Two studies to be presented at ACL and NAACL–held on May and July 2022 respectively
Riiid, a leading AI Education solutions company and a member of Born2global Centre, today announced two of its new AI research studies that will be presented at the upcoming Association for Computational Linguistics (ACL) and the North American Chapter of the Association for Computational Linguistics (NAACL), the world’s leading Natural Language Processing (NLP) conferences. With this ground-breaking research, the company will continue to extend its AI capabilities with NLP technology to build extensive new educational content-aware AI models.
At ACL, Riiid will present ‘Dialogue Summaries as Dialogue States (DS2), Template-Guided Summarization for Few-shot Dialogue State Tracking’, focusing primarily on creating dialogue systems like Chatbots that can deliver meaningful results even with a limited amount of labeled and structured dialogue data by leveraging very large language models. By using only 1% of the training data, Riiid’s new method was able to achieve 66% of the full data performance, whereas other approaches achieved less than 60%.
“Dialogue State Tracking (DST) is an essential element of task-oriented dialogue systems but is infamous for the expensive and difficult data collection process,” said Jay Shin, AI Research Scientist at Riiid, who led the study. “Our study proposes a new method to reformulate DST into dialogue summarization, to minimize the pre-train and fine-tune discrepancies that typically occur”
Riiid researchers provided rule-based summary templates from dialogue states and guided the summarization to conform to these templates. Applying heuristic dialogue state extraction from the generated summaries, researchers were able to create the strongest DST model in the limited label scenario that uses only 1% of training data. “This study can be extended as the proof-of-concept that Riiid can develop new educational features such as AI Tutors with chatting capabilities at a much lower cost with much higher efficiency and accuracy,” said Shin.
Riiid’s second study, to be shared at NAACL as an oral presentation, introduces an efficient training algorithm ‘GRAM (GRadient Accumulation for Multimodality)’ for incorporating content information in Knowledge Tracing (KT).
“Content-based KT can potentially offer personalized item recommendations to users, but its exponential training time has hindered its usage in the industries,” said Yoonseok Yang, AI Research Scientist at Riiid, leader of the study. “In this research, we accelerated the training speed up to 150 times while maintaining state-of-the-art KT performance. In addition, GRAM is especially powerful for newly-added questions that have no interaction history, showing a dramatic accuracy improvement of 40% in cold start predictions compared to the previous models.”
GRAM will be deployed in the A/B testing platform in Santa, the company’s own English proficiency test (TOEIC) prep solution that is the best-selling AI-based smartphone application in Japan and Korea. With 300 new questions added to Santa every month, the company expects its latest model to provide new levels of personalized learning experience for users.
“The impact of this work is not limited to improving the model performance in Santa,” said Yang. “With this technology, Riiid can easily provide high-quality cross-domain KT models, which will open our doors to countless domains with small or no interaction data. Basically, we can apply TOEIC KT models directly for similar domains like GTELP, TOEFL, or even SAT without interactions data from these new domains.”
“The two studies are remarkable as academic research but also in demonstrating how our lab breakthroughs help remove some of the most common limitations of AI for education,” said YJ Jang, CEO of Riiid. “The ultimate goal of Riiid’s research is to find how technology can help to solve practical problems in education and bring more value to actual learners. We will continue to expand the field of research and change the landscape of the education industry in the direction that can best help improve everyone’s learning experience.”
Riiid has been active in publishing academic works at top global conferences since 2016. Based on its proprietary AI technology and the world’s largest data collection set in AI education, the company has published more than 16 papers in leading AI and education technology conferences such as Neural Information Processing Systems (NeurIPS) and Association for the Advancement of Artificial Intelligence (AAAI).
For more such updates and perspectives around Digital Innovation, IoT, Data Infrastructure, AI & Cybersecurity, go to AI-Techpark.com.
