The term-frequency inverse-document(tf-idf) paradigm which is often used in general search engines for ranking the relevance of documents in a corpus to a given user query, is based on the frequency of occurrence of the search key terms in the corpus. These search terms are mostly expressed in natural language thus requiring natural language processing methods. But for domain-speciffic search engines like a software download portal, search terms are usually expressed in forms that does not conform to grammatical rules present in natural language and as such, they cannot be tackled using natural language processing techniques. This thesis proposes named entity recognition using supervised machine learning methods as a means to understanding queries for such domain-speciffic search engines. Particularly, our main objective is to apply machine learning techniques to automatically learn to recognize and classify search terms according to named entity class of predefined categories they belong. By so doing, we are able to understand user intents and rank result sets according to their relevance to detected named entities present in search query. Our approach involved three machine learning algorithms; Hidden Markov Models (HMM), Conditional Random Field(CRF) and Neural Network(NN). We followed the supervised learning approach in training these algorithms using labeled training data from sample queries, we then evaluated their performance on new unseen queries. Our empirical results showed precisions of 93% for NN which was based on distributed representations proposed by Yoshua Bengio, 85.60% for CRF and 82.84% for HMM. CRF 's precision improved to about 2% , achieving 87.40% after we generated gazetteer-based and morphological features. From our results, we were able to prove that machine learning methods for named entity recognition is useful for understanding query intents.