TVPulse: Improvements on detecting TV highlightsin Social Networks using metadata and semanticsimilarity

Abstract

Sharing live experiences in social networks is agrowing trend. That includes posting comments and sentimentsabout TV programs. Automatic detection of messages withcontents related to TV opens new opportunities for the industryof entertainment information.This paper describes a system that detects TV highlights in oneof the most important social networks - Twitter. Combining Twit-ter's messages and information from an Electronic ProgrammingGuide (EPG) enriched with external metadata we built a modelthat matches tweets with TV programs with an accuracy over80{\%}. Our model required the construction of semantic profilesfor the Portuguese language. These semantic profiles are usedto identify the most representative tweets as highlights of a TVprogram. Measuring semantic similarity with those tweets it ispossible to gather other messages within the same context. Thisstrategy improves the recall of the detection. In addition wedeveloped a method to automatically gather other related webresources, namely Youtube videos. TVPulse: Improvements on detecting TV highlights in Social Networks using metadata and semantic similarity.