Online content providers recommend content based on the user’s viewing history, e.g. to sustain user interest and promote content that users are likely to be interested in. Current recommendation systems generally focus on one signal, e.g., content viewed by others who also viewed the present content (“co-watch signal”), to make recommendations. Co-watch signal-based recommendations tend to skew towards lowest-common-denominator content and are not capable of accurately recommending niche content. Niche content recommendations that are poorly targeted result in low click-through / view-through rates (CTR/VTR) and miss truly interested audiences. This disclosure describes recommendation systems that take contextual information into account. Context is established, with user permission, based on content that the user has interacted with in the past. Per techniques of this disclosure, content is recommended if the co-watch signal is strong and the context-matching score is high.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Lewis, Justin and Davies, Scott, "Dynamic user-affinity based curation", Technical Disclosure Commons, (May 09, 2017)