The thesis addresses the distinction between human-written texts and AI-generated texts in two different stages. After an exploratory analysis and essential pre-processing, topic modeling and topic mining techniques are applied to describe the corpus’s thematic structure and assess the possible presence of differences between the two groups of texts. Subsequently, using classification models, the aim is to discriminate between human and generated texts by comparing lexical representations (frequency-based) and embeddings; interpretability is supported by SHAP values, used to highlight which features are most relevant in the models’ decisions.
La tesi affronta la distinzione tra testi umani e testi generati dall’AI in due passaggi differenti. Dopo un’analisi esplorativa e un essenziale pre-processing, si applicano tecniche di topic modeling e topic mining per descrivere la struttura tematica del corpus e valutare l’eventuale presenza di differenze tra i due gruppi di testi. In seguito, mediante modelli di classificazione si mira a discriminare testi umani e generati, confrontando rappresentazioni lessicali (basate su frequenze) ed embedding; l’interpretabilità è supportata dai SHAP values, impiegati per evidenziare quali caratteristiche risultino più rilevanti nelle decisioni dei modelli.
Analisi tematica e classificazione di testi umani e generati dall'AI
CONTERNO, JACOPO
2024/2025
Abstract
The thesis addresses the distinction between human-written texts and AI-generated texts in two different stages. After an exploratory analysis and essential pre-processing, topic modeling and topic mining techniques are applied to describe the corpus’s thematic structure and assess the possible presence of differences between the two groups of texts. Subsequently, using classification models, the aim is to discriminate between human and generated texts by comparing lexical representations (frequency-based) and embeddings; interpretability is supported by SHAP values, used to highlight which features are most relevant in the models’ decisions.| File | Dimensione | Formato | |
|---|---|---|---|
|
Conterno_Jacopo.pdf
accesso aperto
Dimensione
6.14 MB
Formato
Adobe PDF
|
6.14 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/98936