Sorry for the late delivery (Had to much workload on other stuff) - The API is now ready. 5 Requests / day free. I've added a paid plan(0.01 USD per request - I think i cant go cheaper on rapidapi), just for servercost compensation.
@Sven Maybe it would be an advantage to implement a generic rapidapi.com integration and the user can add specific rapidapi.com apis themself, because there are many good api for SEO available on rapidapi.com.
Thanks for adding the option @Sven. How does it work if I have a list of 100 words for example, I have to enter them one by one in the software first? It may be long (import then deletion of the words already present then) I should have better explained my thought, it's my fault.
I was thinking of importing a list of words or copied / pasted in a window for example.
Maybe you can import them as before (but in bulk) then test them. Or do not import them into the software but just into the window and get the result back.
I mean that it could be a tool that doesn't necessarily disrupt the list of keywords already sorted and inserted, unless the user wants it.
Or I didn't test well the option ^^
EDIT: It could be added/modified here without much change, maybe add checkboxes?
@TOPtActics thats no problem...but how to integrate it to GSA Keyword Research? I mean what should the user do with it. Select keywords, apply the script and get back what!?
create good content silos automatically (assign page content or generated content from GSE CG to existing taxonomies, categories or tags automatically if scores > threshold (0.5 default) for that label)
automatically generate a good interlinking structure on the hole site (e.g. extract keywords from all sites of a specific domain and use a set of all collected keywords as labels and then interlink keywords (in-content) to the site with the highest categorization score if score is > threshold (0.5 default))
detect spammy content (generated from autocontent generators or on competitors sites) if you chose your labellabels right (drugs, spam, sex, crime, etc...)
...
This is a very flexible API, because u can chose your labels on your own and this are my first thoughts what a SEO could do with it (This cases are better integrated in GSA Content Generator than GSA Keyword Research, but we discussed it here so far). I think you guys will find better use cases.
A more sophisticated way to integrate this in GSA Keyword Research would be to:
Extract all Keywords from a competitor page by page (x levels deep)
Generate a keyword set out of all pages
categorize every content on that keyword set
for every page look at the percentage of coverage this page has internal links to pages scored higher than x% (>30%?) on the categorizer labels.
E.g. a page (we call it focuspage now) has categoriesed as drills and bits with +30% score and we have 2 pages on that site with keywords bits and 3 pages on that site with drills as keywords extracted (you can also use the categorizer scores here instead of extracted keywords) and the focuspage links to 4 out of this 5 pages --> 80% coverage. Now calulate this for ever page and average and you have a good score for interlinking quality.
I tested the example sentence of their API: "After living abroad for such a long time, seeing my family was the best present I could have ever wished for." Thats what they detect:
If you ask me i think content categorizer (not fine tuned on this task) beats their specialized emotion analyser out of the box, because suprise is clearly misclassified in their model.
@TOPtActics - Could your API be used to improve search intent classification further?
For example, could it sort keywords into the following categories?
Research – looking for information Answer – looking for a quick answer Transactional – potentially looking to make a purchase Local – a locally relevant query Visual – looking for inspiration or ideas Video – tutorials or video-based guides News/Fresh – queries relating to a trending topic Branded – queries contain brand relevant terms
well this is indeed more something for Content Generator and create texts based on certain emotions or categories. I will keep an eye on it and add support unless someone thinks it has it's place for KeywordResearch.
@TOPtActics - Could your API be used to improve search intent classification further?
For example, could it sort keywords into the following categories?
Research – looking for information Answer – looking for a quick answer Transactional – potentially looking to make a purchase Local – a locally relevant query Visual – looking for inspiration or ideas Video – tutorials or video-based guides News/Fresh – queries relating to a trending topic Branded – queries contain brand relevant terms
I think the performance is OK for not fine tuned on that task and for a quick usable model. I the long run i can build a custom model for that use case if needed, because this seems to be a pretty important topic.
Maybe with good "tuned" labels we can get better performance here:
I added "buy" as a alternative for transactional, because maybe "transactional" is not as clear to the model as "buy" and in this case the sample is better classified. As a quick and dirty solution you could come up with more than one descriptiv word on each intent and average them out in an additional step.
The model is case sensititive, this could be tested too. I have also some ideas for a quick adjustments in the code to get the model running better on that specific task. I can test this adjustments this week and check if it works even better. If so, i would add another parameter to the API where someone can specify 'intent' to tell the model that an intent prediction is the task.
You can even categories the intent of pages with analysing the description or the hole text corpus:
https://www.nike.com/de/ --> Description: 'Inspiring the world's athletes, Nike delivers innovative products, experiences and services.'
I think visual is high because of the "Anthrazit Stoff" description. Some additional keywords for the intent "visual" that describes more what a SEO understand under visual would fix this.
@z3r exporting the domains by right click->copy. Do you really want to export things via file? Anyway, I have added export for next update now.
There is a bug when you try to export, after selecting the names and it ask what to export if you press cancel it crashes. Is possible to add export only available domains with full/select option? Thanks
@TOPtActics - It seems like it works pretty good for intent. The software already handles KW intent, but it's not separated into as many categories as I listed above. Categorizing content pages is also a neat idea. Are you working on any other APIs?
I was trying to throw out an idea for the API, but I would not spend a lot of time improving the model for KW intent unless @Sven confirms he wants to integrate it.
Comments
Content Categorizer:
https://rapidapi.com/Optimalize/api/content-categorizer/endpoints
Or I didn't test well the option ^^
EDIT: It could be added/modified here without much change, maybe add checkboxes?
- rapidapi-key
- rapidapi-host
- rapidapi-url
- a mapping from required parameters to variables in your software
- request method (POST, GET)
and code a standard method for requesting for POST and GET.For the response the user could build a own script.ini to process.
Or do everything in one go with one *.ini file per rapidapi. The User has to build the *.ini file like in SER for custom sites.
- create good content silos automatically (assign page content or generated content from GSE CG to existing taxonomies, categories or tags automatically if scores > threshold (0.5 default) for that label)
- automatically generate a good interlinking structure on the hole site (e.g. extract keywords from all sites of a specific domain and use a set of all collected keywords as labels and then interlink keywords (in-content) to the site with the highest categorization score if score is > threshold (0.5 default))
- detect spammy content (generated from autocontent generators or on competitors sites) if you chose your
- ...
This is a very flexible API, because u can chose your labels on your own and this are my first thoughts what a SEO could do with it (This cases are better integrated in GSA Content Generator than GSA Keyword Research, but we discussed it here so far). I think you guys will find better use cases.labellabels right (drugs, spam, sex, crime, etc...)
E.g. a page (we call it focuspage now) has categoriesed as drills and bits with +30% score and we have 2 pages on that site with keywords bits and 3 pages on that site with drills as keywords extracted (you can also use the categorizer scores here instead of extracted keywords) and the focuspage links to 4 out of this 5 pages --> 80% coverage. Now calulate this for ever page and average and you have a good score for interlinking quality.
I tested the example sentence of their API: "After living abroad for such a long time, seeing my family was the best present I could have ever wished for." Thats what they detect:
I pasted the categories in the content categorizer and some more from (https://simple.wikipedia.org/wiki/List_of_emotions) and thats what we got
If you ask me i think content categorizer (not fine tuned on this task) beats their specialized emotion analyser out of the box, because suprise is clearly misclassified in their model.
Answer – looking for a quick answer
Transactional – potentially looking to make a purchase
Local – a locally relevant query
Visual – looking for inspiration or ideas
Video – tutorials or video-based guides
News/Fresh – queries relating to a trending topic
Branded – queries contain brand relevant terms
I added "buy" as a alternative for transactional, because maybe "transactional" is not as clear to the model as "buy" and in this case the sample is better classified. As a quick and dirty solution you could come up with more than one descriptiv word on each intent and average them out in an additional step.
The model is case sensititive, this could be tested too. I have also some ideas for a quick adjustments in the code to get the model running better on that specific task. I can test this adjustments this week and check if it works even better. If so, i would add another parameter to the API where someone can specify 'intent' to tell the model that an intent prediction is the task.
You can even categories the intent of pages with analysing the description or the hole text corpus:
https://www.nike.com/de/ --> Description: 'Inspiring the world's athletes, Nike delivers innovative products, experiences and services.'
German Amazon Dot Echo (https://www.amazon.de/Echo-Dot-3-Gen-Intelligenter-Lautsprecher-mit-Alexa-Anthrazit-Stoff/dp/B07PHPXHQS) --> Description: 'Echo Dot (3. Gen.) Intelligenter Lautsprecher mit Alexa, Anthrazit Stoff: Amazon.de: Alle Produkte'
I think visual is high because of the "Anthrazit Stoff" description. Some additional keywords for the intent "visual" that describes more what a SEO understand under visual would fix this.
I was trying to throw out an idea for the API, but I would not spend a lot of time improving the model for KW intent unless @Sven confirms he wants to integrate it.
It's great.