HOME ›

BLOGS ›

Common Mistakes In Text Analytics

Common-mistakes-in-text-analytics

There are common mistakes made when a data analyst performs text mining (or text analytics). Text analytics is the study of written text for information- mining purposes. This includes linguistic, statistical, and machine learning techniques that can model human language in order to extract insights from it. Text analytics is a powerful tool and can be transformational for businesses.

Not Asking Target Questions

In order to get the most out of your survey, you need an initial question. Even if it is a simple and straightforward “What do my guests think about these rooms?” or even something more complex like “How could I improve this lobby for future visitors?” Asking a targeted question will get the useful information that is desired.

Isolated Data

Last year, 25 guests complained about the bathrooms. Is this better or worse than your competition? Is this better or worse than last year? You need to make comparisons either across competitors or time (preferably both) in order for you to know if there is anything wrong and what needs redressing.

Acting Slowly On Individual Complaints

Along with getting a view of all discussions related to your brand, you can also use text analytics and sentiment analysis in order to throw an alert if someone has Tweeted that they’re having a poor experience. The first case (the tweet) is vital because oftentimes this type of feedback happens right now.

Acting Too Quickly On Broad Analysis

Don’t just do a surface-level analysis. Spend time analyzing the feedback and why it matters to you. Dig deeper if needed, but don’t forget that this is also an investment of your valuable marketing resources. Look at trends for the broad analysis. Evaluate the complaints against the competition and see if they demand immediate attention.

Assuming Clean Data

Data quality is always an issue and should not be assumed to be good quality. Even with the help of a review analysis system, it might be worth looking into the incoming data stream to make sure weird stuff is not making its way in.

Unrealistic Expectations Of The Text Analytics System

According to one university study, humans will disagree about at least one sentence out of five as to its sentiment classification. Is that statement’s mood to be seen as good, neutral, or bad? The accuracy of a text analytic system will be dependent on its fine-tuning. It is also dependent on the system’s inter-rate agreement. Resolve to have only 85% accuracy.

Not Appreciating Multiple Conversion Steps

Hand-written comments and speech-to-text conversions are not easy. If you want the most accurate data, make your own transcriptions. Use an Optical Character Recognition (OCR) system or convert from one type of file format to another with relative accuracy for each step.

Not Allocating Any Tuning Time

Time should be spent fine tuning the text mining system for details and itemized concerns, such as restaurant menu choices or the brand of furniture used in a hotel.

Thinking Too Small

After the initial question targeted question, mentioned above, there should be an expansion out of more questions. Mine the text for what people are saying about you and about your competitors.

Conclusion

Text mining is a valuable practice for any data-driven business. Text mining with the smart disciplines of asking the right questions, awareness of limitations, and fine tuning your mining habits will prove even more valuable.

Post This Article

Facebook
Twitter
LinkedIn

Related Articles