Adding more keywords into your SEM campaigns should be a routine if you want to grow and optimise your campaign. It's part of the evolution where new keywords need to be added in while bad ones get removed.
Finding keywords to expand into can be fun. There are many ways to do it. We can either dive into the search query report, do another round of deliberate expansion using AdWords Keyword Planner or Ubersuggest, or we can use my favourite method (post to come in the future).
After generating a new list of keywords, let's say you have 15,000 of them. You currently have 50 ad groups for each market in a regional account. How do you channel or classify those new keywords into the existing ad groups?
That is where the Keyword Grouping Tool comes into play and makes your job extremely easy.
The vectorisation, latent semantic analysis and singular value decomposition are similar. However, the difference comes after that.
Keyword Clustering Tool, as the name implies, uses a clustering methodology that is one of the unsupervised learning methods of machine learning.
While for the Keyword Grouping Tool, it's a classification method that is part of the supervised learning methods of machine learning.
The tool uses multiple algorithms to output 4 decisions based on 4 different classification algorithms.
Moving forward, you can either
Note that the quality of the grouping depends on the quality of your training set. If your ad groups are already structured badly, the new keywords will be placed wrongly as well.
To structure your ad groups correctly, I recommend you use the Keyword Clustering Tool to restructure your ad groups before using this tool. From my experience, the Keyword Grouping Tool's accuracy improves to about 80-90% accuracy range when it learns from a structure created by the Keyword Clustering Tool.
Those who have learnt about machine learning will know about hyperparameter tuning.
Currently, the tool doesn't tune its hyperparameters to your dataset. It's currently tuned for a general-purpose dataset.
That is currently in the pipeline, but hyperparameter tuning is expensive and resource-intensive as I would probably need to use Amazon Web Services, Google Cloud Platform or MicrosoftAzure to power that.
If you would like a bespoke solution to your dataset, do contact me via Twitter and we can discuss from there.
"Keyword Grouping Tool for Search Engine Marketing (SEM)"
Python, Machine Learning, Natural Language Processing