Customer long term propensity driven Prime Video page composition
The Prime Video Homepage provides customers with several carousels to explore the diverse catalog. Each of these carousels is constructed around a certain theme. It’s not only important to compose the page with individual carousels relevant to the customer, but also balance different customer and business aspects. The Prime Video business positions itself as an entertainment hub with diverse content types such as Movie/TV Shows/Sports etc. with different offers such as free with Prime membership, individual title purchase and channel subscriptions. From a customer perspective, we would like to recommend a certain offer/content type only if they have high propensity to that offer/content type. Also, when customer looks at a page, we would like them to see diverse content in the adjacent carousels. To address these requirements we first develop a long term customer propensity model. Then we linearly combine propensity with carousel relevance scores so that we balance between customer’s immediate streaming needs and long-term interests. On top of the propensities we apply upper confidence bound (a multiarmed bandit method) so that customers can explore the unexplored offer/content types. Next we make use of customized maximum marginal relevance criterion so that neighboring carousels are not from the same offer/content type. The balance between different offer/ content types with customer relevance is analyzed using Pareto front so that we can select appropriate treatments with best balance for an A/B experiment. The best performing treatment from A/B experiment is found to have significant improvement (+4.6%) in customer engagement metrics as compared to control.