الفهرس 
Related WorkOnly 14 pages are availabe for public view
 |  | from | 91 |  |  |
| | | from | 91 | | |
Abstract
Abstract
The rapid use of social media made location prediction the key to research studies based on-location services such as, advertising, recommendations, climatological forecast, and security system. Locations are the center of information for these applications. According to millions of users who post tweets every day, the geographical coordinates are often hidden in Twitter due to privacy reasons. Identifying the home location of Twitter users is very important in many business community applications.
Therefore, many approaches have been developed to automatically geolocate Twitter users using their tweets. Depending on the importance of catching the location of the users and the rapid usage of Twitter, Location prediction on Twitter has been a point of research in many studies.
This thesis work provides a comprehensive overview of the prediction of the user’s location on Twitter, which focuses on the home location prediction and tweet location prediction. This is achieved by defining the inputs of these two research views that are content, network, and context, and then proposing two new location prediction models.
The First proposed model is to predict the tweet location based on the KNN-Sentimental Analysis (KNNSA) model. Predicting the tweet location based on the KNN-sentiment analysis (KNNSA) extracts text features from the tweet in addition to the date and time features. Then, applying sentimental analysis and classifying the data by K-nearest neighbors (KNN) classifier. The (KNNSA) is evaluated and compared to the previous work and it achieves better performance in terms of root mean squared error (RMSE) and of the mean absolute error (MAE).
The second proposed work is to predict home location for Twitter users based on sentiment analysis (Pre-HLSA). It predicts the users’ home location using only their tweets, by analysing some of the tweet’s features. Achieving this goal allows providing geospatial services, especially in the epidemic dispersion. The Pre-HLSA represents user tweets as a set of extracted features and predicts the users’ home locations by analysing their tweets to find sentiments and polarities, even in the absence of geospatial clues. Then, different classifiers are applied by applying sentimental analysis The experimental results show a promising performance compared to the previous methods in terms of accuracy, mean and median performance measures. It achieves up to 85% accuracy, 223 km mean, and 96 km median