Real-world Conversational AI: Context is King!
It is a human's tendency to express their intent through a very fluid and potentially ambiguous multiturn dialogue, taking most of the context for granted, rather than a sequence of fully specified commands. Humans communicate their needs to a Conversational AI in a similar manner. The attention-based language models have proved that to understand the correct context of a sentence it is important to focus on certain parts of a sequence and the relationship between them. Drawing parallels from the same we have learned that to perform accurate actions at any given time we have to learn the relations between the current user query and some valuable parts from the previous turns in a multiturn dialogue. With this talk, we will like to focus on various aspects of conversational and retail-oriented context.
Key Takeaways :
• Difficulties in Conversational Retail.
• Forms of context for retail assistants
• Importance of designing NLU components with context to attain friction-less and goal-oriented retail AI
Komal is a former Masters' student from Stony Brook University and an active enthusiast of Natural Language Processing, Search Engines, Recommendations, and Deep Learning. She is currently working as a Senior Data Scientist-NLP Research at Walmart Labs in the Emerging Tech org. She works extensively on context oriented and multimodal NLU components of Walmart’s Conversational Assistant. Prior to working at Walmart Labs, she worked at DMAI developing voice-based Educational AI for young children to help with their reading and speech abilities.