Gustavo Penha

I am a research scientist at Spotify working with search and recommendation. I am also passionate about photography and am working on my first project ๐Ÿ–ผ๏ธ.

news

May 26, 2023 Defended my PhD on Conversational Search and Recommendation ๐ŸŽ“
Apr 19, 2022 We won the best paper award with our ECIRโ€™22 paper ๐Ÿ† ๐ŸŽ‰.
Aug 27, 2021 Finished my research internship at Amazon ๐Ÿ“ฆ: CHIโ€™22 & CHIIRโ€™22.

research interests

Representation learning
Encoders learn representations for queries and retrieval units (e.g. documents), which are then used to calculate a relevance score. The goal in contrastive learning for example is thatrelevant documentsget close to thequeryandnon-relevant documentsget far from the query in the embedding space. I am interested in many aspects of representation learning, including negative sampling techniques, disentanglement, interpretability, and fairness.
Explainability and model understanding
Information filtering systems, such as rankers and recommender systems, have a large impact on what we are able to find, what we are exposed to, which content is discoverable on online platforms, and the decisions we make. A better understanding of models---for example, when they fail, how robust they are, and why they are recommending certain items over others---is crucial for stakeholders such as machine learning practitioners, end users, and content providers.

selected publications

  1. TheWebConf
    Improving Content Retrievability in Search with Controllable Query Generation
    Penha, Gustavo, Palumbo, Enrico, Aziz, Maryam, Wang, Alice, and Bouchard, Hugues
  2. ECIR ๐Ÿ† best paper
    Evaluating the Robustness of Retrieval Pipelines with Query Variation Generators
    Penha, Gustavo, Cรขmara, Arthur, and Hauff, Claudia
  3. RecSys
    What does BERT know about books, movies and music? Probing BERT for Conversational Recommendation
    Penha, Gustavo, and Hauff, Claudia

presentations

Slides for our TheWebConf 2023 paper on entities' retrievability. video ๐ŸŽฌ.
Slides for our ECIR 2022 paper on query variations. video ๐ŸŽฌ.

academic services

  • Reviewer for ECIR (19, 20, 21, 22, 23), CIKM (19, 21, 22, 23), SIGIR (21, 22, 23), RecSys (21, 22, 23), TheWebConf (20,23), MICROS'21, CHIIR'22 and CUI'22.
  • Organizer at the Delft eXplainable AI Summer School 2022 (XAISS).
  • Organizer at Search-Oriented Conversatinal AI workshop (SCAI) at SIGIR'22.

one page cv