Automated Hierarchical Keyword Taxonomy for Google Ads
JOT Internet Media, a leader in acquiring and monetizing web traffic, is faced with the challenge of ranking millions of keywords in Google Ads. Google's keyword structure is organized into a seven-level hierarchy and encompasses thousands of categories, from general terms such as “trips” to specific terms such as “ERP systems”. The task of classifying these keywords in a precise hierarchy represents a significant challenge both in modeling and in data processing.
At WhiteBox, we have designed a custom-tailored solution to address this challenge, implementing a deep learning model with a unique architecture that processes information from each level and uses it to inform and structure the next level. This model, built with the TensorFlow library, starts with a pre-trained base and is fine-tuned using a large dataset of keywords manually classified by JOT. Finally, we encapsulate the model in MLFlow and Docker, making it easy to implement it in the Google Cloud Platform infrastructure, where it integrates seamlessly with Google's digital marketing tools.
Our solution allows JOT to classify keywords in real time, transforming a slow and error-prone manual process into a fully automated system that classifies thousands of keywords per second. This significantly optimizes ad placement, allowing large volumes of keywords to be processed with unprecedented efficiency. In addition, our model has been shown to improve human classifications, sometimes exceeding human performance.
“By collaborating with WhiteBox, at JOT we have developed analytical solutions that allow us to integrate performance prediction services into the management of our digital marketing campaigns”
Targeted Opinion Polling & Social Listening
