Work Behavior Analysis of Indonesian Civil Servants Using Social Media Interactions

This research examined Indonesian civil servants’ work behavior using social media interactions. A hermeneutic approach was used to ascertain the meaning of an individual’s actions us-ing text in social media interactions. The results showed that the work behavior of Indonesian civil servants tends to be negative and is primarily due to the unequal workload distribution. First, this study uses a collection of data obtained from Twitter. Second, the collected data is based on the topic of work behavior. This research implies that work behavior needs to be monitored and managed appropriately to enhance public sector organizations’ achievement and keep employees in proper conditions. Furthermore, work behavior management can be improved by evenly distributing the workload among employees. Negative work behavior leads to decreased employee performance, causing dissatisfaction in public services. Work behavior analysis using social media interactions in the public sector is a new theme needing exploration in public administration practice. This is a critical research topic as it can affect an organization’s achievements supported by rapid technological developments.


INTRODUCTION
The invention of the Internet has made global communication limitless.Everyone can connect through cyberspace using social media (Clark et al., 2018;Hanna et al., 2011;Yohanna, 2020).Furthermore, most people use social media to establish communication with friends and family, find trending topics, post their personal lives, share opinions, and carry out work-related activities (Data Reportal, 2022).According to "Digital 2022: Indonesia," the number of active social media users in Indonesia totaled 191 million (Data Reportal, 2022), up 12% from the previous year.
Openness in social media has also led to the emergence of anonymous accounts (Ma et al., 2016;Ong & Weiss, 2000), including in government circles.Owners of this kind of account disguise their real identities to be freer in expressing their personal opinions (Sharon & John, 2018).Anonymous accounts are present because of a lack of security in conveying aspirations related to the emergence of negative consequences, including the risk of losing a job, the emergence of opposing views, the risk of damaging promotion opportunities, and matters relating to sensitive issues (Mao & DeAndrea, 2019;Scott & Rains, 2005).
One of the anonymous Twitter users with the most interactions within the government circles is the @PNS_Ababil account, which has 140,200 followers and 89,700 tweets.@PNS_Ababil is an account that conveys experiences, shares opinions, and criticizes issues or internal government policies.Various messages written by that ac-count were responded to by its followers, resulting in enormous interactions with differing opinions.Therefore, these interactions give rise to observable behavior.
The emergence of a massive social media phenomenon creates new opportunities for assessing civil servant work behavior.Previous research focused more on evaluating work performance and job satisfaction, including Aisy (2022), Rokhmawati (2013), Epita (2013), andBahar (2019).Furthermore, Puspita Rokhmawati (2013), in the "Analysis of Employee Performance Evaluation," observed a bias in ASN work performance assessments carried out by superiors.This bias was due to the absence of clear criteria for assessing work performance and behavior, causing the assessments to be mainly subjective.The accuracy of these assessments could be enhanced, as research showed that several appraisers need to be firmer in giving grades to their subordinates.
However, current technological development is more effectively reducing assessment subjectivity.Technology can be used to support human resource analysis of employee emotions and interactions using real-time data.The technology, which includes artificial intelligence, data mining, machine learning, and IoT, has been implemented in several large companies such as Google and IBM (Gelbard et al., 2018).Previous research by Gelbard et al. (2018), Young & Gavade (2018), and Saraff et al. (2020) analyzed social media interactions using technology.Meanwhile, from a public policy perspective, Rathore et al. (2021), Driss et al. (2019), and Grubmuller et al. (2013) emphasized that the use of technology can also support the policymaking process.Previous studies have analyzed social media data more broadly, namely from the public opinion used as datasets.Retrieving data through social media can provide a good picture and increase the information value for policymakers.Opinions expressed in social media interactions can also be researched from a public point of view to detect trends and infer prognoses.
Research on Indonesian civil servants' (ASN) work behavior needs to be conducted, and this fills in the knowledge gap through the observation of social media.This is crucial to improve policies related to human resources management.Furthermore, this research contributes to mapping the ASN work behavior to formulate a general description.It provides empirical evidence associated with ASN work behavior and the related triggers.

Organizational Behavior
Organizational behavior is a form of action or interaction that directly and indirectly affects employees' effectiveness (Chirumbolo, 2017).It relates to human behavior in an organization or a particular group (Thoha, 2010).It arises from individual and group behavior, ultimately affecting the organization.
Work behavior is a significant aspect of organizational behavior.It is an individual response or reaction in the form of actions, attitudes, and assumptions about work, working conditions, and the treatment of employees by a leader (Maulana, 2013).It is also described as communication between an individual and groups in his work environment.Meanwhile, according to Robbins, work behavior is how a person actualizes themselves through attitudes at work (Maulana, 2013;Sidharta & Alexander, 2006).There are two types of work behavior in organizations: positive and negative work behaviors (Maulana, 2013).The output at the individual level is divided into four aspects, namely performance, job attitude and job stress, Organizational Citizenship Behavior (OCB), and withdrawal behavior.

Performance
Performance is one of the indicators used to evaluate employee achievement.Furthermore, it illustrates employee efficiency in performing professional tasks (Caillier, 2010).Performance is work assessment based on several aspects, including the quality of work, speed, initiative, ability, and communication (Kadarisman, 2019).Robbins & Judge (2017) emphasized that performance is a way to demonstrate an employee's ability to achieve goals within an allotted time.
In a government circle, performance is often defined as an employee's ability and responsibility to provide good service and fulfill administrative work (Kreitner & Kinicki, 2010).Moreover, performance comes from a combination of abilities, efforts, and opportunities in the work environment.Based on these definitions, performance is the result of work that describes an employee's ability and professional expertise, evaluated according to specific factors.The work results will directly impact individual and organizational success through organizational productivity (Coleman & Borman, 2000).Moreover, performance can also influence decisions regarding promotion, termination of employment, and increased performance to bonuses within the organization (Caillier, 2010).

Job Attitude
Attitude is a learned tendency to respond to something based on what an individual knows, believes, and feels.Furthermore, attitude can be evaluative and may depend on pleasant and unpleasant statements about objects, people, or events (Robbins & Judge, 2017).In organizations, it constitutes an essential component of behavior (Robbins & Judge, 2017).
Attitudes in the work environment are called job attitudes.These lead to positive or negative evaluations of the work environment, affect performance, and influence withdrawal behavior.Furthermore, positive job attitudes improve employee performance, while negative job attitudes can cause high withdrawal behavior within the organization (Robbins & Judge, 2017).

Job Stress
Stress is an adaptive response to external actions, situations, or events attacking an individual's physical or psychological state (Kreitner & Kinicki, 2010).Robbins & Judge (2017) stated that stress is a psychological process of desired opportunity, demand, or resource.
Most stresses originate from work and often occur due to excessive workload and hours (Robbins & Judge, 2017).These conditions cumulatively affect the mind and psychology, causing job stress (Soep, 2012).In addition, low employee abilities, unskilled superior workers, excessively short deadlines, lack of trust, role ambiguity, job changes that are too fast, and role conflicts at work also cause job stress to employees (Rivai & Mulyadi, 2013).Kreitner & Kinicki (2010) defined stress causes at the individual level to include job demands, excessive work, insufficient and monotonous work, role conflict, role ambiguity, and job security.Anitha J (2014) added that the most influential causative factors are role ambiguity and the lack of employment opportunities in career development.
The consequence of job stress includes psychological disorders and reduced work productivity.Psychological disorders can be in the form of low job satisfaction, decreased organizational commitment, decreased employee involvement in work, selfesteem, burnout, and depression (Kreitner &

Figure 3. Word cloud visualization
Kinicki, 2010).Furthermore, stress can also affect an individual's physique, causing dangerous diseases.Stress is an unavoidable condition in the work environment.However, stress is not only considered a negative factor; it can also have positive consequences.

Withdrawal Behavior
Withdrawal behavior is a collection of actions causing an individual to separate themselves from the organization (Robbins & Judge, 2017).There are several forms of withdrawal behavior, such as tardiness, absence, and turnover.Eder & Eisenberger (2008) explicitly defined it as actions carried out by employees, such as frequent lateness, negligence in assignments, talking about matters outside of work during working hours, and extending rest periods from the approved provisions.Withdrawal behavior begins with an intention and is realized when the opportunity is presented (Hastuti, 2015).Some triggers include a lack of tasks/unclear tasks, poor work systems and culture, and unexecuted rewards and punishments (Epita, 2013).Consequently, withdrawal behaviors reduce organizational productivity (Adam, 2018).

Affective Event Theory (AET)
AET is a model developed from organizational behavior in 1996 by Weiss and Cropanzano.The AET involves the environment, work events, affective reactions, and consequent attitudes.According to KBBI, affective is related to feelings or can affect feelings and emotional states.Meanwhile, Robbins & Judge (2017) defined affection as a broad understanding of the feelings experienced by an individual, including emotions and moods.The AET was developed to determine the response of emotions to events within the organization's scope.Therefore, this theory defines the events that affect an individual's feelings, emotions, and moods, influencing the consequent behavior.
Events occurring in an organizational environment are known as work events.According to Rentsch (1990), these broadly happen within an organization and are connected to its members' roles.The process of work events is triggered by superior support and employee welfare (Ghasemy et al., 2021) and occurs differently depending on the type of organization.In the profit organization, the events may include bankruptcy and customer satisfaction.However, work events can also take different forms in public organizations, as shown in Table 1.
Work events trigger emotional reactions in the form of both positive and negative emotions.It also triggers conflict and cohesion (Ghasemy et al., 2020).These reactions should not be ignored because they can accumulate and ultimately affect the organization (Robbins & Judge, 2017).

METHOD
This is an exploratory-descriptive and qualitative research, and the approach is hermeneutic.Furthermore, this research considers the meaning of an individual's actions in the tweets about public policies and issues related to government, especially work behavior, through text in the interactions with the @PNS Ababil account.Indonesia was chosen as the location of the Twitter platform due to the broad reach of social media, enabling a broader ASN perspective.Twitter was selected due to its superiority, especially in terms of anonymity compared to other social media platforms.Anonymous users are often active in discussions and more willing to express their points of view (Peddinti et al., 2014).In addition, Twitter focuses more on text, enabling a wide range of users to easily express opinions on various topics (Hughes et al., 2012).The information obtained on Twitter is more up-to-date and can cover significant issues in general (Laor, 2022).
For this research, data was collected by scrapping on @PNS_Ababil using the Python programming language and the snscrape library.The data collected includes the tweet posting time, tweet ID, and the number of retweets, likes, replies, tweet texts, and usernames.Accounts interacting with @PNS_Ababil are partly anonymous, but we collect data sets over a fairly long period to ensure that anonymous accounts are indeed from ASN circles.Data was collected for this research in June-August 2022.Furthermore, work behavior cannot be con-cluded within this time frame, as this period reflects employee behavior.The selection of the research period also increased data validity while reducing data sourced from accounts other than ASN.In addition, the tweets used as the research sample are tweets originating from accounts that have had a minimum number of interactions for 2 months from the period of data collection.In addition, the tweets used as the research sample originates from accounts that have had replying tweets for @PNS_Ababil for two months from the data collection period.In addition, the tweets used as research samples were also reduced based on the research context, which is work behavior.These terms are not part of the civil servant's work, so they do not reflect the work behavior of the civil servant.
The data analysis method used in this research is sentiment analysis, followed by qualitative content.Sentiment analysis is a text assessment method combining Natural Language Processing (NLP) and machine learning (Vitadhani, 2021).It also analyzes public responses to events and past or future phenomena (Kanugrahan, 2021).Sentiment analysis in this research includes: In preprocessing phase, data is prepared, including case folding, tokenization, stemming, and stop-word removal.The case folding stage removed mentions, characters, and numbers and changed the letters to lowercase to ensure the tweet data had a more uniform shape.Meanwhile, the tokenization stage was carried out to break sentences into words, and the stemming process was used to transform each word into its basic form.The stop words removal process deleted words without meaning in sentences, such as affixes, conjunctions, or greetings.Furthermore, data preparation included changing slang words to normalize data.This step is crucial as it influences the final result of the work behavior analysis.
Further we label the data.This process is notating data according to the sentiment used in the tweet.This ensures the computer can recognize and process data for the next step (Kanugrahan, 2021).The data labeling process was carried out manually, using positive and negative sentiment groups.
Finally, data classification phase is to classify work behavior based on sentiments used in social media interactions.A classification model was built using two algorithms, Naive Bayes and Support Vector Machine (SVM).These are the most popular classification algorithms producing a good level of accuracy.The classes used in this classification only consist of positive and negative categories.SVM determines the best hyperplane to divide classes and classifies patterns not in the classes to become data train.This algorithm also has good generalization abilities that can be implemented quickly.In comparison, Naive Bayes ascertained each class's most frequent probability value.These algorithms were then compared to obtain the best accuracy of the model.Model evaluation is also measured using a confusion matrix to determine accuracy, precision, recall, and F-1 values.
Furthermore, this research applied qualitative content analysis to observe both positive and negative aspects of ASN work behavior.The qualitative analysis mapped the opinions expressed and the factors influencing the formation of an individual's attitude (Ependi, 2014).Qualitative content analysis also investigated the context and process of the writing presented to produce in-depth information related to the social media context (Sumarno, 2020).

FINDINGS AND DISCUSSION
The data collected in the last three months totaled 43,436 tweets.The data is then mapped according to the occurring interactions.
Interactions on the @PNS_Ababil account in the last 3-month period have increased.Interactions in August had the largest number, amounting to 69,778.The interaction of the last three months is shown in Figure 1.
Furthermore, Figure 2 shows that interactions mostly occur on Sundays.The trend shown in Figure 2 tends to decrease from Sunday to Saturday.Wednesdays had the lowest interactions, with an average of 1,126 interactions.However, Figure 2 also shows that on weekdays, followers of the @PNS_Ababil account continue to interact as usual, especially on Mondays and Tuesdays.
The data was then visualized using the word cloud technique.The word cloud visu-

Naïve Bayes SVM
alization technique used to ascertain the frequency of word occurrences showed that the most frequent words were kerja, kantor, pns, and esmelon.These words can be combined as work behavior characteristics.The word cloud visualization is shown in Figure 3.The data was then reduced and mapped as work events.A reduction minimizes the outlier of the data posted by buzzer or non-ASN accounts.The collected data amounted to 7,114 tweets, and the results are shown in Table 2.
Based on the data above, the screenshots of work events with the highest interaction are shown in Figure 4.During the last three months, the work events receiving the most interaction were tweets about a government official.One complained about PT Taspen's Chairman, suspected of having an affair, while the other complained about a public official's statement on application blocking.
These work events were further classified based on the sentiment used.The accuracy was obtained using the ratio or comparison between correct predictions (both positive and negative) with the actual data.Based on experiments conducted using Naive Bayes and SVM algorithms, the model with the best accuracy was Naive Bayes, which was tested using 4,979 previous research data.Based on its accuracy, Naive Bayes has 72.88%, better than SVM, which has 72.51%.Furthermore, the model was tested using a confusion matrix, as shown in Figure 3.The confusion matrix will show the precision, recall, and f-1 score.Based on the results, Naive Bayes also has a relatively better value than SVM.
Of the 2,135 data tested, the number of data predicted to be correct was 1,556, with 1,204 indicated as negative and 352 as positive sentiments.On the other hand, 579 data had the opposite value, consisting of 209 predicted as positive and 307 as negative.Moreover, the total accuracy obtained from the evaluation model was 72.88%, as shown in Table 3.
From the results of the sentiment analysis, the final number of negative sentiments on ASN work behavior was 4,708 data, while the positive totaled 2,406.Therefore, most sentences have negative sentiments due to the satirical words used in the tweets.
From the word cloud visualization of tweets with negative sentiment, the words appearing the most were esmelon, umbi, suruh, and banget.The word esmelon means eselon or the higher position in the governmental office.Then, the word umbi means a staff that should obediently belong to the superior.These words can represent discussions about orders and attitudes from superiors or officials in the organization.Meanwhile, in the word cloud visualization of tweets with positive sentiment, the words that appeared the most were atur, latsar, and gaji.The word atur was found in a tweet discussing rules, while latsar involved the basic training of a civil servant candidate.The word cloud visualization of negative and positive sentiment is shown in Figure 5.
Furthermore, the qualitative content analysis used 1,000 research data on perfor-mance, job attitudes, job stress, and employee withdrawal behavior.The results are shown in Table 4, which contains allegations of increased performance in 92 events, positive job attitudes in 141 events, decreased job stress in 28, and decreased withdrawal behavior in 13 events.Conversely, there was a decreased performance in 165 events, negative job attitude in 287, increased job stress in 218, and increased withdrawal behavior in 164 work events.

Reduced performance caused by workload
This was determined using a collection of user tweets describing performance, including quality of work, speed, initiative, and ASN capabilities.The discussion about ASN performance also concerns compensation, including salaries, benefits, and facilities.Regarding the initiative and speed of work, the discussion was related to the workload division.ASN tends to avoid aspects of attitude as well as initiative.The more initiative the employee has, the more the workload is increased.This additional workload carried out by employees is often outside the position's primary duties, including treasurer roles, video editing, and a master of ceremony.Finally, the additional workload tends to cause employees to be reluctant.
Most employees expressed their opinion about the amount of compensation needed to be more active and depend on their performance.Some tweets stated, "No matter how much work needs to be done, my compensation is not increased."Therefore, employees prefer to relax in their jobs.On the other hand, employees who are more committed to the job and give the best performance are not adequately appreciated.This is based on a tweet with the sentence, "Professional work demands a sincere salary."The statement implied that the compensation is not equivalent to the workload.However, it forms a paradigm for normalizing lousy behavior at work due to inadequate compensation.

The emergence of job attitude
The results regarding job attitudes are closely related to work culture within the organization.Job attitude is mainly a negative attitude that reacts to a disturbing work environment.These negative attitudes are influenced by acts of violence, both physical and verbal, committed by co-workers and superiors.Negative job attitudes also include laziness at work, bribery and corruption, collusion and nepotism, abuse of office, as well as opposition to orders given by superiors.
A positive job attitude also occurs in various interactions and discussions.Some support a particular policy, co-workers experiencing difficulties, mutual respect between co-workers, as well as the adaptive behavior of employees towards changes due to policies and the work environment.

Reflection on job stress
Tweets discussing job stress often reflected a complaint and were present due to the workload, policies, association, culture,

Withdrawal behavior during working hours
Withdrawal behavior often manifests in absenteeism, delays in attending activities, and resignation when accepted as an employee.These behaviors are highlighted by tweets discussing "ceremonies instead of tweeting," buying coffee or praying until lunch break, sleeping while working, freelancing during working hours, playing games on cell phones, as well as downloading and watching movies.
The public consistently monitors ASN behavior, which helps to shape work behavior and impact organizational achievement.Based on this research results, ASN behavior is classified as unfavorable and is a response to various work events in the government.Tweets with negative sentiments lead to negative work behavior, including decreased performance, negative job attitudes, increased job stress, and withdrawal behavior.Conversely, tweets with positive sentiment led to positive work behavior, including improved performance, positive job attitude, decreased job stress, and decreased withdrawal behavior.
Based on Table 4, per 1,000 events, increased performance occurs in 92 events, positive job attitude appears in 141 events, decreased job stress occurs in 28 events, and withdrawal behavior falls in 13 events.Conversely, decreased performance was found in 165 events, a negative job atti-tude was shown in 287 events, increased job stress in 218 events, and increased withdrawal behavior in 164 work events.
The categories of work events triggering negative emotions, thereby reducing employee performance, are mostly found in the excessive workload category.This is due to the experiences conveyed through tweets, which are often unpleasant experiences or something related to less-than-optimal work results.This is also related to the unequal distribution of workload in government, which causes a person to be busier and vice versa.These results support Silva et al. ( 2022) and Ali et al. ( 2021) that the inappropriate distribution of workload will cause employee emotional exhaustion.In addition, this research supports the results of Poulose & Dhal (2018) that the distribution of workload is uneven, and excessive workload for employees can cause a decrease in one's career commitment.One of the causes of the uneven workload distribution discussed in the tweet is the competence or ability of each employee.This is indicated by the tweets stating that the more competent the employee is, the more work the employee is given to carry out.Therefore, employees tend to hide their competence.This result is also in line with research conducted by Wang (2022).
Conversely, the lack of roles and workload for employees can also trigger an increase in job stress.Shultz et al. (2010) stated that lacking employee roles could trigger increased stress and boredom, as well as an increase in withdrawal behavior.According to Tetteh et al. ( 2020), job stress provokes employees to engage in withdrawal behavior, which is also supported by Epita  (2013).An example of withdrawal behavior is absenteeism, which occurs because employees often do not have tasks, the government's work system and culture are poor, as well as the reward and punishment system does not apply.
The negative emotions correlating with negative job attitudes occur due to an abuse of position/authority, an uncooperative work environment, and a lack of respect and appreciation for the job.These results support Asbari et al.'s (2019) finding that positive organizational culture correlates with innovative employee behavior and vice versa.However, negative job attitudes are insufficient to trigger withdrawal behavior.
On the other hand, the categories of work events that lead to positive emotions can improve performance significantly when compensation increases.This is in line with Do (2018), Arnolds &Boshoff (2011), andDwianto et al. (2019).Regarding positive emotions related to job stress, reductions are triggered by the adjustment of workload to the qualifications, competencies, and principal tasks and functions.Meanwhile, a supportive work environment can trigger a positive job attitude.The results of this mapping can be used as a guideline for improving ASN work behavior.

CONCLUSION
The ASN showed negative work behavior due to emotions in response to various work events.This research showed that the most frequent work events involve low employee workload, leading to withdrawal behavior.This negative work behavior can also happen in other work events, causing decreased performance, negative job attitudes, increased job stress, and withdrawal behavior intentions.Meanwhile, work events that generate positive emotions can shape positive work behavior, including improved performance, positive job attitude, decreased job stress, and decreased withdrawal behavior.
Despite being conducted as well as possible, this research still has several limitations that need to be considered so that it can be improved in future research.First, researchers used social media data from interactions with the @PNS_Ababil account.Social media, especially Twitter in Indonesia, tends to be used to express various feelings one experiences.However, the feelings expressed are often disappointments or complaints.Tweets containing controversy often get a lot of interaction, causing them to go viral.Additional data needs to be collected from hashtags or a combination of several keywords to increase the obtained data and diversity of emotions.Second, this research would be considerably better if it could show the networking interactions arising from the @PNS_Ababil account.

Figure 4 .
Figure 4. Screenshots of work events that get the most interactions

Figure 6 .
Figure 6.Sentiment analysis on ASN's work behavior

Figure 7 .
Figure 7. Word cloud visualization of negative and positive sentiments

Table 2 . Data collection and reduction process
Interaction in this research is formulated as the sum of likes, retweets, and replies.