Discovering Connotations as Labels for Weakly Supervised Image-Sentence Data
Published: 2018 April
Buchtitel: The Web Conference (Cognitive Computing Track)
We address the task of labeling image-sentence pair at large-scale with varied concepts representing connotations. That is for any given query image-sentence, we aim to annotate them with the connotations that capture intrinsic intension. To achieve it, we pro- pose a Connotation multimodal embedding model (CMEM) with a novel loss function. Its unique characteristics over previous models include (i) can leverage multimodal data as opposed to only visual information, (ii) robust to outlier labels in a multi-label scenario and (iii) works well with large-scale weakly supervised data. With extensive quantitative evaluation, we exhibit the effectiveness of CMEM for detection of multiple labels over other state-of-the-art approaches. Also, we show that in addition to annotation of images with connotation labels, our byproduct of the model inherently supports cross-modal retrieval.