| Information Retrieval (IR) is one of the most hot research areas in computer science, and its main role is to represent, organize and store the useful information items for the aim of future querying. By avail of modern computer and network technologies, we can efficiently retrieve much information. But along with the increasing of the total sum of information more and more rapidly, how to enforce information retrieval under new circumstance challenge and achieve better performance, is a research issue unresolved to researchers.Modern information retrieval is always discussed on the basis of certain kinds of model or indexing mechanism. This paper mainly proposes some model based on Inter-Relevant Successive Trees (IRST), which is developed for many years in our research group. In this paper, we have adapted IRST successfully better in full-text domain and gained good performance in the area of XML related semi-structured data.Firstly this paper discusses the model of IRST under the context of text retrieval. To enhance the speed of querying, it alters two dimensional IRST to better sort order vectors, which is called DOST (Double-Order Successive Trees). Then it further investigates the relationship between DOST and PAT array to make out the theory basis and abilities of the IRST family. As the common foundation of text retrieval and XML retrieval, Two-Dimensional Successive Trees (TWST) is important to support both non-structured and semi-structured data. So this paper investigate some critical topics about TWST, such as sub-db and db merging, inserting and deleting performance, code compression algorithms.How to avail the structures included in semi-structure data to promote both normal and full-text retrieval is still a problem to be improved. This paper proposes XML Inter-Relevant successive Trees ( Xistree), then it discusses the properties and related algorithms concerning Xistree, at last it introduces the organization of the experiments referring to XISS, Timber and XIndice.Concept assisted information retrieval is a new research area in recent years, it's concerned by people for its accuracy and good user interface so as to promote user usability, so it will be an important trend in the near future. This paper mainly focuses on the concept querying problems, which includes SLCA algorithm on document structure with node name, concept semantic model creating, and relevance numbering& ranking methods, etc.This article also builds a robust algorithm by methodically combining two different mining algorithms on FP-tree while adjusting the mining strategy dynamically and automatically during a complete process of Frequent Pattern Mining. This article firstly proposes the Naive Depth First Search algorithm (NDFS) that is based on FP-tree, and then briefly analyzes its performance on different datasets. After that, a new self-adaptive algorithm (SAFP) is proposed, which combines the NDFS with the FP-growth by a dynamic mining strategy on conditional FP-trees. Experiments demonstrate that the SAFP is more robust and efficient than both the NDFS and the FP-growth on various datasets.Finally, this paper summarizes the work till now, and tries to combine the technologies discussed in former chapters to constitute a retrieving system that includes structured, semi-structured and non-structured data. At last the demo system and the project that applying TWST is briefly introduced. |
