Lucene.NET Search Highlighting that respects HTML Tags

Question

I am trying to highlight search terms in a block of HTML, the problem is if a user does a search for "color", this:

<span style='color: white'>White</span>

becomes: <span style='<b>color</b>: white'><b>White</b></span>

and obviously, messing up my style is not a good idea.

Here is the code I am using:

        Query parsedQuery = parser.Parse(luceneQuery);
        StandardAnalyzer Analyzer = new StandardAnalyzer();
        SimpleHTMLFormatter formatter = new SimpleHTMLFormatter("<b class='search'>", "</b>");

        QueryScorer scorer = new QueryScorer(parsedQuery);
        Highlighter highlighter = new Highlighter(formatter, scorer);

        highlighter.SetTextFragmenter(new SimpleFragmenter());
        Highlighter.GetBestFragment(Analyzer, propertyName, invocation.ReturnValue.ToString())

I'm guessing the problem is that i need a different Fragmenter, but I'm not sure. Any help would be appreciated.

Answer 1

I think I figured it out...

I subclassed StandardAnalyzer and changed TokenStream to this:

public override Lucene.Net.Analysis.TokenStream TokenStream(string fieldName, System.IO.TextReader reader)
    {
        var start = base.TokenStream(fieldName, reader);
        HtmlStripCharFilter filter = new HtmlStripCharFilter(reader);
        TokenStream result = new StandardFilter(filter);
        return new StopFilter(new LowerCaseFilter(result), this.stopSet);
    }

and Implemented HtmlStripCharFilter as:

public class HtmlStripCharFilter : Lucene.Net.Analysis.CharTokenizer
{
    private bool inTag = false;

    public HtmlStripCharFilter(TextReader input)
        : base(input)
    {
    }

    protected override bool IsTokenChar(char c)
    {
        if (c == '<' && inTag == false)
        {
            inTag = true;
            return false;
        }
        if (c == '>' && inTag)
        {
            inTag = false;
            return false;
        }

        return !inTag && !Char.IsWhiteSpace(c);
    }
}

It's headed in the right direction, but still needs a lot more work before it's done. If anyone has a better solution (read "TESTED" solution) I would love to hear it.