지원사업
학술연구/단체지원/교육 등 연구자 활동을 지속하도록 DBpia가 지원하고 있어요.
커뮤니티
연구자들이 자신의 연구와 전문성을 널리 알리고, 새로운 협력의 기회를 만들 수 있는 네트워킹 공간이에요.
논문 기본 정보
- 자료유형
- 학위논문
- 저자정보
- 지도교수
- 최동구
- 발행연도
- 2023
- 저작권
- 포항공과대학교 논문은 저작권에 의해 보호받습니다.
이용수0
이 논문의 연구 히스토리 (3)
2023
Compatibility Learning and Recommendation between Clothing and Accessories Using Siamese Structure
김우찬
,
최동구
한국경영과학회 학술대회논문집
2023.05
학술대회자료
Compatibility Learning and Recommendation between Clothing and Accessories Using Siamese Structure
김우찬
,
최동구
대한산업공학회 춘계공동학술대회 논문집
2023.05
학술대회자료
Compatibility Learning for Recommendation between Accessory and Clothing with Siamese Architecture
김우찬
산업경영공학과
2023.01
학위논문
- 이 논문의 후속연구가 궁금하신가요?
- 연관 학술논문 또는 학술발표를 통해 보다 발전된 연구결과를 확인하실 수 있습니다.
- 이 논문의 연구 히스토리 확인하기
초록· 키워드
오류제보하기
Recently, the importance of fashion recommendation has been increasing. From the consumer’s point of view, they can develop their fashion through compatible items, and from the seller’s point of view, they can expect additional sales. However, recommendations between clothing are quite implemented, but recommendations between accessory and clothing are insufficient. Most of existing fashion recommendation related studies are also conducted on the entire outfit, so the proportion of accessory is small.
In this paper, we introduce metric learning based method that uses the Siamese architecture to learn compatibility between image pairs in existing studies. As a result of experimenting with our data based on metric learning, the performance was not high. Therefore, we propose binary classification-based method that uses the Siamese architecture, but changes the learning method. The feature extraction process of images of accessory and clothing is carried out via Siamese CNN. We learn the compatibility through binary classification-based method that concatenates the two embeddings and adds classifier. Additionally, we use subcategory combination information in the binary classification-based method.
We set a few pairs which the expert determined compatible to ‘positive’, and other randomly sampled pairs to ‘negative’. As a result of experiment using this data, the binary classification-based method makes higher classification accuracy and recommend performance than existing metric learning based method. In addition, using the combination information show higher performance than not using it.
In this paper, we introduce metric learning based method that uses the Siamese architecture to learn compatibility between image pairs in existing studies. As a result of experimenting with our data based on metric learning, the performance was not high. Therefore, we propose binary classification-based method that uses the Siamese architecture, but changes the learning method. The feature extraction process of images of accessory and clothing is carried out via Siamese CNN. We learn the compatibility through binary classification-based method that concatenates the two embeddings and adds classifier. Additionally, we use subcategory combination information in the binary classification-based method.
We set a few pairs which the expert determined compatible to ‘positive’, and other randomly sampled pairs to ‘negative’. As a result of experiment using this data, the binary classification-based method makes higher classification accuracy and recommend performance than existing metric learning based method. In addition, using the combination information show higher performance than not using it.
목차
I. Introduction - 1 -II. Literature Review - 4 -2.1 Compatibility Learning for Fashion Recommendations - 4 -2.2 Siamese Convolutional Neural Network - 7 -2.3 Metric Learning - 9 -2.4 Loss functions for metric learning - 10 -Ⅲ. Methodology - 11 -3.1 Metric Learning Based Method - 12 -3.2 Binary Classification Based Method - 16 -3.3 Binary Classification Based Method with Combination Information - 19 -Ⅳ. Experimental Setup - 22 -4.1 Data Description - 22 -4.2 Performance Measure - 23 -4.3 Training Detail - 25 -Ⅴ. Results and Discussion - 26 -5.1 Performance Results of Metric Learning Based Method - 26 -5.2 Classification Performance of Binary Classification Based Method - 28 -5.3 Comparison of Recommend Performance - 30 -5.4 Examples of recommendation - 32 -VI. Conclusion - 34 -요 약 문 - 36 -REFERENCES - 37 -Acknowledgments - 39 -Curriculum Vitae - 40 -
최근 본 자료
댓글(0)
0