## Executive Summary
The project aims to optimize the average browsing time. We tested the four factors - Tile Size, Match Score, Preview Length, and Preview Type for their interaction and main effects. We carried out the factorial experiments including hypothesis tests on four-way, three-way and two-way interactions and finally the main effects, and followed by a response surface experiment to locate the optimal condition and corresponding response value, which is presented as follows:
optimal condition - Preview Length = 70, Match Score = 76, Preview Type = "Teaser/Trailer", Tile Size could be any number between 0.1 and 0.5
predicted browse time - 11.0770
95% prediction interval for browse time - [10.8160, 11.3380]
## Introduction
The interest of this problem lies into minimizing the user’s browsing time before clicking into any one of the recommended movies presented to them. Four design factors were chosen for this experiment:
First, we started off with the series of factorial experiments to learn which factors influence the browsing time and how that may be exploited to minimize the browsing time. We tested on the 4-way interaction, then the 3-way interaction, then the 2-way interaction, and then the main effect. At each step, as the interactions are not significant, we keep on reducing the models by dropping these interactions. Finally, we used a surface response experiment to find the optimal value of the influential factors.
In this report, we will present the details of the design of a series of factorial experiments on the influence of multiple design factors and the final model reduced by the factorial experiments; a central composite design for response surface experiment; and a detailed process of deriving the optimal condition and final predictions of visitor's browse time using the response surface experiment.
## Factorial Experiments
In order to determine among all the main effects of four design factors and all interactions between them what are significant to the response variable, i.e. visitor's browse time, we perform a series of factorial experiments with linear regression.
Please note that for the sake of simplicity, we will address the name of design factors as their abbreviations starting from this section, namely,