Reducing photo-tagging spam

摘要

A photo spam detector detects illegitimate non-natively captured images through extracting image features and feeding the extracted features into a probabilistic model. The probabilistic model categorizes the photo as legitimate or illegitimate. Requests to tag one or more users in a photo are analyzed by a tag analyzer that assesses relationships between the tag requests themselves, social relationships between the tagged users, and the presence or absence of faces within the regions specified by the tag requests. Based on the classification of images or tags as illegitimate, a social networking system applies one or more social media distribution policies to the image or tags to suppress or prohibit distribution.

主权项

1. A method comprising, by one or more computing systems:
accessing an image, the image being posted on a social-networking system, wherein the image is accessed in response to an action being requested on the image on the social-networking system; extracting a predetermined number of features from the image; processing the extracted features using a machine-learning model to determine a probability that the image is an illegitimate image; extracting metadata from the image; comparing, using the machine-learning model, the extracted metadata to one or more of the extracted features; adjusting, using the machine-learning model, the determined probability that the image is an illegitimate image based on the comparison; categorizing, based on the adjusted probability, the image as a natively-captured image or as an illegitimate image; and applying one or more distribution policies to the image based on the categorization, wherein the distribution policies limit distribution of the image on the social-networking system, and wherein at least one of the distribution policies comprises prohibiting actions on the image by users of the social-networking system.