Popularity of social media in Bangladesh is prodigious. 80 percent of internet users are on social networking websites like Facebook, Twitter (a2i.pmo.gov.bd). That is over 16 million people and counting. The rate of new Facebook users is outpacing the country’s birth rate as one new Bangladeshi Facebook account is opened every 20 seconds. This makes social media a great platform for government to reach out to citizens and stay up-to-date with current events and trends in society. That is why, a Facebook group named “Public Service Innovation Bangladesh” has been created. In this group, discussions related to public service innovation, public service related problems and solutions, decision making in administrative works etc. are being prioritized. These discussions are reaching across the hierarchy among all ministries, departments and agencies. There is an organization named as Organization for Economic Co-operation and Development (OECD). Member countries of this organization deliver better public services via ICT based platforms such as institutional websites, social media and public service related apps. There are 34 OECD member countries and among them, 26 countries have their own Facebook page and twitter account for various ministries (Mickoleit, 2014). Lansing is the capital of the US state of Michigan and they have their own verified twitter account for better public service. Bangladesh govt. is also trying to do that in Facebook. Cabinet division of People’s Republic of Bangladesh enforced a propaganda to join all the government officials in this Facebook group (www.rdcd.gov.bd). The focus of this study is to construct complex network from posts given by the members of this Facebook group, analyze features of the complex network including degree distribution, assortative mixing and betweenness centrality. It is important to detect influence networks or communities of that Facebook group to examine, is there any principal agent problem occurring in that group in decision making (Dahlin J, Svenson P). The principal agent problem arises when one party (agent) decides to work in favor of another party (principal) in return for some incentives (Sanford J., Oliver D., 1983).
We have obtained data from that Facebook group using Scrapy which is an open source and concerted framework for extracting data from websites. Apart from data collection, data cleaning, data integration and some prepossessing has been carried out. In this study, relevant data fields include status id, status message, status type, status link, status published date and time, link name, number of reactions, number of comments, number of shares, comment author, comment message, comment likes, comment published date and time. We have analyzed group data from April, 2016 to January, 2017 and generated a report which has given some interesting insights about that group. During this time frame, 601 posts have been posted and most amazingly, majority of these posts have been posted on December, 2016 and January, 2017. So, it can be said that, this group is growing now. Also, we have analyzed post distribution by day and hour. We have found out that government officials give posts mainly on Sunday and Monday around at 5 pm. We have categorized the posts into 4 classes, namely, Social Welfare, Motivational, Problem Statements and Government Information Distribution. We have observed that most of the posts fall into the Social Welfare category. Then, we have generated our data file in gdf format to be analyzed by Gephi, an open-source network analysis and visualization tool. In the network graph, a node represents a post and the person who gives that post whereas an undirected edge represents likes/comments to a particular post. We have gathered last 1000 posts which contains 5408 nodes and 32768 edges. 4408 members gave those 1000 posts which equals 5408 nodes. The average degree distribution, both in-degree and out-degree of the network are calculated to be 32. Furthermore, the centralization value of in-degree and out-degree are calculated to be 0.876 and 0.874, correspondingly. These values indicate that members are directly connected to one another via their posts. The value of Pearson correlation coefficient is very near to 1 which designates that the network is assortative. That is, nodes having many connections tend to be connected with other highly connected nodes. In our constructed network, we have seen that the people who give more posts, get more likes and comments. That is how, they tend to be connected with other highly connected people. If a person who has many connections, gives a post, gets more attention meaning likes and comments than other. So, if s/he gives a post related to infrastructure problems, chances of getting solutions of that particular problem are high. Also, junior officials can share their views, opinions, perceptions in a common platform like Facebook to draw attention to the senior officers (Kapucu N. 2015). In that way, uncertainty in decision making can be reduced. Our study helps to understand this complex network of government officials.