The challenge
The client is a large media company that owns TV and radio channels, audio podcasts, magazines and newspapers. It wanted a personalised recommendation system for its existing users and content consumers, to increase engagement, reduce churn and raise customer lifetime value.
The content is diverse, from TV programmes and shows to news articles, and every user has their own preferences. The system had to recognise individual consumption patterns, understand content preferences and recommend new items that users are likely to consume.
Our solution
We developed a recommendation system that combines several factors. It estimates each user’s consumption patterns, understands their content preferences such as topics of interest and content type, estimates their demographics and the technical means they use to access content, and assesses how similar content items are from different perspectives.
A hybrid recommender brings together the outputs of several analytical components.
- NLP-based topic discovery and content tagging module
- Content item similarity analytics
- Consumption patterns extractor
- Collaborative filtering-based recommender
- Item-to-item recommender
- Hybrid recommender that combines all of the above
The outcome
The solution increased the diversity of content consumed by users by 5%, and with it the lifetime value of a customer.
The client can also identify similar groups (clusters) of consumers directly, target them with the content they are most likely to enjoy, and estimate potential demand for specific content.
