Abstract:
With the invasion of electronic devices all areas of our life, sp eeding up the pro cess of
entering information and facilitating it is now an imp erative. We can contribute to this
by predicting the next word.
In this work we discuss the effective metho ds of predicting the next word, esp ecially
given the magnitude of current data. This work aims to apply deep learning to this problem, sp ecifically recurrent neural networks and temp oral convolutional networks, with a
primitive comparison b etween their results.
In this pro ject we use three databas es: the first one is Coursera Swiftkey, the s econd is
the b o ok: Nietzsche Writings: Volume1 by Friedrich Nietzsche and the third is the News
category from the Brown corpus in the nltk library. We prepare and insert them into
RNN and TCN mo dels.
The results were satisfactory according to the state of the art res ults and the platform
and data set used, as we reached an accuracy of 71.51% for the RNN model and 65.20% for
TCN model using the third database when taking into account the three previous words
to predict the next word. Given the results, we can say that the temporary convolutional
network competes with the recurrent neural network in the field of language modeling.
Although we obtain satisfactory results, they would have been better had it not been
for the inefficiency of the devices and the limited work environment due to the restrictions
imposed by Google Colab and Kaggle, as well as the circumstances that we faced while
we were in the process of completing this work due to the pandemic Covid19. We will
develop the research in the future by using platforms that meet the requirements of most
deep learning
