In a world where people rely to an ever-greater extent on digital technology to access documents, it has become of central concern to ensure that all categories of users benefit from accessibility standards in their interaction with digital files. Although many improvements have been made in the field of content accessibility, especially for the web, research shows that PDFs---despite representing the de-facto standard in a variety of contexts--- still pose significant challenges and barriers for blind and visually impaired (BVI) users. Building on literature on human-centered computing, this thesis aims to explore issues and potential solutions for PDF document accessibility through a twofold approach. Firstly, it analyzes the limitations and hindrances of current PDF documents, as well as promising accessibility solutions, as per the existing literature. Subsequently, it investigates novel opportunities and design approaches to support BVI users in their interactions with PDFs. More specifically, drawing from related research on the design of reading interfaces, this work conducts an initial step of feature review and analysis. Starting from a recent study proposing a reference set of innovative features, originally conceived to improve the reading experience of academic papers, this thesis extends and adapts these initial features to the challenges faced by BVI users with their everyday PDF interactions, through a process of analysis and comparison with relevant literature. The findings of this gradual refinement finally culminate in a new list of features, reformulated and tailored to BVI accessibility needs and use cases. Additionally, following a flourishing line of research on the use of conversational agents (CA) for accessibility, this thesis evaluates the effectiveness and reliability of widely adopted LLM-based chatbots, when used to access and parse an average PDF document through conversation. Eventually, consistent with the intended outcomes, the resulting benchmark and evaluation can serve as an actionable premise for further exploration of this growing design space.
In un mondo sempre più dipendente dal digitale per la fruizione di documenti, è diventato prioritario assicurare che tutte le categorie di utenti possano beneficiare di standard di accessibilità nell’interazione con i contenuti dei file. Nonostante i numerosi progressi svolti nel campo dell’accessibilità, specialmente in ambito web, la ricerca mostra che i PDF – per quanto siano lo standard de-facto in molti contesti – ancora pongono significative barriere per gli utenti non vedenti e ipovedenti (“utenti BVI”). Basandosi sulla letteratura di \textit{human-centered computing}, questa tesi si propone di indagare problematiche e potenziali soluzioni per l’accessibilità di documenti PDF, tramite un approccio duplice. In prima istanza, vengono analizzate le limitazioni di accessibilità presenti negli attuali PDF, così come alcune soluzioni promettenti. Successivamente, vengono esplorate nuove opportunità e approcci di design per gli utenti BVI. Più nello specifico, partendo dalla ricerca sul design di interfacce di lettura, questo lavoro svolge un primo step di analisi e valutazione di features. Partendo da un recente studio di riferimento che propone un insieme di funzionalità originariamente pensate per l’esperienza di lettura di articoli accademici, questa tesi, attraverso un processo di analisi e comparazione con la letteratura, estende e adatta queste features iniziali alle sfide affrontate da utenti BVI nelle loro interazioni quotidiane con i PDF. I risultati di questo graduale processo di raffinamento culminano infine in una nuova lista di funzionalità, riformulate e adeguate ai bisogni e casi d’uso dell’accessibilità BVI. Successivamente, seguendo una fiorente linea di ricerca sull’impiego di agenti conversazionali (CA) a scopo di accessibilità, questa tesi valuta l’efficacia di alcuni popolari chatbots basati su Large Language Models, nel loro utilizzo per rendere accessibili documenti PDF tramite conversazione. In linea con gli obiettivi di questo lavoro, il benchmark risultante da questa analisi si propone infine come una valida premessa per successive esplorazioni di questo design space.
Accessibility approaches for interaction with PDF documents through conversational agents and innovative reading features
Barisani, Francesco
2024/2025
Abstract
In a world where people rely to an ever-greater extent on digital technology to access documents, it has become of central concern to ensure that all categories of users benefit from accessibility standards in their interaction with digital files. Although many improvements have been made in the field of content accessibility, especially for the web, research shows that PDFs---despite representing the de-facto standard in a variety of contexts--- still pose significant challenges and barriers for blind and visually impaired (BVI) users. Building on literature on human-centered computing, this thesis aims to explore issues and potential solutions for PDF document accessibility through a twofold approach. Firstly, it analyzes the limitations and hindrances of current PDF documents, as well as promising accessibility solutions, as per the existing literature. Subsequently, it investigates novel opportunities and design approaches to support BVI users in their interactions with PDFs. More specifically, drawing from related research on the design of reading interfaces, this work conducts an initial step of feature review and analysis. Starting from a recent study proposing a reference set of innovative features, originally conceived to improve the reading experience of academic papers, this thesis extends and adapts these initial features to the challenges faced by BVI users with their everyday PDF interactions, through a process of analysis and comparison with relevant literature. The findings of this gradual refinement finally culminate in a new list of features, reformulated and tailored to BVI accessibility needs and use cases. Additionally, following a flourishing line of research on the use of conversational agents (CA) for accessibility, this thesis evaluates the effectiveness and reliability of widely adopted LLM-based chatbots, when used to access and parse an average PDF document through conversation. Eventually, consistent with the intended outcomes, the resulting benchmark and evaluation can serve as an actionable premise for further exploration of this growing design space.| File | Dimensione | Formato | |
|---|---|---|---|
|
MSc_Thesis_Francesco_Barisani.pdf
non accessibile
Descrizione: MSc Thesis by Francesco Barisani
Dimensione
3.72 MB
Formato
Adobe PDF
|
3.72 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/247652