An algorithmic framework for reducing bias and polarization in online media

Online media is an important part of modern information society, offering a podium for public discourse and hosting the opinions of hundreds of millions of individuals. Online media are often credited for providing a technological means to break information barriers and promote diversity and democracy. In practice, however, the opposite effect is often observed: users tend to favor content that agrees with their existing world-view, get less exposure to conflicting viewpoints, and eventually create information silos and increased polarization. Arguably, without any kind of mediation, current social-media platforms gravitate towards a state in which net-citizens are constantly reinforcing their existing opinions.

In this project we will develop theoretical foundations and a concrete set of algorithmic techniques to address deficiencies in today’s online media. We will develop methods to discover structure and patterns of segregation, conflict, and closeness in social-media systems. We will address the issues of reducing bias and polarization, breaking information silos, and creating awareness of users to explore alternative viewpoints. We will also study the effect of different design features to the willingness of the users to explore viewpoints that conflict their opinion.

The project is structured along three intertwined research thrusts: knowledge discovery, exploration, and content recommendation. To accomplish its aims the project will formulate novel problem representations that provide a deeper understanding of the undesirable phenomena observed in online media and allow for effective remedial actions. Strong emphasis will be given on designing algorithms that are scalable to large data, are able to deal with uncertainty, and offer theoretical guarantees. The end result will be a set of new methods and tools that will contribute to increasing exposure to diverse ideas and improving online deliberation.

During the first half period of the REBOUND project we have pursued several different research directions and have made progress towards the objectives outlined in the project proposal. First, we have introduced novel problem formulations and developed efficient algorithms for discovering polarized structure in signed networks. We have designed efficient methods with quality guarantees that can be used to find conflicting communities in streaming data, or find local communities using only sublinear preprocessing and query time. Second, we considered diversity in data summarization, and introduced novel formulations for diversity-aware data summarization in clustering and ranking problems. We also considered the concept of diversity when information propagates in a social network, and proposed efficient methods to maximize novel diversity indices, or maximize the balance of exposure of different view points of a social debate. In addition, we studied opinion dynamics in social networks, designed methods to optimize different polarization and disagreement indices, and proposed a novel model to integrate information propagation with opinion formation. Finally, we studied edge-recommendation problems, and developed novel methods to minimize distances in a graph with a given number of edge additions. The results of the project were disseminated via publications and tutorials in top journals and conferences.

In the second half of the project we expect to improve upon the results we have obtained so far, either by resolving some of the open questions we have encountered, or by studying more general settings and problem formulations. We also plan to pursue further research on forthcoming project themes, namely, analysis of temporal networks, and design of recommendation methods for optimizing novel indices of polarization and diversity.

Periodic Reporting for period 2 - REBOUND (An algorithmic framework for reducing bias and polarization in online media)

Condividi questa pagina

Scarica