We propose a novel method of constructing causal networks to clarify the relationships among events. Since events and their relationships continually change, our method works in an incremental manner. There are two problems in the conventional methods of constructing causal networks that use keywords representing events: 1) similar events detection to network construction is a time-consuming task, and 2) because the merge operation depends on appearing order of events there is a consistency issue in the incremental construction. In this paper, as the representation model, we propose a Topic-Event Causal network model (TEC model) in which the topic and details of an event are separately represented by using structured keywords. We cluster events by using the topic keyword and then detect similar events per each cluster. This fashion will reduce the comparison times of events. When we compute the similarity of two events in a topic, since we compare the pair of SVO tuples of vertices based on WordNet, we solve the difference in the word used and the lexical ambiguity and keep the consistency with a causal network. We also show experimental results to demonstrate its usefulness.
[1]
Satoshi Sekine,et al.
Extracting Causal Knowledge Using Clue Phrases and Syntactic Patterns
,
2008,
PAKM.
[2]
Graeme Hirst,et al.
Lexical chains as representations of context for the detection and correction of malapropisms
,
1995
.
[3]
Masahide Horita,et al.
ASSESSING THE PLAUSIBILITY OF INFERENCE BASED ON AUTOMATED CONSTRUCTION OF CAUSAL NETWORKS USING WEB-MINING
,
2006
.
[4]
Yuji Matsumoto,et al.
Acquiring causal knowledge from text using the connective marker tame
,
2005,
TALIP.
[5]
James Allan,et al.
Finding and linking incidents in news
,
2007,
CIKM '07.
[6]
Roxana Gîrju,et al.
Automatic Detection of Causal Relations for Question Answering
,
2003,
ACL 2003.
[7]
Christiane Fellbaum,et al.
Book Reviews: WordNet: An Electronic Lexical Database
,
1999,
CL.
[8]
Hiraku Okada,et al.
Technical Report of IEICE
,
2000
.
[9]
Yuji Matsumoto,et al.
Japanese Dependency Analysis using Cascaded Chunking
,
2002,
CoNLL.