Open Call Winner at Strata 2013
In September, we offered the opportunity to win free passes to the 2013 Strata Conference in New York by submitting work to our gallery on data, connectivity, and society. Here, Sepand Ansari, one of the winners for his project Wikistalker, talks about his experience and insights he discovered at the conference.
Having a background in computer science and a long-time interest in machine learning, I have always been very interested to see how breakthroughs in data mining infrastructures would change the methods of data mining and visualization. With this in mind, I submitted my project Wikistalker, which visualizes the semantic relevance of different Wikipedia articles, to the open call on Visualizing for passes to Strata 2013 and won. In addition to the more technical and basic workshops and talks on data mining, a section of the conference was dedicated to data visualization. Here are notes from some of the sessions that I attended.
Sean Kandel, an alumni of Stanford Data Visualization Lab (the same lab that gave birth to protovis.js), did a presentation on “Interactive Data Visualization of Big Data” (slides here). Suppose for a visualization task you have millions or billions of records and assume that each record has 2 properties, so you would need a scatter plot (2 properties could be latitude and longitude of a geolocation). In such a case, representing each record with a point, knowing that the number of points is more than the number of pixels of the visualization canvas, is not very useful since all the points will cover a large area of the scatter plot, if not all of it. In the best case, this approach will create a shade of a color. The challenge here is to find the best ‘visual encoding’ to communicate the structure of data. Kandel proposed three solutions: sampling, binning and modeling.
Naomi Robbins, a statistician talked about best practices in visualizing data from the perspective of the cognitive capabilities of the viewers. In a section of her talk Naomi emphasized avoiding bar charts because human perception is stronger in comparing lengths than angles, especially when angles are oriented toward different directions, as in a pie chart. Instead she introduced diverging stacked bar charts as an alternative for multiple pie charts.
Noah Illinsky, a data visualization researcher at IBM and the co-author of Designing Data Visualizations also did a very insightful presentation on clarity in data visualizations. Talking about the importance of using stories in data visualization and comparing the objectivity of data with the subjective aspect of story, he made a very great point that it is the story that connects data together and makes a coherent totality out of it and without it data is nothing but some numbers.
By Sepand Ansari