Characteristics of human route descriptions and their improvement by machine landmark extraction

摘要

Any commercial sites, be it a hotel, a restaurant, or a shop, have to be reached properly by potential customers, and thus an adequate information on the route to the site is important. Route information is represented as a textual description and/or as a map, and in either case, adequate mentions of landmarks, a building or a site that is located in the midst of the route which plays an intermediate marker, are essential. However, human route descriptions or maps based on them may not contain a sufficient number of landmarks that is needed for potential customers, and an automatic determination of necessary landmarks is desirable. In this study, as a preliminary one, we collected more than thousands of route descriptions written by commercial sites, and made an extensive observation to find characteristics of textual route descriptions. Based on the observation, we conducted two experiments: to automatically extract potential landmarks from textual route descriptions and to evaluate the legitimacy of them using a simulative agent. The result indicates that more than half of the textual route descriptions accumulated needs improving, and the extracted potential landmarks can be effectively added for improving existing route descriptions.