Off the Shelf online
In November we were excited to launch our latest exhibition Off the Shelf. This experience encourages readers to browse around 650 books from the Library’s collection. We built this exhibition to promote access to the Library’s physical books that are hidden down in the many levels of stacks under the Library.
The exhibition app was released first as an in-gallery website, we then used the same code to build the public website, which is now ready. One of the DX Lab’s design principles is to Iterate which is exactly what we have done with Off the Shelf. It is based on our previous exhibition #NewSelfWales from a technical back end and infrastructure point of view. We wanted to spin up another exhibition fairly quickly but with new content and with a slightly different user-experience.
If a reader wants to read any of the books, they can simply take the call number to our Librarians in the Reading Rooms and request to have it retrieved from the stacks.
Whilst working on Off the Shelf we decided that the content would make great cards for our vending machine. We were also looking to add new items to the @VendingLibrary which is located in our Macquarie St foyer. Initially a three-week experiment for the 2017 Spark Festival, it became so popular that we had been keeping it full of content for over two years now. It has needed a change so we decided to update the content to reflect our recent exhibition, Off the Shelf. This involved making new content for the printed items inside the envelopes, but also updating the original Twitter bot.
Above is a sample of the new content that is now in the @VendingLibrary.
This new content is very different to what has been in the vending machine before. Previously, the bot analysed your tweet history and chose one of seven themes, then randomly chose one of eight items in the theme. With Off the Shelf content we upgraded the bot to choose one personalised book from all 56 items. The Twitter bot uses the text of the catalogue entry (title, author, date, publisher, subjects and so on) as training data.
Some technical information about the how the original version of the Vending Library and its Twitter bot work can be read here but the new version of the bot has a few differences.
As before we used the Twit package to simplify interaction with the Twitter API, however we switched to using Natural as the Natural Language Processing text classifier as it was much faster to train. Previously, we simply trained the bot with text for seven themes – but now we have 56 items to train for, and so efficiency was important.
Since the new items are all books chosen for their interesting-looking covers, we thought it was a good idea to include the image of the cover in the tweet. The bot now uploads the image via Twit before constructing the rest of the tweet. Since there is a web presence for each book cover in the Off the Shelf website, we also link to that in the tweet. From there users can of course click through to the catalogue record.
