implement feedback in lucene - lucene

Question

Can someone give me hints to apply pseudo-feedback in lucene. I can not find much help on google. I am using Similarity classes.
Is there any class in lucene which I can extend to implement feedback ?
thanks.

Answer 1

Assuming you are referring to this relevance feedback method, Once you have the TopDocs for your original query, iterate through how ever many (let's say we'll want the top 25 terms for the top 25 docs of the original query) records you desire, and call IndexReader.getTermVectors(int), which will grab the information you need. Iterate through each. while storing the term frequencies in a hash map would be the implementation the immediately occurs to me.
Something like:
//Get the original results
TopDocs docs = indexsearcher.search(query,25);
HashMap<String,ScorePair> map = new HashMap<String,ScorePair>();
for (int i = 0; i < docs.scoreDocs.length; i++) {
//Iterate fields for each result
FieldsEnum fields = indexreader.getTermVectors(docs.scoreDocs[i].doc).iterator();
String fieldname;
while (fieldname = fields.next()) {
//For each field, iterate it's terms
TermsEnum terms = fields.terms().iterator();
while (terms.next()) {
//and store it
putTermInMap(fieldname, terms.term(), terms.docFreq(), map);
}
}
}
List<ScorePair> byScore = new ArrayList<ScorePair>(map.values());
Collections.sort(byScore);
BooleanQuery bq = new BooleanQuery();
//Perhaps we want to give the original query a bit of a boost
query.setBoost(5);
bq.add(query,BooleanClause.Occur.SHOULD);
for (int i = 0; i < 25; i++) {
//Add all our found terms to the final query
ScorePair pair = byScore.get(i);
bq.add(new TermQuery(new Term(pair.field,pair.term)),BooleanClause.Occur.SHOULD);
}
}
//Say, we want to score based on tf/idf
void putTermInMap(String field, String term, int freq, Map<String,ScorePair> map) {
String key = field + ":" + term;
if (map.containsKey(key))
map.get(key).increment();
else
map.put(key,new ScorePair(freq,field,term));
}
private class ScorePair implements Comparable{
int count = 0;
double idf;
String field;
String term;
ScorePair(int docfreq, String field, String term) {
count++;
//Standard Lucene idf calculation. This is calculated once per field:term
idf = (1 + Math.log(indexreader.numDocs()/((double)docfreq + 1))) ^ 2;
this.field = field;
this.term = term;
}
void increment() { count++; }
double score() {
return Math.sqrt(count) * idf;
}
//Standard Lucene TF/IDF calculation, if I'm not mistaken about it.
int compareTo(ScorePair pair) {
if (this.score() < pair.score()) return -1;
else return 1;
}
}
(I make no claim that this is functional code, in it's current state.)