Prediction of users' facial attractiveness on an online dating website
| Autor: | Toshihiko Yamasaki, Shinichi Egami, Tetsuhiro Nakamoto, Kiyoharu Aizawa, Eitaro Kuwabara, Yusuke Fuchida, Xiaoxue Zang |
|---|---|
| Rok vydání: | 2017 |
| Předmět: |
Attractiveness
Information retrieval Multimedia Computer science media_common.quotation_subject Rank (computer programming) 020207 software engineering 02 engineering and technology computer.software_genre Convolutional neural network Perception 0202 electrical engineering electronic engineering information engineering Facial attractiveness 020201 artificial intelligence & image processing computer media_common |
| Zdroj: | ICME Workshops |
| DOI: | 10.1109/icmew.2017.8026293 |
| Popis: | Online dating websites are popular platforms for adults to search for their life partners. Because on online dating websites, a user's profile image is an important factor determining other's impressions, we focus on profile images and analyze users' visual attractiveness in this study. Facial attractiveness is strongly related to our perception of aesthetics and therefore we believe our investigation can somewhat contribute to artwork analysis. We use pre-trained convolutional neural networks (CNN) to extract visual features and propose a new method to rank users' attractiveness from their online dating interactions. For both genders, we predict users' facial attractiveness by supervised machine learning. Our experimental results show that deep representations of profile images are powerful to capture facial attributes' differences and perform well in predicting users' attractiveness. The correlation coefficient of 0.462 for male users and the correlation coefficient of 0.387 for females users is obtained for regression. The accuracy of 75% for females and the accuracy of 78.8% for males is obtained for 2-level classification. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|