Re:infer's property filters are now conveniently grouped by category (Email, Thread, or User).
Icons now distinguish between string and number value types, and Re:infer also provides examples for string user properties.
We're proud to reveal a completely new model for all trainable entities. You can expect better accuracy, faster training, better validation statistics, and many exciting opportunities for fine-grained entities in the future!
Our new entity model is built on a transformer-based architecture - a type of deep learning that learns to only pay attention to the most relevant parts of an input.
We've massively improved our capabilities for filtering by label predictions. You can now apply multiple label filters at once, exclude certain labels with a filter, search with label filters, and more!
This major update coincides with some closely linked improvements in Explore and Reports for selecting labels to train and choosing which labels appear on charts.
We're excited to announce general availability of 14-day free trials of Re:infer. No credit card required. Includes access to all features and dedicated Customer Success and Technical Support.
You can upload your own communications data, including using our pre-built connectors. Alternatively, get started using our sample datasets.
In this blog post we discuss a tool we build to ensure data used for training is balanced and has minimal labelling bias.
Training a machine learning model from scratch is hard. It is a common misconception that building the model itself is the hardest part of training a model. In reality, the data collection and annotation process is far harder than people assume.
In this blog post, we talk about how Active Learning, which sits at the heart of the Re:infer platform, reduces the amount of data required to train a high quality model.