Image Circulation of Russian State-Controlled and Independent Media on Google.com and Yandex.ru

Team Members

Alberto Federico Olivieri, Alice Noris, Giovanni Daniele Starita, Richard Rogers

Contents

Summary of Key Findings

1. Russian-language, state-controlled and Russian-language, independent media seem, at least in our sample, to offer "muffled" and less dramatic images of the war/special operation on their social media pages than those of western media.
2. Yandex seems to filter images published by Russian-language, independent media. On Yandex the number of images from independent media is significantly lower than the ones that have been published by state-controlled. The finding is buttressed by the fact that the images from Russian-language, independent media do circulate on Google Images.
3. Google and Yandex seem to act as two separate ecologies when it comes to image circulation of the Russian-Ukrainian war/special operation.

1. Introduction

The aim of this project is to understand how images extracted from Russian-language, state-controlled and Russian-language, independent media compare as well as circulate on the search engines Yandex.ru and Google.com, using each engine's reverse image search. The aim to provide insights into the type of imagery deployed in the selected media sources as well as their circulation ecologies.

2. Research Questions

The research aims to answer the following questions:

1. To what extent do state-controlled, Russian-language media and independent, Russian-language media have separate kinds of images that depict the war/special operation in Ukraine?
2. Which images circulate on Yandex and do not circulate on Google, and vice versa? (presence and absence)
3. Do state-controlled images have a particular network of sources where they are published distinctive to that of independent media?
4. Are Google and/or Yandex suppressing particular images (or their sources) of the war (or special military operation)? To what extent?

3. Methodology and initial datasets

The methodology carried out to conduct the research is mostly quantitative, with the addition of Open Source Intelligence Techniques (OSINT). Our first step is to select three Russian-language state-controlled and four independent media sources. In order to choose the images under analysus, we download daily the top images (by engagement score) from their VK and Instagram accounts. Subsequently, each image has been loaded in reverse image search: Google.com from the Netherlands and Yandex.ru from Russia - with a VPN located in Moscow (CyberGhost VPN has been utilized). To make sure that our own machines could not influence the outcome of the reverse image search we used the incognito mode present in Google Browser. (For further studies we suggest using a clean virtual machine that will add more layer of protection from tracking cookies.) Once each image was loaded, the presence and absence of images per engine was determined.

A spreadsheet with the following categories “Media name”, “Image name”, “Presence on Google”, “Google Link Number”, “Links from Google”, “Presence on Yandex”, “Yandex Link Number”, “Links from Yandex” was created. Afterwards, the outputs have been presented through a series of graphs that show respectively Google and Yandex Link Number per Media Outlet and the aggregated data of State Controlled vs Independent media. The outputs have been also represented through a Bi-partite network of sources per image set.

We also performed a network analysis of the sources, displaying a bi-partite graph showing the media ecologies of state-controlled and independent media according to their circulation on Google and Yandex.

4. Findings

The research has led to three main results. The first refers to the image semantics of Russian-language state-controlled and independent media, which seem, in our sample, to offer a "muffled" and less dramatic images of the war/special military operation on their social pages than their western counterparts (which we however did not study).

Fig. 1 visually represents the presence/absence of images on Google.com and Yandex.ru. It appears that State Controlled media seems to present the “special military operation” showing the power of Russian Federation through a representation of armaments, military vehicles and soldiers while Independent media seems more willing to represent the destruction of buildings and cities and the desperation of civilians, though to an extent that appears tamer than in western sources.

Our second finding involves the filtering of images, which seems to be performed by Yandex.ru around the images retrieved from the independent media. In fact, the number of images returned from those outlets is significantly lower on Yandex.ru than it is on Google.com.

Fig. 2 presents the temporal evolution of the presence/absence of the images retrieved by each media outlet on Google.com and Yandex.com in the following timeframe 16-22 March 2022. Green cells show the presence of each image, red cells the absence, while yellow cells with the wording ‘no relevant image found for the day’ show that no images could be found on IG with the predefined criteria determined for the analysis. As is shown, state- controlled and independent media provide different images, and seems to represent two different ecologies not only in terms of the war/special military operation but also in terms of image circulation on Google.com and Yandex.ru.

Our third and last finding revolves around the ecologies of the selected images, their circulation sphere, determined by the number of common domains. Google and Yandex seem to act as two separate ecologies when it comes to image circulation of the Russian-Ukrainian war/special operation. In order to demonstrate the separate source ecologies, we extracted the domains from the links of the images retrieved from the reverse image search, and compared how much they share common ground online.

Fig. 3, and 4 present respectively the network of Google and Yandex in relation to their domains. The second representation also shows the distribution of domains related to state-controlled and independent media.

Regarding the first research question, we found that state-controlled and independent media do not share common image spaces. No identical images of the conflict were found during the seven days under consideration.

Concerning the second question, it can be stated that all the images of the state-controlled media circulate on Google.com, while three images of the state-controlled media are not present on Yandex.ru. Moreover, 4 images of the Independent media do not circulate on Google.com and 14 images do not circulate on Yandex.ru.

Fig. 5 and 6 show respectively the number of Yandex and Google links obtained by uploading on Google reverse image search and Yandex reverse image search each image retrieved from state-controlled media (Argumenty i Fakty, Kosmomol’skaja Pravda, Ria Novosti) and from Independent media (BBC Russia, Meduza, Novaya Gazeta, Radio Svaboda).

Fig. 7 presents the aggregated data of the link retrieved on Google.com vs Yandex.ru divided according to state-controlled vs independent media. The graph suggests that Yandex.ru provides fewer results for independent media in terms of image circulation and in terms of links retrieved from the image search.

Regarding the third research question, fig. 1 (presented above) shows which images circulated in Yandex and not in Google, and vice versa and fig 3-4 shows the two distinctive networks of Google.com and Yandex.ru.

Finally, with regard to the last research question concerning possible suppression of image circulation by Yandex/Google, the study seems to suggest that Yandex.ru filters image circulation more than Google.com (see fig. 2), but further work is needed.

5.1. Limitations

Our preliminary research project has some inherent limitations that must be taken into account for a complete understanding of the results. Starting with the technical constraints posed by the tools that we have used during this research, we have noticed how a few images with superimposed text did not return accurate results on Google Images and Yandex. The platforms we used (Instagram and VK) had their own image format (adding a box with text under the image, or slightly superimposed) and some images did not return results on the reverse image search. We suppose the algorithm is too strict and tries to return the image with the specific platform format and is unable to recognise the visual representation of the image when the textual representation is highly preeminent.

Our data collection of conflict-related images on independent media was influenced by the limited number of images directly related to the war that they publish in comparison to the ones published by state-controlled media. Moreover, the preliminary study was carried out considering only 40 images (21 from State Controlled and 19 from Independent media), a sample that is small.

5.2. Future Studies

Given the issues we just exposed, future studies should consider adding Google.com reverse image search to the sample related to Russia, and adding Yandex.com reverse image search to the sample related to the Netherlands. This procedure would be useful to determine if there are some unseen biases in the search engine that we have not considered during our work (and allow us to gather insights on that part of the Russian population that uses Google Images). We should also develop an in-depth qualitative analysis, employing in a more significant manner OSINT techniques. The sample size issue could be addressed by increasing the number of images collected (a sample of 706 would provide 94% confidence interval with ± 5% margin of error). In order to achieve such a figure, the number of media outlets could be expanded, and using images with superimposed text should be avoided.

6. Eternal Links

Table for Figure 1

Table for Figure 2

Raw Data Table

Raw Data Presentation

7. References

Babbie, E. R. (2011). Introduction to social research. Wadsworth Cengage learning.

Bazzell, M. (2016). Open source intelligence techniques: resources for searching and analyzing online information. IntelTechniques.com

Carlo de Gaetano, (2019). https://digitalsocietyschool.org/wp/wp-content/uploads/2020/09/1907_ESIEA_Gephi-1.pdf

Euronews (2002). https://www.euronews.com/2022/03/04/russia-restricts-access-to-foreign-and-independent-media-including-local-bbc

Calvert Journal (2014). https://www.calvertjournal.com/features/show/2228/russian-media-guide-to-the-troubled-world-of-independent-journalis

Field, A., Miles, J., & Field, Z. (2012). Discovering statistics using R. Sage publications.

8. Addendum: Bucha

Given the ongoing situation in Bucha, we have decided to utilize the methodology presented above to provide a brief overview on how the images selected from the same independent and state controlled media are shown on Google.com and Yandex.ru reverse image search. The research has been conducted on the 6th of April, with April 5 as a point of reference to retrieve the images. The table below shows the presence/absence of the selected images on Google.com and Yandex.ru.

Media	Image Name	Source	Presence on Google	Google Link Number	Presence on Yandex	Yandex Link Number
Independent	IND_bbcrussia_bucha	bbc_russia	1	155	0	0
Independent	IND_meduza_bucha	meduza	1	176	0	0
Independent	IND_radiosvaboda_bucha	radio_svaboda	1	193	0	0
state_controlled	SC_argumenty_bucha	argumenty	1	93	0	0
state_controlled	SC_kosmo_bucha	kosmo	1	135	0	0
state_controlled	SC_ria_novosti_bucha	ria_novosti	1	150	0	0
				SC Total	524
				IND Total	378

As you can see from the table, no images were returned by Yandex.ru in our searches, but we should consider that Yandex might take longer to display the images compared to Google. At the moment we can not state if this outcome is the result of a deliberate choice.

From a qualitative perspective we must note that the state controlled media mainly utilize images from Reuters writing about Bucha. These images might not even have been taken in Bucha. Most importantly, these same images have also been used in relation to other event coverages from the same State Controlled media. This choice could impact the Google results for state controlled media since the results from the search do not represent that specific event.

Another aspect to be noted is that this time images used by the Independent media, in particular Meduza, seems to present the same type of images than the Western ones, while State Controlled media keep a more “muffled” representation.

Finally, the ratio of the returns between State Controlled and Independent is much higher than the one noticed for the main research, since both sample sizes are small and not statistically significant, we cannot confirm if this is a statistical anomaly caused by the uniqueness of this event or if it is inside the margin of error.

Figg. 1, 2, 3 IND_bbcrussia_bucha, IND_meduza_bucha, IND_radiosvaboda_bucha

Figg. 4, 5, 6 SC_argumenty_bucha, SC_kosmo_bucha, SC_ria_novosti_bucha