Abstract:
Sentiment analysis is involved in almost every domain and it is a powerful feedback tool for advertisers and manufacturers. Advertisements play an important role in promoting products on social media platforms. These ads require careful execution as dire consequences await in case things go wrong.
During this research, three possible datasets of user reviews were extracted from standard and unorthodox (emotional and non-stereotypical) advertising campaigns. A corpus containing 13000 YouTube comments and manually annotated using 3 basic sentiment label (positive, negative, neutral) and compared with two auto labelling tools: textblob and sentiment intensity analyzer. Self-dictionaries were made in order to convert Roman-Urdu comments into English language. Term Frequency Inverse Document Frequency (TF − IDF) was then used to transform the textual description of comments/reviews into feature vectors. Various machine learning techniques were applied such as Naive Bayes classifier, Support vector machine, Decision tree, Random forest and Logistic regression to train a model based on positive, negative and neutral sentiment. A comparative analysis was performed to identify the techniques, which can give good result. Performance of five ML algorithms was evaluated for each dataset using 10-fold cross validation with the ratio of training data and test data: 90:10. F-measure score was used for performance evaluation. As the result, SVM classifier shows higher F-Measure score while considering for all three datasets with manually annotated dataset.
