Weight Estimation and Evaluation of User Suggestions in Mobile Browsing

Abstract: This study investigates the suggestion system of a mobile browser. The goal of a suggestion system is to assist the user by presenting relevant suggestions in an ordered list. By weighting the different types of suggestions presented to the user, such as history, bookmarks etc., it is investigated how this affects the performance of the suggestion sys- tem. The performance is measured using the position, error and Mean Reciprocal Rank of the chosen suggestion as well as the number of written characters. It is also measured if the user chose to not use the suggestion system, by searching or entering the entire URL. The weights were estimated using a Genetic Algorithm. The evaluation was done by performing an A/B test, were the control group used an unweighted system and the test group used the weights estimated by the genetic algorithm. The results from the A/B test were statistically analyzed using BEST and Bootstrap. The results showed an improvement of position, number of written characters, MMR and the error. There was no change in how much the user used the suggestion system. The thesis concluded that there is a correlation between the position of the desired suggestion and when the user stops typing, and that weighting types is a way to improve said position. The thesis also concludes that there is a need for future work in regards to evaluation of the optimization algorithm and error measurement.