Artigo versão em Português

Birdie’s platform has multiple proprietary and third-party AI models - both generative and discriminative - that are used to process and analyze content.

We use Natural Language Understanding (NLU), a complex combination of algorithms and artificial intelligence backed by large libraries of information, to understand human language.

Rather than relying on computer language syntax, our NLU models enable comprehension of different layers of expression in natural language text, such as topics, root causes, sentiment, intention, and intensity.

Here are the key models and aspects of our AI:

Conversations are summarized to have a concise text representing the essential content being exchanged, usually involving one customer and someone from your company. The summaries help users to quickly understand what is going on in each conversation and “set the stage” for the other AI algorithms to run and analyze the data.

Not all ingested text has useful content. The signal classification aims to filter out the noise from the signal, distinguishing useless content from the one that has more context or detail. This makes it easier for Birdie's users to focus on what really matters.

Birdie extracts the most relevant topics from each conversation/feedback. A topic is usually a product-related subject mentioned in the feedback that has enough relevance to be extracted. Very similar keywords are grouped together to form a topic (topic clustering) - and Birdie’s users can merge or delete topics to make them even more relevant.

Birdie classifies the sentiment of a given topic in the context of a conversation/feedback. A sentiment can be Positive, Negative, or Neutral.

Similarly to sentiment classification, the intention is also related to a given topic in the context of a conversation/feedback. The possible values of an intention today are Issue, Compliment, Request, and Question.