Adthena seeks out millions of search terms each day. Each term is examined multiple times to provide an accurate view of the auction landscape it belongs to. We use this observed data to estimate the click-through-rate (CTR) and cost-per-click (CPC) for every competitor we have seen bidding on a term.
Because we have significant pools of data on thousands of competitors, collected over many years, we have an excellent baseline for calculating the CTR and CPC for each search term. These figures are optimised for each competitor by a combination of supervised and unsupervised machine learning. Our models use observations from the search results pages, such as the number of adverts competing for a term, the position of each advert and components on competing pages, as well as features that are derived through statistical natural language processing.
What elements of Adthena are Machine Learned?
- Unsupervised Learning: dense vector embeddings of search terms. This essentially allows similar words to be ‘close’ to each other, e.g. ‘king’ and ‘queen’.
- NLP, Neural Networks (1): search term categorisation. This supervised learning algorithm maps search terms to one of up to 1500 categories, e.g: “cricket nets” = Sports & Fitness → Sporting Goods → Cricket Equipment
- NLP, Neural Networks (2): finding Whole Market search terms that are meaningful to clients.
- Supervised Learning: using search term categorisation within the regression model to predict CTR/CPC from the indexed data.
- Online Learning: we use online learning to handle large amounts of indexed and search term categorisation data.
Each value is re-estimated daily to keep it up to date with our newly learned knowledge of each continually changing individual auction.
Automatically Mapping to Google’s 8K Search Term Categories:
Using an extensive in-house data set and the latest statistical natural language processing techniques, we have constructed a supervised machine learning model to accurately categorise search terms. The model provides a high level of granularity and can accurately assign search terms to one of thousands of possible categories. This model allows us to attach valuable meta-data to our extensive set of indexed data, greatly enhancing our ability to provide insights from it. Adthena uses its own neural network model, built using Google’s new TensorFlow open source machine learning library, with online learning handling the large amounts of data.
How does the Adthena "automatic categorisation of search terms" work?
Classification is a standard strain of Artificial Intelligence. For instance, if you follow this link you will see Course 3, "Machine Learning Classification".
AI classification is used wherever you want to assign a label or group to, for example, some text. You might want to determine positive or negative sentiments,, or perhaps predict that a loan is risky or safe for a bank.
We have written our own classification AI that can classify words or phrases into industries - and this sets Adthena apart from competing solutions. This powerful AI classification system has nothing to do with Google keyword planner, and should not be associated with that.
This is how the AI works:
If I asked you how much a pint of milk costs in Newcastle, you would be able to make a fairly accurate guess – even if you have never been there. You can do it because you have learned many things that influence the price of milk, from knowing which cities are more expensive or cheaper to live in and what average incomes are like in different parts of the country.
In the same way, machine learning develops an understanding of what influences costs and clicks, what relationships exist between search terms and brands, along with many other insights and comparisons across billions of rows of data.
Note: Adthena only searches the first page of the SERP (where the action is!)