Bias in search: Evaluating search results through rank and relevance based measures

Official URL: https://risc01.sabanciuniv.edu/record=b2767124_(Table of contents)

Search is ubiquitous. People continue to seek information through popular search engines, Bing and Google as well as online search platforms, YouTube. Nonetheless, they tend to think that these platforms are objective by only displaying information without injecting any bias. Since users are more susceptible to bias when they are unaware of it, it is important to evaluate the retrieved search results of the aforementioned platforms with respect to bias. This thesis analyses two main things as search engine bias towards controversial issues and gender bias in the context of online education. For evaluating specifically search engine bias, three novel rank and relevance-based measures have been proposed and search results of two widely-used search engines Google and Bing have been analysed through web documents’ content with respect to stance (in support or against), and ideological bias (conservative or liberal). Then, the impact of geolocation on the bias has been investigated. Lastly, in the scope of search engine bias, the source of bias has been tracked, to check whether the bias (if exists) comes from the input data, or the ranking algorithm. For assessing gender bias in online education, two new rank and relevance based measures that are more suitable in the scope of gender bias have been proposed. Further, video search results returned by YouTube towards the queries in STEM and NON-STEM fields have been analysed using narrators’ information. Lastly, the source of gender bias has been investigated by proposing the specifically-curated gender bias measures