The Brain Systems, Connections, Associations, and Network Relationships (a phrase with more words than strictly necessary in order to bootstrap a good acronym) assumes that somewhere in all the chaos and noise of the more than 20 million papers on PubMed, there must be some order and rationality.

To that end, we have created a dictionary of hundreds of brain region names, cognitive and behavioral functions, and diseases (and their synonyms!) to find how often any two phrases co-occur in the scientific literature. We assume that the more often two terms occur together (at the exclusion of those words by themselves, without each other), the more likely they are to be associated.

We began by creating a list of ~150 neuroscience terms identified as the broadest, most common search terms that are also relatively unique (and thus likely not to lead to spurious connections). Using Google App Engine and distributed processing via the "task queue" and asynchronous JavaScript http requests to the ESearch utility provided by PubMed we calculated three key numbers based on the co-occurrance of each term to each other term (and their synonyms) in the title and abstract of publications: the "conjunction", "term a AND term b", the exclusive disjunction "term a NOT term b", and the probability of association the conjunction divided by the exclusive disjunction. In this way we were able to leverage Google's massive computing power to quickly, accurately, and very cheaply (we spend about $11 developing the first prototype) model these connections.

Progress is ongoing. We are continuously collecting suggestions from the more than 6,000 visitors to the site thus far. We have begun synthesizing and implementing these suggestions iteratively. Our "Beta 2" version (released about 2 weeks after the initial launch) included improvements to the search interface allowing for case-insensitive search, and the ability for visitors to search for common acronyms (such as "MEG" for "magnetoencephalography") that should not be used in the association calculations, better use of asynchronous server requests to provide feedback to users when the page loads slowly, and an informative key listing the association of color to term category and association type.

Future work includes additional features requested by visitors: an expanded dictionary of terms, the ability to add terms on-the-fly, the ability to filter the network graph based on term category type of strength of probable association, additional interactive data visualizations that would researchers to get a different view of the material (a tree diagram with descendants and ancestors rather than a circular diagram representing the network as a whole). We are also submitting an article for peer review that describes this project and what we believe is a novel method for determining association between terms in neuroscience.

Cookie Booth Finder

GSUSA, the governing body of all Girl Scout Councils in the US prohibits the sale of Girl Scout cookies online. This posed a problem for councils catering to a market that expects to find anything and everything online. The online Cookie Booth Locator allowed troops to advertise their booth sales online and allowed visitors to find the sales when and where they were happening. In 2008 and 2009 this Cookie Booth Finder attracted over 20,000 unique visitors.

Graphic Design

In the Fall of 2008 the Girl Scouts of San Francisco Bay Area merged with 4 other area Girl Scout councils creating the Girl Scouts of Northern California. By Spring of 2009 the once disparate councils were becoming more cohesive and a new website that served to focus the new council was desired. The goals of the design was to create a fresh, bright, informative website that highlighted the variety of activities available to Girl Scouts in this new council, and to present a cohesive face without compromising the individual characteristics of the legacy council areas. The screen shot above highlights these features on the homepage in the Spring of 2010.

Online Volunteer Training

In order to facilitate the creation of troops I developed an online training system in February of 2008. Through cooperation with our adult development department I was able to implement one of the first online training systems for Girl Scouts in the country. The system tracks participants’ progress, integrates with our membership database to record participation, and allows volunteers to be trained at any time on any day. It also incorporates a custom content management system that allows non-technical staff members to update and create new trainings. In contrast to Flash-based systems, volunteers do not need to install any proprietary software to view the basic trainings. This also ensure that file size stay very small to accomodate our dial-up volunteers.

Volunteers have flocked to the website with over 10,000 trainings recorded. My system also addressed a need in rural areas where in-person training was either inaccessible due to the distances to the training site, or to the infrequency of training sessions. A significan percentage of trainings done in rural counties are now done online. As a result, new troops can be created in a matter of days instead of weeks.

Content Management System

In 2009 I created a custom content management system to allow non-technical staff members to update pages on our website. The system uses tinyMCE to provide staff members with MS Word-like formatting tools for editing the content of certain portions of web pages. The system integrates with a custom authentication system to restrict access to certain pages to certain staff members. Staff members are able to browse our website as normal and when they find a page they want to edit, login and edit the page directly in the browser window. There is no need for them to learn any HTML or how to use FTP or SSH. They can see the changes they're making to the page in real time, save and edit pages without making their changes "live" until they're completely satisfied, or their managers approve the changes.

System Description

MUSE is a mobile application that encourages visitors to interact with art museum collections in ways that are not yet widely available. It is an affordable way for museum staff to easily create, organize, and publish authoritative content to augment physical museum artifacts. In addition, visitors can create their own content and “attach” it to artifacts. Users can save and comment on artifacts, review other visitors’ comments, explore supplementary material associated with an artifact, and leave feedback for museum staff.

In contrast to traditional audio guides, MUSE re-imagines the museum guide in a way that emphasizes timely and alternative content, rather than production value. Visitors and curators alike can converse through their interpretations of the art pieces in dynamic and interesting ways. MUSE blurs the line between curator and consumer in a way that will be beneficial to the visitor as well as the museum. Constantly changing, timely, and interesting content will be available to augment the museum experience for visitors, which will in turn attract new and repeat visitors alike.

Problems and Solutions

Interactive museum tours are an increasingly necessary addition to physical collections. Enriching a museum’s physical offerings with multimedia tools may increase consumption by a broader audience. While not a replacement for traditional pedagogical content, interactive tours can augment existing content with novel and memorable interactions.

Traditional audio guides are expensive and time-consuming to create. As a result the guides must remain relevant for a long time. Other “do-it-yourself” approaches, including downloadable podcasts, have received lukewarm reviews because they are difficult to use.

The resulting collection of museum audio guides is diminutive and static. Moreover, the feature set common to these guides is predictable. A typical device consists of a large blocky pendant (hard to steal or forget) worn around the neck. Interaction with the device is limited to typing in code numbers and listening to the recorded content. These devices are only available at large museums and tend to play the same content about the same pieces for years. They do little to stimulate interest in new collections or foster timely interpretation of pieces.

MUSE allows museum staff to create and constantly update the information they provide about a piece or collection. It allows them to create timely “tours” through a collection that can be used to encourage repeat visitors. It also encourages visitors to leave their mark on a piece (record an audio comment), share it with their friends, and interact with the piece outside of the museum.

The working prototype (pictured) allowed users to post 4 specific emotions using what we determined were the “most natural” gestures for that emotion. Emlo can also receive messages. When Emlo receives a message that it can understand from a user’s friend, it will buzz and light up in the color corresponding to that friend’s emotional message. To determine a set of natural gestures we began by choosing a set of emotions. Robert Plutchik’s “Wheel of Emotions” (1980) provided a structure for our inquiries. The Wheel describes 8 cardinal emotions that when combined in varying strengths (according the Plutchik) describe the all possible human emotions. Using these 8 cardinal emotions allowed us to focus on a subset of all possible human emotions.

Testing

We performed two phases of user testing. The first was open ended, meant to narrow down the set of all possible gestures to certain gestures that may be more natural and could be tested further in phase 2. In phase one we asked our participants to model the 8 cardinal emotions using one or more of 4 household objects (a ping pong ball, a hair band, a hair roller, and slightly wet sponge). Each of the objects had different physical affordances, the ping pong ball lent itself to being tossed or bounced, the hair band was soft to the touch, and stretched and bent, the hair roller was spiky, and the sponge was soft and to certain participants a bit unappealing. By comparing the participants’ responses we were able to devise 10 possible emotion-gesture pairs for further study.

In phase 2 we showed participants the following video describing the 10 possible gestures and then had them match each gesture to one of the 8 cardinal emotions (there were two gestures left unmatched). The result was significant agreement on 4 emotion-gesture pairs. We then built these 4 emotion-gesture pairs into our final working prototype.

Prototypes

We created several non-functional form-factor prototypes out of clay and one functional but over-sized prototype. The functional prototype was able to express joy, tossing it in the air, anger, hitting down on something, sadness, bending the “arms” down, and surprise, covering and uncovering the “face.” It became very anthropomorphic throughout the process acquiring a “face” the place where the photo-diode existed, “arms” one of which contained a flex-sensor, and hair which was more a vestigial organ from our first phase of testing. In the first phase of testing, the hair roller was used most often to describe “fear.” We deduced that it might have something to do with its odd texture, this connection between spiky texture and “fear” proved inconclusive in phase two, but we decided to keep the hair to add some texture.

Mobile User Interface Development

We paid particular attention to the challenges and opportunities of developing for the mobile environment. Of particular concern were mobile users’ short attention spans, frequent distractions, and small screen real estate. Certain opportunities present themselves as well, including the possibility of automatically attaching metadata such as time and location to notes and highlights and using the phone’s built-in microphone to allow for voice notes as well as text notes.

Prototype Refinement

We began with an information discovery phase or “contextual inquiry”. We observed target users in context (a coffee shop, their shared living space, an office, on public transportation) interacting with their preferred annotation devices (usually paper, a pen, and a highlighter). By studying these users in real world environments we were able to better inform our design. For instance, although all of our users had heard of people using multiple colors of highlighter for different types of notes, none actually used multiple colors. They preferred to focus on the reading rather than deciding which color highlighter to use.

Next we conducted actual user testing with a low-fidelity paper prototype which confirmed out suspicion that a finger-drag action for highlighting content is most intuitive for users. Following the paper prototyping phase we created a working mock-up of the system and conducted more usability tests. Please view our final process book for more information about the final product.