The end goal of any Text Analytics exercise is to extract insights from structured or unstructured text. In general, text analytics is focused on 3 main aspects.
1) Sentiment – Or more specifically – capturing the emotion shown in the text. This is done via an automated machine learning algorithm. TalkAboutIT uses the Sentiment algorithm of Google, which is the recognized top performer both during internal as well as external tests. Our own algorithms build upon it in order to capture industry- or context-specific sentiment.
2) Categories/Topics – Or what the person is talking about. If a telco customer is complaining about the speed of their broadband at home, this is something which the telco would like to know. We are extracting this information with a mix between human and automatic coding – automated coding alone is accurate, but not actionable. In our example automated categorization might lump together Mobile internet speed and Home Broadband speed in the topic “internet speed” whereas these are two distinct issues the telco would like to deal with separately. This is why a Text Analyst creates a set of comprehensive Boolean rules that are designed to identify specific themes and make them actionable.
The rules are tested on a small Training dataset with 20-30,000 comments, and once the Code Frame has been finalized, the rules are applied to the full set of comments (hundreds of thousands).
3) Metadata – This software incorporates additional data that has been supplied together with the source text into its dashboards. For example, we can combine the existing NPS score of each comment that is entered into the system with the Sentiment and Category extracted and visualise them on a single Dashboard.