"False"
Skip to content
printicon
Main menu hidden.
Published: 2024-05-02

Deep-learning models provides relevant ads without tracking users

NEWS New research from the Industrial Doctoral School at Umeå university offers a promising solution to serving ads without compromising user privacy. Arezoo Hatefi’s doctoral thesis shows how deep learning can be used for effectively categorising news articles and displaying ads based on the content of the page rather than user behavior or personal data.

With the growing volume of daily news online, automated processing such as categorisation and summarisation has become essential. In her doctoral thesis, Arezoo Hatefi has developed innovative models that can be trained to recognize the content of news articles, even with unlabeled or partially labeled datasets. This is particularly useful in real-world applications where obtaining a completely labeled dataset can be expensive or impractical.

The methods can be used to improve contextual advertising, which is a type of advertising that addresses privacy concerns associated with cookie-based advertising by placing ads solely based on web page content, without tracking users or their online behavior.

“Since news media heavily rely on advertising, there is a substantial market for contextual advertising strategies”, says Arezoo Hatefi, a doctoral student at the Industrial Doctoral School and Department of Computing Science at Umeå University.

Mimics the brain

The past decade has seen major advancements in deep learning. Deep learning is a subset of machine learning that uses multi-layered neural networks to mimic human brain function, allowing it to perform complex tasks like text recognition or image identification. Training involves giving the model lots of data and adjusting it to make better decisions over time.

“The thesis proposes new methods for sorting text data into topics, even when only a few examples are available for each topic. Additionally, it introduces techniques to group topics in news and discover new ones across various sources”, says Arezoo Hatefi.

Synergy between text and images

Given that online news reporting often includes different elements like text, images, video, and audio to convey information, the thesis also investigates the synergy between these in news article analysis.

“The proposed models are mainly designed for news monitoring and contextual advertising, but they also introduce new ways to categorise text, group similar articles, and find emerging news stories. Comparison with state-of-the-art baseline models demonstrates their effectiveness in addressing the respective objectives,” says Arezoo Hatefi.

Arezoo Hatefi’s doctoral project has been funded by the Industrial Doctoral School at Umeå University, Codemill and Aeterna Labs.

Read the full doctoral thesis

About the Industrial Doctoral School

The Industrial Doctoral School is based on collaboration between the University, researchers and businesses or organisations. The aim is to combine benefits for both society and the external party while training new high-quality researchers. The doctoral student also receives a tailored academic course package. The doctoral school is open to all disciplines and the doctoral student is employed at Umeå University.

Read more on the Industrial Doctoral School website