Once Upon a Vis
A study of top GoodReads books and development of visualizations with the intent of showing recommendations. Used task analysis, created personas, conducted user interviews, ran user testing, and concluded our work with a final research paper.
The setup
Despite the availability of e-readers such as Kindles, reading has become less and less common. One way this
shift can
be reversed is by giving people resources to find new books. The general area of interest for our project is
the ratings
and rankings of books, and how they compare to one another. We intend to address the comparison of popular
books and
allow book readers to get a sense of similarities and differences between books. This will potentially give
them new
books to read based on ones they have enjoyed in the past. We believe this visualization will address an
important topic
as reading improves one's focus,memory, empathy, and communication skills.
A human-centered interpretation of data from the top 100 GoodReads books. Learning based in design
principles. Data cleaning and exploration in Jupyter notebooks.
Data Collection & Exploration
Data was collected from GoodReads, cleaned, merged, and explored. When a sentiment analysis was carried out on the dataset using VADER, some notable statistics we found in our initial review of this dataset include the mean rating being 3.7/5 and the mean Vader neutral score being -551.95. We found that some of the best performing books in terms of financial success are also some of the worst performing in terms of reviews.
Task Analysis & User Personas
Users may want to know; which genres are highest rated, the general public's rating of a specific popular book, find the most popular book in a series, or compare two books based off of their statistics and public opinion. Secondary users such as the publishing industry may be interested in the trends. Our tasks were consume and discover based. The search tasks involved browse, locate, and lookup; while analytic tasks included summarize, identify, and compare.
Visualization & Interaction
We planned a linked bubble chart on 1- and 5-star reviews, sized by number of reviews. Our intent for a review bar chart included, bars showing the number of 1-through-5 star ratings, colored green/gray/red to correspond to the percent positive/neutral/negative reviews for each star rating. Data obtained from sentiment analysis. Another graph on book sales over the years includes brushing and linking between the points highlighted and corresponding genres.
Results
We created a site with four main visuals that best captured the ideas and intentions of our group. We may have overestimated our abilities to work quickly in a JavaScript tool that was new to all of us, but the content we created still felt like a representation of our planning and user testing results. Much of the success of our presentation came from from having clear vision for future improvements and outlining a realistic path toward achieving them.
Reflections
The most impactful part of this project was the ending. After spending so much time on initial planning and testing, the time required to code in D3.js how much we were able to accomplish. Having to be so purposeful with our design decisions taught us a lot about what we could add to improve the site's interactivity. I think rebuilding the site with different tools could be rewarding and could significantly reduce development time.
If I took time to extend or redo this project, I would focus on refining the current visuals by adding responsive components. We never got to incorporate the sentiment analysis data, and the tiered bar charts we intended to add could provide a great deal of information with relatively minimal changes to the existing design.
