| Abstrakt: |
Internet and social media explosive growth has led to the rapid and widespread dissemination of information, which often takes place anonymously. This anonymity has fostered the rise of uncredited copying, posing a significant threat of copyright infringement and raising serious concerns in fields where verifying information's authenticity is paramount. Authorship Attribution (AA), a critical classification task within Natural Language Processing (NLP), aims to mitigate these concerns by identifying the original source of content. Although extensive research exists for longer texts, AA for short texts, namely informal texts like tweets, remains challenging due to the latter's brevity and stylistic variation. Thus, this study aims to investigate and measure the performance of various Machine Learning (ML) and Deep Learning (DL) methods deployed for feature extraction from short text data, using tweets. The employed feature extraction methods were: Bag-of-Words (BoW), TF-IDF, n-grams, word-level, and character-level features. These methods were evaluated in conjunction with six ML classifiers, i.e. Naive Bayes (NB), Support Vector Machine (SVM), Decision Tree (DT), Logistic Regression (LR), K-Nearest Neighbors (KNN), and Random Forest (RF) along with two DL architectures, i.e. Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs). The highest accuracy achieved with an ML model was 92.34%, using an SVM with TF-IDF features. Even though the basic CNN DL model reached 88% accuracy, this outcome still surpassed the previously established baseline for this task. The findings of this research not only advance the technical capabilities of AA, but also extend its practical applications, providing tools that can be adapted across various domains to ensure proper attribution and expose copyright infringement. [ABSTRACT FROM AUTHOR] |
