Sensor Tower Blog

Everything about App Store Optimization

Introducing the Sensor Tower Intelligent Keyword Suggestions Tool

We want to introduce the Intelligent Keywords Suggestion Tool that we’ve been building and testing for the last few months. Although we’ve made understanding keywords in the App Store more transparent by unveiling the keywords of all Apps via our Keyword Spy tool, selecting good keywords still takes a lot of time and research for App Developers.

With the Intelligent Keyword Suggestion Tool you can get high quality related keyword suggestions for your current App or for the App that you’re in the process of making, giving you the option of getting more exposure and more installs through the use of smart keywords.

Before we talk about how exactly it works or how it can be customized to produce even better results, let’s see an example of it in action. We’ve taken an App with good related keywords: iWhiskey – 600 reviews of the world’s whiskey, and whisky by F. Paul Pacult and put it through the Intelligent Suggestions tool, tweaking the weight values of the keywords a bit.

According to our Keyword Spy Tool, iWhiskey has these keywords:

1
2
alcohol, whisky, world, cocktail, review, flip, bourbon, scotch, drink,
600, jack, daniels, iwhiskey, reviews, whiskey, paul, pacult

What kind of keyword suggestions can the Intelligent Keyword Suggestion tool give based on the keywords above?

iWhiskey keyword suggestions

Pretty good! The Intelligent Keyword Suggestions is giving high quality related keyword suggestions (spirit, liquor, gin, martini, etc. ). This is because we went to the advanced settings and configured the weight values of the most important words to be high.

How Does It Work?

How keyword suggestions works

By combining the data we have on keywords used by Apps in the App Store along with an n-gram based database of word relations derived from Common Crawl and Project Gutenberg data along with other sources, we’ve build a collection of word relations designed specifically for generating related keywords.

With a bit of tweaking on top of the n-gram database that makes the results more useful by removing common words and non-information carrying parts of speech such as prepositions, we’ve built a tool capable of producing highly relevant and useful keywords given a set of seed keywords.

The Intelligent Keyword Suggestions relies on a set of relevant and inter-related seed keywords to generate new keyword suggestions. These seed keywords are used to generate the single n-gram sets and the multiply-related n-gram sets. An ideal situation is a medium sized set of seed keywords (about 15-20) that are highly related.

By default, we place our best estimate of your current keywords and the words in your app title as your seed keywords. This basic setting usually produces highly relevant results, though by enabling Advanced Settings and adding extra seed keywords and/or disabling less relevant ones further increases the power and flexibility of the tool.

Disabling less relevant keywords and keywords that you intend to replace usually yields better results as well. If your keywords are highly specialized, you will generally see really good suggestions, while Apps with very generic keywords will need to further tweak their seed keywords to see relevant suggestions.

Advanced Settings

This opens up the possibility of tweaking the seed keywords used to generate the new keywords, but it also gives the possibility of assigning weight values to individual keywords. For example, we can give the words “alcohol” and whisky higher seed weight by sliding the slider to the right.

Fine tune your keyword recommendations

The general filter strength controls the relevancy of generated keywords. A higher filter strength will result in less generated keywords, but the keywords will be more closely correlated with the seed keywords.

Finally, an experimental phrase based feature based on two word seed keyword combinations is available to generate even more keywords. We pre-generate the two word seed phrases based on your current keywords, but make sure to look over them and tweak them for good and relevant results.

The phrase based keyword generation, due to the nature of being seeded with two word combinations, tends to generate results that will fill in the long-tail searches for phrases based on your current keywords. It can throw up some gems, just there is a bit more work involved with tweaking the seed phrases.

Conclusion

We’ve been testing the Intelligent Keyword Suggestion tool internally for a while now, and with some human oversight it is capable of greatly simplifying the keyword research and generation process. For a lot of Apps, it will generate a couple of very good keywords without any intervention, and for Apps in more specific niches it is capable of producing really good results after tweaking the seed keywords and weight values.

The Intelligent Keyword Suggestion tool is capable of improving keywords for almost any App — sign up for Sensor Tower and try it out.

UPDATE: For more tips on how to choose the right keywords for your app, read our tutorial on choosing the right keywords.

Alex Malafeev

Alex Malafeev

Founder at SensorTower

Comments