M2PI Case Studies Virtual Seminar: Irushi Jayathunga and Hiva Gheisari

In this talk, we will share a comparison of how well different search engines find relevant articles, focusing on nautical crimes. The project involved collecting and cleaning data, looking at how many articles each engine retrieved, when those articles were published, and how relevant information was. A machine learning model to help categorize the articles is used.

To measure how relevant each article was, we used two techniques: TF-IDF and Cosine Similarity. We will explain the mathematical background of these techniques and how they work using easy-to-follow examples, and show how they help us judge the performance of each search engine.

The results reveal notable differences among search engines in terms of the quantity, recency, and relevance of retrieved information. In particular, Google tends to retrieve older but more relevant articles, while Bing returns more recent content with comparatively lower relevance.

This project was carried out as part of the M2PI 2024 workshop in partnership with NCIS. It involved a collaborative effort from Clotilde Djuikem (University of Manitoba), Hiva Gheisari (University of Lethbridge), Irushi Jayathunga (University of Calgary), and Sumin Leem (University of Calgary), with guidance from industry mentor Sogol Ghattan (NCIS).