This project grew out of a partnership my former company Dynamic Signal had with twitter and their “firehose” feature, providing access to millions of tweets in realtime. Our clients had large following lists already and were interested in specific topics related to their brand. Using selected topics and seed accounts, we looked at this intersection for insights from users, networks, tweets, and published links.
The basis for a ranking was seed accounts, keywords, and exclusion (tuning). A seed account could be a company account, group of employees, whatever was representative of the client or userbase. The keywords were topics and hashtags that made up an area of focus. The tuning consisted of what the ranking wasn’t - excluding accounts and keywords in order to refine the existing set.
Once we had a handle on creating rankings with account and keyword tuning, we looked at how lists of users would appear. Rankings would support ways of sorting, filtering, and search, along with related demographic information in a side panel. Selecting a user card would open a card on top of the dashboard where the specific user would have their ranked tweets shown.
Moving from users into sites and publishers, we setup tuning based on topical relevance to the keywords provided. There was also an audience component in how generalized or niche the content was. We also experimented with different visualizations for categories as they were often made up of many publishers of varying significance.
The rankings product was configured on a client by client basis and used as an add-on feature for major clients. While useful for understanding communities on twitter, it strayed from the eventual mission of internal employees and their engagement at work. An initial seed ranking also took up to 24 hr to complete and was performance heavy to run at the time. While we continued to refine and improve the rankings interface, it eventually was shelved in order to focus on products which fit the enterprise model more clearly.