Luminoso Technologies: Programming the Enterprise to Adapt to the “Human” Way of Life
Artificial Intelligence or AI presents several prospects which scientists and researchers have been trying to harness and bring to reality for decades. While AI is being applied in several verticals today, most of its applications are still in the prototype phase. However one of the most major advancements in this territory of science was developed about a decade ago at MIT Media Labs. The advanced machine learning technology that was created therein dubbed ConceptNet later became the premise for the inception of Luminoso Technologies. From the beginning ConceptNet’s development was fueled by the desire to replicate how humans learn and understand language and leverage the information gained thereby. Today, Luminoso can process this growing volume of unstructured customer feedback data clients are collecting and enable them to analyze it at scale. “These insights become the foundation of knowledge that clients use to make decisions on how to address these customer opportunities to improve their experience,” explains Catherine Havasi, CEO, Luminoso.
The science behind Luminoso’s technology is intricate and incorporates cutting-edge research on artificial intelligence, word embeddings, and machine learning. On the back-end, once textual data is uploaded into Luminoso’s products, the software stems and parses the text before applying machine learning and natural language processing to identify the most relevant terms, vectorizes each term using a combination of semantic knowledge bases and word embeddings (including, but not limited to, MIT’s ConceptNet and Stanford’s GloVe), and maps out those vectors into a multidimensional space.
“While our solution is complex, our priority has been to ensure that the software is approachable, comprehensible, and – above all – usable for our clients,” says Catherine. On the front end, text data can be uploaded from a spreadsheet or another source. Luminoso’s software applies machine learning and automatically sets appropriate parameters; the user does not see any of this. Once the project is ready (typically fewer than 5 minutes later), users see an interactive visualization of the data that can be explored by simply clicking on specific terms. Luminoso’s software includes built-in dashboards to make it easy for key findings to be displayed and shared across the user’s organization.
Luminoso technology automatically processes text data faster than other resources – 100,000+ messages in 10 minutes as opposed to weeks – saving companies time and labor in getting insights.
Using these AI-based techniques means that it takes less time, is less biased, and is thus more scalable than historical approaches. Using machine learning to replace analyst-built ontologies and dictionaries means that data can be processed in minutes rather than weeks. It also removes the bias inherent in ontologies created by people, who come to the table with preconceived hypotheses and biases.
Finally, Luminoso is scalable to any size dataset (whether it contains 100 or 1,000,000+ verbatims) and can process unstructured data from any source, including open-ended survey responses and call center transcripts. These improvements are critical for companies who need to be able to integrate data, gather insights, and respond quickly in order to remain competitive in the marketplace. A fine example of this would be the firm’s projects with one of its clients, a software provider, which transitioned to a new online checkout system vendor and quickly uncovered a major issue. By quickly finding the issue and acting on it, which Luminoso enabled it to do the company saved $11.5 million (USD).
Luminoso is deeply involved in NLP research. In addition to keeping track of and incorporating research being done in this field, its own researchers continue to conduct experiments and push the boundaries of NLP. Any new findings, once tested and vetted, are incorporated into Luminoso’s product suite. As an example, the semantic knowledge bases used to enable domain-specific learning and term embedding were upgraded this year to reflect new research conducted by Luminoso’s R&D team. Luminoso’s researchers developed and tested an ensemble of four different knowledge bases. This new representation is the most accurate in the world on many standard evaluations on word relatedness, and automatically interprets the meaning of rare words at a 16% higher level of accuracy than the previous industry standard.
Luminoso continues to make it's language understand more human-like and adaptable by delving more deeply into the complexities inherent in language - both to continue to help companies understand their customers. Especially as people are becoming more comfortable with interacting through chat-bots the organization seeks to help make conversational agents life-like and more capable of handling human spontaneity. “For Luminoso, global reach is very important and we'll continue to add languages to our platform and increase our multilingual capability,” says Catherine.