Various researchers proposed different methods of detecting cyberbullying; however, they were largely based on text and user-defined functions. Most of the studies found in the literature aim to improve the detection by introducing new features. However, with an increase in the number of elements, the selection and selection steps of the elements get complicated. An input dataset is sent for data preprocessing and is used to improve the quality of the input data. Data preprocessing also includes the removal of keywords and special characters. After preliminary data processing, the output data is sent to the feature extraction process where the optimal method is selected through classification. Then the oppositional grasshopper optimization with convolutional neural network (OGHOCNN) classification algorithm is used to detect the cyberbullying words in tweets. The key characteristics are determined using the reduction method and combined into a single model that offers the best detection accuracy. The proposed Twitter data based cyberbullying detection method shown in Figure 10.2.


This section focuses specifically on the features built into the current study’s cyber threat detection model. This includes user personalities that focus on Big Five, Dark Triad models, Twitter-based emotions, passions, and features. Social network data is serious due to which preprocessing is applied to improve the accuracy of input data. This includes the removal of stop words. Stop words are generally “a,” “like,” “have,” “is,” “the,” “o,” etc., which are nothing but “Stop using words,” a phenomenon important for memory space and processing time.

Proposed twitter data-based cyberbullying detection

FIGURE 10.2 Proposed twitter data-based cyberbullying detection.

Feature Extraction

Twitter includes Twitter API as extracted to identify digital vulnerabilities, while additional coding has been created to remove highlights, such as lowercase and uppercase letters (i.e., content features). The highlights removed are:

  • Text/contentfeatures—including font sizes, uppercase, lowercase, hash tags, images, client references, URLs, and multimedia content.
  • User features—functionality removed from each person’s profile, for example, age by age, number of status checks (i.e., the number of tweets—number of retweets performed by a customer), number of registrations (i.e., open registrations that are part of a customer), and the number of customer’s best choices (a customer with his number) bill greeted by the amount of tweets (Chatzakou et al., 2017a).
  • Network features—Customer-specific steps involve social measurement such as the amount of followers, devotees, and their reputation (the extent of support for supporters).

Feature Selection Using Ranking Method

Based on a supervised methodology, tweets can be obtained as a basic unit or a modified one to find the average scores for a language. The estimate mainly shows how common is the amount of irritating words in the progress of tweets, when deciding the progress of previous tweets. When a tweet contains quick words, they are available at that time, whereas the level of importance determines the scores. It is now indistinguishable from the inclination of the multinomial network, in which all the tweets of a class are converted into individual tweets. After this process, the feasibility of these classy tweets is assessed. To the extent of administrative significance, the parameter class addresses the correct tweets and also the full set of readiness. It is recognized that a part shows the time in dataset and the time in class tweets. The length of the information collection (e.g., the planning set) is class and individual based on the frequency of the words collected. The data shown in Figure 10.1 denotes the length of the set and the speed of the class. The glory of false alarms is taken into account and is defined in equation (10.4).

The meaning of phrase win set j c is as follows:

When trying to rearrange the estimated formula, it is possible to edit the following:

The meaning of a sentence in a set of j c is that a given word can be understood as gradually meaningful, important, or refined for that class. Phrases, with a very high average score, are associated with keywords that gradually illuminate the specific set. Again, it is expected that a class score is added to the brand selection and the selection of best R features. This grouping focuses on making the strategy. Placement factors are sorted using significance scores for each class. For example, the main component of each configured reorder denotes the word reference indicator. Now, part evaluation is used in each class in contrast to their centralized scores. When a summary of the solitaire components that depend on this class is documented, the largest position for all items is selected [6].

< Prev   CONTENTS   Source   Next >