Analysis of the Corpus with Naïve Bayes in Determining Sentiment Labeling

Complexity of the data is a major problem in producing in-depth understanding. So that the help of artificial intelligence is needed, namely Natural Languange Processing (NLP) which is the key to the problem because it can allow us to process text more effectively and increase the opportunity to allow computer interaction with human language to be more natural. Sentiment analysis is one of the main applications in the automation process to find out a person's opinion on a subject based on the text they write, one of which is a comment on cars in Indonesia such as Toyota, Daihatsu, Honda and Suzuki. This research utilizes a quantitative methodology and uses a machine learning model, namely the Naive Bayes algorithm, to classify the text into positive or negative based on the calculation of the probability of words in the text. A corpus is also used as a dictionary/word pattern using InsetLexicon to analyze the sentiment of the text based on keywords with predetermined weights that will be pre-processing first. This was done to assess the performance of Naive Bayes analysis and the effectiveness of InsetLexicon in improving the accuracy of sentiment on text. The research results from Table. 21 that the total value of comments on the highest aspect of all aspects on 4 types of cars and each brand is the aspect of "Comfortable" or comfort around 91% with the acquisition of average Recall, Precision and F1-Score around 0.83, 0.85 and 0.87 and the overall accuracy of the dataset is around 71%.