Even with the recent development of new types of social networking services such as microblogs, Bulletin Board Systems (BBS) remains popular for local communities and vertical discussions. These BBS sites have high volume of traffic everyday with user discussions on a variety of topics. Therefore it is difficult for BBS visitors to find the posts that they are interested in from the large amount of discussion threads. We attempt to explore several main characteristics of BBS, including organizational flexibility of BBS texts, high data volume and aging characteristic of BBS topics. Based on these characteristics, we propose a novel method of Online Topic Detection (OTD) on BBS, which mainly includes a representative post selection procedure based on Markov chain model and an efficient topic clustering algorithm with candidate topic set generation based on Aging Theory. Experimental results show that our method improves the performance of OTD in BBS environment in both detection accuracy and time efficiency. In addition, analysis on the aging characteristic of discussion topics shows that the generation and aging of topics on BBS is very fast, so it is wise to introduce candidate topic set generation strategy based on Aging Theory into the topic clustering algorithm.

An efficient, scalable speech recognition architecture combining topic detection and topic-dependent language modeling is proposed for multi-domain spoken language systems. In the proposed approach, the inferred topic is automatically detected from the user's utterance, and speech recognition is then performed by applying an appropriate topic-dependent language model. This approach enables users to freely switch between domains while maintaining high recognition accuracy. As topic detection is performed on a single utterance, detection errors may occur and propagate through the system. To improve robustness, a hierarchical back-off mechanism is introduced where detailed topic models are applied when topic detection is confident and wider models that cover multiple topics are applied in cases of uncertainty. The performance of the proposed architecture is evaluated when combined with two topic detection methods: unigram likelihood and SVMs (Support Vector Machines). On the ATR Basic Travel Expression Corpus, both methods provide a significant reduction in WER (9.7% and 10.3%, respectively) compared to a single language model system. Furthermore, recognition accuracy is comparable to performing decoding with all topic-dependent models in parallel, while the required computational cost is much reduced.