(IJACSA) International Journal of Advanced Computer Science and Applications, Vol. 9, No. 5, 2018 439 | Page www.ijacsa.thesai.org Framework for Rumors Detection in Social Media Rehana Moin, *Zahoor-ur-Rehman, Khalid Mahmood Department of Computer Science COMSATS University Islamabad, Attock Campus, Pakistan Mohammad Eid Alzahrani, Muhammad Qaiser Saleem College of Computer Science and Information Technology, Al Baha University, Al Baha, Saudi Arabia Abstract—The development of social networks has led the public in general to find easy accessibility for communication with respect to rapid communication to each other at any time. Such services provide the quick transmission of information which is its positive side but its negative side needs to be kept in mind thereby misinformation can spread. Nowadays, in this era of digitalization, the validation of such information has become a real challenge, due to lack of information authentication method. In this paper, we design a framework for the rumors detection from the Facebook events data, which is based on inquiry comments. The proposed Inquiry Comments Detection Model (ICDM) identifies inquiry comments utilizing a rule-based approach which entails regular expressions to categorize the sentences as an inquiry into those starting with an intransitive verb (like is, am, was, will, would and so on) and also those sentences ending with a question mark. We set the threshold value to compare with the ratio of Inquiry to English comments and identify the rumors. We verified the proposed ICDM on labeled data, collected from snopes.com. Our experiments revealed that the proposed method achieved considerably well in comparison to the existing machine learning techniques. The proposed ICDM approach attained better results of 89% precision, 77% recall, and 82% -measure. We are of the opinion that our experimental findings of this study will be useful for the worldwide adoption. Keywords—Social networks; rumors; inquiry comments; question identification I. INTRODUCTION A rumor is an unverified claim about any event, transmitting from person to person. It may refer to an incident, object or problem of public concern. It may prove to be a social destructive phenomenon in any human culture. Usually, the social media rapidly transmits the unverified statements that may be harmful for anybody. Nowadays, social networks like Twitter and Facebook are more popular with regards to acquiring and propagating information. On social networks everybody is free to obtain and share information, anywhere at any time [1]. Besides, it has been reported that these social sites are capable to spread rumors [2]. In general, a rumor refers to the information that lacks source and its truthfulness. Ordinarily, it is generated in an emergency situation, leading to anxiety, disruption of social activities; thus, reducing the government credibility, even endangering the national security, for instance, on March 2011, after Japan Earthquake followed by tsunami and nuclear disaster. A rumor was propagated by microblog platforms, advising use of iodized salt for protection of people by nuclear radiation. Consequently, the public in general rushed to markets to buy salt, which was totally untrue and unnecessary practice. In the future, to avoid such unfruitful happenings, at the earliest, rumor detection is essential. Earlier, much work has been done on rumor detection using the Twitter. We did work on Facebook to address the problem of rumor detection. We selected Facebook reason being the most popular social network. In Oct 2012, Facebook was having one billion users per month. Cameron Marlow, one of the research scientists, considered Facebook as world’s most powerful instrument for studying human society [3]. A framework diagram is developed for rumors detection, starting from Facebook data collection, preprocessing of data, extraction of English text, apply TopicRank to obtain keyphrases and based on those keyphrases (topics) extract the event data and detect assertion to filter assertive event posts and finally detecting the inquiry comments on assertive posts using our proposed ICDM approach . We used labeled data from snopes.com to check the validity of our proposed ICDM approach and to make comparison with machine learning techniques. We aim to tackle the rumors detection problem using inquiry comments identification through ICDM approach. This comprises two steps. In the first phase, we identify questionable statements named as “inquiry comments”. We adopt both machine learning supervised approach like classifiers to detect questions and rule-based method to detect question marks, 5W1H words and regular expressions [4] which utilizes patterns to filter inquiries. In the second phase, we extract inquiry comments asking question about the event. We define the threshold to identify the rumors and test our ICDM model using labeled data from snopes.com. Consequently, following research questions are formulated:  How English text is separated from different languages?  How to develop rumors detection framework that can correctly identify the rumors?  How can we verify our proposed ICDM (Inquiry Comments Detection Model)? Remaining part of this paper is organized as follows: Section II provides the related work; in Section III, methodology is presented; in Section IV, the results are presented: Section V concludes the whole work and addresses the future research directions of this study.