Spelling Checking with Deep Learning Model in Analysis of Tweet Data for Word Classification Process

The process of preprocessing produces data that still has writing and spelling errors that refer to Indonesian words so there are problems in the next process, namely in classifying formal and non-formal words. if the preprocessed tweet data is corrected manually, it will take a lot of time with 17,378 data. The solution to this problem in this context will be to improve spelling starting from the preprocessing results and then use a dictionary of slang words and the big Indonesian dictionary which in the process data from the Big Indonesian Dictionary and the slang word dictionary will be studied for patterns with models from deep learning which is used later in the process is converted into vector embedding then from the vector vectors formed the cosine similarity technique is used which measures the similarity between two text data. In this study, the model of deep learning is the Bi-LSTM method. The results obtained show that the spell checker produces better data with an accuracy of up to 82,5% for words that have irregular sentence structures.