A time sensitive system for recommendations

Major Telecom providers offer their consumers a wide range of services, including video content both with live television and video on demand. This kind of content is consumed on different types of devices, e.g. mobile or landline connected internet boxes (setup box). Some households might even have multiple boxes.

Due to nature of data and, in many cases, multiple hardware devices per household, aggregating data per client or household usually means multiple users per profile.

We developed highly distributed workflows using Apache Hadoop and Oozie for a major Telecom operator in order to provide clustering and segmentation of its consumers. Analysis of households consumption and pattern identification enables specific and targeted advertising as well as content recommendation. We were able to provide a time sensitive content recommendation with cyclic patterns detection to better match the multiplicity of users for a single profile.