70 0

온라인 몰 추천을 위한 사용자 광고 클릭 예측 기반의 협업 필터링

Title
온라인 몰 추천을 위한 사용자 광고 클릭 예측 기반의 협업 필터링
Other Titles
Collaborative Filtering based on User Ad Click Prediction for Online Mall Recommendations
Author
한영진
Alternative Author(s)
Han, Young Jin
Advisor(s)
조인휘
Issue Date
2020-02
Publisher
한양대학교
Degree
Master
Abstract
오늘날 4차 산업혁명의 시대에 이르러 매일 방대한 데이터가 누적되어 가정, 산업, 직장, 사회 전반에 걸쳐 하루에도 수많은 새로운 데이터가 생성되고 있다. 온라인 몰 또한 사용자의 편의에 따라 재 방문시 기존의 광고 클릭과 구매 이력으로 상품을 추천한다. 이는 추천 시스템이라는 검색 알고리즘으로 기반이 갖춰지면서 다양한 방식으로 분화되어 있어서 가능했다. 응용 분야도 뉴스 서비스, 쇼핑, 아마존, 넷플릭스, 멜론, 교보문고 등 사용자들은 추천 시스템 알고리즘에 노출돼 있다고 봐도 무방하다. 문제는 사용자가 경험하지 않은 국내, 국외 온라인 몰을 웹 서핑할 때에는 주위의 의견과 사용자 후기 등으로 판단할 수밖에 없는 상태이다. 사용자 정보의 안정성, 상품 거래, 정보 조작 등 여러 가지 변수들이 내재되어 있다. 이 논문에서는 온라인 몰을 광고 클릭으로 머신러닝 하여 메타 데이터를 생성한 후 협업 필터링으로 추천하는 새롭고 창의적인 알고리즘을 제시하고자 한다. 본 논문에서 새롭게 제시할 알고리즘은 사용자를 위한 몰(Mall for User
MFU)이다. 이것은 온라인 몰의 데이터를 메모리 최적화를 한 후 지도학습인 로지스틱 회귀(logistic regression)와 의사 결정 트리(decision tree)로 온라인 몰 데이터를 트레이닝 세트(training set)와 검증 세트(test set)로 모델링 한다. 예측값을 혼동행렬(confusion matrix)로 분석하여 분류 성능 평가를 하고 ROC curve와 AUC로 시각화(visualize)를 한다. 그리고 온라인 몰에서 사용자 유형 분석은 계층적 클러스터링으로 클러스터를 하여 상향식 접근 방법으로 비지도 학습을 한다. 성능 평가는 첫째, 데이터 최적화 실험 결과는 샘플링을 추출하여 측정하고 둘째, 로지스틱 회귀 모델링과 의사 결정 트리 모델링의 예측 값으로 비교 측정 평가하고 셋째, 파이썬(Python) 기계 학습 사이킷런(scikit-learn) 라이브러리의 precision, recall, f-score로 예측 값과 모델링 예측 값을 비교한다. 재 검증은 분류 보고서 표시기로 모델링 예측 값과 비교하여 다시 검증한다. 넷째, 협업 필터링의 피어슨 상관계수로 선호도를 추측 예측하고 성능 평가로 검증한다. | Today is the 4th revolution, and there is a lot of data being generated every day throughout the home, industry, and the workplace. Most online malls also recommend products through existing ad clicks and purchasing history when the buyer revisits the store according to user convenience. This was possible because it was based on a search algorithm called a recommendation system and was differentiated in various ways. Application area users of news services, shopping, Amazon, Netflix, Melon, Kyobo Books,etc. may be exposed to recommendation system algorithms. The problem is that when surfing the web on domestic or foreign online malls that the user has not experienced, they have no choice but to judge the surrounding opinions and user reviews. Various variables are inherent, such as stability of user information, commodity trading, and information manipulation. In this paper, we propose a new and creative algorithm that machine-learns an online mall with ad clicks to generate metadata and recommend it as collaborative filtering. The new algorithm proposed in this paper is Mall for User (MFU). After optimizing the memory of online mall data, this algorithm can modelling the datas into training set and test set with logistic regression and deceision tree. It analyzes the predicted values with CONFUSION MATRIX to confirm the classification performance evaluation. And it visualizes with ROC curve and AUC. Visualize with AUC And in the online mall, user type analysis is clustered by hierarchical clustering, which leads to unsupervised learning from a bottom-up approach. In the performance evaluation, First, the results of data optimization experiments are sampled and measured. Second, the results are compared and evaluated with the predictive values of logistic regression modeling and decision tree modeling. Thirdly, This Algorithm compare the predicted values and the forecasted modelling values with the precision, recall, f-score of Python scikit-learn machine learning libraries. Re-validation is re-validated against the modeling predictions by the classification report indicator. Fourth, the Pearson correlation coefficient of collaborative filtering is used to speculatively predict and verify the performance.
Today is the 4th revolution, and there is a lot of data being generated every day throughout the home, industry, and the workplace. Most online malls also recommend products through existing ad clicks and purchasing history when the buyer revisits the store according to user convenience. This was possible because it was based on a search algorithm called a recommendation system and was differentiated in various ways. Application area users of news services, shopping, Amazon, Netflix, Melon, Kyobo Books,etc. may be exposed to recommendation system algorithms. The problem is that when surfing the web on domestic or foreign online malls that the user has not experienced, they have no choice but to judge the surrounding opinions and user reviews. Various variables are inherent, such as stability of user information, commodity trading, and information manipulation. In this paper, we propose a new and creative algorithm that machine-learns an online mall with ad clicks to generate metadata and recommend it as collaborative filtering. The new algorithm proposed in this paper is Mall for User (MFU). After optimizing the memory of online mall data, this algorithm can modelling the datas into training set and test set with logistic regression and deceision tree. It analyzes the predicted values with CONFUSION MATRIX to confirm the classification performance evaluation. And it visualizes with ROC curve and AUC. Visualize with AUC And in the online mall, user type analysis is clustered by hierarchical clustering, which leads to unsupervised learning from a bottom-up approach. In the performance evaluation, First, the results of data optimization experiments are sampled and measured. Second, the results are compared and evaluated with the predictive values of logistic regression modeling and decision tree modeling. Thirdly, This Algorithm compare the predicted values and the forecasted modelling values with the precision, recall, f-score of Python scikit-learn machine learning libraries. Re-validation is re-validated against the modeling predictions by the classification report indicator. Fourth, the Pearson correlation coefficient of collaborative filtering is used to speculatively predict and verify the performance.
URI
http://dcollection.hanyang.ac.kr/common/orgView/000000111015https://repository.hanyang.ac.kr/handle/20.500.11754/122960
Appears in Collections:
GRADUATE SCHOOL OF ENGINEERING[S](공학대학원) > ELECTRICAL ENGINEERING AND COMPUTER SCIENCE(전기ㆍ전자ㆍ컴퓨터공학과) > Theses (Master)
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