Package org.apache.lucene.search.grouping

This module enables search result grouping with Lucene, where hits with the same value in the specified single-valued group field are grouped together.

See:
Description

Class Summary

AbstractAllGroupHeadsCollector<GH extends AbstractAllGroupHeadsCollector.GroupHead>	This collector specializes in collecting the most relevant document (group head) for each group that match the query.
AbstractAllGroupHeadsCollector.GroupHead<GROUP_VALUE_TYPE>	Represents a group head.
AbstractAllGroupsCollector<GROUP_VALUE_TYPE>	A collector that collects all groups that match the query.
AbstractDistinctValuesCollector<GC extends AbstractDistinctValuesCollector.GroupCount<?>>	A second pass grouping collector that keeps track of distinct values for a specified field for the top N group.
AbstractDistinctValuesCollector.GroupCount<GROUP_VALUE_TYPE>	Returned by AbstractDistinctValuesCollector.getGroups(), representing the value and set of distinct values for the group.
AbstractFirstPassGroupingCollector<GROUP_VALUE_TYPE>	FirstPassGroupingCollector is the first of two passes necessary to collect grouped hits.
AbstractGroupFacetCollector	Base class for computing grouped facets.
AbstractGroupFacetCollector.FacetEntry	Represents a facet entry with a value and a count.
AbstractGroupFacetCollector.GroupedFacetResult	The grouped facet result.
AbstractGroupFacetCollector.SegmentResult	Contains the local grouped segment counts for a particular segment.
AbstractSecondPassGroupingCollector<GROUP_VALUE_TYPE>	SecondPassGroupingCollector is the second of two passes necessary to collect grouped docs.
BlockGroupingCollector	BlockGroupingCollector performs grouping with a single pass collector, as long as you are grouping by a doc block field, ie all documents sharing a given group value were indexed as a doc block using the atomic IndexWriter.addDocuments() or IndexWriter.updateDocuments() API.
CollectedSearchGroup<T>	Expert: representation of a group in AbstractFirstPassGroupingCollector, tracking the top doc and FieldComparator slot.
GroupDocs<GROUP_VALUE_TYPE>	Represents one group in the results.
GroupingSearch	Convenience class to perform grouping in a non distributed environment.
SearchGroup<GROUP_VALUE_TYPE>	Represents a group that is found during the first pass search.
TopGroups<GROUP_VALUE_TYPE>	Represents result returned by a grouping search.

Enum Summary

TopGroups.ScoreMergeMode

How the GroupDocs score (if any) should be merged.

Package org.apache.lucene.search.grouping Description

This module enables search result grouping with Lucene, where hits with the same value in the specified single-valued group field are grouped together. For example, if you group by the author field, then all documents with the same value in the author field fall into a single group.

Grouping requires a number of inputs:

• groupField: this is the field used for grouping. For example, if you use the author field then each group has all books by the same author. Documents that don't have this field are grouped under a single group with a null group value.
• groupSort: how the groups are sorted. For sorting purposes, each group is "represented" by the highest-sorted document according to the groupSort within it. For example, if you specify "price" (ascending) then the first group is the one with the lowest price book within it. Or if you specify relevance group sort, then the first group is the one containing the highest scoring book.
• topNGroups: how many top groups to keep. For example, 10 means the top 10 groups are computed.
• groupOffset: which "slice" of top groups you want to retrieve. For example, 3 means you'll get 7 groups back (assuming topNGroups is 10). This is useful for paging, where you might show 5 groups per page.
• withinGroupSort: how the documents within each group are sorted. This can be different from the group sort.
• maxDocsPerGroup: how many top documents within each group to keep.
• withinGroupOffset: which "slice" of top documents you want to retrieve from each group.

The implementation is two-pass: the first pass (TermFirstPassGroupingCollector) gathers the top groups, and the second pass (TermSecondPassGroupingCollector) gathers documents within those groups. If the search is costly to run you may want to use the CachingCollector class, which caches hits and can (quickly) replay them for the second pass. This way you only run the query once, but you pay a RAM cost to (briefly) hold all hits. Results are returned as a TopGroups instance.

This module abstracts away what defines group and how it is collected. All grouping collectors are abstract and have currently term based implementations. One can implement collectors that for example group on multiple fields.

Known limitations:

• For the two-pass grouping search, the group field must be a single-valued indexed field (or indexed as a SortedDocValuesField). FieldCache is used to load the SortedDocValues for this field.
• Although Solr support grouping by function and this module has abstraction of what a group is, there are currently only implementations for grouping based on terms.
• Sharding is not directly supported, though is not too difficult, if you can merge the top groups and top documents per group yourself.

Typical usage for the generic two-pass grouping search looks like this using the grouping convenience utility (optionally using caching for the second pass search):

  GroupingSearch groupingSearch = new GroupingSearch("author");
  groupingSearch.setGroupSort(groupSort);
  groupingSearch.setFillSortFields(fillFields);

  if (useCache) {
    // Sets cache in MB
    groupingSearch.setCachingInMB(4.0, true);
  }

  if (requiredTotalGroupCount) {
    groupingSearch.setAllGroups(true);
  }

  TermQuery query = new TermQuery(new Term("content", searchTerm));
  TopGroups<BytesRef> result = groupingSearch.search(indexSearcher, query, groupOffset, groupLimit);

  // Render groupsResult...
  if (requiredTotalGroupCount) {
    int totalGroupCount = result.totalGroupCount;
  }

To use the single-pass BlockGroupingCollector, first, at indexing time, you must ensure all docs in each group are added as a block, and you have some way to find the last document of each group. One simple way to do this is to add a marker binary field:

  // Create Documents from your source:
  List<Document> oneGroup = ...;
  
  Field groupEndField = new Field("groupEnd", "x", Field.Store.NO, Field.Index.NOT_ANALYZED);
  groupEndField.setIndexOptions(IndexOptions.DOCS_ONLY);
  groupEndField.setOmitNorms(true);
  oneGroup.get(oneGroup.size()-1).add(groupEndField);

  // You can also use writer.updateDocuments(); just be sure you
  // replace an entire previous doc block with this new one.  For
  // example, each group could have a "groupID" field, with the same
  // value for all docs in this group:
  writer.addDocuments(oneGroup);

Then, at search time, do this up front:

  // Set this once in your app & save away for reusing across all queries:
  Filter groupEndDocs = new CachingWrapperFilter(new QueryWrapperFilter(new TermQuery(new Term("groupEnd", "x"))));

Finally, do this per search:

  // Per search:
  BlockGroupingCollector c = new BlockGroupingCollector(groupSort, groupOffset+topNGroups, needsScores, groupEndDocs);
  s.search(new TermQuery(new Term("content", searchTerm)), c);
  TopGroups groupsResult = c.getTopGroups(withinGroupSort, groupOffset, docOffset, docOffset+docsPerGroup, fillFields);

  // Render groupsResult...

Or alternatively use the GroupingSearch convenience utility:

  // Per search:
  GroupingSearch groupingSearch = new GroupingSearch(groupEndDocs);
  groupingSearch.setGroupSort(groupSort);
  groupingSearch.setIncludeScores(needsScores);
  TermQuery query = new TermQuery(new Term("content", searchTerm));
  TopGroups groupsResult = groupingSearch.search(indexSearcher, query, groupOffset, groupLimit);

  // Render groupsResult...

Note that the groupValue of each GroupDocs will be null, so if you need to present this value you'll have to separately retrieve it (for example using stored fields, FieldCache, etc.).

Another collector is the TermAllGroupHeadsCollector that can be used to retrieve all most relevant documents per group. Also known as group heads. This can be useful in situations when one wants to compute group based facets / statistics on the complete query result. The collector can be executed during the first or second phase. This collector can also be used with the GroupingSearch convenience utility, but when if one only wants to compute the most relevant documents per group it is better to just use the collector as done here below.

  AbstractAllGroupHeadsCollector c = TermAllGroupHeadsCollector.create(groupField, sortWithinGroup);
  s.search(new TermQuery(new Term("content", searchTerm)), c);
  // Return all group heads as int array
  int[] groupHeadsArray = c.retrieveGroupHeads()
  // Return all group heads as FixedBitSet.
  int maxDoc = s.maxDoc();
  FixedBitSet groupHeadsBitSet = c.retrieveGroupHeads(maxDoc)

For each of the above collector types there is also a variant that works with ValueSource instead of of fields. Concretely this means that these variants can work with functions. These variants are slower than there term based counter parts. These implementations are located in the org.apache.lucene.search.grouping.function package, but can also be used with the GroupingSearch convenience utility