MEETINGSENTIMENTNLPMLSentiment Analysis - What is it? And its use cases.

John Jacob

John Jacob

November 16, 2022·4 min read

What is Sentiment Analysis?

Sentiment Analysis is the task of using Machine Learning (ML) Algorithms to detect sentiments of spoken text by identifying, extracting, and classifying the texts into sentiments they express. Most Sentiment Analysis systems categorize text into positive, negative or neutral sentiments.

In Natural Language Processing (NLP), there are similar concepts such as Intention Analysis and Emotion Detection. Intention Analysis is the process of identifying the intent of a conversation. For Example: Feedback or a complaint registered by a participant in a meeting.

Emotion Detection is the process of detecting variation in a speaker’s pitch, the speed and volume of which they speak, can help determine the speaker’s emotion, and segregate it into categories such as happy, satisfied, angry etc. For Example: In a recorded meet, if a speaker says “This product is so cool, and the best I’ve used so far!” - The corresponding Emotion detected would be ‘Happy’ or ‘Thrilled’. But if a speaker says ”This product is buggy, unresponsive, and hard to use.” - Then, the corresponding Emotion detected would be ‘dissatisfied’ or ‘angry’

How does it work?

Sentiment Analysis is typically achieved by fine-tuning a transformer based architecture model on a ton of conversational data. In order to calculate the overall sentiment of a meeting, the sentiment of each or specific paragraphs in the conversation are considered.

A confidence score is used to represent the likelihood that the text belongs to a certain class (positive, negative, or neutral).

Example output from a Sentiment Analysis engine run on a single input

• label: POSITIVE, with score: 0.9998
• label: NEGATIVE, with score: 0.5309
• label: NEUTRAL, with score: 0.0102

Here, we see that the POSITIVE sentiment is very highly scored, and NEGATIVE very low, and this reflects that the engine is confident that the sentiment of the text is POSITIVE

Given the scores of an output being 50% positive, 5% negative, and 45% neutral. we can assign the text a positive sentiment, but we can store the additional class scores, so we can know that this is a mildly positive sentiment and likely even neutral.

What are the Use Cases of Sentiment Analysis?

Sentiment Analysis is an impressive tool used by teams all around the world to extract sentiment from conversations. A few use cases are:

Customer Service
Sentiment Analysis help businesses understand their customers. Teams can track customer sentiment and feedback on their product meetings to help with customer service, and analyze customer behavior. Generate sentiment score to understand customer emotions. For Ex: Find out whether the customers are optimistic or pessimistic about their future.

Call Center
With Emotion detection technology, call center agents can enhance the customer’s call center experience by understanding the sentiment behind the customer’s conversation.

Medical Service
By detecting sentiment, Doctors can monitor and track a patient’s emotion from their medical conversations.

Marketing Campaigns
Sentiment analysis can help you understand your market better by extracting valuable information from your recorded meetings that benefit your marketing campaigns.

Live Sentiment during Virtual Meetings
Use live sentiment generation in virtual meetings to understand your customers emotions during the call.

Product Launch
Help you monitor and analyze customer reviews at product events such as a product launch.

Research Projects
Use sentiment to analyze and understand how well the research is progressing in projects.

Employee Feedback
Understand employee emotion via accessing employee feedback during meetings.

What are the Limitations?

People always talk in different shades of meaning. It’s not always easy to understand what was exactly implied by them in their conversation.

There are a few limitations, which are:

Sarcasm
Sarcasm basically conveys the opposite of what is being said. It is used as an irony to mock people, which is hard for sentiment analysis to detect. The problem lies with understanding the truth behind the context of a sarcastic remark. Furthermore, these engines typically work on textual data, which don't characterize the tone of the speaker. With increased training data, and supplying additional context to the model it is possible to reduce mislabeling sentiments and reduce such errors.

Access Sentiment via Exemplary AI’s API

Use Exemplary AI’s simple API to easily transcribe your conversations, capture and analyze the sentiments by use of WebSockets. Steps:

• Log into Exemplary AI.
• Generate an AUTH_TOKEN (authentication token key).
• Access Playground.
• Upload a file of your choice to transcribe.
• Select a Speech-to-text provider and a preferred Language.
• Select Sentiment under Transcript NLP/NLU;
  Choose from these sentiment providers:
  • Google
  • AWS Comprehend
• Execute and run the API Call.
• Access Sentiment via Exemplary AI's Transcript Editor.

Conclusion

Sentiment analysis is helpful in certain domains and use-cases - in product feedback meetings to go to critical moments, to help improve your customer experience or analyzing customer service or success calls and emails.

With Exemplary AI’s approach to managing sentiments analysis engines, and standardizing their responses regardless of the provider, you can use the cutting-edge ML techniques out there, to seamlessly extract knowledge based on your needs.

Check out our other blogs for more interesting content.