Presented at the South Australian State History Conference, Adelaide, 23 September 2013.
- Trove provides access to more than 350 million resources
- Including more that 100 million newspaper articles
- Trove receives more than 70,000 unique visits per day
- And has more than 100,000 registered users
- Trove users have added more than 2 million tags
- And corrected more than 100 million lines of text
They’re all a bit too big to be comprehensible aren’t they?
I want to try try and bring some of these numbers down to a local scale, a human scale, to see beyond the numbers and to try and explore Trove as a platform for making meanings.
Bringing it all together
Trove is a discovery interface – it helps you find stuff. But obviously we’re not trying to compete with Google. Indeed, most of our traffic comes from Google.
Our aim to is work with other organisations to bring content by Australians, about Australia, or of interest to Australians to the surface. To make it easier to find and to use.
To achieve this we aggregate metadata about collections. We collaborate with all sorts of organisations to find a way of harvesting the information that they have online and bringing it into our databases.
That’s not just libraries, but:
- Research projects or agencies
- Historical societies
- Local councils
- ? ask us
From small – for example, the Bega Historical Museum
To the large – the National Museum of Australia.
While a large amount of our resources are books and journal articles, reflecting Trove’s bibliographic roots, most of our collecting effort goes to unique Australian content. That’s the stuff that gets us excited. That’s the stuff that we want to help make as accessible as possible.
I’ll admit that aggregation in itself doesn’t seem particularly exciting – we’ve been doing this stuff for many years, generations of portals have come and gone... But what does get me excited are some of the things that aggregation makes possible or easier – possibilities beyond the search box, beyond simple discovery.
The possibilities of scale
I’m sure you’ll have no trouble guessing what the most heavily used part of Trove is…
Yep, the digitised newspapers.
The digitised newspapers are the result of another collaborative effort between the NLA, the State and Territory libraries, and, increasingly, local and regional organisations that are raising funds to aid the digitisation of local titles.
The digitised newspapers are also an example of the power of scale. Technologies like OCR, full text indexing and the humble keyword search are not very exciting on their own, but apply them to 100 million newspaper articles across 150 years of Australian history and…WHAM! Something different happens. I’m sure many of you here today have a story about how access to digitised newspapers filled in a crucial gap in your research, or alerted you to subtleties or complexities in your story that you hadn’t appreciated, or lead you in a totally new direction.
This form of access enables us to re-route our inquiries around the assumptions of indexes and the politics of received memory, without having to chain ourselves to a microfilm reader for years.
But with the power of keyword searching comes new challenges – we have to be prepared to ask questions about what we’re searching and how.
What’s in the Trove newspaper database? This was the picture last week…
This graph simply shows the number of newspaper articles by state and year. You can see the effects of investment in digitisation by the State Library of NSW and Victoria's emphasis on the World War I period.
How do these sort of digitisation priorities and investments affect what you get back in your search results?
What about the quality of the OCR?
Did you notice that the Renmark Pioneer was recently added to Trove? I love it because the early editions are handwritten. This is fantastic, but of course the OCR results are less than stellar, so until someone jumps in and corrects it, these articles will difficult to find.
The overall accuracy of the uncorrected OCR sits at around 80%, but this varies considerably by title, date and issue. While our text correctors have done an amazing job, much of their work has been focused on ‘family notices’, reflecting their importance in genealogical research. The vast majority of news articles have had no corrections at all.
What does this mean? It means that the database itself, its development, its assumptions, its operation, its content is not immune to the workings of history.
There is no perfect, unqualified means of access. Beyond the huge numbers and overwhelming possibilities are events that are local, particular and contingent. There are stories to be told about the newspapers themselves, and how they came to be in our web browser.
This a scatterplot of number of words in news articles on the front pages of newspapers.
This tells a story about the changing nature of the front page, or more specifically, when the news moved to the front page. That’s the big picture. But underneath that are all sorts of complexities, played out across hundreds of separate newspaper titles. Here’s just one - when did the Advertiser change?
One on particular day, the newspaper turned inside out. Why?
There are stories to be told, even around the structure of newspapers themselves.
Interpreting the newspaper corpus is challenging, but the nature of the aggregation gives us the opportunity to play with scale – to twiddle the knobs, to zoom in and out, to look for large-scale patterns and then dig down to a particular day in a particular newspaper in a particular place.
That’s what I really like about a tool I built called QueryPic. QueryPic helps you look for patterns or changes across your newspaper search results. So instead of a list of 100,000 matching results, you get a graph, showing the number of articles that match your query over time.
This QueryPic graph show the number of newspaper articles matching the query 'city of churches' AND Adelaide over time. You can also do quite specific searches – limited by title, date, or state.
This is a comparison of the occurance of 'german' in South Australia and NSW newspapers.
But it’s not just the big picture, you can drill down – find out what’s under each point.
AS QueryPic demonstrates, I think, the scale of aggregated content offers both interpretative challenges and the opportunity to shift perspective – from big pictures to small stories.
Many small acts of meaning
Trove is often used as an example of a successful crowdsourcing project. Certainly the numbers are astonishing – more than 100 millions lines of text corrected; with the top corrector personally responsible for 2 million lines!
Academics are now exploring the motivations of people involved in Trove’s text correction – weighing up altruism and self-interest, rewards and recognition. We’re obviously interested in this sort of research, particularly as the rate of correction seems to be flattening out.
But I also wonder about the context of each of those individual acts of enrichment and engagement, of the meanings that are gained and given. It’s clear that Trove has changed lives, but even those who aren’t paid up members of the Trove army can make connections.
Text correction is not the only way that people can enrich or engage with the content in Trove – they can add tags and comments. They can also create lists. Lists are collections of resources and provide a convenient way for people to save and share their research.
And they do.
There are currently more than 27,911 public lists created by 5,796 users. Those lists contain almost half a million items.
Unlike text corrections, Lists leave plenty of contextual clues. We can examine their titles, descriptive notes, and of course their contents. So I took those 27,911 public lists, grabbed all the titles and did a quick word frequency visualisation using VoyantTools.
Given what we know about the interests of Trove users, the dominance of the word ‘Family’ is hardly surprising, though it does give us a clearer picture of how Lists are being used. What is more tantalising are words like ‘mowers’. If we drill down a bit we find that there are indeed 107 lists about lawn mowers, created by one user. That, to me, is just wonderful. I just love to be surprised by the passions of our users.
Lets’s get a bit more specific. Instead of just looking at everything, I decided to focus on Lists that included ‘South Australia’ in their title.
- 109 lists
- 42 users
- 4587 items – comprising …
This time, instead of just examining the titles of the lists, I harvested the details of all of the 4000 items:
Billiards, whales, earthquakes, oysters, archery, motoring – again it’s a tantalising glimpse into the questions that lead people to Trove.
Now, I should explain that I’m not using any special Trove Manager powers to undertake these analyses. This data is all public, and is accessible in machine-readable formats via the Trove API. Anyone could start to play around with the user-generated content in Trove – I’ll even make available the scripts I used.
What this means is that as well as being a platform for the discovery of Australian resources, it’s also a platform for exploring the connections, the meanings that people find within, between and around those resources. It’s a living laboratory.
Distributed modes of meaning making
For all its impressive statistics, Trove is still just one little corner of the web.
That phrase ‘of the web’ is an important one, because packed inside it are a series of standards, expectations and philosophies. The days of standalone databases, of walled gardens and silos are going, if not gone. We expect links to go somewhere, we expect resources to be connected. We expect to be able to share, re-use and annotate, not just in the spaces controlled by cultural institutions, but in our own social realms.
Trove Tuesday is a weekly event that has developed without any sort of ‘official’ support, simply powered by the enthusiasm of users. Every Tuesday bloggers share their latest stories or discoveries sourced from Trove.
Tools like Pinterest or Storify show that we have a passion to collect across the whole web. So while Trove’s Lists provide an easy way for people to organise their research, they’re only the beginning. We need to be able to support the process of discovery, enrichment and meaning-making wherever it might happen.
We need to find better and smarter ways of linking cultural heritage resources to the interpretations, narratives and arguments they support. Every footnote should be a portal, every article a collection interface.
One way we can start to build these sorts of frameworks is through the Trove API. An API is just a way for computer programs to talk to each other in a language they both understand. The Trove API sits around waiting for programs to ask it to do something, and then it delivers results from the database in a machine-readable form, without all the fancy bits that we humans seem to like.
My analysis of front pages was built using the API. Every time you create a new graph in QueryPic you are accessing the Trove API.
Instead of expecting you to come to a particular website, at a particular address, the API puts the data where you are. It gives you the tools to create your own interfaces, your own tools, your own ways of displaying and using the data.
TroveMosaic, created by Mitchell Whitelaw, is an example of a new discovery interface, focused on pictures, using the API.
Pixtory is an iPhone app developed over the course of a weekend during the GovHack competition. It harvested images from the State Library of WA using the Trove API and uses them as a way of exploring the changing meanings of place.
Look out too for Advintage, another GovHack prize winner, that explores social history through advertisements extracted from the newspaper database via the API.
The library writer and thinker Lorcan Dempsey coined the phrase ‘discovery happens elsewhere’ to raise the horizons of cultural institutions beyond their own carefully managed discovery services. Meaning-making also happens elsewhere and we have to find better ways of supporting it.
Europeana, which aggregates content from across Europe, talks about their API putting collections ‘in your workflow’. What about in your life? Instead of the awe inspiring statistics I want Trove’s success to be measured by the small, the local, the individual acts of meaning making wherever they might happen. I want Trove to help us find new, exciting, and challenging ways of bringing our pasts into conversation with our present.