With the number of publishers growing every year it becomes more and more complicated to deal manually with campaigns management. It is easy to get lost for humans with managing campaigns for thousands of publishers, so the usual reaction is to concentrate on some most perspective or largest and well-known of them, having the rest majority overboard. It is evident that such a strategy is far from optimal and some sophisticated tools needed to deal with such high cardinality features as publishers are.
There are several possibilities for dimensionality reduction, here we will uncover the most effective approach, suitable in the case of OpenRTB ecosystem…
This report intends to contribute a more precise understanding of publishers embeddings and their impact on improving performance oriented campaigns.
1. Evaluation of major neural network embedding models
o Word2Vec model
o Negative sampling
o Hierarchical SoftMax
2. Practical application of Neural networks embeddings in optimizing spends with DSP based on data retrieved from Fiksu DSP