The project in five facts.
Challenge, approach and results.
The diploma thesis addresses the comparison of classical and modern NLP processing methods and their impact on accuracy, processing speed, comprehensibility of results and user experience. Natural language processing represents one of the key areas of artificial intelligence, as it enables computer systems to understand, analyse and interpret texts in natural language.
The theoretical part presents fundamental NLP approaches, from classical statistical methods to modern transformer models, and the importance of explainability in the use of intelligent systems. The experimental part includes a comparison of the TF-IDF method with logistic regression and the BERT model.
Users tested both methods in an interactive system in Google Colab, and the system measured processing time, displayed the most important words and performed sentiment analysis. The results of the questionnaire in Google Forms captured the subjective assessment of comprehensibility and user experience.
The research showed that the classical method works exceptionally quickly and offers clear explanations that enable users to understand the results well. BERT without language adaptation often returns incorrect classifications for Slovenian texts, and processing is significantly slower. Due to the inexplicability of the results, the user experience with the modern method was worse. Analysis of the results showed that the complexity of modern models does not necessarily guarantee better practical usability.
Classical methods, although simple, offer stable, fast and explainable results that are often more useful for the end user. Modern models only become effective when they are adapted to the language and dataset, which requires additional resources and knowledge.
The research thus emphasises the importance of explainability, transparency and model adaptation in the development of NLP systems.
Based on the results, we can conclude that modern methods are not always the best choice in practical environments, especially when they are not adapted to the language or when transparency of results is crucial for the user. Classical methods remain important due to their speed, stability and explainability.
The diploma thesis confirms that the choice of method depends on the context of use and user expectations, and that explainability represents a key element of modern NLP systems.
Module, programme and stories.
Your diploma project for a real company.
A diploma thesis at Academia solves a challenge from a company – often the one where you do your 800 hours of placement. Read about the Razvoj aplikacij module or book a consultation.