Applying recommendation to network fault detection
Recommender algorithms, which support recommender systems, seem like tools for retail services and this is a way that they have been used with great success. Mobile network managers are a long way from this: they are looking in real time at how their customers are using their networks and how they provide a continually good service to them. This is even in locations of very high demand, or conversely in areas of very poor coverage. Because the world is not uniform, mobile networks are not perfect and things do break down. Mobile network monitoring centres are designed to receive alerts rapidly and deal with them. However, not all alerts are real, and some are more important than others. Prioritisation is needed. This is where recommendation comes in. For mobile network fault detection there is the advantage (compared to the retail sector) that there is a huge amount of historical data at every location (cell) in the network, and thus a potential or actual fault can be prioritised in relation to this. I worked as a contractor for a major mobile service infrastructure company who wanted to enhance the services they provided to their mobile network operator clients. Recommendation allowed them to do this.