Share
نظام توصيات قائم على الويب لانتشار الاوبئة == Web - Based Recommender System for Spread Epidemics
Author name:
حيدر محمد حبيب مجيد
Supervisor name:
نبيل هاشم الاعرجي
General topic:
Computer Science
Specific topic:
Computer Science
Degree:
Doctorate
University:
University of Babylon - College Of Science - Department Of Computer Science
Language:
English
University location:
Babylon
First pages:
28T765 - p.pdf
Abstract:
The usage of Online Social Networks, such as Facebook and Twitterbecomes more and more popular in order to exchange and disseminate news andinformation in real - time. Twitter in particular allows the instant dissemination ofshort messages in the form of microblogs to followers. This dissertation exploresand examine the usage of how social networks, such as the microblogging toolTwitter, can help in the detection of spreading epidemics and reducing time delaybetween the emergence of disease and report sick to the health authorities suchas World Health Organization (WHO).Text classification has been used to classify the patients and non - patients(positive / negative). Sentiment Analysis (SA) and Linear Support VectorClassifier (LSVC) have been applied in the classification patients. In thisdissertation, four diseases have examined. Diseases that have most similarity intheir symptoms have been taken in order to classify patients based on theirsymptoms by applying a recommendation system techniques. Symptoms - basedHealthcare Recommender System is new approach in this work. It uses patientsymptominstead of user - item in traditional Collaborative Recommender System.Collaborative Filtering (CF) has been applied in order to recommend whichdisease the patient may has. CF shows an indicator that users on Social Networkshave not enough knowledge to mention all symptoms for specific disease, that’sled to classify patients to more than one disease according to common symptomsthat mentioned by patients.Geolocation of users that classified as patients has been extracted in orderto recommend health authorities that there is a certain area might has a beginningof spread disease. An implicit geocoding of users has been extracted by usingGoogle Maps Geocoding API to avoid neglecting those who don’t have explicitgeolocation.IISuspected areas has been weighted by computing a Confidence Factor(cFactor) of Tweet source whatever it comes from mobile or desktop. cFactorhelp in reducing time consuming into 29% of collecting and processing data.Weighted and Geographic Symptoms - based Recommender (WGSR) model hasbeen created detect, classify and visualize patients on the map.The accuracy of WGSR model reached to %94 in the classification andmore than %80 with the real reports of World Health Organization (WHO) whichrefers as a very good and can be improved for better results.
