Idea navigation is a first-of-a-kind approach to textual information access that uses semantic relationships between terms and supports exploratory search by providing a view of the most common ideas in a given result set. This paper shows how this strategy can effectively solve realistic search tasks and evaluates its efficiency.
The main objective of the reported research is a next-generation interface for exploratory search that makes use of linguistic knowledge about subject-verb-object relationships to construct queries (that is, to identify ideas that are found in the searched texts). The proposed method is based on the premise that the usual search box is insufficient for supporting many common information-seeking tasks, and it is often complemented by faceted browsing interfaces that offer users the option to narrow their queries to available, relevant choices (facets). This is very convenient for exploratory search tasks, where the offerings of the searched collection are unknown. The idea navigation approach allows the creation of facets for abstract and subjective concepts, common particularly to scientific, legal, and patent searches, such as, “Find molecules that target a particular cell,” or “Find inventions that burn solids for locomotion.”
The approach is demonstrated through a prototype, a user interface that provides enough supporting functionality to allow users to answer real questions, in order to evaluate the usability of the idea navigation module. The interface displays the subject-verb-object relationships available in the searched texts based on an initial nominal query (taking the subject position) and the number of matches for each proposed combination. In addition, the interface offers the capability to search over all ideas in the collection, allowing the use of search and idea navigation together, to narrow the results. The evaluation of the prototype is reported through a formative evaluation, to see whether users would: understand the navigation interface after a brief introduction; choose to use that interface, when given the option alongside a standard search box; and successfully complete tasks with its help.
The evaluators were given tasks inspired by the kinds of queries that are not fully supported by search boxes and faceted metadata, such as: “In October 2000, a Yankee pitcher named Roger Clemens threw a bat at an opposing player. Find the opposing player’s name”; or “Find something about George W. Bush in October 2000 that is interesting to you in the context of the modern day”; or “Find a quotation that could be considered controversial or offensive.” A regular search box was also available to evaluators.
The test results show that evaluators adopted the idea navigation over time. Overall, 79 percent of the tasks were completed with idea navigation as the final search step, thus demonstrating that idea navigation indeed provides a significant improvement over standard search boxes for the types of queries tested. Furthermore, the precision of the detection of subject-verb-object ideas of the prototype is 100 percent. The recall is less, but the authors do not provide actual figures for the recall levels.
This is a descriptive paper, without much technical detail, showcasing an application and its test results. It emphasizes future work that hints at usability in industrial applications. It outlines a promising natural next step in accessing information from texts, as exploratory search and idea navigation principles are a very plausible bridge between current technological capabilities and the users’ demands for more powerful means of human-computer interaction. Defining search terms with subject-verb-object relationships has already been explored in industrial settings in the past year by Powerset’s Powerlabs search engine.
