Event Detection and Violence Recognition from Textual News through Multilayer Perceptron and Supervised Learning

Abstract

An unprecedented way is accomplished by using concept words derived from statistical context analysis between sentences which is better than traditional methods that use only keyword representation. Through scaling to a very large dataset we proposed an algorithm which discovers, and describes events with effective keyword networks, based on their coexisting peripheral co-occurrences. In our experiment, we used real-world news, and supervised them into paraphrases by weighting for the all attempted events. We evaluated our scheme by a set of terms that maximally discriminated the percussion in news and which also keep the evidences. Here we are classifying the events with a multilayer perceptron by executing auto-convolution methodology in back propagation.