Changes in the discourse of online hate blogs: The effect of Barack Obama's election in 2008
First Monday

Changes in the discourse of online hate blogs: The effect of Barack Obama's election in 2008 by Shlomi Sela, Tsvi Kuflik, and Gustavo S. Mesch



Abstract
This study examines the narrative strategies that the blogs of hate groups adopted before and after a central political event, namely, the 2008 election of President Obama in the U.S. Using data from a large number of hate blogs (N=600), and sentiment analysis and data mining, we tested two alternative hypotheses derived from social identification theory. We found that there were major differences between the content of these blogs before the election and immediately after the 2008 election, with the latter evincing an increase in the advocacy of violence and hostility. We also determined that faced with this new change, the hate groups adopted a social competition strategy rather than a creativity strategy to manage their identity. Our findings imply that since the election of Barack Obama as President, the worldview of online hate groups has become more violent. The implications of the findings are discussed.

Contents

Introduction
Literature review
Research goals and questions
Data and methods
Findings
Discussion and conclusion

 


 

Introduction

Together with the emergence of the Internet and its rapid adoption in every domain of our lives, there has been an increasing presence of hate groups, groups in which members are socialized and indoctrinated into attitudes and norms that devalue other groups based on their religion, race, ethnicity and gender [1]. According to the Southern Poverty Law Center (SPLC, 2005; see http://www.splcenter.org/) the number of hate groups operating in the United States increased by more than four percent in 2008 alone and has grown by 58 percent since 2000, reaching 1,002 active hate groups in 2010. Hate groups often use weblogs to establish a salient presence online, spread their worldview, and recruit new members to their activities (Douglas, et al., 2005; Gerstenfeld, et al., 2003). The groups most often targeted in these blogs are Blacks, Jews, Muslims, gays, women and immigrants (Ballard, et al., 2002; Jaishankar, 2010; Stanton, 2002).

The purpose of this study is to understand how the discourse in weblogs of hate groups has changed over time. Our study can be seen as a natural experiment that takes advantage of a unique historical event, namely, the election of President Barack Obama, the first African American president in the U.S. Many saw his election as potentially affecting attitudes toward racial and ethnic groups. As The American Civil Rights Education Fund (LCCREF; see http://www.civilrights.org/) emphasized, for many, the election of President Obama appeared to be a central event in the long struggle for racial equality in the U.S. (see http://www.civilrights.org/publications/hatecrimes/executive-summary.html). Our study investigates whether the election results changed the strategies that hate groups adopt in online communication, specifically, their weblogs. We used sentiment analysis techniques to distinguish changes in wording and emotions expressed in the textual content of hate blogs, one year before the election and immediately after the 2008 election.

 

++++++++++

Literature review

Barack Obama’s election as U.S. President marks a historical shift in the political life of the United States (Harris and Davidson, 2009). During the campaign and the election, the media highlighted the characteristics of the candidate in both positive and negative ways. On the positive side, in describing Obama as hard working, intelligent, family oriented and respectful of the law, the media presented him as a counterexample to the common stereotypes about African Americans (Welch and Sigelman, 2011). On the negative side, some made efforts to highlight controversial signals by referring to him (with or without emphasis) by his full name, including his middle name, Hussein. His campaign opponents used this strategy to play on concerns about a black presidential candidate with a foreign–sounding name, a non–American father, and some periods of residency outside the United States (Waltman and Haas, 2011). Some hate groups depicted the candidacy of an African American Presidential candidate as yet another sign that their country was under siege by non–whites (Beirich and Potok, 2009; Sternberg and Sternberg, 2008; Willis–Esqueda, 2008), Given the emphasis during the campaign and afterwards on both the positive and negative aspects of the candidate, an important question to ask is whether the narrative of online hate groups changed and if so, in what direction.

From a social–psychological perspective, social identity theory (SIT) provides a useful framework to examine the process of group identity creation and change (Blanz, et al., 1998; Haslam, 2001; Tajfel and Turner, 1979; van Knippenberg, 1989). A social identity is the portion of an individual’s self–concept derived from perceived membership in a relevant social group. SIT has been used to explain the process of identification with social groups, including ethnic and social groups, and intergroup behavior in terms of loyalty to social identities. Thus, SIT is best described as a perspective that seeks to explain intergroup behaviors on the basis of the perceived status, legitimacy and permeability of the inter–group environment.

According to SIT, social change affects inter–group relations. Groups may adopt different strategies for self–enhancement and self–distinctiveness in relation to other groups, but the selection of a strategy depends on their status, norms, and political events. Hence, SIT explains individual responses to changes in the social status of the group vis–à–vis other groups. SIT posits that there are two main identity management strategies that individuals and groups can use to adapt to social changes: social competition and social creativity (Douglas, et al., 2005).

The social competition strategy is an identity management technique that tries to change the status positions of the in–group and/or the out–group in order to protect the status of the in–group (Smith, 1991). Social competition is expressed by the use of language that includes terms such as conflict, (i.e., “active self–defense”) open hostility, and antagonism. Examples of such activities are calls for political action, protests, and actions through which the groups directly challenge the out–group and its social position (e.g., Milgram and Toch, 1969; Mummendey, et al., 1999; Reicher, 2001; Tyler and Smith, 1998).

As an alternative to social competition, hate groups can adopt a social creativity strategy to maintain their positive social identity and enhance the group distinctiveness. Groups compare themselves with the in/out–groups and may change the criteria for comparison so that comparisons that were previously negative are now perceived as positive. The classic example of social creativity is that of Lemaine (1974). He found that children’s groups that could not compare themselves favorably with others in terms of constructing a hut because they had been assigned poorer building materials than the out–group, tended to seek out other criteria for comparison involving new constructions in the hut’s surroundings. Hence, the social creativity strategy has the effect of redefining the position of the group without directly challenging the out–group. As Douglas, et al. (2005) suggest, the term “creativity” does not convey any positive connotations. It simply means that the groups will engage in inter–group comparisons using unorthodox criteria that tend to favor their own group.

According to Haslam’s (2001) interpretation of the group’s strategy, the extent to which groups adopt these strategies depends on the status (low or high) of the in–group as well as on the threat from the out–group. Haslam (2001) suggests that “when status is secured, groups are likely to display magnanimity toward the out–group or covert discrimination (refers to actions that are subtle), while members of a high–status group who feel their relative advantage is under threat may band together to resist change” [2]. This resistance to change is much more likely to involve social competition than creativity, because it sets the high–status group against the out–group and directly challenges the out–group instead of accepting the status quo or circumventing the possibility of confrontation. Haslam (2001) and others (Douglas, et al., 2005; Seger, et al., 2008) argue that it is possible to predict that online white supremacist groups would engage in more social competition strategies than social creativity strategies. Their perception that their high–status position is under threat makes it more likely that they will adopt competition strategies such as conflict and violence in order to maintain the status quo or promote their group as superior to the out–group.

Douglas, et al. (2005) present one of the few studies that investigated the competition strategy in online communication. Using data from 43 white supremacist Web sites, they conclude that the level of violence advocated was minimal. Using content analysis, the study found that the social competition strategy might be associated with support for violent language. Some of the sentences they found were

“Now all the black community will be satisfied with a first black presidency.”
“To provoke your fate is a matter of time until we’re white we will return to the White House.”
“President White! Humane efforts towards separation and self–determination are better for us than endless repression, tension, and racial violence.” (post I.D 576)

The researchers discovered that white supremacist Web sites were more likely to feature social creativity strategies against targeted out–groups. Essentially, they contend that the Web site content is centered more on reframing the audience’s perceptions about the white supremacist group and less on overt campaigns against other out–groups.

An alternative hypothesis posited that after the election, even hate groups would change their view and present a more positive view of the out–group or parts of it. According to the exemplars exposure hypothesis, exposure to positive and negative exemplars has the ability to reduce automatic racial associations (Dasgupta and Greenwald, 2001). In other words, when members of an in–group are exposed to positive, high achieving, well liked members of a stigmatized out–group, their negative view of the out–group might be moderated or become less salient (Bodenhausen and Macrae, 1998). Thus, according to this view, the high level of media exposure to the President as a positive African American exemplar and a source of counter–stereotypic information could have produced positive shifts in attitudes toward African Americans. There is some empirical support for this perspective. A longitudinal study of the perceptions of white people about blacks with regard to how hardworking and intelligent they were found that before the election, their mean assessment of black people’s work ethic was slightly on the “lazy” side of the neutral point. However, in the election year, it crossed the midpoint to the hardworking side of the scale. The study concludes that white attitudes about the work ethic of blacks as well as their assessment of their intelligence increased slightly from before the elections to after the election of Obama as President (Welch and Sigelman, 2011).

Another study found that changes in negative attitudes to African Americans might depend on the perception of the election results as an indication of a lack of racial discrimination. Individuals who perceived the election as not affecting racial discrimination reported lower levels of racial resentment than individuals who perceived the result of the election as an indication of a dramatic reduction in the discrimination against minorities (Valentino and Brader, 2011). Thus, the exemplar exposure hypothesis and these studies indicate the possibility that the discourse of hate blogs might also have changed after the election, resulting in the use of more positive expressions or less negative expressions about the out–group (African Americans). These expressions relate to both the emotions and behaviors of the hate groups.

 

++++++++++

Research goals and questions

This study examines the type of strategy that the blogs of hate groups adopt, using Barack Obama’s election as a case study. Using sentiment analysis, we assess the strategies that the groups used in their discourse one year before the election and immediately after the election of Obama as President. In our assessment, we seek to determine whether changes occurred in their discourse after the election.

Sentiment analysis is the processing of ”a set of search results for a given item, generating a list of product attributes (quality features, etc.) and aggregating opinions about each of them“ (Eguchi and Lavrenko, 2006). It deals with the identification, extraction and classification of opinions, sentiments and emotions that are expressed through natural language (Alm, et al., 2005; Yang, et al., 2007). Since 2002, scholars have focused their efforts on the task of analyzing how positive or negative sentiments expressed in documents are by analyzing the words that are used in the text (Sood and Churchill, 2010).

Based on the literature survey above, Table 1 illustrates the main strategies that various hate groups adopt and provides some examples of their statements. The first column presents the strategy type, the second column indicates the time frame as before or after the election, the third column presents the expected type of strategy and some explicit examples, and the last column shows the linguistic features that might be related to each strategy.

 

Table 1: The expected reflection of social competition and social creativity in hate blogs.
Type of strategySituationExampleContent exampleLinguistic features
(This example taken from WordNet 2.1)
Social sreativity (out–group strategy, inferior group)Before the 2008 electionGroups compared themselves with the in/out–groups. Groups using low–prestige language or dialect or a once derogatory word (e.g., nigger/niggard, queer, dyke, etc.)“We, the blacks are more superior that the white people”.
“We have our president in the White House”.
The group will emphasize negative emotions, but not negative behavior: conflict, offend, debate, infract.
Social competition (in–group strategy, superior group)After the 2008 electionPolitical action, protest, debates, support for violence, advocate violence.“Active self–defense is the answer ... .”
“The white people must be awakened”.
“We are at war because of Obama”.
The group will emphasize negative emotions and direct negative behavior: kill, crimes, attack, racist, violence etc.

 

Our goal is to explore the possibility of detecting a hate group’s strategies by analyzing the textual content of their blogs. The specific research questions address the group’s identity management strategies within the hate blogs:

RQ1: What are the terms (words) that represent and characterize hate blog posts one year before the election and immediately after Barack Obama’s election in 2008?
RQ2: What type of strategy (creativity or competition or other scenarios as described above) did hate groups adopt one year before the election and immediately after Barack Obama’s election in 2008?

 

++++++++++

Data and methods

We created the blog corpus using data collected from hate blogs. We chose blogs as a source because they are rich in emotions and attitudes. The data include content from hate blogs, regardless of the ethnic origin of the blogger. In other words, the data focus on the content associated with the U.S. election, its results and the Presidency. The data include the content of 600 hate blogs collected from blogging communities (at www.xanga.com) that are considered to be a good representation of hate blogs (Chau, et al., 2009). Following the method of Chau, et al. (2009), we used the search feature in Xanga to semi–automate the task. Examples of the terms used to identify the relevant content include “2008 Presidency”, “Obama’s Presidency”, “President Barack Obama”, “Hussein Obama”, “white pride”, and “white supremacist”. We used these terms to search for groups (blog rings) in Xanga that had any of these words in their group name or description. In addition, we also investigated them manually and included only those that explicitly mentioned hatred through phrases such as “we hate blacks”, “we hate Obama”, “Hussein Obama”, “Kill Obama”, “Obama’s election”, “Our president”, “we hate white people”, “we hate black people”, “hate the black race”, etc. In addition, we collected data at two different periods of time: before the Presidential elections (from July 2007 to July 2008, N=305 blogs) and after the election (from August 2008 to January 2009, N=295 blogs). We also used IBM–SPSS Text Mining and Data Mining technology (PASW–Predictive Analytics Software). As Figure 1 shows, the experimental system is comprised of two main components: a hate groups lexicon and a natural language processing (NLP) component.

 

Hate blogs analysis framework
Figure 1: Hate blogs analysis framework.

 

The built–in lexicon

The system has a built–in lexicon that contains terms specific to the area of interest and their attributes, which are used later in the NLP task (for analyzing the posts in our case). The lexicon includes a dictionary containing a list of base forms of terms with a part–of–speech code (noun, verb, adjective, adverb, participle, coordinator, determiner, or preposition). The lexicon also includes reserved, built–in types, where “type” refers to a semantic grouping of terms. For example, the ”teacher“ type group’s terms are: “professor”, “faculty”, “instructor”, and “trainer” (Yu, 2009).

In our study, the built–in lexicon contains 12,363 terms that are associated with eight types of attributes: Positive/Negative emotions (9,678 terms), Positive/Negative behavior (562 terms), Location (642 terms), Organization (874 terms), and Person (611 terms). Table 2 includes examples of the relevant terms for our study. As it shows, in the built–in lexicon “violence” and “hostility” are tagged as negative behavior. We also provide some examples of the words classified as negative emotions. The built–in lexicon is suitable for our task because it contains a wide variety of terms that measure positive and negative emotions and behaviors.

 

Table 2: Linguistic features of hate blogs.
Main entryPart of speechDefinitionTypeSynonyms
ViolenceNounExtreme force, IntensityNegative behaviorabandon, acuteness, assault, attack, bestiality, bloodshed, blowup, brutality, brute force, clash, coercion, compulsion, confusion, constraint, cruelty, destructiveness, disorder, disturbance, duress, ferocity, fervor, fierceness, fighting, flap, foul play, frenzy, fury, fuss, harshness, murderousness, onslaught, passion, power, raging, rampage, roughness, ruckus, rumble, savagery, severity, sharpness, storminess, struggle, terrorism, tumult, turbulence, uproar, vehemence, wildness.
HostilityNounAntagonism, meannessNegative behaviorabhorrence, aggression, animosity, animus, antipathy, aversion, bad blood, bellicosity, belligerence, bitterness, detestation, disaffection, enmity, estrangement, grudge, hatred, ill will, inimicality, malevolence, malice, opposition, resentment, spite, spleen, unfriendliness, venom, virulence, war, warpath.
DebateNounDiscussion of issues; considerationNegative emotionagitation, altercation, argument, argumentation, blah–blah, cogitation, contention, contest, controversy, controverting, deliberation, dialectic, disputation, dispute, forensic, hassle, match, meditation, mooting, polemic, rebutting, reflection, refuting, tiff, words, wrangle

 

The second component of the experimental system (see Figure 1) is the NLP component that includes pre–processing, concept extraction and semantic grouping. The main purpose of this stage is to extract the categories that represent the sampling corpora. In the pre–processing step, data were converted to a uniform format (free text), and text was converted into individual elements from which concepts were extracted (Feldman and Sanger, 2007).

As an input at this stage, we used the data sets: “before” and “after” elections. Next, we executed the concept extraction. A typical extraction process returns two kinds of results: terms and types. Terms are words or phrases that carry important connotations, and types are semantic groupings of terms. For example, consider the “hate” type groups terms: “victim,” “lesbian,” “minority,” and “Jew”. Hence, concepts are a collection of key terms that are actually found in the data (Yu, 2009).

Finally, semantic grouping is the process of bundling concepts with similar content into categories. Concept extraction found a large number of concepts, so we categorized them in an attempt to discover combinations of emotionally charged topics and the relationships among them. Grouping by category allowed us to investigate which categories appeared together and to determine their links with one another. To create the categories, we used a semantic network technique that links related terms using parent/child and peer/sibling grouping relations (Yu, 2009). This technique builds categories using a built–in network of word relationships based on WordNet (see http://wordnet.princeton.edu/).

At the end of this stage, we extracted two sets of categories: one for the “before the election” and the other for the “after the election”.

Table 3 presents a summary of the two–step experimental process we used to determine the identity management strategy of various hate groups as expressed in their blogs. In accordance with SIT, we then compared them one year before the election and immediately after the election. Using sentiment analysis, we sought to determine whether there was a change in the content of the hate blogs before and after the election, and if the content analysis of the blogs reflected changes in the groups’ strategies over time.

 

Table 3: Summary of research steps.
StepGoalPartTechnique
1Identifying the concepts that represent social competition vs. social creativityPart 1: Distribution of concepts in hate blogs before and after election
Part 2: Comparing the distribution of emotions and behavior type
NLP
2Grouping the concepts into categoriesPart 1: Grouping concepts into categories and distribution of emotions
Part 2: Comparing the distribution of emotions and behavior type, with respect to categories
NLP

 

 

++++++++++

Findings

The first step was to examine how the concepts that represent hostility and violent language appear in hate blogs and the changes in the content, before and after the election. In Table 4, we present a few examples of the concepts extracted (due to the limitations of space, we are not able to present all of the results). The second step was to group the concepts into the various categories. This step had two parts. First, we examined the differences between hate blogs, before and after the election, with respect to emotional categories. Then, we compared the types of emotional and behavioral categories identified in the first part. This comparison allowed a more precise differentiation of the emotional world of hate groups, before and after the election.

 

Table 4: Distribution of concepts before and after the election.
 Hate groups before the electionHate groups after the election
ConceptGlobalDocsTypeConceptGlobalDocsType
1like8343
(26%)
2576
(44%)
P.Ewe13222
(20%)
6755
(54%)
P.E
2we3032
(9.4%)
1605
(29%)
P.Ethey9901
(15%)
4763
(39%)
N.E
3well2655
(8.2%)
1567
(28%)
P.Edislike7633
(11.6%)
2458
(32%)
N.E
4they2133
(6.6%)
1360
(24%)
N.Elike6744
(10%)
2145
(28%)
P.E
5good1689
(5.2%)
1188
(21%)
P.Egroups5466
(8.3%)
1326
(17%)
N.E
6bad1618
(5%)
1119
(20%)
N.Epeople4322
(6.5%)
1274
(16%)
N.E
7dislike1442
(4.5%)
993
(18%)
N.Eviolent2744
(4.1%)
1166
(15%)
N.B
8stars1389
(4.3%)
1034
(18%)
Unknownexcellent2321
(3.5%)
985
(13%)
P.E
9ok1358
(4.2%)
674
(12%)
P.Enew2201
(3.3%)
912
(12%)
Unknown
10people1321
(4.1%)
743
(13%)
Unknownmore1975
(3%)
792
(8%)
N.E
11excellent1316
(4%)
713
(13%)
Unknownright1875
(2.8%)
787
(8%)
P.E
12life1301
(4%)
668
(12%)
Unknowngood1733
(2.6%)
733
(8%)
P.E
13don’t cognitive1223
(3.8%)
680
(12%)
Unknownlife1564
(2.3%)
687
(7%)
N.B
14more1120
(3.5%)
549
(10%)
N.Ehostility1463
(2.2%)
688
(7%)
N.B
15horrible1098
(3.4%)
512
(9%)
N.Erace1344
(2%)
661
(7%)
N.E
16black1023
(3%)
351
(6%)
N.Efunction1201
(1.8%)
646
(7%)
Unknown
 Index: N.E — negative emotion; P.E — positive emotion; N.B — negative behavior; P.B — positive behavior; Unknown

 

Table 4 lists the distribution of the 16 most frequent concepts in the hate blogs before and after the election. By concepts, we mean the words or phrases belonging to a group of terms with the same meaning. For each period of time, the table shows the name of the concept, the global frequency of the concept (the total number of times a certain word appears within the entire set of posts), the document (Docs) frequency of the concept, (the total number of posts in which a certain concept appears) and its emotional meaning, (positive/negative emotion, positive/negative behavior) as defined by the system (based on the built–in lexicon described earlier). For example, the “like” concept has a global frequency of 8,343, a document frequency of 2,576 and a positive emotional (P.E) meaning.

As Table 4 demonstrates, the two corpora differ in the frequency of their concepts. The concepts “we” and “they” appear both before and after the election as do the concepts “good”, “people”, “excellent”, “life”, and “dislike”. In addition, it is interesting that the concepts “violent” and “hostility”, which are associated with the advocation of violence, appear only after the election. It may be reasonable to conclude that violent and hostile content, and the desire to isolate oneself from society is more frequent after the election. The literature also supports these findings (Douglas and McGarty, 2002; Douglas, et al., 2005; Haslam, 2001; Tajfel and Turner, 1979) and provides some indication that social competition occurred because the election changed the group’s status. To confirm the discrepancy between the two corpora in terms of concept frequency, we conducted a two–sided t–test. We found a significant difference between the two corpora (p > 0.05) before and after the 2008 election.

Next, we compared the distribution of types of emotions expressed by the concepts before and after the election in more detail. The goal of this step was to measure the emotional and behavioral concepts in each corpus and assess the differences between the corpora. To this end, we collected and aggregated the types of emotional and behavioral concepts that the system suggested during the analysis process (presented in Table 4). In the hate group’s corpus, before the election, six out of 16 concepts (37.5 percent) were classified as negative emotions, and six out of 16 concepts (37.5 percent) were classified as positive emotions (37.5 percent), while four (25 percent) were unknown. After the election, six out of 16 concepts (37.5 percent) were classified as negative emotions, and five out of 16 were classified as positive emotions (25 percent), two as negative behavior (12.5 percent) and three (18.7 percent) as unknown. Figure 2 illustrates the distribution of emotions and behavior in hate blogs before and after the election.

 

Distribution of concepts with positive and negative emotions and behavior
 
Figure 2: Distribution of concepts with positive and negative emotions and behavior.

 

According to Figure 2, 14 percent of the hate posts after the election are tagged as negative emotions in comparison to the 8.2 percent of negative emotions that appeared in the before–the–election corpus. Moreover, the after–the–election corpus contained 10 percent positive emotions in comparison to 17 percent positive emotions in the hate blogs before the election. The positive and negative behaviors in all of the corpora also differ significantly depending on the time period. The after–the–election corpora are more associated with attitudes expressing negative behavior (7.2 percent) than the before–the–election corpus in which only two percent of postings related to negative behavior. To summarize, Figure 2 clearly shows that there are differences between the before and after election data in terms of the expression of positive and negative emotions and behavior. After the election, we can see many more postings about negative behavior emotions and far fewer expressions of positive behavior and emotions than before the election.

Grouping the concepts into categories

The goal of the second step was to group the concepts together into - categories of terms having similar emotional meaning in order to determine differences in those characterized by hostility and violence, or conflict language before and after the election. First, we grouped the individual terms identified in the previous step into categories. As Table 5 demonstrates, the before–the–election data included 22 categories, while the after–the–election data included 29 categories. In the before–the–election data, there were 10 categories expressing negative emotions, 10 categories expressing positive emotions, one category expressing negative behavior, and one category expressing a mixture of positive and negative emotions. However, after the elections, there were nine categories expressing negative emotions, nine categories expressing positive emotions, one category expressing positive behavior and 10 categories expressing negative behavior. Comparing the two sets of results shows us a drastic tilt towards negative behavior from before the election to after. Figure 3 shows the differences between the before and after election data in terms of negative behavior.

In both cases, the most frequent category is ”person of color“, which is classified as a negative emotion type. In the before–the–election data, this term is associated with 92 concepts that appear in 502 documents. The concept is associated with “poor black”, “black violence drives”, “terror of blacks”, “black skin background”, “black racists negative emotion”, “black president”, “etherification of blacks”, etc. It is important to note that the category “violence” appears only in the after–the–election data and includes 57 concepts that appear in 300 documents. Examples of such categories include expressions such as “violated election laws”, “victims of violence”, “sparked Muslim violence”, “nonsense violence”, “fear of violence”, “rate of violence”, etc. The category “kill” is associated with concepts such as “kill American citizens”, “kill Obama”, “afghan kill”. From this description, it seems that after the election, the hate groups became more aggressive, evident in their increased expression of negative behavior.

 

Table 5: Distribution of concepts before and after the election.
 Hate groups before the electionHate groups after the election
#CategoryTypeConceptsDocsCategoryTypeConceptsDocs
1person of colorN.E92
(19.8%)
502person of colorN.E308
(15.4%)
851
2badN.E45
(9.7%)
1232racistN.E159
(7.9%)
493
3satisfiedP.E35
(7.5%)
563freeP.E123
(6.1%)
495
4freeP.E25
(5.3%)
143attackN.B118
(5.9%)
381
5competitionP.E+N.E25
(5.3%)
83killN.B109
(5.4%)
901
6niceP.E25
(5.3%)
263dishonorN.E107
(5.3%)
399
7helpP.B24
(5.1%)
184rightP.E87
(4.3%)
799
8horribleN.E24
(5.1%)
556crimesN.B59
(2.9%)
502
9terminationN.B22
(4.7%)
386weP.E59
(2.9%)
3546
10racistN.E22
(4.7%)
71problemN.B59
(2.9%)
683
11crazyN.E15
(3.2%)
181violenceN.B57
(2.8%)
300
12rightP.E15
(3.2%)
316hostilityN.B54
(2.7%)
98
13interestP.E14
(3%)
113wrongN.E52
(2.6%)
268
14sweetP.E13
(2.8%)
178popularP.E52
(2.6%)
268
15wittinessP.E12
(2.5%)
385favoriteP.E51
(2.5%)
163
16wrongN.E12
(2.5%)
208legalP.E50
(2.4%)
144
17goodP.E12
(2.5%)
1277fearN.E47
(2.3%)
214
18conservativeN.E10
(2.1%)
40horribleN.E47
(2.3%)
179
19raceN.E10
(2.1%)
103economicN.E46
(2.2%)
160
20dirtyN.E10
(2.1%)
39interestP.E45
(2.1%)
688
21theyN.E1
(0.2%)
1360terrorismN.B43
(1.9%)
113
22weP.E1
(0.2%)
1605unfortunate personN.B43
(1.9%)
102
23 fastP.E39
(1.6%)
256
24proP.E38
(1.5%)
106
25activeP.B37
(1.4%)
270
26releaseN.B36
(1.3%)
260
27poorN.E35
(1.2%)
84
28destructionN.B33
(0.9%)
73
29theyN.E1
(0.05%)
2553

 

To confirm the differences between the two corpora in terms of the frequency of the categories, we conduct a two–sided t–test. We found a significant difference between the two corpora (p > 0.05) before and after the 2008 election. Note that in Table 5, which presents this comparison, there are some common categories (marked in yellow) in both data sets: “we”, “they”, “horrible”, “right”, “person of color” and “racist”.

Next, we compared the distribution of types of emotions expressed by the categories before and after the election in more detail. Looking at Figure 3, we did not find a significant change between positive and negative emotions, positive behavior, before the elections and after the election. However, we found significant change between negative behavior before the elections (one percent) Compared to after the elections (10 percent).

 

Distribution of categories Positive and negative emotions and behavior
 
Figure 3: Distribution of categories: Positive and negative emotions and behavior.

 

At this point, having assessed the frequency of the various categories before and after the election, we can identify the strategy used in the blogs before and after the election. Here are two examples of the posts in the data associated with the competition strategy and the creativity strategy.

The competition strategy after the election

Barack Obama Has Soviet Russian Czars on his Staff?
Barack Obama has destroyed our economy, health care system and long–standing societal accord with his lecherous presidency, and there is one thing many of us have missed: Obama is also a Soviet communist. It is one thing to have socialist tendencies, such as duplicitous leaders like Franklin Trotsky Roosevelt and Bill Clinton, but an entirely different ballgame when a sitting president creates SOVIET POSITIONS right into our government. I pray for you to sit down before a reveal this, in case you missed it and it is coming new to you. Barack Hussein Obama has created the position of ‘Czar’ within his presidential staff. Seriously. Czars. It is necessary to take him down (September 20, 2009).

The creativity strategy before the election

True Core Agenda is to Stop the Census Bureau’s Prediction of Majority Minority Country by 2040
The key factor behind the ANTI Immigration Reformers agenda is this: The Census Bureau calculates that by 2042, Americans who identify themselves as Hispanic, Black, Asian, American Indian, Native Hawaiian and Pacific Islander will together outnumber non–Hispanic whites. In other words, America is rapidly evolving into a multi–ethnic, multi–racial, multi–cultural society. The ANTI Core Agenda is to STOP this prediction from coming true and maintain a White, Northern European ethnicity majority as long as possible. The ANTIs are fully aware their ability to stop this change would be like stopping a runaway train. Four years ago, officials had projected the shift would come in 2050. The main reason for the accelerating change is significantly. It is our interest and our right to protect our country (January 29, 2009 9:41 PM).

Thus, Figure 3 supports our claim that after the election, the hate groups adopted a narrative with more expressions of negative behavior.

 

++++++++++

Discussion and conclusion

The purpose of this study was to examine the narrative strategies used by hate groups that are active in the blogosphere. For this study, we collected data on racial attitudes expressed in a large number of blogs at two points in time: before the 2008 election and immediately after the election. Using the innovative techniques of sentiment analysis, we determined the distribution of words used to express positive and negative emotions and behaviors about racial and ethnic relations as expressed in the blogs of hate groups. Our study design allowed us to investigate changes in these expressions over time.

Our first aim was to investigate whether the emotions expressed in the blogosphere changed as a result of the election of an African American President. According to the exemplars exposure hypothesis, exposure of an in–group to the positive image of a member of the stigmatized out–group may lead the former to develop a more positive view of the latter (Bodenhausen and Macrae, 1998; Welch and Sigelman, 2011). Our study showed no support for this argument. Indeed, immediately after the election, we found a moderate increase in the frequency of words appearing in the hate blogs denoting negative emotions and even negative behaviors. After the election, the content included concepts associated with and advocating violence and negative behavior.

After observing this change in negative emotions, we relied on social identification theory to determine whether in the face of the change in society that the election results implied, the blogs adopted a social competition strategy or a social creativity strategy to manage their identity. We showed that the characteristics of the textual context of hate blogs after the election were similar to the description of the social competition strategy that appears in the literature (Douglas, et al., 2005; Haslam, 2001; Tajfel and Turner, 1979). While we found evidence for both social competition and social creativity (evident in the examples we presented), the appearance of the concepts that called for hostility and advocated violence after the election is clear evidence of the preference for the social competition strategy. Scholars theorize that in–groups who have suffered a social change may adopt this strategy, which leads to the exacerbation of the relationship between in–groups and out–groups.

It is important to note that our findings differ slightly from those of a previous study that investigated the competition strategy among 43 white supremacist Web sites and found minimal levels of the advocacy of violence (Douglas, et al., 2005). The difference between our results and those in that study might result from differences in sample size and the dimension of time. Douglas and colleagues studied only a few Web sites, while we included text from a large sample of bloggers. In addition, our study examined blogs at different points in time, so we were able to trace changes over time. In fact, the issue of time seems to be particularly relevant, given that the post–election expressions are significantly different from the political atmosphere before the election.

Limitations of this study

Our study has several limitations. First, it covers only the blogosphere. However, hate groups use other platforms as well such as Web sites, and textual and multimedia social network sites. Given that the detection of hate groups and the analysis of the user generated content in other platforms require different tools, it is difficult to generalize our findings to other platforms. Nevertheless, we suspect that such techniques might lead to even stronger support for our results.

Another limitation of this study relates to the fact that we relied on commercial tools (IBM–SPSS Text Mining and Data Mining technology). One of the common limitations of commercial technology lies in the fact that their algorithms are proprietary, making it difficult to understand how the machine obtained certain results, particularly when comparing the machine’s results with human judgment. Nevertheless, in practice, our study presents an innovative framework and methodology for the domain of behavioral studies. The main contribution of this research is the ability to detect the type of identity management strategy that hate groups adopt based on the textual content of their blogs. In the future, we look forward to comparative studies on other Web–based content sources. It should be interesting to see if we can apply our system to other online sources besides blogs. End of article

 

About the authors

Shlomi Sela is a Ph.D. student at the Faculty of Social Sciences at the University of Haifa. His reseach interests are machine learning, data mining, Web mining, sentiments analysis and natural language processiong. His dissertation deals with the automatic identification of hate blogs.
E–mail: selash [at] gmail [dot] com

Tsvi Kuflik is a senior lecturer and the chair of the Information Systems Department at the University of Haifa, where he leads a group that focuses on research on ubiquitous user modeling and on personalization and intelligent user interface for cultural heritage. He has worked on personalization and intelligent user interfaces for cultural heritage over the past 10 years. He has authored over a hundred technical papers and has edited several books.
E–mail: tsvikak [at] is [dot] haifa [dot] ac [dot] il

Gustavo S. Mesch is a professor of Sociology at the University of Haifa. His research interests are the social effects of Internet use, social networking and computer mediated communication and the link between online and off–line social capital..
E–mail: gustavo [dot] mesch [at] gmail [dot] com

 

Notes

1. Anderson, et al., 2002, p. 141.

2. Haslam, 2001, p. 40.

3. Douglas, et al., 2005, p. 71.

 

References

C. Alm, D. Roth and R. Sproat, 2005. “Emotions from text: Machine learning for text–based emotion prediction,” HLT ’05: Proceedings of the Conference on Human Language Technology and Empirical Methods in Natural Language Processing, pp. 347–354.

J. Anderson, L. Dyson and W. Brooks, Jr., 2002. “Preventing hate crime and profiling hate crime offenders,” Western Journal of Black Studies, volume 26, number 3, pp. 140–148.

J. Ballard, J. Hornik and D. McKenzie, 2002. “Technological facilitation of terrorism: Definitional, legal, and policy issues,” American Behavioral Scientist, volume 45, number 6, pp. 989–1,016.

H. Beirich and M. Potok, 2009. “USA: Hate groups, radical–right violence, on the rise,” Policing, volume 3, number 3, pp. 255–263.http://dx.doi.org/10.1093/police/pap020

G. Bodenhausen and N. Macrae, 1998. “Stereotype activation and inhibition,” In: R. Wyer, Jr. (editor). Stereotype activation and inhibition: Advances in social cognition. Hillsdale, N.J.: Erlbaum, volume 11, pp. 1–52.

M. Blanz, A. Mummendey, R. Mielke and A. Klink, 1998. “Responding to negative social identity: A taxonomy of identity management strategies,” European Journal of Social Psychology, volume 28, number 5, pp. 697–729.http://dx.doi.org/10.1002/(SICI)1099-0992(199809/10)28:5<697::AID-EJSP889>3.0.CO;2-#

N. Dasgupta and A. Greenwald, 2001. “On the malleability of automatic attitudes: Combating automatic prejudice with images of admired and disliked individuals,” Journal of Personality and Social Psychology, volume 81, number 5, pp. 800–814.http://dx.doi.org/10.1037/0022-3514.81.5.800

K. Douglas and C. McGarty, 2002. “Internet identifiability and beyond: A model of the effects of identifiability on communicative behavior,” Group Dynamics, volume 6, number 1, pp. 17–26.http://dx.doi.org/10.1037/1089-2699.6.1.17

K. Douglas, C. Mcgarty, A.–M. Bliuc and G. Lala, 2005. “Understanding cyberhate: Social competition and social creativity in online white supremacist groups,” Social Science Computer Review, volume 23, number 1, pp. 68–76.http://dx.doi.org/10.1177/0894439304271538

K. Eguchi and V. Lavrenko, 2006. “Sentiment retrieval using generative models,” EMNLP '06: Proceedings of the 2006 Conference on Empirical Methods in Natural Language Processing, pp. 345–354.

R. Feldman and J. Sanger 2007. The text mining handbook: Advanced approaches to analyzing unstructured data. Cambridge: Cambridge University Press, 2007.

P. Gerstenfeld, D. Grant and C.–P. Chiang, 2003. “Hate online: A content analysis of extremist Internet sites,” Analysis of Social Issues and Public Policy, volume 3, number 1, pp. 29–44.http://dx.doi.org/10.1111/j.1530-2415.2003.00013.x

J. Harris and C. Davidson, 2009. “Obama: The new contours of power,” Race & Class, volume 50, number 4, pp. 1–19.http://dx.doi.org/10.1177/0306396809102993

S. Haslam, 2001. Psychology in organizations: The social identity approach. London: Sage.

G. Lemaine, 1974. “Social differentiation and social originality,” European Journal of Social Psychology, volume 4, number 1, pp. 17–52.http://dx.doi.org/10.1002/ejsp.2420040103

S. Milgram and H. Toch, 1969. “Collective behavior: Crowds and social movements,” In: G. Lindzey and E. Aronson (editors). Handbook of social psychology, volume 4. Second edition. Reading, Mass.: Addison–Wesley, pp. 507–610.

S. Reicher, 2001. “The psychology of crowd dynamics,” In: M. Hogg and R. Tindale (editors). Blackwell handbook of social psychology: Group processes. Oxford: Blackwell, pp. 182-207.

C. Seger, E. Smith, Z. Kinias, and D. Mackie, 2008. “Knowing how they feel: Perceiving emotions felt by outgroups,” Journal of Experimental Social Psychology, volume 45, number 1, pp. 80–89.http://dx.doi.org/10.1016/j.jesp.2008.08.019

E. Smith, 1991. “Ethnic identity development: Toward the development of a theory within the context of majority/minority status,” Journal of Counseling and Development, volume 70, number 1, pp. 181–188.http://dx.doi.org/10.1002/j.1556-6676.1991.tb01581.x

S. Sood and E. Churchill 2010. “Anger management: Using sentiment analysis to manage online communities,” at http://research.yahoo.com/pub/3352, accessed 22 October 2012.

Southern Poverty Law Center (SPLC), 2005 “Hate groups, militias on rise as extremists stage comeback,” at http://www.splcenter.org/, accessed 12 March 2005.

J. Stanton, 2002. “Terrorism in cyberspace: Terrorists will exploit and widen the gap between governing structures and the public,” American Behavioral Scientist, volume 45, number 6, pp. 1,017–1,032.

R. Sternberg and K. Sternberg, 2008. The nature of hate. Cambridge: Cambridge University Press.

H. Tajfel and J. Turner, 1979. “An integrative theory of intergroup conflict,” In W. Austin and S. Worchel (editors). The social psychology of intergroup relations. Monterey, Calif.: Brooks/Cole, pp. 33–47.

T. Tyler and H. Smith, 1998. “Social justice and social movements,” In: D. Gilbert, S. Fiske, and G. Lindzey (editors). The handbook of social psychology, volume 2. Fourth edition. Boston: McGraw–Hill, pp. 595–632.

N. Valentino and T. Brader, 2011. “The sword’s other edge: Perceptions of discrimination and racial policy opinion after Obama,” Public Opinion Quarterly, volume 75, number 2, pp. 201–226.http://dx.doi.org/10.1093/poq/nfr010

A. van Knippenberg, 1989. “Strategies of identity management,” In: J. van Oudenhoven and T. Willemsen (editors). Ethnic minorities: Social psychological perspectives. Amsterdam: Swets & Zeitlinger, pp. 59–76.

M. Waltman and J. Haas, 2011. The communication of hate. New York: Peter Lang.

S. Welch and L. Sigelman, 2011. “The ‘Obama Effect’ and white racial attitudes,” Annals of the American Academy of Political and Social Science, volume 634, number 1, pp. 207–220.http://dx.doi.org/10.1177/0002716210386302

C. Willis–Esqueda, 2008. “Introduction: Motivational aspects of prejudice and racism,” In: C. Willis–Esqueda (editor). Motivational aspects of prejudice and racism. New York: Springer, pp. 1–6.

C. Yang, K. Yin, and H.–H. Chen, 2007. “Building emotion lexicon from weblog corpora,” ACL ’07: Proceedings of the 45th Annual Meeting of the Association for Computational Linguistics (ACL), pp. 133–136.

N. Yu, 2009. “Doctoral qualifying paper: Opinion detection for Web content,” at http://ningyu.wikidot.com/start, accessed 22 October 2012.

 


Editorial history

Received 1 July 2012; accepted 15 October 2012.


Copyright © 2012, First Monday.
Copyright © 2012, Shlomi Sela, Tsvi Kuflik, and Gustavo S. Mesch. All rights reserved.

Changes in the discourse of online hate blogs: The effect of Barack Obama’s election in 2008
by Shlomi Sela, Tsvi Kuflik, and Gustavo S. Mesch
First Monday, Volume 17, Number 11 - 5 November 2012
http://firstmonday.org/ojs/index.php/fm/article/view/4154/3354
doi:10.5210/fm.v17i11.4154





A Great Cities Initiative of the University of Illinois at Chicago University Library.

© First Monday, 1995-2017. ISSN 1396-0466.