Improving Entity Based Information Retrieval In Video

This PhD by project portfolio advances entity-based information retrieval in video, enabling deep search across multi-modal media in the e-commerce sector. Based in applied industrial research, the resulting work led to several granted technology patents that form a commercial system. The system allows organizations to identify and extract information from video content that was previously challenging to access. The portfolio describes three interconnected projects within a single system designed to extract and provide actionable commercial and industry intelligence at scale. The first project focuses on automatically segmenting and extracting data from webpages, creating a mapping database essential for entity normalization. The second project identifies and classifies key components within video, such as objects, people and brands; these are then aligned across different modalities. Having such topics of interest pre-identified helps reduce the subsequent search space. Finally, the third project details a process for entity normalization through a multi-modal approach to brand and product identification, significantly improving search accuracy. Collectively, these projects enhance brand, logo, sentiment and topic detection, offering a robust foundation for future development in video-based information retrieval.

Name dc:type.qualificationname

Doctoral Thesis

Level dc:type.qualificationlevel

Student thesis

Year dc:date.issued

2026

Author dc:creator

• O'Neill, Allen

Advisors dc:contributor.advisor

• Wilson, Ian
• Ware, Jonathan

Language dc:language

eng

Identifier

oai:pure.atira.dk:studenttheses/9d52ba01-2abe-41f7-9aa2-2c437e8454fc

OAI identifier oai:identifier

oai:pure.atira.dk:studenttheses/9d52ba01-2abe-41f7-9aa2-2c437e8454fc