Keyword Cluster Finder

This project is a proof of concept to demonstrate the programmatic computation of related keywords with respect to search engine results.

Packages

data

Manages interacting with the product database to fetch SERP data. If you don't have access to the product database I'm using, this isn't very interesting to you.

graph

Computes the Markov cluster from a graph of keywords with edges weighted by the similarity scores among them.

rankings

Logic for parsing rankings data -- either from stored JSON files or from a database -- and combining them into a SERP containing prominent results for a search as well as the keyword that yielded those results.

rbo

A Go port of a Python implementation of the rank-biased overlap algorithm designed for SERP structures defined in the rankings package.

Credit to dlukes/rbo for the original implementation.

Programs

build-from-disk

Computes keyword clusters based on SERP data stored in JSON files. It accepts a single path to a directory containing SERP data. See the program comments for the required data schema or for sample data to use.

build-from-db

Requires access and credentials to the product database. You probably don't want to try this if we don't work together.

Computes keyword clustes based on SERP data obtained from the product database.

It accepts as arguments parameters for both the RBO algorithm for computing similarity as well as the Markov cluster algorithm to determine neighbor selection.

For more information about RBO parameters, read documentation of an implementation in R or in the original paper.

For more information about Markov cluster parameters, read "Demystifying Markov Clustering".

Usage of build-from-db:
  -config string
    	app JSON config
  -domainID int
    	Domain ID
  -inf int
    	Cluster inflation (default 2)
  -p float
    	RBO p value (default 0.9)
  -pow int
    	Cluster power (default 5)