Aditya Vashishta, Sam Sudar {adityav, sudars}@uw.edu
Millions of people cannot access services using conventional graphical user interfaces. These users include the visually impaired, the low-literate, speakers of tribal languages without font support, and those without access to computers or mobile devices. Interactive voice response (IVR) systems provide a way to interact with users without the need for graphics. Callers are read options and use the keypad to navigate the content, lowering the barriers for access and providing the opportunity for engagement. Current IVR systems provide a wide spectrum of services, including health information, citizen journalism, and social media. However, little work has been done to inform the organization of IVR content. We compare list, shallow and deepr hierarchies to organize content in an IVR system, expanding on similar work that has focused on graphical user interfaces. We find that both flat lists and deep hierarchies are inappropriate, and that shallow hierarchies create a more effective IVR structure.
Below are the hiearchy structures we employed. The list structure is simply a flat list of the leaf nodes.
Poster is available here: https://github.com/CSE512-14W/fp-adityav-sudars/blob/master/poster-adityav-sudars.ppt?raw=true
Final Paper is availble here: https://github.com/CSE512-14W/fp-adityav-sudars/blob/master/final/paper-adityav-sudars.pdf
The code in this project relies on the software Voxeo Prophecy, Express Talk,
SQLServer, and IIS. The code is written in C# and generates aspx files. The
directories must be served using IIS, at which point the Start.aspx files may
be run. These files are largely Voice XML files that define the IVR hierachy.
You will have to point Express Talk at these pages to access the IVR.
Please see the 7 page detailed documentation on how to set up our code and an IVR server on a Windows machine. The documentation is available here: https://github.com/CSE512-14W/fp-adityav-sudars/blob/master/Running%20Instructions.pdf
To create the hierarchy png files from the Graphviz dot files, run a command similar to the following:
dot -Tpng shallowHierarchy.dot -o shallowHierarchy.png
We have conducted 3 between-subjects experiment design to compare three designs of IVR interfaces: linear list, shallow hierarchical interface, and deep hierarchical interface. We have used the \textit{Household items} dataset shown above. In our experiment, each structure is read aloud by the system. The list user interface is a sequential list of the household items. The shallow hierarchical interface has a branching factor of five and the depth of the tree is two (see figure 1). The branching factor for the deep hierarchical interface is up to four, and the depth of the tree is four (see figure 2). We have implemented the design structures in VoiceXML using Voxeo Prophecy and IVR Junction. We have also logged user's interaction with the IVR system while performing tasks to review the actions user took while performing the tasks.
We recruited 18 participants (12 Male, 6 Female) to evaluate the three design structures. Participants were randomly divided into 3 groups where each group contained equal number of participants. Each participant performed five tasks, one after another. Each task consisted of selecting a target item from the assigned design structure. All participants searched for five target items (Shirt, Blouse, Rings, Football, and Plate) in the same order.
We have also analyzed the experiments and shared our analysis and recommendations in the paper.
This work is closely tied to Aditya's active area of research on IVR systems. The project was conceived out of his research. Given that Aditya has a significant amount of experience designing IVR systems, he also was responsible for coding the system.
Sam wrote the Graphviz files visualizing the hierarchies and conducted the majority of the user experiments. He constructed the summary figures using Tableau and managed the github submissions.
Transforming the experimental design from the GUI-based work to an IVR format was done by both Sam and Aditya. Both also contributed to the writeup.