Tabularized Search Results

Abstract

This thesis focuses on the problem of generating search engine results as a table. These are tables describing a list of entities in terms of their features. The goal of making these tables is to make it easier and faster for users to finding information related to their search queries. In resent years we have stated to see changes to the Search Engine Result Pages (SERP). These changes are often centered around presenting entities to the user.

We introduce the term "entity table" to describe such tables, and the term is further explained. The task of making entity tables is formalized, and we identify and explore the challenges that this task introduces. The problem is broken down in to two main task: identifying what rows the table should have, and identifying the most appropriate columns. For the former of these two problems, we explore the task of filtering out inappropriate entities. For the latter, we explore ways of ranking the properties in terms of how well it would fit as a column in the table. Approaches to these task are explained and implemented.

To evaluate our approaches, a gold standard is made for both tasks, listing the ideal output for a set of test queries. An evaluation of the approaches shows us which work the best for our tasks. Lastly, we demonstrate how we can use our implementations to visualise the "entity tables" on a web page.