Following the rise in the use of digital assistants such as Amazon’s Alexa, Apple’s Siri, and Google Assistant Murray Husband Head of ICT for East Renfrewshire Council suggested the idea of building the ability to allow digital assistant users to ask questions where the source data is held by local authorities. Dubbed the Citizen Interaction project.
Looking at the media’s reaction to voice assistants we can see the familiar pattern of: “this is new”, moving over to the 6 to 12 months of “this is overrated” before we finally turn towards the “we knew this was brilliant all along”. The overrated period which we are currently in is the part of the tech journey that sees the building of standards, infrastructure and developing a dedicated following.
The fact that we see the public sector and specifically East Renfrewshire Council interested this early on in the journey towards the wide-scale public acceptance of a technology is amazing. Being a part of that journey even more so.
The larger project of building the infrastructure to work with digital assistants involves many steps:
- First we will need to define the questions and variations of those questions that will be asked of the assistant
- Second we need to determine what answers are required for those questions
- Third we need to define the source of the information to support those answers
- Fourth we need to build this knowledge into the formats behind the digital assistants
- Finally we need to disseminate and market these products
Wittin’s work in our project with East Renfrewshire Council focuses on the first and second steps of this process. Our goals are to:
- Arrive at a series of questions citizens ask of their local councils
- Derive a series of variations on those questions
- Build a series of answers responding to each variation and question
- Finally help to track down the source(s) behind each answer
We are aiming to pull together the foundations to develop the ability to support digital assistants with public sector data in East Renfrewshire. By extension, this work can be extrapolated and used as a basis to move outward and support other local authorities in Scotland and the UK as a whole.
Our first goal was to list any and all data sources that recorded some aspect of an interaction between citizens and East Renfrewshire council. To that end we looked at Social Media, the website live chat software, their CRM system, the telephony system (recordings of all phone calls), google analytics to see what visitors to the website wanted to look at, FOI requests, and then finally as the process at East Renfrewshire dictated that issues that could be fixed within the required system done so these requests weren’t being recorded within their CRM system and pointed us to their benefits and council tax systems which held details of interactions with citizens.
We are currently going through the process of scarping this data together to get a good look at what is available, how much there is, and what the quality of that data is.
This data holds all kinds of personal information and will be difficult to anonymise. Speech to text will have its own problems and then leave us with all kinds of unstructured text to work with that includes all kinds of personal details. Our approach is to view this work in-house at Wittin bringing in individuals from our Community that can be suitably trained, contracted up and prepared for access to personal data.
Stage two of our project focuses on writing a privacy impact assessment and re-evaluating our contract with East Renfrewshire based on the data actually found and how we propose to process it moving forward through the project.
What do we want to know?
We sat down and tried to figure out those questions we needed to ask the data that we could gather together. We built a list of
- Can we identify the topics behind groups of questions?
- What are the actual questions being asked?
- What are the variations that those questions are asked in?
- Where are the answers provided to those questions?
- What are the variations of those answers?
- Where are the sources or data that feeds those answers?
- Can we cross-reference questions/topics being asked about and a list of transactional services?
- Can we identify a preamble and postamble in conversations between the council and the public?
- How can we identify variations in preamble and postamble?
- How can we map the complexity to processing questions?
A big debate on the day was whether or not postamble was a word. My computer doesn’t think so, but if you know what we mean and have a better word let us know.
How to get involved if you are local
If you are close enough by to Dundee, Scotland, have a few data skills you’d like to stretch out particularly in the natural language processing and k-means clustering arenas we’d love to hear from you. We are aiming to take up the analysis stage in September / October.
How to get involved if you are not local
We are aiming to release the processes we develop with mocked up or synthetic datasets that will allow people to learn from our work. This will be in a use case blog as well as in a more in-depth tutorial. Get involved with giving this a go yourself.
Another task we need is to be held to account, we’ll give as many details as we can about our process and thinking, but we are only watertight if we can stand up to scrutiny.
Finally though if you have data you’d like explored then talk to us and we can undertake a project like this for your organisation. Give us a buzz or send through an email and we can chat!