[SOLR-2939] Clustering of multilingual search results

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 3.6, 4.0-ALPHA
• Component/s: contrib - Clustering
• Assignee: Stanislaw Osinski
• Reporter: Stanislaw Osinski
• Created: 02/Dec/11 13:25
• Updated: 10/May/13 10:40
• Resolved: 17/Dec/11 13:43

Description

Carrot2 internally supports clustering of multilingual search results. The clustering component should allow passing a language field to Carrot2. This feature would need at least two new parameters: carrot.lang for the name of Solr field that contains the language code (ISO 639) and a carrot.lcmap field similar to the one in language recognizer to map arbitrary strings to ISO 639 codes.

Another feature of language recognizer we should mirror is the expansion of the {{

{lang}

}} token in field names into the language code of the document (in case of multiple languages per document – the first Carrot2-supported language code). The feature seems easy to implement in the non-distributed setting of Solr, but the simple implementation isn't going to work in the distributed setting because the name of the specific field to be fetched depends on the content (language) of each matching document. Looking at the SearchClusteringEngine.getFieldsToLoad(SolrQueryRequest) method, a quick but costly solution would be to load the contents of all stored fields. I'm not too strong in distributed-mode Solr, but maybe this could be optimized so that only the required fields get fetched?