Engagement and Performance Studies of Media Agencies Publications on Social Networks

This paper presents a study that contributes to the existing work on the social diffusion and interaction strategy in social media. The aim is to know the most shared post by some electronic media in the world from end to end social network, and also to know post nature of the most successful one, and the link between different kind of interaction these are main objectives of this study. Our work is also considered as a ground and a base for social network analysis researchers in all social networks in order to allow them to benefit and help in their future research work from all information collected and results found via this study. An empirical analysis using multiple methods is conducted based on 275 Facebook publications gathered from the Facebook pages of 5 electronics journals the best one in its original country represented 5 countries in the world. This contribution discovered a set of important information and it is also projected to confirm hypothesis addressed in pre-existing studies. Keywords—Social network, Posts, Facebook, Interaction.


Introduction
Since the advent of web 2.0, internet users have taken control of all the uses, tools and features of the web that now allow them to contribute easily to promote exchange of information and make interaction (discuss, share , exchange, etc.) in a simple and easy way, both with the content published on the internet but also between them, thus creating a new web said social media or social network. On the other hand, social networks cover different activities that integrate technology, social interaction (between individuals or groups of individuals), and creation of content. Social networks consider themselves as a group of online applications (blogs, micro blogs, sharing sites, social networks, etc.) that are based on Web technology, we talk here about participatory web, social web and collective intelligence [1] and allow creating and exchange the content generated by users.
This web evolution has made it possible to build different types of large social networks, which are now recognized as an important means to share information [2]. Social network is a social media that concern different social relationships between individuals and the way that they are structured; these different relationships help to understand the behavior of individuals [11]. Thus, it is represented by a well-defined structure Experimental Methodology Facebook is an online social network that allows its users to publish "status", "photo", "video" or "link", and to share "files" or "documents" in groups that it offers. It is also a space for exchanging messages, to join and create pages and to use the set of applications circulating in this famous social network.
For its posts publishing strategy, it uses the algorithm called Edge Rank which aims is to choose and to order the content appearing on the news feed of users [10], the principle of this algorithm is very simple: for each user and each content, the goal is to define a score of relevance, which will classify the different messages and display the best. The score is calculated according to three parameters, which we will note A, T and F.
• A: Is the affinity, which measures how the user and the author of the content are close. • T: Is a measure of attractiveness of the content (more for photos than for texts for example). • F: Is the freshness, weak if the content is old, strong if it has just been posted.
The score is the result of multiplication of all parameters: To go deeper in our experience we have chosen five existing Facebook pages (B, C, E, H and M) which represent respectively five countries (United Kingdom, USA, Spain, Morocco and France), we talk about the official Facebook pages of the most famous media platforms in their original countries and having a number of fans that exceed 4 million one. Then, we extracted all the necessary data concerning their publications posted during Wednesday 02/05/2018, and we also collected all the interactions made in this day during 24 hours (number of posts, type of posts, number of comments, number of shares, and number of interactions via empathy button).
The selected pages are: • In this experiment, we did not cite examples of the media platforms of the major economic countries like China and Russia for the simple reason that they have other social networking tools that are used much more than Facebook.

Data analysis
We should further recall that the data was extracted in Wednesday, May 02, 2018 from the news feeds by just collecting the publications published in the same day above (and taking into consideration UTC+1 as time zone).
Thus, all data (number of posts, type of posts, number of comments, number of shares, and number of interactions via empathy button) are extracted from the HTML tags that carry the necessary information that we have need: <article> Tag <div> Tag that the <header> Tag has "data-sigil" as attribute with "m-feed-voice-subtitle" value.
<abbr> Tag that follow the tag above. <div> Tag from <footer> Tag that contains "_1fnt" as value in class attribute It should be noted that the extract of data from news feed is done manually (nonautomatically). After collecting the data, we found some results that we put it in as form and diagrams come in the following paragraphs. Figure 1, represents the number of publications published by type and by Facebook page, which are in total 275 (234 as links, 1 as photo, 40 as videos). The figure 2, represents the percentages of all publications posted by type for the five Facebook pages. Figures 1 and 2 describe the various kinds of publications, and the one that dominates or the most shared on the Facebook pages for these 5 media companies. it is link type publication.  Note: It should be noted that since the status type is not present in data extracted from the content of these five pages, so, it will be excluded in the analysis table.
Case (1): Likes buttons' performance   Figure 3 shows a summary of the performance, which clearly shows that the like button is the most expressive as type of interaction in all publications, and we also see that page E is the most effective of the other pages in term of this rate whether for the rate of performance of like or that of share. while for the rate of performance of the comments page C remains the leader.    For the publications posted on page C which constitute 57 publications with a total of 446,080 as the number of reactions between (Likes, shares and comments), and according to Figure 5, we see that the reactions increase clearly in 2 periods who take the majority of all reactions. Period 1; between 08h to 14h with a decrease around 11h UTC+1 of which 21 publications are posted. Period 2; between 19h to 23h UTC+1 with a drop to 21h including 11 publications are posted.
Since this page is from USA, so the time used by the fans of this page follows UTC-4 as time zone   Figure 7, we see that the reactions are clearly increasing in a long period that starts at 9am and arrives at 23h but with a decrease in 2 occasions one around 15h UTC+1 and the other around 17h UTC+1 which 54 publications are posted in this long period. For the publications posted on page L which constitute 46 publications with a total of 26600 as the number of reactions between (Likes, shares and comments), and according to Figure 8, we see that the reactions increase clearly in 3 periods, who take the majority of all reactions. Period 1; between 06h to 08h UTC+1 including 7 publications are posted. Period 2; between 10h to 12h UTC+1 including 10 publications are posted. Period 3; between 18h to 20h UTC+1 including 6 publications are posted. Since this page is native to France, suddenly the time used by the fans of the page follows UTC+2 as time zone.

Correlations results
Here we discuss the subject of correlation, to know its presence between the set of reaction tools (Shares, Comments and likes) for the five pages studied.
The correlation coefficient between two real random variables X and Y each having a variance, denoted by Cor (X, Y). Or sometimes simply r, that is defined by: where Cov (X, Y) is the covariance of the variables X and Y, σ X and σ Y their standard deviations.
After computations made on our data collected and following the previous formula, we found the results presented in Figures 10, 11, 12, 13 and 14. and that evidence the existence of a strong correlation between the set of interactions tools especially for pages C, E and H.  Figure 10 shows the correlation results for page B, which clearly states that there is a medium correlation between "Shares" and "Likes". Figure 11 shows the correlation results for page C, which says that there is a strong correlation between the set of interaction tools and less strong in both "Shares" and "Comments". Figure 12 shows the correlation results for page E, which says that there is a strong correlation between the set of interaction tools and less strong in both "Shares" and "Comments". Figure 13 shows the correlation results for page H, which says that there is a strong correlation between the set of interaction tools "Likes", "Shares" and "Comments". Figure 14 shows the correlation results for page L, which says that there is a medium correlation between the two pair ("Likes", "Comments") and ("Shares", "Comments"). while a strong correlation for the duo ("Shares", "Likes").

Conclusion and Perspectives
From what we have studied previously, and according to the results found in section 3, we can say that the all interaction mechanisms usually have a strong link that takes an important correlation score for the duo "Likes" and " shares ". We have also discovered that the media agencies we have just studied through their Facebook page always prefer to publish link-type publications which are of course links towards topics published on their website. The publications as video types come in 2nd place in terms of sharing and lastly, we find the publications as types images.
We have also noticed that the publications as status type doesn't exist among publications in all of these 5 pages.
Regarding to the most attractive publications type, it appears that there are many links-type publications which are mostly liked and shared, but we cannot confirm this hypothesis because there is no equality in terms of sharing numbers between all types of publications.
It is true that all data extracted from posts published in a single day are not enough to make definite conclusions and peremptory judgments, but we must not ignore the mass of these data and the information that can bring with the large number of interactions that presents a subject to be studied by researchers in the field via different approaches. The importance of what we have collected for the five Facebook pages lies in the fact that these are huge numbers, we are talking about 104,066,851 fans, 617,440 likes, 139,614 shares, and 148,488 comments, are the totals of the said pages.
Extracting data was done with a manual way, so as a perspective, our future work is to do it automatically, especially because that we have already identified the set of tags and attributes that bear what we have needs as data. Automatic data extraction will facilitate the collection of all information not only for a single day, but may be spread over several consecutive or sporadic days, or even weeks and months. This next work will surely help us to extract huge data that will allow via Big Data concept, to have good results and to make good decisions.