Role of semantic links in performance of Information Retrieval on graph-based multimodal collections

Currently there is a proliferation of data in different modalities—text, audio, video, and image. Various platforms such as social networks or different online repositories facilitate this trend. In addition, users search more for information from different modalities. This trend has created multimodal Information Retrieval (IR) as a challenge. However, information objects are not independent and there are different types of relations (e.g. semantic or similarity links) between them. In this paper, we evaluate the role of adding semantic links in performance of our graph-based model for multimodal IR. The result analysis shows that we gain 13% recall increase by using inter-lingual semantic links.