Our updates blog is live!
Our new UiPath Communications Mining (formerly Re:infer) updates blog is now live online and in the platform!
This will keep you updated with all of the latest updates and improvements we make to the platform, as well as the brilliant blog articles written by our team on their research and various areas of Machine Learning.
From now on, we'll be updating that blog instead of this article - so make sure to visit updates.reinfer.io, or access it via the platform by clicking the '!' icon next to your username.
Historical updates
Learn about some of our most important historical updates to product features and functionality below. Click on any of the links included to find out more about the specific update / feature.
If you have any ideas for features or updates that you'd really like to see, we'd love to hear them. There's a section for product feature ideas in your community forum - please post in here!
August 2021
New Advanced Prediction Filters
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 applied, and more!
This major update does mean some other improvements have been made in Explore and Reports when it comes to selecting labels for training or choosing which labels appear on charts.
Click here to find out what the new filters can do and what the additional improvements are for Explore and Reports.
July 2021
New 'Rebalance' training mode
Alongside the brand new Model Rating functionality, we've released a new training mode called 'Rebalance'. This mode is intended to help users remove labelling bias from their models and ensure their training data is a representative sample of the dataset as a whole.
Click here to see how to use it.
Model Rating live in Validation!
Our new Model Rating functionality changes the game for model training in the Re:infer platform! It provides a comprehensive overview of your model's performance in 4 key areas, and guides you on how to improve performance in each of them.
Click here to find out how to understand it and start using it!
New Dataset settings page
We've create a new and improved dataset settings page - all the same functionality is available, it's just laid out nicely in tabs for you to access the information you need to quickly and easily.
Click here to see how it looks now.
June 2021
Multi-lingual data sources and datasets available
After a fantastic effort from our team, we're now able to support sources and datasets containing multiple different European languages, without requiring any translation for Re:infer's models to be able to understand them. This means users can train models on data that may have a mix of European languages if that's how their team or company does business.
We'll also be adding to the number of languages we support over time.
Click here to find out how they work and to see which languages we currently support.
Re:infer trial account sign-up!
Prospective customers can now sign up for a trial version of Re:infer! This will give them free access to all of Re:infer's core features with demo datasets, allowing them to explore the platform and test its functionality before they become a Re:infer customer.
This is definitely one to share with friends or colleagues who you think might also benefit from Re:infer's capabilities.
Click here to see where to sign up.
Updated Compare page
We've re-designed the Compare page slightly - all the same great functionality is there, it just looks better!
Click here to see the new design.
Improved validation for entities
Entities are an important component of any model intended to facilitate automations. We've now made more updates to the Validation page for entities to give you better insights into how well Re:infer is able to predict all of the entities enabled for your model.
Click here to find out more.
May 2021
Updated Organisations page
Our Organisations page also has a new revamped look!
Take a peak here.
April 2021
Delete verbatims from a source via the UI
Users with the necessary permissions are now able to delete verbatims from a Re:infer source within the UI (not just via the API), allowing you to quickly remove data from sources if it shouldn't be there.
Click here to find out how to use it.
March 2021
API now returns label and dataset health feedback
We've updated our API to return important information on label and model health, including the MAP, coverage, dataset quality, and the number of labels with performance warnings.
Click here to find out more.
February 2021
User defined and fully trainable entities released
One of our most exciting feature releases yet, user defined and fully trainable entities give our users complete freedom to create and train their own bespoke entities (structured data points in communications) for Re:infer to predict and extract.
Click here to find out how to use them.
New 'Check label' and 'Missed label' training modes
It's important for users to be able to easily correct any potential mistakes or inconsistencies in their training data. We've added two new training modes instead of 'Teach label' for reviewed verbatims, 'Check label' and 'Missed label'.
Click here to find out how to use them.
December 2020
Updated CSV export
We've made it easier to export the data you need right from the UI in the platform with updates to our CSV export that allow you to easily filter the data before exporting it.
Click here to find out how.
November 2020
Next best actions live in Validation
We previously released Model Warnings in Validation to alert users of potential performance issues with individual labels. We've now coupled those warnings with Next Best Actions to help you fix these issues and improve the performance of your labels. These actions act as links that take you directly to the suggested training mode in Explore!
Click here to find out more about next best actions and how they can help you.
New entity filtering and updated 'reviewed' vs. 'unreviewed' filter
You're now able to filter to verbatims containing both assigned and predicted entities, just like you can for labels.
At the same time, we've adjusted the filter for 'reviewed' vs. 'unreviewed' verbatims, which previously related to verbatims with assigned or predicted labels only, to be explicitly part of the label filter.
Click here to see how entity filtering works, or here for the new label filters.
Updated format for verbatims with assigned labels
When training in Re:infer, you can assign both labels and entities (if enabled) to each verbatim. As we've now significantly expanded our entity-related features, we've updated how we present 'reviewed' verbatims in the platform.
This is in order to reflect the differences between verbatims that have been reviewed by assigning labels, or by assigning entities, or both.
Click here to see the new format for verbatims that have labels that have been assigned by a person.
October 2020
New CSV upload feature
You asked and we listened: you can now upload data from a CSV file directly into a source in Re:infer's user interface in a simple step by step process! To learn more about how to upload a CSV file to Re:infer, see here.
Improved dataset creation and editing flows
We've recently updated the process for creating and editing datasets, to make it much smoother and more intuitive. We hope you agree it's a dramatic improvement!
To see how to create a new dataset click here, to update an existing dataset click here.
September 2020
Introducing validation for entities
If you're looking to extract entities (structured data points like emails, IDs, or dates) from your data, you want to also know how well Re:infer is able to do so - are Re:infer's entity predictions correct? Are entities being missed?
So, similarly for labels, we've released validation for entities, which helps you understand how accurately Re:infer is able to predict each entity that you've enabled. Click here to understand how validation for entities works.
New dataset status feature
When you're labelling data in Re:infer, you want to know whether your training has been incorporated into the model and the predictions have been updated, and if not, how long will it be until they are.
To help you keep tabs on how your model is retraining and re-predicting, we've released a new dataset status feature. Click here to see how to understand the status of your dataset.
August 2020
Easily interpret clusters in Discover
Discover now highlights the parts of a verbatim that most contribute towards it being included in the cluster, helping you identify the common themes quicker! Click here to see how this works when labelling clusters in Discover.
July 2020
Exclusion filters
When filtering different string-format metadata properties, you can now choose to exclude certain values that you're not interested in! This makes it much easier to filter out certain values for a metadata field that has a large number of potential values.
You'll notice that the filtering interface now looks a bit different for these properties, but we hope you find it pretty intuitive.
Follow these links to see how the new filters work in Explore and Reports.
Model warnings in Validation
The platform now provides model warnings in Validation, whenever you select an individual label that is not performing as well as it should be. Re:infer will also start to suggest the next-best actions to try and improve them, and this is a feature we will continue to develop to make these suggestions as helpful as possible.
To see how they help you understand and improve model performance, click here and scroll down to the 'Model warnings and next-best actions' section.
June 2020
Email recipient filters updated
We've updated how our filters work for recipient related user properties for email datasets. You are now able filter on number of recipients or number of recipient domains, as well as being able to filter to a specific recipient or recipient domain present in any of the 'To' 'CC' or 'BCC' fields of an email.
Follow these links to see how filters work in Explore and Reports.
Verbatims with dismissed labels return to 'Unreviewed'
Previously if you added labels to a verbatim and then decided to remove them later, they would remain in the same 'Reviewed' format, but have no labels associated with them.
We've updated this now so that if a verbatim has all of its labels dismissed, it reverts back to 'Unreviewed' and the platform will start to make predictions for it when it retrains. This acts as a sort of 'undo' feature, letting you correct things or remove labels more easily, without having to apply new ones.
'Introduction to Re:infer' video available
The latest of our explainer videos is now available. This video gives users a high-level introduction to Re:infer, covering:
- Who we are,
- Why we exist as a company
- What we do
- How we do it
- How we're different
- How our clients are using our platform
To watch the video, check it out here.
May 2020
Intelligent label performance warnings
To help you easily understand and improve label performance, we've introduced intelligent label performance warnings in the platform. These will warn you when a label is not performing as well as it should be.
We'll be regularly updating the inputs to these and creating more and more intelligent 'next best actions' to help you improve labels quickly and effectively.
To see an explanation of these performance warnings, click here and scroll down to the 'Label performance' section.
Intelligent searching in Explore
We've updated how searching works in the platform. You can now add quotation marks (" ") to your search terms in the search bar to force an exact match of the terms within them. If you don't add quotation marks, Re:infer will return exact matches first, and then partial matches once you have passed all of the exact matches.
Explore will also give you an approximate count of how many verbatims match your exact search terms, giving you a rough idea how many times a concept or theme might appear in your dataset.
To learn more about how search works in Re:infer, click here.
'Platform Overview', 'Entities Training' and 'Using Discover' videos available
The first of our training videos are now available on the Knowledge Base. These will be the first of many, so keep an eye out for more.
To get a feel for navigating the platform and an overview of all the main features, check out the Platform Overview video.
To understand more about Entities and how to use them in the platform, watch the Entities Training Guide video here.
To learn about the first phase of the model training process, 'Using Discover', watch the video here.
April 2020
Discover looks for poorly covered clusters
In February, we updated Discover to take into account your taxonomy when re-training. We've now updated it so that when it retrains, it tries to find clusters of verbatims that are not well covered by your taxonomy (meaning that predictions for each of the labels are low).
Finding and presenting clusters is always a balancing act - we want to balance finding good clusters with clear connections between the verbatims, with finding new things that don't fit your current taxonomy. There will always be a mix of some clusters with strong predictions, and others with less strong or no predictions, depending on how your taxonomy has been trained so far.
To find out more about Discover and how it works, click here.
February 2020
Knowledge Base and Support Centre goes live!
You'll know that we've launched or new Knowledge Base and Support Centre (support.reinfer.io), as you're currently in it!
Now you've signed up, you can continue to access this site from the platform by clicking the 'Support' option in the Global Action action dropdown menu, found in the top right-hand corner of your screen at all times.
Discover takes into account your taxonomy when re-training
When it retrains, Discover now takes into account your taxonomy and the training that you've completed to-date, as well as looking for clusters of semantically similar verbatims, in order to increase the chances that the new clusters it creates are interesting to you.
To understand more about how Discover works and how to use it, click here.
The Datasets page now lets you filter by organisation
If you belong to multiple organisations in Re:infer, the Datasets page can get a bit crowded with all of the different datasets that you have access to. We've added a simple filter bar at the top to let you filter to datasets within a specific organisation, to cut out some of the noise.
To understand more about the Datasets page, click here.
January 2020
Customisable dashboards
The Dashboard page now lets users create fully customisable dashboards. Users can add or remove any charts they want with only a few clicks, and fully adjust their size and arrangement.
To find out more about how to set up and edit your own dashboard, click here.
'Low confidence' mode
Within Explore you can enable the 'Low confidence' mode, which will present you with verbatims in the dataset for which the model has very low confidence that any of the labels in your taxonomy applies to them.
This, as well as 'Shuffle' mode, is a useful way to see how well your taxonomy covers the verbatims in the dataset, and may show you where you need to create new labels to capture interesting intents or concepts you may have missed.
To understand more about how to use 'Low confidence' mode, click here.
Updated Organisations page
The Organisations page has now been updated to show you all of the organisations that you and other users belong to (as long as you have permissions to see them), and lets you switch between them easily by simply clicking on the one you want.
To understand a bit more about the Organisations page click here. Or to understand more about how to manage users, click here.
December 2019
Discover now re-trains
We've updated Discover to show you new clusters, once you've done a significant amount of training or added a significant amount of data to your dataset. This way Discover continues to be useful and hopefully shows you new and interesting things.
To understand more about how Discover works and how to use it, click here.