Recently, a large amount of attention has been paid to efficient document retrieval from a gigantic data collection, such as in web search areas. For a keywords-search, the top-K computation scans the documents from multiple involved inverted lists and the computation should be stopped when there is no other unseen documents better than the top-K documents being seen so far, instead of traversing the whole lists. In this thesis, we give a comprehensive study and analyze the pros and cons of the state-of-the-art indexing structures and the top-K document retrieval techniques. Furthermore, we propose a new top-K evaluation framework from the following aspects. 1. Reorganizing inverted index– we study a new indexing structure such that more promising documents appear to the beginning of an inverted index. 2. Refining execution strategies – we study a new processing strategy such that a faster evaluation can be achieved. The superior of our proposed techniques has been demonstrated by a thorough experimental evaluation which compares our proposed techniques to the state-of-the-art approaches in the final section. It turns out that, on average, the response time of our methodologies is about 80% less than that of the reviewed approaches.