Conflicting Statements Detection Using Bi-Directional LSTM

Abstract

As data sources become more and more diversified, the importance of data mismatch detection increases. Contradiction is a semantic relation where two sentences cannot be true simultaneously. Conflicts can happen because of lexical features, such as negation, antonyms or numerical mismatch, or can be adapted upon time or potentially physiological status. There are many methods to detect conflicting statements. The method based on Deep Learning, Artificial Neural Networks (ANN), Bi-directional long short-term memory, and Global vectors for word representation (GloVe) is used for conflicting text statements detection. The Bidirectional Long short-term memory (LSTM) model is built to increase the accuracy of detecting conflicting statements. GloVe is a count model, looking at how frequently a word appears in another word context (co-occurrence probabilities), within the whole corpus. Thus, the GloVe model learns the word representation vector by performing dimensionality reduction on a co-occurrence counts matrix. Google colab is used for implementing this. Tensorflow machine learning framework is utilized for training and testing the predictive model for confliction detection in texts. The main aim is to build a system to identify inconsistencies and defiance in text. And also improve the performance of the model which can be evaluated using confusion matrix and accuracy.