Sentiment Analysis Taxonomy
Sentiment analysis, or Customer Sentiment Analysis, has become an important component for monitoring attitudes or feelings about your business, products, or reputation on the Internet. With social media vehicles such as Twitter, Facebook, product review sites, comment sections, and more, your company or product could be mentioned all over the web. It's important to understand how people feel about your company, and take action to respond if necessary.
WAND has created a taxonomy specifically for this purpose. Our new customer sentiment taxonomy has hundreds of terms including, verbs, adverbs, superlatives, adjectives, and more that all correspond to an intensity of feeling from Extremely dissatisfied, Dissatisfied, Neutral, Satisfied, and Extremely Satisfied.
On twitter, a user might mention "I went to XYZ restaurant and the service was incredible". Using a taxonomy approach, one could understand that "incredible" indicates extreme satisfaction with the restaurant. Or, imagine a product review that says "When I received ABC product, the case was cracked". This would indicate that the product was in disrepair when it was received.
It's easy to understand and comprehend the meaning of the above examples by just reading them. However, with a huge volume of social media and other commentary on the Internet, a company cannot manage to read everything posted about it or its products/services. Leveraging a taxonomy that is designed to identify different levels of sentiment can be a valuable tool for any company to keep tabs on what people are saying about it online.
Sentiment taxonomies are not just valuable in monitoring online and text sentiment, however. They can be equally powerful in helping to analyze vasts amount of customer service or other call center communication. Transcriptions of phone calls can be run against a taxonomy to tag phone calls with levels of satisfaction. This allows a manager to identify an overall level of satisfaction experienced by customer the call-reps are speaking with, but also to identify specific calls that should be listened to and perhaps followed up with.
A recorded customer service call could conceivably last 5-10 minutes. It would be inefficient to listen to the entire call to find the exact point where a customer became upset. Instead, if the customer said "Your company's product is terrible", the exact point in the call where this took place could be tagged by the sentiment taxonomy and a manager could start listening at that point in the call.
For training purposes, a manager could identify all calls where the words "not helpful" were spoken by the customer. The manager could listen to these calls and determine why the representative was not being helpful and identify possible areas of improvement.
Call transcripts or social media could also be monitored and tagged with other taxonomies, such as product names, product features, or other elements to provide the ability to drill through the monitored data from several dimensions.
Please reach out if you'd like to learn more about sentiment analysis taxonomies or how the solutions described above can be implemented.