ML-based Date Format Identification in File Names

3 posts · May 2021 – Oct 2023

What this trail covers

Dropbox developed an ML model to accurately identify various date formats within file names, improving the functionality of its automated naming conventions feature. This system moves beyond rule-based approaches to handle inconsistent and ambiguous date representations, enabling more effective file organization and retrieval. The ML model uses supervised learning for multi-class classification, with components for annotation, tokenization (using SentencePiece), and classification, leveraging transfer learning and synthesized data generation.

All Dropbox trails

Oct 2023Putting everything in its right place with ML-powered file organizationThis post details the development of the 'smart move' feature, which uses machine learning to suggest file organization within a user's existing subfolder structure. It describes the challenges in defining 'organization' from a user perspective, the need for a human-in-the-loop workflow, and the development of a new ML model. The model is trained by treating existing organized folders as the desired end state, creating hypothetical pre-move cases from this structure. The post also highlights the importance of user research in shaping the feature and the limitations of model reuse for this specific problem.
Sep 2023Is this a date? Using ML to identify date formats in file namesThis post details the development of an ML model for identifying date formats in file names. It covers the challenges of inconsistent date representations, the limitations of rule-based approaches, and the ML-based solution using supervised learning for multi-class classification. Key components discussed include manual data annotation with Doccano, synthesized data generation to improve coverage, subword tokenization using SentencePiece, and IOB tagging for entity recognition. The model leverages transfer learning for cost-effectiveness.
May 2021How image search works at DropboxThis post details the technical approach to implementing image content search at Dropbox. It explains the core idea of using a relevance function based on image classification and word vectors. The post describes how image classifiers (EfficientNet trained on OpenImages) generate category scores, and how word vectors (ConceptNet Numberbatch) are used to represent search queries in the same category space. It outlines the mathematical formulation for projecting query word vectors into category space and calculating relevance scores. The production architecture leverages the Nautilus search engine, explaining how inverted indexes are used to efficiently retrieve relevant images based on image content metadata.
Advertisement