Finding and tracking subjects within an ongoing debate

This paper describes a new algorithm for finding and tracking different subjects within an ongoing debate. The algorithm finds blocks of co-occurring terms, representing subjects, including blocks for which the term co-occurrence pattern forms a ring topology. We used short online debate forum data and longer summary bulletins to assess the extent to which the algorithm could correctly detect subjects, according to the judgements of human evaluators. The results show that it could normally detect subject-shifting and track different subjects over time in online debate forums and with adjustments could find subjects in bulletins, but could not track the subjects in the bulletins because the interlinking between subjects was too dense in the longer documents.