Story Upgrade is a specialized search application for building a collection of similar stories around a particular topic or activity context. It has gone through several iterations of development starting from work done by my advisor Andrew Gordon at ICT. I developed a more advanced version for my Masters project and there are now currently two separate versions; one maintained by myself and the other by Andrew.
In the current implementation, the user supplies a sample boring story (a seed) and the application suggests a new set of relevant stories to upgrade the existing collection. When a query is made the application uses the same story collection as Say Anything. The user manually browses the list of suggestions and annotates a suggestion as either relevant to the topic or not. The application uses a simple relevance feedback mechanism to update the search query and then provides a more refined set of candidates to the user. There are nearly a dozen active users and over 40 topics that have been created.
To build a topic collection the user browses the stories retrieved in the left hand column. Selecting the relevant radio button will automatically move the story to the relevant stories list in the right hand column, while selecting not relevant will move it to the not relevant list. If the text is not a story, a duplicate or not appropriate for the collection for any other reason you can select skip, which will remove the document from the retrieved list but not update the query. Only a small excerpt of a document is shown on this page. If it is not clear whether the document is relevant or not the user can view the entire story by clicking the magnifying glass.
This shows the entire document using the original HTML markup. When viewing the story the user is also encouraged to provide some additional annotations to help improve research involving the corpus. These include hard coded annotations, such as whether the document is actually a personal story or not, but also user generated tags that can be applied to the document.
The interface also allows the user to filter the results to build topics with specialized properties. For example, stories that don't have any images (which often disrupt the discourse and require non linguistic context to interpret) or to specify a minimum and maximum length to the documents. It is also possible to only show relevant documents that have been assigned a tag by at least one user. For example, to select only stories marked as funny or embarrassing.