<%
// @author Dawid Weiss
//
// PERFORMANCE/USER INTERFACE NOTE:
//
// What I do here is merely a demonstration. In real life the clustering
// process should be done in a separate "processing" stream, most likely
// a separate HTML frame that the user's browser requests data to.
// We don't want the user to wait with plain snippets until the clusters
// are created.
//
// Also: clustering is resource consuming, so a cache of recent queries
// would be in place. Besides, such cache would also be beneficial for the
// purpose of re-querying existing clusters (remember that the
// clustering extension may be a heuristic returning a DIFFERENT set of
// clusters for an identical input).
// See www.vivisimo.com for details of how this can be done using frames, or
// http://carrot.cs.put.poznan.pl for an example of a Javascript solution.
// cluster the hits
HitsCluster [] clusters = null;
if (clusterer != null) {
final long clusteringStart = System.currentTimeMillis();
try {
clusters = clusterer.clusterHits( details, Summary.toStrings(summaries) );
final long clusteringDuration = System.currentTimeMillis() - clusteringStart;
bean.LOG.info("Clustering took: " + clusteringDuration + " milliseconds.");
} catch (Exception e) {
// failed to do clustering (see below)
}
}
if (clusterer == null) {
%>No clustering extension found.<%
} else {
if (clusters == null) {
%>Unable to do clustering.<%
} else if (clusters.length == 0) {
%>No clusters found.<%
} else {
// display top N clusters and top Q documents inside them.
int N = 10;
int Q = 3;
int maxLabels = 2;
int displayCounter = 0;
N = Math.min(N, clusters.length );
for (int clusterIndex = 0 ; clusterIndex < N ; clusterIndex++) {
HitsCluster cluster = clusters[ clusterIndex ];
String [] clusterLabels = cluster.getDescriptionLabels();
// probably leave it on for now
//if (cluster.isJunkCluster()) continue;
// output cluster label.
%><%
for (int k=0;k0) out.print(", ");
out.print( Entities.encode(clusterLabels[k]) );
}
%>
<%
// now output sample documents from the inside
HitDetails[] documents = cluster.getHits();
if (documents.length > 0) {
%><%
for (int k = 0; k < Q && k < documents.length; k++) {
HitDetails detail = documents[ k ];
String title = detail.getValue("title");
String url = detail.getValue("url");
if (title == null || title.equals("")) title = url;
if (title.length() > 35) title = title.substring(0,35) + "...";
%>
- <%= Entities.encode(title) %>
<%
}
%>
<%
}
// ignore subclusters for now, ALTHOUGH HIERARCHICAL CLUSTERING
// METHODS DO EXIST AND ARE VERY USEFUL
// HitsCluster [] subclusters = cluster.getSubclusters();
}
}
}
%>