Research ArticleSOCIAL SCIENCES

The memory remains: Understanding collective memory in the digital age

+ See all authors and affiliations

Science Advances  05 Apr 2017:
Vol. 3, no. 4, e1602368
DOI: 10.1126/sciadv.1602368

Figures

  • Fig. 1 View flow.

    Left: Daily Wikipedia article view count on a logarithmic scale for the Wikipedia articles representing Germanwings Flight 9525 (source) and American Airlines Flight 587 (target). The colored area measures the increase in views relative to the daily average of the previous year (dashed line). Right: View flow from 98 sources (20082016) to all 123 target events from the period 20002007. The color of the pixels shows the strength of the view flow on a logarithmic scale. Both axes are sorted according to the date of the accident such that going down or going right brings the reader to more recent events. Some source events, like Germanwings Flight 9525 (see pointer), trigger a lot of target events. We also point to the articles for the 9/11 crashes, which are triggered often and always in unity.

  • Fig. 2 Triggering factors for view flow.

    Left: Average view flow among pairs belonging to different groups according to different factors. The black bar labeled “All” includes all pairs, the bars labeled “Recent” and “Old” split the source-target pairs into those that are separated by more or less than 29 years (the median separation between pairs). The bars “Many deaths” and “Few deaths” split the pairs according to the number of deaths of the target event (at the median value of 22 deaths). The next two bars split the pairs according to the prior activity of the target article. The bins in dark gray area are based on whether the source and target flights were operated by companies located on the same continent, whether the operating company is located in Europe, Australia, or North America (Western), whether the source and target articles belong to the same article categories, and whether there is a direct hyperlink from the source article to the target article. Right: The same as in the left panel, but pairs with a hyperlink from source to target have been removed from the sample.

  • Fig. 3 Detailed analysis of triggering factors for view flow.

    Left: Average view flow against the separation in years between source and target event. Center: Average view flow against the number of deaths involved in the target event. Right: Average daily views of the target article during the year before the source event. A power law fit with an exponent of 1.23 is also shown.

  • Fig. 4 Filtering.

    Left: View curves of all articles from 2008 to 2016. The vertical axis measures the average views during the 1-year sampling period. The horizontal axis determines the period used for sampling. Specifically, it determines the days of separation between the incident and the beginning of the sampling period. Note that the computation relaxes after approximately 1 year of cooling. We therefore require that sources and targets are separated by at least 2 years because this ensures that the target has relaxed before sampling its average view rate. Right: We show the average view flow from each source against the views of that source during the same period. The dark dots represent the 11 largest source events, which were used in Triggering factors: Malaysia Airlines flight 370, Malaysia Airlines flight 17, Air France flight 447, Germanwings flight 9525, 2010 Polish Air Force Tu-154 crash, Indonesia AirAsia flight 8501, Asiana Airlines flight 214, 2011 Lokomotiv Yaroslavl air disaster, Metrojet flight 9268, and Colgan Air Flight 3407. The smaller source events (light blue) have not been included because noise dominates in this region.

Tables

  • Table 1 Model parameters.

    Least square fit of the parameters in the model to the data. Error bars are estimated using bootstrapping.

    ahistoryalinkadeathsayears
    0.83 ± 0.040.05 ± 0.023.3 × 10−9 ± 2.2 × 10−9−2.0 × 10−8 ± –1.5 × 10−8
    acategoryalinkeda0
    0.0 × 10−7 ± 8.7 × 108.2 × 10−6 ± 3.9 × 10−61.5 × 10−6 ± 0.8 × 10−6