| Summary: | The exponential growth of Internet penetration in recent years has transformed the media
landscape. A growing number of people consume their media online, a trend paralleled by a
proportionate, and now exceeding increase in the amount of content produced. Exempted from
barriers of cost and reach, users now consume from a bevy of publishers. This paradigm has
necessitated content aggregators - platforms that offer content in bundles from different publishers.
Given the multiplicity of voices on the platform, aggregators face a novel challenge. Mere
aggregation doesn't guarantee retention - aggregated content now needs to be reliable, relevant and
personalized. To have repeat use, the aggregators need to predict a user's disposition, and collect
content that appeals to the user at a deeper level, than just her 'interests'. This study introduces a
machine learning approach using principles of recommender systems to model and predict user
behavior for a content-aggregation platform. Unlike traditional recommender systems, this study
incorporates content related meta-data in a predictive model. The study evaluates the performance
of the model through a comparison of popular classification algorithms. Finally, based on the
results of the predictive model the study undertakes a series of investigations which indicate key
bottlenecks in the approach and paves the way for future work.
|
|---|