使用深度學習之Hashtag推薦系統

Abstract

此研究主要是針對 Instagram 社群平台而設計的自動化 Hashtag 推薦系統。當使用者在使用 Instagram 上傳新圖片時，可簡化在操作上與手動標籤 Hashtag 的動作。我們使用了卷積神經網絡 (Convolutional Neural Network) 圖像辨識技術來達成圖像歸類 (image classification) 的效果。但現今，Hashtag 的歸類整理還是個未能被完全解決的問題。社群平台上有著層出不窮和大量的 Hashtag 正在推陳出新，如果只使用單一機器學習方法，將會無法達成完整學習的效果。 在這研究中，我們將會探討不同 Hashtag 的語義程度，而導出並不是所有的 Hashtag 都適合推薦給使用者使用的結論。此外，為了擴充 Hashtag 的詞量，我們結合了圖像辨識和語義嵌入 (semantic embedding) 的模型來建構此推薦系統。透過定期詞彙和語義嵌入模型的更新，此系統將會推薦最符合現今所流行的 Hashtag 詞彙供使用者選擇。我們使用 Instagram 上的圖片做實驗，並完整呈現此推薦系統可以有效的推薦和圖像所呼應的 Hashtag 詞彙。

The goal of this research is to design a system that can predict and recommend hashtags to users when they upload new images on Instagram. The system will simplify the picture tagging process for users by automatically suggesting relevant hashtag options when an image is uploaded. Using the image recognition technology like Convolutional Neural Network (CNN), we could achieve the image classification function. However, hashtag prediction is still an open problem due to the large amount of media contents and hashtag categories; using single machine learning method will not be sufficient. In this research, we show that not all hashtags are equally meaningful, and some are not suitable in recommendation learning. In addition, we combine image classification and semantic embedding models to gain the expansion of recommended hashtags. At the same time, by periodically updating semantic embedding model, we ensure that the hashtags being recommended follow the latest trends. We apply the design to existing image-hashtag pairs on Instagram and demonstrate that the capability of the system can successfully recommend hashtags that are more relevant to the images.