Consumer reviews are very instrumental in projecting the development and direction of businesses. The wealth of information provided with online reviews can be analyzed to access the general sentiment of the consumer towards the product/service. This sentiment analysis will provide general idea and quantitative measure of the consumer’s sentiment.
The required data set, in .tsv format, consist of a text review and a binary input of 0 or 1 indicating whether the consumer liked the item or not. These will be balanced using “ROSE” package, which attempts to generate synthetic data by randomly over-sampling examples in case of imbalance data.
Then using a text mining package “tm”, it will mine for the words/word base of interest such as keywords that conveys sentiment based on dictionary list provided by the package. This “stream-lined” tokens that were independent variables on the raw dataset is then joined in a matrix through either of the three algorithms of the user’s choice such as Bag of Words, Tfdfi, or Bigram and will be analyzed based on its sentiment. Next, using “caTools” package, we split the data set to 0.2 data and 0.8 training set split ratio to be fed to different classification models of provided by the existing machine learning algorithms training models.
The package also compares the accuracy of the prediction model based on the machine learning algorithm which the user can choose from: decision tree, naïve bayes, or random forest. The result is different parameters of the sentiment prediction that is built in a confusion matrix: Accuracy, Precision, Recall, and F1 score. Lastly, these parameters will be visualized as a wordcloud as well as presented in an interactive Shiny app. Overall, this package will put together different algorithms and visualization of a version of natural language processing (NLP) and will allow a broader audience of user to access this tool.