Multimodal metaphor detection based on distinguishing concreteness

摘要: Abstract Metaphors are a common linguistic phenomenon, and metaphor identification plays an essential role in processing. Most existing computing techniques use only texts to gain features, but we acquire additional knowledge from other modalities. At present, the multimodal model field is exploratory stage, few models available still relatively crude. We propose detection method according idea that different types of words suitable for modality calculations. First, our proposed framework uses fine-grained concreteness calculation based on part speech distinguish abstract concrete words. then choose appropriate modal feature computational with concreteness. Additionally, also improve image features detection.