Identifier

etd-01162014-201105

Degree

Doctor of Philosophy (PhD)

Department

Computer Science

Document Type

Dissertation

Abstract

Let D = {d_1, d_2, ...} be a collection of string documents of n characters in total, which are drawn from an alphabet set Sigma =[sigma] ={1,2,3,...sigma}. The top-k document retrieval problem is to maintain D as a data structure, such that when ever a query Q=(P, k) comes, we can report (the identifiers of) those k documents that are most relevant to the pattern P (of p characters). The relevance of a document d_r with respect to a pattern P is captured by score(P, d_r), which can be any function of the set of locations where P occurs in d_r. Finding the most relevant documents to the user query is the central task of any web-search engine. In the case of web-data, the documents can be demarcated along word boundaries. All the search engines use inverted index as the back-bone data structure. For each word occurring in the document collection, the inverted index stores the list of documents where it appears. It is often augmented with relevance score and/or positional information. However, when data consists of strings (e.g., in bioinformatics or Asian language texts), there are no word demarcation boundaries and the queries are arbitrary substrings instead of being proper valid words. In this case, string data structures have to be used and central approach is to use suffix tree (or string B-tree) with appropriate augmenting data structures. The work by Hon, Shah and Vitter [FOCS 2009], and Navarro and Nekrich [SODA 2012] resulted in a linear space data structure with optimal O(p+k) query time solution for this problem. This was based on geometric interpretation of the query. We extend this central problem, in two important areas of massive data sets. First, we consider an external memory disk based index, where we give near optimal results. Next, we consider compression aspects of data structure, reducing the storage space. This is central goal of the active research field of succinct data structures. We present several results, which improve upon several previous results, and are currently the best known space-time trade-offs in this area.

Date

2014

Document Availability at the Time of Submission

Release the entire work immediately for access worldwide.

Committee Chair

Shah, Rahul

Share

COinS