Geo-mining: Discovery of Road and Transport Networks Using Directional Patterns

Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 267–275, Singapore, 6-7 August 2009. c⃝2009 ACL and AFNLP Geo-mining: Discovery of Road and Transport Networks Using Directional Patterns Dmitry Davidov ICNC The Hebrew University of Jerusalem dmitry@alice.nc.huji.ac.il Ari Rappoport Institute of Computer Science The Hebrew University of Jerusalem arir@cs.huji.ac.il Abstract One of the most desired information types when planning a trip to some place is the knowledge of transport, roads and geographical connectedness of prominent sites in this place. While some transport companies or repositories make some of this information accessible, it is not easy to find, and the majority of information about uncommon places can only be found in web free text such as blogs and fo- rums. In this paper we present an algo- rithmic framework which allows an auto- mated acquisition of map-like information from the web, based on surface patterns like “from X to Y”. Given a set of loca- tions as initial seeds, we retrieve from the web an extended set of locations and pro- duce a map-like network which connects these locations using transport type edges. We evaluate our framework in several set- tings, producing meaningful and precise connection sets. 1 Introduction Textual geographical information such as location descriptions, directions, travel guides and trans- port tables is extensively used by people. Dis- covering such information automatically can as- sist NLP applications such as question answering (Santos and Cardoso, 2008), and can be useful for a variety of other applications, including au- tomatic map annotation. Some textual geographical information can be found in web sites of transport companies, tourist sites and repositories such as Wikipedia. Such sites usually utilize structured information such as machine-readable meta-data, tables, schedule forms or lists, which are relatively convenient for processing. However, automatic utilization of such information is limited. Even on these sites, only a small fraction of the available geographi- cal information is stored in a well-structured and freely accessible form. With the growth of the web, information can be frequently found in ‘or- dinary’ web pages such as forums, travelogues or news. In such sites, information is usually noisy, unstructured and present in the form of free text. This type of information can be addressed by lexical patterns. Patterns were shown to be very useful in all sorts of lexical acquisition tasks, giv- ing high precision results at relatively low com- putational costs (Pantel et al., 2004). Pattern- driven search engine queries allow to access such information and gather the required data very effi- ciently (Davidov et al., 2007). In this paper we present a framework that given a few seed locations as a specification of a region, discovers additional locations (including alternate location names) and map-like travel paths through this region labeled by transport type labels. The type of output produced by our framework here differs from that in previous pattern-based studies. Unlike mainstream pattern-based web mining, it does not target some specific two-slot relationship and attempts to extract word tuples for this …

Content truncated to 500 words