ABSTRACT
Online forum is a web space for people to share their knowledge and experiences. Discussions in online forum often refer to particular geographical locations and places. To extract the geographical knowledge embedded in these conversations, the first step is to understand the geographical references in the content. Although this problem has been investigated, online forum data presents new challenges to existing geotagging approaches. In particular, the linguistic context of online forum discussion threads available for disambiguating vague geographical references is not well understood and is not exploited in existing geocoding algorithms.
In this thesis, we analyzed a number of episodes of online forum discussions to understand the intricacies of geocoding disambiguation of geotagging and the inadequacy of existing approaches in resolving geographical references in online forum discussion thread. We found that the development of forum discussion is closely related to its participants’ intentions and attention’s saliency, which are constantly changing during the course of the discussion. These characteristics of online forum discussion have not been considered in the geocoding disambiguation researches. Based on this finding, we propose a method that replies upon the intentional discourse model to improve existing geocoding approaches’ performance on online forum discussion contents.
CHAPTER 1 INTRODUCTION
1.1 MOTIVATION
The rise of online social media has brought an information explosion to the 21th century and has changed the way people communicate and share information. Online forums are a form of social media and often have geographic references embedded within, especially when the discussion involved spatial related topics [61] . Furthermore, online forum threaded discussion characterizes its reader in terms of communities that are relevant to specific geographic locations.
Online forums can be important sources of information as their discussion content accumulate for long periods of time and from contributors with varied knowledge and expertise[12] . In order to make use of these discussion contents, we must first understand the information presented in it. Our interest is to extract and retrieve the geographical information embedded in them. This process is known asgeotagging. The process of automatic geotagging involves identifying all textual reference to real world geographic entities, locations and toponyms, with assigning the corresponding geospatial coordinate values to these references [3,37,45,60] . However, automatic geotagging turns out to be challenging because many location references (or toponyms) are ambiguous, e.g. toponyms have same names. There are geocoding methods working on resolved the ambiguities of toponyms using various geocoding heuristics. However, these geocoding methods did poorly to disambiguate the toponyms in online forum discussion threads as we demonstrated the inadequacy of these methods in chapter three.
One of the motivations for conducting this research is to support the deliberative dialogue on spatially defined issues. Research in field aims to expand public participation and civic engagement of local communities in decision making of spatial related problems[61] , e.g., spatial planning[32,50] . Participant engages deliberative dialogues often facilitated by the professional moderator. Meanwhile, the geodeliberation participants need to share their personal knowledge and experience to others. However, these shared experience and knowledge contains toponyms and location references with vagueness. The vagueness is due to the participant’s heterogeneous conception of space [35,76,77] and usage of spatial language [41,85] , which brings additional difficulties for others to understand the spatial concepts and location references appeared in discussion. For example, the participants may use different cognitive reference points[78] to describe the community they live, which is hard to be understand by others who are not familiar with (e.g. “the construction plan will risk people who lives inCapital building area).
Automatic geotagging with geocoding disambiguation reduces the difficulties and efforts for online forum discussion participants to understand the spatial meaning in discussion. On the other hand, the facilitator of geodeliberation can benefit from this improved geocoding disambiguation method as well. For example, facilitators can monitor the focus of the deliberative discussion using the toponyms appeared in discussion threads, and provide related information and materials to the participants to improve the quality of deliberation.
In this thesis, we adopted the characteristics of online forum discussion thread with existing geocoding heuristics to improve the toponym disambiguation’s accuracy. There are several potential applications that could benefit from improved geocoding disambiguation in online forum:
- Geographical Information Retrieval: Successful geocoding disambiguation can increase the precision of information retrieval from online forums by enabling a search by location function. For example, search for all the discussions related to state of Paris, Texas, rather than Paris, France with the textual reference “Paris”.
- Event Detection and Tracking: Assuming the toponyms have been extracted from discussion thread and spatially grounded to the correct real world location, it can
help not only to identify events from discussion contents, but also organize these events base on their locations. This may allow for tracking in both space and time. These geotagged event with explicit locations can be used for spatial pattern extraction, such as crime pattern identification [27] .
The goal of this thesis is to advance the research in geocoding disambiguation with a new heuristic. To achieve this goal, we applied the intention based discourse theory [23] to model the development of discussion thread in online forum. We hypothesized that the discussion thread’s interruption discourse segment is the main reason underlying the inadequacy of existing geocoding methods. We adopted the concept of focusing space stack from the same discourse theory[23] to provide a new heuristic for geocoding of online forum discussion content.
1.2 STATEMENT OF THE PROBLEM
Existing geocoding disambiguation strategies are on the assumption that all the toponyms appeared in the same document’s context are related. However, this assumption could be in under the environment of online forum threaded discussion. This is because online forum discussion threads are composed by multiple authors with various background and interests. Meanwhile, there are very few geocoding researches focusing on geotagging with geocoding disambiguation of the contents of online forum threaded discussion[6,61] . In addition, online forum threaded discussion has unique characteristics that makes it different to the document studied in existing geocoding methods, such as news article and reports. The characteristics of online threaded discussion forum, as summarized from the information retrieval research work[14,26,33,81] , are:
- Online forum’s discussion threads are composed of individual posts from discussion participants with heterogeneous knowledge background and experience
- A single discussion post is oftentimes related to other posts in the same discussion thread as a reply post
- The reply posts in a discussion thread could be separated in time to which it responds, leading to a fragmented discussion structure and therefore not all the reply posts are related to the root post. Thus, adjacent reply posts might not be semantically related, and this phenomenon is termed as “discourse disentanglement”[19,34,42] and
“interruption”[23] .
These characteristics have not been exploited in existing geocoding disambiguation researches. Existing geocoding researches focusing on resolve the toponyms using the geographical context only (i.e. the other toponyms in the same document), but ignore the intrarelationship between the discussion posts in the same thread that related to the participants’ intention. In this thesis we are aiming on exploring impact of the discussion participant’s intention and attention on geocoding disambiguation in online forum discussion thread. To address this research goal, we raise several research questions that need to be answered.
- Are existing geocoding methods adequate to disambiguate toponyms in online forum discussion thread?
Existing geocoding methods do not take the characteristics of discussion threads into the toponyms disambiguation. On the other hand, online forum discussion threads has characteristic that may violated the assumption that all the toponyms appeared in the same document are related to each other, such as the existence of “ discourse interruption”[23] . It is important to understand how the interruptions in discussion thread influencing existing geocoding disambiguation methods.
- How to model the online forum discussion thread using participants’ intention to solve the difficulties of existing geocoding methods encounter?
The constantly changing of intention and attention of discussion participants during the development of a discussion thread caused the discourse interruption, which could possibly violates the assumption taken by existing geocoding methods that all the toponyms appeared in the same document are related to each other. Therefore, the discussion participants’ intention, which is an integral component of discussion thread, needs to be considered and combine with geocoding methods for identifying the interruptions in discussion thread and avoid mixing the geographical context from main discussion and interruptions in toponyms’ geocoding disambiguation.
1.3 THESIS STRUCTURE
The rest of this thesis is structured as follows. In chapter two we reviewed the background and related work of geotagging, with both geoparsing and emphasizing on geocoding disambiguation methods and heuristics. Four major automatic geocoding disambiguation methods and their related research works have been reviewed.
In chapter three we presented two scenarios of discussion thread to illustrate the characteristics of online forum threaded discussion. The two scenarios are built based on real world online forum’s discussion thread with the interruption phenomenon. The purpose for these two scenarios is to asses existing geocoding methods’ accuracy and their inadequacy on online forum discussion thread. Through the assessment, we claim that the existence of interruption caused the inadequacy of geocoding methods in discussion thread’s toponym disambiguation.
In chapter four we introduced the theory of modeling online forum threaded discussion with its participants’ intention structure and attentional state, based on Grosz and Sidner’s work of intention based discourse modeling[23] . We discussed the application of using discourse’s attentional state’s focusing space stack to model the interruption in discussion thread and therefore to improve existing geocoding methods’ accuracy by separating unrelated toponyms appeared in the same document.
The final chapter reviewed what we have done in this thesis, and discussed potential future research.
GEOCODING DISAMBIGUATION FOR ONLINE FORUM DISCUSSION THREAD