WWW/Internet - Portal do Software Público Brasileiro

IADIS International Conference WWW/Internet 2010measure the number of connected resources over the resource-specific query tag clouds generated byQueryCloud system.To measure this metric, we calculated the (strongly) connected components (Levine 1965) for the wholeQueryCloud resource-specific tag cloud network (cf. Strohmaier et al. 2010). Essentially, we could observethat 98.58% of all resources within Austria-Forum are reachable via a resource-specific query tag cloudgenerated by QueryCloud system (cf. Trattner et al. 2010c). Contrary to this, the Austria-Forum taggersgenerate a resource-specific tag cloud network which is to 93.97% connected (cf. Trattner et al. 2010c).Figure 5. Number of network components: QC dataset(blue line) vs. AF dataset (red line).Figure 6. Distribution of shortest path pair lengths: QCdataset (blue line) vs. AF dataset (red line).4.4 Measuring Retrieval QualityLast but not least we measured the retrieval quality of QueryCloud system, i.e. we examined theeffectiveness of the tool to navigate related documents within a Web based encyclopedia system (cf. Helic etal. 2010). In (Helic et al. 2010) we have shown that navigable tag cloud networks have certain properties. Cf.Kleinberg (Kleinberg 2000a, Kleinberg 200b and Kleinberg 2001) a navigable network can be formallydefined as network with a low diameter (Newman 2003) bounded polylogarithmically, i.e. by a polynomial inlog(N), where N is the number of nodes in the network, and an existing giant component, i.e. a stronglyconnected component containing almost all nodes. Additionally, Kleinberg defined an “efficiently” navigablenetwork as a network possessing certain structural properties so that it is possible to design efficientdecentralized search algorithms (algorithms that only have local knowledge of the network) (Kleinberg2000a, Kleinberg 200b and Kleinberg 2001). The delivery time (the expected number of steps to reach anarbitrary target node) of such algorithms is polylogarithmic or at most sub-linear in N.Thus, as a first step, we examined the structural properties defined by the resource-specific tag cloudnetwork of QueryCloud system. First, we investigated the tag cloud networks “connected component”distribution. As shown in Figure 5 QueryCloud generates a resource-specific tag cloud network whose largestconnected component contains almost all nodes (99%) of the network. Contrary to this, the AF datasetgenerates a resource-specific tag cloud network which is “only” connected to 93.97%.Thereafter, we calculated the number of short path pairs within QueryCloud’s resource-specific tag cloudnetwork. As Figure 6 shows, QueryCloud generates a resource-specific tag cloud network whose effectivediameter is around 6.3 hops while the AF dataset generates a resource-specific tag cloud network with aneffective diameter of around 9.9 hops. Putting the results of these two experiments together we can say thatQueryCloud produces a navigable tag cloud network (Helic et al. 2010, Kleinberg 2000a, Kleinberg 2000b,Kleinberg 2001).Now, since we have shown that the resource-specific tag cloud network of QueryCloud system isnavigable (cf. Kleinberg 2001), we implemented a decentralized searcher (see Algorithm 1) based on the239