基於社群媒體內容之個人化 Hashtag 推薦模型

Abstract

為了提升貼文能見度以及讓貼文呈現個人化或品牌形象，在本研究中，我們提出一個個人化主題標籤推薦模型，我們提出的模型包括四個階段，首先，我們萃取每篇貼文的視覺與文字特徵，接著，我們設計一個貼文與標籤的注意力機制以取得使用者標注的習慣以及貼文與標籤的關係，然後，我們開發一個六重共注意力機制產生貼文表示式，最後，我們利用關係圖卷積網絡來強化貼文表示式，並利用強化後的貼文表示式推薦主題標籤。實驗結果顯示，我們的模型在命中率、精確度、召回率和F1分數等指標上均優於現有模型。我們的模型可幫助使用者使用個人化的主題標籤，亦可幫助企業加速主題標籤標注的工作，進而提升貼文的能見度、社群參與度以及使用者間的互動。

To increase the post visibility and present personalized or brand image, in this study, we propose a novel personalized hashtag recommendation framework to recommend hashtags for user posts. The proposed framework contains four phases. First, we extract the visual and textual features from each post. Second, we devise the post-tag attention mechanism to acquire user tagging habits and post-tag relationships through similar posts. Third, we employ the sextuplet co-attention mechanism to derive the representation of each post. Finally, we adopt the relational graph convolution network (R-GCN) to construct the interaction graph between posts and update the post representations by propagating information among neighboring nodes. Then, we use the convoluted post representations to recommend hashtags for users. The results from the experiments demonstrate that the suggested framework surpasses the performance of current state-of-the-art models, achieving higher scores in hit rate, precision, recall, and F1 score. Our proposed framework may offer an effective tool for users to quickly make personalized hashtags on their posts and for businesses to accelerate the task of creating hashtags, which in turn increases post visibility, fosters community engagement, and enhances user interactions.