Instagram Hashtag Prediction Using Sequential Analysis with Deep Neural Networks

Instagram is one of the most popular photos sharing service. For more convenient content search people use hashtags (#nature, #love, etc.) in posts with photos. The author’s aim is to make hashtag prediction possible and convenient for users. The paper provides a reader with a detailed theoretical overview of Multi-Label Image Classification, Knowledge Distillation, and overview of ResNet architecture. Next author proposes improvements on ResNet architecture allowing the model to boost quality and converge faster. Finally, the model type Self-Improving-Modified- Resnet (SIMR) is presented. Their main feature is the additional bottleneck block used as the tool incorporating benefits from novel self distillation (a combination of self training and knowledge distillation, introduced in 2019). The author thoroughly describes parameters of own dataset collected from Instagram. com web page. The dataset is considered as challenging due to its humangenerated tags. People are using a wide range of tags, some of them are noisy, not exhaustive, sometimes random, or even not corresponding to the photo at all. For Instagram users’ convenience for testing model author develops Telegram Chatbot that predicts hashtags by the user’s sent image.