*Classification Model for Bullying Posts Detection DOI: http://dx.doi.org/10.5772/intechopen.88633*

**MOTIVE = SEXUAL**

**160**

**TOPIC 30**

**CRUDE** 

Sexy Kickit FuckforOL' Getdown dirty

Slap Hump Screw Predators: Victims

P1: V1 P2: v2 P3: V3 **MOTIVE = OUTRAGE**

**TOPIC 60**

**ANGER** Bitterness

Hard Storm Irritation

Wrath

Fury

0.0365 0.0354 0.0321 0.0306 0.0268 0.0251

Marquee

Mate Minor Moot

MP MUM

 Prob 0.0737 0.0552 0.0324

P1: V3

0.0367

**MOTIVE =** 

**TOPIC 70**

Make out

0.0267 0.0235 0.0223 0.0215 0.0209 0.0201 0.0189

Patient

PC Period

Plant

Pass out

Outhouse

Pant

0.0246

0.0232

0.0214

0.0208

0.0179

0.0165

0.0152

**IRRELEVANT**

P1: V3

0.0276 **MOTIVE =** 

**TOPIC 90**

**UNKNOWN**

P4: V5

0.0354

P1: V3

0.0211

P1: V2

0.0428

P2: V6

0.0321

P5: V7

0.0467

P2: V6

0.0241

P4: V4

0.0541

P3: V5

0.0452

P1: V 6

 Predators: Victims

 0.0421 0.0316

0.0307

0.0201

 Getlucky

 0.0142

 Prob

 Predators: Victims

 Raunchy

 0.0147

 Prob

 Predators: Victims

 Prob 0.0595

P3: V5

0.0354

 Predators: Victims

 Prob

 Giveitup

 0.0154

 doublebag

 0.0165

 Intimacy

Cottage

0.0191

Jellosex

Score

0.0104

Trim

0.0107

0.0135

Smush

0.0124

 0.0206

Donasty

 0.0208

 Sexytime

 0.0319

0.0432

 Encounter

 0.0215

 Monkeylove

 0.0328

 Homerun

Smack

Serve

0.0271

Ride

0.0154

0.0284

 Jiffystiffy

 0.0209

 0.0307

 Hempedup

 0.0245

0.0569 0.0445

 Kneedeep

 0.0241

Givebusiness

 0.0465

 Poundduck

 0.0324

Sleep

0.0282

 Hangnow

 0.0485

Happyhappy

 0.0341

Randy

Juicy

0.0284

0.0307

**LANGUAGE**

**TOPIC 35**

**IMPLICIT**

**LANGUAGE**

**INDECENT**

**PROPOSALS**

**UNREFINED**

**LANGUAGE**

**SLANG WORDS**

**TOPIC 40**

**TOPIC 45**

**TOPIC 50**

*Cyberspace*

Better simplification functioning is designated by means of a lesser perplexity on a held-out document. The derivation of the likelihood of a collection of texts specified the predator is a uncomplicated computation in Bully-LDA model.

indicate that B-LDA better generalizes performance than ATM and LDA. The improvement in generalization performance of B-LDA can be explained by its ability to better model when comparing with LDA and ATM model. If a word which has small probability in the bullying topics of training document, then it will cause an increase in perplexity. As the number of bullying topics increase, then the probabilities assigned to words get smaller in each bullying topic. Even though ATM models the roles of authors, does not show promising results and it is originally designed for the scenario where each document has multiple authors. It is clear that B-LDA achieves superior performances among all the adopted models. The perplexity of LDA, ATM, and B-LDA are closer and they decrease steadily with the increase of topics. According to human judgments, perplexity is not easy to correlate the results. So, it is necessary to compare the models using simple metrics like Precision, Recall, and F1 measure. The standard supervised classifier, i.e., Support Vector Machine (SVM), is adopted with B-LDA for classification. LibSVM was applied to the two-class classification problem using a linear kernel. Each post is an instance; positive classes contain bullying messages and negative classes contain non-bullying messages. A 10-fold cross-validation was performed in which the complete dataset was partitioned 10 times into 10 samples; in every round, nine portions were employed for exercising and the enduring section was applied for

*Classification Model for Bullying Posts Detection DOI: http://dx.doi.org/10.5772/intechopen.88633*

The functioning of the classifier was appraised on precision, recall and F-1 measure and these measures depend on the top-ranked features produced through B-LDA method against the truth set as tested on the datasets. Precision: The Aggregate number of accurately distinguished genuine harassing posts out of recovered tormenting cases. Recall: Number of effectively distinguished tormenting cases from an aggregate number of genuine harassing cases. F-1 measure: the equally weighted harmonic mean of precision and recall. **Table 4** shows the classifier

The weighted B-TFIDF method is compared with the work done in a content analysis in a web on four different datasets. The new feature selection method using

cataloged in **Table 5** and also indicate a very high precision, recall and F-1 measure on Twitter. In Kongregate precision fell down at the top 2000 features. In most of

weighted B-TFIDF proved that it is better than baseline. The outcomes are

**3.5 Comparison of weighted B-TFIDF with baseline method**

*Classifier performances based on different feature reduction methods.*

trial (**Figure 8**).

performance.

**Figure 8.**

**163**

$$p(wd|pd) = \int d\theta \int d\phi p(\theta|\text{Dtrain}) p(\phi|\text{Dtrain}) \ast \prod\_{m=1}^{Nd} \left[ \frac{1}{Ad} \sum\_{i \in pd, j} \theta \text{ij} \phi wmj \right] \tag{12}$$

The term in the brackets is merely the probability for the word wm specified the pair of predators pd. The detailed results are exposed in **Figure 7**. These results

**Figure 6.** *Predators activity for bullying topic 30.*

**Figure 7.** *Comparisons of different models in terms of perplexity.*

#### *Classification Model for Bullying Posts Detection DOI: http://dx.doi.org/10.5772/intechopen.88633*

Better simplification functioning is designated by means of a lesser perplexity on a held-out document. The derivation of the likelihood of a collection of texts spec-

The term in the brackets is merely the probability for the word wm specified the

pair of predators pd. The detailed results are exposed in **Figure 7**. These results

Y *Nd*

1 *Ad* X *i*∈*pd*,*j*

*θijϕwmj*

(12)

� �

*m*¼1

ified the predator is a uncomplicated computation in Bully-LDA model.

*dϕp*ð Þ *θ*j*Dtrain p*ð Þ *ϕ*j*Dtrain* ∗

*p wd* ð Þ¼ j*pd*

*Cyberspace*

**Figure 6.**

**Figure 7.**

**162**

*Comparisons of different models in terms of perplexity.*

*Predators activity for bullying topic 30.*

ð *dθ* ð indicate that B-LDA better generalizes performance than ATM and LDA. The improvement in generalization performance of B-LDA can be explained by its ability to better model when comparing with LDA and ATM model. If a word which has small probability in the bullying topics of training document, then it will cause an increase in perplexity. As the number of bullying topics increase, then the probabilities assigned to words get smaller in each bullying topic. Even though ATM models the roles of authors, does not show promising results and it is originally designed for the scenario where each document has multiple authors. It is clear that B-LDA achieves superior performances among all the adopted models. The perplexity of LDA, ATM, and B-LDA are closer and they decrease steadily with the increase of topics. According to human judgments, perplexity is not easy to correlate the results. So, it is necessary to compare the models using simple metrics like Precision, Recall, and F1 measure. The standard supervised classifier, i.e., Support Vector Machine (SVM), is adopted with B-LDA for classification. LibSVM was applied to the two-class classification problem using a linear kernel. Each post is an instance; positive classes contain bullying messages and negative classes contain non-bullying messages. A 10-fold cross-validation was performed in which the complete dataset was partitioned 10 times into 10 samples; in every round, nine portions were employed for exercising and the enduring section was applied for trial (**Figure 8**).

The functioning of the classifier was appraised on precision, recall and F-1 measure and these measures depend on the top-ranked features produced through B-LDA method against the truth set as tested on the datasets. Precision: The Aggregate number of accurately distinguished genuine harassing posts out of recovered tormenting cases. Recall: Number of effectively distinguished tormenting cases from an aggregate number of genuine harassing cases. F-1 measure: the equally weighted harmonic mean of precision and recall. **Table 4** shows the classifier performance.

### **3.5 Comparison of weighted B-TFIDF with baseline method**

The weighted B-TFIDF method is compared with the work done in a content analysis in a web on four different datasets. The new feature selection method using weighted B-TFIDF proved that it is better than baseline. The outcomes are cataloged in **Table 5** and also indicate a very high precision, recall and F-1 measure on Twitter. In Kongregate precision fell down at the top 2000 features. In most of

**Figure 8.** *Classifier performances based on different feature reduction methods.*

the cases, the classifier performed almost similar, that is between 80 and 100%. On Myspace dataset recall is moderate nearing to 1. However, precision varies between 76 and 87% except at feature value 18,000 when it reaches 91%. Unlike other datasets, Slashdot performance is very low. Although recall is moderate, precision and F-1 measures decomposed while component set was low. Also, poor performance is observed at feature value 18,000. From this discussion, the performance of weighted B-TFIDF shows the best result (**Figure 9**).
