Final year PhD Candidate @UniLeiden | Ex- @bookingcom | Ex-Microsoft | EX-VR @Irlab_amsterdam | Interested in IR, NLP, Large Language Models
Joined April 2018
- Tweets196
- Following801
- Followers534
- Likes1.9K
Pinned Tweet
🎉 Thrilled to share that my internship at Microsoft, titled "MAGIC: Generating Self-Correction Guideline for In-Context Text-to-SQL" has been accepted at the at AAAI 2025 (Main)! together w/ Christian Pölitz and Xinye Tang! #AAAI2025 #AAAI #NLP #Microsoft #SelfCorrection
1/9
Arian Askari retweeted
Excited to share that our paper "Query Performance Prediction using Relevance Judgments Generated by Large Language Models" has been accepted by ACM Transactions on Information Systems (TOIS), with @NegarEmpr, @arian_ask, @maliannejadi, @mdr.
dl.acm.org/doi/10.1145/37364… – at Amsterdam, The Netherlands
🎉Our paper “HotelMatch-LLM: Joint Multi-Task Training of Small and Large Language Models for Efficient Multimodal Hotel Retrieval” has been accepted into #ACL2025 (main). The result of my internship at Booking with Emmanouil Stergiadis, Ilya Gusev & Moran Beladev
🎉 Excited to share that our full paper titled "Self-seeding and Multi-intent Self-instructing LLMs for Generating Intent-aware Information-Seeking dialogs" has been accpeted at #NAACL2025 together w/
@rmpetcu, @ChuanMg, @maliannejadi, @aminvenv, Evangelos Kanoulas, and
@Suzan
- Manager agent in MAGIC:
- 🔄 Controls the feedback-correction cycle
- ✏️ Revises instructions
- 📝 Generates guidelines
- 🔍 Using the guideline generated by MAGIC enhances interpretability of corrections, offering valuable insights into LLM failures and successes.
8/9
🔹 Agent interactions, code, and updated pre-print will be available soon.
Paper link: lnkd.in/eMvhgFqq
9/9
📦 To make SOLID more accessible, we’ve also developed a Python library called SOLID-Conv, designed to generate synthetic intent-aware dialogues with LLMs, based on the SOLID method from our paper. You can explore it on github.com/arian-askari/SOLI….
Using SOLID is quite easy! You could see a quick code example in below figure. SOLID can be used for generating intent-aware dialogues. This can enhance your conversational AI projects in various aspects including intent detection.
- 💡 SOLID-RL: Further refines this by generating dialogs in one step and improving quality through a length-based estimation mechanism.
- 📊 Impact: Over 300k intent-aware dialogs generated, improving intent prediction quality compared to human-annotated data.
Effortlessly generate synthetic dialogues given a sequence of intents ! 🤖 Our new ACL 2024 NLP for Conversational AI workshop paper w/ @rmpetcu, @ChuanMg , @maliannejadi , @aminvenv , @ekanou , and @suzan ! 🎉
#ACL2024 #NLP #ConversationalAI #LLMs
What is SOLID?
- 🔍 Intent-aware Dialogs: SOLID uses large language models (LLMs) to generate dialogs that are aware of user intents, addressing the challenge of manual intent annotation.
