On 03.02.2010, at 13:54, Lukas Kahwe Smith wrote:
> The issue is that I have multiple fields of data (names, address etc) that
> should all be relevant for the auto suggest. Furthermore a "phrase" entered
> can either match on one field or any combination of fields. Phrase in this
> context means separated by spaces or dash. For this I found the above
> approach the only feasible solution.
Actually I plan to write a bigger blog post about the approach. In order to
match the different fields I actually have a separate core with an index
dedicated to auto suggest alone where I merge all fields together via some
javascript code:
function splitEntity(row) {
var name = row.get('name');
var telefon = row.get('telefon');
var city = row.get('city');
var arr = new java.util.ArrayList();
arr = splitField(arr, name);
arr = splitField(arr, firstname);
arr = splitField(arr, city);
if (telefon) {
arr.add(telefon);
}
row.put('all', arr);
row.remove('name');
row.remove('telefon');
row.remove('city');
return row;
}
function splitField(arr, value) {
if (value) {
arr.add(value);
var pieces = value.toLowerCase();
arr.add(pieces);
if (pieces.indexOf('-') != -1 || pieces.indexOf(' ') != -1) {
value = value.replaceAll(" ", "_");
pieces = pieces.replaceAll(" ", "-");
pieces = pieces.split('-');
// TODO: generate more/all permutations? drop stop words?
for (var i = 0; i < pieces.length; i++) {
arr.add('{'+pieces[i]+'}'+value);
}
}
}
return arr;
}
This way I can then use terms for a single word entered and a facet prefix
search with the last term as the prefix and the rest as the query for multi
term entries into the auto suggest box.
The idea is that I can then enter any part of any of the fields, but I will
then be suggested the entire phrase in that field:
So if I have a field:
Foo Bar Ding Dong
and I enter "ding" into the search box, I would get a suggestion of "Foo Bar
Ding Dong"
regards,
Lukas Kahwe Smith
m...@pooteeweet.org
