UNSUPERVISED AUTOMATED TOPIC DETECTION, SEGMENTATION AND LABELING OF CONVERSATIONS

A method for information processing includes receiving in a computer a corpus of recorded conversations, with two or more speakers participating in each conversation. Respective frequencies of occurrence of multiple words in each of a plurality of chunks in each of the recorded conversations are computed. Based on the frequencies of occurrence of the words over the conversations in the corpus, an optimal set of topics to which the chunks can be assigned is derived, such that the optimal set maximizes a likelihood that the chunks will be generated by the topics in the set. A recorded conversation from the corpus is segmented using the derived topics into a plurality of segments, such that each segment is classified as belonging to a particular topic in the optimal set.