Disclaimer: This is a simple proof of concept that still contains some bugs, but can be used nonetheless. Please have patience.
Please check out the [slide deck] (http://skallinen.github.io/assignment_deck) about this project made as part of a Coursera course on Developing Data Products.
Let's you explore data about how different content items are shared on social media. The data is collected in real time from Twitter, and made available for the app in near real-time. You can use different filters and explore different aspects of the trending item. Filters include the following:
- Defining minimum and maximum values for item share count.
- Defining share velocity, ie. how many times the url has been shared per hour.
- Defining time period by setting how many hours will be displayed.
- Selecting sources. Narrow down the selection to certain predefined types of tweeters.
- From the narrowed down selection you can pick one item and isolate its trend plus forecast how it will be shared in the future.
Step 1
Use the filters to find a batch of items that you are interested in. By toying with the filters you can isolate content items that are extremely popular or that are bubbling under, about to hit big. Use the "Forecast individual items" list to examine a particular item more closely.
Pro tip:
If you want to have finer controls in the filters you can use the keyboard arrow buttons on the slider controls.
Please note that the legend listing the titles of the urls is shown only when they can fit the screen, i.e. when you have narrowed down the selection to 20 or less urls.
Step 2
Pring out a list of the selected items at the bottom including link, short description and image, when availalbe. You can do this by clicking the "Review Selected Content" button below. Loading the content might take a moment.
The forecasting feature is experimental. We are using exponential smoothing from the forecast R-package with all settings on automatic. We are plotting both observed and fitted values, as well as the calculated forecast including both 95% and 80% confidence intervals.
The data is collected by a different back end. The backend currently follows three different Twitter lists. Collects all the url's that have been mentioned in the list and starts tracking their share data. Currently the source lists are:
- [The best mindcasters I know] (https://twitter.com/jayrosen_nyu/lists/best-mindcasters-i-know)-twitter list by Jay Rosen is tha source for the "Journalism and Politics" selection in the app.
- [Tech News Brands] (https://twitter.com/Scobleizer/lists/tech-news-brands)-twitter list by Robert Scoble is the source for the "Tech Media" selection.
- [Data Etcetera] (https://twitter.com/sakalli/lists/dada-etcetera)-list by Sami Kallinen is the source for the "CS & Data" selection.
Have fun!
- Relplace ggplot as plot library to something where user can use explore lines using the mouse.
- Fix "argument is of length zero" bug.
- Fix Top10 visually.
- Enable user to explore the content item (like in top10) directly from the forecast.
For bugs, comments and questions contact notjustsilicon-at-gmail-dot-com