---
id: "doduo"
category: "detector"
url: "http://registry.apicrafter.io/tool/doduo"
---

# Doduo (`doduo`)

Doduo is a research model from Megagon Labs, published at SIGMOD 2022, that predicts the semantic type of every column of a table and the relation between pairs of columns. It fine-tunes a BERT language model on the whole table written out as one sequence, so the values of the other columns help to type each column. The repository ships two pretrained models: one trained on WikiTable data from TURL, with 255 Freebase column types, and one trained on VizNet data from Sato, with the same 78 semantic types as Sherlock and Sato, such as birthDate or fileSize. The registry maps the VizNet types through mappings.sherlock, so this list is the same as the one of the sherlock and sato tools. The code is released under the Apache 2.0 license.

Website: <https://github.com/megagonlabs/doduo>

## Supported datatypes (30)

- `address`
- `age`
- `birthday`
- `birthplace`
- `brand`
- `category`
- `city`
- `comment`
- `continent`
- `country`
- `currency`
- `dc_creator`
- `dc_publisher`
- `description`
- `duration`
- `elevation`
- `filesize`
- `gender`
- `genre`
- `industry`
- `isbn`
- `language`
- `name`
- `nationality`
- `orgname`
- `person_fullname`
- `religion`
- `species`
- `weight`
- `year`