TASHEEH wins the EDBT Best Paper Award

Our paper on repairing malformed rows in raw CSV files received the EDBT 2024 Best Paper Award!

Repairing the structure of messy files

If you have previously heard my research pitch, you know that CSV looks like a very simple format, but real files have a talent for breaking simple assumptions. These problems have been at the center of my PhD thesis. Rows can have missing or additional fields, quotes can be inconsistent, and metadata can appear where a parser expects a table. Before we can clean the values in such a file, we first have to recover its structure.

In TASHEEH, we study how to repair malformed rows by learning from the dominant patterns in the file itself. The system identifies structurally inconsistent records, aligns them with well-formed patterns, and applies transformations that bring the rows back into a consistent shape.

A very welcome recognition

I am very honored that this work received the EDBT 2024 Best Paper Award. The paper was led by Mazhar Hameed, and we worked together with Fabian Panse and Felix Naumann, and it grew out of a longer line of research on making messy files usable without asking people to repair every corner case by hand.

More than the award itself, I am happy that data preparation research is gaining attention. File repair may not be the most glamorous step of a data pipeline, but nothing downstream works particularly well if the data cannot be loaded correctly in the first place!