Currently, Asia faces a narcotic drug addiction problem. In social networking services, such as Twitter, some drug addicted users
converse about behaviours related to narcotic drugs. This research proposes a new Narcotic-related Tweet Classification Model
(NTCM) that uses data preprocessing. Two new data preprocessing methods, Sentence Vector of Word Embedding (SVWE) and
Sentence Vector of Word Embedding with Feature Extension (SWEF), are introduced to prepare data for the NTCM. The proposed
data preprocessing method uses the reduction of the dataset to produce an SVWE. Word embedding is generated by deep neural
networks using the skip-gram model. The authors further extended some features to SVWE to produce a new dataset called SWEF;
these datasets were used for the dataset in the NTCM. The authors collected data with keywords related to narcotic drugs from Twitter
in Asia. The authors investigated a text classification model using a Support Vector Machine, Logistic Regression, a Decision Tree,
and a Convolutional Neural Network. Logistic Regression with the SWEF provided the best approach for the NTCM compared with
state-of-the-art methods. The proposed NTCM showed correctness and fitness by accuracy (0.8964), F-Measure (0.895), AUC (0.949),
Kappa (0.7131), MCC (0.714), and low running time performance (1.04 seconds).
Keywords
Data mining, Data preprocessing, Feature reduction, Narcotic drug, Text classification, Text vectorization, Word embedding