The topics for the fourth year of the Conversational Assistance Track (CAsT) have been released: https://t.co/ePIC1WePYV
New this year: (1) Different topic format, (2) Mixed-Initiative system responses (3) Complex discourse structures
@JeffD@JTrippas@maliannejadi@leifos
Check out the first session of @scai_workshop 2022: "Conversational Search Evaluation", where I present a way to evaluate conversational systems by user simulator that predicts users' satisfaction scores along with actions and utterances 🤖🤖
Joint work with @AldoLipani
Excited to announce that I've been promoted to Associate Professor (Senior Lecturer) at @GlasgowCS.
My group, @grill_lab_ai has open positions for PhD/Postdocs on interactive and multimodal foundation models -- https://t.co/FNIG6Y9ayj #conversationalai
📢 3 new research-focused faculty positions @HeriotWattUni computer science. We're looking to strengthen our team in #NLProc and #ConversationalAI. DM me if interested
🥁🥁🥁 @IgluContest is about to start! The focus is on building an interactive agent that can understand the instructions in natural language to build structures in the #Minecraft environment! In case of uncertainty, ask clarifying questions! Interested? #SIGIR2022#NAACL2022
user simulator for answering clarification questions
evaluation: automated NLG & document retrieval metrics, human annotations of naturalness & usefulness
@puding_p@scai_workshop#sigir2022#convsearch
https://t.co/8w7JnjuRgo
risk-aware policy trained with RL deep q-learning decides to ask a clarifying question based on a user simulator response & future results
session-based evaluation metrics: recall, MRR, decision error rate
@Zhenduo_W@scai_workshop#sigir2022#convsearch
https://t.co/ux7jZnE6PY
limitations of user simulators for conversational search? @JeffD
* lack of multi-turn data with natural long-term trajectories
* personalisation, cultural & language variations
* need for taxonomy & understanding of different system initiative actions
@scai_workshop#sigir2022
In the perspectives session, I enjoyed the talk by @krisztianbalog on on natural language user profiles for recommendation, especially how they used a few-shot prompt template to turn a large language model (LaMDA) into a recommender. https://t.co/OUAdVh5nw0
Valeriia Bolotova et al. got the best paper award at #SIGIR2022 for their very extensive paper titled "A Non-Factoid Question-Answering Taxonomy", which I really like because it is such a timeless topic and remains relevant https://t.co/taap1MJxJO
🤵♂️I can see myself in the corner :-)
Jokes aside it was a wonderful session. Thanks @scai_workshop and @SIGIRConf for the opportunity, my fellow presenters, session chair, and all the audience for the wonderful discussion at the end of the session.
#SCAI#SIGIR2022
Chairing the mixed initiative session at @scai_workshop. Sandeep (@lolthyself) from Amazon presenting The Effects of System Initiative during Conversational Collaborative Search https://t.co/chMkjzAb69
#SIGIR2022#scai2022
main challenges for conversational QA? @AldoLipani
* difficulties in evaluation:
* lack of metrics for answer naturalness & facet discovery
* need to measure user satisfaction
* incorporating personalised context
* leveraging non-parametric information sources
GRILLBot winner of 2022 Alexa Prize Challenge: cooking & DIY assistant
TaskGraph representation: branching for interaction & ingredient requirements
LLM prompts for contextual QA & question taxonomy
@carlos_gemmell#sigir2022 https://t.co/TbfkcQTymJ https://t.co/8IwFhP1vIz
fine-tuned REL on AMT annotations of concepts and entities in conversations helpful to answer follow-up questions
@hideaki_joko#sigir2022#convsearch
https://t.co/vFqe8eJbqw
https://t.co/m2g3rn97eP
https://t.co/Y7gYrLnrzA