Research
BONDHON: An Integrated Digital Platform for Bridging Gaps between General Public, Donors, and Social Welfare Organizations Working for Street Children in Bangladesh
In this project, we aim for the effective engagement between the stakeholders (small NGOs, donors, volunteers) working for the street children of Bangladesh. Along with renowned established NGOs numerous small scaled NGOs run by the university students also operate in this aspect. However, potential donors as well as common people hardly know about them and these NGOs face severe accountability, and credibility issues. In this context, we conduct two extensive interviews from the sample organizations to dig out the actual scenario. We have designed an interactive platform BONDHON after the interview and ethnographic study. The initial success of the projects -- a) A significant amount of the donors along with 5 different organizations. b) The system is scaled up to the similar type of organizations like orphanage and achieved success. c) This is the first study to the context of Bangladesh and we have achieved interesting human factors in this context that eventually leads to the revision of the well established HCI theories. In our paper, we have also discussed how the model can be modified to tackle down the similar types of problems in developing countries. Check out the web portal of BONDHON: http://bondhon.org/StreetChild/start
Towards the development of the Bengali Language Corpus for Hate Speech Research
In this project, we aim for the development of hateful speech corpus in Bengali language from the public Facebook pages. Public Facebook pages are potential source for haters for spreading hate speech, however, there is little research work is accomplished in this aspect. We have first collected around 5,000 comments using Facebook Graph API. By considering the social, cultural, historical, political, religious factors we have categorized the hateful and non hateful comments into six different classes. The hateful speeches are classified into four different categories: Political hate, religious hate, communal hate, insightful, political comments, religious comments while non hateful comments are classified into two categories: political and religious comments. The extensive categorization will definitely help the researcher from different discipline to acquire deep insights regarding hateful speeches. The research work is accepted at the Asian CHI Symposium held in conjunction with ACM CHI 2019
Hateful Speech Detection in Public Facebook Pages for the Bengali Language
In this project, we aim for devising a methodology for detecting hateful speeches in Bengali language. Since there was little work in this aspect, we have developed a data set around 5000 comments. We go for extensive linguistic feature analysis for identifying the correct feature sets for detection. Along side word N-gram, character N-gram feature, we have also worked for non conventional readability features and showed which feature sets are prominent in hateful speech. We have used different ML as well as GRU based deep neural network for detection process. We have attained 51% accuracy in the case of Random Forest for categorizing the comments into six classes. The accuracy is leaped up to 72% in the case of GRU based model. The research work is accepted at ICMLA 2019.
uReporter: An Anonymous One Stop Reporting Platform
In this project, we designed, developed, and deployed an anonymous reporting platform named uReporter, where a user can directly report to the proper authority for specific concerns. The concerned authority also interacts with the platform by describing their steps. The reporting platform is deployed in a university campus broadly where students can report about the campus problems. It is also deployed in other cases in a limited scope. However, with our finding, we have shown how human behaviors are acting towards the impediments of the scalability and maturity of such systems. As a result, the system fails to achieve its accepted result. The work is accepted at NSysS 2017. link: http://cse.buet.ac.bd/ureporter/
Note: Recently, the portal is facing state imposed ban since the common students from the campus are posting reports in the platform against the ruling powered backed student body's mischiefs, vandalism, physical harassment, and torture in the dorm.