LG Ranks First in Korean AI Evaluation... Plans to Apply Customer Service Chatbot

CoQuad 1.0 Machine Reading Comprehension Evaluation Score 95.39
Answers Through AI Repetitive Learning to Understand Context

Source: LG Electronics Official Blog

[Asia Economy Reporter Dongwoo Lee] LG has ranked first in the Korean language artificial intelligence (AI) machine reading comprehension evaluation. LG plans to apply this technology to customer consultation chatbots.

On the 19th, LG announced that the AI program developed by LG Science Park scored 95.39 points in the AI training Korean standard dataset ‘KoQuAD1.0’ machine reading comprehension evaluation, ranking first. This result surpasses the score (91.2 points) obtained by humans solving the same reading comprehension problems.

Machine reading comprehension is a technology where AI repeatedly learns ‘questioning and answering’ to analyze problems independently and find optimized answers to questions.

For example, given a passage such as “Jeju Island is an island located in the southern sea of the Republic of Korea and is the largest island in Korea. Administratively, it belongs to Jeju Special Self-Governing Province. The island’s area is 1,833.2㎢, which corresponds to 1.83% of South Korea’s area. The resident registration population in 2020 was about 670,000, making it the most populous island in Korea,” and the question “How many people live on Jeju Island?”, the AI understands the context and answers “670,000.”

The reading comprehension evaluation is conducted by having the AI answer thousands of unstructured subjective questions over about 10 minutes, with scores calculated based on the precision and recall of the answers.

LG plans to apply this technology to customer consultation chatbots. Unlike existing chatbots that respond by selectively choosing predetermined answers according to question types, the advanced chatbot service understands the meaning of questions and can provide natural answers as if a human were consulting.

With AI technology, it can quickly and accurately understand vast amounts of long and complex documents such as Korean product manuals and web pages, and find the necessary key information in a short time.

The LG Science Park AI Promotion Team, which achieved this result, also ranked first last month in the ‘Continual Learning Technology Competition’ held by the world’s most prestigious computer vision conference, ‘2020 CVPR.’

Continual learning is a technology that improves accuracy by learning new tasks without forgetting previously learned tasks and is considered one of the challenges in the AI field. LG Science Park participated as a joint research team with the University of Toronto and won in this field, beating a total of 79 teams including Amazon, the Chinese Academy of Sciences, and the University of Tokyo.

The LG Science Park AI Promotion Team is a group-level dedicated AI organization that establishes AI strategies and conducts advanced technology research.

Baek Gyeong-hoon, head of the LG Science Park AI Promotion Team, said, “Through the consecutive first-place achievements in credible domestic and international AI evaluations, we have proven LG’s outstanding AI technology. We will increase research and development investment and actively promote bold open innovation and talent recruitment to secure global competitiveness in the AI field.”

This content was produced with the assistance of AI translation services.