We’re constantly improving the Re:infer Conversational Data Intelligence platform, extending its capabilities and making it easier to use across the enterprise.
Take a look at some of our biggest launches and improvements from the last quarter…
Quality of Service
This quarter, we were very proud to launch our new platform extension - Quality of Service. Building off our recent work on tone analysis, Quality of Service gives our customers the power to quantify, analyse and improve the service they deliver - with unprecedented accuracy and insight into customer intents.
Re:infer Quality of Service provides businesses with complete real-time visibility into service levels and the capabilities to extract sentiment and emotion data from service conversations en masse.
Unlike simplistic sentiment or effort analysis, the Quality of Service Score factors in sentiment, phrasing and intent - making Quality of Service signals hyper-relevant to the business context in which they are deployed.
New Validation Dropdown and Starred Models
We've updated our model Validation functionality with a new and improved dropdown menu, helping users find the models they are looking for faster.
The new dropdown lets you see all validation scores on a given dataset, rather than just pinned models. You can also prioritise or 'star' them so that they appear at the top of the list in future.
We have created headings for 'All' and 'Recent' models as well, for faster and easier navigation.
Improved Calls and Chats UI
We have greatly enhanced our user interface for viewing calls and chats in the platform.
When listening to calls in Re:infer, the conversation will scroll automatically so you can easily follow the chat. Users can also take direct control of the scrolling as preferred.
We’ve implemented a new audio control button that lets users skip to the next speaker or chat in the conversation. Clicking on a speaker’s name will also immediately jump to their turn in the conversation and start playing their audio.
When labelling a comment in chats or calls, Re:infer now automatically marks the previous one as ‘uninformative’. We also allow users to explicitly mark comments as uninformative while they label. This means Re:infer will not try to use these for training.
Thread-aware Labeller for Chats & Calls
Re:infer supports a wide range of different message types and communications. However, we wanted to improve the platform’s accuracy and ability to make more granular predictions for calls and chats.
Phone calls and chat exchanges are broken up into ‘turns’, in which a different speaker has their turn to speak in the conversation. To provide predictions for each turn, we had to treat each turn in the conversation as if it was its own unique message (or ‘verbatim’). However a single turn in a chat often doesn't make sense when taken without its context.
We therefore worked hard this quarter to modify the platform’s labeller - enabling it to analyse and extract context from the previous verbatims that are in the same chat when trying to predict labels.
Re:infer’s new labeller is now capable of analysing the previous five turns in a chat or call when it tries to decide whether a label applies or not. This work has made the platform better than ever for chats and phone calls - able to extract valuable insights from every relevant turn in the conversation.
Editable Label Description Fields
When Re:infer users create labels for their messages these are saved in the Settings menu in a collection known as a taxonomy. Now users can add additional context and explanation to their labels through editable description fields.
This is a great feature for teams working together to label a dataset, helping them understand what each other’s labels mean. The descriptions are also a useful reminder for people labelling alone.
Work in progress
This isn’t the last time we’ll be improving Quality of Service. Currently, we’re building an Alerts feature for the new platform extension. This will allow users to create their own customised alerts when an event happens - such as when service quality falls below a chosen acceptable level. As part of this, we’re creating an Alerts user interface that will allow users to see and respond to alerts and notifications as they happen.