Discovering unpredictably related words from logs of scholarly repositories for grouping similar queries

As the number of institutional repositories is increasing, more and more people, including non-researchers, are accessing academic contents on them via search engines. User models of non-researchers are not well understood yet, unlike researchers, although non-researchers may use quite different queries from researchers. For understanding their search behavior, it is a good way to categorize search queries of non-researchers into groups. This chapter is devoted to finding related query words at the first step from logs of scholarly repositories. In particular, we try to find words which are related from the viewpoint of non-researchers. In this sense, these words are unpredictably related. A simple method to do this using the access log is that we treat queries which lead to the same paper as related. However, it is challenging because one academic paper generally has a small amount of accesses while accesses to one paper bring many kinds of query words. Instead, we expand relationships between query words and papers, and use a graph-based algorithm in which query words and papers are vertices to find related words. As experiments, we usemore than 400,000 accesses recorded at amajor portal site of Japanese theses, and show that we can find related words with respect to specific disciplines if these words appear frequently. There words seems to be interested in non-researchers and hencewe can’t say they are not related in a usual manner. This result implicates that we can obtain related words if we enrich relationships between technical terminologies using background knowledge, such as dictionaries.