254 0

Trip Planning Considering Personal Preferences and Environmental Variables

Title
Trip Planning Considering Personal Preferences and Environmental Variables
Other Titles
개인 선호도 및 환경 변수를 고려한 여행 계획 방법
Author
김민구
Alternative Author(s)
김민구
Advisor(s)
서일홍
Issue Date
2020-02
Publisher
한양대학교
Degree
Doctor
Abstract
개인화된 여행 계획의 생성은 사용자 선호도에 따라 관광지를 선택하고 이를 다양한 제약조건을 만족시킬 수 있도록 순차적으로 배열해야하는 어려운 문제이다. 본 논문에서는 여행 계획 중에 고려해야할 불확실성을 갖는 다양한 변수들을 모델링하고 선호도 및 제약조건에 따라 사용자 만족도를 최대화 하는 개인화된 여행 계획을 효율적으로 생성할 수 있도록 하는 것을 목표로 하였다. 구체적으로는 선호도, 제약조건, 불확실한 환경 변수를 고려하여 주어진 여정의 확률적 평가를 가능하게 하는 하이브리드 템퍼럴 베이지안 네트워크(Hybrid temporal Bayesian network)를 제안하고, 해당 모델을 이용하여 주어진 여정을 시간 가중 사용자 만족도를 목적 함수로써 평가할 수 있도록 하였다. 그리고, 개미 군집 최적화(Ant colony optimization)를 이용하여 다수의 여행 일정 후보군을 생성하고, 생성된 여정을 제안한 모델을 사용해 평가하여, 얻은 평가 결과를 바탕으로 이전보다 향상된 새로운 후보군을 생성하는 과정을 반복함으로써 앞서 언급된 목적 함수를 최대화하는 여행 계획을 생성할 수 있도록 하였다. 제안한 여행 계획 방법의 검증은 실험 참가자들에게 사람이 작성한 여행 계획과 제안한 방법을 사용하여 생성된 여행 계획 중 더 선호하는 여행 계획을 선택하도록하는 사용자 연구를 통해 이루어졌다. 실험 결과 약 57%의 참가자들이 본 논문에서 제안한 여행 계획 방법을 통해 생성된 여행 계획을 선택하였고, 이를 통해 제안한 여행 계획 방법이 사람과 비슷한 수준의 여행 계획을 생성할 수 있음을 확인할 수 있었다. 추가로, 다른 최적화 방법과의 성능 비교를 통해 여행 계획에 사용된 개미 군집 최적화 알고리즘의 효율성 또한 입증하였다. 제안한 여행 계획 방법과 더불어, 여행 계획에 있어 중요한 역할을 하는 관광 명소의 평점을 소셜 미디어의 텍스트 데이터를 분석하여 예측하는 시스템을 제안하였다. 여행 명소에 대한 평점 정보는 여행 계획에 대한 객관적이고 포괄적인 정보를 제공한다. 그러나, 이러한 평점 정보는 리뷰가 매우 적을 때는 높은 신뢰성을 보장하기 어렵다는 단점이 있다. 따라서 제안된 시스템은 지정된 관광 명소의 이름을 키워드로 사용하여 소셜 미디어에서 많은 게시물을 검색 및 수집하고, 필터링 모델을 사용하여 수집한 게시물을 리뷰 및 리뷰가 아닌 게시물로 분류 후 스코어링 모델을 사용하여 각 리뷰의 해당 관광지에 대한 점수를 예측하도록 하였다. 그 후 예측 된 점수들을 평균내어 주어진 여행지의 전체 평점을 예측하였다. 이러한 여행지 평점 예측 시스템의 성능을 향상시키기 위해 본 논문에서는 딥 러닝 기반 필터링, 스코어링 모델 구조를 제안하였고, 두 가지의 한국어 데이터 증강 방법도 함께 제안하였다. 제안한 모델 구조는 두 개의 다른 비교 모델들과 실험을 통해 비교해 보았을 때 더 우수한 성능을 보임을 확인할 수 있었다. 또한 제안한 데이터 증강 방법의 효과를 검증하기위한 실험에서 제안한 방법을 사용하면 필터링 및 스코어링 모델의 정확도가 각각 약 8% 및 3% 증가하는 것을 확인할 수 있었다. | Personalized trip planning is a challenging problem given that places of interest should be selected according to user preferences and sequentially arranged while satisfying various constraints. In this thesis, we aimed to model various uncertain aspects that should be considered during trip planning and efficiently generate personalized plans that maximize user satisfaction based on preferences and constraints. Specifically, we propose a probabilistic itinerary evaluation model based on a hybrid temporal Bayesian network that determines suitable itineraries considering preferences, constraints, and uncertain environmental variables. The model retrieves the sum of time-weighted user satisfaction of a given itinerary as an objective function, and ant colony optimization generates the trip plan that maximizes the objective function. First, the optimization algorithm generates candidate itineraries and evaluates them using the proposed evaluation model. Then, it improves candidate itineraries based on the evaluation results of previous itineraries. To validate the proposed trip planning approach, we conducted an extensive user study by asking participants to choose their preferred trip plans from options created by a human planner and our approach. The results show that our approach provides human-like trip plans, as participants selected our generated plans in 57\% of the pairs. We also evaluated the efficiency of the employed ant colony optimization algorithm for trip planning by performance comparisons with other optimization methods. We also propose a system for predicting the ratings of tourist attractions using textual data from social media. Average rating information on travel attractions provides objective and comprehensive information for trip planning. However, such rating information has a disadvantage that it is difficult to guarantee high reliability when there are only a few reviews. Therefore, the proposed system searches and gather a large number of posts from social media by using the name of a given tourist attraction as a keyword. Then, it classifies the posts into reviews and non-reviews using a filtering model and predicts the rating of each review using a scoring model. Predicted ratings are then averaged to predict the overall rating of the given travel attraction. To improve the performance of the system, deep learning-based filtering and scoring model structures are proposed along with two different Korean data argumentation methods. The proposed model structures are compared with the two different model structures, and the experimental results show that the proposed structures outperform the other two structures. Also, in the experiments to validate the effectiveness of the proposed data argumentation methods, we found that using the proposed methods increases the accuracy of the filtering and scoring models by about 8\% and 3\%, respectively.; Personalized trip planning is a challenging problem given that places of interest should be selected according to user preferences and sequentially arranged while satisfying various constraints. In this thesis, we aimed to model various uncertain aspects that should be considered during trip planning and efficiently generate personalized plans that maximize user satisfaction based on preferences and constraints. Specifically, we propose a probabilistic itinerary evaluation model based on a hybrid temporal Bayesian network that determines suitable itineraries considering preferences, constraints, and uncertain environmental variables. The model retrieves the sum of time-weighted user satisfaction of a given itinerary as an objective function, and ant colony optimization generates the trip plan that maximizes the objective function. First, the optimization algorithm generates candidate itineraries and evaluates them using the proposed evaluation model. Then, it improves candidate itineraries based on the evaluation results of previous itineraries. To validate the proposed trip planning approach, we conducted an extensive user study by asking participants to choose their preferred trip plans from options created by a human planner and our approach. The results show that our approach provides human-like trip plans, as participants selected our generated plans in 57\% of the pairs. We also evaluated the efficiency of the employed ant colony optimization algorithm for trip planning by performance comparisons with other optimization methods. We also propose a system for predicting the ratings of tourist attractions using textual data from social media. Average rating information on travel attractions provides objective and comprehensive information for trip planning. However, such rating information has a disadvantage that it is difficult to guarantee high reliability when there are only a few reviews. Therefore, the proposed system searches and gather a large number of posts from social media by using the name of a given tourist attraction as a keyword. Then, it classifies the posts into reviews and non-reviews using a filtering model and predicts the rating of each review using a scoring model. Predicted ratings are then averaged to predict the overall rating of the given travel attraction. To improve the performance of the system, deep learning-based filtering and scoring model structures are proposed along with two different Korean data argumentation methods. The proposed model structures are compared with the two different model structures, and the experimental results show that the proposed structures outperform the other two structures. Also, in the experiments to validate the effectiveness of the proposed data argumentation methods, we found that using the proposed methods increases the accuracy of the filtering and scoring models by about 8\% and 3\%, respectively.
URI
https://repository.hanyang.ac.kr/handle/20.500.11754/123769http://hanyang.dcollection.net/common/orgView/200000437124
Appears in Collections:
GRADUATE SCHOOL[S](대학원) > ELECTRONICS AND COMPUTER ENGINEERING(전자컴퓨터통신공학과) > Theses (Ph.D.)
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE