This thesis addresses difficult challenges in distributed document clustering and cluster summarization. Mining large document collections poses many challenges, one of which is the extraction of topics or summaries from documents for the purpose of interpretation of clustering results. Another important challenge, which is caused by new trends in distributed repositories and peer-to-peer computing, is that document data is becoming more distributed. We introduce a solution for interpreting document clusters using keyphrase extraction from multiple documents simultaneously. We also introduce two solutions for the problem of distributed document clustering in peer-to-peer environments, each satisfying a different goal: maximizing local clustering quality through collaboration, and maximizing global clustering quality through cooperation. The keyphrase extraction algorithm efficiently extracts and scores candidate keyphrases from a document cluster. The algorithm is called CorePhrase and is based on modeling document collections as a graph upon which we can leverage graph mining to extract frequent and significant phrases, which are used to label the clusters. Results show that CorePhrase can extract keyphrases relevant to documents in a cluster with very high accuracy. Although this algorithm can be used to summarize centralized clusters, it is specifically employed within distributed clustering to both boost distributed clustering accuracy, and to provide summaries for distributed clusters. The first method for distributed document clustering is called collaborative peer-to-peer document clustering, which models nodes in a peer-to-peer network as collaborative nodes with the goal of improving the quality of individual local clustering solutions. This is achieved through the exchange of local cluster summaries between peers, followed by recommendation of documents to be merged into remote clusters. Results on large sets of distributed document collections show that: (i) such collaboration technique achieves significant improvement in the final clustering of individual nodes; (ii) networks with larger number of nodes generally achieve greater improvements in clustering after collaboration relative to the initial clustering before collaboration, while on the other hand they tend to achieve lower absolute clustering quality than networks with fewer number of nodes; and (iii) as more overlap of the data is introduced across the nodes, collaboration tends to have little effect on improving clustering quality. The second method for distributed document clustering is called hierarchically-distributed document clustering. Unlike the collaborative model, this model aims at producing one clustering solution across the whole network. It specifically addresses scalability of network size, and consequently the distributed clustering complexity, by modeling the distributed clustering problem as a hierarchy of node neighborhoods. Summarization of the global distributed clusters is achieved through a distributed version of the CorePhrase algorithm. Results on large document sets show that: (i) distributed clustering accuracy is not affected by increasing the number of nodes for networks of single level; (ii) we can achieve decent speedup by making the hierarchy taller, but on the expense of clustering quality which degrades as we go up the hierarchy; (iii) in networks that grow arbitrarily, data gets more fragmented across neighborhoods causing poor centroid generation, thus suggesting we should not increase the number of nodes in the network beyond a certain level without increasing the data set size; and (iv) distributed cluster summarization can produce accurate summaries similar to those produced by centralized summarization. The proposed algorithms offer high degree of flexibility, scalability, and interpretability of large distributed document collections. Achieving the same results using current methodologies require centralization of the data first, which is sometimes not feasible.

Automatic text summarization and keyphrase extraction are two interesting areas of research which extend along natural language processing and information retrieval. They have recently become very popular because of their wide applicability. Devising generic techniques for these tasks is challenging due to several issues. Yet we have a good number of intelligent systems performing the tasks. As different systems are designed with different perspectives, evaluating their performances with a generic strategy is crucial. It has also become immensely important to evaluate the performances with minimal human effort. In our work, we focus on designing a relativized scale for evaluating different algorithms. This is our major contribution which challenges the traditional approach of working with an absolute scale. We consider the impact of some of the environment variables (length of the document, references, and system-generated outputs) on the performance. Instead of defining some rigid lengths, we show how to adjust to their variations. We prove a mathematically sound baseline that should work for all kinds of documents. We emphasize automatically determining the syntactic well-formedness of the structures (sentences). We also propose defining an equivalence class for each unit (e.g. word) instead of the exact string matching strategy. We show an evaluation approach that considers the weighted relatedness of multiple references to adjust to the degree of disagreements between the gold standards. We publish the proposed approach as a free tool so that other systems can use it. We have also accumulated a dataset (scientific articles) with a reference summary and keyphrases for each document. Our approach is applicable not only for evaluating single-document based tasks but also for evaluating multiple-document based tasks. We have tested our evaluation method for three intrinsic tasks (taken from DUC 2004 conference), and in all three cases, it correlates positively with ROUGE. Based on our experiments for DUC 2004 Question-Answering task, it correlates with the human decision (extrinsic task) with 36.008% of accuracy. In general, we can state that the proposed relativized scale performs as well as the popular technique (ROUGE) with flexibility for the length of the output. As part of the evaluation we have also devised a new graph-based algorithm focusing on sentiment analysis. The proposed model can extract units (e.g. words or sentences) from the original text belonging either to the positive sentiment-pole or to the negative sentiment-pole. It embeds both (positive and negative) types of sentiment-flow into a single text-graph. The text-graph is composed with words or phrases as nodes, and their relations as edges. By recursively calling two mutually exclusive relations the model builds the final rank of the nodes. Based on the final rank, it splits two segments from the article: one with highly positive sentiment and the other with highly negative sentiments. The output of this model was tested with the non-polar TextRank generated output to quantify how much of the polar summaries actually covers the fact along with sentiment.

No description available.