TMCnet News

AI Application Startup mycelebs Develops Revolutionary Taste-based AI Voice Service Powered by Its Keytalk Technology Surpassing Old Keyword-based Approaches
[July 09, 2018]

AI Application Startup mycelebs Develops Revolutionary Taste-based AI Voice Service Powered by Its Keytalk Technology Surpassing Old Keyword-based Approaches


mycelebs, Korean AI application startup, is making waves for releasing 'Malhae', a new voice-based service that brings a new approach to the smart speaker and virtual assistant market. 'Malhae' means 'Just Say It' in Korean.

"Existing voice services are designed on a scenario basis, requiring users to speak with perfect grammar and to speak according to preset, preprogrammed scenarios and dialogues to be understood," says mycelebs CEO Estelle Shin, "however, with mycelebs' Malhae, you get recommendations tailored to your tastes regardless of whether you speak in complete sentences, curse or use slang."

New AI-based data classification system, Dynamic Ontology (News - Alert)

mycelebs designed a new AI-based cognitive classification system, Dynamic Ontology, that uses machine learning technologies, such as color analysis, image recognition and other high-level machine/deep learning technologies along with natural language understanding technologies to classify data by people's tastes, preferences, sentiments, context (time, place and occasion), and categorical characteristics, such as clor aesthetics, texture and material. Dynamic Ontology makes it possible to analyze and derive insights from lifelog data found online, on platforms such as social media, and classify those insights in terms of natural language that can be easily understood. This new technology is used to create Keytalks, the next generation in keyword tagging that forms the basis of Malhae's voice recognition system.



From Keywords to Keytalks

Users can speak naturally, the same way they would in conversation, when using Malhae thanks to Keytalks, which understand the core intent and context of user utterances and provide results based on user tastes. A Keytalk is the term used to classify basic units of search information from voice-based interface services. While expressed in a similar form to keywords, Keytalks are different in that they do not just match words to the documents, but include a structure consisting of closeness and similarity between words and expressions, as well as attribute data such as color aesthetics, texture and material.


Malhae has been taking the Korean market by storm, surpassing 1 million downloads in less than 2 months while receiving top ratings from users.

"Malhae reflects mycelebs' core philosophy of using data science to curate people's tastes, help them discover tastes they never knew they had, and promote communication between people with similar tastes," says Ms. Shin, "Malhae is the first service to overcome the limitations of existing voice-based services by using our revolutionary new approach, and we expect to continue to grow into a service that provides information that matches people's tastes and circumstances."

About mycelebs, Inc.

mycelebs, Inc. (www.mycelebs.com) is an AI application group that uses AI technologies to develop services that can be immediately adapted to your business. mycelebs created the Big Data Studio, an all-in-one tool for automating data collection/processing, visualization, machine learning and real-time updates, and has worked with top Korean businesses in sectors such as hotels, tv/streaming, restaurants and recipes, as well as others to help them undergo an AI transformation and boost business competitiveness. Recently, mycelebs has proposed a new paradigm in portal services with the release of Malhae ('Just Say It' in Korean), a voice-based taste portal that conveniently provides taste-based recommendations across over 40 categories. mycelebs is preparing to enter the US market in the 2nd half of 2018 and is looking for partners that are looking to bring AI technologies and voice-based search to their businesses. For partnership inquiries please contact [email protected]


[ Back To TMCnet.com's Homepage ]