Multilingual multimodal content mining for hashtag recommendation and popularity prediction in social networks

Abstract

Social networking services have profoundly reshaped human interaction and information exchange transcending geographical, cultural, and temporal boundaries. These platforms empower users to both consume and produce content, expressing themselves in diverse languages and modalities, referred to as multilingualism and multimodality, respectively. This resulting surge in user-generated content, characterized by multilingualism and multimodality, has introduced a significant challenge of information overload, which hinders content discoverability and reachability. To mitigate these challenges and manage content efficiently, this thesis investigates hashtag recommendation and popularity prediction as effective solutions. Hashtag recommendation is the process of assigning hashtags to uploaded content, thereby facilitating thematic organization of vast volumes of content. However, existing methods overlook crucial aspects such as multilingualism and multimodality. In this thesis, we propose hashtag recommendation methods tailored for various linguistic contexts and content modalities. These methods encompass monolingual content, multilingual content, multimodal content comprising texts and images, and micro-videos, acknowledging the diverse levels of user engagement they elicit.