Geekly Articles each Day
📜 ⬆️ ⬇️

Comparison of keyword automatic clustering services for SEO


In a small SEO department of a large content project where I work, we decided to increase the staff. It was planned to recruit people with little or no experience. For this reason, it was decided to create a guide that would serve as a comprehensive guide to writing new articles. The manual turned out really detailed and complete, one of its important blocks is the clustering of requests.


Why do we need clustering services?


Only such requests that have a good chance of getting into the top 10 search engines with a common relevant page should be combined into one cluster. That is, if the two requests for the issuance of all pages of sites are different and there are no intersections, then they should be attributed to different clusters. And vice versa: if two requests can be promoted on one article, then you should not spread them to different clusters so as not to write too much - the budget for the content is not rubber.


The general scheme for drawing up the TOR for writing an SEO article is as follows:

  1. Collect semantics - search engine statistics, semantics bases, internal project statistics;
  2. Automatic clustering - a service or a program for clustering similar to the tops;
  3. “Postclustering” manual - processing of what could not be automatically clustered;
  4. Prioritization - determining the importance of the received requests in each cluster;
  5. Design of TK for a copywriter - lemmatization, LSI and various instructions for writing articles, on the article for each cluster.

That's it for the second item, you had to choose the most suitable automatic clustering service. For this purpose, I conducted a comparative analysis of the most famous, in my opinion, services.


Clustering Methods


Of the ways that are automated in some well-known services or programs, we can distinguish two:


Based on the task - writing SEO articles, a method was chosen in the likeness of the tops. The search engine for the traffic we are targeting is Yandex, so the top 10 Yandex was used for clustering. This method has two types:


as well as such a parameter as “link strength” - the number of common URLs in search results for queries.

According to the recommendations of the creators of clustering services, for our case, the Soft option was chosen with a communication strength of 4. This is an important point, because for an online store, for example, it would be necessary to select other options.


Comparison method


The essence of the comparison of services is as follows: select a perfectly clustered list of requests — the reference core. Compare the results of the clustering of each service with the reference.


It was important to make such a reference core well. Since we have a content project and most of the content is the questions and answers of users, there is plenty of material for collecting statistics on the project.


A core of 2500+ key phrases was taken, which has been tracked for many months. From it, only requests that are in the top 5 of Yandex are selected. And only those that have a relevant page from one of the broad sections (category of the question, topic of the question, category of the document, a page with the form “ask a question”), and not a narrow page of the question with answers are taken from them. Requests were grouped by relevant page. Left only groups in which more than 4 requests. The result was 292 queries divided into 22 clusters.


Looking ahead, I will say that the results of clustering on the Moscow issue of Yandex were compared without geo-referencing. Regional Moscow issue proved to be better, so we will continue to talk about it.


Comparison of services


In the search for the most popular services, the report of Alexander Ozhgibesov on BDD-2017 helped a lot; in addition to the fact that he had added several more services, the following list came out:
')

  1. Top viewer
  2. Pixelplus
  3. Serpstat
  4. Rush Analytics
  5. Just magic
  6. Key collector
  7. MindSerp
  8. Semparser
  9. KeyAssort
  10. coolakov.ru

The first thing that was checked was the result of the clustering of the reference core for these services of the group - whether this makes the service too broad groups. Namely, did the requests from different groups of the reference core fall into one cluster according to the service version?


But only such a comparison is not enough. Services are divided into two approaches to a nonclustered remainder of phrases:


Because of clause 2, it became necessary to look at the number of phrases that are in the same group of the reference core and fall into different services.

In comparison, I used both of these parameters in the form of a ratio - what percentage of the total number of phrases was not in my group.


The results of the comparison:


Results


As the optimal solution for our project, the KeyAssort program was chosen - this is the program, not the online service, the license is purchased once, attached to the hardware.


Quite good results were shown by the popular online service Serpstat, but for our case a little worse, and also much more expensive. If you take large volumes of requests per month and use it only for clustering, it is not profitable. Perhaps, if you use the clusterizer together with its other tools, then it will be affordable.


The worst indicators of the Key Collector program, which still does not eliminate the need to have it in your arsenal for any SEO.


I was very surprised by the MindSerp service, through whose website I could not get any feedback about the demo. If the representatives of this service read the article, maybe I will add to it the comparison)

Source: https://habr.com/ru/post/354986/

More articles:


All Articles