It's also important to handle random requests (tails) - for many stores they drive traffic. It is important to do a probabilistic check on requests without statistics to test for finding randomly requesting sites.
In addition, the speaker considers the complexity of compiling a semantic core from the collected queries. At the first stage of work, it makes sense to search for requests from the popularity Brazil WhatsApp Number List of competitors. Marketers must compile a list of competitors, which then needs to be expanded.
Different projects have different semantic core sizes:
Clustering is performed after compiling a semantic core based on collected queries. Grouping can be done manually or using automation. Automatic clustering is implemented as follows:
An example calculation might look like this:
A stepwise clustering algorithm might look like this: we keep softening the clustering criteria; select the query sequentially from the semantics; we refer the query to an already existing cluster (the query is compatible with all or lucky words in the cluster). Next, create a new cluster.
At the same time, it is important to understand is there no point in clustering without follow-up monitoring? The monitoring tasks are as follows.