A study of topic modeling techniques for spoken document retrieval

This paper focuses on comparison of two common categories of topic modeling techniques for spoken document retrieval (SDR), namely document topic model (DTM) and word topic model (WTM). Apart from using the conventional unsupervised training strategy, we explore a supervised training strategy for estimating these topic models, assuming that user query logs along with click-through information of relevant documents can be utilized when building an SDR system. This attempt has the potential to associate relevant documents with queries even if they do not share any of the query words. Moreover, in order to lessen SDR performance degradation caused by imperfect speech recognition, we also leverage different levels of index features for topic modeling, including words, syllable-level units, and their combination. Experiments conducted on the TDT-2 SDR task show that the methods deduced from our proposed modeling framework are very promising when compared with a few existing retrieval approaches.