Back office ops · Production

Dropbox uses ML model to identify date formats in file names for automated naming conventions

The problem

Dropbox's naming conventions feature needed to detect dates already present in file names before renaming them, but the wide variety of date formats, abbreviations, and inconsistent separators across files made reliable date identification very difficult.

First attempt

A rule-based approach to date identification was tried first but could not handle the breadth of formats encountered at Dropbox's scale without requiring impractical enumeration of every possible format.

Workflow diagram · grounded in source
1
File upload triggers renaming
trigger
“files uploaded to a specific folder will automatically be renamed to match the preferred convention”
2
SentencePiece tokenization
ai_action
“We chose the SentencePiece tokenizer, which is a subword tokenizer and based on the BPE and Unigram algorithms. SentencePiece provides different options based on the specific requirements of the tokenization process—which is useful for u…”
3
IOB tag classification
ai_action
“In our classification task, the transformer-based model DistilRoberta is our backbone for predicting IOB tags.”
4
File renamed per convention
output
“a machine learning model that can accurately identify dates in a file name so that files can be renamed more effectively”
5
Auto-suggest conventions
feedback_loop
“we started to automatically suggest potential naming conventions based on the naming conventions of existing files already in a folder”
Reported outcome

The ML model achieved a 40% increase in renamed files over the rule-based baseline, and following rollout in August 2022 naming conventions were applied to more than one million files in the feature's first few weeks.

Reported metrics
Renamed files vs rule-based baseline40%
Files renamed in first weeks of availabilitymore than one million
Inference latency reduction via model pruningmore than 30%
Reported stack
DistilRobertaSentencePieceDoccano
Source
https://dropbox.tech/machine-learning/using-ml-to-identify-date-formats-in-file-names
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The ML model achieved a 40% increase in renamed files over the rule-based baseline, and following rollout in August 2022 naming conventions were applied to more than one million files in the feature's first few weeks.

What tools did this team use?

DistilRoberta, SentencePiece, Doccano.

What results were reported?

Renamed files vs rule-based baseline: 40%; Files renamed in first weeks of availability: more than one million; Inference latency reduction via model pruning: more than 30% (source-reported, not independently verified).

What failed first in this deployment?

A rule-based approach to date identification was tried first but could not handle the breadth of formats encountered at Dropbox's scale without requiring impractical enumeration of every possible format.

How is this back office ops AI workflow structured?

File upload triggers renaming → SentencePiece tokenization → IOB tag classification → File renamed per convention → Auto-suggest conventions.

Related back office ops cases