Context is key: Language and Data Science

This post might be an interesting read if you want to know more about how Appsedia uses Natural Language Processing to analyze app reviews.

What is Natural Language Processing (NLP)? NLP is a useful set of Data Science tools and techniques that enable computer algorithms to understand and respond to human language. At Appsedia, NLP methods are at the heart of what we do — whether it’s sentiment analysis to find out how people are talking about your app online, topic modeling to find out what kinds of things they’re saying, or our Social Health Score which combines these and other data sources into a single easy-to-understand metric.

But the very power of NLP models means they can easily lead you astray if wrongly used. If trained on the wrong dataset, they can be susceptible to bias or overfitting, thinking they’ve found patterns that aren’t actually there. They don’t always understand that words can have different meanings in different contexts, or that people might communicate differently when they're on different platforms. And NLP models that perform brilliantly in the domain they were designed for can fall on their faces as soon as they’re taken out of their comfort zone and applied to a different kind of data. 

This is why Appsedia doesn’t just use out-of-the-box NLP models: we understand that context and genre matter. For example, a word like addictive can be associated with very different kinds of sentiment when used in a game review than it is in a medical context. And angry may sound like a very negative word, but you definitely wouldn’t want a model that classified every tweet about this game you may have heard of as a negative one!

To make sure our models understand this kind of nuance, we teach them domain-specific linguistic knowledge — not only at the level of app categories like games or music apps, but down to individual app content. They even understand emojis!

Sign up to Appsedia today to use our customized NLP tools for deeper insights into your apps’ performance.