Amir Soleimani

I am passionate about AI because it aids humans in processing complex and large amounts of data. I am an NLP researcher received academic and self-training to continuously learn new knowledge and skills. I like facing new challenges and using my experience to find creative and effective solutions. I am highly skilled in research and problem-solving, from defining problems to collecting data, employing existing and new methods, and proposing innovative solutions. I am capable of working independently while also being committed to collaborating and contributing effectively as a team member.

I am a PhD candidate at Language Technology Lab (LTL). I worked on the application of Language Models with Christof Monz and Marcel Worring. My research interests are Language Representation Models, Document Understanding, Document Summarization, Question Answering, Question Generation, Factuality Evaluation, and Claim Verification.

Highlights:

• I defended my PhD on 3 April, 2024. You can find my PhD thesis here

• Our long paper "NonFactS: Nonfactual Summary Generation for Factuality Evaluation in Document Summarization" has been accepted at #ACL2023 Findings. [github]

• I finished my internship at Naver Labs Europe (Jun-Nov 2021). I worked with my supervisors: Vassilina Nikoulina, Benoit Favre and Salah Aït-Mokhtar. We worked on summarizing scientific papers given users desired queries or perspectives when there is limited labelled data or no data available.

  • Our paper "Zero-Shot Aspect-Based Scientific Document Summarization using Self-Supervised Pre-training" has been accepted at BioNLP2022 [link]!

• I started writing simplified blog posts at asoleimanib.medium.com. 

• NLQuAD: A Non-Factoid Long Question Answering Data Set

  • We introduced NLQuAD, a non-factoid long question answering dataset from BBC news articles. NLQuAD’s question types and the long length of its context documents as well as answers, make it a challenging real-world task. NLQuAD consists of news articles as context documents, interrogative sub-headings in the articles as questions, and body paragraphs corresponding to the sub-headings as contiguous answers to the questions. NLQuAD contains 31k non-factoid questions and long answers collected from 13k BBC news articles. Check github for the dataset and codes! 

• Longformer for MS MARCO Document Re-ranking Task

  • Cooperation with Ivan Sekulić and Mohammad Aliannejadi  to check the performance of Longformer on the new MS MARCO  document retrieval dataset

• BERT for Evidence Retrieval and Claim Verification

  • We showed how BERT can be used in a two-step verification system. It firstly retrieves evidence sentence and then verifies a claim against top retrieved evidence. We also showed the good effect of Hard Negative Mining in training the BERT model.