Smart Monitoring and Controlling of Government Policies Using Social Media and Cloud Computing

Governments around the world are promoting multiple services to serve its citizens in a better and transparent way. E-Government is one such mechanism that enables the government to perform its day to day tasks and provide a better service to its citizens (Mohammed et al. 2016; Zwattendorfer et al. 2013). E-Government is the use of ICT and other web technologies to provide access to effective, efficient and transparent public services to its citizens and employees (Jeong 2006; Rana et al. 2013). Although, use of ICT can provide several advantages, yet the required technical infrastructure, implementation cost and requirement of skilled staff becomes major obstacle towards E-Government implementation (Rana et al. 2013). With the emergence of cloud computing, these challenges can be addressed up to a fair degree of satisfaction for all stake-holders (Mohammed et al. 2016). Cloud computing consists of large shared pool of computer resources which provide features like on-demand scalability and pay-as-use (Sadiku et al. 2014). These advantages have played a decisive factor in motivating the governments of many countries to migrate from traditional costly E-Government model to cost efficient as well as scalable cloud based E- Government model (Sharma et al. 2012). Due to this adoption cloud computing has become a new channel for delivering improved government services (Liang et al. 2011; Smitha et al. 2012). Cloud based E-Government services provides advantages like reduced operational cost, distributed data storage, scalability and finally security management (Smitha et al. 2012), which indeed improves the relationship between government and public. Not only this, but cloud based E-Government model is also building a strong foundation for smart cities (Clohessy et al. 2014).

Traditionally, policy making had always been based upon official statistics (data) generated by government agencies and international bodies (Severo et al. 2016). However, decision makers normally reported flaws in this form of data because of publication delays, top down approach and insufficient topic of interest etc. (Pfeffermann et al. 2015). Therefore to overcome this traditional problem of data collection, political scientist and policy makers moved their attention towards social media platforms like Twitter and Facebook, to accumulate more authenticated user data and that too in real time. Hence, social media becomes an efficient tool for enhancing transparency in government working and increasing the communication between citizens and policy makers by bringing out transparent data analytic statistics (Chung and Zeng 2016; Lee and Kwak 2012). Nowadays, social media have become an integral part of everyday life, irrespective of the status of any individual (AlAlwan et al. 2017; Dwivedi et al. 2015; Kapoor et al. 2018; Shiau et al. 2017, 2018). This virtual world provides a perfect platform for people from all around the world to discuss topics of common interest such as sports, entertainment and even politics. Talking about politics, at least 33% of social media users comments, discusses or post about politics on these platforms (Hossain et al. 2018). Even governments have realized the potential of social media. Consequently, various government agencies have started using various social media platforms to connect and engage with general public (Aladwani and Dwivedi 2018; Alryalat et al. 2017; Bertot et al. 2012; Rana et al. 2015). As social media is helping to increase the interaction between public and government, it is indeed facilitating public participation (Ceron and Negri 2016). Generally, people post something regarding government, politics or policies which might be intentional or unintentional (Hossain et al. 2018), which can be utilized by government for formulating more effective public policy as well as designing and delivering better services to its citizens (Androutsopoulou et al. 2018; Joseph et al. 2017; Park et al. 2016).

As represented in both the above subsections that considerable work has been done in the field of cloud computing based e-government and use of social media for policy making having their own advantages and benefits. Hence, cloud computing based e-government and social media can be considered as an influential paradigm to be collaborated with controlling and monitoring of public policies. With large number of people engaging on social media sites for discussions related to public policies, resulting in generation of huge amount of data. There is a need for a cloud based system that can utilize this large data and map useful public opinion in form of issues, concerns, solutions, proposals, advantages and disadvantages (Androutsopoulou et al. 2018) towards a public policy at an early stage so that appropriate steps can be taken in order to please the public for which the policy has been formulated (Bertot et al. 2012). To the best of our knowledge, till date no effort has been made to combine both these services. Hence, this paper aims to unitize the capabilities of cloud computing and social media analytics for efficient monitoring and controlling of public policies. With this research, we try to answer the following questions that may erupt in the mind of the readers of this paper:

Data for experimentation has been collected from Twitter. Though the proposed system (See Section 3) is capable enough to use multiple social media platforms for data retrieval, however for this experimentation only Twitter data is used. As discussed above data was fetched based upon specific #hashtags (#GST, #GSTForNewIndia and #OneNationOneTax). For identification of #hashtags, an expert team consisting of three independent experts was constituted. These experts sorted #hashtags that were linked with GST. The common #hashtags identified by three experts were selected for performing tweet fetching operation (Singh et al. 2018c). The Twitter API used for fetching tweet provide us with various search parameters such as, language of tweets, type of tweets (Original, Re-Tweets or Both) and date range between which fetching operation need to be performed. Using these search parameters, we perform the fetching operation in more appropriate and desired manner.

In total 41,823 tweets were collected over a span of 24 days starting from June 23, 2017 to July 16, 2017. Since data was fetched using APIs, it was in unstructured form i.e. json format. In order to use this data for further analysis it was necessary to convert this unstructured data into structured data i.e. excel format. Table 2 shows the various attributes that were fetched during data collected from Twitter API.

In total 41,823 tweets were collected from 35,400 different users from India. In total 2873 users were detected who tweeted more than one tweet, accounting a total of 6423 tweets. For better interpretation of the results tweet collection period was broken into 3-phases (Pre-GST, In-GST and Post-GST) of eight days each. The breakup of these phases is shown in Fig. 4, while Fig. 5 shows daily tweet collection.

The shaded part (orange color) in Fig. 5 indicates the In-GST period when Twitter traffic was highest as compared to other two phases. The details of tweet statistics are given in Table 3.

Hashtag analysis deals with the various hashtags that occurs among the collected tweets. In total 11,499 unique hashtags were found, that appeared 81,054 times. From the total 41,823 tweets, 19,114 tweets contained more than one hashtag. The analysis shows that hashtag “#GST” had the maximum occurrences i.e. 18,925. Figure 6 shows top 15 hashtags which had maximum occurrences. For better visualization “#GST” was excluded from the results in Fig. 6. Figure 7 shows association among popular hashtags. The association is plotted using Fruchterman-Reingold layout (Fruchterman and Reingold 1991; Csardi and Nepusz 2006). Similarly Table 3 shows the hashtag adjacency matrix. Both Fig. 7 and Table 4 the relationship among different hashtags.

Sentiment analysis is often regarded as most effective tool to map public response towards an entity (Joseph et al. 2017). Sentiment analysis is defined as a text analytic technique which deals with extraction of sentiment from given piece of text (Liu 2012; Mishra and Singh 2016). Unlike the previous analysis, the upcoming analysis requires preprocessing of data. Data preprocessing is an important task, which aims to prepare data for various data mining task (Van Broeck et al. 2005; García et al. 2016; Liu et al. 2015). Since Twitter data contained lot of noise and unwanted stuff, task of data preprocessing was performed (Haddi et al. 2013). These tweets preprocessed by performing various task conversion to lower case, removing punctuations, removing special characters and finally removing web links. Sentiment analysis consists of two sub-operations: (a) E-Motion Analysis (Mohammad and Turney 2010; Ou et al. 2014) and (b) Polarity Analysis (Saif et al. 2013; Yuan et al. 2017).

Topic modelling identifies the main themes among the captured tweets (Blei 2012). It is defined by the common set of words that have high probability of belonging to a particular topic. For performing the process of topic modelling latent dirichlet allocation (LDA) is used (Arun et al. 2010; Deveaud et al. 2014). For deciding optimum number of topics we calculated the coherence score of topics (2 to 30). The results of these are shown in Table 6, while the scores are shown pictorially in Fig. 13. The results show that 29 topics are optimum. The results of topic modelling are shown in Fig. 14, in form of intertopic distance map.

Community analysis is part of network analysis, which detects various communities taking active part in a particular discussion (Ding 2011). The polarity (positive and negative) wise results of community analysis is shown in Fig. 17 for all three phases. In this the green color nodes indicates community which is happy with GST implementation, while red color nodes indicates the community which has negative opinion towards GST. The results clearly indicates that the number of people with negative opinion towards GST increased during phase-2 (In-GST Period). But once all preventive steps were taken positive community again had upper hand.

The overall results of community analysis is shown in Fig. 18. The results shows that there are four prominent communities. Community-1 (Blue) indicates people who are supporting GST, making it the largest community. Community-2 (Red) indicates people which are against GST. Finallay, community-3 (Green) and community-4 (Yellow) showing people who have some confusion and requries some clarrification regarding GST.

Location based analysis is a very crucial tool for gathering information, while mapping public response towards an entity (Amirkhanyan and Meinel 2017). Although all tweets do not contain location from where they were tweeted, yet we can’t ignore them as they provide important information about the actual location, hence helping policy makers to target the audience while finding solution to their problem. Since in the previous section we detected that during phase-2 (In-GST Period) the overall sentiment was on negative side, it is essential to see which states and cities are worst affected so that appropriate solution can be found. Figure 19 shows results of zone wise analysis, while Fig. 20 shows results of state wise analysis and finally Fig. 21 shows location wise analysis of two markets of Mumbai city from where negative tweets were sent.

The main reason for integrating cloud computing in our proposed system was cost cutting, as we only pay for what we use and initial cost of installation and later cost of maintenance are all eliminated. Since cost plays an important role in implementation of such projects and often becomes a major bottleneck in developing countries (Like India). Hence is becomes extremely important to evaluate economic feasibly of our proposed system. The very basic instance (t2.nano with 1 processor and 0.5GB RAM) available at Amazon cost around $0.0062 per hour (Amazon EC2 Instance Price 2018). While the instance used by us for the entire experimentation c4:large instance (containing 2 processors and 8GB of RAM) cost around $0.10 per hour. Depending upon the hardware requirements for computation we can switch to any instance. Another cost that needs to be considered here is social media sites API cost. Again depending upon the use the government might opt for monthly or annual contract (Twitter API Price 2018). Paid APIs gives access to various additional parameter like location of all tweets, which is not available in free APIs. So from above we can conclude that the proposed system is highly cost effective.

The implication of the study for practice can be divided into three subsections (a) Government which will be implementing any policy (b) General Public for which this policy will be implemented and (c) cloud service providers. These subsections are discussed briefly below.

For this government can create a page on Facebook, or highlight the advantages of new policy through tweets on Twitter. Early discussion on social media platform will help the government to better monitor the policy before actual implementation by considering various suggestions highlighted by people on social media. Another important insight that the government can gain are, what all factors lead to polarization among various social media groups towards a policy (Grover et al. 2018). In case the government receives too much negative opinion towards the policy, then government might consider delaying its implementation or might take expert opinion for its betterment. Once the policy is implemented and if afterwards government starts receiving negative opinion towards the policy as it was in case of GST. Then government should take appropriate steps to overcome the hardship faced by the public.

Though, the proposed system has given us some encouraging results yet it suffers from certain limitations that needed to be addressed. Firstly, data (Tweets) for experimentation was fetched only from Twitter, while other social media platforms like Facebook, Instagram etc. were not utilized. Secondly, only original tweets were considered for analysis, while no emphasis was given to re-tweets. Re-tweets as we know constitute up to third of the entire Twitter traffic (Holmberg 2014), hence considerable amount of Twitter traffic was ignored by us in our analysis. Similarly, bot detection (Chu et al. 2010) was also not performed in our system. Social media bots can adversely affect the analysis by making the results biased. Hence, it is very important to address them. Though there are many advantages of cloud computing, still cloud computing has its own limitations such as network communication cost, unreliable latency, lack of mobility support etc. (Stojmenovic and Wen 2014). We can overcome these limitations by extending our system to Fog computing (Yi et al. 2015). All above stated issues can be addressed in future.