IIS Lab University of Tehran

Intelligent Information Systems Lab

آزمایشگاه سیستم‌های هوشمند اطلاعات

School of Electrical and Computer Engineering

We study how people search for and understand text, in Persian and across languages. The lab is part of the School of Electrical and Computer Engineering at the University of Tehran and is led by Associate Professor Azadeh Shakery.

Our work covers retrieval models, cross-language retrieval, Persian NLP and LLM evaluation. It appears at SIGIR, ECIR, CIKM, EMNLP, NAACL and LREC, and we release our datasets and models openly.

Publications Datasets & models

آزمایشگاه سیستم‌های هوشمند اطلاعات دانشگاه تهران: پژوهش در بازیابی اطلاعات، پردازش زبان طبیعی فارسی و ارزیابی مدل‌های زبانی بزرگ.

NewsAll news →

  1. Our paper “ParsVoice: A Large-Scale Multi-Speaker Persian Speech Corpus for Text-to-Speech Synthesis” has been accepted at the EMNLP 2026 Main Conference, congratulations!
  2. ParsVoice is now public, with 2,178 hours of Persian speech from 1,803 speakers on Hugging Face, the corpus-building pipeline on GitHub, and a fine-tuned XTTS model.
  3. The PersianPunc dataset and ParsBERT punctuation model are released.
  4. Our paper “PersianPunc: A Corpus for Persian Punctuation Restoration” has been accepted at SilkRoad NLP Workshop @ EACL 2026, congratulations!
  5. Our paper “PolitiSky24: U.S. Political Bluesky Dataset with User Stance Labels” has been accepted at Findings of EMNLP 2025, congratulations!

ResearchRepresentative papers →

  • Retrieval models

    Relevance feedback, query expansion and axiomatic analysis of ranking functions.

  • Cross-language retrieval

    Persian–English search with comparable corpora, query translation and learning to rank, plus cross-lingual plagiarism detection.

  • Persian NLP

    Corpora and models for Persian summarization, punctuation restoration and speech.

  • LLM evaluation

    Cultural and medical benchmarks for language models in Persian.

  • Conversational search

    Question retrieval in community QA, expert finding and retrieval-based dialogue.

  • Credibility & social media

    Fake news, stance and offensive language detection in social and news streams.

Latest releasesAll resources →

  • EMNLP 2026 Main

    ParsVoice

    A Large-Scale Multi-Speaker Persian Speech Corpus for Text-to-Speech Synthesis

    2,178 hours of speech · 1.36M aligned segments · 1,803 speakers

  • SilkRoadNLP @ EACL 2026

    PersianPunc

    A Large-Scale Dataset for Persian Punctuation Restoration

    17M samples · 91.3 macro-F1

  • LREC 2026

    PersianMedQA

    A Persian-English Bilingual Medical QA Benchmark for LLMs

    20,785 questions · 23 specialties