EFFICIENT DISCOVERY OF MALEVOLENT USERS IN COMPLEX SOCIAL NETWORKS
V Mohana Radhika, Dsaidan · IJITR International Journal of Innovative Technology and Research - IJITR International Journal of Innovative Technology and Research · 2018
According to an eMarketer survey, about 75% of the content published by Facebook users contains photos.The appropriate data of different modalities will often have semantic correlations.Most of the existing works use a pocket of words to model text information.Since we proposed to use a Fisher kernel framework to represent the textual information, we use it to add the SIFT descriptors of the images.We suggest that you include examples of continuous words to treat semantic textual agreements and to adopt them for the recovery of mixed media.Your building block of the network used in the work can be the Boltzmann-limited Gaussian machine.However, fisher vectors are often high dimension and dense.This limits the use of the FV for large scale applications, where computer requirements need to be studied.Finally, the Hamming distance can be used to determine the similarities between the converted FV hash codes as well as other image hate codes.We have evaluated the proposed SCMH method in three sets of general data.SCMH achieves better results than state-of-the-art methods with multiple hash code lengths.A Skip-gram model was used to create these 300-dimensional vectors for 3 million sentences and words.To generate Fisher vectors, we use the implementation of INRIA.In this paper we compare the important duration of the proposed approach with other methods of learning hash.Although the off-line stage of the proposed framework requires massive computing costs, the computer complexity of the internet phase is small or similar to other hassle methods.