Funnelback phrase search result snippets

[quimby]

Have noticed a behaviour with phrase searches in Funnelback. Firstly, phrase searching is working as expected, in other words a phrase in double-quotes is returning records that have the full string only.

However, when the results are generated the snippets (including the parts that are highlighted i.e. bolidicised) are not full text matches. It seems to boldicise any words from the query, not the full text of the phrase being searched on.

The result is users think that phrase searching is not working as it should (even though it actually is - it's just not generating snippets as you'd expect).

Anyone seen this before? Our FB version is 12.2.1.

Example is on the following URL:
http://economicdevelopment.vic.gov.au/search?query=%22education+and+training%22&collection=meta-web-ecodev

Phrase search but 3rd and 4th results show boldicising on 'and', or 'education and' rather than the full phrase only.

[gordongrace]

You've encountered a limitation on the default snippet highlighting mode within Funnelback. Looking at the data model:

transacation.response.resultPacket.queryHighlightRegex

Will contain the regular expression used for snippet highlighting - the default approach is to highlight all the terms, and ignore the phrase based arrangement.

You may need to modify the queryHighlightRegex value as a hook script to remove the 'OR'ing nature of the terms:

queryHighlightRegex": "(?i)\\bmickey\\b\\s\\bmouse\\b",

[quimby]

Thanks Gordon

Had a look at what you suggested and could definitely see the OR's (pipe characters), but simply removing them wasn't going to cut it (I found that on our system the regex was duplicating the terms multiple times - not sure if that's expected).

Regardless, what I settled on was creating the regex from the query itself rather than trying to massage what was already there. Also needed to add logic so it only used the 'AND'-ing version if there were actually double-quotes. End result was this (for those interested):

if (transaction.question.query.length() > transaction.question.query.replace('"','').length()) { 
  transaction.response.resultPacket.queryHighlightRegex = "(i?)\\b" + transaction.question.query.replace('"','').replaceAll(" ", "\\\\b\\\\s\\\\b") + "\\b";
}

Seems to be working ok. Am currently getting it tested/reviewed.