![]() | Only 14 pages are availabe for public view |
Abstract Twitter, an online micro blog that enables its users to send and read text-based posts known as ”tweets”, became one of the most commonly used social networks. However, it is not available to find a summary of tweets about a certain Arabic hot topic. This thesis presents an algorithm for summarizing Arabic micro blogging posts. Algorithm processes collections of tweets on specific topics and creates short summaries from these collections. The goal is to produce summaries that are similar to what a human would produce for the same collection of tweets on a specific topic. The problem is formulated as a regression problem, not a binary classification based problem. Instead of classifying the tweets to be important and not important, each tweet is given a score that determines to which extent this tweet is candidate to be in the summary. This helps generate the summary according to the user predefined compression rate. The proposed system is evaluated by using two methods namely, the Precision and Recall method and the Normalized Discounted Cumulative Gain (NDCG). The results of summarization of the proposed system are compared with that obtained by the well-known multi-document summarization algorithms including; SumBasic, TF-IDF, PageRank, and MEAD summarizer. Also, three volunteers are asked to make summarization of several groups of tweets and the results obtained by the proposed system are compared with that obtained by the manual summarization experiments. Demonstrated two ways were used in the evaluation Precision/Recall and NDCG indicate that our proposed system produced good level of summarization performance compared to the other systems and compared to the manual summarization. the proposed strategy provides best results of 50% compared to the manual summarization of 53% using Precision / Recall method and 91% compared to the manual summarization of %95 using NDCG method. |