Deep Learning Based Skincare Product Recommendation System Using a Wide & Deep Architecture
Abstrak
This research describes a skincare product that is built using Wide & Deep architecture which is an efficient hybrid deep learning model. The architecture uses a wide component to learn feature interactions that often occur in the data, together with a deep component to identify the complex relationships between users and skincare products. The Sephora Products and Skincare Reviews dataset retrieved from Kaggle was used to conduct the study. It contains information on skincare products and reviews provided by users. The best training run was achieved with a Learning Rate of 0.001 and the RMSprop Optimizer using a batch size of 128 over 20 epochs. The AUC of the validation was 0.9659 and test was 0.9658. The system effectively identifies patterns within the Sephora data for products and reviews, demonstrated by the presented figures. The model achieves values of 0.91 for precision, recall, and F1-score at the same time. This performance metrics balance proves the system is robust, and it works. The system demonstrated its effectiveness in differentiating between classes of both evaluation. These results indicate that our model can be useful, for Sephora Products and Skincare Reviews analysis. This system is implemented using Streamlit for real-time inference to filter, preprocess, predict, and rank products based on matching score. The systems take user and product information through several filters and stages which include Pre-Processing, Prediction, and Filtering. The product is ranked based on Suitable Score. According to the findings, the Wide & Deep approach is capable of developing a stable and accurate recommendation system that is able to provide a skincare product recommendation based on the user needs.
Identifier
- DOI: https://doi.org/10.67561/jaids.v2i1.2
- OpenAlex: https://openalex.org/W7169905611