TILEC Work In Progress: Madina Kurmangaliyeva

main presenter:  Madina Kurmangaliyeva. others: Tobias Klein; Jens Prüfer, Patricia Prüfer

Do search engines really need a lot of user data to provide good results? We asked a search engine to produce counterfactual sets of search results for a representative sample of queries, using different amounts of user data. We compare these search results against each other  automatically based on the overlap of top results for each query with those produced with the most user data. We also compare the quality of search results against those of Google and Bing. In addition, we ask human scorers to assess the quality for a random sample of queries. Our preliminary results indicate that user data is crucial for quality, especially for rare queries, which represent 75% of all queries searched on the internet. We proceed with the conclusions about the degree of data-drivenness of the market for internet search.