Ot match the reputation prediction theme, leaving us with 11 articles. Inside the Scopus database, we obtained 573 papers with 547 excluded because of the 3 exclusion criteria, and, in the ACM Digital Library of 606 articles, 576 were GS-626510 manufacturer discarded. All articles chosen in the three bases added as much as a total of 67 papers to be studied. We analyze and opt for the most applied approaches that will be explained within this manuscript. This section presents the taxonomy constructed in the solutions involved in Recognition Prediction. We present definitions, the operation of popularity predictions, the sorts of content material, and a taxonomy to classify the models studied.Sensors 2021, 21,7 of3.1. Taxonomy To structure the study and presentation, we divided the methods of predicting popularity in accordance with the issue definition and the prediction task, as follows: Regression Techniques. These procedures carry out a numerical prediction, quantifying the popularity according to the defined metric. One of the most popular target attributes are the quantity of views, number of shares, variety of GNE-371 supplier tweets, and comments. These predictive procedures use Regression and are generally referred to as regressors [9,22]. Classification Procedures. Recognition classes are defined; the predictive model allocates the content material in among the list of defined classes. The goal is usually to predict no matter whether content material will develop into well known or not; in most circumstances, only two classes are used: well known and non-popular. These predictive techniques use Classification and are frequently named classifiers [13,15,16].Moreover to the above division, we are able to group the prediction procedures as outlined by the attributes used: Textual Attributes. These attributes are extracted in the content material employing NLP tactics. The extraction can be direct from the content material. In news articles, it can be from the description presented on the web, as in videos and pictures, and in some cases taking advantage of social media components, for instance comments published by customers. Visual Attributes. These attributes are extracted from videos and images applying ML strategies (ANN, for example) or manually choosing functions from the frames representative on the content material. Metadata Attributes. These attributes are offered by the web site where the content material was published and inherent for the Internet. Nevertheless, they don’t belong to any previous groups, like the supply with the content material, category, number of views, and publication date. This taxonomy is shown in Figure 1: Prediction MethodsClassification Textual Features Visual Features Metadata FeaturesRegressionTextual Features Visual Features Metadata FeaturesFigure 1. Taxonomy as outlined by the prediction procedures and attributes made use of.Here, we classify the research that present reputation prediction models applying the proposed taxonomy. We show the results in Table 1, where C in predictive tasks indicates that the model studied makes use of Classification and R indicates a predictor that utilizes Regression. These surveys have been also classified in accordance with the attributes utilized. This classification is not exclusive. Some models are positioned in greater than 1 category. These articles present a number of techniques and models that may be deemed state of the art for predicting popularity. Table two shows the top models for each study as well as the functionality earned. It is necessary to pay consideration for the metrics employed to validate the comparisons. We observed that the classifiers that use textual attributes usually achieved the most beneficial outcomes. In contrast, the regressors using the finest results utilised visual f.