Creating products that satisfy the market is critical to companies as it determines their success and revenue. Online services, such as Amazon, eBay, and Netflix, use data-driven approaches to recommend products to their customers. Currently, experts use their judgment to estimate solutions to designing a new product that will satisfy customers, but this does not scale or allow leveraging massive datasets.
Prof. Danai Koutra and her colleagues sought to identify how they can design new movies with features tailored to a specific user population. Their goal was to design a successful movie that will attract the interest of a targeted demographic by leveraging the large amounts of available data like user ratings, reviews, and product characteristics.
Their research differs from previous studies because it does not aim to find the best existing movies that the target users are likely to enjoy, which is the goal of typical recommendation systems. Instead, given the user preferences, it determines the features (e.g., actors, directors, genre) for a new movie that is likely to attract the largest number of users.
The researchers used real-world data with 5 million movie ratings and 175,000 movie-features, and approached the problem by separating the solution into two phases: user-feature preference inference, and model-based design.
The results showed that their path-based method is highly scalable and effectively provides movie designs oriented towards different groups of users, including men, women, and adolescents. Also, they showed that their algorithm is superior to baseline methods in terms of movie preference predictions.
Even though they tailored their approach to the design of movies, it can be generalized to other products as long as reviews and product features can be identified. They hope to extend their work in the future to include other elements of the creative movie process, like the plot of the movie, the screenplay, and the soundtrack.