Semantic Sticker Classification based on Convolutional Neural Network
Hsin-Ying Lin, Chin-Hung Teng, Yi‐Jheng Huang · 2022
In instant messaging applications, stickers can convey more emotion than text, and thus, they are frequently and heavily used in instant messaging applications and play a very important role in conversations. However, because of the richness and diversity of stickers, it sometimes takes a lot of time for users to find a suitable one. To facilitate users to find the desired stickers quickly, this study uses a deep learning network, i.e., CNN, for the semantic classification of stickers. We classify the stickers into nine semantic aspects, and each aspect is classified by a deep learning network model with classification accuracy between 85.2% and 98.7%. Based on these nine aspects, we implement a sticker recommendation system for users to find the most suitable stickers according to their application scenarios.