A Study of the Drivers of Movie Rating Recommendations : Optimal Movie Prediction via Linear Regression

In recent years, online retailers have increasingly relied on state of the art algorithms to help predict user satisfaction and recommend movies to consumers. The major limitation of advanced state of the art computational algorithms such as SVD and latent factor models using implicit feedback is that these conceal the marginal effects of the most significant independent variables. This paper uses traditional linear regression to develop a recommender system which is as good as more advanced computational algorithms. Results reveal that the most important variables in predicting movie recommendations are ratings of user with extreme differences on other movies, age specific (conditioned) effects of movie ratings per movie and user average ratings conditioned on genres. These 3 elements allow regression models to match black box and matrix algorithms in recommender movie rating performance. This furthers the science behind recommender and also discusses a shock expectation shock minimization theory for enhancing movie watching experiences. Priming the users with appropriate expectations or frames for watching a movie should enhance user satisfaction. This supports query theory which states preferences are not recalled but dynamically built and can be influenced