Managing Customer Interactions on Twitter

In this lesson you are going to learn how to use text analytics to efficiently manage customer interactions on Twitter.

Uplus is a major telecommunications company interested in improving its customer engagement on social media. It has decided to monitor relevant Twitter data and post appropriate responses where necessary. Here’s a sample of recent tweets about Uplus.

Of course, the entire list may be much longer than this. It’s not practical for customer service representatives to go through every tweet and respond to the relevant ones in a timely manner. Uplus has automated this process. They run text analysis on the Twitter data. Based on the analysis, only certain tweets are forwarded to a customer service representative, who then reviews them and responds.

Text analysis output has many aspects. The following analysis shows how it works with tweets.

This is the result of text analysis on a customer’s remark about a billing error. Language is an output of text analysis. It has detected English for this text. Sentiment refers to the general attitude of the author towards a subject. It can be positive, negative, or neutral. The Overall sentiment refers to the sentiment analysis outcome for the entire text, which in this case is negative. The corresponding score represents the degree of polarity. As the score approaches 1 (or -1), the degree increases. The sign indicates the direction of polarity. So +1 is extremely positive, and -1 is extremely negative. When the score is closer to 0 it is considered neutral. The input text is color coded to show which parts are evaluated as positive, negative, or neutral.

Classification is the next result of text analysis.

A piece of text is classified into one or more possible categories. The list of valid categories is known as a taxonomy. This list varies across organizations and industries. You can define your own rules or use a model developed using machine learning techniques to classify text. The Classification output also contains the sentiment analysis for each of the recognized categories. For example, the sentence that corresponds to the ‘Billing/Payment > Accuracy’ category has a Negative sentiment with a score of -0.61. The Confidence score is an indication of the reliability of the categorization. The closer the score is to 1, the more reliable the categorization. For example, the system is more confident categorizing the first sentence than the last two.

The next output of text analysis is known as Entity extraction.

Entities and Topics are proper nouns found in the input text. They are names of organizations, people, places, products, services, dates and times, quantities, tracking codes, etc. You can manually define the rules for extraction and the list of topics to look for in the input text. Similar to Classification, you can also use models developed using machine learning for Entity extraction. You can also use a combination of both approaches.

Uplus decides to use only the overall Sentiment to determine if a tweet should receive a response or not. If the overall Sentiment is negative, a work item is automatically created and assigned to a customer service representative.

The representative then reviews and responds to the tweet appropriately.

A dashboard shows consolidated reports of the text analysis output. This shows the volume per sentiment analysis outcome.

This graph shows the volume by Category detected.

You have now reached the end of this lesson, which described how to use text analytics to efficiently manage customer interactions on Twitter.