Training image-recognition systems using a joint embedding model on online social networks

摘要

In one embodiment, a method includes identifying a shared visual concept in visual-media items based on shared visual features in images of the visual-media items; extracting, for each of the visual-media items, n-grams from communications associated with the visual-media item; generating, in a d-dimensional space, an embedding for each of the visual-media items at a location based on the visual concepts included in the visual-media item; generating, in the d-dimensional space, an embedding for each of the extracted n-grams at a location based on a frequency of occurrence of the n-gram in the communications associated with the visual-media items; and associating, with the shared visual concept, the extracted n-grams that have embeddings within a threshold area of the embeddings for the identified visual-media items.