چكيده به لاتين
Finding trends in social networks has always been one of the most interested topics among scientific communities. Since Twitter is one of the most important and widespread social networks in terms of the number of users, it has devoted a lot of research in this field. Considering that trends arise from the content of messages exchanged between users, it is natural that most studies are based on content analysis. In these analyses, the methods based on counting words, weighting their importance, syntactic analysis of the message text in order to identify and extract and obvious and latent topics in it, using vectorization methods in order to calculate the similarity between the meaning of the messages and finally using data mining techniques to find trends. In addition, the methods of analyzing and discovering sentiment from messages are also among the methods studied in this field. The existing limitations of Twitter, including the limitation of the number of message characters, the use of unusual, colloquial and abbreviated discourses, the use of emoticons, the change of discourse literature over time, and the strong dependence of the past methods on language of message, are the major challenges of content analysis-based studies. In order to avoid these challenges and find trends, regardless of the language used in the message, we focused on another aspect of social network analysis, and that is, without referring to the text, the extraction of features that occur when trends and events occur. For this purpose, by examining the metadata characteristics of the message, the related user, and the network structure of that user, as well as using the studies that have been conducted to investigate the role of metadata regarding the popularity of tweets, define measurable and functional characteristics. We used the potential function, which indicates the potential trending index of a tweet, then with the help of this function, we were able to identify and announce trends in the initial stages of their occurrence. By comparing the results of this method with the content-based methods, we came to the conclusion that the structural and metadata analysis of tweets, will lead to similar results in order to discover the ongoing trends with methods based on content analysis.
Keywords:
Trend Detection, Structural Analysis, Trend Potential Function, Anomaly Detection
