Abstract-This paper investigates how to integrate multi-modal features for story boundary detection in broadcast news. The detection problem is formulated as a classification task, i.e., classifying each candidate into boundary/non-boundary based on a set of features. We use a diverse collection of features from text, audio and video modalities: lexical features capturing the semantic shifts of news topics and audio/video features reflecting the editorial rules of broadcast news. We perform a comprehensive evaluation on boundary detection performance for six popular classifiers, including decision tree (DT), Bayesian network (BN), naive Bayesian (NB) classifier, multi-layer peceptron (MLP), support vector machines (SVM) and maximum entropy (ME) classifier. Results show that BN and DT can generally achieve superior performances over other classifiers and BN offers the best F1-measure. Analysis of BN and DT reveals important inter-feature dependencies and complementarities that contribute significantly to the performance gain.

SUMMARYIn this paper, we propose integration of multimodal features using conditional random fields (CRFs) for the segmentation of broadcast news stories. We study story boundary cues from lexical, audio and video modalities, where lexical features consist of lexical similarity, chain strength and overall cohesiveness; acoustic features involve pause duration, pitch, speaker change and audio event type; and visual features contain shot boundaries, anchor faces and news title captions. These features are extracted in a sequence of boundary candidate positions in the broadcast news. A linear-chain CRF is used to detect each candidate as boundary/non-boundary tags based on the multimodal features. Important interlabel relations and contextual feature information are effectively captured by the sequential learning framework of CRFs. Story segmentation experiments show that the CRF approach outperforms other popular classifiers, including decision trees (DTs), Bayesian networks (BNs), naive Bayesian classifiers (NBs), multilayer perception (MLP), support vector machines (SVMs) and maximum entropy (ME) classifiers.

This paper proposes to perform probabilistic latent semantic analysis (PLSA) for broadcast news (BN) story segmentation. PLSA exploits a deeper underlying relation among terms beyond their occurrences thus conceptual matching can be employed to replace literal term matching. Different from text segmentation, lexical based BN story segmentation has to be carried out over LVCSR transcripts, where the incorrect recognition of out-of-vocabulary words inevitably impacts the semantic relation. We use phoneme subwords as the basic term units to address this problem. We integrate a cross entropy measurement with PLSA to depict lexical cohesion and compare its performance with the widely used cosine similarity metric. Furthermore, we evaluate two approaches, namely TextTiling and dynamic programming (DP), for story boundary identification. Experimental results show that the PLSA based methods bring a significant performance boost to story segmentation and the cross entropy based DP approach provides the best performance.