Functional Hierarchical Search Results Data Analysis

NASA Ames, in collaboration with JSC ISS MOD, has been developing novel systems for searching across multiple heterogeneous data systems in order to facilitate rapid retrieval of relevant information for flight controllers on console and in their off-console duties. Traditional data mining techniques to identify document similarity rely upon brute-force O(n²) cosine similarity of vectors representing text contained in the documents - an expensive and time-consuming operation that becomes unfeasible on large data sets such as those in use by the Shuttle and Station programs at NASA. This paper details a technique developed as part of this activity to identify documents similar to documents relevant and of interest to the user by leveraging the prior search results to produce a weighted graph of document relationships based upon search terms. This paper details the algorithms used, the user interface prototypes developed to evaluate the usefulness of similarity to the browsing activities of users, and the evaluation of the accuracy of this technique as compared with traditional data mining techniques.