Finding an λ-representative subset from massive data

Retrieving representative information from large-scale data becomes an important research issue nowadays. This paper focuses on certain aspects of representativeness in database queries and web search, and proposes an approach to extracting a subset of results from original search results in light of high coverage and low redundancy. In the paper, the notion of λ-Represent is introduced based on similarities and related fuzzy operations, which enables us to describe the λ-Represent relationship between the sets of data objects. Then, the λ-Representative problem is formulated as an extension of the typical set covering problem, which leads to developing a heuristic algorithm (namely, LamRep) to cope with the problem effectively. In LamRep, a “vote” mechanism is proposed to overcome the limitation of the naive greedy algorithm. Data experiments on benchmark data show that LamRep outperforms the other approaches.