Lunch at 12:30pm, (virtual) talk at 1pm, in 148 Fitzpatrick

Title: New Challenges in Question Answering: (Part I) From News/Wiki to Emotional Support

Abstract: This talk will present a few new challenges on language models and question answering tasks that Jiang’s Lab is making efforts to address. Knowledge plays an essential role in answering questions from news/Wikipedia domains. Machines may be able to suggest some texts or edits of the answers if they have knowledge. Knowledge may be retrieved from entity-based texts, entity-based representations, and entity-based knowledge graphs, so Retrieval-Augmented natural language Generation (RAG) is rising as a powerful method. This talk will discuss the challenges and potential solutions, when we apply the methods for emotional support with social media data: Entity is no longer the central player, but structured knowledge is still needed. The second part of the talk (series) will be “from accuracy to diversity” of the machine-generated answers. Stay tuned to see if Prof. Jiang will be invited another time.

Bio: Dr. Jiang’s research is in mining and learning from text and graph data, the core topics of data mining and natural language processing. He believes intelligent systems need knowledge and can find knowledge for themselves. He focuses mainly on information extraction, text generation, and question answering.