Social media and the web have provided a foundation where users can easily access diverse information from around the world. However, over the years, various factors, such as user homophily (social network structure), and algorithmic filtering (e.g., news feeds and recommendations) have narrowed the content that a user consumes. Consequently, users on different ends of the ideological spectrum live in their own information bubbles, oblivious to the views on the other side and creating their own world-view of truth. This phenomenon has led to the polarization of viewpoints, intolerance towards opposing views, and ideological segregation. The goal of the project is to use machine learning techniques to mitigate the information filter bubblem.
This demo is complimentary to our WSDM paper, which tackles two main challenges: (i) learning the ideological latent factors from Twitter user-content bipartite graph; and (ii) embedding the discovered factors in a common latent space. Our pipeline is completely algorithmic, and no human judgements are involved in the process.
In this demo we demonstrate the utility of the joint latent space learning model in real-world scenarios. We illustrate how we can learn ideological latent space on Twitter, as well use these results to develop exploratory and interactive interfaces that can help users in diffusing their information filter bubble. The experience focuses on the ideological positioning of a user and the ideological positioning of the content consumed by her. There is a link between a user and a media source, if the user has interacted with the said media source. Size of a data point depicts the volume of interaction between user and the corresponding media source.
Motivated by the observation that (i) a user’s ideological stance on a topic depends on both his social network as well as the media sources that users consume their information from and (ii) A news media channel's ideological stance depends on the ideological stance of its audience. We exploited the inherent connection between the two data types. We developed a joint model that simultaneously explores both user-to-user and user-to-content relation in order to learn the ideological leaning of both Twitter user's as well as news media channels. More details about the methodology can be found in our WSDM 2018 paper .
This demo is built using the mpld3, an interactive D3 js based viewer for matplotlib.
For questions/comments, please contact Preethi Lahoti (plahoti(at)mpi-inf(dot)mpg(dot)de).