Efficient Keyword Searching in Large-scale Social Network Service

Abstract

Different from traditional web searching, the relevant information for a social network system (SNS) is commonly the content from his/her friends. Such a difference makes content indexing extremely difficult for an online social network (OSN) search system because every user has an individual view during searching. Building such a per-user view index over existing SNS Key-Value stores raises a large amount of communication cost due to the complex interconnections among OSN users, making the search system unscalable. To address the problem, we propose a novel protocol called summary index to support keyword searching. In the protocol, each user keeps a directory of the succinct summaries of his/her neighbors, and checks these summaries for potential hits before sending any queries. Two factors contribute to the low overhead of our design: the summary index representations are memory efficient, and the summary dissemination for index updating is communication efficient. First, we design an incremental scalable Bloom filter for summarizing the content constantly generated by a neighbor of a user. For an issued query by a user, the search system first checks against the summary index for a user's neighbor to predict the neighbors likely having desired content. Thus, the search system saves a significant inter-server communication cost by avoiding exhaustively transmitting the query to all the neighbors. Second, to further reduce the overhead for maintaining the social index, we leverage the piggyback strategy which exploits the links with high social strengths to avoid redundant messages during updating the per-user view summary index. We conduct comprehensive simulations using traces from real world systems to evaluate this design. Results show that our scheme significantly outperforms existing schemes for OSN searching in terms of inter-sever traffic by 98 percent.