mohammadali javadzade

Graphs are data that describe complex relationships between different things in the real world, such as the Internet, social network, bibliographic network, and so on. One of the things that many people deal with today is online social networks. The graph display of online social networks such as Twitter, WeChat and Facebook is not possible today with less than billions of nodes, and for this reason, the study of large-scale network data has become a necessity for researchers. Regarding social networks, online users often have limited information; But for social media service providers, user node information such as interest, beliefs, or other characteristics are very important to customize their services for users in many applications such as recommendations and personalized search, making it a challenge for service providers. An effective way to deal with this challenge is to infer missing user information using pervasive network structures in social media. One of the most important inferences in data mining and network analysis is node classification, which aims to infer the missing labels of nodes based on labeled nodes and network structure. In this research, we have performed the task of node classification on the PubMedDiabetes, CiteSeer and Cora citation network datasets using GraphSAGE, GCN and GAT neural networks and we have generally concluded that the GraphSAGE neural network on the network datasets The cited reference works well for the node classification task.

Today, the interest in predicting and detecting events using the data available on social networks has increased. Social networks can be called the sensors of society, because the users always express their positive and negative opinions about the events of the world around them, which results in an environment full of real-time reactions to real-world events. Social networks are one of the best tools for assessing the society and predicting upcoming events. Although the automatic detection and classification of events, especially social anomalies such as riots, is a trivial task, it is of great value to governments and security organizations that need to respond quickly and appropriately; because the costs and damages caused by these unrests can be reduced. For this challenge, we have developed an event predicting framework that can distinguish "disruptive events" that threaten social security and order from daily events. To do this, we have used natural language processing techniques to comprehend texts, remove the limitations of human language, and perform sentiment analysis and topic detection. We have classified the events using machine learning techniques such as the Naïve Bayes and Support Vector Machines. Finally, we have evaluated our framework in a large and real data set from Twitter to show the efficiency and effectiveness of our system in predicting future events. The results show that the proposed framework has the ability to detect tweets reflecting dissatisfaction with 79% accuracy. We have also managed to extract the useful information related to an event with 40% accuracy from these tweets in the form of 5 topics namely, the place, time, people, goals and event related factors.

Although the lack of data is one of the important challenges for research in the field of natural language processing, but this challenge is more acute in the Persian language, so finding a high-quality and comprehensive dataset in the Persian language is a difficult task. In addition to that, having some problems such as the ability to categorize and not complying with the storage standard are among the problems of the existing datasets, each of which can affect the learning rate of the model, the results, and the error rate in the experiments. For this reason, all these reasons made us seek to collect and prepare a dataset that covers all such problems and reduces the amount of error when using data in different models. In this research, we have designed and used a crawler to collect textual data. By crawling on one of the news bases, it has been able to collect data sets in five columns: title, summary, text, tag, and publication date. The textual data has been normalized with the help of one of the Persian language libraries in the Python programming language and stored in csv and xml formats and made available to fellow researchers. The tags in this dataset include 13 main tags of sports, art and media, culture, science and progress, political, foreign policy, life, family, society, education and training, international, economic and provinces. Among the tasks that can be done on this data set are text classification, text extraction, text summarization and title recognition. Also, one of the prominent features of this data set is its comprehensiveness, the amount of suitable data, the existence of useful features, having unique features, as well as storage in a standard format. This dataset is a product of the Language Processing Department of Imam Hossein Comprehensive University and can be downloaded and used through the link mentioned in the footnote of the next page and with respect to copyright.

With the development of technology, the use of social networks has become more popular. One of the most popular social networks is the Twitter platform. Sentiment analysis of users' tweets plays an important role in showing users' feelings about the existing conditions of society. In recent years, due to the fact that the text of users' tweets have become more conversational, sentiment analysis has become problematic and reduced its accuracy; It also makes natural language processing difficult. In this research, a method for sentiment analysis of Persian tweets based on the combination of semantic relations and support vector machine classification has been presented. FastText semantic relation is used to extract features. Considering that a large number of features have been extracted; They should be reduced, which is done by using long-short-term memory (LSTM) neural network. In the last part of the proposed method for the classification of sentiments in tweets, the support vector machine model is used. The evaluation criteria used in this research were precision, accuracy, recall and F criterion, and the evaluation results were 83.9, 84.3, 83.9 and 84, respectively. The results of the experiments show the applicability of the proposed method in analyzing the emotions of Persian tweets into six classes of anger, sad, joy, disgust, surprise and fear.