There are two types of text analytics on the platform
- Sentiment Analysis using Natural Langauge Processing (NLP)
- Text Categorization using Machine Learning (ML)
1. Sentiment Analysis
What is Sentiment Analysis and how does it work?
Sentiment analysis is the interpretation and classification of emotions (positive, negative and neutral) within text data using text analysis techniques. Sentiment analysis allows you to identify customer sentiment toward products, brands or services from direct customer feedback.
Sentiment analysis can be enabled per verbatim/feedback question and uses Natural Language Processing (NLP) to score and assign sentiment to text.
The sentiment score goes from -1 (Negative) to + 1 (Positive) and with 0 being neutral. It's possible to have a decimal fraction score i.e. +0.4.
Enabling Sentiment Analysis
Click the verbatim feedback question.
- Toggle NLP to be ON
- Save Changes.
- Remember to Save and Next to save all changes.
Once enabled text collected by the verbatim question will be processed in realtime and the charts will become available when sufficient data is present. Sentiment doesn't need any model to be created - once enabled it starts working.
Our NLP implementation currently supports two data extraction/analysis categories:
- Keyword (Returns important keywords in the content)
- Entities (e.g. people, places, things, an organisation in the content)
Both of these are enabled when you enabled NLP.
NLP will analyse a piece of feedback from a customer and automatically determine keywords or entities and assign a sentiment and score. A piece of feedback may have several keywords/entities. NLP assigns the score based on analysing the whole feedback and the context of the keyword. You can choose to chart either Keyword or Entities in the Insights Manager.
Keywords/Entities can have Positive and Negative sentiment. If this occurs we show the keyword to have a red and green bar with scores to present the Positive and Negative sentiment.
Frequency
When calculating and presenting the overall score we look at the frequency of the Keyword/Entity.
Example
For example, from a sample of 10 comments
- 3 customers commented on 'efficiency' in a positive way and the NLP system assigned scores of +0.55, +0.87 and +0.53 to the keyword the overall positive score would be +1.94. We would represent this as a green bar and a score of 1.95.
- 2 of the comments on 'efficiency' in a negative way and the NLP system assigned scores of -0.4 and -0.3 the overall negative score would be -0.7 we would represent this as a red bar and a score of -0.7.
The system also adds a positive and negative score together to give the summed score. The chart displays the top 45 keywords/entities starting with the highest summed score first.
Example Keyword Sentiment Chart
2. Machine Learning
How can machine learning be applied to the verbatim response data?
Before we enable a Machine Learning model we need to build a Classifier model. The classifier model is built in conjunction with the client. Categories are established that best meet the requirements of the client and feedback is then categorized (manually) to create a training data set (usually in an Excel file). An example of categories would be:
- People
- Price
- Product
- Technology
It is recommended to keep the number of classifications low (<10) in order to reduce the work in building a training dataset.
Once built, the training dataset is uploaded to the platform and tested for accuracy. If the accuracy is high enough then the Machine Learning Model is enabled as a Trained Classifier Model. All new feedback text will pass through the classifier model and be classified. The classification step involves a statistical model.
A piece of text may have several classifications but the Classifier model will choose the most likely i.e. the most certain based on the previous classification.
The ML classification chart shows the count of the classification.
An important difference between NLP and ML is that with ML new data and classification are added to the ML model - i.e. it is continuously improving with new data.
An example chart for ML Classification
The ML chart can be viewed as a trend chart over time.