Extracting Temporal Patterns From Large-Scale Text Corpus

Knowledge, in practice, is time-variant and many relations are only valid for a certain period of time. This phenomenon highlights the importance of designing temporal patterns, i.e., indicating phrases and their temporal meanings, for temporal knowledge harvesting. However, pattern design is extremely laborious and time consuming even for a single relation. Therefore, in this work, we study the problem of temporal pattern extraction by automatically analysing a large-scale text corpus with a small number of seed temporal facts. The problem is challenging considering the ambiguous nature of natural language and the huge amount of documents we need to analyse in order to obtain highly representative temporal patterns. To this end, we introduce various techniques, including corpus annotation, pattern generation, scoring and clustering, to reduce ambiguity in the text corpus and improve both accuracy and coverage of the extracted patterns. We conduct extensive experiments on real world datasets and the experimental results verify the effectiveness of our proposals.