تصنيف الجنس بناءً على تحليل الأسلوب اللغوي باستخدام تقنيات التعلم الآلي والتعلم العميق

Gender Classification Based on Linguistic Style Analysis Using Combine Machine Learning and Deep Learning Techniques

Abstract

The enormous amount of textual data available around the world, including articles and social media content, has gave the importance for media platforms such as Twitter to employ this data within gender classification. This is an interesting topic for many practical applications such as marketing, recommendation systems and cybercrime. Gender classification in text refers to the process of classifying individuals into one of two genders, male or female, typically based on observable linguistic characteristics.
Natural language processing (NLP) has gained popularity in machine learning field. NLP techniques automatically apply gender classification by using linguistic and stylistic features. This leads to greater participation and satisfaction, improved customer support, and customized content delivery. The dynamic nature and huge vocabulary of the language makes it difficult to determine an author’s gender based on linguistic style, it has been a challenge, while the proposed feature extraction method has great importance in overcoming this problem and creating a precise distinction between males and females.
The aim of this thesis is to improve gender classification accuracy based on her/his linguistic style in general domain dataset and specific domain dataset. To achieve this aim, extracting gender textual nuances using text similarity for gender classification improvement and three models have been applied. The first model was applied by three machine learning classifiers Random Forest (RF), Logistic Regression (LR) and Support Vector Machine (SVM) to obtain gender labels and prediction probabilities of them. The second model was applied through the pre-used successful structures of the CNN models to get gender labels and their probabilities. Finally, linear combination mechanism was used by combining additional weights with the label probability results of the previous two models to compute the final prediction probability.
The highest accuracy results of the two datasets (Twitter and TripAdvisor) were acquired. The machine learning model achieved 87.8% on Twitter, while it achieved 75.1% on TripAdvisor. The deep learning model obtained 89.1% on Twitter and 76.3% on TripAdvisor. Finally, the linear combination method achieved (89.6%, 77%) on Twitter and TripAdvisor respectively.
The utilization of the suggested feature extraction technique was crucial in achieving superior outcomes in comparison to earlier researches. In addition, the fact that general domain dataset Twitter’s vocabulary is more diverse that helped us outperform TripAdvisor, a specific domain dataset with less accuracy because of its language related only to hotels and restaurants. In order to achieve the maximum accuracy feasible, the using of linear combination strategy utilizing deep learning and machine learning was very important.