Narratives of Conflict: What the 2014 Gaza War Can Tell Us About Discourse on the Internet.

Co-authored with Sands Fish

This essay first appeared in the Internet Monitor project’s second annual
Internet Monitor 2014: Reflections on the Digital World. The report, published by the Berkman Center for Internet & Society, is a collection of roughly three dozen short contributions that highlight and discuss some of the most compelling events and trends in the digitally networked environment over the past year.

This summer, the war in Gaza and Israel dominated coverage on many news sites. The war lasted for 50 days and resulted in over 2,000 deaths (mostly civilians), thousands of injuries, and overwhelming destruction. Media reports varied in their coverage, with the stories spanning from Hamas’ rockets being fired into Israel to the toll this war has had on Palestinian children, the destruction of UNRWA schools and more. Using digital media analysis platform Media Cloud we tracked media coverage of the war to gain a better understanding of the different topics that emerged and the frames that were adopted by different media outlets.

Through this analysis we are able to draw a map of the media landscape, highlighting the relationship between different media sources based on the similarities and differences in the language that they adopt. We are able to identify five meaningfully distinct clusters: one cluster includes large media sources from the US and UK that align with Israeli narratives, while another cluster of western media hews more closely with Palestinian perspectives. Three other clusters are comprised of the Israeli military, human rights organizations, and relief organizations.

For our analysis, we drew on a corpus of 80,000 stories that were collected for this study. From this corpus of stories, we focused our attention on the stories published by the 50 most prominent English language sites, as measured by the number of links from other media sources. We employed a network structure that clusters media sources by the content of their stories and creates a visual representation of the media landscape in which influential media sources are connected to one another based on the words that they use most frequently within the bounds of a given topic. Using network theory, we constructed and analyzed the ways in which media sources are affiliated via shared vocabulary. We drew a network map based on the 100 most frequently used words from each media source. A link was created from a media source node to a target node representing the word the media source used frequently, appropriately weighting the link based on the frequency of usage. We then used a community detection algorithm, allowing us to see patterns of term usage in the text of online discourse. Media sources, in this way, are connected via shared language and can be laid out based on their connective strength. The resulting map, in this case, generated five distinct community clusters.

Image by Sands Fish

Focused on Gaza, this network map revealed communities of media sources and vocabulary that, when analyzed, uncovered a number of frames deployed in the overall coverage of the war.

We were able to discern that, while the average reader might assume there is simply one Israeli frame and one Palestinian frame, the media landscape is much more complex.

Upon further investigating the communities, one can see what makes each of them unique is based on the media source’s frequent use of certain terms. The fact that Relief Web (a UN affiliated site) is one of the only media sources to use ambulances frequently in their coverage describes the predominant narrative this media source focuses on. Assertions can also be made about media source similarity within and across these communities by analyzing the co-occurrence of word usage in pairs or clusters of media sources. An illustrative example of this is the use of distinct terms such as breaches, crimes and disproportionate linked to Amnesty International and Human Rights Watch. These terms help define the “Human Rights” community.

Two of the clusters (Palestinian Focus and Israel Focus in the figure above) are primarily comprised of mainstream media sources. The split in these two clusters is evidently based on the affiliation with Israeli or Palestinian sources and exposes a slant in the coverage of the war associated with different media sources. The Palestinian Focus cluster contains both Maan News, a Palestinian mainstream media source, and Mondoweiss, a pro-Palestinian left-leaning American Jewish site, in addition to sites such as Al Jazeera and the UK’s The Independent. The language used more frequently by these sites include words such as evacuate, children, and casualties, terms that focus on the toll of the war. The Israel Focus cluster includes Israeli media sites such as Jerusalem Post and the National News of Israel, along with CNN and the US right-wing site Breitbart. The media sources in this cluster more frequently use terms such as tunnels, sirens, and arms, terms indicating a greater focus on Hamas’ actions against Israel. Both clusters are comprised primarily of US and UK mainstream media sites and use terms common across all coverage of the conflict; however, the divide between clusters indicates a slight slant towards either perspective.

Two topical clusters revolve around humanitarian and human rights issues, respectively. The Humanitarian cluster includes three sources: the United Nations, UNRWA (UN Palestinian Refugee Agency), and Gisha, an Israeli legal center for freedom of movement. These sources used words such as psychosocial, hygiene, sanitation, truckloads, and shortage. This cluster is distinct from the Human Rights cluster that includes Amnesty International, Human Rights Watch, and B’Tselem, an Israeli human rights organization. This cluster more frequently uses a separate set of terms, including breaches, detainees, shrapnel, and violations, that are more commonly used in the human rights community.

In addition to the Israel Focus cluster, there is also a separate cluster comprised of Israeli military and Israeli mainstream media sources. This cluster is dominated by the IDF (Israeli military) blog and spokesperson site alongside news sites such as Ynet and Times of Israel. The terms used more commonly in this cluster, such as soldiers, battlefield, and brigade, carry military connotations.

Notably absent from the map is a cluster comprised primarily of Palestinian media sources.

This stems in large part from the relatively small number of Palestinian sites that were represented in the top 50 media sites contributing to coverage of this issue; three pro-Palestinian sites compared to eight Israeli sites. The Palestinian media sites that are included in the network map are included because they drew attention from the Western mainstream media, as demonstrated by the number of inlinks to these sites

Using texts from thousands of online stories, we were able to detect slants in mainstream media’s coverage toward Israeli and Palestinian narratives. The Gaza war provided an ideal case study for better understanding how to analyze discourse online. These are initial findings resulting from the exploration of both story and tool. Moving forward, we plan to increase the robustness of the visualization and our confidence in its representation of substantive discursive trends. With the growth of analytical tools and methods such as this, the insight that can be gained from study of the large-scale text sourced from the networked public sphere will only expand. The opportunity for qualitative analysis of discourse online using automated quantitative tools provides a deeper view into human communication than has ever been available.