Public facilities recommendation system based on structured and unstructured data extraction from multi-channel data sources

Nowadays social media data has grown very rapidly by producing a huge amount and variety of data everyday. Those data can be analyzed and processed to deliver useful information especially for public needs. However, most of the data available in social media are unstructured. This paper proposes a recommendation system for public facilities by utilizing both structured and unstructured data gathered from multi-channel data sources. The system uses single-criteria rating, multi-criteria-rating, and text data as the inputs. The challenge is how to handle data variety such that any kind of data from any channel can be integrated. The second challenge is how to extract location-related data from the raw data. There are four data channels used in the system. Three of them are social media channels, i.e. Twitter, Instagram, and Foursquare, while the other is internal data channel built as a part of the system itself. The system deals with three categories of public facility, i.e. park, hospital, and mosque. The whole system consists of two sub systems, i.e. the extractor system including the rating input module and the recommendation system. The recommendation system is implemented as end-user mobile application such that the users are able to use it anytime and anywhere. The system successfully integrate data from different social media channels and in different format to provide users with useful information concerning public facilities in the form of recommendation (rating) and popularity of the facilities. The experiment has shown that above 90% of the data collected from the social media contains location-related information that is useful for further processing. The system has been tested using usability test, and it obtained an average users score 3.9 on a scale of 1 to 5.