Luminoso, the company that turns unstructured text data into business-critical insights, announced the next generation of its proprietary natural language modeling system, QuickLearn 2.0, which is the first commercially-available solution that reduces biases in AI-powered text analysis.
“QuickLearn 2.0 is a significant advancement in our method of transfer learning,” said Robyn Speer, chief science officer at Luminoso. “More so than any other natural language understanding system, it can solve the difficult problem of how to learn about a new domain quickly, without bias, and without the need for huge amounts of training data.”
First introduced in 2016, QuickLearn® automatically learns domain-specific terminology without any training, setup, or ontology-building — enabling organizations to quickly discover insights from text-based data such as surveys, product reviews, and call center transcripts.
Also read: Does Google Care for User Privacy?
QuickLearn 2.0 represents a significant evolution of Luminoso’s natural language modeling system, with advancements including:
- Reducing bias: QuickLearn’s background space is based on a combination of ConceptNet and word embeddings learned from the text on the Web, which provides a general understanding of a large vocabulary of words. Some vocabulary learned from the Web may display biases in words referring to groups of people, a reflection of the discrimination that occurs in the real world. Now, when QuickLearn 2.0 analyzes a dataset, it includes a step that identifies and counteracts biased language to mitigate the effect of bias on the project. Reducing bias in an analysis of customer feedback lets companies uncover more objective insights to help drive their businesses forward.
- Enhancing conceptual matches: By expanding QuickLearn’s background space to take advantage of new advancements in natural language understanding (NLU) research, and doubling the dimensions of relationships across concepts, conceptual matches are now more intuitive and of higher quality. Improved conceptual matches help users identify the true impact of a conversation topic and are particularly beneficial for domain-specific language, like acronyms and abbreviations across different industries.
Also read: Gen Z Consumers Want Values to be Followed
- Improving understanding across languages: QuickLearn 2.0 includes updates to non-English language background knowledge, which strengthens conceptual matches in all 15 languages. This is especially relevant for global companies that receive customer feedback in multiple languages.
QuickLearn 2.0 will power both Luminoso Daylight and Express for Luminoso Daylight, applications for analyzing conversational text, including support tickets, open-ended survey responses, and product reviews. QuickLearn 2.0 will also power Luminoso Compass, a platform for processing, categorizing, and tagging streaming text data like call and chat transcripts.
Also read: IoT to Revolutionize B2B Marketing