ermiasYosti/sentiment_analysis_amharic

# Sample TV and RAdio Audionce Sentiment Analysis (benchmarck tweeter sentiment anlaysis) # Author: Ermias Sene(supported by Michael Erb) # Problem Company A will soon be releasing a new mobile phone. They are concerned about its reception in the market and would like a way to monitor it. I will be building a Sentiment Analysis model to classify tweets as positive, negative or neutral. # Data The data is provided by CrowdFlower and is available for download from data.world It contains about 8500 instances. # Methodology * Exploratory Data Analysis to understand the data. * Preprocess the text data: including removing unwanted punctuation, changing to lowercase and tokenization. * Model Selection: A series of classification models were run on the data and evaluated on three metrics in descending order of importance: Average Macro Recall, a balanced Recall score for each class, and overall accuracy. # Conclusion * Using a Naive Bayes Classifier, I achieved an macro average recall score of 0.61, balanced recall scores of 0.61, 0.56 and 0.67 for Negative, Neutral and Positive classes respectively, and an overall accuracy of 60% * While not the greatest accuracy, automating Twitter sentiment analysis will be a step in the right direction in terms of efficiently monitoring the sentiment of Twitter users towards Company A's new mobile phone. # Recommendations * I recommend that Company A use Twitter's API to filter tweets with hashtags and text deemed to be related to their mobile phone. These tweets can then be classified by the model and monitored to keep track of the current sentiment regarding their phone. * Building upon the previous recommendation, an alert system can be created to monitor for changes in sentiment so that they can be addressed quickly. * Use the model to monitor sentiment regarding the mobile phone industry in general, as well as the sentiment towards competing products. # Future Work * Acquire more labeled Tweets to improve …