Silent Users on Social Media-Why they are Silent? How to Infer Their Opinion?
Silent Majority and Vocal Minority are Significantly Different
- “We present results of data analysis that compares two groups of different users: the vocal minority (users who tweet very often) and the silent majority (users who tweeted only once). We discover that the content generated by these two groups is significantly different“
- Difference in ditigal age: “members of the Silent group have been using Twitter for a longer time than those of the Vocal group”
- Difference in online behaviour: “The vocal minority users link more to outside content, use more hashtags, and retweet more, all activities intended to broaden the impact and reach of tweets”
Why Silent? Key Theories
Spiral of Silence
What is spiral of silence
- “To the individual, not isolating himself is more important than his own judgment.” – “More frightened of isolation than of committing an error, they joined the masses even though they did not agree with them”
- “Thus the tendency of the one to speak up and the other to be silent starts off a spiraling process which increasingly establishes one opinion as the prevailing one.”
Key assumptions
- Threat of isolation
- Fear of isolation: conformity
- Quasi-statistical sense: “As a result of fear of isolation, individuals constantly monitor their environment to check on the distribution of opinions as well as the future trend of opinion” (Scheufle & Moy, 2000)
- Willingness to speak out and tendency to remain silent: “Individuals tend to publicly express their opinions and attitudes when they perceive their view to be dominant or on the rise. In contrast, when people sense their view is in the minority or on the decline, they become cautious and silent” (Scheufle & Moy, 2000)
Quantitative evidence
Association with social capital
- individual-level indicators (civic engagement, trust, and neighborliness) of social capital are associated with willingness to express opinions
- individual-level indicators of social capital are associated with the perception that others support one’s opinions
- perceived support for one’s opinions mediates the proposed relationship between individual-level social capital and willingness to express opinions
Association with cultural predispositions and news attention
- The concept of face refers to one’s social worth and image in social interactions
- fear of isolation and saving face were negatively associated with individuals’ willingness to express their opinion on the issue
- whereas news attention and issue salience were positively associated
- fear of isolation was negatively associated with individuals’ willingness to offer a rationale for their opinion
Association with types of public (situational theory)
- active publics are more likely than other types of publics to express their opinions
- journalists who perceived a greater discrepancy between their opinions and the opinions of Twitter users about controversial issues in South Korea were less willing to voice their opinions on Twitter
- politically conservative journalists were more likely to perceive that their opinions were in the minority; therefore, they were less likely than politically liberal journalists to discuss their opinions on Twitter because use of that particular technology is generally regarded in Korea as favored by liberals
Association with opinion climate
- those who hold strong opinions are more likely to comment when they perceive the opinion climate to be oppositional rather than supportive to their worldview
Spiral of silence and social media
- in contemporary social networking websites, wherein users commonly face a personally relevant audience, people are prone to hold back their opinion as they expect losing control over the reactions of their audience
- In both personal settings and online settings, people were more willing to share their views if they thought their audience agreed with them.
- Previous ‘spiral of silence’ findings as to people’s willingness to speak up in various settings also apply to social media users. Those who use Facebook were more willing to share their views if they thought their followers agreed with them.
- This survey shows how the social and political climate in which people share opinions depends on several other things:
- Their confidence in how much they know. Those who felt they knew a lot about the issues were more likely than others to say they would join conversations.
- The intensity of their opinions. Those who said they had strong feelings about the Snowden-NSA matter were more willing
- Their level of interest. Those who said they were very interested in the Snowden-NSA story were more likely
- fear of social isolation has an indirect effect on discouraging disagreeing opinion expression but not supporting opinion expression
- fear of social isolation also has an effect on encouraging withdrawl behaviors through enhancing willingness to self-censor
Self-Censorship
- Self-censorship: the withholding of one’s opinion around an audience perceived to disagree with that opinion
- some people rely on information about the climate of opinion more so than do others when they decide whether or not to voice their opinion publicly
- People who are relatively more influenced by the climate of opinion when choosing whether or not to voice an opinion, are also “relatively less likely to engage in public political activities.”
Other Mechanisms
Censorship
- people tend to stay silent when the censorship in the global environment is intensive, whereas they tend to “rebel” against censorship by voicing their opinions, when they experience censorship themselves or witness censorship occurring to their friends or reference persons
- Outspoken crowd could shield individuals from the fear of punishment and outspoken friends could mitigate individuals’ anger against censorship
Network structure
- being active or silent in the social media environment is largely conditional on the surrounding network structures
- individuals who are well-connected directly or indirectly with others, and/or merely belong to well-connected local communities are likely to post more messages
- Please note that this conclusion reflects correlation, not causation
How to Infer their Opinion?
Use the Network Structure
Label Propagation
- use graph methods to propagate the found opinions of users to silent participants
- we can re-introduce the stance inference problem in our dataset as identifying the label of the node based on the influence it receives from surrounding labels of a graph
- step 1 seed label: “we use weak labels coming from the participating users as they have linguistic clues”; hashtags are used as contexts
- step 2 label propagation
- Data: Twitter - misinformation relating to important medical facts: masks and vaccinations
- Performance: 0.94 F1 score vs 0.49 F1 score (baseline - clustering)
- we infer the opinions of silent users by leveraging the text content posted by active users and their relationships between silent users
- Theoretical base: existing studies on social science theories have verified that users that are connected (become friends) are prone to exhibit similar opinions on a certain topic
- Data:
- content features: message posted by oneself and other users in the same community
- network topology
- Method:
- Clustering: users are clustered into communities
- intra-community structural features: degree, shortest path, average distance, Katz centrality
- inter-community structural features
- Matrix Factorization: we design a coupled sparse matrix factorization (CSMF) model to capture the complex relations among these features
- Clustering: users are clustered into communities
- Performance: our CSMF model achieves over 80% accuracy for the inference of silent users’ opinions
- we propose a novel framework, named SOCIALSENSE, that leverages a large language model to induce a belief-centered graph on top of an existent social network
- Method:
- building a belief-centered network on top of the existing social network
- propagating information across multiple levels
Collaborative Filtering (User Homophily)
- this study proposes and evaluates a new approach for silent users’ stance prediction based on collaborative filtering and Graph Convolutional Networks, which exploits multiple relationships between users and topics
- Data: we examine user attitudes leading to the Chilean constitutional referendums in 2020 and 2022 through extensive Twitter datasets
- Performance: outperforms the baselines by over 9% at the edge- and the user level
- Method: identifying silent users’ engagement patterns with different topics and similarities to other, more active users
- collaborative filtering: predict users’ association with a topic and the various perspectives within each discussion based on hashtags
- Weighted-LightGCN: the model jointly updates the representations of users and hashtags by aggregating the neighbors’ features
Use the Text on Other Topics
- Opinion Survey
- we conduct an opinion survey on a set of users for two popular social media platforms, Twitter and Facebook
- more than half of our users who are interested in issue i are i-silent users in Twitter
- i-silent users are likely to have different opinion distribution from the users who post about i
- Opinion Prediction
- Data: sentiment features (from user’s content) and opinion features (from user’s predicted opinion or opinion on other issues)
- Performance: predicting i-silent users’ opinions can achieve reasonably good accuracy from user posted content that is not related to issue i, and achieve better accuracy when we make use of user opinions on other issues
Anomaly Detection
- This work focuses on predicting the absent links in the retweet graph
- They formulate a joint fact-finding and silence interpretation problem, and shows that the joint formulation significantly improves our ability to distinguish true and false claims
- An unsupervised algorithm, Joint Network Embedding and Maximum Likelihood (JNEML) framework, is developed to solve this problem
- Input: dataset of posts - a content (assertion), a type (retweet-a measure of joint interest), a source (user)
- content(post/retweet) relations
- semantic similarity between assertions
- Anomaly detection: the target is to identify normal or anomalous silence
- The target is interpreted as a network embedding problem
- To find what is “normal” for a user to post on, and what is “normal” for a user to retweet
All articles on this blog are licensed under CC BY-NC-SA 4.0 unless otherwise stated.
Comments