Abstract:
Because of the rapid development of the information communication technology, an enormous
amount of data is produced, shared across the internet and other media. Opinion mining, also known
as Sentiment Analysis, is a technique, which can be used to detect the opinion of a given sentence or
to make a judgement based the given sentence, can play a major role in automatically analysing this
data. In addition, the development of Natural Language Processing in Sri Lanka leads the Sri Lankan
native user to browse web in their native language and to express their opinions in their mother
tongue. But in most of the cases Sinhala language was named as the morphological rich,less
resourced language. An automatic solution for the text categorisation and opinion mining could be
very useful for analysing sentences from Sinhala language. This work explores a Convolution Neural
Network (CNN) based sentiment analysis techniqe, where, each word of a sentence is converted into
a numerical representation using a pretrained FastText word embedding model. These numerical
representations obtained for each word of the sentences are then used to train the CNN, in order to
predict the opinion of the given sentences at test time. This CNN model is trained and tested on a
Sinhala news comments dataset which consists of 5010 comments. There were 2520 negative
comments and 2490 positive comments. Dataset used in this project is collected by crawling Sinhala
online news sites, mainly www.lankadeepa.lk.This data was initially preprocessed by removing non-
Sinhala characters, punctuation marks and stop-words. Our model was trained and tested on 70% and
30% of the data respectively. Experiments report a testing accuracy of 85%.
Keywords: Convolution Neural Network; Natural Language Processing; Sentiment Analysis; Text
analysis for Sinhala Language
