A NLP tool for relations extraction from a training set, including over 1000 of sentences. From each sentence, we extract one kind of relation: "Living In", mean, that there is some realtion between a person and a living place.
The tool consist two methods to deal with this problem:
-
Deep Learning style - an attentions network was builded with 1 layer of LSTM and a regular feed-forrowed neural network to predict the relation (existing or non-existent) One of the problems that showed up is how do we train a deep neural network with only 1000 training examples. So, we used an augumented data
-
Feature Selection style - a simple SVM using a vector of features to predict the relation (existing or non-existent).
Including report