Research On The Identification Method Of News Event Timing Relationship Based On Cross-event

Information extraction is an important problem that increasingly attracts researchers' attention.ACL(Association for Computational Linguistics)sets a branch ACE specially for automatic content extraction.Event relation extraction is extremely valuable in information extraction,it contains temporal relation,causal relation,etc.Temporal relation extraction is widespread used in automatic QA(question and answering)system and automatic summarization system,hence temporal relation extraction become a hot study site all over the world.Corpus is an important element of the measure.Using corpus TempEval-2010corpus of meetings,Chinese temporal relation identification field of the corpus is currently one of the most authoritative corpus.Aiming at the problems that exist in corpus and temporal relation identification task,around corpus,signal word extraction and cross-event temporal relation identification method of the theory of fusion related research,mainly the following characteristics of the work.1.Analysing the characteristics of corpus.Using a corpus base expansion method based on temporal logical inference to handle non-equilibrium problem caused by data sparsity.The number of expanded corpus is 5588.Experiments proved that the total F value increased 0.1%,but precision and recall of "after" ralation increased a lot.2.Signal word is an important element in TimeML label system,it is also plays a key role in temporal relation extraction.This paper aimed at the difficulty of chinese signal word extraction,transformed the signal word identification to sequence labeling,automaticly identified signal word by feature extraction and CRF model construction,and then add signal word,an improtant feature,to mechine learning algorithm for increasing its property.3.Cross-event theory is a new theory which attracts academia's attention since it is proposed in 2011,related research focus on event type identification and missing role rebuilding.This paper bring cross-event theory into temporal relation extraction,choosing feature space and construct maximum entropy classifier to finish sentence—level temporal relation extraction.Setting a threshold value,keep the high probability part as the classification result,then construct document-level classifier to handle the part that with low probability.4.Integration the results of the above research,design and implementation the pro totype system of temporal relation identification.