RT @krisztianbalog: Interested in LLMs and recommender systems? We have a fully funded PhD position at @IAI_group #phd #vacancy.

RT @helloiamleonie: RAG is 80% retrieval, 20% generation. The quality of retrieval dictates the quality of generation. If you do anything….

RT @jobergum: 9 out of 10 RAG applications would be better off by establishing a simple baseline before introducing text embeddings. https….

RT @jobergum: Happy Friday 😅.

We compare our initial fine-tuning method to in-context learning approach for LLM-based user simulation. We also find patterns in the types of generated utterances (shown below) and quantify their prevalence. This is an extension of our WSDM'22 paper:.

I forgot to tweet this out! 😅. "Analysing Utterances in LLM-based User Simulation for Conversational Search" with @maliannejadi and Fabio Crestani accepted to TIST! 🥳. Full paper here:

Thanks, Weronika, for presenting our work! :). I gotta admit, I'm a bit jelly of y'all hanging out in Glasgow @ecir2024. enjoy and greetings to everyone!. Full paper:

RT @JeffD: My group, @grill_lab_ai at @InfAtEd has openings for multiple postdocs and PhD students to work on the future of interactive ass….

Youngest participant at last week's #EACL2024 @eaclmeeting :D

RT @JeffD: I look forward to seeing you at #ecir2024 in Glasgow soon! As my current base, I know it well. I'm happy to share two food maps:….

Today at SCI-CHAT at #EACL24, I'll present our paper on user simulation for task-oriented dialogue systems. Come and say hi if you're around, we're on the 7th floor of Radisson!. Check out the full paper:

Crowdsourcing-based results show the preferance of rewritten answers, compared to the original one. However, they indicate high subjectivity in preference of rewerite type and entities explained, thus calling for further personalization in conv search.

We thus investigate the presence of salient entities in conv search answers and propose two rewriting strategies:.1) explicit in-line explanations, making the answer self-contained;.2) asking a follow-up question, offering more info on salient entities in the answer.

We argue that conv search limits the user to explore entities that they might be unfamiliar with, but are crucial for understanding the system's answer. This is easier in Web search, as users can easily follow hyperlinks or consult other SERP elements.

This week at @ACM_CHIIR, I'm presenting our full paper titled "Towards Self-Contained Answers: Entity-Based Answer Rewriting in Conversational Search", in collaboration with @krisztianbalog and Fabio Crestani. Preprint: Resources:

Behind the scenes of my recent PhD defence 🎓, recorded by my amazing wife! 🧡.

I am greatful for the oppurtunity to work with and learn from so many great researchers. Some of them became collaborators, some of them friends 😊. Most importantly, I want to thank my (growing) family (@yoga4academics) for keeping me grounded and loved throughout the journey 🧡.

PhD dissertation defended!! 🥳. It's been wonderful 4 years, full of life, hard work, and personal growth in all possible directions :)

That's me! 🙋 :). Mixture of excitement, nervousness, tiredness, and relief is hitting hard 😅.

Starting in ~1 hour, sign up below to get the Zoom link :). See ya soon!.