SMS based FAQ Retrieval using Theme Matching Scheme

As a participant of FIRE 2012 monolingual English SMS based FAQ Retrieval Task, we proposed a theme matching scheme [1]. Once again the scheme is implemented for the same task in FIRE 2013 having different set of SMS and FAQ queries. An SMS text usually consists of certain noisy terms due to the limitations of characters allowed in an SMS, lack of screen space and unintended typographical errors; thereby retrieving the relevant FAQ query/queries corresponding to an SMS query is a very challenging task. We observed that depending on the parts of speech, different terms in a text carry different significance values in representing the theme of the text. The said scheme is mostly based on this observation. In order to recognize the noisy terms of an SMS, three approximate matching measures were derived in the retrieval procedure. Although the FAQ and test queries in FIRE 2013 consist of not only SMS text queries but also speech transcribed queries, our theme matching scheme is found to be very effective for the first category i.e., SMS queries. The relevance set published for the speech transcribed query set is completely erroneous as there is absolutely no match between the test queries and the relevant FAQ queries. Accordingly, our scheme recognized the complete speech transcribed set as out of domain.