Academia.eduAcademia.edu

Figure 2 – uploaded by IRJET Journal

See full PDFdownloadDownload figure
6. FUTURE SCOPE

Table 2 6. FUTURE SCOPE

Related Figures (7)

The 2d step of the proposed Model is the capabilities extraction step. In this step, the textual information is transformed right into a appropriate layout relevant to feed into gadget getting to know algorithms. First, we extract the capabilities of the enter information the usage of TFIDF and placed them in a capabilities listing. The key concept of TFIDF is that it really works at the textual content and receives the weights of the phrases regarding the record or sentence. In Addition to TFIDF, we use the sentiment evaluation technique to extract the polarity of the sentences and upload them as a function to the capabilities listing owning the TFIDF capabilities. The polarity of the sentences method that if the sentence is assessed as high-quality or negative. For that purpose, we extract the polarity the usage of the Text Blob library That’s a pre-educated version for film reviews. In more to The proposed technique, as visible in Fig. 1, consists of 3 principal steps: Preprocessing, capabilities extraction, and category step. In the preprocessing step we smooth the information through deducting the noise and disproportionate textual content. The preprocessing step is achieved within side the following: - Tokenization: In this part, we take the textual content as sentences or complete paragraphs after which output the entered textual content as separated phrases in a listing. - Lowering textual content: This takes the listing of phrases that were given out of the tokenization after which lowers all of the letters Like: THIS IS AWESOME’ goes to be ’that is awesome’. - Stop phrases and encoding cleaning: This is a essential a part of the preprocessing wherein we smooth the textual content from the ones forestall phrases and encode characters like n or t which do now no longer offer a bit of significant data to the classifiers. - Word Correction: In this part, we used Microsoft Bing phrase correction API that takes a phrase after which returns a JSON item with the maximum comparable phrases and the space among those phrases and the unique phrase
The 2d step of the proposed Model is the capabilities extraction step. In this step, the textual information is transformed right into a appropriate layout relevant to feed into gadget getting to know algorithms. First, we extract the capabilities of the enter information the usage of TFIDF and placed them in a capabilities listing. The key concept of TFIDF is that it really works at the textual content and receives the weights of the phrases regarding the record or sentence. In Addition to TFIDF, we use the sentiment evaluation technique to extract the polarity of the sentences and upload them as a function to the capabilities listing owning the TFIDF capabilities. The polarity of the sentences method that if the sentence is assessed as high-quality or negative. For that purpose, we extract the polarity the usage of the Text Blob library That’s a pre-educated version for film reviews. In more to The proposed technique, as visible in Fig. 1, consists of 3 principal steps: Preprocessing, capabilities extraction, and category step. In the preprocessing step we smooth the information through deducting the noise and disproportionate textual content. The preprocessing step is achieved within side the following: - Tokenization: In this part, we take the textual content as sentences or complete paragraphs after which output the entered textual content as separated phrases in a listing. - Lowering textual content: This takes the listing of phrases that were given out of the tokenization after which lowers all of the letters Like: THIS IS AWESOME’ goes to be ’that is awesome’. - Stop phrases and encoding cleaning: This is a essential a part of the preprocessing wherein we smooth the textual content from the ones forestall phrases and encode characters like n or t which do now no longer offer a bit of significant data to the classifiers. - Word Correction: In this part, we used Microsoft Bing phrase correction API that takes a phrase after which returns a JSON item with the maximum comparable phrases and the space among those phrases and the unique phrase
Where TP represents the number of true positive, TN represents the number of true negatives, FP represents the number of false positives, and FN represents the number of false negatives classes. the function extraction the usage of TFIDF and sentiment polarity extraction, the proposed method makes use of NGram[28] to keep in mind the one of a kind assortments of the phrases for the duration of the assessment of the version. Particularly, we use used 2- Gram, 3-Gram, and 4- Gram. The remaining step withinside the proposed technique is the category step wherein the extracted capabilities are fed right into a category set of rules to train, and check the classifier and consequently use it withinside the prediction phase. We used _ classifiers, namely, SVM (Support Vector Machine) and Neural Network. The neural community consists of 3 layers: Input, hidden, and output layers. In the enter layer, it includes 128 nodes. In the hidden layer, it consists of sixty four neurons. The output layer is a Boolean output. Generally, the assessment of classifiers is achieved the usage of numerous assessment matrices relying at the confusion matrix. Among the ones standards are Accuracy, precision, recall, and f-score. They are calculated consistent with the subsequent equations:
Where TP represents the number of true positive, TN represents the number of true negatives, FP represents the number of false positives, and FN represents the number of false negatives classes. the function extraction the usage of TFIDF and sentiment polarity extraction, the proposed method makes use of NGram[28] to keep in mind the one of a kind assortments of the phrases for the duration of the assessment of the version. Particularly, we use used 2- Gram, 3-Gram, and 4- Gram. The remaining step withinside the proposed technique is the category step wherein the extracted capabilities are fed right into a category set of rules to train, and check the classifier and consequently use it withinside the prediction phase. We used _ classifiers, namely, SVM (Support Vector Machine) and Neural Network. The neural community consists of 3 layers: Input, hidden, and output layers. In the enter layer, it includes 128 nodes. In the hidden layer, it consists of sixty four neurons. The output layer is a Boolean output. Generally, the assessment of classifiers is achieved the usage of numerous assessment matrices relying at the confusion matrix. Among the ones standards are Accuracy, precision, recall, and f-score. They are calculated consistent with the subsequent equations:
TABLE I. STATISTICS OF THE DATASET B. Results with either cyberbullying or not. The annotation classes were unbalanced distributed such 1038 question- answering instances out of 12773 belongs to the category cyberbullying, while 11735 belongs to the opposite class. First, to remedy the information unbalancing, we take the identical number instances of both classes to live the accuracy. We also faraway from the information big size conversations and remove the noisy data. We ended up with total 1608 instance conversations where 804 instances belongs to every class. Table I summarizes the statistics of dataset. After preprocessing the dataset, we follow the identical step presented in Section III to extract the features. We then split the dataset into ratios (0.8,0.2) for train and test. Accuracy, recall and precision, and f-score are taken as a performance measure to judge the classifiers. We apply SVM likewise as Neural Network (NN) as they're among the most effective performance classifiers within the literature. We run several experiments on different n- gram language model. particularly, we take into consideration 2-gram, 3-gram, and 4-gram during the evaluation of the model produced by the classifiers. Table I] summarizes the accuracy of both SVM and NN. The SVM classifier achieved the best percentage using 4-Gram with accuracy 90.3% while the NN achieved highest accuracy using 3-Gram with accuracy 92.8%. it's found that the common accuracy of all n-gram models of NN achieves 91.76%, while the average accuracy of all n-gram models of SVM achieves 89.87%. Fig. 2 depicts the accuracy results of both classifiers
TABLE I. STATISTICS OF THE DATASET B. Results with either cyberbullying or not. The annotation classes were unbalanced distributed such 1038 question- answering instances out of 12773 belongs to the category cyberbullying, while 11735 belongs to the opposite class. First, to remedy the information unbalancing, we take the identical number instances of both classes to live the accuracy. We also faraway from the information big size conversations and remove the noisy data. We ended up with total 1608 instance conversations where 804 instances belongs to every class. Table I summarizes the statistics of dataset. After preprocessing the dataset, we follow the identical step presented in Section III to extract the features. We then split the dataset into ratios (0.8,0.2) for train and test. Accuracy, recall and precision, and f-score are taken as a performance measure to judge the classifiers. We apply SVM likewise as Neural Network (NN) as they're among the most effective performance classifiers within the literature. We run several experiments on different n- gram language model. particularly, we take into consideration 2-gram, 3-gram, and 4-gram during the evaluation of the model produced by the classifiers. Table I] summarizes the accuracy of both SVM and NN. The SVM classifier achieved the best percentage using 4-Gram with accuracy 90.3% while the NN achieved highest accuracy using 3-Gram with accuracy 92.8%. it's found that the common accuracy of all n-gram models of NN achieves 91.76%, while the average accuracy of all n-gram models of SVM achieves 89.87%. Fig. 2 depicts the accuracy results of both classifiers
TABLE II. THE ACCURACY OF SVM AND NN IN DIFFERENT LANGUAGE MODEL Fig. 2. Comparison between SVM and Neural Network in Terms of Accuracy n addition to accuracy, Table III and Table IV show the evaluations of both classifiers in terms of precision and recall respectively for every language model. The trade-off between recall and precision is shown in Table V which represents the fscore of both classifiers within the different language model. Table V summarizes the f-score of both SVM and NN. The SVM classifier achieved the very best f-measure using 4-Gram with f-score 90.3% while the NN achieved highest f-measure using 2-Gram with f-score 92.2%. it's found that the common f-score of all n-gram models of NN achieves 91.9%, while the typical f-score of all n-gram models of SVM achieves 89.8%. Fig. 3 summarizes the f-score of the classification of the SVM and Neural Network. The results of average accuracy furthermore because the average f-score indicate that NN performs better than SVM. l International Research Journal of Engineering and Technology (IRJET) — e- -T Volume: 09 Issue: 05 | May 2022 www.irjet.net p-l
TABLE II. THE ACCURACY OF SVM AND NN IN DIFFERENT LANGUAGE MODEL Fig. 2. Comparison between SVM and Neural Network in Terms of Accuracy n addition to accuracy, Table III and Table IV show the evaluations of both classifiers in terms of precision and recall respectively for every language model. The trade-off between recall and precision is shown in Table V which represents the fscore of both classifiers within the different language model. Table V summarizes the f-score of both SVM and NN. The SVM classifier achieved the very best f-measure using 4-Gram with f-score 90.3% while the NN achieved highest f-measure using 2-Gram with f-score 92.2%. it's found that the common f-score of all n-gram models of NN achieves 91.9%, while the typical f-score of all n-gram models of SVM achieves 89.8%. Fig. 3 summarizes the f-score of the classification of the SVM and Neural Network. The results of average accuracy furthermore because the average f-score indicate that NN performs better than SVM. l International Research Journal of Engineering and Technology (IRJET) — e- -T Volume: 09 Issue: 05 | May 2022 www.irjet.net p-l
TABLE VI. COMPARISON WITH RELATED WORK Fig. 4. Comparison between the Best Classifiers in Terms of Accuracy
TABLE VI. COMPARISON WITH RELATED WORK Fig. 4. Comparison between the Best Classifiers in Terms of Accuracy
Fig. 3. Comparison between SVM and Neural Network in Terms of F- Measure
Fig. 3. Comparison between SVM and Neural Network in Terms of F- Measure
Fig. 5. Comparison between the Best Classifiers in Terms of F-Measure
Fig. 5. Comparison between the Best Classifiers in Terms of F-Measure
Related topics:
Engineering
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved