Creative Commons

This dissertation was written as a part of the MSc in Data Science at the International Hellenic University. This thesis deals with a classification problem concerning prediction of users’ intent using Natural Language Processing (NLP) and machine learning techniques. Intent determination is a crucial part of Spoken Language Understanding systems. Recurrent Neural Networks are particularly effective in this task because they capture the order of the words in the text, which is one of its most essential characteristics. This work tackles this problem using pre-trained word embeddings and an LSTM neural model to extract the features for intent prediction. Other classifiers like SVM, Logistic Regression and MultiLayer Perceptron were tested too, without achieving the performance of the LSTM approach. Those methods are evaluated on the benchmark ATIS dataset. Compared to the current state of the art methods, the approach of this thesis achieves the best results, using a lightweight model containing a single LSTM layer which outperforms more complicated approaches that may also be slower or have overfitting issues. Specifically, it gave 0.44% absolute error reduction compared to the current state-of-the-art.