Efficient Combined Text and Spatial Search

Abstract

We present a search engine called TexSpaSearch that can search text documents with associated locations in space. We defined three search queries denoted as Q1(t), Q2(t, r) and Q3(p, r) for finding documents containing text t intersecting a disc centered at position p with radius r. Testing was performed using the UNB Connell Memorial Herbarium database whose records normally contain the location where plant specimens were collected along with associated textual data. The sample herbarium database of size $N= 40,791$ records with associated locations was indexed using a novel R*-tree and suffix tree data structure to achieve efficient search for the defined queries. Significant preprocessing was performed to transform the database into the index data structure used by TexSpaSearch. Testing was performed with 20 example Q1 text only queries to compare TexSpaSearch to a Google Search Appliance, as well as a significant number of example Q2 and Q3 queries. TexSpaSearch search results are ranked by a modified Lucene scoring algorithm, and combined with a spatial rank for Q2 search. A theoretical analysis shows that TexSpaSearch requires $O(A^{2}\overline{|b|})$ average time for Q1 search, where A is the number of single words in the query string t, and $\overline{|b|}$ is the average length of a subphrase in t. Q2 and Q3 queries require $O(A^{2}\overline{|b|} + Z\log _{\mathcal {M}}\mathcal {D}_N + y)$ and $O(\log _{\mathcal {M}}\mathcal {D}_N + y)$ time, respectively, where Z is the number of point records in the list $\mathcal {P}$ of text search results, $\mathcal {D}_N$ is the number of data objects indexed in the R*-tree for N records, $\mathcal {M}$ is the maximum number of entries of an interior node in the R*-tree, and y is the number of R*-tree leaf nodes found in range in a Q3 query.