On the use of statistical semantics for metadata-based social image retrieval

We revisit text-based image retrieval for social media, exploring the opportunities offered by statistical semantics. We assess the performance and limitation of several complementary corpus-based semantic text similarity methods in combination with word representations. We compare results with state-of-the-art text search engines. Our deep learning-based semantic retrieval methods show a statistically significant improvement in comparison to a best practice Solr search engine, at the expense of a significant increase in processing time. We provide a solution for reducing the semantic processing time up to 48% compared to the standard approach, while achieving the same performance.