Abstract: In this topic there are large numbers of documents which are cover more information about any topic. We are extracting one keyword from that document, when we are extracting this keyword can easily retrieve whole document. However, even a small piece contains a variety of words, which are potentially related to several topics; more- over, using an automatic speech recognition (ASR) system introduce errors among them. There for, it is dif?cult to infer precisely the in sequence requirements of the discussion participants. We ?rst propose an algorithm to extract keywords from the output of an ASR system which makes use of topic modeling techniques and of a sub modular reward function which favors range in the keyword set, to match the possible range of topics and reduce ASR noise. This method is to derive many topically divided queries starting this keyword set, in organize to take full advantage of the probability of making at least one related reference when with these queries to search over the English Wikipedia. Examples like Fisher, AMI, and ELEA conversational corpora.

Keywords: Document recommendation, information retrieval, keyword extraction, meeting analysis, topic modeling.