[
https://issues.apache.org/jira/browse/LUCENE-579?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12512743
]
Shahan Khatchadourian commented on LUCENE-579:
----------------------------------------------
DocumentWriter seems to be the culprit in adding 1 to the previous token's
endOffset. It may not be possible to provide token offsets that "undo" this
operation since it is not possible to determine the order in which tokens are
handled as they are grouped by field which doesn't necessarily correspond to
document-order. This would also interfere with custom synonym tokens since
custom token offsets are no longer guaranteed.
I suggest that there be a flag in Fieldable or IndexWriter which allows exact
provided offsets to be stored rather than increased by one. There does not seem
to be any sanity checks on offset values during reading/writing the term vector.
A current workaround to this issue is to store the correct startOffset, but
leave the endOffset as -1. This has the effect of undoing the +1 of the
previous token's endOffset but prevents endOffset information from being
available without retokenizing/reparsing the original document.
> TermPositionVector offsets incorrect if indexed field has multiple values and
> one ends with non-term chars
> ----------------------------------------------------------------------------------------------------------
>
> Key: LUCENE-579
> URL: https://issues.apache.org/jira/browse/LUCENE-579
> Project: Lucene - Java
> Issue Type: Bug
> Components: Analysis
> Affects Versions: 1.9
> Reporter: Keiron McCammon
>
> If you add multiple values for a field with term vector positions and offsets
> enabled and one of the values ends with a non-term then the offsets for the
> terms from subsequent values are wrong. For example (note the '.' in the
> first value):
> IndexWriter writer = new IndexWriter(directory, new SimpleAnalyzer(),
> true);
> Document doc = new Document();
> doc.add(new Field("", "one.", Field.Store.YES, Field.Index.TOKENIZED,
> Field.TermVector.WITH_POSITIONS_OFFSETS));
> doc.add(new Field("", "two", Field.Store.YES, Field.Index.TOKENIZED,
> Field.TermVector.WITH_POSITIONS_OFFSETS));
> writer.addDocument(doc);
> writer.optimize();
> writer.close();
> IndexSearcher searcher = new IndexSearcher(directory);
> Hits hits = searcher.search(new MatchAllDocsQuery());
> Highlighter highlighter = new Highlighter(new SimpleHTMLFormatter(),
> new QueryScorer(new TermQuery(new Term("", "camera")),
> searcher.getIndexReader(), ""));
> for (int i = 0; i < hits.length(); ++i) {
> TermPositionVector v = (TermPositionVector)
> searcher.getIndexReader().getTermFreqVector(
> hits.id(i), "");
> StringBuilder str = new StringBuilder();
> for (String s : hits.doc(i).getValues("")) {
> str.append(s);
> str.append(" ");
> }
>
> System.out.println(str);
> TokenStream tokenStream = TokenSources.getTokenStream(v, false);
> String[] terms = v.getTerms();
> int[] freq = v.getTermFrequencies();
> for (int j = 0; j < terms.length; ++j) {
> System.out.print(terms[j] + ":" + freq[j] + ":");
>
> int[] pos = v.getTermPositions(j);
>
> System.out.print(Arrays.toString(pos));
>
> TermVectorOffsetInfo[] offset = v.getOffsets(j);
> for (int k = 0; k < offset.length; ++k) {
>
> System.out.print(":");
>
> System.out.print(str.substring(offset[k].getStartOffset(),
> offset[k].getEndOffset()));
> }
>
> System.out.println();
> }
> }
> searcher.close();
> If I run the above I get:
> one:1:[0]:one
> two:1:[1]: tw
> Note that the offsets for the second term are off by 1.
> It seems to be that the length of the value that is stored is not taken into
> account when calculating the offset for the fields of the next value.
> I noticed ths problem when using the highlight contrib package which can make
> use of term vectors for highlighting. I also noticed that the offset for the
> second string is +1 the end of the previous value, so when concatenating the
> fields values to pass to the hgighlighter I add to append a ' ' character
> after each string...which is quite useful, but not documented anywhere.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
