Towards optimal keyword-based content dissemination in DHT-based P2P networks

Keyword-based content alert services, e.g., Google Alerts and Microsoft Live Alerts, empower the end users with the ability to automatically receive useful and most recent content. In this paper, we leverage the favorable properties of DHTs, such as scalability, and propose a design of a scalable keyword-based content alert service. The DHT-based architecture matches textual documents with queries based on document terms: For each term, the implementation assigns a home node that is responsible for handling documents and queries that contain the term. The main challenge of this keyword-based matching scheme is the high number of terms that appear in a typical document resulting in a high publication cost. Fortunately, a document can be forwarded to the home nodes of a carefully selected subset of terms without incurring false negatives. In this paper we focus on the MTAF problem of minimizing the number of selected terms to forward the published content. We show that the problem is NP-hardness, and consider centralized and DHT-based solutions. Experimental results based on real datasets indicate that the proposed solutions are efficient compared to existing approaches. In particular, the similarity-based replication of filters that is a key element of our solution is shown to mitigate the effect of hotspots that arise due to the fact that some document terms are substantially more popular than the others, both inside documents and queries.