In real-world information retrieval (IR) tasks, the searched items and/or the users' queries often have implicit information associated with them -- information that describes unspecified aspects of the items or queries. For example, in web search tasks, web pages are often pointed to by hyperlinks (known as anchors) from other pages, and thus have human-generated succinct descriptions of their content (anchor text) associated with them. This indirectly available information has been shown to improve search effectiveness for different retrieval tasks. However, in many real-world IR challenges this information is sparse in the data; i.e., it is incomplete or missing in a large portion of the data. In this work, we explore how to discover and use implicit information in large amounts of data in the context of IR. We present a general perspective for discovering implicit information and demonstrate how to use the discovered data in four specific IR challenges: (1) finding relevant records in semi-structured databases where many records contain incomplete or empty fields; (2) searching web pages that have little or no associated anchor text; (3) using click-through records in web query logs to help search pages that have no or very few clicks; and (4) discovering plausible geographic locations for web queries that contain no explicit geographic information. The intuition behind our approach is that data similar in some aspects are often similar in other aspects. Thus we can (a) use the observed information of queries/documents to find similar queries/documents, and then (b) utilize those similar queries/documents to reconstruct plausible implicit information for the original queries/documents. We develop language modeling based techniques to effectively use content similarity among data for our work. Using the four different search tasks on large-scale noisy datasets, we empirically demonstrate the effectiveness of our approach. We further discuss the advantages and weaknesses of two complementary approaches within our general perspective of handling implicit information for retrieval purpose. Taken together, we describe a general perspective that uses contextual similarity among data to discover implicit information for IR challenges. Using this general perspective, we formally present two language modeling based information discovery approaches. We empirically evaluate our approaches using different IR challenges. Our research shows that supporting information discovery tailored to different search tasks can enhance IR systems' search performance and improve users' search experience.

Relevance is an essential concept in information retrieval (IR) and relevance estimation is a fundamental IR task. It involves not only document relevance estimation, but also estimation of user's information need. Relevance-based language model aims to estimate a relevance model (i.e., a relevant query term distribution) from relevance feedback documents. The true relevance model should be generated from truly relevant documents. The ideal estimation of the true relevance model is expected to be not only effective in terms of mean retrieval performance (e.g., Mean Average Precision) over all the queries, but also stable in the sense that the performance is stable across different individual queries. In practice, however, in approximating/estimating the true relevance model, the improvement of retrieval effectiveness often sacrifices the retrieval stability, and vice versa. In this thesis, we propose to explore and analyze such effectiveness-stability tradeoff from a new perspective, i.e., the bias-variance tradeoff that is a fundamental theory in statistical estimation. We first formulate the bias, variance and the trade-off between them for retrieval performance as well as for query model estimation. We then analytically and empirically study a number of factors (e.g., query model complexity, query model combination, document weight smoothness and irrelevant documents removal) that can affect the bias and variance. Our study shows that the proposed bias-variance trade-off analysis can serve as an analytical framework for query model estimation. We then investigate in depth on two particular key factors: document weight smoothness and removal of irrelevant documents, in query model estimation, by proposing novel methods for document weight smoothing and irrelevance distribution separation, respectively. Systematic experimental evaluation on TREC collections shows that the proposed methods can improve both retrieval effectiveness and retrieval stability of query model estimation. In addition to the above main contributions, we also carry out initial exploration on two further directions: the formulation of bias-variance in personalization and looking at the query model estimation via a novel theoretical angle (i.e., Quantum theory) that has partially inspired our research.