Internet forum is a web community that brings people in different geographical locations together. Members of the forum exchange ideas and expertise and as a result generate huge amount of content on different topics on daily basis. A good percentage of human generated content of Internet forums have been found to be question-answer (QA) pairs. These QA pairs are useful for automating question answering system. Mining these QA pairs has become a hot issue in the research community. Effective mining of the QA pairs is being hindered by a number of factors. Lexical chasm that renders some Information Retrieval (IR) techniques less effective, casual language that creates noisy data; multiple authors that bring about unfocused topics are some of the issues that need to be addressed. In this paper, an extensive overview of the strategies and findings relevant to these three challenges are addressed. The survey revealed that researchers are adopting non-lexical features as against lexical to resolve the issue of data sparseness. Noise level is mostly controlled using conventional dictionary rather than using domain-specific dictionary.