated, the state decoding and determination module performs state decoding on the travel behavior observation sequence based on this distribution, obtaining the current travel stage identification result and stage duration characteristics, and generating a dwell depth interval identifier. The core here is not only determining which stage the user is currently in, but also further determining the extent to which that stage has lasted. This is particularly crucial for advertising systems. For example, even within the same waiting-related stage, users who have just entered the stage and those who have stayed for a longer period have different levels of demand intensity, and traditional systems often fail to reflect this difference, easily leading to triggering timing deviations. This invention further refines the internal structure of the same stage through stage duration characteristics and dwell depth interval identifiers, thereby providing a more accurate basis for subsequent advertising trigger control.

[0033] After obtaining the current travel stage identification result and dwell depth range identifier, the ad triggering control module does not simply select ads directly from the ad resource pool. Instead, it first performs triggering control processing on the candidate ad set based on the current travel stage identification result and dwell depth range identifier, and then generates the ad delivery decision result. Specifically, when the ads are actually delivered, the system first filters them according to the degree of stage adaptation and dwell depth matching, suppressing ad entries that do not match the current travel stage or the current dwell depth. Then, it adjusts the delivery order and delivery configuration of the remaining ad entries, thereby making the ad triggering timing more consistent with the user's actual travel status.

[0034] After the ad campaign is completed, the feedback iteration update module collects the ad feedback data corresponding to the ad campaign decision results, and updates the processing results of the stage boundary gating, duration elastic adjustment and trigger control based on the ad feedback data, forming a closed-loop optimization process for business travel scenarios. In this embodiment, the ad feedback data is not only used for general statistics, nor is it just a simple analysis at the click and conversion level, but it also participates in the segmentation of business travel stages, duration adjustment and ad trigger control.

[0035] To further verify the technical effects of the present invention, under the conditions of the same advertising resource pool, the same budget, the same travel service category, and the same user scale, the system of the present invention was compared with the original advertising delivery system. The original system used traditional order status recognition, historical interest tag matching, and basic location fence matching to complete the advertising recommendation. The specific comparison data is shown in Table 1: Table 1. Comparison of Overall Performance of Advertising Precision Targeting Optimization Systems for Business Travel Scenarios

[0036] As shown in Table 1, under the same conditions of travel session sample size, complete travel link sample size, advertising campaign item size, and observation data unit size, the system of the present invention significantly outperforms the original system in terms of travel stage identification accuracy, stage boundary positioning accuracy, duration estimation capability, and advertising delivery effect. Specifically, the accuracy rate of stage switching point identification increased from 78.6% to 91.8%, and the average absolute error of travel stage dwell time decreased from 27.3 minutes to 9.8 minutes, indicating that the duration modeling and elastic correction mechanism based on the hidden semi-Markov model of the present invention can more accurately depict the change process of travel stages. Meanwhile, the advertising trigger effectiveness rate, overall click-through rate, and overall conversion rate increased to 84.7%, 5.11%, and 2.07%, respectively, while the invalid exposure rate of high-perturbation samples and the average number of advertising interferences per user decreased to 19.6% and 3.1 times, respectively, demonstrating that the present invention can effectively improve advertising trigger accuracy, delivery matching, and user experience in complex travel scenarios.

[0037] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An AI-based advertising precision targeting and optimization system for business travel scenarios, characterized in that, include: The multi-source data acquisition module is used to collect multi-source observation data from business travelers in business travel scenarios, and to perform data preprocessing to generate business travel behavior observation sequences. The boundary segmentation processing module is used to perform phase boundary gating, buffer confirmation, and boundary rollback on the travel phase observation sequence, and output the travel phase segmented sequence; The duration distribution modeling module is used to construct a hidden semi-Markov model based on the segmented sequence of travel stages. The travel behavior observation sequence is input into the hidden semi-Markov model to establish the initial duration distribution corresponding to the travel stages. The duration elastic correction module is used to perform duration elastic correction processing on the hidden semi-Markov model. It generates duration elastic correction parameters based on multi-source observation data and corrects the initial duration distribution to generate the corrected duration distribution. The state decoding and determination module is used to perform state decoding on the travel behavior observation sequence based on the corrected duration distribution, obtain the current travel stage identification result and stage duration characteristics, and generate the stay depth interval identifier. The advertising trigger control module is used to perform trigger control processing on the candidate advertising set based on the current travel stage identification result and the dwell depth interval identifier, and generate advertising placement decision results; The feedback iteration update module is used to collect advertising feedback data corresponding to the advertising placement decision results, and update the processing results of stage boundary gating, duration elastic correction and trigger control based on the advertising feedback data.

2. The AI-based advertising precision targeting optimization system for business travel scenarios according to claim 1, characterized in that, The multi-source data acquisition module includes: Collect multi-source observation data from business travelers in business travel scenarios, aggregate the data according to the observation time, and establish a correspondence between various types of data corresponding to the same observation time to form raw observation data units; Perform data preprocessing on the original observation data units, and then re-integrate the processed data to generate cleaned observation data units; The cleaned observation data units are processed by sequence arrangement and state association, and then connected in chronological order to form a difference travel observation sequence.

3. The AI-based advertising precision targeting optimization system for business travel scenarios according to claim 1, characterized in that, The boundary segmentation processing module includes: The read difference travel is the observation sequence. Order status change information, location trajectory change information, page interaction change information, and external disturbance event change information are extracted sequentially between adjacent observation data units. The change information is weighted and summed to generate candidate boundary judgment values. At the same time, the candidate boundary judgment values ​​are compared with the preset gate threshold to generate candidate boundary judgment results. The candidate boundary determination results are buffered and confirmed. Candidate boundary positions that meet the preset gate threshold are marked as boundary positions to be confirmed. A preset number of observation data units are read after the current boundary position to be confirmed. The corresponding candidate boundary determination values ​​are averaged to generate buffered confirmation values. The buffered confirmation values ​​are compared with the preset confirmation threshold to confirm the effective stage boundary. Boundary rollback processing is performed on the effective stage boundaries. A preset number of observation data units are read from each effective stage boundary. The corresponding candidate boundary judgment values ​​are averaged to generate rollback verification values. The rollback verification values ​​are compared with preset rollback thresholds to cancel or retain the corresponding effective stage boundaries. Based on the retained effective stage boundaries, the travel behavior observation sequence is segmented to output the travel stage segment sequence.

4. The AI-based advertising precision targeting optimization system for business travel scenarios according to claim 1, characterized in that, The duration distribution modeling module includes: Read the travel stage segment sequence and travel behavior observation sequence, number each stage segment in the travel stage segment sequence according to the time sequence, and arrange each observation data unit in the travel behavior observation sequence according to the time sequence. Align each stage segment with the observation data unit within the corresponding time range, so that each stage segment corresponds to a set of continuous observation data units, forming stage input data. Based on the stage input data, a hidden state system of a hidden semi-Markov model is constructed. According to the connection order of each stage segment in the travel stage segment sequence, the connection relationship between the previous stage segment and the next stage segment is extracted, and each connection relationship is written into the corresponding state transition relationship in the hidden semi-Markov model. At the same time, the connection relationships between all stage segments are summarized and organized to generate the state transition relationship between travel stages. Based on the stage input data, the continuous length of the stage segment corresponding to each hidden state is statistically determined. The number of observation data units that appear consecutively in the difference behavior observation sequence or the length of time that they are continuously covered is taken as the duration value of the current hidden state. All duration values ​​under the same hidden state are collected, sorted and frequency counted. Based on the statistical results, an initial duration distribution corresponding to each hidden state is generated. The state transition relationship and the initial duration distribution are written into the hidden semi-Markov model.

5. The AI-based advertising precision targeting optimization system for business travel scenarios according to claim 1, characterized in that, The duration elastic correction module includes: Read the initial duration distribution of the hidden semi-Markov model and simultaneously read multi-source observation data. Extract order status change information, location trajectory change information, time node change information, page interaction change information, advertising feedback change information, and external disturbance event change information corresponding to each travel stage according to the observation time sequence. Collect and associate the extracted change information to generate duration correction input data corresponding to each travel stage. Based on the continuous duration correction input data, the continuous duration elastic correction parameter generation process is performed on each travel stage. For each travel stage, the corresponding delay disturbance intensity, stay fluctuation intensity, itinerary change intensity, interaction activity intensity, and feedback response intensity are extracted, and the intensity information is combined according to the preset correspondence to generate the continuous duration elastic correction parameters corresponding to each travel stage. Based on the flexible duration correction parameter, the initial duration distribution corresponding to each travel stage is elastically corrected. The duration intervals in the initial duration distribution are lengthened, shortened, or the distribution center of gravity is shifted to generate the corrected duration distribution corresponding to each travel stage.

6. The AI-based advertising precision targeting optimization system for business travel scenarios according to claim 1, characterized in that, The state decoding and determination module includes: The travel journey observation sequence, state transition relationship, and corrected duration distribution are read. Each observation data unit in the travel journey observation sequence is input into the hidden semi-Markov model in chronological order. The hidden semi-Markov model matches the travel journey stage to which each observation data unit belongs at the current time position according to the state transition relationship and corrected duration distribution corresponding to each travel stage, and generates stage decoding results corresponding to each continuous observation interval. The path filtering process is performed on the stage decoding results. According to the sequential connection of each continuous observation interval in the travel behavior observation sequence, the continuity verification and stage affiliation verification of each stage decoding result are performed. The target stage decoding path corresponding to the current observation position is determined from all stage decoding results. The current travel stage identification result corresponding to the current observation position is determined based on the target stage decoding path. Based on the current travel phase identification results, the continuous coverage range of the current travel phase in the travel behavior observation sequence is extracted, the number of observation data units corresponding to the current travel phase and the duration of continuous coverage are counted, the actual duration result of the current travel phase is generated, and the duration values ​​in the corrected duration distribution are extracted sequentially and statistically distributed to generate the distributed duration result of the current travel phase. Based on the actual duration result and the distributed duration result, the phase duration feature is generated. The stage persistence features are processed to determine the dwell depth. The stage persistence features are mapped to the corresponding dwell depth intervals according to the preset dwell depth division rules, and a dwell depth interval identifier is generated based on the mapped dwell depth interval.

7. The AI-based advertising precision targeting optimization system for business travel scenarios according to claim 1, characterized in that, The advertisement triggering control module includes: Extract the current travel stage identification result and the dwell depth interval identifier, match the current travel stage identification result with the stage adaptation information in the advertising resource pool, match the dwell depth interval identifier with the depth adaptation information in the advertising resource pool, and filter to obtain a candidate advertising set; Trigger control processing is performed on the candidate ad set. Based on the stage matching result, depth matching result and feedback constraint result corresponding to each ad item, suppression processing is performed on ad items that do not meet the trigger conditions, and retention processing is performed on ad items that meet the trigger conditions, generating a trigger release set and the corresponding trigger control result. Based on the trigger release set and trigger control results, the system performs placement sorting and placement configuration processing on each ad item to determine the placement order, timing, and content of each ad item, and then integrates these to generate ad placement decision results.

8. The AI-based advertising precision targeting optimization system for business travel scenarios according to claim 1, characterized in that, The advertisement triggering control module includes: Collect advertising feedback data corresponding to the advertising placement decision results, and associate and sequentially collect various advertising feedback data to generate feedback data units that correspond one-to-one with each advertising placement decision result, forming an advertising feedback data sequence. Based on the advertising feedback data sequence, each feedback data unit is back-matched with the corresponding travel stage segment sequence and the corrected duration distribution, and correction processing is performed within the corresponding matching range to generate the updated travel stage segment sequence and the updated corrected duration distribution. Based on the advertising feedback data sequence, each feedback data unit is back-matched with the advertising entries in the corresponding advertising placement decision results, and correction processing is performed within the corresponding matching range to generate updated advertising placement decision results. The updated advertising placement decision results are then used as the basis for the next round of advertising placement decision processing.

Images