Effective use of artificial intelligence methods in natural language processing

The abstract and dynamic nature of language lead to certain limitations in modelling language with artificial intelligence. To minimize these limitations, it is essential to approach language and context as multidimensional structures and adopt data-specific methodological strategies for modelling. This thesis focuses on the content generation performance of artificial intelligence systems, the analysis of content, and content classification. The performance of artificial intelligence was evaluated within the scope of educational sciences, focusing on how their content generation capacities can be enhanced through effective prompt design. Questions were designed in accordance with Bloom's Taxonomy in educational sciences, and reference points guiding the preparation of taxonomy-compliant questions were carefully considered. For content analysis, the content of films was used to predict movie ratings and genres through machine learning models. Therefore, feature selection and input formation were addressed, and regression models were applied to predict movie ratings using selected features. Additionally, genre classification based on film summaries was explored. Contemporary mental health issues such as anxiety, depression, and suicidal tendencies have become identifiable through related social media posts.The impact of text preprocessing techniques and hyperparameter optimization methods on classification performance was examined. Furthermore, different methods for text vector representations were analyzed, and classification performance was evaluated. A Roberta model, applied after preprocessing by removing repetitive expressions, achieved 87.2% accuracy. It was observed that hybrid vector classification, created by combining pre-trained Roberta, word2vec, and TF-IDF vectors in training the Bi-LSTM model, increased classification accuracy by approximately 13%. Three different problems were evaluated using various methods and techniques for the effective use of artificial intelligence methods in natural language processing. The applications revealed that the diversity of textual content, the sequence and execution of preprocessing steps, and the implementation of hyperparameter methods significantly influence performance.