Corvidae’s Machine Learning Process
At the centre of Corvidae’s attribution model is LSTM (Long-Short Term Memory). LSTM is a type of recurrent neural network capable of learning order dependence in sequence prediction problems.
We undertook a two-year project with Edinburgh University which looked at applying Machine Learning techniques (Random Forest) and the principles of LTSM to create technology that applies the deep learning techniques to the issue of marketing attribution.
How does the Machine Learning component work?
Once your analytics data has been cleansed and ordered, Corvidae will perform Machine Learning processing on the data to unlock insight around the customers and the patterns of their behaviour.
The Machine Learning component of Corvidae is used to understand the value of an individual visit. This then allows the relative value to be attributed to the relevant media channels in the journey. By identifying the types of journeys which lead to conversion, the value to individual channels in the process can be accurately attributed.
Throughout the Corvidae development process, the Machine Learning techniques have evolved, due to the research undertaken, to improve the accuracy and value of the output it is able to generate.
The computationally expensive machine learning process is undertaken in Databricks. This is a platform which manages the setup and configuration of an Apache Spark framework, with all required components, such as clustering, data streaming, and the computational management to most efficiently handle the processing.
How is Corvidae’s Machine Learning technique different to other attribution models?
We replace these models in two fundamental ways:
- Our patented machine learning process rebuilds the raw, cross-channel analytics data.
- We then unify online and offline data before applying neural network processing to calculate attribution accurately.