Browsing by Author "Weerakoon, W.M.H.G.T.C.K."

Now showing 1 - 2 of 2
  • No Thumbnail Available
    Item
    A Comparative Study: Best Machine Learning Algorithm for Social Media Sentiment Analysis
    (Uva Wellassa University of Sri Lanka, 2020) Manthrirathna, M.A.L.; Weerakoon, W.M.H.G.T.C.K.; Rathnayaka, R.M.K.T.
    Sentiment analysis is a field of study that aims to derive the sentiment or the opinion of a text using natural language processing techniques. Performing sentiment analysis on Twitter data has a vast number of applications including predicting stock market prices, product recommendations, etc. Sentiment analysis can be done in lexicon-based, machine learning-based, or hybrid approaches. K Nearest Neighbor, Support Vector Machine, Logistic Regression, Naïve Bayes, K Means Clustering, Decision Trees, and Random Forest are the few most popular machine learning algorithms. This study aims to conduct a comparative analysis among the usage of K Nearest Neighbor, Support Vector Machine, Logistic Regression, and Multinomial Naïve Bayes machine learning algorithms combined with sentword net lexicon to suggest which one provides the best accuracy in sentiment classification of Twitter data. A data set of 1028 tweets was acquired using the Twitter Standard Search API (Application Programming Interface) and Tweepy python library. The name of a popular brand of mobile phones was used to search for tweets. 570 tweets remained after the duplication removal and cleaning process. Then the remaining data was classified as positive, negative, or neutral using sentiword net lexicon and used to train selected machine learning algorithms.80% of the data was used for training and 20% was used for testing. Word counts in the tweets were used as features. Multinomial Naïve Bayes is proved to be the best machine learning algorithm with a model accuracy of 74.56% and K Nearest Neighbor (k=3) is the worst-performing algorithm with an accuracy of 54.38%. Logistic Regression and Support Vector Machine (linear kernel) respectively had accuracies: 72.80% and 70.17%. The result of this research proves Multinomial Naïve Bayes performs relatively better in Twitter sentiment analysis than K Nearest Neighbor, Support Vector Machine, Logistic Regression. This is because two basic assumptions for applying the Multinomial Naïve Bayes algorithm: feature independency and multinomial distribution are well satisfied by the features selected for this study. Also, Multinomial Naïve Bayes can perform well with high dimensional data like tweet text. On the other hand, the poor performance of the K Nearest Neighbor is due to the same reason. K Nearest Neighbor cannot handle a large number of features very well. Keywords: Sentiment analysis, Twitter, Hybrid approach, Machine learning algorithms, Comparative analysis.
  • No Thumbnail Available
    Item
    Survey on Supplying the Sri Lankan Demand for Sand in Construction Industry Aided with a Sustainable Approach
    (Uva Wellassa University of Sri Lanka, 2020) Weerakoon, W.M.H.G.T.C.K.; Bandula, W.G.S.; Kokila, W.G.S.
    Demand for sand is ever increasing, as enormous construction projects are initiated in the country’s suburbs. Mining sand and all other construction raw materials have to be increased to extreme levels to match the demand. But, the raw materials for constructions, mainly sand, are natural resources that cannot be renewed and regeneration of those natural resources need millions of years. Hence, the sustainability in mining sand and related raw materials are in a critical stage, if extraction happens in an ill-monitored manner; natural hazards will occur. Therefore, the governing body, Geological Survey, and Mines Bureau (GSMB) took initiatives to introduce mining licenses, so the authorities (the Police) can identify the illegal miners of raw materials. Starting from 2017, a sudden drop can be observed in the sand extracted from sources tracked at GSMB. When the analysis was carried out to determine the theoretical sand consumption in Sri Lanka using sand to cement ratio, the consumption during the 2017 and after are 69% more than the supply permitted from GSMB. This means that the only possible way of supplying the demand is to use illegal means in sand mining and transporting, which is thereby untraceable to GSMB. Furthermore, when identifying the factors which trigger this cause, it was determined that the current licensing process is cumbersome. The sudden shutdown at mining sites due to legal cases and civil unrest also increased illegal sand mining. The miners had a demotivation on pricing the sand market value above reasons, political intervention in the industry, and rise of the royalty charges and thereby used illegal means to extract and mine sand. As per this study, it could be concluded to have a scheme of interrelated concepts that aided using ICT so that GSMB can regain the total control of mining and transportation and eliminate illegal sand mining, thereby meeting the supply to the national demand without compromising the nature. Keywords: E-licensing, Environment protection, Information Communication Technology, Illegal mining, Sustainable governance