| Abstrakt: |
People express their opinions through social media sites like Twitter, Instagram, and Facebook. Tweets can be used to analyze opinions expressed towards various keywords, which can be anything from the names of products, companies or famous people. Consumers can utilize the insights of analyzed opinions and companies that want to monitor the public sentiment of their brands or any other situation where data on public opinions can be useful. Training data is abundantly available and can be obtained through automated means, we've obtained our data from twitter. The goal of our project is to implement the most effective algorithms and compare their accuracy. We use machine learning techniques like Max entropy, Naive Bayes, SVM and DCNN (Deep Convolution Neural Network). The data is analyzed for the presence of emoticons and keywords that either signify positive or negative expression towards the query term by Max, NB, and SVM methods. In DCNN, latent contextual semantics relationships and co-occurrence of statistical characteristics are used to obtain word embeddings between the words. Finally, word embedding is combined with n-grams and a polarity score is issued for the tweets. Our research focuses on comparing the accuracy of these algorithms to determine which are the most effective and accurate. [ABSTRACT FROM AUTHOR] |
