Identifying Documents and Experts in Corporate Environments

This was my Ph.D. project. It began in February 2004, and the dissertation was defended in May 2009 under the supervision of Professor Edson Satoshi Gomi. The title of the dissertation is: "Information Retrieval of Documents and People in Corporate Environments through Decision Trees".

Thesis abstract

This work evaluates decision trees as ranking functions for documents and people in corporate environments. Relevant attributes of the entities to retrieve were identified by analyzing: (i) the dynamics of information production and consumption in a corporate environment; (ii) algorithms in the literature for retrieving documents and people; and (iii) concepts used in ranking functions for general domains. An evaluation environment based on the Cerc reference collection was built to assess the applicability of the C4.5 algorithm to produce ranking functions for the corporate domain. C4.5 was partially effective: it did not yield good results for document retrieval, but made it possible to control the ranking function to optimize precision at the top of the ranking or mean average precision (MAP). For people retrieval, C4.5 produced a decision tree that outperformed all other evaluated ranking functions, with MAP of 0.83 versus an average of 0.74. The decision tree also helped explain the attributes used to characterize documents and people. Its analysis showed that a person is considered an expert on a topic when they appear in many documents, appear frequently within them, and those documents are highly relevant to the query.